The Crosslinguistic Study of Language Acquisition: Theoretical Issues [2]

Citation preview

THE CROSSLINGUISTIC STUDY OF LANGUAGE ACQUISITION Volume 2: Theoretical Issues Edited by DAN ISAAC SLOBIN University of California, Berkeley

2

Copyright © 1985 by Lawrence Erlbaum Associates, Inc. All rights reserved. No part of this book may be reproduced in any form, by photostat, microform, retrieval system, or any other means, without the prior written permission of the publisher. Lawrence Erlbaum Associates, Inc., Publishers 365 Broadway Hillsdale, New Jersey 07642

Library of Congress Cataloging-in-Publication Data Main entry under title:

3

The crosslinguistic study of language acquisition. Includes bibliographies and indexes. Contents: v. 1. The data — v.2. Theoretical issues. 1. Language acquisition. I. Slobin, Dan Issac 1939– P118.C69 1985 401′.9 85-27411 ISBN 0-89859-367-0 (set) Printed in the United States of America 10 9 8 7 6 5 4

4

Contents Volume 2: Theoretical Issues Format and Abbreviations for Glosses 11. Cognitive Prerequisites: The Evidence from Children Learning English Judith R. Johnston Introduction Thought as the Content of Language Language and the Structure of Thought Cognitive Prerequisites and Crosslinguistic Studies: Theoretical and Methodological Guidelines Coda References 12. Function, Structure, and Language Acquisition Talmy Givón Preamble 5

Some Reflections on “Error” Analysis Typology of What: Reflections on the Input Typology and Functional Domains Structure vs. Function in Language Acquisition Syntactic Complexity Cultural Context References 13. Language Segmentation: Operating Principles for the Perception and Analysis of Language Ann M. Peters Introduction Specific OPs for Extraction and Segmentation Related Issues Research Questions Relevant to Segmentation Heuristics References Appendix: Condensed List of OPs 14. Hungarian Language Acquisition as an Exemplification of a General Model of Grammatical

6

Development Brian MacWhinney Descriptive Sketch of Hungarian Sources of Data The Overall Course of Development The Competition Model Basic Operating Principles of the Competition Model Competition in the Lexicon Applications of the Model to the Development of Phonology Applications of the Model to the Acquisition of Syntax Applications of the Model to the Acquisition of Lexical Semantics Suggestions for Further Study Theoretical Implications References 15. Crosslinguistic Evidence Language-Making Capacity Dan I. Slobin

7

for

the

Operating Principles for the Construction of Language Basic Child Grammar Attention to Speech Entering and Tagging Information in Storage Semantic Space and Grammatical Morphemes Grouping Information in Storage General Problem-Solving Strategies Strategies for Grammatical Organization of Stored Information Conclusion References Appendix: Operating Principles for the Construction of Language 16. What Shapes Children’s Grammars? Melissa Bowerman The Processing Approach in Historical Perspective Some Methodological Problems Testing Proposed Operating Principles

8

Basic Child Grammar: Is Only One Way of Structuring Semantic Space “Basic”? References Subject Index Author Index

Volume 1: The Data

Introduction: Why Crosslinguistically? Dan 1. Slobin

Study

SPOKEN LANGUAGES 1. The Acquisition of English Jill G. de Villiers and Peter A. de Villiers 2. The Acquisition of German Anne E. Mills 3. The Acquisition of Hebrew Ruth A. Berman

9

Acquisition

4. The Acquisition of Japanese Patricia M. Clancy 5. The Acquisition of Kaluli Bambi B. Schieffelin 6. The Acquisition of Polish Magdalena Smoczyńska 7. The Acquisition of Romance, with Special Reference to French Eve V. Clark 8. Variation and Error: A Sociolinguistic Approach to Language Acquisition in Samoa Elinor Ochs 9. The Acquisition of Turkish Ayhan A. Asku-Koç and Dan I. Slobin SIGNED LANGUAGES 10. The Acquisition of American Sign Language Elissa L Newport and Richard P. Meier

10

Format and Abbreviations for Glosses*

All foreign language examples are given in Italics. (Small caps are used for emphasis and other usual functions of Italics.) In running text, English glosses and grammatical codes are given in single quotes, and optional free translations follow in parentheses, indicated by an equal sign and single quotes. Grammatical codes are always given in capital letters (see list, below). For example: gel-me-di-n ‘come-NEG-PAST-2SG’ (= ‘you didn’t come’).

11

In interlinear format, translation equivalents appear below each foreign element, and the free translation is placed below in single quotes: gel -me -di -n come NEG PAST 2SG ‘you didn’t come’ Hyphens in a gloss always correspond to hyphens in the foreign example. If one foreign element corresponds to more than one English element and/or grammatical code, the collection of meaning equivalents is joined by colons; e.g. gel-medin ‘come-NEG:PAST:2SG’, or even gelmedin ‘come:NEG:PAST:2SG’. If it is relevant to indicate the possibility of segmentation, equal signs can be used in place of colons. The preceding example consists of segmentable morphemes, and could also be glossed, for example, as gel-medin ‘come-NEG+PAST+2SG’. Use of colons is neutral in regard to the possibility of segmentation, and in most instances either colons or hyphens are used. (The degree of precision of segmentation and glossing of an example, of course, depends on the role it plays in the exposition.) If a grammatical code consists of two words or abbreviations they are joined by a period; e.g. DEF.ART means “definite article.” Combining the principles for use of colons and periods, consider the gloss for the German definite article in its masculine singular accusative form: den ‘DEF.ART:MASC:SG:ACC’.

12

LIST OF GRAMMATICAL CODES 1 First Person 2 Second Person 3 Third Person ABESS Abessive (‘without X’) ABL Ablative (‘from X’) ABS Absolutive ACC Accusative ACT Active ADESS Adessive (‘toward X’) ADJ Adjective, Adjectival ADV Adverb(ial) AFFIRM Affirmative AGR Agreement AGENT Agent ALLAT Allative (‘to(wards) X’) AN Animate

13

AORIST Aorist ART Article ASP Aspect AUG Augmentative AUX Auxiliary BEN Benefactive CAUS Causative CL Clitic CMPLR Complementizer COMIT Comitative (‘(together) with X’) COMPAR Comparative COMPL Completive CONC Concessive COND Conditional CONJ Conjunction CONN Connective CONSEC Consecutive

14

CONT Continuous, Continuative COP Copula DAT Dative DECL Declarative DEF Definite DEICT Deictic DEM Demonstrative DESID Desiderative DIM Diminutive DIREC Directional DO Direct Object DYN Dynamic (Nonstative) ELAT Elative(‘out of X’) EMPH Emphatic ERG Ergative ESS Essive (‘as X’) EVID Evidential

15

EXCL Exclusive EXIS Existential FACT Factive FEM Feminine FIN Finite FOC Focus FUT Future GEN Genitive HAB Habitual HON Honorific HUM Human ILL Illative (‘into X’) IMP Imperative INAN Inanimate INCH Inchoative INCL Inclusive INDEF Indefinite

16

INESS Inessive (‘in X’) INF Infinitive INFER Inferential INSTR Instrumental INT Interrogative INTENT Intentive INTERJ Interjection INTRANS Intransitive IO Indirect Object IPFV Imperfective IRR Irrealis ITER Iterative LOC Locative MASC Masculine MOD Modal N Noun NEG Negative

17

NEUT Neuter NEUTRAL Neutral NOM Nominative NOML Nominal NONPAST Non-past NONVIR Non-virile NUM Numeral, Numeric OBJ Object OBLIG Obligatory OPT Optative PART Participle PARTIT Partitive PASS Passive PAST Past PAT Patient PERF Perfect PFV Perfective

18

PL Plural POL Polite POSS Possessive POST Postposition POT Potential PP Past Participle PRE Prefix PREP Preposition PRES Present PRESUM Presumptive PRO Pronoun PROG Progressive PROL Prolative (‘along X’) PTL Particle PURP Purposive Q Question QUANT Quantifier

19

QUOT Quotative RECENT Recent RECIP Reciprocal REFL Reflexive REL Relative REM Remote REPET Repetition REPORT Reportative RES Resultative SG Singular SIMUL Simultaneous STAT Stative SUBJ Subject SUBJV Subjunctive SUBL Sublative (‘onto X’) SUFF Suffix SUPER Superessive (‘on X’)

20

SUPERL Superlative TEMP Temporal TNS Tense TOP Topic TRANS Transitive TRANSL Translative (‘becoming X’) V Verb VIR Virile VN Verbal noun VOC Vocative *The abbreviations are adapted from a list used by Bernard Comrie (The languages of the Soviet Union, Cambridge University Press, 1981, p xv). The format is based on useful suggestions offered by Christian Lehmann in “Guidelines for interlinear morphemic translations: A proposal for a standardization” (Institut für Sprachwissenschaft, Universität Köln, Arbeitspapier Nr. 37, 1980). The system presented here is offered as a proposal for standardization in child language studies.

21

VOLUME 2: THEORETICAL ISSUES

22

23

11

Cognitive Prerequisites: The Evidence from Children Learning English

Judith R. Johnston Indiana University

Contents Introduction Thought as the Content of Language Linguistic Evidence of Conceptual Constraints Nonlinguistic Evidence of Conceptual Constraints Locating Evidence of Prerequisite Knowledge Summary

24

Language and the Structure of Thought Use of the Passive and Operative Thought Conclusion and Summary Cognitive Prerequisites and Crosslinguistic Studies: Theoretical and Methodological Guidelines Diversity in Language-Thought Relationships Appropriate Candidates for Control Variables Assessment of Conceptual Prerequisites Coda

1. INTRODUCTION Children use language to communicate ideas, to direct attention, to symbolize imagined worlds, and to reason about complex states and events. It is scarcely surprising then, that the child language literature is filled with claims about the relationships between language development and cognition. The intimate ties between language and thought invite such claims; the facts of language acquisition demand them. We have seen “strong” cognitive hypotheses, i.e., those which assert that the facts of cognitive development are sufficient to explain language learning, “weak” cognitive hypotheses, i.e., those which assert that the facts of cognitive

25

development only partially explain language learning, and even recent questions as to whether nonlinguistic cognitive achievements “promote or determine or account for language in any way” (Dore, 1979, p. 129; also, Erreich, Valian, & Winzemer, 1980). Language is obviously the product of complex mental processes and is thus in some sense cognitive. There is, however, no way to characterize language learning by exclusive appeal to nonverbal mechanisms or knowledge. Our theories of language acquisition must take into account the nature of the language object as it resists and shapes the child’s organizing efforts, and the nature of the child’s prior linguistic knowledge as it leads to further acquisition and systematization of rules. We need no longer ask whether language learning is a cognitive phenomenon, nor whether language learning may be explained solely by non-linguistic constructs. The challenge instead is to identify the specific points, if any, at which developments in nonverbal cognition help determine the course of language acquisition. The issue of cognitive determinism is particularly relevant to crosslinguistic investigations of language acquisition. As outlined by Slobin (1973) the research strategy involves comparing children learning different languages to see which types of linguistic patterns are “easy” or “difficult” to learn. From this characterization of language patterns we can then infer the nature of the rule discovery processes themselves. The notion of COGNITIVE PREREQUISITE is an integral and crucial component of this methodology. Before we can attribute 26

differences in success or speed of rule discovery to the specific nature of the linguistic patterns being learned, we must be able to argue that children come to their learning tasks with comparable mental resources. Slobin suggests that the pertinent resources are of two sorts: the conceptual and factual knowledge which gives rise to communicative intentions, and the cognitive processing mechanisms which participate in rule formation. The latter are assumed to remain constant and are in fact the object of study; the former are assumed to change with experience and intellectual growth, and hence must be treated as control variables. Slobin’s original plan was to match children at the point at which they first attempted to communicate a given notion, and then to observe the period during which they worked out the linguistic patterns which encode that notion in their particular language. Patterns which took longer to acquire were to be considered as more difficult. In practice, this point-of-first-intention strategy has been difficult to operationalize, and crosslinguistic studies to date have tended to rely instead on simple age-matching. If it is true that specific nonverbal cognitive achievements promote or determine the discovery of specific linguistic patterns, age is at best an approximate index of the true critical variables. The purpose of this chapter is thus twofold, to assess the current viability of “cognitive” hypotheses, and to indicate the implications of this assessment for crosslinguistic methodology. If we discover that aspects of nonverbal cognition indeed regulate linguistic discovery, it may be possible to match children not only 27

at the point of first intention, but earlier, when nonverbal performance reveals the existence of prerequisite concepts and mechanisms. If such evidence does not exist, we may need to revise our approach to crosslinguistic studies and our understanding of the acquisition process itself. Since I plan to pursue these goals on several fronts, a general orientation to the chapter may prove helpful. The macrostructure is provided by two specific and quite different cognitive hypotheses: first, the claim that children only can talk about what they already understand, and secondly, the claim that children’s linguistic rules can only be as complex as their cognitive structural capacity will allow. For each of these hypotheses I survey accounts of children learning English and indicate the types of evidence which appear to be supportive. Selected pieces of this evidence are discussed in some depth, utilizing studies of cognitive development and studies which explicitly test language-thought dependencies. These sections will focus on children’s learning of spatial locatives, more/ less, dimensional adjectives, and the passive voice. Finally, I return to the general issue of cognitive determinism and its relationship to crosslinguistic methodology.

2. HYPOTHESIS ONE: THOUGHT AS THE CONTENT OF LANGUAGE Current views of language acquisition hold that children come to the language learning task with both factual and

28

conceptual resources; that this knowledge is necessary if they are to interpret the utterances they hear and establish the mapping from meaning to form; and further, that it is the desire to express new ideas which energizes the search for specific linguistic patterns. Implicit in this view is the strong claim that language acquisition is constrained by the child’s available conceptual and factual knowledge. The child, for example, who does not understand the relationship of support or proximity should not be able to discover what in or next to mean because he cannot interpret an object configuration in those terms. He is thus unlikely to use the words in or next to, but if he does, he will use them to express non-adult meanings that lie within his scope of knowledge. The relationship between language and thought being proposed here is clearly that of symbol to ideational content—a relationship which implies both logical and temporal asymmetry. Unless language is viewed as potential nonsense, conceptual notions must be acquired prior to their verbal expression.

2.1. Linguistic Evidence of Conceptual Constraints Observations of children learning English lend support to this view of language acquisition. The pertinent evidence is of several different sorts, some involving direct tests of language-thought asynchrony, some only indirect. I begin this review by characterizing and illustrating five types of child language evidence which indirectly implicate the role of conceptual development in language learning. 29

2.1.1. Nonlinguistic Comprehension Strategies A number of recent studies suggest that children who have not yet mastered a given language pattern make use of nonverbal conceptual and factual knowledge to interpret the meaning of utterances. (a) Two-year old children interpret semantically reversible SVO sentences according to their estimate of specific event probability while 3-year old children use word order rules (Chapman & Kohn, 1978; Strohner & Nelson, 1974; Bever, 1970 a). (b) Young children interpret instructions to place objects in, on, or under according to their understanding of the relational implications of reference object properties or according to their knowledge of conventional and specific object relations. If the reference object has a flat, supporting surface, young children tend to place a second object ON it; if it is a container, the second object is placed IN it, etc. (E. Clark, 1973a, 1977; Cook, 1978). Alternatively, if the reference object is a bridge, boats are placed UNDER it; if it is a house, roofs are placed ON it (Hoogenraad, Grieve, Baldwin, & Campbell, 1978; Wilcox & Palermo, 1974/5; Cook, 1978). (c) Before mastering temporal and causal conjunctions, children use order-of-mention to determine relational meaning, In the absence of contextual cues, first-mentioned events are assumed to be temporally prior (E. Clark, 1971;Coker, 1978; Barrie-Blackley, 1973; Johnson, 1975; Bever, 1970b; but also, French & Brown, 1977) and/or causal (Bebout, Segalowitz, &

30

White, 1980; Emerson, 1979). Such interpretations seem to entail some nonverbal understanding of temporal order (See Bullock & Gelman, 1979, and Brown & Murphy, 1975, for evidence that 3- to 5-year-olds also use temporal order judgments in nonverbal causal and memory tasks.) (d) In certain comprehension tasks where they are asked to judge quantities as being more or less, children respond to less by choosing the greater quantity (Donaldson & Balfour, 1968; Palermo, 1973; Holland & Palermo, 1975; Weiner, 1974). Although earlier interpretations of this finding invoked linguistic factors such as erroneous or partial semantic mapping (Donaldson & Wales, 1970; E. Clark, 1973b), more recent arguments assume some nonlinguistic bias towards objects/sets with greater extent (E. Clark, 1977; Carey, 1978b).1 The role of nonlinguistic comprehension strategies in the acquisition of new linguistic patterns is not well understood (Cromer, 1977). In some cases, e.g., the probable-event strategies, the information used in the initial comprehension of a linguistic form is only distantly related to the eventual meaning of that form. Even when a strategy does entail conceptual notions which are later to be incorporated into the adult rule, the child must still at some point formulate that rule. Early comprehension strategies may or may not facilitate this latter step. In any case, these findings clearly indicate the role of conceptual and factual knowledge in language use and show that this knowledge precedes the mastery of linguistic form. 31

2.1.2. Immature Expression of New Intentions Young children’s utterances frequently suggest that they know what they want to say but have not yet learned how to say it. This gap between evident intention and mastery of linguistic means is particularly striking when children resort to idiosyncratic forms, but it can also be seen when they draw on familiar, standard forms to express new meanings agrammatically. Observations of this sort are open to debates over “rich interpretation,” but they do suggest the developmental priority of conceptual content. (a) Some children’s first locative expressions consist of a general-purpose functor such as schwa /ə/ or syllabic n/n/ (E. Clark, 1978). (b) When children lack a label for some action or object category, they often invent one. E. Clark (1980) presents numerous examples of children making use of English word formation devices to fill lexical gaps, e.g., using denominal verbs such as cracker ‘to put crackers in’ or scale ‘to weigh’ and compound agentive nouns such as garden-man or fix-man. (c) When Adam and Sarah began to talk about the current relevance of past events they first used past tense forms along with the adverbs now or yet, forms they had known for some time, rather than using the perfect tense (Cromer, 1974), e.g. You didn’t peek yet?

32

2.1.3. Simultaneous Emergence of Meanings across Form A given concept may participate in the meaning of more than one language form. If conceptual achievements prompt linguistic discovery, we might expect the various expressions of a new notion to be acquired in near synchrony regardless of formal diversity. Although studies of child language are usually too focused on particular forms to yield evidence of this sort, a few cases have been reported. (a) Cromer (1974) observes that Adam and Sarah first used “present tense” forms ø or -s,to express notions of habitual aspect or timelessness at about the same time as the adverbs always or never were first used to express similar ideas. (b) In Hood and Bloom’s (1979) study of causal expressions, children’s first productive use of causal conjunctions tended to occur at about the same time as their first why questions. (c) About the time that children begin to use periphrastic causative constructions, they also overregularize certain verbs and use them as if they had causal meanings, e.g. Fall the block ‘Make the block fall’ (Bowerman, 1974).

2.1.4. Protracted Acquisition despite Formal Similarity Members of certain closed linguistic form classes, e.g. prepositions, appear to be acquired over a span of years rather than simultaneously as their common syntactic 33

properties might suggest. Moreover, they are acquired in largely consistent orders and at similar ages across children. This protracted and regular course of acquisition may well reflect the gradual emergence of conceptual prerequisites. (a) In the auxiliary class, forms of BE precede HAVE by some two years despite the availability of the surface form have in its main verb function (Cromer, 1976; Brown, 1973; Nussbaum & Naremore, 1975; Menyuk, 1969). (b) Studies of children’s spontaneous Wh-questions indicate that What and Where questions consistently precede How, Why, and When questions, and that among the latter types, When questions tend to appear last. The usual course of acquisition spans 1 to 2 years (Smith, 1933; Ingram, 1972; Tyack & Ingram, 1977; Lee, 1974). (c) Experimental data indicate that locative prepositions generally appear over a span of 2 or more years in the following order:2 IN, ON, UNDER < NEXT TO < BETWEEN < IN BACK OF, IN FRONT OF (Holmes, 1932; Washington & Naremore, 1978; Johnston, 1984; Johnston & Slobin, 1979). Early vocabulary records corroborate this account, except that the acquisition point for UNDER seems somewhat variable (Ames & Learned, 1948; Bateman, 1914, 1915; Bohn, 1914; Boyd, 1914; Gale & Gale, 1902; Grant, 1915; Pelsma, 1910). Experimental studies of comprehension also confirm this picture (E. Clark, 1972b, 1977).

34

(d) Dimensional adjectives, while not comprising a linguistic form class per se, do constitute a semantic field and have common syntactic properties. Acquisition studies using comprehension and antonym-naming paradigms indicate that these terms are acquired in the order: big/little < tall/short, long/short < high/low < thick/thin, deep/shallow, wide/narrow, again over a period of 2 to 3 years (Bartlett, 1976; Coots, 1975; Carev. 1978a: Berndt & Caramazza, 1978; Clark, 1972a; Brewer & Stone, 1975; Tashiro, in H. Clark, 1973). Vocabulary studies that report on young children’s conversational use of dimensional adjectives tend to corroborate this order (Court, 1920; Gale & Gale, 1902; Bohn, 1914; Boyd, 1914; Grant, 1915; Belsma, 1910), as does a recent investigation of adjective ordering where overall size and length were correctly ordered earlier than girth (Richards, 1979).

2.1.5. Systematic Changes in Meaning When children first learn a word they frequently use it in situations where an adult would not, or fail to use it where an adult would. Some of these over- or under-extensions appear to be idiosyncratic and momentary; others are more systematic in that they characterize the speech of many children, persist, and evolve with age in predictable ways. Such patterns of word use imply changes in meaning that may well reflect conceptual growth. (a) Children’s early uses of IN BACK OF occur in contexts where the reference object obscures the topic object. Later uses of IN BACK OF and IN FRONT OF occur where the

35

reference object has inherent BACK or FRONT features, and finally in contexts where only the listener’s position determines the spatial relationship (Johnston, 1984; Johnston & Slobin, 1979). Comprehension of these terms seems to follow a similar course of development (Clark, 1972a, 1977; Harris & Strommen, 1972; Kuczaj & Maratsos, 1975; see Webb & Abrahamson, 1976, for related observations of deictic this and that). (b) Early uses of the past tense occur in the presence of visible, lasting end-states and are largely confined to verbs which express change-of-state meanings, e.g. break or find (Antinucci & Miller, 1976; Bloom, Lifter, Hafitz, 1980). Later extensions of past tense forms to experience and Activity verbs suggest decreasing attention to specific event characteristics and a new concern with locating diverse events along a temporal continuum. (c) (c) Studies of children’s understanding of big and little indicate that meaning may change even after a period of adultlike use. Two- and 3-year-old children correctly respond to big or little with global assessments of size, but four and five year old children respond as if big/little had the dimensional meanings of tall/short (Maratsos, 1973; Cook, 1976; Wales & Campbell, 1970). The literature on children learning English thus contains five initial sorts of evidence that point to the role of prior conceptual and factual knowledge in language acquisition: use of nonlinguistic comprehension strategies, immature expression of new intentions, 36

simultaneous emergence of meanings across form, protracted acquisition of form class members despite formal similarity, and systematic changes of meaning in already acquired forms. It is certainly true that other factors participate in the acquisition phenomena surveyed above. Communicative salience, i.e. the likelihood of needing to talk about a particular notion, seems influential in the early uses of IN BACK OF versus IN FRONT OF (Johnston, 1984; Tanz, 1980). Input frequency, now recognized as a more potent factor than was once believed (e.g. Newport, Gleitman & Gleitman, 1977; Block & Kessel, 1980; Nelson, Carskoddon, & Bonvillian, 1973), seems to affect the learning of dimensional adjectives and auxiliaries. Likewise, linguistic factors such as the number of synonomous forms, the type of semantic components, or the scope and nature of entailed syntactic markers, have been invoked to explain acquisition patterns for locatives, spatial adjectives and Wh-question words respectively (Johnston & Slobin, 1979; E. Clark, 1973b; Ervin-Tripp, 1970). Our search here, however, is not for a single sufficient explanation. The crosslinguistic methodology and indeed virtually all current views of language learning derive from multicausal models. The question is whether ceterus paribus nonlinguistic conceptual growth can be seen to influence the course of language acquisition. Unfortunately, with language as with life, all other things are seldom equal, which leaves the evidence presented thus far somewhat equivocal.

37

2.2. Nonlinguistic Evidence of Conceptual Constraints To better inform our interpretation of the linguistic observations and gain a clearer picture of the role of conceptual or factual prerequisites, I turn next to studies of nonverbal cognitive growth and to studies which assess both verbal and nonverbal development in the same children, using either cross-sectional or training study paradigms. Here we will be looking for three additional sorts of evidence: parallels in patterns of nonlinguistic and linguistic growth; temporal priority of conceptual versus linguistic achievements; and asymmetries in the effects of verbal versus conceptual training. Although evidence of this type is increasingly available for a number of domains, I have chosen to focus on two areas where the literature is particularly rich, i.e. spatial and quantificational concepts as they relate to the child’s acquisition of locative prepositions, morel/less, and dimensional adjectives. This selectivity should allow me to explore these domains in sufficient depth to evaluate specific evidence while still illustrating the range of theoretical arguments which such data invite.

2.2.1. Spatial Concepts and the Language of Location As noted earlier there is a set of locative prepositions which English speaking children learn in a predictable sequence between the years of 2 and 5. To understand the ways in which nonverbal conceptual growth might induce this learning, we need a characterization of

38

spatial conceptualization during these preschool years. The Piagetian account provides a reasonable starting point.

Parallels in Conceptual and Linguistic Growth Piaget and Inhelder argue that “spatial notions do not derive directly from perception” but must be constructed operationally on the plane of reflective, nonpresentational thought (Inhelder, 1969, p. 35). This development of representational space begins with the objectivization of the physical world and the child’s knowledge of inherent object properties (Piaget & Inhelder, 1969). The earliest spatial notions are thus closely bound to object functions such as containment or support, and to the child’s concern with object permanence. Recall here the toddler’s pleasure with pots and pans, towers and hiding games. In the next phase, children construct the spatial notions of proximity, separation, surrounding and order. Such concepts do not entail perspectivism or measurement and in that sense resemble the relations found in a topological geometry. Piaget and Inhelder observe here, for example, that young children, when asked to place lampposts along an imaginary street, will create a meandering row with all elements touching, creating proximities but failing to construct the projective axis. Finally, children coordinate topological relations with their schemes for quantifying sets and imagining alternative points of view, arriving at spatial notions of a projective and Euclidean character (Piaget & Inhelder, 1967). At this level children can, for example, imagine what an object array would look like from some viewing point other than their own. In sum,

39

the Piagetian account of spatial conceptualization during the preschool years proposes a developmental progression from functional to topological to projective-Euclidean representations of space. Over the past decade much of the Genevan inspired research on spatial concepts has consisted of attempts to replicate specific observations rather than assess or elaborate the broader framework. Nevertheless, some evidence for the priority of topological notions can be found in the work of Laurendeau and Pinard (1970), and support for the early dominance of functional schemes can be found in recent studies of object placement. Sugarman (1983), reporting on her observations of early block play, notes that many 12- to 36-month-old subjects made use of IN and ON relationships in building one-to-one correspondence arrays, but only the 30- to 36-month-olds incorporated NEXT TO relationships. Likewise, the 1 ½- to 3-year-olds in Clark’s (1972b) study tended to substitute IN, ON, or UNDER placements for NEXT TO placements when asked to replicate object configurations. It is in the area of perspectivism, however, that we find particularly significant recent observations of early spatial knowledge. Flavell and his associates have demonstrated that children can conceptualize the path of sight and decide what someone else is looking at by age three (Lempers, Flavell & Flavell, 1977; Masongkay et al., 1974). Judgments of object orientation and inter-object relations from imagined perspectives emerge during the fourth year (Flavell, Shipstead, & Craft, 1978) and are quite evident by age four (Shantz & Watson, 1970; Shatz & Gelman, 40

1973; Masongkay et al., 1974; Fishbein, Lewis, & Keiffer, 1972). Such studies make it clear that while projective spatial concepts may evolve from earlier notions of proximity and order, they emerge at much younger ages than the original Piagetian observations indicated. The hypothesized progression from functional to topological to projective/Euclidean space remains a viable heuristic scheme, but one which needs further investigation. From the perspective of child language, what is notable about the Piagetian account is the striking way in which it parallels locative acquisition facts. In, on and under could well express functional concepts; next to, between, and in back of/in front of(used with reference objects which have identifiable back and front features) seem to express notions of proximity; and, in back of /in front of (used with non-featured reference objects) seem to express order relationships in a projective, perspectivistic space. Of course, as the child develops, early locatives assume increasingly sophisticated spatial meanings. For the adult, prepositions such as on are used to express projective as well as functional meanings, e.g. The arrow flew on target vs. The book lay on the shelf Parallels between developing spatial concepts and locative acquisition are thus clearest when we consider plausible child meanings rather than the full range of meanings assigned to these words by the adult grammar. Failure to recognize this distinction has yielded mixed results in those studies which have contrasted children’s understanding of “topological” vs. “projective” or “Euclidean” locatives (Walkerdine & Sinha, 1974; Windmiller, 1973). 41

Temporal Priority of Conceptual Achievements Parallelism, however provocative, does not establish the prerequisite status of conceptual knowledge. The strongest evidence would come from investigations of conceptual and linguistic development using longitudinal or training study paradigms. Data of this sort do not yet seem to exist within the spatial domain. My own research (Johnston, 1979) does, however, provide pertinent cross sectional data. Rather than pursue the broad distinction between functional, topological, and projective-Euclidean meanings, I chose to investigate children’s use of in front of/in back of as it related to their understanding of the various spatial notions which comprise the meanings of these terms. I hypothesized that early uses of these prepositions (with featured reference objects) might express meanings like NEXT-TO-THE-BACK/FRONT-OF while later uses (with nonfeatured reference objects) might express meanings like FIRST/SECOND-IN-THE-LINE-OF-SIGHT. If so, children at the earlier stage should control notions of PROXIMITY and OBJECT-FEATURE, while children in the later stage should additionally understand ORDER and PROJECTIVE relations. Moreover, if these nonverbal spatial notions were truly prerequisite, children who did not have such knowledge should not use in back of/in front of in their featured or deictic senses respectively. On the other hand, children who did control these concepts might or might not demonstrate the associated prepositional use since factors other than conceptual development are known to influence language learning. To test these hypotheses I asked 33 children ages 31 to 54 months to perform in a

42

variety of spatial tasks. They told puppets where to look for lost objects, placed animals where a doll could not see them (PROJECTIVE RELATIONS), replicated linear arrays of toys (ORDER RELATIONS) and induced the rule that Ernie was always hiding at the back of the truck (PROXIMITY-OBJECT FEATURE). Comparisons of performance across these tasks confirmed my initial hypotheses. First, Guttman Scale analyses indicated that spatial notions and prepositional uses emerged in the predicted parallel orders: PROXIMITY, OBJECT FEATURE < ORDER, PROJECTIVE RELATIONS and in front of/in back of (featured) < in front of/in back of (nonfeatured). Second, children who used these prepositions in their featured and/or deictic senses could indeed solve nonverbal spatial problems requiring application of the pertinent spatial notions, i.e. PROXIMITY/OBJECT FEATURE for featured meanings and ORDER/PROJECTIVE RELATIONS for deictic meanings. The contingency tables for each hypothesized relationship reveal a good fit between expected and observed values (See Table 11.1). Note that for each comparison the number of children in the pass-concept/fail-language cell is substantially larger than that in the fail-concept/ pass-language tell; moreover, given the vagaries of testing young children, there are remarkably few disconfirming, i.e. fail-concept/pass-language, cases. A similar recent study with 2- and 3-year-olds (N = 36) looked at children’s understanding of back and front as it related to knowledge of object features (Levine & Carey, 1982). In the nonverbal tasks these children were asked to place objects with inherent fronts and backs in a 43

“parade” line, and also to place these objects so that they could “talk” to a doll. Virtually all of the children oriented the objects appropriately in the nonverbal tasks though many of the children subsequently failed to point to the front or back of these same objects on command. None of the children who failed the orientation tasks passed the language comprehension tasks. TABLE 11.1 Number of Subjects Passing or Failing Nonverbal Tests of Spatial Concepts and Related Locative Production Tasks Requiring IN BACK/FRONT OF in Johnston (1979)

Research of this sort is open to considerable methodological criticism, particularly concerning task appropriateness—a topic I shall return to later. Assuming task validity however, data from these two studies attest to the temporal priority of nonverbal spatial concepts and thus lend credence to their presumed role in locative learning. To summarize briefly, patterns of acquisition for English locatives suggest the role of conceptual prerequisites in a number of ways. Spatial prepositions are learned in a constant order, at generally predictable ages, over a protracted period of time, and show predictable patterns

44

of meaning change. The meanings they express emerge in a sequence that resembles the one hypothesized for nonverbal spatial concepts. And finally, for at least two groups of children, specific uses of spatial terms appear to follow the acquisition of related nonverbal concepts.

2.2.2. Quantity Concepts and the Language of Comparison Since comparative judgments are basic to virtually all human reasoning it is not surprising to find a rich and varied set of linguistic tools for talking about comparison. In English this set of forms includes too, enough, like, exceed, same, -er, and others with equally diverse syntactic properties and semantic complexity. Acquisition studies in this domain have been numerous and often perplexing, but they are eminently pertinent to the question of conceptual prerequisites. In this section I will focus on that portion of the literature which looks at children expressing judgments of quantifiable extent, with special attention to judgments of inequality. 2.2.2.1. More and Less. Studies of more suggest that this adjective has a long and fascinating developmental history. In the first stages of language learning, children use more to request, or comment upon, the recurrence of an object or event (Bloom, 1970, 1975; Brown, 1973). Slightly later uses occur in the presence of two or more similar objects (Bloom, 1970) and seem to be primitive expressions of numerosity. Next, more appears in requests for the addition of an object (or amount) to an already present set (or substance) (Bloom, 1970). And finally, around age 3, children use more to talk about

45

differences in the quantity or extent of two object sets or in the amounts of two substances (e.g. Bullock & Gelman, 1977). These observations suggest that the meaning of more may be progressively glossed as ANOTHER/AGAIN < MORE-THAN-ONE < MORE-THAN-THERE-WERE < MORE-HERE-THAN-THERE. Although this sequence is thus far based on the conversational records of only a few children, comprehension studies (Beilin, 1968; Brush, 1976; Weiner, 1974) lend credence to the account. Some of Weiner’s 2-year-olds, for example, when asked which row had more objects, would pick up an object from either row and say more x. There was also a tendency for her 2-year-olds to pick the row to which objects had been added regardless of its magnitude. By age three, on the other hand, children clearly comprehend more as an expression of relative quantity/extent across a variety of tasks (e.g. Weiner, 1974; Beilin, 1975a; Wales, Garman, & Griffiths, 1976). As was suggested in the earlier discussion of comprehension strategies, the acquisition of less remains puzzling despite intense research effort. Ignoring certain knots in this line of investigation, the broad facts appear to be these. Less, when used to express judgments of relative quantity/extent between sets, is understood later than the comparable sense of more, with many 4 year olds still having difficulty (Wales et al., 1976; Holland & Palermo, 1975; Weiner, 1974; Palermo, 1973; Donaldson & Balfour, 1968). Studies which report successful comprehension of comparative less by 3 year olds appear to be those that ask consecutive more and 46

less questions about the same sets and/or make use of a separate standard to which given sets may be compared (Wannamacher & Ryan, 1978; Griffiths, Shantz, & Siegel, 1967). Although less may have an early “removal” meaning similar to the “additive” meaning of more, i.e. LESS-THAN-THERE-WERE (Carey, 1978b; Beilin, 1968), this term seems to have no further history of meaning change. Given our interest in conceptual prerequisites, the acquisition of more and less is noteworthy on three counts. First, there is the protracted and regular history of changes in the meaning of more which could easily reflect the influence of conceptual growth. Second, although we think of more as expressing judgments of relative quantity/extent, the child’s more is at first nonquantitative and non-comparative (Donaldson & Wales, 1970; Weiner, 1974). Such semantic changes invite us to look at the ontogenesis of number concepts and their coordination into ordinal, comparative schemes. Third, although the factors which create differences in the acquisition of more and less may ultimately prove to be language (or culture) specific (Wales et al., 1976), it is also feasible that some conceptual asymmetry is responsible.

Parallels in Conceptual and Linguistic Growth The Piagetian account of number conservation has for years been a pivot point for discussion of early quantity concepts. The presumed developmental scenario is by now familiar. The 4- to 5-year-old child, lacking a meaningful counting system and unable to construct

47

one-to-one correspondences, judges quantity by noting spatial properties such as density or extent. Only as thought becomes “reversible” and the child can coordinate compensating spatial dimensions (shorter but denser), build correspondences, or imagine transformations which return arrays to their original form will he argue for invariance of quantity across physical transformations. The literature which documents this developmental sequence, while fascinating in its own right, does not greatly illuminate child language. The account provides little insight into conceptual achievements during the early preschool years, and the data at least in part reflect children’s initial nonadult meanings for words such as more (Griffiths et al., 1967). We can well imagine, for example, that Mehler and Bever’s (1967) 4 year olds did less well than 2½ year olds in judging which array had more because they had moved beyond the stage where more means only MORE-NOW-THAN-BEFORE. Finally, it is not at all clear that conservation reasoning per se constitutes a knowledge base which should be prerequisite to the learning of any particular linguistic form (Beilin, 1975a). For these reasons, I focus here on a smaller and more controversial literature which explores numerical reasoning in 2- to 4-year-olds. As in space, number conceptualizations begin with knowledge of object properties, particularly those which help to define a spatio-temporal set. Although 12–18 month olds successively choose similar objects for play, it is at 24–30 months that we begin to see objects grouped spatially according to their physical properties 48

and thus comprising sets which may be quantified (Sugarman, 1983; Ricuitti, 1965). At first children seem unable to coordinate successive judgments of similarity between elements in the set with any predication on the set as a whole. Saxe (1979a, 1979b) notes, for example, that young children may count an array without recognizing the cardinal significance of the final number, behavior which amounts to iterative labeling, or may base their quantity judgments upon spatial extent, a global quality-of-the-whole. Since judgments of numerosity demand both types of notions, i.e., three is at once the third in a series of interative correspondences and a comment on the series, this early developmental phase remains essentially prequantitative. The next phase of development begins with the child’s first judgments of numerosity derived from the use of conventional or idiosyncratic counting schemes. The age for which this achievement is claimed varies according to the task and the investigator’s beliefs about perceptually-based subitizing. Even generous observers, however, find only infrequent quantity judgments among 2 year olds (Gelman & Gallistel, 1978). For 3- to 4-year-olds, the evidence is clearer. Children of this age, in a variety of nonverbal, concept discovery tasks, have demonstrated their ability to judge the quantity of objects in a set and to compare quantities between sets while ignoring irrelevant spatial displacements (Estes, 1976; Siegel, 1974, 1976, 1978; Bullock & Gelman, 1977; Gelman & Gallistel, 1978). Interestingly enough, these studies do not reveal the asymmetry between judgments of MORE and LESS that the verbal findings suggest.

49

The final steps from simple judgments of relative quantity to conservation reasoning are not yet well understood. Gelman and Gallistel (1978) argue that the child moves from NUMERICAL thinking about specific numerosities to ALGEBRAIC reasoning about unspecified quantities, and that only the latter entails any appreciation of one-to-one correspondences. Conservation tasks may be approached and solved with either sort of thought. Saxe and Cohen (1980) on the other hand, argue that the use of numerical notation is itself an instance of one-to-one correspondence and hence that these relations are an integral component of conservation reasoning. All investigators seem to agree that none of the early quantity notions discussed thus far will quite suffice to explain the conservation facts. This picture of early quantificational thought is as yet sketchy, but gives promise of the evidence we seek. First, in contrast to the traditional conservation literature, it indicates a surprising degree of early sophistication in quantity conceptualization among 3- and 4-year-olds which could well provide a foundation for language learning. It also points to an essentially prequantificational period and thus to a sequence of development in nonverbal cognition which could have language acquisition consequences. Although it is difficult as yet to draw for number the sort of broad parallels between nonverbal conceptual development and language learning that we found in the domain of space, Saxe’s proposal that quantity judgments evolve from notions of iteration and class intention nicely mirrors the early meanings that are proposed for more, i.e. AGAIN or ANOTHER. There have been a few studies which explicitly 50

test such developmental relationships between nonverbal knowledge of quantity and linguistic knowledge of comparison terms, and I turn now for a brief look at that literature.

Temporal Priority of Conceptual Achievements and Training Asymmetries The first finding comes from Bullock and Gelman’s (1977) study of numerical reasoning. In a concept attainment task, their 2- to 4-year-old subjects were asked to pick the “winner” (greater or lesser magnitude) from two object arrays and to explain their choices. Although some 80–85% of the 3- and 4-year-olds were able to choose the correct array despite changes in numerosity, and hence were making judgments of relative quantity, only 6/50 children used relational language, i.e. morel less, to justify their choices. Half of the remaining children appealed to absolute numerosity; the others gave irrelevant answers. These findings suggest not only that thinking about quantity comparisons occurs prior to talking about them, but that children acquire the language of absolute quantity prior to comparative terms. To my knowledge there has been no nonverbal test of this latter sequence though it seems quite reasonable. In a recent training study Thorns (1977) looked at relationships between children’s performance on conservation tasks and their knowledge of the meanings of more and less. Thirty children between the ages of 31/2 and 6 who had been identified as performing at chance levels on conservation (continuous quantity,

51

number and mass) and more/less comprehension tasks were trained in one or the other of these areas. Conservation training was done nonverbally. Children watched while one of two quantitatively equivalent sets (substances) was transformed spatially and/or quantitatively, then pushed a button to indicate their judgments of set (substance) equivalence. They received prizes if correct, but no verbal explanations. Children in the more/less training group were shown unequal amounts of objects (substances) and were asked which array/container/ mass had more/less. Again feedback was provided. Posttests indicated that children trained in nonverbal conservation judgments improved both on the traditional conservation tasks and on the more/less comprehension task. Children trained to make more/less judgments, on the other hand, improved only on the linguistic task. Since the two tasks entailed different sorts of conceptual judgments, i.e. dynamic equivalence across spatial transormations vs. direction of nonequivalence in static displays, it is not surprising that more/less training did not induce conservation. Indeed, Holland and Palermo (1975) report a similar lack of effect for less training on conservation. What does seem noteworthy in the Thorns study is that nonverbal training in quantity equivalence led to improved performance on the more/less task. Whatever conceptual growth occurred apparently induced some new ability to comprehend these terms. Even stronger evidence for conceptual priority can be found in Siegel’s (1976) study of quantificational language. She administered a nonverbal concept

52

attainment task and a language comprehension task to 102 3- and 4-year-olds. In the conceptual task children were shown a pair of cards, each of which displayed from 1–9 dots, and were asked to decide which card was “correct.” For some children the greater quantity was correct, for others the lesser. In the language task children viewed the same cards and were asked to choose the big (or little) array. Although the use of big/little to describe varying numbers of dots seems slightly foreign to adult ears, the arrays were of equal density and thus could be assessed in spatial (length) as well as more purely quantificational terms. Patterns of performance across the two tasks indicated that success on the nonverbal quantification task preceded comprehension of the quanity-related terms. Table 11.2 summarizes the findings. TABLE 11.2 Number of Children Passing or Failing a Nonverbal Test of Quantity Concepts and a Related Language Comprehension Task in (Siegel, 1976) Comprehension Pass Pass

Fail

3 yr.

15

20

4 yr.

37

17

3 yr.

2

8

4 yr.

1

2

Concept Fail

53

In sum, several lines of evidence suggest that children’s learning of more/less is influenced by conceptual growth. For more we find a long and systematic history of meaning change which, at least in its initial stages, runs parallel to developments in quantity conceptualization. Quantity concepts arise out of notions of iteration and class intention just as more has early prequantificational meanings of ANOTHER or AGAIN, and nonverbal concepts of quantity comparison emerge during the age range within which morel less take on quantificational and comparative meanings. Studies which have compared conceptual and linguistic knowledge in this domain at several levels of development so far indicate the temporal priority of nonverbal conceptualization. 2.2.2.2. Dimensional Adjectives. During the later preschool years children learn that set of attributive adjectives which refers to the spatial, dimensional properties of objects, e.g. big, tall, wide, etc. As reviewed earlier, these terms seem to be acquired in a consistent order over a protracted period of time despite the fact that they are drawn from a common semantic field and have similar syntactic properties. Such facts suggest the possible influence of conceptual growth on language learning. Before we consider the pertinent cognitive development literature or review studies which explicitly test language-thought relationships, there are further facts about the acquisition of spatial adjectives which invite comment. A number of child language studies have focused on this domain and together they indicate a developmental picture of considerable complexity.

54

Adult speakers of English use spatial adjectives to refer to particular, differentiated dimensions, e.g. vertical, horizontal, secondary, etc. and to express judgments of extent along these dimensions. The terms most often express comparative reasoning. With or without the inflection -er, they are used to talk about the relative position of two or more objects along a continuum of dimensional extent, e.g., It was the taller midget who … , or to compare one object’s degree of extent to some implicit standard, e.g., John is quite tall ‘has greater height than the average male’. Note that in both of these cases tall refers to the direction of deviation away from some reference point rather than to any absolute height value. Finally, a pair (or pairs) of spatial adjectives is mapped asymmetrically onto each dimension. The “positive” term is used to refer to the dimension itself, e.g., How tall is John?, as well as to judgments of GREATER-EXTENT, while the “negative” term is usually reserved for comparative, LESSER-EXTENT meanings. For the adult then, spatial adjectives express notions of DIMENSION, RELATIVE QUANTITY, and DIRECTIONALITY. What does the language acquisition literature reveal about the child’s mastery of these parameters of meaning? Early reports of children’s comprehension of relational terms (e.g. Donaldson & Wales, 1970) led H. Clark (1970, 1973) to hypothesize that contrasting adjective pairs like tall/short are initially understood as synonyms referring to a given spatial dimension in a noncomparative fashion. The child knows that the adjective pair refers to an attribute such as size, height, or width, but hasn’t yet discovered the directional and 55

relative extent aspects of their meanings. Assuming that objects with greater extent are better examples of a given dimension, children will choose the taller (wider, bigger, etc.) object in tests of adjective comprehension and thus seem to master the positive before the negative terms. Further, this congruence between focus on greater extent and the meaning of the positive term may actually facilitate its acquisition (E. Clark, 1973b, 1977). Initial support for this hypothesis would come from data which indicated that positive terms were consistently easier for children to understand; stronger support would come from within-subject error analyses that revealed consistent positive interpretations of negative terms, or indicated that dimensional components of meaning were learned prior to polarity. Evidence of the first sort has been mixed. Investigators have reported positive terms easier (Donaldson & Wales, 1970; Tashiro in H. Clark, 1973; Bartlett, 1976; Brewer & Stone, 1975; Townsend, 1976), negative terms easier (Eilers, Oiler, & Ellington, 1974), or no polarity effect at all (Bartlett, 1976; Carey, 1978a; Coots, 1975). Evidence of the second two sorts has been sparse. Investigators have tended to use stimuli which differ in only one dimension or have reported only grouped results. However, recent studies which do provide pertinent data tend not to support Clark’s hypothesis. Coots (1975) and Townsend (1976) find systematic misinterpretation of negative terms in only a small minority (approximately 10%) of possible cases; Carey (1978a) and Brewer and Stone (1975) report a predominance of dimensionality errors over polarity errors when both are possible. Moreover, it appears that 56

polarity effects are quite task specific (Coots, 1975; Bartlett, 1976; Carey, 1978a), suggesting the influence of non-linguistic factors such as object preference (Eilers et al., 1974) or information processing constraints (Bartlett, 1976). Children’s acquisition of directionality remains poorly understood, but polar features do not seem to be the final step in the mastery of spatial adjectives. What about other aspects of meaning such as dimensionality or relative extent? Few acquisition studies have focused on these features, but the scattered findings are provocative. First, children seem to have difficulty determining the relevant spatial dimension. Their task is considerable. Not only must they distinguish between different dimensional notions and map them to the appropriate terms, but they must first discover that it is single dimensions which are to be mapped. Child language data reveals both of these steps. The earliest spatial terms, big and little, express global, nondimensional judgments of size; terms for specific dimensions are learned later, over a period of years. Moreover, the exact dimensional meaning of a given term is apparently mastered only after it is understood as spatial and polar (Carey, 1978a; Brewer & Stone, 1975; Coots, 1975). Second, even when children know the global-size or dimensional meaning of a spatial adjective, they may fashion this meaning differently than an adult. Words like big seem to refer first to absolute values rather than to directions of difference. These values are undoubtedly ranges and they certainly are specific to object class, but 57

they seem absolute nevertheless. Two-year-old children can indicate whether a shoe is big—for a shoe (Carey, 1978a), but this fact alone does not reveal whether the child sees the shoe as similar in size to a prototypical big shoe or as deviating in a positive direction from the average shoe. Observations of slightly older children using nonspatial dimensional terms suggest the former interpretation. H. Clark (1970) cites examples of children who understand the phrase four years younger to mean four years young, and of other children who, given two trees with equal numbers of apples, argue that both trees have more apples. Recall too that Bullock and Gelman’s (1977) subjects explained their choice of GREATER-MAGNITUDE arrays by appealing to absolute values, e.g., It’s three, and Siegel (1976) reports similar justifications for length comparisons, i.e., It’s big. In a recent study of dimensional adjective use, Gitterman (1979) asked 4½- to 7½-year-old children to describe the differences between two objects in a dimensionally graded array vis-à-vis a third standard object. Children at lower developmental levels (defined by age, seriation ability and restricted use of comparative inflections) used explanations of a categorical sort: This one’s hard, this one’s not hard; or, Smooth, soft, hard. Ehri (1976) reports similar response patterns. Children in these experiments seem to have constructed contrastive dimensional categories which although they may derive from primitive two-term comparisons, are treated as absolutes, i.e. A-LITTLE-X vs. A-LOT-OF-X. Such categories may be marked for polarity, but they do not yet incorporate the notion of relative extent.

58

As long as young children fashion dimensional meanings categorically, there is little need for comparative morphology. Indeed, some uses of -er must seem quite anomolous. How could one small shoe be bigger than another small shoe, or a single object be both bigger and smaller? The superlative -est apparently finds its initial role as an intensifier. Extreme dimensional values highlight the categorical contrasts. Three year olds will pick the extremes when asked to indicate an object which is very big (Berndt & Caramazza, 1978) or the biggest (Layton & Stick, 1979). Success with the comparative -er, however, comes somewhat later (Townsend, 1976; Layton & Stick, 1979) and seems to mark the child’s growing appreciation that spatial adjectives can refer merely to the direction of difference along a dimensional continuum. Such uses of the -er inflection occur first in contexts where a single object is changing in extent. Children will talk about towers getting bigger or tape measures getting longer before they will use these same inflected terms to explain the differences between two objects vis-à-vis a standard (Gitterman & Johnston, 1983). Early meanings for the -er inflection thus parallel the “additive” meanings used for more. Such meanings are newly directional but not yet fully comparative, since they do not entail judgments about the relative extent of two end-states. In the adult grammar, terms such as bigger, taller, and more can convey “three-state comparisons” (Donaldson & Wales, 1970): X has extent,, Y has extent2, extent, exceeds extent2. For the child these terms assume comparative significance only after a period in which they convey more primitive categorical and/or additive meanings.

59

Our understanding of the acquisition of spatial adjectives is far from complete and the view presented here needs further experimental support. The major developmental themes do seem clear, i.e. the emergence of specific dimensions and the evolution of purely relative comparative meanings. The relationships between these themes do not. Significant age/dimension/polarity interactions in recent experiments (e.g. Coots, 1975; Townsend, 1976) suggest that different adjective pairs may be at different evolutionary points at any given time. Faced with objects varying along a relatively unfamiliar dimension, children fail to mark polarity or inflectional distinctions they otherwise control (Gitterman, 1979; Bartlett, 1976).

Relative Extent: A Semantic Mapping Problem? Parts of this developmental picture seem likely to reflect conceptual development while others reflect the process of semantic mapping. Children’s early preference for categorical meanings is a good case in point. Such meanings may result from an initial but erroneous assumption that the meanings of dimensional adjectives are analogous to those of already familiar nouns, color adjectives, spatial prepositions, and possessive adjectives. Alternatively, children may be incapable of constructing and reasoning with purely ordinal relations. Studies of early cognitive development make this second, “conceptual” explanation improbable. As we have seen, the literature on numerical reasoning reveals facility with nonverbal judgments of relative quantity among 3 year olds. Similarly, observation of children’s play with nesting cups (Greenfield, Nelson, &

60

Saltz-man, 1972) reveals control of ordinal size relations by age three—at least in contexts where the child works them out in a functional space. Yet 4- and 5-year-olds are still treating new dimensional adjectives categorically. Experimental tests of the relationship between nonverbal conceptual knowledge and the noncategorical use of spatial adjectives have unfortunately focused on a later developmental point—the one marked by success on Piagetian seriation tasks in which children are asked to arrange objects in order by size or length. Such seriation tasks systematically differ from the number or nesting-cup tasks on which 3-year-olds succeed. They involve no functional consequences and they entail comparisons among more than two objects. It is certainly true that advanced ordinal reasoning is expressed in comparative language, and that “senators” have reason to produce more of these terms in more complex syntactic frames. However, early use of words like tall, more, or bigger requires most only the directional comparison of two points on a dimensional scale. The multiterm comparisons inherent in seriation reasoning have no additional lexical or morphological correlates. It is thus not surprising that experiments have largely failed to demonstrate that success on a seriation task occurs prior to comparative uses of spatial adjectives (Heber, 1977; Ehri, 1976; Ehri & Ammon, 1972; but also see Gitterman & Johnston, 1983). The weight of current evidence suggests that it is the child’s semantic hypotheses rather than conceptual development which leads to early categorical meanings 61

for spatial adjectives. Further research may indicate some principled sequence of emergence for ordinal relations in different conceptual domains e.g., comparisons of relative quantity may be easier than comparisons of relative length. Such a sequence might illuminate the child’s progressive shift to non-categorical meanings. For now, we can only observe the young child’s evident competence with ordinal relations and entertain the more “linguistic” explanation.

Dimensionality: A Conceptual Problem? As a second empirical puzzle, consider the protracted learning of the set of spatial adjectives. Terms referring to different dimensions (or sets of dimensions) are learned in a predictable order over a period of several years. These differences in age of acquisition could reflect differences in input frequency (Bartlett, 1976), differences in semantic complexity (H. Clark, 1973), or differences in conceptual “naturalness” which affect the child’s nonverbal discovery of the underlying dimensional notions (Osherson, 1978). At the moment no one of these explanations will fully account for the acquisition sequence. Recent studies of nonverbal classification, however, do suggest that conceptual development plays its role, albeit in an unexpected manner. To set the stage for these new findings we need to look briefly at the spontaneous sorting behavior of very young children. When given a collection of objects differing on one or more parameters, 18- to 24-month-old infants will select similar objects and place them together, attending

62

to size, form, and/or color (Ricciuti, 1965; Nelson, 1973a; Woodward & Hunt, 1972; Sugarman, 1983). At first children construct only single classes, but by 30 months they will separate objects into two contrasting groups, e.g. big ones vs. small ones, in an operating sequence that indicates their simultaneous consideration of both sets (Sugarman, 1983). Such observations seem to reveal not only an early competence with attribute values, but the heirarchical organization of these values into complementary classes which are implicitly linked by superordinate notions of color, size, etc. This interpretation may be too generous. The objects used in these studies of early classification have usually differed in value along a single dimension or in covarying values along two or more dimensions, e.g. red circles vs. blue squares. When confronted with an object assortment in which dimensional values are crossed, e.g. big/little circles vs. big/little squares, 2-year-olds are less likely to form spatial groupings (Denny, 1972) and have difficulty discovering the categorization rule which determines “winners” (Watson, Hayes, & Vietze, 1979). When they do construct groupings, the objects in each set are often defined by two dimensional values, e.g. big AND circle. Although there is some disagreement over the interpretation of this last fact (Sugarman, 1983), young children may be making use of unanalyzed global estimates of physical similarity rather than coordinating dimensional notions per se. (See Cook, 1976, for further evidence on this point.) When applied to objects which differ in value along a single dimension, these global estimates yield dimensional groupings by default.

63

By 4- to 5-years-of-age, children are able to construct simple classification matrices incorporating two dimensions (Roberts & Fisher, 1979), but again these experimental “facts” may be misleading. Recent experiments by Smith (1979; Smith & Kemler, 1977) reveal the persistence of global similarity strategies well into the school years. In one task (Smith, 1979), kindergarten, second grade, and fifth grade children were shown two sets of objects and then were asked to place new objects in one (or none) of the original sets. Unlike the classification studies reviewed earlier, Smith’s tasks utilized objects which represented one of eight possible values along each of two dimensions, i.e. height and color. Instead of presenting the child with discrete and maximally contrasting dimensional values, e.g. green vs. yellow or big vs. small, these materials presented “continuously” varying values, i.e. small triangles with equal area, but differing heights (½″ to 2”) and colors (yellow to red). It was thus possible to determine whether children grouped objects according to overall similarity or according to identity of dimensional value. Given two sets whose members were identical on one dimension and similar on the second (three ¾″triangles of various yellow shades and three 1⅜″ triangles of various red shades) where would children place a ¾″ red triangle, or a ⅝″yellow one? Organizing by dimension, the child could place the ¾″ red triangle with items of identical size, and reject the ⅝″ triangle. Organizing by similarity, the child could reject the ¾″ red triangle and place the ⅝″ yellow triangle with other smallish yellowish items. In fact, Smith’s findings reveal a clear developmental pattern: Kindergarteners tended to

64

construct similarity groupings while fifth graders organized dimensionally. It would be foolish to argue that spontaneous dimensional organization of the sort revealed in Smith’s older group is prerequisite to the learning of spatial adjectives. The facts indicate otherwise. Children have mastered a wide dimensional vocabulary prior to the fifth, or even second, grade. What these findings do indicate is the young child’s proclivity for global, wholistic judgments of object attributes—a proclivity which could impede the construction of semantically pertinent isolated dimensions. In sum, a variety of findings attest to the nondimensional character of young children’s thought: they initially fail to group objects by single dimensions unless the materials necessitate it; they group objects at the level of maximum similarity according to all possible dimensions, ignoring perceptible differences, rather than focusing on single attributes or values; they may be induced to dimensional analyses by task manipulation but prefer the more global similarity strategies. This bias undoubtedly serves the child well for practical purposes; it scarcely facilitates the learning of dimensional language. The protracted acquisition of spatial adjectives may in part reflect the effect of wholistic biases. In contrast to the discussions of spatial locatives and more/less, this exploration of dimensional adjectives has revealed a limited sphere of influence for conceptual development. The mature use of dimensional adjectives clearly depends upon some prior facility for comparative 65

judgments of relative extent. Likewise, the late emergence of adjectives other than big/little may reflect some wholistic bias in young children’s thought. Other facts however, notably the protracted sequence of acquisition and the repeated shifts to noncategorical meanings, seem, at least for the moment, unrelated to parameters of conceptual growth.

2.3. Locating Evidence of Prerequisite Knowledge My review thus far has emphasized a particular set of child language phenomena, i.e. those which in my view were most likely to reveal the influence of important conceptual prerequisites. My selection was based on the developmental level of given linguistic forms and their probable relationships to ideational content. It seems quite clear that all aspects of language learning are not equally dependent upon prior nonverbal knowledge. As cases in point, English forms such as the auxiliary DO, the copula BE, or the infinitive particle to, express no particular ideational content and thus should be minimally affected by the availability of conceptual or factual knowledge. Even when linguistic forms do have clear semantic functions and thus would seem to require prior knowledge, this relationship between language and thought does not always control the visible course of acquisition. In certain periods and in certain linguistic domains, the developmental “action” seems to reflect aspects of rule discovery rather than new conceptual resources.

66

Children who are just beginning to talk already command a rich store of factual and relational knowledge. The first stages of language acquisition involve the mapping of this knowledge to the appropriate linguistic forms. The rapidity and fashion in which children learn basic phrase structure rules and the vocabulary of places, objects, and events, attest to the essentially linguistic nature of this period. It is after the initial phase when language learning has “caught up” with intellectual development and the “waiting room” is relatively empty (Johnston & Slobin, 1979), that we have our best opportunities to see the effects of conceptual growth. Even in later periods of development, however, the evidence demands careful analysis. For some forms, the prerequisite knowledge seems to be available quite early but the linguistic analysis proves slow. General category labels like animal, for example, seem to be learned later than BASIC LEVEL nouns like dog or cat (Rosch et al., 1976), not because their meanings are conceptually more demanding, but because they have fewer physical correlates, are used less often, and have reduced communicative salience. The fact that these general terms eventually assume superordinate status in classification reasoning has no obvious bearing on their initial acquisition. As a second example, children’s difficulty with color labels seems to derive not so much from the sophistication of the concept of color as from the semantic peculiarities of the color lexicon. Indeed, many children know the attribute dimension and learn a large repertoire of color labels prior to working out the sensory range for each term (Cruse, 1977; Bartlett, 1978; 67

however, see Rice, 1980, for evidence that preference for grouping objects by color predicts speed of learning and generalization of color terms). Finally, there are aspects of later language learning which reflect only the most idiosyncratic features of cognitive growth. Children who lack words like okra or dive suffer not so much from general conceptual deficiencies as from the lack of particular experiences and information. Children may indeed need to know something before they can learn to talk about it, but this claim is at times trivial, difficult to test, or inappropriate as an explanation of a given child language fact. The most appropriate and productive tests of this hypothesis will focus on forms which (a) have semantic functions, (b) are learned after the 2–3 word stage, (c) express relational concepts rather than factual information, and (d) express concepts which are achieved during the language learning years. The previous sections of this chapter have surveyed children’s learning of many such forms. How has the content prerequisite hypothesis fared?

2.4. Summary A number of facts about children’s acquisition of English suggest the guiding and constraining role of prerequisite knowledge. We have seen how children learn syntactically equivalent forms in a consistent order over a protracted period of time; how word meanings change in predictable patterns; and how forms that express similar meanings may emerge in synchrony. We have seen too how children stretch their grammatical tools to express new ideas and how they use nonverbal 68

resources to interpret the language they hear. Observations of these sorts occur at many ages and in diverse linguistic domains. Taken together they provide a solid platform from which to argue that nonverbal conceptual development is an important determinant of language growth. This general conclusion may be reassuring to those of us who wish to make use of crosslinguistic data, but it solves no methodological problems. In order to control for the availability of conceptual resources, we need to identify the ways in which particular types of nonverbal knowledge affect the learning of particular language forms. This requires a convergence of evidence for each case which will confirm the role of prerequisite knowledge and disclose interactions with other causal factors. My discussion of locative prepositions, more/ less, and dimensional adjectives was meant to illustrate this process. In each area, the argument began with a child language fact, e.g. a constant order of acquisition, a history of meaning change, or a protracted period of learning, that seemed to indicate the influence of conceptual achievements. Further evidence from child language studies, linguistic analysis, investigations of nonverbal cognitive growth, and experimental tests of linguistic dependency did not always confirm this initial estimate. Children’s learning of locative prepositions, morel/less, and big/little does seem to be affected by the evolution of spatial and quantificational knowledge. But the acquisition of dimensional adjectives other than big/ little is not yet so clearly related to parameters of conceptual growth. In that domain, factors such as

69

frequency of input or semantic complexity may play a more dominant role. In sum, observations of children learning English lend credence to our first cognitive hypothesis. Some aspects of language acquisition do seem to be guided and constrained by the availability of conceptual and factual knowledge. Moreover, it is increasingly possible to identify particular instances of this influence. I have argued the case for spatial and comparative terms; there is the potential for similar arguments in the areas of causal, temporal, and logical expressions.

3. HYPOTHESIS TWO: LANGUAGE AND THE STRUCTURE OF THOUGHT In learning to talk, the child’s fundamental task is to discover regularities in the language he hears. Communicative intentions may energize the search for linguistic patterns, and conceptual knowledge may place limits on discoverable rules, but these factors alone will not suffice in defining the child’s interaction with the language object. In both a physical and a metaphoric sense, language must be manipulated, analyzed, taken apart and reconstructed to be known. We assume that children come to this task with certain organizing capacities. From a process perspective, these capacities depend upon aspects of memory and attention, e.g. recency, novelty, and processing capacity, as well as properties of the sensory apparatus, e.g. its power for temporal resolution. Slobin (1973) seems to have had such basic processing mechanisms in mind when he

70

formulated operating principles like “pay attention to the ends of words,” or “avoid interruption.” These mechanisms appear to be biological givens which operate in a single fashion on linguistic and nonlinguistic information alike. Their status as cognitive prerequisites thus seems secure. Interestingly, recent research suggests that these basic processing mechanisms do not change with age (e.g. Olson, 1973; Huttenlocher & Burke, 1976; Flavell & Wellman, 1977). What we think of as developments in memory and attention occur not because of some fundamental change in capacity or function per se, but because of changes in higher level intellectual structures which potentiate and guide these functions. Such arguments suggest a second perspective on the child’s organizing capacities, one which for lack of a better term, I will call structural. Children come to the language learning task not only with biologically conditioned mechanisms of attention, memory, and perception, but with the potential for creating intellectual structures of increasing complexity, flexibility, generality, and coordination. The determinants of structural level are not well understood, but they certainly include the degree to which component schemes or mental subroutines are “automatic” (Norman, 1976), as well as the relative scope of available “information handling techniques” (Shatz, 1978). We may describe mental structures according to specific content as was seen in earlier sections, but it is their more general features which concern us here. Insofar as a child’s structural potential is similar across 71

domains of thought (a point on which there is considerable debate, e.g. Nelson & Nelson, 1978; Brainerd, 1978), this potential constitutes a likely prerequisite to language learning. The limited complexity and pervasive generality of children’s early linguistic rules led Slobin (1973) to formulate operating principles such as “avoid exceptions” or “the use of grammatical markers should make semantical sense.” Such constraints on language rules may well reflect the character of the young child’s general capacity for intellectual organization. Children may not be able to formulate linguistic rules which entail more complexity, flexibility, generality (specificity), or coordination than they are otherwise capable of. Weaker structural claims are also to be found in the child development literature. Greenfield and her associates (Goodson & Greenfield, 1975; Greenfield et al., 1972), for example, argue that certain organizational principles govern constructive activities in both linguistic and nonlinguistic domains, but further, that the two types of structure evolve independently and without any bond to common underlying schemes. Such structural isomorphism suggests no role for cognitive prerequisites, at least not in the usual sense of that term. General principles of mental organization may allow us to make predictions about the evolution of knowledge within a particular domain, but across domains they imply neither constraint nor preparedness. It is clearly the stronger structural claim which is of interest here. Namely, that due to the influence cf underlying cognitive capacities, the structure of language 72

should prove to be contemporaneously isomorphic to the structure of thought. The child may be able to “do” with words only what he can “do” with objects and actions. This notion of structural prerequisite has been most fully tested in studies of the passive voice, and I turn now to that literature.

3.1. Use of the Passive and Operative Thought 3.1.1. Acquisition Data From many perspectives, the rules governing English sentences in the passive voice should be difficult for children to learn. Full passives, i.e., those which include mention of the agent, are infrequently used in conversational speech, are motivated by considerations of discourse topic, and entail operations on purely linguistic entities such as Subject Noun Phrase. There is in fact much child language evidence to confirm this suspicion. Young three year olds largely fail to comprehend passive sentences in which the Agent and Patient cannot be determined on nonlinguistic grounds (Fraser, Bellugi, & Brown, 1963). Some studies report that older three year olds can enact the meanings of such SEMANTICALLY REVERSIBLE sentences (Baldie, 1976; Bever, 1970a; Lempert, 1978), but other studies find substantial improvement in comprehension of the passive throughout the preschool years (Lempert & Kinsbourne, 1978) with “success” coming only at age five or later (Slobin, 1966; Turner & Rommetveit, 1967a; Gaer, 1969; Sinclair, Sinclair, & De Marcellus, 1971; Beilin, 1975b).

73

The apparent difficulty of comprehending sentences in the passive voice invites explanation. Why do young children misinterpret the meaning of sentences such as John was hit by Mary? It is clearly not because they are unable to interpret linguistic information. At the earliest stage of language learning children may rely on nonlinguistic knowledge to estimate utterance meaning, but in most studies of the passive, children succeed in comprehending semantically reversible active voice sentences while failing to understand the passive items. Could it be that children misinterpret passives because they apply the wrong linguistic rule and treat passives as though they were oddly formed active voice sentences? Some evidence suggests that this is the case. Young children do often interpret the Subject Noun Phrase in passive sentences as if it were the Agent rather than the Patient/Location (Sinclair et al., 1971; Bever, 1970a; Lempert & Kinsbourne, 1978; Fraseret al., 1963). Moreover, the high frequency and intrasubject consistency of these “reversal” errors indicate that they are rule based rather than random in character. Use of the Subject-as-Agent rule may be more apparent than real, however, since studies which control for the animacy of Agent and Patient find no word order effect (Lempert, 1978; Scholnick & Adams, 1973). In any case, these results tell us more about what children do before they know the passive than about how they learn it. To answer this latter question we must look again at the nature of passive sentences. From a pragmatic point of view, passivization allows the speaker to topicalize the Patient (or sometimes, 74

Location) in a causal proposition. Since this function is usually realized in context, isolated passive sentences are somewhat anomalous and at least one study has demonstrated that children’s understanding of passive sentences may be affected by discourse context (Dewart, 1975, cited in Cromer, 1977). Ultimately, however, it is the speaker rather than the listener who must maintain the topical flow in context. This suggests that production data may provide the more useful key to children’s acquisition of the passive. Observers agree that full passives are rare in the spontaneous speech of young children (e.g. Harwood, 1959; Horgan, 1978), but there is considerable disagreement as to children’s proficiency at producing passives in elicitation tasks. The reported performance range for 5-year-olds extends from 5% to 50% success (Hayhurst, 1967; Sinclair et al., 1971; Beilin, 1975b; Turner & Rommetveit, 1967b). A closer look at these studies reveals at least two experimental sources of variance: the nature of the scoring criteria and the nature (or provision) of the linguistic context. Experimenters who count only full, grammatical passives and provide no linguistic prompts report low rates of success; experimenters who accept truncated, agrammatical or colloquial get passives, and provide a topicsetting question report higher rates of success. The appropriateness of a method naturally depends upon the theoretical issue at stake. These data indicate on the one hand that even very young children have some knowledge of the word-order and topicalization properties of the passive, while on the other hand they reveal children’s continuing difficulty with morphology. 75

The literature on children’s use of passive sentences suggests one further developmental trend. Horgan (1978) reports that her younger subjects, ages 2 through 4, used full passives when talking about animate Patients who were being affected by inanimate instrumental “Agents,” e.g. The boy was hit by the car. Confirmation of this trend can be found in Lempert’s (1978) comprehension study where 3- and 4-year-olds chose the inanimate rather than the animate object as the Agent for passive sentences, but did not show this preference for active sentences. Although Lempert’s data could be taken as evidence of non-linguistic comprehension strategy, the two studies taken together suggest otherwise. For young children, passive sentences may not be mere syntactic variants of active sentences; they may emerge independently to serve particular semantic functions and only later be incorporated into a “transformational” syntax. (See Olson & Nickerson, 1978, for related arguments based upon sentence verification data.) To summarize briefly, children evidence increasing control of passive voice sentences in both production and comprehension tasks between the ages of 3½ and 6. The improvement seems to reflect the acquisition and generalization of specific morphological rules rather than any change in nonspecific comprehension or topic focusing abilities. This broadening of linguistic competence could result from the autonomous reorganization of local syntactic rules, or from the availability of some new capacity for complex cognition. This latter, structural hypothesis is of particular interest

76

here because it invokes general cognitive prerequisites rather than prior linguistic knowledge.

3.1.2. Test of Structural Isomorphism The notion that knowledge of the passive reflects new structural capacity has recently been tested in studies which attempt to link advanced control of passive forms to the advent of concrete operational thought. Use of the passive voice, it is argued, requires the same sort of thought processes which enable success on concrete operational tasks: both involve the ability to DECENTER, i.e. to consider a single event from two or more simultaneous perspectives (Sinclair et al., 1971); and both entail REVERSIBILITY, i.e. the ability to operate on a transformation and recover the original state/meaning (Beilin, 1975b; Moore & Harris, 1978). There are serious flaws in both of these arguments. While it is certainly true that active/passive sentence pairs represent two different topical viewpoints, simple use of the passive requires the speaker or listener to assume only one of these perspectives at a time. Moreover, in concrete operational reasoning the equivalence of two configurations is deduced from physical laws and logical necessity while the correspondence between active and passive sentences rests ultimately in linguistic convention. Finally, both lines of argument make the most sense when we view passive sentences as derived transformationally from an active base, but young children’s restricted use of the passive suggests that it may have a quite independent history. Ignoring these problems for the moment,

77

however, let us assume with recent investigators that passive sentences are derived from an active base, and that operations on linguistic and physical objects are analagous. How does the structural hypothesis fare in empirical tests? Tremaine (1975) administered a variety of concrete operational and linguistic tasks to a group of 7- to 9-year-olds. She reports significant correlations in the order of .6 to .7 between knowledge of syntax, as measured by comprehension and synonomy judgments, and operational reasoning, as measured by conservation of mass and numeration tasks. Knowledge of grammatical inflections and success on other conservation and seriation tasks showed weaker patterns of association. Passive sentences were one of four structures which contributed to the significant syntax correlations. Other investigators have focused more particularly on knowledge of the passive rule. Scholnick and Adams (1973) compared 5- to 7-year-old children’s ability to comprehend passive sentences with their ability to construct and transform a classification matrix organized along two dimensions. They report a significant, if modest, correlation of .44 which remained significant with age partialed out. Moore and Harris (1978) looked at comprehension and production of reversible passives by 3½ to 8½ year olds in relationship to their performance on 12 conservation tasks. Rather than noting general correlations, these investigators were particularly interested in the linguistic abilities of subjects who failed all tests of concrete operational 78

thought. Eighteen children fell into this group. Of the eighteen, six comprehended at least 3/5 of the passive sentences and seven produced a passive sentence at least once. Further intrasubject data are provided by Beilin (1975b). Like Moore and Harris,he tested both production and comprehension of reversible passives along with performance on classification, seriation and conservation tasks among 864½- to7½-year-olds. While it was generally true that success in one domain was associated with success in the other, there were also a number of children who evidenced asyncronous patterns of development: Seven children passed the passive comprehension task but did not display operative reasoning while 17 children showed the opposite pattern; likewise, 14 children produced passive sentences without solving the concrete operational tasks while 5 children showed the reverse pattern. Two broad conclusions can be drawn from these data. It does seem to be the case that advanced competence with passive constructions emerges at the same time as concrete operational reasoning skills. Production and comprehension of passive sentences, however, clearly does not depend upon prior success with concrete operational tasks.

3.1.3. Constraint or Epiphenomenon? In order to evaluate these findings, it is important to recall the nature of the relationship between language and thought which is implied by the structural hypothesis. The claim is not that some concept must be acquired nonverbally before it can be expressed in

79

language. Rather, the structural hypothesis asserts that the rules which children derive from their interactions with linguistic or nonlinguistic objects will be only as complex or general as their underlying cognition allows. At one level of analysis, the data from studies of passive sentences and operativity are perfectly compatible with this hypothesis. They suggest that tasks which require similar thought processes have in general the same degree of difficulty for a given child regardless of whether these tasks are linguistic or nonlinguistic in nature. As children achieve some new capacity for structural complexity they may realize this potential in either a nonlinguistic or a linguistic mode. At another level of analysis, these studies remain quite unconvincing. Their pertinence to the structural hypothesis depends upon the assumption that use of the passive and success on concrete operational tasks require similar thought processes. As was argued earlier, it is not at all clear on theoretical ground that simple use of the passive entails either decentered or reversible mental structures. On empirical grounds the assumption appears equally unlikely. If, given a facilitating context and a grammatically uncritical listener, 3½-year-olds can both comprehend and produce passive sentences, we are left wondering why it takes another 2½years for this structural capacity to be realized in a nonlinguistic arena. One clue to this puzzle may be found in the nature of the linguistic tasks which have been used by experimenters who report relatively late acquisition of the passive voice. Comprehension tasks have involved sentence 80

verification tasks (Gaer, 1969; Slobin, 1966; Turner & Rommetveit, 1967a), successive active and passive sentences for single events (Beilin, 1975b), or use of the predicate follow (Sinclair et al., 1971). Production tasks have failed to provide a discourse context (Hayhurst, 1967), or have asked children to produce passive versions of the active sentences they have just used (Sinclair et al., 1971). Such tasks seem to require more than simple knowledge of a passive rule. They entail metalinguistic abilities, mastery of specific predicates, and the systematization of formerly independent syntactic structures. If so, the empirical association between concrete operational reasoning and competence with passive forms may be epiphenominal. The true structural links may lie between concrete operational reasoning and advanced linguistic abilities which have nothing special to do with the passive voice at all.

3.2. Conclusion and Summary Where does this leave the structural hypothesis? It remains a provocative and viable perspective on language-thought relationships which has yet to receive an adequate test. The work of Bates and her colleagues (Bates et al., 1979), e.g. their reports of developmental correlations between tool use and “instrumental” communication schemes, illustrates the possibilities of further research on structural prerequisites. The literature on children’s use of the passive, however, serves primarily to demonstrate the points at which this sort of investigation is vulnerable, i.e., the specification of underlying structural similarities, and the construction of appropriate experimental tasks. 81

Thus far I have reviewed the English child language literature as it pertains to two specific cognitive prerequisite hypotheses. The first hypothesis dealt with thought as the content of language. If children learn language by analyzing context and mapping form to meaning, then their conceptual and factual resources should constrain the acquisition process. Linguistic data provided preliminary support for this view. Children rely on nonverbal strategies to interpret unfamiliar linguistic input and express ideas agrammatically before they learn the appropriate forms. They acquire the members of some form classes in a predictable order over a protracted course despite their formal similarity. Word meanings change in predictable ways across children and new components of meaning are simultaneously incorporated into diverse formal patterns. These observations suggested the priority and guiding influence of nonverbal thought. Even stronger support for the conceptual prerequisite hypothesis was found in studies of conceptual development itself. The literature on spatial and quantificational notions in particular revealed patterns of development which were parallel to, but in advance of, related verbal achievements. Taken together, these lines of evidence provide a firm basis for the theoretical claim that nonverbal knowledge guides and constrains some aspects of language growth. The second hypothesis focused on the structure rather than the content of thought. It may be possible to characterize in general terms the degree of complexity, generality, and coordination which typifies thought at given developmental periods. If so, this structural 82

capacity should constrain the child’s organizing activity in a similar fashion across all domains of learning. To test this hypothesis, researchers have studied the relationship between children’s use of the passive voice and operational reasoning. Although some studies report positive correlations between performance in the two domains, the significance of these findings is arguable on both logical and empirical grounds. Simple use of the passive seems to require neither the coordination of multiple perspectives nor the interpretation of transformations, making this an inappropriate forum for evaluating the structural hypothesis—at least in relationship to operational thought. Evidence from other domains and periods will be needed before we can draw any conclusions about the role of structural capacity in language acquisition.

4. COGNITIVE PREREQUISITES AND CROSSLINGUISTIC STUDIES: THEORETICAL AND METHODOLOGICAL GUIDELINES As outlined in the opening sections of the chapter, this review was motivated by practical as well as theoretical concerns. Not only must we understand the role of cognitive prerequisites in order to build a thorough model of language development, we must be able to use our knowledge of specific prerequisites to design effective crosslinguistic studies of acquisition. The remaining portion of this chapter discusses the theoretical and methodological issues inherent in this second

83

task. We will consider the scope of cognitive hypotheses, the sorts of cognitive prerequisites which will prove most valuable in crosslinguistic studies, and the use of nonverbal assessment in studies of language growth.

4.1. Diversity in Language-Thought Relationships Particularly in the last decade, the language learning child has provided a forum for discussing the relationships between language and thought. Despite periodic lapses into hyperbole, this discussion has yielded at least one significant insight: that there is no single relationship between language and cognition. Language is the product of intellect. It is learned as children apply their knowledge of the world and their organizing capacities to the language object. Once learned, its competent use continues to demand planful and complex thinking, e.g. the ability to present an argument with the listener’s needs in mind. Language is also the tool of intellect. It is used to encode experience and to create imaginary worlds for reflective and expressive purposes. Children’s interpretation of events can be shaped by linguistic descriptions (Rommetveit, 1978; Grieve, Hoogenraad, & Murray, 1977), and their entry into certain conceptual domains is directed by conventional linguistic routines. Counting schemes and honorifics, for example, play significant roles in the construction of quantity concepts (Saxe, 1979b) and in the discovery of social categories (Rice, in press). Any notion of cognitive prerequisites must clearly be balanced by the recognition that language, in fact if not of necessity, facilitates intellectual growth. 84

Although the study of cognitive prerequisites can at best illuminate a portion of the language-thought mosaic, this portion is fascinating and complex. In this chapter, we have considered how the child’s developing knowledge might prepare him to learn new linguistic patterns and how new structural capabilities might open the way for increasingly complex operations on the language “object.” These two hypotheses far from exhaust the possibilities. Cognitive development during infancy certainly fits the child for symbol use, although the nature of this preparation remains unclear (e.g. Bowerman, 1978a; Bates et al., 1979; Miller, Chapman, Branston, & Reichle, 1980). Information processing mechanisms are necessary to linguistic rule discovery, and elements of cognitive style may determine specific approaches to grammar (Nelson, 1973b). These few examples suffice to indicate the diverse character of potential cognitive prerequisites. This diversity has important research consequences since each “cognitive hypothesis” focuses on a specific, often limited, set of linguistic phenomena, and requires a particular sort of confirming evidence (Jamison & Dansky, 1979). Of the two hypotheses discussed in this paper, the content hypothesis focused on the relational lexicon and required evidence of conceptual priority, while the structural hypothesis focused on syntax and required only evidence of correlated development. Failure to recognize such distinctions can only confuse our investigations.

85

4.2. Appropriate Candidates for Control Variables Given this diversity, which sorts of cognitive prerequisites are likely to prove useful in crosslinguistic studies of acquisition? The child language literature again provides some preliminary guidelines. The key factors in this decision are the developmental course of the cognitive variable and its temporal relationship to the language learning years. Three sorts of cognitive prerequisites emerge from this framework. First, there are those cognitive variables which develop during infancy and are essentially “in place” (Brainerd, 1979) prior to the emergence of speech. Such variables may be prerequisite to any language learning, but once the child begins to talk their influence remains constant. Candidates for this category include the relational notion of similarity, the coordination of means-ends schemes, and those mental processes which underlie the symbolic function. The study of such cognitive variables may illuminate the onset of language but will provide no explanations for particular patterns of language growth. As a case in point, consider the child’s acquisition of early noun “class” labels. The relational concept entailed by these terms is the notion of similarity. Beyond this, the words require only information about object functions and appearance. As noted earlier, the concept of similarity is certianly “early in place.” The 18-month-old infant can already use space to constitute object sets, and the 12-month-old will successively select similar objects for play. Thus, when we observe the child’s acquisition of early noun labels, we see not so

86

much a change in conceptual resources as a gradual discovery of conventional semantic maps. The concept of similarity may be prerequisite to this learning, but no child language fact bears witness to the relationship. It may be no accident that recent cautions about the explanatory adequacy of cognitive prerequisites have used verbal categorization evidence (Bowerman, 1981; Schlesinger, 1977; Dore, 1979). Verb meanings on the other hand seem to entail conceptual notions beyond mere similarity, notions which may or may not be available to the child just learning to speak. The child’s vocabulary of verbs certainly grows more slowly than his repertoire of nouns, but whether this reflects conceptual achievements or the fact that verbs refer to classes of events which have few perceptual correlates remains to be seen (Gentner, 1978). The second sort of cognitive variable has an uncertain early history (e.g. Olson, 1973; Huttenlocher & Burke, 1976; Flavell & Wellman, 1977), but may not undergo change even in infancy. Like the first set of cognitive variables, this second group seems to be in place by the time that language learning begins, and thus exerts a constant influence during the learning years. Unlike the first type, however, these variables in interaction with the structural properties of language affect the course of language learning since they constitute the primary tools for pattern discovery. Here I have in mind the basic information processing “hardware” discussed in the literature on memory and attention. Such cognitive mechanisms are, of course, the focus of crosslinguistic studies of language development. Their characters may be inferred by comparing 87

acquisition patterns across languages—as long as other cognitive variables are held constant. This condition is easily met in the case of variables such as means-ends coordination or the symbolic function which are available to all normally speaking children. It is the third type of cognitive variable which presents the methodological challenge. Wherever a cognitive variable is known to emerge DURING the language learning years and to pose constraints on the acquisition process, this variable must be experimentally or statistically controlled before we can draw conclusions about the nature of rule formation mechanisms. The prime candidates for such control variables are the spatial and quantificational concepts that seem now to play a significant role in the acquisition of English locatives and quantifiers. Future work may reveal analagous roles for logical, causal, and temporal notions as well. Viewed thusly, the success of crosslinguistic research will depend ultimately on our understanding of mental growth during the preschool years. With further study we may discover that young children “know” more and reason more complexly than is now evident. Indeed, in comparison with traditional views, the preschooler has already emerged as impressively competent (Gelman, 1978). The research reviewed here, however, indicates that while the two year old approaches language with significant intellectual tools, he must extend these abilities in many directions before mastering the linguistic system. Differences in conceptual resources,

88

left uncontrolled, can only confound our experimental studies of language development.

4.3. Assessment of Conceptual Prerequisites It does seem possible, at least as a long-term goal, to match children for pertinent knowledge resources prior to determining their relative mastery of expressive means. As argued above, such nonlinguistic control variables will be particularly important for studies of relational forms that entail concepts attained during the language learning years. The literature on conceptual prerequisites suggests a number of additional guidelines for the design of appropriately controlled research. First, nonlinguistic control variables should be conceived quite specifically for given linguistic forms. Notions such as EGOCENTRISM or PREOPERATIONAL THOUGHT seem less useful in predicting language knowledge than notions such as PROXIMITY or GREATER MAGNITUDE. Since the cognitive constraint is on meaning rather than form, we can use semantic analyses to estimate the relational concepts of interest. Secondly, nonverbal assessment tasks should require that sort of knowledge which is presumed to contribute to language learning. There has been increasing attention in the child language literature to various “levels” of knowing. Schlesinger (1977) speaks of the child’s early understandings as “vague and imperfect”; Genter (1978) distinguishes between “perceptual” and “conceptual” thought; Sinclair (1969) writes about “prestructures”; and Bowerman (1978b) notes that the child’s early 89

semantic knowledge is “unconscious and inaccessible.” Distinctions such as these invite us to specify the depth, or breadth, of understanding which we believe to be prerequisite to linguistic discovery (Sugarman, 1983; Siegel, 1978). This is not an easy task. Empirical studies should eventually provide some answers, but in the interim we must rely on theoretical, and poorly operationalized, constructs such as REFLECTIVE, CONSTRUCTIVE, or EVOCATIVE thought. Our assessment tools should at least require the child to go beyond the perceptual given to some transforming interpretation of experience. Thirdly, nonverbal assessment should focus differentially on recognitory or evocative processes. However else they may differ, language production seems to require more evocative control of forms than does comprehension. Likewise, some nonverbal tasks require more anticipation and construction of relations in the absence of perceptual cues than do others. If we plan to assess linguistic knowledge through a comprehension task, the appropriate nonverbal control measure would seem to be recognitory; if we plan to use an elicitation task, our nonverbal control measure should have a more constructive character. (See Nelson & Bonvillian, 1978) Finally, nonverbal tests of conceptual knowledge should require only the simplest response necessary to demonstrate the required content and level of thought (Braine, 1962). The literature on object permanence (Corrigan, 1978, 1979), and spatial perspectivism provides ample proof of the importance of this point. If all we are interested in is knowing whether a child can 90

evoke an object in its absence, we needn’t require him to track a toy through three invisible displacements. Likewise, if all we are interested in is whether a child can imagine spatial relations for alternative points of view, we needn’t require him to transform order relations in two dimensions simultaneously. These higher level achievements reveal cognitive growth, but have no obvious pertinence to language learning.

5. CODA My purpose in this chapter has been to review the English child language literature and evaluate the status of putative cognitive prerequisites. This endeavor was motivated by two particular needs: The need for a more fully explanatory account of language acquisition, and the need for a more powerful crosslinguistic methodology. From diverse research sources we have found mounting evidence in support of at least one cognitive hypothesis: that the child’s available knowledge sets limits on his ability to discover linguistic patterns. These constraints may apply only to particular aspects of language learning and have visible consequences only at particular ages; they may not be the sole determinant of language growth nor even the most influential; they constitute neither the only sort of cognitive prerequisite nor the only sort of relationship between thought and language; they are nevertheless important. Insofar as we understand the role of prior knowledge in language learning we can characterize one parameter of a multicausal model of acquisition and incorporate this

91

parameter into crosslinguistic designs. The current literature on conceptual prerequisites both affirms the viability of the crosslinguistic approach and directs us to a refined methodology. As-is-so often the case, however, this conclusion also gives rise to a beginning. The English evidence of conceptual prerequisites invites replication with crosslinguistic data. If these aspects of language learning are truly motivated by conceptual achievements, parallel trends should exist regardless of formal differences. Based on this review, we might expect to find early nonquantificational meanings for more, topological meanings for in back of, and a common order of development for locative expressions. Big/little should be the first spatial adjectives and habitual aspect should be late learned. To the degree that such predictions are confirmed across languages, we will have further reason to interpret these child language facts as evidence that cognitive growth exerts a guiding force in language acquisition.

ACKNOWLEDGMENTS The suggestions and comments of Dorothy Saltzman, Mabel Rice, Susan Carey, and Susan Sugarman were most useful in the preparation of this chapter. I thank them for their time and interest.

Author’s Note This chapter was written in 1981. While I remain committed to the major lines of argument presented here,

92

the literature review is no longer completely current. This fact is most noticeable in the section on early dimensional concepts where recent studies by L. Smith would enrich my account of language-thought relationships.

REFERENCES Ames, L., & Learned, J. The development of verbalized space in the young child. Journal of Genetic Psychology, 1948, 72, 63–84. Antinucci, F., & Miller, R. How children talk about what happened. Journal of Child Language, 1976, 3, 169–189. Baldie, B. The acquisition of the passive voice. Journal of Child Language, 1976, 3, 331–348. Barrie-Blackley, S. Six year old children’s understanding of sentences adjoined with time adverbs. Journal of Psycholinguistic Research, 1973, 2, 153–165. Bartlett, E. Sizing things up: The acquisition of the meaning of dimensional adjectives. Journal of Child Language, 1916, 3, 205–220. Bartlett, E. The acquisition of the meaning of colour terms: A study of lexical development. In R. Campbell & P. Smith (Eds.), Recent advances in the psychology of language. New York: Plenum Press, 1978.

93

Bateman, W. A child’s progress in speech with detailed vocabularies. Journal of Educational Psychology, 1914, 5, 307–320. Bateman, W. Two children’s progress in speech. Journal of Educational Psychology, 1915,6, 475– 493. Bates, E., Benigni, L., Bremerton, I., Camaioni, L., & Volterra, V. From gesture to first word: On cognitive and social prerequisites. In M. Lewis & L. Rosenblum (Eds.), Interaction, conversation and the development of language. New York: Wiley, 1977. Bates, E., Benigni, L., Bremerton, I., Camaioni, L., & Volterra, V. The emergence of symbols: Communication and cognition in infancy. New York: Academic Press, 1979. Bebout, L., Segalowitz, S., & White, G. Children’s comprehension of causal constructions with ‘because’ and ‘so’. Child Development, 1980, 51, 565–568. Beilin, H. Cognitive capacities of young children: A replication. Science, 1968, 162, 920–921. Beilin, H. Development of the number lexicon and number agreement. In H. Beilin (Ed.), Studies in the cognitive basis of language development. New York: Academic Press, 1975. (a) Beilin, H. Experiments on the passive. In H. Beilin (Ed.), Studies in the cognitive basis of language development. New York: Academic Press, 1975. (b)

94

Berndt, R., & Caramazza, A. The development of vague modifiers in the language of pre-school children. Journal of Child Language, 1978,5, 279–294. Bever, T. The cognitive basis for linguistic structures. In J. Hayes (Ed.), Cognition and the development of language. New York: Wiley, 1970. (a) Bever, T. The comprehension and memory of sentences with temporal relations. In G. B. Flores d’Arcais & W. J. M. Levelt (Eds.), Advances in psycholinguistics. Amsterdam: North-Holland Publishing Co., 1970. (b) Block, E., & Kessel, F. Determinants of the acquisition order of grammatical morphemes: A reanalysis and reinterpretation. Journal of Child Language, 1980, 7, 181–188. Bloom, L. Language development: Form and function in emerging grammars. Cambridge, MA: MIT Press, 1970. Bloom, L., One word at a time: The use of single word utterances before syntax. The Hague: Mouton, 1975. Bloom, L., Lifter, K., & Hafitz, J. Semantics of verbs and the development of verb inflection in child language. Language, 1980, 56, 386–412. Bohn, W. First steps in verbal expression. Pedagogical Seminary, 1914, 21, 578–595. Bowerman, M. Learning the structure of causative verbs. Papers and Reports on Child Language Development,

95

Dept. of Linguistics, Stanford University, 1974, 8, 142–178. Bowerman, M. Semantic and syntactic development: A review of what, when and how in language acquisition. In R. Schiefelbusch (Ed.), Bases of language intervention. Baltimore, MD: University Park Press, 1978. (a) Bowerman, M. Systematizing semantic knowledge: Changes over time in the child’s organization of word meaning. Child Development, 1978, 49, 977–987. (b) Bowerman, M. Cross-cultural perspectives on language development. In H. Triandis (Ed.), Handbook of cross-cultural psychology, Vol. 4. Boston: Allyn and Bacon, 1981. Boyd, W. The development of a child’s vocabulary. Pedagogical Seminary, 1914, 21, 95–124. Braine, M. Piaget on reasoning: A methodological critique and alternative proposals. In W. Kessen & C. Kuhlman (Eds.), Thought in the young child. Monographs of the Society for Research in Child Development, 1962, 27. Brainerd, C. Learning research and Piagetian theory. In L. Siegel & C. Brainerd (Eds.), Alternatives to Piaget: Critical essays on the theory. New York: Academic Press, 1978.

96

Brainerd, C. Commentary on Hood, L., & Bloom, L. What, when and how about why: A longitudinal study of early expressions of causality. Monographs of the Society for Research in Child Development, 1979, 44, 42–47. Brewer, W., & Stone, J. Acquisition of spatial antonym pairs. Journal of Experimental Child Psychology, 1975, 19, 299–307. Brown, A., & Murphy, M. Reconstruction of arbitrary versus logical sequences by preschool children. Journal of Experimental Child Psychology, 1975, 20, 307–326. Brown, R. A first language: The early stages. Cambridge, MA: Harvard University Press, 1973. Brush, L. Children’s meanings of ‘more.’ Journal of Child Language, 1976, 3, 287–289. Bullock, M., & Gelman, R. Numerical reasoning in young children: The ordering principle. Child Development, 1977, 48, 427–434. Bullock, M., & Gelman, R. Preschool children’s assumptions about cause and effect: Temporal ordering. Child Development, 1979, 50, 89–96. Carey, S. The child as word learner. In M. Halle, J. Bresnan, & G. Miller (Eds.), Linguistic theory and psychological reality. Cambridge, MA: MIT Press, 1978. (a)

97

Carey, S. “Less” may never mean more. In P. Smith & R. Campbell (Eds.), Recent advances in the psychology of language. New York: Academic Press, 1978. (b) Chapman, R., & Kohn, L. Comprehension strategies in two and three year olds: Animate agents or probable events? Journal of Speech and Hearing Research, 1978, 21, 746–761. Clark, E. On the acquisition of the meaning of “before” and “after.” Journal of Verbal Learning and Verbal Behavior, 1971, 10, 226–275. Clark, E. On the child’s acquisition of semantics in two sematic fields. Journal of Verbal Learning and Verbal Behavior, 1972, 11, 750–758. (a) Clark, E. Some perceptual factors in the acquisition of locative terms by young children. Proceedings of the Chicago Linguistic Society, 1972, 8, 431–439. (b) Clark, E. Non-linguistic strategies and the acquisition of word meanings. Cognition, 1973, 2, 161–182. (a) Clark, E. What’s in a word? On the child’s acquisition of semantics in his first language. In T. Moore (Ed.), Cognitive development and the acquisition of language. New York: Academic Press, 1973. (b) Clark, E. Strategies and the mapping problem in first language acquisition. In J. MacNamara (Ed.), Language learning and thought. New York: Academic Press, 1977.

98

Clark, E. Strategies for communicating. Development, 1978, 49, 953–959.

Child

Clark, E. Lexical innovations: How children learn to create new words. Papers and Reports on Child Language Development, Department of Linguistics, Stanford University, 1980, 18, 1–24. Clark, H. The primative nature of children’s relational concepts. In J. Hayes (Ed.), Cognition and the development of language. New York: Wiley, 1970. Clark, H. Space, time, semantics and the child. In T. Moore (Ed.), Cognitive development and the acquisition of language. New York: Academic Press, 1973. Coker, P. Syntactic and semantic factors in the acquisition of before and after. Journal of Child Language, 1978, 5, 261–277. Cook, N. The acquisition of dimensional adjectives as a function of the underlying perceptual event. Papers and Reports on Child Language Development, Department of Linguistics, Stanford University, 1976, 12, 81–88. Cook, N. In, on and under revisited again. Papers and Reports on Child Language Development, Department of Linguistics, Stanford University, 1978, 15, 38–45. Coots, J. Polarity and complexity in spatial adjective acquisition. Paper presented to The Society for Research in Child Development, 1975. Conigan, R. Language development as related to stage 6 object permanence

99

development. Journal of Child Language, 1978, 5, 173–189. Conrigan, R. Cognitive correlates of language: Differential criteria yield differential results. Child Development, 1979, 50, 617–631. Court, S. Numbers, time and space in the first five years of life. Pedagogical Seminary, 1920, 27, 71–89. Cromer, R. The development of language and cognition: The cognition hypothesis. In B. Foss (Ed.), New perspectives in child development. Baltimore, MD: Penguin Books, 1974. Cromer, R. The cognitive hypothesis of language acquisition and its implications for child language deficiency. In D. Morehead & A. Morehead (Eds.), Normal and deficient child language. Baltimore, MD: University Park Press, 1976. Cromer, R. Developmental strategies for language. In V. Hamilton (Ed.), The development of cognitive processes. New York: Academic Press, 1977. Cruse, P. A note on the learning of colour names. Journal of Child Language, 1977, 4, 305– 312. Denny, N. Free classification in preschool children. Child Development, 1972, 43, 1161–1170.

100

Dewart, H. A psychological investigation of sentence comprehension by children. Unpublished doctoral dissertation, University College, London, 1975. Donaldson, M., & Balfour, G. Less is more: A study of language comprehension in children. British Journal of Psychology, 1968, 59, 461–472. Donaldson, M., & Wales, R. On the acquisition of some relational terms. In J. Hayes (Ed.), Cognition and the development of language. New York: Wiley, 1970. Dore, J. What’s so conceptual about the acquisition of linguistic structures? Journal of Child Language, 1979, 6, 129–137. Ehri, L., & Ammon, P. The development of antonym adjective structures in children. Final Report Project No. O-I-045, U.S. Department of Health, Education and Welfare, Office of Education, 1972. Ehri, L. Comprehension and production of adjectives and seriation. Journal of Child Language, 1976, 3, 369–384. Eilers, R., Oiler, D. K., & Ellington, J. The acquisition of word-meaning for dimensional adjectives: The long and short of it. Journal of Child Language, 1974, 1, 195–204. Emerson, H. Children’s comprehension of ‘because’ in reversible and nonreversible sentences. Journal of Child Language, 1979, 6, 279–300.

101

Erreich, A., Valian, V., & Winzemer, J. Aspects of a theory of language acquisition. Journal of Child Language, 1980, 7, 157–180. Ervin-Tripp, S. Discourse agreement: How children answer questions. In J. Hayes (Ed.), Cognition and the development of language. New York: Wiley, 1970. Estes, K. Nonverbal discrimination of more and fewer elements by children. Journal of Experimental Child Psychology, 1976, 21, 393–405. Fishbein, H., Lewis, S., & Keiffer, K. Children’s understanding of spatial relations: Coordination of perspectives. Developmental Psychology, 1972, 7, 21–33. Flavell, J., & Wellman, H. Metamemory. In R. Kail & J. Hagen (Eds.), Perspectives on the development of memory and cognition. Hillsdale, NJ: Lawrence Erlbaum Associates, 1977. Flavell, J., Shipstead, S., & Craft, K. Young children’s knowledge about visual perception: Hiding objects from others. Child Development, 1978, 49, 1208–1211. Fraser, C., Bellugi, U., & Brown, R. Control of grammar in imitation, comprehension and production. Journal of Verbal Learning and Verbal Behavior, 1963, 2, 121–135.

102

French, L., & Brown, A. Comprehension of ‘before’ and ‘after’ in logical and arbitrary sequences. Journal of Child Language, 1977, 4, 247–256. Gaer, E. Children’s understanding and production of sentences. Journal of Verbal Learning and Verbal Behavior, 1969, 8, 289–294. Gale, M., & Gale, H. The vocabularies of three children in one family at two and three years of age. Pedagogical Seminary, 1902, 9, 422–440. Gelman, R. Cognitive development. In M. Rosenszweig & L. Porter (Eds.), Annual Review of Psychology, 1978, 29, 297–332. Gelman, R., & Gallistel, C. The child’s understanding of number. Cambridge, MA: Harvard University Press, 1978. Gentner, D. On relational meaning: The acquisition of verb meaning. Child Development, 1978,49, 988–998. Gitterman, D. Talking about comparisons: A study of comparative adjective use as tasks and attributes vary. Unpublished masters thesis, Indiana University, 1979. Goodson, B., & Greenfield, P. The search for structural principles in children’s manipulative play: A parallel with linguistic development. Child Development, 1975, 46, 734–746.

103

Grant, J. A child’s vocabulary and its growth. Pedagogical Seminary, 1915, 22, 183–203. Greenfield, P., Nelson, K., & Saltzman, E. The development of rulebound strategies for manipulating seriated cups: A parallel between action and grammar. Cognitive Psychology, 1972, 3, 291– 310. Grieve, R., Hoogenraad, R., & Murray, D. On the young child’s use of lexis and syntax in understanding locative instructions. Cognition, 1977, 5, 235–250. Griffiths, J., Shantz, C., & Siegel, I. A methodological problem in conservation studies: The use of relational terms. Child Development, 1967, 38, 841–848. Harris, L., & Strommen, E. The role of front-back features in children’s ‘front’, ‘back’, and ‘beside’ placements of objects. Merrill-Palmer Quarterly, 1972, 18, 259–271. Harwood, F. Quantitative study of the speech of Australian children. Language and Speech, 1959, 2, 236–270. Hayhurst, H. Some errors of young children in producing passive sentences. Journal of Verbal Learning and Verbal Behavior, 1967, 6, 634–639. Heber, M. The influence of language training on seriation of 5–6 year old children initially at different levels of descriptive competence. British Journal of Psychology, 1977, 68, 85–95.

104

Holland, V., & Palermo, D. On learning ‘less’: Language and cognitive development. Child Development, 1975, 46, 437–443. Holmes, T. Comprehension of some sizes, shapes, and positions by young children. Child Development, 1932, 3, 269–273. Horgan, D. The development of the full passive. Journal of Child Language, 1978, 5, 65–80. Hood, L., & Bloom, L. What, when and how about why: A longitudinal study of early expressions of causality. Monographs of the Society for Research in Child Development, 1979, 44. Hoogenraad, R., Grieve, R., Baldwin, P., & Campbell, R. Comprehension as an interactive process. In R. Campbell & P. Smith (Eds.), Recent Advances in the psychology of language. New York: Plenum Press, 1978. Huttenlocher, J., & Burke, D. Why does memory span increase with age? Cognitive Psychology, 1976, 8, 1–31. Ingram, D. The acquisition of questions and its relation to cognitive development in normal and linguistically deviant children: A pilot study. Papers and Reports in Child Language Development, Committee on Linguistics, Stanford University, 1972, 4, 13–19. Inhelder, B. Some aspects of Piaget’s genetic approach to cognition. In H. Furth (Ed.), Piaget and knowledge. Englewood Cliffs, NJ: Prentice-Hall, Inc., 1969.

105

Jamison, W., & Dansky, J. Identifying developmental prerequisites of cognitive acquisitions. Child Development, 1979, 50, 449–454. Johnson, H. The meaning of ‘before’ and ‘after’ for preschool children. Journal of Experimental Child Psychology, 1975, 19, 88–99. Johnston, J. A study of spatial thought and expression: In back and in front. Unpublished doctoral dissertation, University of California at Berkely, 1979. Johnston, J. Acquisition of locative meanings: Behind and In front of. Journal of Child Language, 1984, 11, 407–422. Johnston, J., & Slobin, D. I. The development of locative expressions in English, Italian, Serbo-Croatian, and Turkish. Journal of Child Language, 1979, 6, 529–545. Kuczaj, S., & Maratsos, M. On the acquisition of front, back and side. Child Development, 1975, 46, 202–210. Laurendeau, M., & Pinard, A. The development of the concept of space in the child. New York: International Universities Press, 1970. Layton, T., & Stick, S. Comprehension and production of comparatives and superlatives. Journal of Child Language. 1979, 6, 511–527. Lee, L. Developmental sentence analysis. Evanston, Northwestern University Press, 1974.

106

IL:

Lempers, J., Flavell, E., & Flavell, J. The development in very young children of tactic knowledge concerning visual perception. Genetic Psychology Monographs, 1977, 95, 3–53. Lempert, H. Extrasyntactic factors affecting passive sentence comprehension by young children. Child Development, 1978, 49, 694–699. Lempert, H., & Kinsbourne, M. Children’s comprehension of word order: A developmental investigation. Child Development, 1978, 49, 1235–1238. Levine, S. & Carey, S. Up front: The acquisition of a concept and a word. Journal of Child Language, 1982, 9, 645–658. Maratsos, M. Decrease in the understanding of the word ‘big’ in preschool children. Child Development, 1973, 44, 747–752. Masongkay, Z., McCluskey, K., McIntyre, C., Sims-Knight, J., Vaughn, B., & Flavell, J. The early development of inferences about the visual percepts of others. Child Development, 1974, 45, 357–366. Mehler, J., & Bever, T. Cognitive capacity of very young children. Science, 1967, 158, 141–142. Menyuk, P. Sentences children use. Cambridge, MA: MIT Press, 1969. Miller, J., Chapman, R., Branston, M., & Reichle, J. Language comprehension in sensorimotor

107

stages V and VI. Journal of Speech and Hearing Research, 1980, 23, 284–311. Moore, T., & Harris, A. Language and thought in Piagetian theory. In L. Siegel & C. Brainerd (Eds.), Alternatives to Piaget: Critical essays on the theory. New York: Academic Press, 1978. Nelson, K. Some evidence for the cognitive primary of categorization and its functional basis. Merrill-Palmer Quarterly, 1973, 19, 21–39. (a) Nelson, K. Structure and strategy in learning to talk. Monographs of the Society for Research in Child Development, 1973, 38. (b) Nelson, K. E., & Bonvillian, J. Early language development: Conceptual growth and related processes between two and four 1/2 years of age. In K. E. Nelson (Ed.), Children’s Language, Vol. 1. New York: Gardner Press, Inc. 1978. Nelson, K. E., Carskoddon, G., & Bonvillian, J. Syntax acquisition impact of experimental variation in adult verbal interaction with the child. Child Development, 1973, 44, 497–504. Nelson, K. E., & Nelson, K. Cognitive pendulums and their linguisitc realization. In K. E. Nelson (Ed.), Children’s Language, Vol. 1. New York: Gardner Press, Inc., 1978.

108

Newport, E., Gleitman, H., & Gleitman, L. Mother, I’d rather do it myself: Some effects and non-effects of maternal speech style. In C. Ferguson & C. Snow (Eds.), Talking to children. Cambridge: Cambridge University Press, 1977. Norman, D. Memory and attention: An introduction to human information processing (2nd edition). New York: Wiley, 1976. Nussbaum, N., & Naremore, R. On the acquisition of present perfect ‘have’ in normal children. Language and Speech, 1975, 3, 219–226. Olson, D., & Nickerson, N. Language development through the school years: Learning to confine interpretation to the information in the text. In K. E. Nelson (Ed.), Children’s Language, Vol. 1. New York: Gardner Press, Inc., 1978. Olson, G. M. Developmental changes in memory and the acquisition of language. In T. Moore (Ed.), Cognitive development and the acquisition of language. New York: Academic Press, 1973. Osherson, D. Three conditions on naturalness. Cognition, 1978, 6, 263–289.

conceptual

Palermo, D. More about less: A study of language comprehension. Journal of Verbal Learning and Verbal Behavior, 1973, 12, 211–221.

109

Pelsma, R. A child’s vocabulary and its development. Pedagogical Seminary, 1910, 17, 328–369. Piaget, J., & Inhelder, B. The child’s conception of space. New York: W. W. Norton and Company, Inc., 1967. Piaget, J., & Inhelder, B. Psychology of the child. New York: Basic Books, 1969. Ricciuti, H. Object grouping and selective ordering behaviors in infants 12 to 24 months old. Merrill-Palmer Quarterly, 1965, 11, 129–148. Rice, M. Cognition to language: Categories, word meanings and training. Baltimore: University Park Press, 1980. Rice, M. Cognitive aspects of communicative development. In R. Schiefelbusch & J. Pickar (Eds.), Communicative competence: Acquisition and intervention. Baltimore: University Park Press, in press. Richards, M. Adjective ordering in the language of young children: An experimental investigation. Journal of Child Language, 1979,6, 253–277. Roberts, R., & Fisher, K. A developmental sequence of classification skills. Paper presented to the Society for Research in Child Development, 1979. Rommetveit, R. On the relationship between children’s mastery of Piagetian cognitive operations and their

110

semantic competence. In R. Campbell & P. Smith (Eds.), Recent advances in the psychology of language. New York: Plenum Press, 1978. Rosch, E., Mervis, C., Gray, W., Johnson, D., & Boyes-Braem, P. Basic objects in natural categories. Cognitive Psychology, 1976, 8, 382–439. Saxe, G. Children’s counting: The early formation of numerical symbols. New Directions for Child Development, 3, 1979, 73–84. (a) Saxe, G. Developmental relations between notational counting and number conservation. Child Development, 1979, 50, 180–187. (b) Saxe, G., & Cohen, W. A priori knowledge and empirical operations: An analysis of developmental relations in the domain of numberical cognition. Unpublished manuscript, City University of New York, 1980. Schlesinger, I. M. The role of cognitive development and linguistic input in language acquisition. Journal of Child Language, 1977, 4, 153–170. Scholnick, E., & Adams, M. Relationships between language and cognitive skills: Passive voice comprehension, backward repetition, and matrix permutation. Child Development, 1973, 44, 741–746.

111

Shantz, C.,& Watson, J. Assessment of spatial egocentrism through expectancy violation. Psychonomic Science, 1970, 18, 93–94. Shatz, M. The relationship between cognitive processes and the development of communication skills. In C. Keasey (Ed.), Nebraska Symposium on Motivation: Social cognitive development, Vol. 25. Lincoln: University of Nebraska Press, 1978. Shatz, M., & Gelman, R. The development of communication skills: Modifications in the speech of young children as a function of listener. Monographs of the Society for Research in Child Development, 1973, 38. Siegel, L. Development of number concepts: Ordering and correspondence: Operations on the role of length cues. Developmental Psychology, 1974, 10, 907–912. Siegel, L. A lot about BIG, LITTLE and SAME: The relationship between quantity discrimination and the comprehension and production of language. Unpublished manuscript, McMaster University, 1976. Siegel, L. The relationship of language and thought in the preoperational child: A reconsideration of nonverbal alternatives to Piagetian tasks. In L. Siegel & C. Brainerd (Eds.), Alternatives to Piaget: Critical essays on the theory. New York: Academic Press, 1978.

112

Sinclair, A., Sinclair, H., & De Marcellus, O. Young children’s comprehension and production of passive sentences. Archives de Psychologic 1971, 41, 1–22. Sinclair, H. Developmental psycholinguistics. In D. Elkind & J. Flavell (Eds.), Studies in cognitive development: Essays in honor of Jean Piaget. New York: Oxford University Press, 1969. Slobin, D. I. Grammatical transformations and sentence comprehension in childhood and adulthood. Journal of Verbal Learning and Verbal Behavior, 1966, 5, 219–227. Slobin, D. I. Cognitive prerequisites for the development of grammar. In C. A. Ferguson & D. I. Slobin (Eds.), Studies of child language development. New York: Holt, Rinehart, and Winston, 1973. Smith, L. Perceptual development and category generalization. Child Development, 1979, 50, 705– 715. Smith, L., & Kemler, D. Developmental trends in free classification: Evidence for a new conceptualization of perceptual development. Journal of Experimental Child Psychology, 1977, 24, 279–298. Smith, M. The influence of age, sex and situation on the frequency, form and function of questions asked by preschool children. Child Development, 1933, 4, 201–213.

113

Strohner, H., & Nelson, K. E. The young child’s development of sentence comprehension: Influence of event probability, nonverbal context, syntactic form, and strategies. Child Development, 1974, 45, 567–576. Sugarman, S. Scheme, order and outcome: The development of classification in children’s early block play. Unpublished doctoral dissertation, University of California, Berkeley, 1979. Sugarman, S. Children’s early thought: Developments in classification. Cambridge, MA: Cambridge University Press, 1983. Tanz, C. Studies in the acquisition of deictic terms. Cambridge: Cambridge University Press, 1980. Tremaine, R. Syntax and Piagetian operational thought. Washington, D.C.: Georgetown University Press, 1975. Thorns, J. Conservation: More or less. Unpublished doctoral dissertation, University of Kansas, 1977. Townsend, D. Do children interpret ‘marked’ comparative adjectives as their opposites. Journal of Child Language, 1976, 3, 385–396. Turner, E., & Rommetveit, R. The acquisition of sentence voice and reversibility. Child Development, 1967, 38, 649–660. (a) Turner, E., & Rommetveit, R. Experimental manipulation of the production of active and passive

114

voice in children. Language and Speech, 1967, 10, 169–180. (b) Tyack, D., & Ingram, D. Children’s production and comprehension of questions. Journal of Child Language, 1977,4, 211–224. Wales, R., & Campbell, R. On the development of comparison and the comparison of development. In G. B. Flores deArcais and W. J. M. Levelt (Eds.), Advances in Psycholinguistics. Amsterdam: North-Holland Publishing Co., 1970. Wales, R., Garman, M., & Griffiths, P. More or less the same: A markedly different view of children’s comparative judgments in three cultures. In E. Walker & R. Wales (Eds.), Approaches to language mechanisms. Amsterdam: North-Holland Press, 1976. Walkerdine, V., & Sinha, C. Spatial and temporal relations in the linguistic and cognitive development of young children. Unpublished manuscript, University of Bristol, 1974. Wannemacher, J., & Ryan, M. ‘Less’ is not ‘more’: A study of children’s comprehension of ‘less’ in various task contexts. Child Development, 1978, 49, 660–668. Washington, D., & Naremore, R. Children’s use of spatial prepositions in two and three dimensional tasks. Journal of Speech and Hearing Research, 1978,21, 151–165.

115

Watson, J., Hayes, L., & Vietze, P. Bidimensional sorting in preschoolers with an instrumental learning task. Child Development, 1979,50, 1178–1183. Webb, P., & Abrahamson, A. Stages of egocentrism in children’s use of ‘this’ and ‘that’: A different point of view. Journal of Child Language, 1976, 3, 349–368. Weiner, S. On the development of ‘more’ and ‘less’. Journal of Experimental Child Psychology, 1974, 17, 271–287. Wilcox, S., & Palermo, D. In, on and under revisited. Cognition, 1974/5, 3, 245–254. Windmiller, M. The relationship between a child’s conception of space and his comprehension and production of spatial locatives. Unpublished doctoral dissertation, University of California, Berkeley, 1973. Woodward, M., & Hunt, M. Exploratory studies of early cognitive development. British Journal of Educational Psychology, 1972, 42, 248–259. 1

Nonverbal concept discovery tasks have, however, yielded mixed results on this point: Siegel (1976) and Bullock and Gelman (1977) found no performance differences between their greater and lesser magnitude conditions, while Estes (1976) found that children in the greater magnitude condition learned the rule faster. Related studies of spatial adjectives have likewise failed to find that positive pole or greater extent adjectives are consistently easier to comprehend (Eilers et al., 1974; 116

Coots, 1975; Bartlett, 1976; Carey, 1978a). Even the original “less equals more” phenomenon disappears with task modifications (Carey, 1978b; Wales et al., 1976), leading some recent investigators to propose that comprehension strategies are quite task-specific (Wannemacher & Ryan, 1978). 2

The use of capital letters here indicates that these terms represent lexical families rather than specific words. Different children may learn different members of a virtually synonomous set, e.g., in back of vs behind.

117

12

Function, Structure, and Language Acquisition*

T. Givón Linguistics Department University of Oregon, Eugene and Ute Language Program, Southern Ute Tribe Ignacio, Colorado

Contents Preamble Some Reflections on “Error” Analysis Typology of What: Reflections on the Input Typology and Functional Domains Types of Linguistically-Coded Message

118

Functional Domains in Syntax Structure vs. Function in Language Acquisition The Nature of the Correlation Between Structure and Function The Independence of Structure and Function Structural Irregularity Syntactic Complexity Functional Compexity Structural Complexity: More vs. Different Cultural Context

1. PREAMBLE Language typologists with any but the most trivial interest in cross-language diversity spend much of their time worrying about the balance between specificity and universality in human language. Their methodology in approaching the study of language universals may be likened to a controlled, multi-variable experiment in science: By comparing individual languages in any particular domain, one rules out gradually but systematically the typologically-unique features of each language, all the while hoping that a coherent residue of language-universal properties will remain behind after the dust has settled. The crosslinguistic study of 119

language acquisition may be conducted likewise, by ruling out what is language-specific about the acquisition process of any particular language, so that universals of language acquisition—and through them universals of human language—may someday emerge from the oft-impenetrable maze of exotic facts. The extension of acquisition studies toward cross-language typological comparison thus marks the growing empirical maturity of an important sub-field of linguistics. Since all children start roughly at the same human-universal cognitive starting-point, the typological study of the INPUT to language acquisition becomes that much more important: It is the INTERACTION between the language specific input and the cognitive human-universals present at the starting point which is, after all, responsible for the acquisition of specific languages. I will therefore assign the first two sections of this paper to discussing some general problems concerning the characterization of the input to language acquisition and its relation to what is acquired.

2. SOME REFLECTIONS ON “ERROR” ANALYSIS If the output produced by the child during various stages of first-language acquisition were identical to adult language, the study of child language acquisition would have never begun. It is, thus, the heuristic effect of the DISCREPANCY between the output and what is TAKEN TO BE the input which gave the initial impetus to the whole investigation. However, it seems to me that there are sufficient reasons for challenging the concept of

120

“erroneous learning” not only on trivial semantic grounds (“discrepancy” or “difference” vs. “error,” a more loaded term), but also on broader methodological grounds. First, and least worrisome, is the problem of methodological EMPHASIS. Characterizing the child’s output as “error” may—though doesn’t logically have to—lead to an overly normative approach to the INPUT. In spite of the fact that this is not a logical necessity, the tendency is still there. In the same vein, conceiving of the output as “error” may lead to de-emphasizing what is systematic and universal about it, in favor of over-emphasizing what is idiosyncratic, language-specific, or irregular. While this again is not a logical necessity, it is an observed tendency nontheless. More worrisome, however, is the nagging feeling that the entire conceptual framework of “erroneous learning” is an unsuitable analytical tool with which to crack the process of first language acquisition. The child’s access to actual input facts and his/her ability to generalize upon them during successive early states of language learning is severely delimited by cognitive development. But at each successive stage the child seems to have arrived at a definite, coherent generalization. That generalization may be then matched to the input to yield two reasonably obvious conclusions, at least potentially: (a) It is DIFFERENT from the generalization made by the adult producer of the input; and

121

(b) It is coherent and “right” with respect to a more limited sub-set of the data that was BELIEVED to have been contributed in the adult input. But there is nothing necessarily “error-like” about either (a) or (b) above. Concerning (a), first: What the adult (and with him/her the linguist) perceives to be his/her generalizations about the input is not necessarily accessible to the child at this particular juncture. The adult generalization is, after all, the end product of protracted sifting through mountains of input data, some highly regular, some less so, some downright chaotic. Such a process of gradual refinement should NOT be expected to serve as model for the child’s generalization at any specific point. But the concept of “erroneous learning” in fact leads one to make precisely this kind of dubious analogy. Concerning (b), next: It is not clear, a priori, what sub-set of the facts present in any individual sequence of adult input is the one which the child specifically extracts and uses to generalize upon at any particular “learning point.” One thus must open the discussion of what exactly is the ACTUAL, RELEVANT input used by the child at any SPECIFIC “instance of learning.” As is only to be expected, this opens a veritable Pandora’s Box, thus leading us into the next question.

3. TYPOLOGY OF WHAT: REFLECTIONS ON THE INPUT The growing interest in typological cross-language comparisons of child language acquisition is due, in

122

large part, to a realization that the input in the acquisition of one’s first language matters ENORMOUSLY. We have, so it seems, put behind us Chomsky’s cavalier trivilization of input: “… The native speaker acquires a grammar on the basis of very restricted and degenerate evidence …” (1968, p. 23). There still remains the danger, however, that the input we are so busy characterizing and typologizing is, in a large measure, a purely theoretical construct. Most obviously—and easily corrected—we tend to typologically characterize some version of an official normative grammar of the adult “language” as described by linguists, then tacitly assume that we have therefore characterized the input. Nothing is further from the truth. Forewarned, we proceed to study the actual adult utterances found in sufficiently high frequency in the child’s learning environment. But soon we find out that some of those are IRRELEVANT as input for the actual “instance of learning.” Those irrelevant portions of the input may be broadly divided into two categories: (i) INPUT THE CHILD HAS ALREADY GENERALIZED UPON: This input may be present at high frquency around the child, but nevertheless has absolutely ZERO relevance to what the child is learning at a PARTICULAR point, except perhaps as residual reinforcement. (ii) INPUT TOO COMPLEX FOR THE CHILD TO GENERALIZE ON: This input may again be present at high frequency around the child, but is nevertheless largely IRRELEVANT to what the child is LIKELY to be able to absorb at that 123

particular point in his/her cognitive and linguistic development. At any particular point, for studying any particular “instance of learning,” one must remain cognizant of the thin band of phenomenological experience where learning can at all take place: New information that is TOO REMOTE from what has already been acquired cannot be integrated. While new information that is TOO CLOSE to already acquired knowledge tends to appear redundant and thus to present little incentive for learning. Learning—the integration of new experience into an existing cognitive network—normally takes place somewhere between these two prohibitive extremes.1 The adult input, so it seems, cannot be characterized once and for all, in a uniform, stable fashion. Rather, at any given point one must discover (a) what the child is ACTUALLY LEARNING, (b) what the child has ALREADY LEARNED, and (c) what the child is yet whatever reason.

INCAPABLE OF LEARNING,

for

It is only (a), and the input pertaining to it, that is actually relevant for the purpose of studying the relation between input and output in the acquisition of language. Our discussion above has already pointed out one obvious fact concerning inputs and outputs: The DISCREPANCY between them COULD NOT BE VERY GREAT

124

This is simply an axiom of any learning process. It is, of course, one more reason why the concept of “erroneous learning” is not terribly viable. But the problem is not purely theoretical. There is a great amount of evidence suggesting that whenever speakers, at whatever level, encounter evidence of incomprehension, they automatically TONE DOWN their input. The evidence comes from foreigner-talk and from pidginization,2 but is also characteristic of the way adults communicate with children,3 What TONING DOWN amounts to, in terms of our discussion, is reducing the discrepancy between the adult input and the child output. But one cannot characterize this toning down across the board. This is so because within the very same adult utterance directed—perhaps didactically—at the child, some portions may pertain to what the child already knows, while other portions may have not been toned down sufficiently and must therefore be counted as FAILED INPUT. Others yet may have been toned down TOO MUCH, and thus teach the child nothing new, and should therefore be also considered FAILED INPUT. If this is indeed true, as I believe it is, it must force upon us a great analytic complexity in terms of localizing and specifying in minute detail the actual “instance of learning” and its RELEVANT INPUT and output. Such specificity and complexity may indeed be mind boggling, but we have little choice but to embark upon it if we are indeed concerned about the interaction between relevant input and output. In terms of typological characterization, then, it is quite possible that the discrepancy between input and output in language acquisition is rather minimal. As a corollary, 125

one must consider the possibility that the great crosslinguistic typological diversity observed in adult language is irrelevant at the earliest stages of child language acquisition, since at POINT ZERO all inputs are presumably toned down to exactly the same place. Only later, gradually, does the adult input begin to diverge, eventually pulling the child speech toward the adult “standard.”

4. TYPOLOGY AND FUNCTIONAL DOMAINS When one undertakes a typological cross-language comparison of any sub-area of language, a tacit assumption that is often left unarticulated is that one makes a comparative typology of structures coding the same FUNCTIONAL DOMAIN. The relation between structure and function in syntax is inherently the same CODING RELATION as between meaning and sound in the lexicon. However, an oft-made mistake is to assume that the notion “function” is the same notion in syntax as it is in the lexicon. It would thus be of value to briefly sketch out the various, types of function one may observe in language.

4.1. Types of linguistically-coded message One may broadly observe four major FUNCTIONAL REALMS in language, each covering a fairly coherent and traditionally recognized terrain, and each involving fairly distinct CODING-PRINCIPLES: Lexicon, prepositional syntax, complex syntax and thematics.

126

(i) Lexicon (a) Function: Expressing the terms/concepts

MEANINGS

of individual

(b) Coding principle: Sound sequences bunched together in fixed order to spell words or morphemes (ii) Propositional syntax (a) Function: Expressing PROPOSITIONAL MEANING in terms of the verb and the various case-role participants in events, actions, or states (b) Coding principle: The use of the three syntactic CODING DEVICES—word order, morphology, and intonation—to create the STRUCTURE of the neutral/ canonical clause-type: The main declarative-affirmative-active clause (iii) Complex syntax (a) Function: Expressing largely DISCOURSE-PRAGMATIC (but partially semantic) variations on the neutral/ canonical clause type (such as embedded, questions/ commands, negative, passive, etc.) (b) Coding principle: The use of the three coding devices—word order, morphology, and intonation—to create various STRUCTURAL VARIATIONS from the neutral/ canonical clause-type (iv) Thematics

127

(a) Function: Express the larger organizational notions of the theme/story (b) Coding principle: Only marginally coded in the normal linguistic terms, most commonly in terms of sequencing, juxtaposition, and CONJUNCTION devices Of these four major functional realms, the lexicon has been most clearly understood, thematics has been largely ceded to literature and rhetorics, and syntax remains the core functional realm studied by linguists.

4.2. Functional domains in syntax Of the two realms normally coded by syntax, (ii) and (iii) above, early syntactic typologists concerned themselves primarily with (ii). It is thus not an accident that their primary interests were WORD ORDER and CASEMARKING typology (ergativity vs. nominativity, etc.). These are indeed the major syntactic devices (alongside with intonation, to a lesser extent) used to code PROPOSITIONAL INFORMATION, i.e. the functional realm of the main-declarative-affirmative-active clause. This is the neutral/canonical clause-type in language, the one that tells us who did what to whom when, where, how or why and what for. It imparts the gist of events/actions/states, and carries the BULK OF NEW 4 INFORMATION in discourse. While this is the SIMPLEST functional realm in syntax, it is already quite complex. It must code the difference between state (no change), event (change over time) and action (change precipitated by an agent). It must code the SEMANTIC role of the various participants. And it most commonly must assign 128

at least one of them a prominent PRAGMATIC role, that of 5 TOPIC/SUBJECT. All these sub-functions must in some way or another be preserved in non-neutral/ non-canonical clauses, where a number of pragmatic and occasionally semantic variations on the neutral theme are performed. The most common variations have to do with the CONVERSES of the four-way definition of the neutral theme: (1) (i) Embedded clauses: Contrasting with main clauses, embedding represents either PRAGMATIC SUBORDINATION (“backgrounding”) in terms of new vs. old information,6 or SEMANTIC SUBORDINATION in terms of the independence of the event of a subordinate verb from that of the main verb.7 (ii) Non-declarative speech acts: Most commonly interrogatives and imperatives, but in many languages also a subtle variety of other intermediate speech acts. (iii) Negative clauses: Contrasting with the canonical/ neutral affirmative both semantically and pragmatically.8 (iv) Passive cluases: Contrasting with the canonical/ neutral active clause. But there are other functional variations with a great deal of crosslinguistic attestation.9 When one moves away from the canonical/neutral clause toward more complex functional domains, the ones involving most commonly the area of discourse-pragmatics, the complexity of the notion 129

“function” multiplies. What are normally taken to be single functions reveal themselves to be: (a) Graded/continuous clines rather than discrete/binary/ categorial entities; (b) Multi-dimensional spaces;

rather

than

unidimensional

(c) Interactive or interdependent across dimensions within the same domain or even, on occasion, between “different” domains. The best way of illustrating the resulting complexity of the notion “syntactic function” is by citing one example, that of the functional domain of PASSIVIZATION.10 The functional domain of the passive clause involves three separate but mutually-dependent dimensions: (2) Topic assignment: The canonical/neutral clause assigns the function of subject/topic to the participant/ argument HIGHER on the following hierarchy:

AGENT>DATIVE>PATIENT>INSTRUMENT>LOCATION>OTHERS In passivization, a participant LOWER on this hierarchy than the original subject/topic is “promoted” to the subject/topic function. (3) Impersonalization: The canonical/neutral clause expresses prominently the identity of the SUBJECT/AGENT responsible for (or involved in) the event. In

130

passivization this information is fashion.

SUPPRESSED

in some

(4) De-transitivization: The canonical/neutral clause tends to construe the proposition as a PROCESS or ACTION, i.e. a CHANGE in the state of the universe. The passive clause tends to focus on the resulting STATE, or in other ways tone down the transitivity of the proposition. Within each one of these sub-domains, further, the passive clause is only one of several SYNTACTIC CODING POINTS. Each domain is thus, potentially, a functional CONTINUUM along which any individual language recognizes a number of discrete functional points it then marks overtly by SYNTACTIC STRUCTURES/ DEVICES. Thus, domain (2) above, that of topic assignment, involves the major graded functional dimension of DEGREE OF SURPRISE or DIFFICULTY OF TOPIC IDENTIFICATION. Along this scalar dimension, the following common syntactic devices distribute in the following hierarchic order:11 (5) Least surprising/most continuity (easiest topic identification) zero anaphora clitic pronouns/verb agreement unstressed pronouns stressed/independent pronouns definite NP left dislocation cleft/focus constructions Most surpising/least continuous (most difficult topic identification)

131

In terms of the syntactic coding-devices used to mark this domain, constructions at the low-surprise end of the scale tend to exhibit less deviation from the neutral/ canonical norms in word order, casemarking, verb agreement, or intonation; while those at the high-surprise end of the scale tend to exhibit MORE deviation from the norm. Within domain (3), that of de-personalization, one must consider the passive alongside with other constructions, such as the use of one, you, or we in English, the third-person plural verb agreement in Hebrew, the German man and the impersonal-passive es, the Dutch impersonal-passive er, the French on or the Spanish se construction. Most languages have more than one syntactic device coding points on this functional continuum. But the structural common denominator involves DELETION of the agent/subject of the active, admittedly a rather iconic expression for the function of suppressing the agent’s identity. Finally, within domain (4), that of de-transitivization, one must consider alongside with the passive also reflexives, reciprocals, middle-voice/statives, as well as adjectives, perfective-resultative verb forms, and even a number of nominalizations. The structural coding devices most commonly associated with this function involve an adjectival/stative/perfective/nominalized form of the verb, thus marking it as STATE, plus a decrease in the number of arguments overtly expressed in the clause—thus marking de-transitivization, as well as the assignment of a less-agentive argument in the subject/ topic position. 132

That three functional sub-domains consistently cross or conflate in the passive construction is not entirely accidental. The rationale for such cross-language conflation may be summed as follows: (6) If the identity of the agent/subject is suppressed, by some means, the language will confer the subject/topic function on another—the likeliest—argument. Hence the conflation of the de-personalization and topic-assignment sub-domains. (7) If the perfective/resultative/stative aspect of the proposition is emphasized, perforce the status of the agent/subject is downgraded. Hence the conflation between the de-transitivization and de-personalization sub-domains. (8) If the clausal topic is a non-agent, it is likely that the more patient-related aspects of the proposition—such as the resulting state—are focused upon. Hence the conflation between the topic-assignment and the de-transitivization sub-domains. We are, it seems, dealing with a multidimensional, scalar and interdependent notion of “function” in syntax, where the various sub-domains may exhibit different mixes of semantic, propositional and pragmatic functions. Further, the syntactic coding of these sub-domains may involve different principles of ICONICITY. When one discusses the child’s acquisition of syntactic function, then, one may as well bear this enormous complexity in mind.

5. STRUCTURE VS. FUNCTION IN LANGUAGE ACQUISITION

133

5.1. The nature of the correlation between structure and function While human language is not a perfect code system, one of the most overwhelming facts about is the degree to which it tends to preserve, by whatever means, the ideal of 1: 1 correlation between the code and the message. In general, one may postulate two different types of situations where this 1: 1 correlation is supposedly violated: (a) AMBIGUITY: One code unit conveys more than one message units (b) SYNONYMY: One message unit is coded by more than one code units Of these two theoretical possibilities, synonymy is all but UNATTESTED in adult human language, either at the lexical or the syntactic level. Further, diachronic evidence shows consistently that whenever two forms (“code units”) approach a near-synonymy point, one of them diverges in its meaning/message, either semantically, pragmatically, stylistically, dialectally, etc., up to and including OBSOLESCENCE of one of the forms. Ambiguity, on the other hand, is widely attested at both the lexical and syntactic levels. However, it does consistently get RESOLVED IN CONTEXT. In other words, the only way one can construe ambiguity as offsetting the 1:1 correlation between code and function in language is by disregarding context. But since language is only used IN context, the actual role of ambiguity in language use is to quite an extent illusory.

134

When one looks at diachronic change in syntax, one finds most commonly that some functional domains are OVER-CODED—with many syntactic structures crowding a “small” domain—while others are UNDER-CODED—with a relatively large domain coded by few syntactic structures. Schematically, within one domain-continuum, one can represent both situations as in:

with A-F being recognized syntactic structures. Let me illustrate this possible discrepancy in CODING DENSITY by contrasting the way two languages, English and Spanish code the same functional domain. Consider first the way Spanish and English code the lowest portion of the domain of the TOPIC-RECOVERABILITY/CONTINUITY functional domain (see (5) above). In Spanish, one cannot have a ZERO expression of the subject, since grammatical agreement is obligatory on the verb. Thus, when coding the LOWEST—most continuous—value on this functional domain, Spanish would match its inflected verbs for the uninflected (“zero” subject) English verbs, as in: a. ENGLISH: …

So then he entered the house, proceeded to the window, took a chair and sat down…

b. SPANISH: … Luego entró en la casa, procedió a la ventana, tomó una silla y se sentó…

135

The English unstressed pronoun he is matched to the subject-agreement (=ó) in Spanish, but the more-continuous use of ZERO in English is ALSO matched to the subject-agreement coding-point in Spanish. Thus, for this lowest portion of the topic-continuity functional domain, Spanish is UNDER-CODED as compared to English, since it codes with ONE point a functional domained covered in English by TWO coding points. On the other hand, at another functional domain, that of the passive-impersonal-intransitive, English seems to have perhaps one less coding point than Spanish, since the English “passive” may be matched by two Spanish constructions, the older “agented” passive, and the “impersonal” =se construction. Thus consider: a. OLDER SPANISH PASSIVE: Las mujeres fueron curado (por los brujos) the women were cured by the sorcerers ‘The women were cured (by the sorcerers)’ b. THE SPANISH: SE-PASSIVE se-curó a las mujeres REFL-cured OBJ the women ‘The women were cured’ When conditions such as schematized in (9) above prevail in a language, the most common diachronic change is that one of the structures A-E which covers an OVER-coded section of the domain shifts its function and “moves” toward covering a point on the UNDER-coded section of the domain. Normally, further, it is the

136

structure that satisfies requirements:

PREFERABLY

two idealized

(a) It is FUNCTIONALLY more similar to the under-coded domain than any of the other structures on the over-coded domain; and (b) It is STRUCTURALLY more similar to one of the structures on the under-coded domain than any of the other structures on the over-coded domain. Ideally, with a high degree of structural ICONICITY, requirements (a) and (b) ought to work in tandem. So that, for example, coding point (E) in (9) above should be STRUCTURALLY the closest to (F) in addition to also being FUNCTIONALLY the closest. And the schematic situation in (9) may yield, as a result of diachronic change (or “analogical change”), the situation in (10):

One must bear in mind, however, that situations do arise where a structure may be functionally close to another but structurally more remote (than other structures), and that under such circumstances humans may sometimes ANALOGIZE by functional similarity and sometimes by structural similarity. While over the long run such conflict situations tend to be zeroed out of the grammar, their existence nevertheless points out to an important fact: That humans are quite capable of perceiving structure and function INDEPENDENTLY of each other, i.e.,

137

that they have both functional intuitions and structural intuitions.

5.2. The independence of structure and function The acquisition of a communicative code is, without any shred of doubt, the acquisition of systematic correlations between structure (“code”) and function (“message”). But for the term “correlation” not to be tautological, one must assume that both structure and function are perceived independently. There is no reason, however, to assume that the sense/intuition about language structure is acquired IN A FUNCTIONAL VACUUM. On the contrary, one overwhelming fact about the structural input in language acquisition is that it can NEVER be presented to the child without functional input, since the linguistic message is ALWAYS present in its structural garb. The same is obviously true of the functional input: It is never presented IN A STRUCTURAL/CODING VACUUM. Nevertheless, I believe there exists enough evidence to suggest that speakers do develop independent intuitions about structure and function. And indeed those separate intuitions— coupled with the ever-present tendency/ imperative to correlate structure and function—are displayed in linguistic behavior. The model of over-coded and under-coded functional domains (or sections of the same domain-continuum), it seems to me, could serve to illuminate similar behavioral facets in the child’s acquisition of language. Thus, for example, it is possible for the child to acquire a certain functional domain with relatively few coding points, i.e. 138

an

UNDER-CODED

domain with a high potential for

AMBIGUITY

The child would then presumably be on the lookout for STRUCTURES that will ENRICH or FURTHER-CODE this domain, exercising his/her sense of both structural and functional similarity. On the other hand, the child could also acquire initially an OVER-CODED domain with a high potential for SYNONYMY. Then presumably the child will be on the lookout for FUNCTIONS that are under-coded, in order to shift the superfluous structure to them. And again the child would presumably be exercising his/her capacity for BOTH structural and functional analogy in this process. Thus, in order to resolve the problem of either potential synonymy (over-coding) or potential ambiguity (under-coding), the child must differentially recognize either FUNCTION or STRUCTURE, respectively. Still, the underlying principle in both cases seems to be the idealized tendency for 1:1 correlation between code and message. In approaching the relation between form and function in language, I believe one must reject either the extreme transformationalist position of the total INDEPENDENCE of structure, or the extreme functionalist position of the total IRRELEVANCE of structure, in favor of a more realistic assessment of the exact role of structure in human language AND its acquisition.12 But above all, I think, one must keep in mind that the child’s motivation in acquiring language is overhwelmingly a FUNCTIONAL motivation.

139

5.3. Structural irregularity I believe there is an unfortunate tendency, in both linguistics at large and in child language studies, to over-represent the extent of irregular morphology at the expense of the immense amount of regularity—defined in terms of transparent 1:1 correlation between structure and function—found in syntax. In syntax, this tendency has resulted in over-weigh ting rules of grammar that can be “broken” at the expense of those that seldom are, to the point where the latter become largely invisible and are never described. In child language studies, it seems to me, a similar tendency persists, with a real prospect of over-generalizaing the irregular MARGINS of the grammatical system at the expense of its overwhelmingly regular core. I would like to approach this topic from three different perspectives.

5.3.1. Syntacticization I have shown elsewhere13 that syntactic structure—characterized by tight constructions under extended intonation contours with attendant grammatical/ inflectional morphology—arises ontogenetically, diachronically, and probably also phylogenetically by syntacticization of another communicative mode, the pragmatic/paratactic mode. The two can be contrasted in the following way: (11)

pragmatic mode

syntactic mode

(a) topic-comment structure

subject-predicate structure

(b) loose coordination

tight subordination

140

(c) slow rate of delivery

fast rate of delivery

(d) small chunks under one intonation contour

large chunks under one intonation contour

(e) lower noun/verb ratio in discourse, with more simple verbs

higher noun/verb ratio in discourse, with more complex verbs

(f) no use of grammatical morphology

extensive use of grammatical morphology

The survey of available evidence suggests, among other things, that: (i) All complex syntactic constructions arise from the syntacticization of paratactic constructions of the pragmatic mode; (ii) Children acquire first the pragmatic mode of communication, then gradually syntacticize it; and (iii) Adults retain a whole range of modes from the pragmatic mode upward, and under appropriate conditions use them appropriately. While the pragmatic mode is a slow means of processing, it has one obvious advantage over the syntactic mode: It is a more TRANSPARENT communicative system, with a simpler 1: 1 correlation between code and message. The syntactic mode, on the other hand, is a FASTER, SEMI-AUTOMATED mode of processing, but it has lost a certain amount of communicative FIDELITY and tends to exhibit either a LOWER correlation between code and message or a more

141

correlation. To illustrate this, consider the difference betvveen “topic” and “subject” constructions, as in the following left-dislocation: COMPLEX

In early childhood, topic-comment structures are not like (12), in that they don’t involve both “topic” and “subject,” but only “topic.” Further, the function of “topic” is undifferentiated or “under-coded,” not distinguishing between NEWLY-INTRODUCED topic and CONTINUING topic. In syntacticized speech, on the other hand, the left-dislocated “topic” is used only to indicate newly-introduced (or re-introduced) participants, while the grammaticalized/syntactic “subject” is used for continuing topics. Thus, in (12) above, the probability of John having been mentioned in the preceding five clauses is low, but the probability of the referent of she having been mentioned within the preceding one or two clauses is exceedingly high.14 In this sense, then, the syntacticized structure has yielded a more COMPLEX correlation between code and message. But from another point of view, an earlier 1: 1 correlation has been offset. Thus, in topic-oriented syntax,15 100% of topics are DEFINITE. But in many languages with grammatical, syntactic subject constructions, only 85–90% of subjects in discourse are definite, the rest indefinite.16 Thus, an “older” correlation between “topic” and “old information” is dis-established in syntactic subject constructions. Two interesting facts about the seeming less-then-perfect overlap between the purely functional category “topic” 142

and the syntacticized category “subject” should be pointed out here: (i) The more syntactic or “automated” mode of processing is reserved for the processing of CONTINUATION topics, i.e. those involving more predictability and less surprise/difficulty; and (ii) Here as elsewhere in syntax, it seems that the discrepancy between function and structure SELDOM EXCEEDS 20%. The figure of 20% discrepancy or 80% code fidelity must be born in mind, since it seems to recur in other areas of language.

5.3.2. Curves of diachronic change One of the most common types of diachronic change in syntax is that of REPLACEMENT. A construction coding a certain point within a functional domain is losing CODE-EFFICIENCY, most commonly due to the erosion of grammatical morphology because of phonological attrition. Another construction, similar enough in either function or structure or both, then gradually begins to pick up the functional load of the original one. The process of invasion and take-over by the new construction seems to proceed gradually on a straight-line time-curve until a certain point, most commonly around 75–80% replacement. But then it takes off EXPONENTIALLY, and replacement is completed in an amazingly short time, as compared to the gradual rate observed at the earlier stages. Thus, when one plots

143

the percentage of coverage of the function by the NEW construction against time, one typically finds a curve such as:17

Most typically, one finds ample intermediate points in the replacement process between 0% and 75–80% coverage, but it is hard to find any intermediate points between 75–80% and 100%. The most likely interpretation of this phenomenon is that somewhere around 75–80% of communicative fidelity of the new generalization, where variation is still a statistical fact for speakers, perceivers or language learners tend to judge the phenomenon as CATEGORIAL and proceed to behave in their output as if it is categorial. I think this points out to a general process of making categorial/rule judgments about data that is variable and lessthan-categorial. The point of 50% is chance variation, so far as a “rule” is concerned. Somewhere between that point and 100%, humans begin to assign categoriality even when the data are not yet categorial. It seems, further, that this is roughly the margin allowed in cognition and communication systems for an IMPERFECT CORRELATION between structure and function. It seems, further, that syntax as a SEMI-AUTOMATED processing system may indeed be interpreted within these bounds. It may sacrifice up to 20–25% code efficiency

144

(“communicative fidelity”) to gain speed in a highly ROUTINIZED system of speech processing. But syntactic and morphological systems on the wane, it seems, deteriorate exponentially when they fall below this point of code efficiency.

5.3.3. Irregular morphology Altogether too much fuss is made, it seems to me, over the study of decaying, decrepit, irregular and indeed bizarre tail-ends of inflectional/grammatical morphology and their acquisition by children. These sub-systems of the grammar have been preserved, rather artificially, by the advent of LITERACY and EDUCATION. Even the most cursory look at non-standard “country” dialects of German ought to convince one that the famous case system is long gone. And the same holds for the celebrated verb conjugation of French or the case system of Albanian or Russian. It is extremely hard to find such a high degree of code-irregularity in the morphology of languages spoken in pre-literate cultures. Children acquiring the language simply won’t stand for it. Such irregularity is zeroed out of the grammar within a couple of generations. While the role of literacy and education in retarding language change is indeed an interesting and valid topic for sociolinguistics, it seems to me that within the study of child language acquisition one ought to tone down the over-preoccupation with relatively insignificant minor processes in order to concentrate on the significant bulk. It simply is not logically necessary that we learn about regularity by contemplating extreme, bizarre irregularity. Rather, we learn about regularity by studying its normal

145

limits. And indeed we have a rough figure to go by, namely the region of 75–80% code-efficiency.

6. SYNTACTIC COMPLEXITY All other things being equal, structural complexity tends to accompany functional complexity in syntax. Of the three coding devices which make up syntactic structure—word order, morphology, and intonation—this iconic principle is easiest to demonstrate for grammatical morphology. So that, given the four major parameters of the simple/neutral clause—main, declarative, affirmative, active—it is in general likely that: (a) Embedded/subordinate clauses will have extra morphemes over the main clause; (b) Nondeclarative clauses will morphology over the declarative;18

exhibit

added

(c) Negatives tend to have an added morpheme over the affirmative; (d) Passive clauses/verbs morphology over the active.

tend

to

exhibit

extra

On the other hand, for word order and intonation in non-neutral clauses it is much harder to establish an ABSOLUTE measure of complexity. Rather, one must resort to a notion of RELATIVE complexity—on the background of the simple clause and its “unmarked,” “neutral,” “normal,” or “more frequent” processing 146

strategy. In this vein, one must also consider DELETION (“absence”) of a constituent found in the simple clause as an element of “complexity.” One must also consider tighter CONSTRAINTS (equi/coreference conditions) and more stringent DISTRIBUTIONAL RESTRICTIONS (in grammar as well as in text) as elements of syntactic complexity. But there are other considerations which contrive to make the subject of syntactic complexity even more arcane.

6.1. Functional complexity If code and message may be construed independently, as I have argued above, then one ought to define functional (“message”) complexity in syntax independently of structural complexity. This can indeed be done, by using a measure of DISCOURSE PRESUPPOSITIONALITY. Roughly speaking, one may assess the amount of presupposed material associated with a clause-type (“construction”) in a well-defined DISCOURSE CONTEXT.19 Alternatively, one may define the discourse-presuppositional complexity of various discourse contexts; then one may proceed to characterize various syntactic constructions according to the discourse contexts in which they normally tend to appear. There are, however, some grave conceptual and methodological problems involved in any attempt to define the discourse-presuppositional complexity of either syntactic constructions or discourse contexts in a non-circular fashion. This is so because when one invokes discouse-text as a relevant consideration in complexity, the amount/degree of complexity must now 147

be expressed PER DISCOURSE UNIT. But what is the relevant unit and how may one discover it in a principled way? Certainly, it could not be the “clause” or “construction,” since that would lead to utter circularity. One may, alternatively, resort to a unit such as “the intonational phase”: (14) INTONATIONAL PHRASE: “A sequence of speech produced under a single intonation contour without a break/pause.” But it is not clear how this will differ from “clause” or “construction” in a way that would bypass the danger of circularity. Another alternative could be to define the discourse unit as an informational unit, such as: (15) ARBITRARY INFORMATIONAL UNIT: “A sequence of speech containing an X number of predications/verbs.” But it is not clear that (15) is not circular either, in view of the strong correlation likely to be found between “verb” and “clause.” One may finally resort to defining TIME as the discourse unit and PREDICATION/VERBS as the information unit, then make the following generalization: (16) FUNCTIONAL COMPLEXITY OF DISCOURSE: “The functional complexity of discourse is expressed in terms of number of predications/verbs per unit of time.” The SYNTACTIC MODE (see 5.3.1. above) then will turn out to be more complex than the PRAGMATIC MODE of speech processing by virtue of packing more 148

information—presumably BOTH asserted and presupposed—into the same unit of time. And some syntactic constructions are more complex than others in terms of information/message because they simply pack in more information per time as compared to other, less-syntacticized constructions.

6.2. Structural complexity: more vs. different Having suggested at least a glimmer of possibility that functional complexity in discourse may be defined in some non-circular and independent way, one would next hope to demonstrate that structural complexity (a) may be defined in nonarbitrary fashion, and

some

independent

and

(b) indeed goes hand-in-hand with functional complexity. That is, that a certain measure of ICONICITY prevails in syntax. Perhaps the best way to demonstrate that this is indeed the case is by citing two simple examples. Relative clauses may be defined FUNCTIONALLY as a sentence modifying a noun for the purpose of helping the hearer identify uniquely the referent. The identification is achieved by using a sentence describing an event/state known to the hearer (“presupposed”), within which the noun to be identified figured as one of the participants (agent, patient, dative, etc.). There are, broadly speaking, two extreme ways of achieving this:

149

(17) PARATACTIC (PREPOSED TOPIC SENTENCE): (a) … you know, that man left our office last week, right? … (b) well, he got a much better job in … (18)

SYNTACTIC

(SUBORDINATE/EMBEDDED

TOPIC

SENTENCE):

… The man WHO LEFT OUR OFFICE LAST WEEK got a much better job in… The paratactic strategy in (17) spreads—presumably—the same information over a longer period of time, with (17a) corresponding to the embedded relative clause in (18). The STRUCTURE in (18) is more complex because of coreference constraints, the deletion of the subject-NP from the embedded clause and the relative-clause morphology (who). In addition, (18) is INFORMATIONALLY more complex, given our non-circular definition (16). But are (17) and (18) otherwise equal in terms of the total amount of information they convey? Not quite. To begin with, (17) ASSERTS both clauses, thus leaving both open to the normal process of TRUTH-NEGOTIATION accompanying asserted clauses.20 But the asserted (17a) is non-negotiable in (18), where it is PRESUPPOSED. While logically nothing has changed—with each clause taken atomically by itself, in terms of information processing (“communicative transaction”) (17) and (18) are not identical. Thus, the packing of “the same information”

150

into more complex syntactic constructions is not a trivial operation. The second example involves “logically-equivalent” causative expressions: (19) PARATACTIC:

three

(a) … John TALKED to Mary sweetly, (b) so she finally SLEPT with him…

(20) SYNTACTIC:

… John SWEET-TALKED Mary into SLEEPING with him…

(21) LEXICALIZED:

… John SEDUCED Mary…

The “same information” is presented as two separate propositions in (19), as a single proposition with two verbs—main and subordinate—in (20), and as a single proposition with a single verb in (21). But only the most language-blind logician would claim that the same information is transacted here. Proposition (19a) and (19b) are asserted independently, thus characterized as two separate— though causally related—events. But syntacticization into a single proposition in (20) already construes the information as A SINGLE EVENT, and this is coded on the subordinate verb by a nonfinite morphology.21 And in (21) this “condensation” of two propositions into one achieves its ultimate STRUCTURAL expression by the CO-LEXICALIZATION of the erstwhile two verbs into one. A parallel change is also observed in terms of the independence of the complement-verb (“result proposition”) SUBJECT/AGENT. In (19) Mary is construed as a SUBJECT/AGENT capable of making some independent judgment, and thusly coded in terms of case-marking. In (20) she is already marked as OBJECT, 151

presumably less capable of independent judgment. While in (21) she is construed as downright helpless. Per time-unit, (21) is more complex informationally than (20) and (20) more than (19). But information-per-time is not the only parameter along which the three renditions differ. Whether pragmatically or semantically, complex structures in syntax assume life of their own. They begin to “mean something else,” above and beyond mere processing complexity. When a child acquires a more complex syntacticized way of organizing the “same” information, it is thus likely that it is NOT quite the same information that gets re-packaged. Rather, the cultural-informational universe of the child undergoes considerable RE-ORGANIZATION. And as one may well suspect, the “same” unit of information is never really the same when construed in the context of a different organizational schema. Information—much like perception data and cognitive primes—has no independent life of its own outside its GENERIC and SPECIFIC contexts. From the generic-cultural context it receives its MEANING value. From the specific-discourse context it receives its MESSAGE value.

7. CULTURAL CONTEXT Most reports describing the linguistic input produced by educated white middle-class parents suggest a rich array of referentially-propositionally oriented language interaction during the early stages of child language acquisition. While the facts are genuine enough, one must bear in mind that other cultures may differ not only in the actual MANNER the parents or other adults may

152

structure their interaction with the child, but sometimes also in the GOALS of language acquisition and its end RESULTS. Let me illustrate this briefly with two examples. A Ute child receives little referential/propositional stimulation from parents or other adults. Like other Northern Plains cultures, the Utes frown upon child-initiated interaction between child and adult. A child is supposed to listen, not talk to adults. Even eye contact between child and adult is discouraged. Eventually, most linguistic and interactional skills are acquired due to peer-group input. And there are grounds for believing that the propositional-referential component of communication is not a major facet of the culturally-sanctioned adult repertoire. A person under the age of 40–50 is not encouraged, traditionally, to speak out in council. Only the oldest and wisest, the few who have become recognized orators, were traditionally expected to indulge in long deliberations. Even there, the very GOAL OF THE DELIBERATION IS PROFOUNDLY DIFFERENT FROM

WHAT

WE

ARE

ACCUSTOMED

CULTURES.

TO

IN

WESTERN

A SPEAKER MAY SPEAK ABOUT ANYTHING UNDER THE SUN, THE CONCEPT OF “SPEAKING TO THE POINT” IS FOREIGN TO THE CULTURE—ANYTHING may be the point. And the goal of the deliberation is NOT to convince your interlocutor by referential/propositional arguments. Rather, it is to create a SPIRITUAL CONSENSUS, to establish one’s spiritual bona fide, to demonstrate to one’s peers—by the use of EVOCATIVE language ALLUDING to shared spiritual experience and invoking commonly-appreciated METAPHORS—how you are indeed made of the same spiritual cloth. Such deliberations have 153

no formal end, no voting procedure and no competition for the floor. And the BIA-introduced Roberts’ Rules of Order are nowadays used within the same traditional framework. Traditionally, when a spiritual consensus is achieved, action is undertaken unanimously. When consensus fails, no general action can be enforced. Since the evocative use of language is oblique and subtle, it is only fitting that it would be acquired in equally oblique and subtle ways, indirectly, by passive observation and a slow process of subliminal generalization, rather than by direct demonstration. The mode indeed cannot be demonstrated directly; that would defeat the whole purpose. The second example involves the Spanish population of the Southwest. Here one finds a great amount of direct interaction between child and adult during language acquisition, and some of it must be referential/ propositional in nature. But again it seems that the bulk is not. The bulk, if one can generalize, seems to be directed at AFFECTIVE interaction. And one of the strongest interactive rituals practiced from an early age is a brand of merciless TEASING, conducted vigorously and ending only when the child is driven to utter frustration, tears, and tantrum. Whereby general mirth explodes and the teasing resumes with vengeance. The cultural goal of this enterprise, if one could venture a guess, is to produce an enduring STOICISM in the child, and a high threshold for frustration. The adaptive value of these traits need not be commented on. But again, when one compares language acquisition across cultural boundaries, one ought to keep in mind that different MANNERS of acquisition are not necessarily different 154

manners of acquiring the same end results. Rather, they may be different manners of acquiring strikingly different communicative MODES. And as may be often observed elsewhere, style is not always separable from substance, but may turn out on occasion to be substance itself.

REFERENCES Chafe, W. Meaning and the Structure of Language, Chicago: University of Chicago Press, 1970. Chomsky, N. Language and Mind. New York: Harcort, Brace and World, 1968. Duranti, A., & Ochs, E. Left-dislocation in spoken Italian. In T. Givón (Ed.), Discourse and syntax: Syntax and semantics Vol. 12. New York: Academic Press, 1979. Ferguson, C. Baby talk in six languages. American Antropologist, 1964, 66. Ferguson, C. Toward a characterization of English foreigner talk. In D. Hymes (Ed.), Pidginization and creolization of language. Cambridge: Cambridge University Press, 1975. Ferguson, C. Simplified register, broken language and Gastarbeiterdeutsch. In C. Mahoney et al. (Eds.), German in contact with other languages. Kronberg Lts: Scriptor Verlag, 1977. (a) Ferguson, C. Baby talk as a simplified register. In C. E. Snow & C. A. Ferguson (Eds.), Talking to children: 155

Language input and acquisition. Cambridge: Cambridge University Press, 1977. (b) García, E. Discourse without syntax. In T. Givón (Ed.), Discourse and syntax: Syntax and semantics Vol. 12. New York: Academic Press, 1979. Givón, T. The drift from VSO to SVO in Biblical Hebrew: The pragmatics of tense-aspect. In C. Li (Ed.), Mechanisms for syntactic change. Austin: University of Texas Press. 1977. Givón, T. On understanding grammar. New York: Academic Press, 1979. Givón, T. The binding hierarchy and the typology of complements. Studies in Language, 1980. (a) Givón, T. Ute reference grammar. Ignacio: Ute Press, 1980. (b) Givón, T. Typology and functional domains, Studies in Language, 1981. Karttunen, L. Presupposition and linguistic context. Theoretical Linguistics, 1974, 1.2. Li, C., & Thompson, S. Subject and topic: A new typology of language. In C. Li (Ed.), Subject and topic. New York: Academic Press, 1976. (a) Li, C, & Thompson, S. Development of the causative in Mandarin Chinese: Interaction of diachronic processes in

156

syntax. In M. Shibantani (Ed.), The grammar of causative constructions: Syntax and semantics Vol. 6. New York: Academic Press, 1976. (b) Long, M. Input, interaction and second language acquisition. Unpublished doctoral dissertation, University of California, Los Angeles, 1980. Wittgenstein, L. Tractatus logico philosophicus [1918], translated by D. F. Pears & B. F. McGuinness. New York: The Humanities Press, 1960. *I have benefited from helpful comments from Dan Slobin and Elinor Ochs, but do not wish to hold them responsible for the final product. 1

The extremes are of course analogs to Wittgenstein’s (1918) TAUTOLOGY and CONTRADICTION, the first characterizing redundant information, the second un-integrable information (given the prevailing assumptions of the system). For further discussion of the pragmatics of knowledge, see Givon (1979, Ch. 8). 2

See Ferguson (1975, 1977a) and Long (1980).

3

See Ferguson (1964, 1977b). Adults may of course choose not to tone down, and thus take the penalty of unsuccessful communication. 4

See Givón (1979, Ch. 2).

157

5

In some languages a SECONDARY pragmatic role is also recognized syntactically, that of DIRECT-OBJECT. For discussion see Givón (1979, Ch. 4). 6

See Chafe (1976).

7

See Givón (1980a).

8

See discussion in Givón (1979, Ch. 3).

9

The most common inventory of major functional domains coded by syntactic structure is the following (in addition to the four enumerated above): tense-aspect-modality, focusing/contrasting constructions, topic-identifying devices, sentence-conjoining and phrase-conjoining devices, de-transitivizing constructions, causative constructions, noun-modifying constructions, possession expressions, existential and other topic-introducting constructions, comparatives and superlatives, adverbial clauses. For a language-specific example, see Givon (1980b). 10

The details of this particular example may be found in Givón ( 1981). 11

For further detail within a single language, see Givón (1980b). The scale is hierarchic and IMPLICATIONAL in the following sense: A language may or may not have a particular coding point A. Buf if A and B are in a particular order on the universal hierarchy in (5) and the language has both, they are going to code functional points on the hierarchy in the same relative order as in (5). 158

12

Chomsky’s approach to both language structure and language acquisition is a represenative of one extreme. The other is elucidated most strongly by Erica García (1979). See discussion in Givón (1979, Ch. 5). 13

Givón (1979, Ch. 5).

14

See details in Duranti and Ochs (1979) as well as Givón (1977). 15

See Li and Thompson (1976a) concerning so-called “topic-prominent” languages. 16

See text counts from English in Givón (1979, Ch. 2).

17

See for example Li and Thompson (1976b) and Givón (1977). 18

This is more likely to be true for interrogatives than for imperatives. In general, the imperative tends to be a more SIMPLE form of the verb, with its “markedness/ complexity” derived from distributional restrictions and the obligatory deletion of the subject/agent. 19

See discussion in Givón (1979, Ch. 2).

20

Asserted clauses are open to denial/challenge. But discourse proceeds incrementally, so that an assertion that is not challenged right away then tacitly is assumed to join the presupposed background when the next assertion is delivered. For a formalism illustrating this, see Karttunen (1974), and for discussion see Givon (1979, Ch. 2,3).

159

21

For the iconicity principles governing the structure of verb complements, see Givón (1980a). Further, while “sweet talk” in (20) is IMPLICATIVE “talked … sweetly” in (19) clearly is not.

160

13

Language Segmentation: Operating Principles for the Perception and Analysis of Language

Ann M. Peters University of Hawaii

Contents Introduction Specific OPs for Extraction and Segmentation Extraction of Early Units Segmentation: Analysis of Acquired Units into Smaller Ones

161

Segmentation of Morpho-Syntactic Frames OPs for Feedback Related Issues Methodological Questions Implications for a Theory of Second Language Learning Research Questions Relevant to Segmentation Heuristics Influence of Segmentation

Language

Structure

on

Extraction

and

Evidence of Malsegmentation Predisposing Factors in the Input

1. INTRODUCTION Language acquisition studies have traditionally focused on such questions as how children go about acquiring certain syntactic structures, or certain semantic (sub)systems, or the system of phonological contrasts in the language they are learning. The evidence on which such studies are based comes from the developing sophistication of utterances children produce. But before a child can produce anything language-like she must have in mind some linguistic target, however rudimentary. In other words, she must have perceived certain pieces of the language she has heard around her

162

since birth which seem relevant to her to try to reproduce. What is this language like? Speech directed to children by middle-class English-speaking parents has been rather carefully studied in recent years (see e.g. a number of the studies in Snow & Ferguson, 1977), and has been variously termed “input speech” or “motherese.” Since, however, it is now becoming evident that in some cultures a child’s primary caretaker (and her primary source of language input) is not her mother (see Ochs & Schieffelin, in press), I will refer to such speech as “caretaker speech.” It is also becoming increasingly clear that not all the language a child hears is of the highly “pre-digested” variety focused on by recent studies of caretaker speech (again, see Snow & Ferguson, 1977)—a variety which might give a child a headstart into the language system by breaking utterances down into easily absorbed bits. But not only does the tendency for caretakers to produce such speech vary from culture to culture (Heath, 1979; Ochs & Schieffelin, in press), it also seems that caretaker speech is much less predigested before children begin to actually produce “words” of their own (Phillips, 1973). Therefore, let us assume that input speech consists, on the whole, of utterances several words in length. We are thus confronted with a perceptual question: how does a child accomplish this necessary first (perceptual) analysis of language? What are the Operating Principles (in Slobin’s sense) that she may discover or otherwise have available for use in solving the “initial Extraction problem”? By Extraction I mean the process of 163

recognizing and remembering recurring chunks of speech out of the continuous speech streams present in the environment. These chunks will provide the material for the child to use in bootstrapping her way into the language system.1 Thus, Extraction is a very rudimentary process, at the level of learning first words, and must be guided by principles that are almost purely phonological, as I shall show. It seems clear that children do not Extract everything they hear said around and to them. Rather they pick and choose what to pay attention to and what to try to remember, and these choices will be affected by factors involving both linguistic and psycholinguistic salience: factors which should be reflected in the Operating Principles (OPs) that guide Extraction. Furthermore, once a child has Extracted some chunks of language, what happens to them? Do they merely lie passively in her lexicon? If these chunks are morphologically complex in the adult language they may contain much information that would be useful to the language learner if she is capable of doing the requisite analysis. How might she go about doing such analysis? She would need to Segment her memorized chunks into smaller linguistic units and take note of such rudimentary syntactic patterns as she might discover in this Segmentation process. Note that I am using the term Segmentation specifically to mean analysis of Extracted chunks of speech, both with respect to discovering the linguistic sub-units of which these chunks are composed, and with respect to becoming

164

aware of the rudimentary syntactic information contained within them. Again, what heuristics are needed to accomplish these tasks? We also need to consider how much of early language acquisition, up to and including the most rudimentary syntax, can reasonably be accounted for by a small set of simple OPs. This leads us to the question of how much of language development is determined by innate abilities and how much is achieved by learning based on those innate abilities. My philosophical preference has been to see how much linguistic knowledge can be accounted for with OPs that are as few and simple as possible even though these OPs may result in apparently inefficient trials and errors. There is no reason to believe that language acquisition is efficient, and in fact we have no metric for determining efficiency because we do not know the mechanisms that the child comes equipped with. We do, on the other hand, have evidence of trials and errors, the nature of which do provide some insight into these mechanisms. In trying to infer a set of Operating Principles that would serve children in perceiving how the language system is put together, i.e. in Extracting, Segmenting, and discovering at least the most rudimentary syntactic patterns inherent in heard speech, we are necessarily on somewhat shaky ground since the evidence is indirect. This is due to the fact that we have only her productions from which to infer what a child has perceived. But there does exist such evidence in utterances that are unexpectedly long (under-Segmented) or unexpectedly short (over-Segmented) or with word boundaries in the 165

wrong places (mis-Segmented), as I have shown in Peters (1983). And the evidence is at least sufficient for us to start to put together a theory of how children must proceed in this universally necessary process. A theoretical question that must be raised here is: how are these initial Extraction and Segmentation heuristics related to those needed for the rest of language acquisition? Is the solution of the Extraction problem merely a matter of isolating the first few “words” and then letting other processes take over— processes which lead to discovering the phonological, morphological, syntactic, and semantic systems of the target language? Or is there a continuously needed Segmentation component to language acquisition which is somehow separate from these other components? Might it be the case that the heuristics involved in the Extraction and Segmentation of linguistic units are also intimately bound up with the perception of morphological, syntactic, and even phonological patterns? In proposing a subset of Operating Principles which seem to be important for Extraction and Segmentation, and in considering the available evidence, I hope to show that this last is indeed the case: that although at the beginning of language acquisition Extraction and then Segmentation are the major components of learning about language, while they diminish in importance as other kinds of linguistic knowledge are acquired, they never disappear entirely, but are alwava available for dealing with newly Extracted chunks of language whose structure is not immediately transparent. I also hope to show that, as language learning proceeds, Segmentation heuristics become increasingly sophisticated. Thus 166

although at first they are extremely crude, as knowledge about the properties of the target language develops this growing knowledge can be used to refine Segmentation procedures, allowing new Operating Principles to come into play.

2. SPECIFIC OPs FOR EXTRACTION AND SEGMENTATION In starting to learn about the system that constitutes language, the beginning learner cannot learn about the whole system at once—she has to start somewhere, with “fuzzy” first approximations that have to be successively adjusted and refined as further information and feedback are acquired. Having to start with a limited piece of the whole system is the nature of any bootstrapping process. How does such “fuzziness” manifest itself in language acquisition? We can draw an analogy with the process of learning to see a complex object through a lens which is movable and which has adjustable focus. Here we have to learn to do three things: (1) decide what is relevant to look at (focus on), including how much to try to look at at once; (2) figure out how to adjust the focus so that the relevant portion is sharp; and (3) discover how to make all the parts fuse into a coherent picture. Of course, because of the influence of various environmental and cognitive factors, even at the earliest stages of such a learning process different paths of progress are possible in terms of what is focused on first, next, etc.

167

In language acquisition, relevant factors influencing how much to focus on at once include how predigested (i.e. presegmented) caretaker speech is (i.e. whether long or short chunks are easily or naturally available), and the learner’s “natural length of focus” (i..e her natural tolerance/preference for long vs. short chunks of speech).2 Learning which details to focus on is also affected by input speech, especially by what purposes commonly-heard language is used for (e.g. naming objects vs. social interaction), and by the occurrence of specific predictably varying routines which would lead to the perception of specific slot-and-frame type sentence patterns. In section 3 I discuss these factors in somewhat more detail. The above considerations have led me to sort out two pairs of oppositions which I will utilize in presenting the subset of OPs that seem to be involved in the early stages of language perception: (1) Extraction of early units from speech streams vs. later Segmentation of these units into their subcomponents, which is comparable to gradual adjusting of focus so that more details become clear; and (2) perception of “slots” (i.e. content words) vs. “frames” (i.e. morpho-syntactic patterns), which corresponds to learning what one is focusing on and how individual elements are interrelated and form a coherent system (picture). The OPs will therefore be presented in three groups: those dealing with Extraction, those dealing with Segmentation, and those having to do with discovery of morpho-syntactic frames.3

168

2.1. Extraction of Early Units The child’s first problem is to get hold of something to work with linguistically. The basic OP to be applied at this state is: EX:EXTRACT. Extract whatever salient chunks of speech you can. Note that at the earliest stages of language acquisition, the question of whether these Extracted chunks are ultimately divisible has not arisen for the child. For the purposes of this discussion, then, we will refer to these chunks as “units.” Nevertheless, the morphological length (as measured in the adult language) of such early units may vary from child to child, depending on such variables as how long words are in the language the child is learning (compare Chinese with Eskimo!), as well as how thoroughly presegmented caretaker speech happens to be. It must also be remembered that, because of limitations on the child’s ability to produce speech, early “perceived units” may be much longer than early “produced units.” Even in English, however, there is evidence that certain oft-recurring phrases are perceived as single units, e.g. look-at-that, open-the-door, what’s-that, I-still-have-some (Peters, 1977, and personal observation). The child also needs two very basic OPs for recognizing whether newly Extracted chunks are the same as or different from ones previously Extracted and for storing the different ones.

169

EX:COMPARE. Determine whether a newly extracted chunk of speech seems to be the same as or different from anything you have already stored. EX:STORE. If it is different then store it separately; if it is the same, take note of this sameness but do not store it separately. We would expect there to be some sort of effect of frequency on Extraction: items that occur more frequently in the environment will, ceteris paribus, get processed more often and thus be absorbed into the child’s growing language system more quickly. We would thus expect that the more frequently a particular chunk of speech is Extracted, the stronger its long-term memory trace will become. Conversely, lack of recurrence of a chunk in the environment will contribute to (though not necessarily lead to) the decay of its memory trace.

2.1.1. Ancillary OPs Based on Salience These first three OPs, while the most basic in one way, cannot operate in the absence of a second set of OPs which must guide the child as to what sorts of stretches of speech are “salient,” i.e. are reasonable candidates for Extraction. This second set of OPs, then, is determined by early limitations on how much information the child is able to pay attention to at once, and by a set of saliency factors, most of which are phonological, which direct the child’s attention to which sorts of information to notice first. These factors focus the child’s attention on selected aspects of input speech, and influence WHICH

170

early utterances are Extracted and HOW LARGE these units are (in terms of the number of morphemes in the adult language they comprise). It must be borne in mind that while salience is to some extent determined by typological characteristics of the language being learned, it is also affected by attentional characteristics of the individual learner (see discussion in 3.1.3). Note, also, that while each of the following OPs contributes to the probability that a particular chunk of speech will be extracted and remembered, no ONE of them MUST apply. One of the potentially most salient features of an utterance in an unknown language is the property of having a clearly identifiable meaning, i.e. of being unambiguously and consistently associated with some salient feature of the environment (thing, situation, action, etc.) EX:MEANING. Pay attention to utterances that have a readily identifiable meaning. Extract and remember sound sequences that have a clear connection to a clear context. The other OPs in this group reflect phonological properties that may lead the child to perceive some part of a speech stream as salient and therefore Extractable as a unit or chunk: EX:SILENCE. An extractable unit is bounded by silence (i.e., it is a whole utterance). EX:SUPRASEG. An extractable unit suprasegmentally delimited stretch of speech. 171

is

a

EX:TUNE. An extractable unit is a speech tune or melody. EX:RHYTHM. An extractable unit is a rhythmic pattern of speech. Such a set of OPs would lead us to expect to find children at fairly early stages of language acquisition extracting and trying to use as linguistic units any of the following: whole sentences, whole phrases, intonation contours or “tunes,” and rhythm or stress patterns. Do they in fact do this? Since such phenomena have rarely been explicitly looked for there is not a great deal of evidence as yet, but a number of suggestive examples do exist. Some evidence for the use of EX:SILENCE comes from a child named Minh whom I studied (Peters, 1977). By 14 months-of-age, Minh had extracted and used as units a number of whole sentences, including Open the door and Look at that. The following example from Clark (1977) shows the use of EX:SUPRASEG as a delimiter of a piece to be Extracted, at least if we can assume that the original adult utterance was produced with two intonation contours: Adult: That’s an elephant, isn’t it? What is it? (A minute or two later) Adam: Intit.

(Adam continued to call elephants intits for several weeks and was impervious to corrections.)

172

While EX:SUPRASEG uses intonational information to determine the boundaries of extractable chunks, EX:TUNE focuses on the tune or melody contained within the chunk. An example of the use of this latter OP can be seen in the strategy I have elsewhere called “learning the tune before the words” (Peters, 1974), where the tune of the particular phrase seems to have been more salient, and hence more memorable, than its segmental phonemes. English phrases which have such characteristic tunes include uh-oh!, look at that!, what’s that?. In tone languages tonal melodies can also be separately Extracted. For instance, in a study of the acquisition of SiSwati, Kunene (1979, pp. 255–256) observed that her young subject had perceived the tone pattern of the copulative construction before she had discovered the proper segmental phonemes to go with it. Instead, she used a “carrier” syllable, /à/( = low tone): TARGET

sì-tfòdmbè

PRODUCED

GLOSS

à-tfòdmbè ‘It’s a picture.’

ngù-m-lùmbì à-lùmbì

‘It’s a whiteman.’

After practicing sentences in Ndebele, I discovered that I would sometimes come away humming the tonal melody of a sentence, having forgotten the phonemes that went with it. EX:RHYTHM claims that rhythmic patterning is one of the saliency factors affecting which early units are 173

focused on at the extraction stage. Furthermore, if EX:RHYTHM operates analogously to EX:TUNE, it would predict that salient rhythmic patterns could be extracted just like intonation contours or tonal melodies. While this might seem improbable on first consideration, there is some evidence from “tip of the tongue” states (Brown & McNeill, 1966, p. 330) that such patterns may indeed be stored and accessed separately from segmental phonological information. Thus it is often possible to remember the number of syllables in an otherwise inaccessible word, and even its stress pattern. There is also evidence from early language production for the importance of a rhythm-based OP. It is manifested in the appearance of “filler syllables,” which seem to be incompletely analyzed syllables which are nevertheless reproduced in a “fuzzy” form in order to maintain the rhythmic envelope of a particular phrase. Such syllables have been observed in the early productions of children learning English (Bloom, 1970; Braine, 1976; Peters, 1977), Spanish and Cakchiquel (Tolbert, 1978), Turkish (Ekmeçi, 1979; Aksu-koç & Slobin, 1985), SiSwati (Kunene, 1979), and German (Stern & Stern, 1907). It is well to remember that all of these “salience OPs” interact with and mutually reinforce eath other. They are also affected by frequency. Thus an an utterance which is at once frequent, often bounded by silence, and possessed of characteristic rhythm and intonation contours is likely to be much more memorable (and thus earlier perceived) than utterance distinguished by only one of these characteristics. 174

2.2. Segmentation: Analysis of Acquired Units into Smaller Ones After a number of early units have been Extracted and memorized, knowledge about these already acquired units can be utilized in further analysis. Returning to our earlier analogy of learning to see through an adjustable lens, we can say that the child is now ready to learn to sharpen the focus so that more details can be perceived, analyzed, and incorporated into the system. She must also learn to shift the focus to different points in an utterance (or an already memorized “word”) so that areas which were at first ignored as being irrelevant (and thus fuzzy) background may suddenly seem interesting and important and worthy of attention and analysis. This can be seen as a sort of figure-ground phenomenon. For example, suppose a sentence such as Open the door has been Extracted as a unit in which no sub-part has been noticed as more prominent than any other, i.e., it is a “figure” in its own right. But in trying to pronounce it more carefully some syllables (e.g. door) may be separately focused on, i.e., the original figure now becomes a “ground” within which a new figure emerges. Another way for such internal figures to emerge is through comparison of utterances; for instance, if Shut the door is also Extracted, it might lead to awareness of those parts of these two utterances which are similar and those which differ. Since these shifts in focus are involved in a process of accumulating knowledge about different aspects of a complex system, a detail once focused on is not forgotten. Thus everything that has had a turn at sharp focus never fades back into the blur of

175

unfocus—the details gradually emerge and stay there, contributing to the growing picture of the larger system as a whole. At this stage of Segmentation the basic OP is: SG:SEGMENT. Attempt to Segment utterances you hear or utterances you have already Extracted into smaller linguistic units. This OP by itself is not, however, sufficient for a child to be able to carry out the necessary Segmentations. She needs some heuristics for finding likely points at which to make cuts. Such heuristics can be based on two types of information: that contained within a particular unit, and that which becomes available in the process of comparing two similar but not identical units. At the initial stages of learning to do Segmentation, I am assuming that the relevant OPs will be based exclusively on phonological information, although I expect that in the process of actually doing Segmentation the child will begin to become aware of distributional information that will eventually lead to a more syntactic type of knowledge that can itself be brought to bear on Segmentation. In order to simplify the presentation of the OPs, however, I defer discussion of distributional analysis to section 2.3.

2.2.1. Internal Cues for Segmentation Internal phonological cues play a role in Segmentation in that they draw attention to likely Segmentation points (SG:RHYTHM, SG:INTONATION) or to likely

176

subunits to Segment (SG:END, SG:BEGIN, SG:STRESS). In particular it seems that these cues draw attention to subunits of single-syllable size. This observation leads to a whole area of phonological perception and analysis that still needs to be worked out. I suspect that on the phonological level the child moves from whole word (i.e. Extracted unit) to syllable to phoneme. Perhaps the kind of Segmentations into syllables which I am proposing here happens when the child’s preferred working level is the syllable. Certainly the syllable is a convenient intermediate size of chunk to work with since it is a minimal pronounceable unit. There is probably a whole parallel process of phonological segmentation which eventually leads to the acquisition of the phonological system of a language just as the Segmentation processes which we are mainly concerned with in this chapter lead to the acquisition of lexicon, morphology, and syntax. I am not prepared to develop the process of phonological segmentation or to speculate on how these two processes are interrelated (although they are probably the same process at the earliest stages). Nevertheless, phonological, or more properly acoustic, cues are the only available ones at the very early stages of Segmentation. On the phonological level I will therefore assume that by the time the child reaches the Segmentation stage she is at least capable of dividing an Extracted unit into a sequence of syllables. Eventually children will have to be able to Segment out units (morphemes) that do not conveniently divide into syllables: they may be sub-syllabic (e.g. 177

English plural or past tense), or they may consist of pieces of more than one syllable (e.g. VC in an open-syllable language). A discussion of how children manage such Segmentations will have to await a more thorough exploration of the phonological side of the process. It should be kept in mind that at this early stage, the syllables being Segmented out are being treated as potential lexical items. It is only later, when sufficient distributional information has been collected, that they will become candidates for grammatical functors, i.e. affixes rather than content words. Morphosyntactic interpretation of such syllables will be discussed in 2.3. As with Extraction, certain saliency factors will also affect the Segmentation of larger units into smaller ones. At the Extraction stage, however, it seems that saliency factors will mutually reinforce each other. That is, if more than one is relevant they will combine to increase the salience of a particular chunk of speech and thus increase its chances of being Extracted. But at the Segmentation, or further analysis stage, although saliency factors may mutually reinforce each other (again enhancing the probability of a cut being made at a particular location), they may also work at cross purposes, scattering different kinds of Segmentation cues across an utterance. These different kinds of interaction possibilities should be kept in mind. It is probably also the case that in learning a particular language children gradually discover that some kinds of cues are more dependable than others in that they more consistently

178

highlight reusable subunits. Thus, some OPs will be reinforced while others will gradually fade away. Saliency factors at this stage are reflected in the following OPs: SG:INTONATION.* Segment intonationally salient places.

utterances

at

SG:RHYTHM.* Segment utterances at rhythmically salient places. SG:STRESS. Segment off a stressed syllable of an Extracted unit and store it separately. SG:END.* Segment off the last syllable of an Extracted unit and store it separately. SG:BEGIN.* Segment off the first syllable of an Extracted unit and store it separately. SG:REPETITION. Segment off subunits that are repeated (in terms of segmentals or rhythm or intonation) within an Extracted unit and store them separately. I do not have direct evidence of the operation of SG:RHYTHM and SG:INTONATION other than examples such as the intit one quoted above. And yet it seems reasonable that rhythm and intonation should play a role in determining Segmentation points. This would be an area in which psycholinguistic experiments with older subjects might provide some evidence. For

179

instance, one could give recognition tasks involving strings of nonsense syllables with clear rhythmic and/or intonational patterns. If the strings were too long to remember whole, perhaps suprasegmentally delimited subparts would be better recognized. The other OPs (SG:STRESS, SG:END, SG:BEGIN, SG:REPETITION) make a set of purely mechanical (i.e. phonological) predictions, namely that children will Segment and store as separate units single syllables with particularly salient properties. Is this plausible? I think it is quite probable as a first step toward language learning. What is the evidence? I do not think it has ever been looked for as such. It could turn out to be the case that some of children’s early utterances are incomprehensible to adults because they involve units Segmented according to such purely phonological hypotheses. This is another area in which psycholinguistic experimentation with miniature artificial languages (MALs) could yield some evidence, at least for older subjects. Certainly children will eventually have to discover that operating on such a purely mechanical basis is not a very efficient strategy, and they will have to learn to pay attention to other kinds of information (semantic and syntactic) along with the phonological. I am assuming that these sorts of discoveries will be guided by OPs for evaluation of Segmentations (see below). Pye (1980, 1981) has shown for children learning Quiche Mayan that stress is a particularly important perceptual determinant of which parts of morphologically complex words are acquired first. 180

SG:STRESS works particularly well in languages where stress is fixed with respect to word boundaries. In such languages children rarely make Segmentation errors which cross word boundaries, e.g. in Hungarian and Finnish which have word-initial stress. (See Peters, 1981, for a discussion.) The syllables at the ends (SG:END) and beginnings (SG:BEGIN) of utterances have particular phonological salience since they are adjacent to silence (EX:SILENCE). The ability to remember such syllables may be enhanced by the tendency for items at the end and beginning of a series (especially at the end) to be remembered better than items located in the middle. (See Kintsch, 1977, for a review of research on serial recall by adults, and Hagen & Stanovich, 1977, for work with children.) At noted above, two OPs operating in conjunction would be expected to be more salient, and hence more predictive of potential Segmentations, than any one acting alone. Thus, syllables that are simultaneously stressed and word- or utterance-final would be more salient than syllables that had only one of these characteristics. Many children learning English do opt to try to produce only a single syllable of a multisyllable target, and it is always stressed and often initial or final. E.g. (from Velten, 1971, pp. 86–97): TARGET PRODUCED coffee

daf

faucet

fas

181

towel

daw

high chair

hats

Weir observed that at about 2½ year her son Anthony over-Segmented the two words whistle and measure: “… the child owned a whistle and apparently did not like to use the same phonetic sequence for noun and verb, and the action of whistling he termed [wIs]. The same apocopation was performed on measure where the verb became [mEž]” (1962, p. 74). In both these cases Anthony seems to have segmented off the stressed syllables (which also happen to be the initial syllables of their respective adult words. Whether he was also invoking a noun:verb analogy such as helper.help = measure.mezh is hard to say. In his study of the acquisition of Quiche, Pye (1980, 1981) makes a good case for the perceptual salience of stressed word-final syllables. Note, however, that while SG:STRESS and SG:END would reinforce each other in a language with word-final stress, stress and recency would conflict in a language with word-initial stress. In such a case the child would have to discover that primacy was more crucial than recency—a discovery that would be aided by the regularity of initial stress. SG:REPETITION is based on the fact that repetition of elements can increase their salience enough to overcome natural loss of accessibility due to memory overload. Certainly even quite young children are sensitive to alliteration and rhyme, as witnessed by their language play (Weeks, 1979; Weir, 1962). Languages where this 182

OP would be useful are those in which gender or number agreement is marked with highly frequent, phonologically repetitive, morphemes which consist of full syllables. For instance, in Hebrew, plurality is marked on both noun and modifying adjective by suffixation of -im (MASC) or -ot (FEM). Thus, a plausible Extracted unit such as yeladot tovot ‘good girls’ would contain the repeated sub-unit -ot. Similarly, in Bantu languages gender-class agreement is marked on both noun and modifying adjective (as well as on the verb if the noun in question is the subject of the verb) by phonologically repetitive prefixes. Thus, in SiSwati, ‘The cat is eating’ would be (from Kunene, 1979, p. 247; PR = prefix, TN = tense): li-kati li-ya-dla PR-cat PR-TN-eat

It seems plausible that the phonological salience of such repetition would aid the child when she was ready to focus on such agreement morphemes (i.e., when it was their turn to emerge as figures out of the ground of Extracted chunks such as yeladot tovot or likati liyadla.) On the other hand, since semantic repetition within an utterance is uncommon (except for emphasis), the occurrence of phonological repetition might at first cause the child to IGNORE this information as somehow redundant or less relevant (back)ground and to focus first on the nonrepetitive parts. Evidence for this comes from one child’s learning of SiSwati where she first produced bare (unprefixed) stems and only later added prefixes (Kunene, 1979, p. 70). 183

2.2.2. External Cues for Segmentation Another kind of phonological salience that may lead to Segmentation results from phonological similarity between two or more Extracted units. This suggests a comparative heuristic, which involves a process of “factoring” out matching parts. This heuristic is here stated in two parts, the second being a generalization of the first: SG:MATCH1. When the initial or final portion of an Extracted unit phonologically matches the whole of another Extracted unit, the remainder of the larger unit is a candidate for storage as a new unit. SG:MATCH2. More generally, when two Extracted units share ANY phonologically similar portion, the shared portion can be Segmented and stored as a unit, and so can the residue. SG:MATCH1 is a relatively simple heuristic which Segments an utterance into exactly two pieces, one of which is known while the other may not be known. As children’s linguistic knowledge and processing capacity grow they will be able to handle Segmentations into more than two units. This generalization is considered in more detail below. For now let us confine our discussion to two-unit Segmentations. When making phonological comparisons such as those described in the MATCH OPs, the material to be compared may be drawn from two sources: previously Extracted and memorized speech (units in long-term

184

memory) or more recently heard speech (units in short-term memory). Any combination of sources is presumed to be possible: either the shorter unit or the longer one in the match can be drawn from either long-term or short-term memory. Note the likelihood that a truly invariant form will be consistently perceived as a unit, whereas a form that has one or more points of variation (substitution slots) will be much more amenable to Segmentation. The effects of applying a principle such as SG:MATCH1 can easily be seen in the frequently attested tendency of children learning English to over-Segment the word behave. Assuming that admonitions such as I want you to be good and I want you to behave are frequently heard, such a factoring principle could lead children to say such things as I’m going to be very very /heyv/.4 The existence of sub-parts of utterances which can occur in different orders can also lead to matching and Segmentation. For instance, the realization that put your shoes on and put on your shoes can both occur would prevent both shoes-on and put-on from being fossilized as units.5 It may be objected that SG:MATCH1 would lead to too much analysis. I think that, in fact, children DO do much more overanalysis than we are aware of—that the few instances we notice are only the tip of a much larger iceberg. Moreover, this overanalysis may sometimes seem quite semantically unmotivated to us— which adds evidence in support of the hypothesis that, at least at first, such analysis may proceed on rather narrow phonological grounds. To illustrate: when Adrian, a 185

2-year-old friend of mine, first encountered cappucino ice cream he was repeatedly coaxed to say cappucino. But he refused, turning away and saying Not “no”. Eventually his mother realized that Adrian had perceived the -no at the end of cappucino as some sort of negation (i.e. a match with No!) and that he was afraid that if he said Cappuci, no he wouldn’t get any!6 On the other hand, the MATCH OPs may be constrained in various ways in order to prevent so much overanalysis. An early phonological constraint could be based on syllabicity, e.g.: Do not consider matches that result in units that do not contain integral numbers of syllables. This would prevent a child at the dinner table from trying to match up pepper mill and milk and Segmenting the latter into mill+k.7 Such a constraint could only operate for a limited time, however, since a child will eventually need to be able to match mill with mills and Segment the latter into mill+s. My preference is to assume that if there is indeed an early syllabicity constraint it is dictated by the child’s progress on the phonological front, i.e., that it is in effect because she has only progressed to being able to handle syllable-sized distinctions, and that this constraint will lift once she learns how to make distinctions at the phonemic level. Examples of children performing Segmentations that respect syllabic boundaries but violate morphological boundaries can be found in Pye’s study of Quiche (1980). For instance, the word kawiloh ‘you see it’ can be analyzed as follows (‘ = stress):

186

Morphologically: k aw il’oh Syllabically: ka wi l’oh It was Extracted as: wil’oh splitting the morpheme aw. There is, however, another way to handle too much overanalysis other than by assuming the existence of constraints upon the Segmentation process itself. This is to assume that there are mechanisms for the perception and correction of erroneous Segmentations—mechanisms which I have grouped together as OPs for feedback and which will be discussed in 2.4.

2.3. Segmentation of Morpho-Syntactic Frames8 So far we have primarily been looking at cognitive processes (OPs) involved in perception of phonologically, and to some extent, semantically salient units. But merely Segmenting out content words does not constitute acquisition of a language. It is also necessary to figure out the grammatical system (the linguistic “glue”) that holds the content words together and allows meanings to be combined in more subtle ways than simple juxtaposition would permit. In order to learn about these grammatical aspects of language it is necessary to shift one’s focus away from the content words and onto their (morpho-syntactic) environments. This approach is used in the technique of foreign language teaching with “pattern practice” and “substitution drills” where a sentence frame is provided and at a given place in the sentence (a “slot”) a set of 187

substitutions is made. Sometimes the focus is on what sorts of content words go in a particular slot, sometimes it is on the shape of the surrounding grammatical frame. Such a shift from slots to frames must also happen, although unconsciously, in first language acquisition. For example, certain functors (e.g. inflections at the ends of words) may start out as being part of an uninteresting ground during the process of root/stem identification. These will eventually, however, have a turn at being figures as their place in the system is worked out. Once this has been accomplished, some other set of functors will have a turn. Learning to shift focus in this way entails dealing with a different sort of figure-ground problem than that involved in perceiving more and more detail: sometimes the focus needs to be on identifying content words and sometimes it needs to be on perceiving the frames that surround them. I am assuming that awareness of complex structural patterns will not emerge all at once. Rather, there must be an iterative process in which at first a single slot will manifest itself in an otherwise unpatterned (unsegmented) block, then within that block another substitution point will be found, etc. Such repeated finer and finer analysis of long chunks of language has been well documented for children acquiring second languages. The following example, from Wong Fillmore (1979) shows three stages in such an iterative analysis by a 5-year-old Spanish-speaking girl learning English as her second language.

188

1. After about 2 months of exposure to English her English productions included the unanalyzed unit: How-do-you-do-dese? 2. After 5 months of learning and breaking down English sentences she had extracted enough underlying structure from sentences containing this chunk in order to have evolved the following productive sentence pattern: How-do-you-do-dese where X could be a noun phrase or a +X prepositional phrase.

E.g. How-do-you-do-dese? How-do-you-do-dese flower power? How-do-you-do-dese in English?

She then used the language she heard or knew in order to Segment do-dese off of how-do-you, and used the latter as the base for another sentence pattern: How-do-you How-did-you

}

+ X where X could be a verb phrase.

E.g. How-do-you make the flower? How-did-you make it?

3. After 7 months of English she had Segmented you off, and had the following sentence pattern:

189

How-do How-does How-did

}

X where X could be a whole clause.

E.g. How-do cut it? How-does this color is? How-did dese work? (= ‘How does this work?’)

Sometime later how was also Segmented off and used in constructing questions. E.g. How you make it? How will take off paste?

What OPs will be needed in achieving such analysis? So far in our consideration of OPs needed for Segmentation, a “word” has been any unit the child has been able to Segment from language she has heard or language she has memorized. For the move into grammar, however, she needs ways to distinguish content from grammatical units. In particular she needs to be able to recognize possible substitution points, i.e. to distinguish frames from slots. FR:FRAME. If two (or more) units, after Segmentation by any of the SG: heuristics, appear to share a common sub-unit, A, followed or preceded by alternative sub-units, B or C, etc., take note of this fact, namely that there is a pattern in which A can be followed (or preceded) by either B or C, etc. FR:SLOT. When you have a list of items, e.g. B, C, etc., that can co-occur with a given unit, A, notice properties 190

common to the members of the list and assume that other items that have those properties can also occur in that slot. If the child assumes that such frames, which have been generalized from repeated instances of particular constructions, will continue to recur, she can use this knowledge as a Segmentation aid: FR:SEGMENT. Use known frames as templates in attempting to Segment new utterances. I am not prepared to go very far in speculating about how much of these last three OPs is innate and how much is learned, although I think that at least some aspects of them may slowly emerge as the child becomes aware that some chunks of speech have different properties (of occurrence and combination) than others. For instance, the application of FR:SLOT would lead to growing awareness of various types of form-classes such as nouns, verbs, adpositions, inflections, etc. The terms “pattern,” “frame,” and “slot” are thus a linguist’s shorthand for clusters of properties that the child could gradually come to know about. It would seem probable that the drive to try to assign realworld meaning would be innate. Moreover, Slobin has suggested (personal communication) that there might also be an innate set of relational concepts that could facilitate the perception of those functors that mark such concepts (e.g. X is IN/ON/UNDER Y).9 The innateness of a propensity to be sensitive to such gravity-related relational concepts does not seem at all unreasonable. 191

Other types of functors, however, seem much more problematic. Perhaps they are dealt with by an innate drive to try to assign structure to perceived sequences. An argument for the innateness of such a drive is made in Newport (1981). It is interesting to note that in many of the world’s languages the most important functors tend to be found in one of these two perceptually salient positions—either as suffixes or as prefixes to lexical roots. Might such positioning be somehow tied up with human cognitive processing principles? Moreover, it may be significant that of a sample of 444 of the world’s languages surveyed in Hyman (1977), 406 have fixed stress with respect to word boundaries. Of these 114 have word-initial stress, 97 have word-final stress, and 77 have penultimate stress. In this last category it would be interesting to know for how many stress falls on the root/ stem, with the final unstressed, but otherwise perceptually salient syllable being some kind of grammatical ending. It would seem that such a system would be very helpful for the language learner, since the fixed penultimate stress would help in root/stem (content word) identification, and once this was accomplished the recency effect would help in processing of grammatical endings.10 Evidence that children do indeed recognize linguistic frames is found in their production of two-unit utterances consisting of an invariant chunk plus a variable part drawn from some sort of a lexico-grammatical class. E.g.

192

That’ s-a + label I’m-unna + action For at least some children such frames are very easy to find in the input speech they hear, since it includes many expressions accompanying various kinds of labels. For instance, the kind of father-to-son speech described by Thomas (1980) provides the learner with many frames for labels. These occur within what she calls syntactic-pragmatic (SYN-PRAG) routines, and include not only the more commonly noticed noun- and attribute-introducers such as That’s(-a) — , but also such verb-introducers as whyncha + Verb

(why don’t you ______)

let’s + Verb wammeda + Verb

(want me to ______)

can you + Verb ya hafta + Verb

The strategy of using such frames as Segmentation aids may work so well that adult observers never notice it unless it leads to under-Segmentations such as: Mother: What is this whole thing? Minh: Whole thing.

193

(Peters, 1983)

Mother: What’s the cat’s name? Adam: Cat name.

(Clark, 1977)

At first these frames will probably be maximally simple, e.g. one fixed part and one variable part, as in the examples just given. But as a child’s processing capabilities grow we would expect that she will learn to handle more and more complex frames, such as discontinuous frames with one or more slots, and frames in which the slots can be filled with other frames rather than single units. In this chapter I confine myself to some remarks about discontinuous frames and refer the reader to Peters (1983, Ch. 3) for a more extensive discussion of generalization of frames. An example of a discontinuous frame with two slots can be found in the productions of another of Wong Fillmore’s Spanish-speaking subjects (1976, p. 350). After approximately 5 months of exposure to English 7-year-old Jesus had the following sentence frame: Is + Verb + it + Noun-phrase. which led to such productions as Is putting it dese. Is making it the car. Is got

194

it dese one.

Is got

it un truck.

Slobin (1973, 1981) and MacWhinney (1978) have repeatedly suggested that discontinuity is difficult for a child to deal with. Their evidence comes from certain discontinuous structures (e.g. French negative ne … pas; Arabic negative ma- … -sh; English progressive be … ing; Russian and Hungarian case marking involving both prepositions and case suffixes) which children seem to deal with by first producing only the second of the two elements in question. Saliency factors surely play some role in the particular cases cited. Thus French ne is phonologically almost nonexistent. Is the same true for Arabic ma-? The English -ing suffix is certainly the most constant aspect of the progressive, since the form of the verb be is not only variable (is, am, are) but can be phonologically much reduced (’s, ’m, ’re). As for the case marking constructions (e.g. Russian locative), it may be the case that the prepositions are more variable than the case inflections (i.e. that several different prepositions take the same inflection). If this is true, then the more constant inflections might be more salient. Of course, the greater salience of word ends may have an effect here, too. In other words, the cited examples do not provide a real test of the effects of discontinuity since the two ends are not equally salient. All of these cases, moreover, involve discontinuous MORPHEMES, where the two pieces are at once semantically and functionally related. These relationships do not necessarily hold for discontinuous FRAMES. Thus, although at the morphological level there 195

may be justification for Slobin’s “Avoid interruption … of linguistic units” (1973, p. 199) or MacWhinney’s “the child avoids acquisition of discontinuous morphemes” (1978, p. 11), discontinuous frames do not need to be avoided altogether. Evidence from the success of pattern practice and substitution drill shows that adult language learners can handle discontinuous frames pretty easily. And the example cited above of Wong Fillmore’s subject Jesus shows this child using a discontinuous frame with two slots in it. (See Peters, 1983, for further discussion and examples.) On the other hand, Slobin’s and MacWhinney’s claims do reflect two common sense processing constraints (suggested in Slobin, 1981): since it is hard to process too many elements at once (1) one should attempt to keep the total number of elements in any structure to a minimum, and (2) one should try not to be processing more than one element at a time (i.e. one should try to proceed as if structures are not embedded). But if the situation seems to warrant it, discontinuity is acceptable (within the learner’s processing capabilities).

2.4. OPs for Feedback As has been repeatedly suggested, the child also needs OPs for dealing with feedback, i.e. in evaluating her linguistic hypotheses. The first three evaluative OPs are based on three types of perceptual salience with respect to potential results of Segmentation: convergence of several Segmentation heuristics, frequency of resulting sub-units, and transparency of meaning.

196

SG:CONVERGE. If several Segmentation heuristics result in the same cut or sub-unit, the result is a better one than if only one Segmentation heuristic could have achieved it. SG:FREQUENCY. If a particular sub-unit resulting from a Segmentation occurs frequently, especially over a short span of time, it is better than one which occurs less frequently. SG:MEANING. If a clear meaning can be associated with a particular sub-unit resulting from a Segmentation, then the cut is better than one that does not result in a unit with a clear meaning. The different kinds of interaction possibilities (reinforcement vs. conflict) among different saliency factors are reflected in SG:CONVERGE. The application of SG:FREQUENCY may cause the persistence of many of the mis-Segmentations that language learners make. An example of such a mis-Segmentation in English is the over-Segmentation of behave through the false Segmentation of be. Similar mis-Segmentation errors, this time involving forms of the definite article, occur in French acquisition, e.g. l’ avion may be perceived as la vion, l’ électricité as les lectricités (Eve Clark, personal communication). This strategy of trying to match familiar items persists well on into later life as is attested by older children’s misanalyses of such later-learned language as The Star Spangled Banner (e.g. “Jose can you see”), as well as by

197

folk-etymologies (e.g. asparagus = “spare grass” or “sparrow grass”). Any bootstrapping operation depends on a feedback cycle to help provide information needed for making necessary adjustments to the developing system. In language acquisition a production-perception feedback cycle provides such information. (See MacWhinney [1978] for discussion of such a cycle.) As the child’s linguistic system develops, awareness of her own speech may become an increasing source of salience for (re)analysis and Segmentation. This suggests another OP: SG:MONITOR* Monitor focal aspects of your own speech and compare them with what you hear and remember of the speech of others. If a novel utterance based on a presumed Segmentation is not acceptable (not understood, not heeded, or somehow sounds funny) then such negative feedback casts suspicion on the Segmentation. It must be borne in mind, however, that while it must be the case that language learners monitor their own speech and become aware of discrepancies between what they produce and what they hear, it is not the case that all such discrepancies are equally accessible at a given stage of language development. This latter claim is attested to by the oft-repeated anecdotes of children whose language is impervious to repeated corrections by adult caretakers (e.g. Braine, 1971, pp. 160-161). Returning to the lens analogy, I would like to propose that the monitoring and awareness of discrepancy that does take 198

place occurs at those points in the system that are currently being most clearly focused on, since this is where the most active development is taking place. Any discrepancies which occur in the fuzzy background of detail not yet focused on will be ignored. A Segmentation which is rejected by these evaluation OPs may simply be quietly forgotten (i.e. the sub-units fade from storage) or it may be actively repaired. SG:REANALYZE. When a Segmentation fares poorly with respect to these evaluation heuristics, reanalyze the original unit. The following sequence, reported to me by Steven Schoen is a good example of SG:REANALYZE in action: Christine is a 4-year-old girl, Steven is an adult male. They are riding in the back seat of a car. Christine is acting rowdy. Steven tells Christine she “must behave” if she wants Steven to read her a book. He is, however, paying more attention to a cassette tape which is playing music than he is to Christine. A couple of minutes later: C: Steven I am /heyv/. S: What? You hate? What do you hate? C: /heyv/. I am /heyv/. S: You hate? You hate me? The music? What? C: No, I am /heyv/. /heyv/. S: I don’t know what you are talking about.

199

Silence. A bit later: C: I /heyv/. S: You hate me? C: (shakes her head no) S: Who do you hate? Silence. A bit later: C: I am behaving.

In this example, Steven’s repeated lack of understanding forced Christine twice to try and reanalyze the word behave. It was only when she reached the correct (adult) analysis that Steven became aware of the nature of her difficulty. There is another kind of feedback that must also be involved as Segmentation proceeds. Thus the child will discover that in the particular language she is learning certain OPs are more useful than others in that they apply more often and that their results are more reliable. For instance, SG:REPETITION will be of little use to a child learning English and therefore may be abandoned, whereas it may prove quite useful to a child learning a Bantu language. Dan Slobin has suggested (personal communication) that the pattern of success and failures in the application of such perceptial OPs may help a child to build up an increasingly precise “typological sense” of what sorts of units and sub-units to expect in the language she is learning.

3. RELATED ISSUES

200

3.1. Methodological Questions In proposing a theory like this one of how children perceive language, there is a major assumption that we would like to be able to make, namely that this set of OPs is valid for all children. This assumption entails three sub-assumptions. One is that the theory holds for children acquiring any natural human language, i.e. a language of ANY HUMANLY EVOLVABLE LINGUISTIC STRUCTURE. (I include here acquisition of Sign as a first language.) A second sub-assumption is that the theory holds for children successfully acquiring their community’s language in ANY POSSIBLE SOCIO-CULTURAL 11 CONFIGURATION. A third assumption is that this theory is valid for children with A WIDE RANGE OF COGNITIVE ABILITIES and A RANGE OF PREFERRED INDIVIDUAL STYLES of interacting with external data (e.g. children who have been labelled gestalt as well as analytic, field-dependent as well as field-independent, expressive as well as referential). Although the scope of this paper precludes extensive discussion of these three sub-assumptions, I would like briefly to say a few words about each, both as a way of setting this theory in a larger context and as a way of indicating further research that needs to be done.

3.1.1. Crosslinguistic Validity In Peters (1981) I have begun to look at how different sorts of linguistic properties (e.g. stress, tone, morphophonology) affect the segmentability of particular languages. There are two basic problems in this area. One is that language acquisition studies have not been done (or at least published) on a very wide or

201

systematic sample of the languages of the world. In particular, almost nothing has been published about languages that can be characterized as “polysynthetic”—the very languages in which one might predict that Segmentation would prove to be most difficult. The second problem is that, even for those languages for which published acquisition studies are available, because of the tendency to focus on later stages of acquisition, information about ease or difficulty of Extraction and Segmentation is often not included. In the hope of obtaining more such information I have included in the next section a list of various types of information that it would be useful to have. Section 4.1 deals with information about linguistic characteristics, while 4.2 deals with information about the child’s successes or failures. I hope that whenever possible researchers will include such information in their publications.

3.1.2. Cross-Cultural Validity Space considerations preclude detailed consideration of the possibilities of cultural influences on children’s Segmentation strategies. Therefore, I briefly summarize here the important issues in this area, which concern the kind of language the child hears, i.e. the nature of the input. The characteristics of caretaker speech are particularly important because such speech is the primary data the child has upon which to do Extraction and Segmentation. Recent American studies of “motherese” have primarily been conducted with middle-class “mainstream” families

202

(e.g. a number of the studies in Snow & Ferguson, 1977). Their findings were that speech to very young children tends to be highly adapted to facilitate communication with such young children: it is broken down (or “pre-digested”) to fit the caretaker’s perception of the child’s competence. It is also often characterized by repetitiveness, often involving expansions, both of previous caretaker utterances and also of utterances by the child. (Expansions of caretaker utterances are usually made to enhance comprehension by the child, whereas repetitions and expansions of child utterances tend to serve as confirmation checks on what the child intended to say.) As the child’s linguistic capabilities develop, caretakers “raise the ante” (in Bruner’s, 1978, terms) by increasing the length and complexity of the speech directed to the child just enough to keep challenging without frustrating the young learner. Studies of “baby talk” in a number of languages (e.g. some of the other studies in Snow & Ferguson, 1977) found similar enough characteristics that Ferguson, in his 1978 paper “Talking to children: A search for universals” felt confident in claiming that in all communities speech to young children is modified along these lines. But then three new, ethnographically based language acquisition studies, carried out in very different cultural settings, began to show that these generalizations had been made too soon and that input speech could have very different characteristics. These studies include Schieffelin’s (1979) study done among the Kaluli of New Guinea, Ochs’s study done in America Samoa (Ochs & Schieffelin, in press), and Heath’s (1980, 1982) study of a black community in the Piedmont Carolinas. 203

Schieffelin (1979; Ochs & Schieffelin, in press) found that the Kaluli are very conscious of the language development of their children and deliberately try to teach them to speak by direct instruction, that is, by telling them appropriate things to say in specific circumstances. Such speech, however, is normal adult speech—in fact the Kaluli find the idea of “baby talk” dismaying. Even young babies are seen as “individuals with intentions, ideas, and identities” (1979, p. 108), and it is important for mothers to help these identities emerge, at first by speaking for their infants, then by instructing their toddlers how to interact in appropriate social and linguistic ways. Although some Kaluli instruction takes place in dyadic situations (when only mother and child are interacting), a much larger portion takes place in triadic situations where the mother coaches the child on what to say to a third party. What are the effects of such language on Segmentation? Since Schieffelin deliberately chose children who were beginning to put words together, they were all beyond the initial Extraction stage. Thus she gives no data on their successes or failures in Segmenting adult-sized words out of such speech. I wonder, however, whether this unmodified type of input might not be somewhat harder to Segment than “mainstream” American input. One thing that struck Ochs (Ochs & Schiefflin, in press) about caretaker speech in American Samoa was the absence of expansions of utterances made by the children. After comparing Samoan and American middle-class beliefs about status, rights to speak, and capability of acting intentionally, she concludes that Samoan caretakers do not expand their charges’ 204

utterances because young children are not considered really capable of communicating intentionally. It is also not appropriate for a (higher status) caregiver to try to accommodate to a (lower status) child’s perspective by offering confirmation checks (in the form of expansions) about what the child might have intended to say. On the contrary, much of caretaker speech consists in providing social instruction for the young child, including elicitation of imitations of people’s names and socially appropriate greetings, questions, etc. (Ochs, 1980, p. 32). In this way Samoan caretaker speech seems similar to that among the Kaluli, although the Kaluli have a strong language-instruction component in their caretaker talk which seems to be absent among the Samoans (p. 37). Samoan input is also similar to Kaluli input in that it is not marked by “baby talk lexicon, special morphological modifications (diminutives, etc.), simpler syntactic constructions or constructions of reduced length” (p. 36). These features might make Segmentation more difficult for the Samoan child. On the other hand Ochs notes that “features such as self-repetition and paraphase do appear in caregiver speech, but they are not exclusive to caregiver-charge interaction” (p. 36). One might suspect that the fact that “self-repetition and paraphrase abound in talk, both formal and informal between adults in traditional Samoan society” (p. 36) would somewhat facilitate the Segmentation task. Heath’s study (1980, 1982) includes observation of language learning in a small all-black residential community in the Piedmont Carolinas. In this community,

205

which she calls Trackton, adults do not think of children as conversational partners until they are skilled enough to be “see as realistic sources of information and competent partners in talk” (1982, p. 114). Since Trackton mothers, like the Kaluli and Samoan mothers, usually have other more skilled potential conversational partners, they are not limited to talking with toddlers—and they don’t. But neither do they exclude young children from adult talk. Direct communication with infants is primarily nonverbal since they are carried about all day. Linguistically the children are exposed to streams of talk which continually flow past them. And their first attempts at speaking (between 12 and 24 months) tend to consist of imitations of the ends of overheard phrases or sentences. Heath’s examples of such imitations show children picking up in this way chunks which are several morphemes long. It seems, not surprisingly, that given such input they Extract out, not single words, but longer chunks which are probably intonationally salient as well as intonationally delimited. As a child’s linguistic skills develop he is heard not only imitating a phrase, but also playing with it, manipulating the parts in various ways. Heath calls this stage “repetition with variation” (1980, p. 33). This process seems quite similar to the formulaic breakdown observed in child second language learners by Wong Fillmore (1976). It seems quite significant that in this community, as in Samoa, such repetition and paraphrase is common among adults. In the black community of Trackton it is part of a verbal style that is heard in storytelling and prayers (Heath, personal communication). Heath’s description of a language acquisition process such as the one in Trackton, where children start with long chunks 206

which they then learn to vary, without specially tailored input language from their caretakers, seems particularly important for a theory of Segmentation (and language acquisition in general). This is because now we have documentation that children do not need to have language pre-digested, nor do they have to start at the level of single words in order to acquire language satisfactorily. Even more importantly, Extraction and Segmentation heuristics such as those presented in this chapter would enable the child to discover a linguistic system from any of these types of input. We do not as yet know, however, whether there might not be other kinds of cultural influences on early language acquisition processes— this line of inquiry still needs to be pursued much further. The research questions in 4.3 are addressed to such cross-cultural aspects of language acquisition.

3.1.3. Individual Variation Among children who are all learning the same language and living within culturally similar families there nevertheless seems to exist a range of preferred styles of interacting with language. Thus, among the 18 English-learning children she studied, Nelson (1973) felt that there were two basic styles, which she called Expressive and Referential. Expressive children are characterized as using language more in social interactions, while Referential children use language more for labeling objects and attributes. What concerns us here is that her Expressive children were more likely to learn and use longer chunks of language (whole sentences or phrases)

207

as single “words” or units, whereas the Referential children were more likely to operate with single adult words. This may imply the existence of a preference, or at least a tolerance for long unsegmented chunks of speech among the Expressive learners and a preference for short chunks of speech among Referential learners. Such preferences have not been systematically studied, but any cognitively based theory of language perception needs to be able to account for them. In Peters (1977) I called the large-chunk approach Gestalt and the small-chunk approach Analytic. With respect to our earlier discussion (section 2.3 above) of analysis of sentences into morpho-syntactic frames with lexical slots, I wonder whether Gestalt learners do not tend to concentrate more on the general frame that goes to make up a whole sentence, whereas the Analytic learners would tend to focus first on the lexical items that go in the slots. This would tie in with the cognitive categorizations Field Dependent and Field Independent. The research questions in section 4.2 are relevant to the question of individual differences in cognitive style.

3.2. Implications for a Theory of Second Language Learning It seems like that second language learners must use basically the same sort of OPs as first language learners, at least in naturalistic situations (as opposed to classroom learning). There is evidence from studies such as those of Wong Fillmore (1976), Huang & Hatch (1978), and Vihman (1980), that child second language learners often start by Extracting large chunks which they then 208

Segment into smaller pieces. This raises the following sets of questions: 1. What is the effect of the more advanced state of cognitive deveopment of the older language learner on the use of OPs for learning a new language? Is this effect solely in the form of fewer processing limitations in terms of, say, memory span? Does it affect which OPs are used more than others? What is the effect of already acquired linguistic knowledge on the use of particular OPs? Contrastive analysis of the two linguistic systems involved might yield predictions about use of OPs: which ones would successfully transfer and which ones would lead to problems. 2. Do adults learning a second language in a naturalistic situation approach Segmentation in the same way as child second language learners? i.e. do they use the same OPs? (One could control at least for same pairs of languages. What about input characteristics?) 3. What are the implications for classroom teaching of second languages? If Segmentation is a “natural” ability/ process for language learners, should more advantage be taken of it in the classroom? e.g. though scaffolded routines? alternate focusing on slots and frames? incorporated pattern practice and substitution drills?

4. RESEARCH QUESTIONS RELEVANT TO SEGMENTATION HEURISTICS In trying to work out how children perceive, extract, and segment their first linguistic units, I have collected a long list of questions in three main areas for which more data is needed. The three basic questions are:

209

1. What is the influence of the typlogical structure of the language a child is learning on the size and shape of extracted units, and does this typological structure have an increasing influence on the OPs a child will use? 2. Within a particular language (so that typological structure is fixed) what sorts of extraction and segmentation errors do children actually make, and what can we infer from these errors about OPs they must be using? 3. Within a particular language (again, so that structure is fixed) how do various kinds of linguistic input affect extraction and segmentation? i.e. do some kinds of input tend to reinforce the use of certain OPs (e.g. resulting in the extraction of smaller vs. larger units)? The availability of more data in all of these areas should help us to be more precise about the OPs children use in extraction and segmentation, helping us to determine which OPs seem to be used most frequently, which ones seem to be used earliest, how and when new OPs are added to a child’s repertoire, whether some are tried out and then abandoned, etc.

4.1. Influence of Language Structure on Extraction and Segmentation The issues underlying the following questions about linguistic structure (cf. 3.1.1.) are discussed in more detail in Peters (1981). In general, these questions request that specific types of information be supplied about the structure of the particular language being 210

acquired. Whenever possible I have also suggested specific predictions which can be tested against the data. Another approach to this question of the influence of language structure on the perception of linguistic units would be to conduct experiments using a set of miniature artificial languages designed to investigate the proposed parameters (stress, rhythm, intonation, morphophonemic boundaries, canonical shapes of morphemes, etc.) in a systematic way.

A. Stress and Rhythm 1. Is location of stress with respect to word boundary predictable? Prediction: The existence of predictable stress will tend to prevent the extraction of units longer than a word, except in cases where word stress is subordinated to phrase stress. 2. If stress is not totally predictable, is it nearly so? What are the exceptions like? Might these be responsible for any segmentation errors (due to overgeneralization)? Prediction: If a language has predictable location of stress for almost all words, the few exceptions are likely to be involved in segmentation errors. 3. Are there characteristic stress envelopes for words? Prediction: At the very earliest stages of speech production, a child learning a language with strong

211

rhythmic patterns for words may extract and produce such rhythms (which may sound to us like some kind of pseudo-speech). Prediction: There will be few segmentation errors that involve word boundaries in languages that have such regular rhythms. 4. Are there clitics which affect stress patterns? Is word-stress reduced in favor of phrase stress? Do phonologically- and lexically-defined “words” tend to be overlapping or coextensive? Do these properties seem to have any effect on segmentation? Prediction: A clitic will tend to be perceived as an integral part of the host word. Prediction: Phrasal articulation of lexicalized chunks is particularly likely to cause segmentation problems. 5. Is there stress-timing, so that some syllables become much reduced, or syllable-timing so that each syllable has approximately the same duration and clarity? Prediction: In a stress-timed language single (stressed) syllables are more likely to be extracted than unstressed ones. In a syllable-timed language there should be no such preference. 6. Is there any kind of rhythmic structuring of whole words (stems plus affixes) along purely syllabic or metric lines that may tend to obscure internal morpheme boundaries?

212

Prediction: In languages with these properties children are likely to make segmentations (either correct or incorrect) that respect syllabic rather than morphological boundaries.

B. Pitch and Tone 1. Is there a pitch-accent system in which there is one pitch peak per word? If so, is the location of this pitch peak predictable with respect to the word boundary? Prediction: The existence of predictable accent will tend to prevent the extraction of units longer than a word. 2. Are there characteristic pitch envelopes for words? Prediction: At the very earliest stages of speech production, a child learning a language with strong pitch patterns for words may extract and produce such tunes (which may sound to us like some kind of pseudo-speech). 3. If the language has tones, do they occur on every syllable, or is there also a “neutral tone”? If so, what are its properties vis-à-vis word boundaries (e.g. always final or always initial)? Prediction: Predictability of location of neutral-toned syllables with respect to word boundaries will aid in correct extraction and segmentation.

213

C. Phonemics and morphophonemics 1. Are there characteristic word-initial or word-final allophones? 2. Is there word-internal vowel harmony? If so, are there affixes which do not harmonize, or free functors which do harmonize? Prediction: If there are affixes which do not harmonize they will tend to be perceived as separate words. Prediction: Conversely, if there are free functors which do harmonize, they will tend to be perceived as parts of the words they occur with. 3. Is there “heavy” sandhi (e.g. metathesis, epenthesis, deletion, diphthongization or fusion of vowels, etc.) at morpheme boundaries, which would tend to obscure such boundaries (e.g. French liaison)? Prediction: Languages which are highly agglutinative, i.e. with no sandhi at morpheme boundaries, will be easier to segment than more fusional languages. Prediction: In languages which are highly fusional, segmentation will tend to occur at first along syllabic rather than morpheme boundaries. 4. Are there resyllabification processes which cross word or morpheme boundaries? i.e. are there prefixes or infixes which cause resyllabification of stems in ways so as to make them hard to recognize in different contexts? E.g., are there both consonant- and vowel-initial stems as well as consonant- and vowel-initial prefixes which lead

214

to morphemes straddling syllable boundaries (e.g. vc-VCV > v.cV.CV)? Or if there are infixes, do they split up the syllable structure of stems (e.g. vc infixed in CVCV > Cv.cV.CV)? Prediction: In languages with these properties children are likely to make segmentation errors that respect syllabic rather than morphological boundaries. Prediction: It will be easier to locate the boundaries of morphemes which consist of integral numbers of complete syllables than of those which consist of only parts of syllables.

D. Morpheme and word structure 1. Are canonical shapes of morphemes simple (e.g. CVCV) or complex (as in English)? Prediction: Words which have rare or exceptional canonical shapes (within a particular language) are more likely to be involved in segmentation errors. 2. Are stems and affixes easily distinguishable from each other by canonical shape? In general are affixes syllabic? Are there exceptions which might cause segmentation problems? Prediction: If stems and affixes have distinct canonical shapes, word-internal segmentation will be easy for the child to discover, i.e. there will be few errors. 3. Are there morphological agreement phenomena which lead either to alliteration (as in Bantu languages) or

215

rhyming (as in Hebrew or Spanish) concord? (Note that not all morphological agreement leads to such phonological repetitions.) 4. Do words tend to be composed of three or more obligatory morpheme slots, so that a child might be aware of this obligatory frame but not know (at first) exactly what goes in some of the slots, thus leading her to use “filler syllables” (as e.g. in Turkish verbs)? 5. Is it possible for morphemes to be interdigitated (either as in Semitic languages, with tri-consonantal roots plus interdigitated vowels, or as in Philippine languages, with infixes?) Prediction: Interdigitation of morphemes will preclude correct segmentation until enough interdigitated chunks have been extracted that the pattern can be perceived. Prediction: Infixing which radically alters the syllable pattern of a stem can be expected to cause errors and/or delay in recognition of the correct morphological segmentation. 6. Do grammatical (functor) morphemes tend to be semantically and phonologically transparent (one sound-sequence: one function) or are there portmanteau or fused morphemes? Prediction: Portmanteau or fused morphemes (where an unsegmentable (sequence of) phoneme(s) carries multiple semantic functions) may be particularly confusing to deal with and lead to numerous errors. On

216

the other hand, the one morpheme-one meaning situation will involve few errors. Prediction. Affixes which are obligatorily and regularly attached, to pro-forms as well as substantives, will be easily perceived and segmented.

4.2. Evidence of Malsegmentation The issues addressed by this list of questions are discussed in more detail in Peters (1983). The information requested here deals with evidence about the child’s failures (and successes) at segmentation (cf. 3.1.1. and 3.1.3). A. Undersegmentation. Are there chunks of speech the child uses that are composed of more than one morpheme but which seem to be used as if they were single unanalyzed units? The following are some (not necessarily mutually exclusive) criteria for detecting such chunks: 1. Are they unrelated to any productive patterns in the child’s speech? 2. Are they used repeatedly and always in the same form (within the child’s phonological limits)? 3. Are they sometimes morphologically inappropriate for the context? E.g. I-carry-you said when the child wants to be carried; or doubly inflected, as in feets or broked.

217

4. Does it cohere phonologically, i.e. is it articulated in a single smooth contour (rather than being marked by hesitations)? 5. Is it situationally bound? i.e. does it occur only in certain situations? B. Mis-segmentation. 1. Are there lexical items used by the child that show evidence of mis-segmentation of words or phrases by splitting morphemes in the middle? If so, do these segmentations respect some sort of morpheme structure constraint such as that morphemes begin with a consonant? E.g. Rachel Scollon heard the story of the Wizard of Oz (and the land of Oz) and declared she was going to Voz: wizardofoz > wizarda voz landofoz

> landa voz

How are such segmentations related to stress? 2. If children’s early productions are monosyllabic, do they correspond to final syllables in the adult targets? to initial syllables? to stressed syllables? What characterizes the syllables that children extract in syllable-timed languages? C. Oversegmentation. (1) Are there unitary items in the adult language that seem to have been segmented into two or more “morphemes” by the child? E.g.

218

behave into be + have (/heyv/, or Anthony Weir’s oversegmentation of whistle and measure which gave him wIs (= the act of whistling) and mEž (= the act of measuring) (Weir, 1962, p. 74).

4.3. Predisposing Factors in the Input The information requested here deals with cultural patterns of interacting with children (cf. 3.1.2.). A. What are caretaker interactions like? 1. Who are the primary caretakers of young children? What is a child’s primary social interaction network (i.e. the people from whom she hears the greatest amount of language)? Are the child plus caretaker generally at home alone all day (a dyadic situation) or are there other potential interactants usually present? 2. At what age or stage is a child perceived as a potential conversational partner? e.g. from birth? as a babbling infant? at the one word stage? when she can say certain words? when she is able to initiate conversational interactions on her own? 3. Is a significant amount of speech directed to the child, or does speech primarily flow past the child? Prediction: Larger chunks of speech will be extracted when speech flows past the child than when it is addressed to the child. 4. If there are conversational interactions with the child are they primarily dyadic (caretaker plus child) or triadic

219

(caretaker plus child plus other)? Is their nature conversational? playful? instructional (for language? or behavior?)? Prediction: Single words will tend to be extracted in name-instruction situations, whereas formulaic chunks will tend to be extracted in behavior-instruction situations. 5. Do caretakers provide the child with linguistic instruction? If so, is it in the form of (a) direct instruction, where the child is told to “Say X,” or (b) scaffolded routines which offer implicit support structures for performance of certain kinds of interactions (and which “self-destruct” as the caretaker “raises the ante” and the child learns to play first one then the other role)? Does such linguistic instruction typically occur in dyadic or triadic situations? Prediction: In general, larger chunks of speech will be extracted in triadic than in dyadic interactions. 6. Is speech which is directed to the child prosodically marked in attention-getting ways, such as by use of high pitch, exaggerated intonation contours, special voice quality (e.g. breathy, creaky), or changes in tempo (either slower or faster than talk to adults)? Prediction: Prosodically marked speech is more likely to be extracted than prosodically unmarked speech. 7. Are any of the following alterations made from normal adult speech in order to make it “easier” for the 220

child to understand: Is pronunciation distorted? Are morphophonological processes “undone” (as in Hidatsa, cf. Voegelin & Robinett, 1954)? Is syntax simplified to approximate the perceived level of the child’s competence? Prediction: Speech altered for the purpose of enhanced comprehension by the child is more likely to be segmented correctly than unaltered speech. 8. Are successive utterances often characterized by repetitions, expansions, or deletions which may provide segmentation cues? Is the intent of such repetitions to facilitate comprehension by the child? or do they occur as part of the normal adult speech style in the community? Prediction: The presence of repetitions, expansions and/ or deletions in the input will facilitate segmentation. B. Does input speech contain recurrent chunks, segmentation cues? Does the speech the child regularly hears contain recurrent pieces that might lead the child to perform extractions, segmentations, and to recognize morpho-syntactic frames (either correctly or wrongly)? Such kinds of speech include: 1. Common everyday phrases that may be extracted as units, e.g. all gone, lookit.

221

2. Verbal routines that accompany everyday activities such as bathing, feeding, etc., where recurrent chunks are likely to be extracted. 3. Verbal routines that offer support for learning about particular grammatical classes or acquiring particular linguistic constructions. E.g. what’s that? and the learning of nouns; where’s Daddy? and the learning of locatives; what color is this? and the learning of color words; tell what happened and the learning of past tense; Turkish story telling and the learning of non-witnessed modality (Aksu-koç & Slobin, 1985); SiSwati “whose X is it?”/“Y’s” as a way of learning possessives (Kunene, 1979). 4. Repetition of crucial words within a single utterance, such as Finnish and Portuguese imperatives (‘Kiss Mommy kiss’) (Argoff, 1976; Slobin seminar, summer 1980). 5. Expansion, rephrasing, etc. of the caretaker’s or the child’s speech which may help the child to recognize the constitutent morphemes. 6. Phrases from books frequently read to the child which provide supporting frameworks for particular constructions. E.g. Eve’s “Remember When” book as a

222

support for past tense constructions (Roger Brown data, analyzed in Slobin seminar, summer, 1980).

ACKNOWLEDGMENTS This chapter is a revised and expanded version of “Initial heuristics in early language acquisition,” Cognition and Brain Theory, V.2, Spring 1982. I would like to acknowledge the helpful comments and suggestions which contributed to the development of the ideas presented here. Many of these cane from the participants in Dan Slobin’s seminar at UC Berkeley during the Winter quarter of 1981. Others were provided by Elizabeth Barber, George Grace, Brien Hallett, Robert Hsu, Ted Rodgers, Sheldon Rosenberg, Dan Slobin, and Amy Strage. This work was supported in part by the Social Science Research Institute of the University of Hawaii.

REFERENCES Aksu-koç, A. A., & Slobin, D. I. The acquisition of Turkish. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Argoff, H. The Acquisition of Finnish inflectional morphology. Unpublished doctoral dissertation, University of California, Berkeley, 1976. (University Microfilms)

223

Berman, R. A. Regularity vs. anomaly: the acquisition of Hebrew inflectional morphology. Journal of Child Language, 1981, 8, 265–282. Berman, R. A. The acquisition of Hebrew. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Bloom, L. Language development: Form and function in emerging grammars. Cambridge, MA.: MIT Press, 1970. Braine, M. D. S. On two types of models in the internalization of grammars. In D. I. Slobin (Ed.), The ontogenesis of grammar. New York: Academic Press, 1971. Braine, M. D. S. Children’s first word combinations. Monographs of the Society for Research in Child Development, 1976, 41 (1, Serial No. 164). Brown, R., & McNeill, D. The “tip of the tongue” phenomenon. Journal of Verbal Learning and Verbal Behavior, 1966, 5, 325-337. Bruner, J. S. The role of dialogue in language acquisition. In A. Sinclair, R. J. Jarvella, & W. J. M. Levelt (Eds.), The child’s conception of language. New York: Springer-Verlag, 1979. Clark, R. What’s the use of imitation? Journal of Child Language, 1977, 4, 341–358.

224

Ekmeçi, O. F. Acquisition of Turkish. Unpublished doctoral dissertation, University of Texas, Austin, 1979. (University Microfilms No. 7928282) Ferguson, C. A. Talking to children: A search for universals. In J. H. Greenberg (Ed.), Universals of human language, Vol. 1: Method and theory. Stanford: Stanford University Press, 1978. Hagen, J. W., & Stanovich K. E. Memory: Strategies of acquisition. In R. V. Kail & J. W. Hagen (Eds.), Perspectives on the development of memory and cognition. New York: Wiley, 1977. Heath, S. B. What no bedtime story means: Narrative skills at home and at school. Paper prepared for the Terman Conference, Stanford University, Nov. 20–22, 1980. Heath, S. B. Questioning at home and at school: A comparative study. In G. Spindler (Ed.), Doing the ethnography of schooling: Educational anthropology in action. New York: Holt, Rinehart & Winston, 1982. Huang, J., & Hatch, E. M. A Chinese child’s acquisition of English. In E. M. Hatch (Ed.), Second language acquisition: A book of readings. Rowley, MA.: Newbury House, 1978. Hyman, L. M. On the nature of linguistic stress. In L. M. Hyman (Ed.), Studies in stress and accent. Southern California Occasional Papers in Linguistics, No. 4. Los Angeles: University of Southern California, 1977. 225

Jordan, A. C. A practical Johannesburg: Longmans, 1966.

course

in

Xhosa.

Kintsch, W. Memory and cognition (2nd ed.). New York: Wiley, 1977. Kunene, E. C. L. The acquisition of SiSwati as a first language. Unpublished doctoral dissertation, University of California, Los Angeles, 1979. (University Microfilms No. 7926038) MacWhinney, B. The acquisition of morphophonology. Monographs of the Society for Research in Child Development. 1978,45, (1–2 Serial No. 1974). Nelson, K. Structure and strategy in learning to talk. Monographs of the Society for Research in Child Development, 1973, 38 (1–2, Serial No. 149). Newport, E. L. Constraints on structure: Evidence from American Sign Language and language learning. In W. Collins (Ed.), Minnesota Symposium on Child Psychology (Vol. 14). Hillsdale, NJ: Lawrence Erlbaum Associates, 1981. Ochs, E. Talking to children in Western Samoa. Unpublished manuscript, 1980. Ochs, E., & Schieffelin, B. B. Language acquisition and socialization: Three developmental stories and their implications. In R. Schweder & R. LeVine (Eds.), Culture and its acquisition. Chicago: University of Chicago Press, in press.

226

Peters, A. M. The beginnings of speech. Papers and Reports on Child Language Development (Dept. of Linguistics, Stanford University), 1974. Peters, A. M. Language learning strategies: Does the whole equal the sum of the parts? Language, 1977, 53, 560–573. Peters, A. M. The units of language acquisition. University of Hawaii Working Papers in Linguistics, 1980, 12, 1–72. Peters, A. M. Language typology and the segmentation problem in early child language acquisition. Proceedings of the Seventh Annual Meeting of the Berkeley Linguistics Society, 1981, 236–248. Peters, A. M. The units of language acquisition. Cambridge Series of Monographs and Texts in Applied Psycholinguistics. New York: Cambridge University Press, 1983. Phillips, J. R. Syntax and vocabulary of mother’s speech to young children: Age and sex comparisons. Child Development, 1973,44, 182–185. Pye, C. The acquisition of grammatical morphemes in Quiche Mayan. Unpublished doctoral dissertation, University of Pittsburgh, 1980. Pye, C. Mayan telegraphese: Intonational determinants of inflectional development in Quiche Mayan. Paper presented at Second International Congress for the

227

Study of Child Language, Vancouver, B.C., August, 1981. Schieffelin, B. B. How Kaluli children learn what to say, what to do, and how to feel: An ethnographic study of the development of communicative competence. Unpublished doctoral dissertation, Columbia University, 1979. [Revised version in press, Cambridge University Press.] Slobin, D. I. Cognitive prerequisites for the development of grammar. In C. A. Ferguson & D. I. Slobin (Eds.), Studies of child language development. New York: Holt, Rinehart & Winston, 1973. Slobin, D. I. Materials for the crosslinguistic study of language acquisition. Unpublished manuscript, 1981. Slobin, D. I. Crosslinguistic evidence for the Language-Making Capacity. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. II). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Snow, C. E. & Ferguson, C. A. (Eds.). Talking to children: Language input and acquisition. Cambridge: Cambridge University Press, 1977. Stern, C., & Stern, W. Die Kindersprache: Eine psychologische und sprach-theoretische Unter-suchung. Leipzig: Barth, 1907.

228

Thomas, E. K. It’s all routine: A redefinition of routines as a central factor in language acquisition. Unpublished manuscript, 1980. Tolbert, M. K. The acquisition of grammatical morphemes: A cross-linguistic study with reference to Mayan (Cakchiquel) and Spanish. Unpublished doctoral dissertation, Harvard University, 1978. Velten, H. V. The growth of phonemic and lexical patterns in infant language. In A. Bar-Adon & W. Leopold, (Eds.), Child language: A book of readings. Englewood Cliffs, NJ: Prentice-Hall, 1971. Vihman, M. M. Formulas in first and second language acquisition. Papers and Reports on Child Language Development (Department of Linguistics, Stanford University), 1980. Voegelin, C. F., & Robinett, F. M. “Mother language” in Hidatsa. International Journal of American Linguistics, 1954, 20, 65–70. Weeks, T. E. Born to talk. Rowley, House, 1979.

MA.:

Newbury

Weir, R. H. Language in the crib. The Hague: Mouton, 1962. Wong Fillmore, L. The second time around: Cognitive and social strategies in second language acquisition. Unpublished doctoral dissertation, Stanford University, 1976.

229

Wong Fillmore, L. Individual differences in second language acquisition. In C. J. Fillmore, W. S-Y. Wang, & D. Kempler (Eds.), Individual differences in language ability and language behavior. New York: Academic Press, 1979.

Appendix: Condensed List of OPs

I. EXTRACTION OF EARLY UNITS EX :EXTRACT. Extract whatever salient chunks of speech you can. 230

EX :COMPARE. Determine whether a newly extracted chunk of speech seems to be the same as or different from anything you have already stored. EX:STORE. If it is different, then store it separately; if it is the same, take note of this sameness but do not store it separately.

Ancillary OPs Based on Salience EX :MEANING. Pay attention to utterances that have a readily identifiable meaning. Extract and remember sound sequences that have a clear connection to a clear context. ES:SILENCE. An extractable unit is bounded by silence (i.e., it is a whole utterance). ES:SUPRASEG. An extractable unit suprasegmentally delimited stretch of speech.

is

a

EX :TUNE. An extractable unit is a speech tune or melody. EX :RHYTHM. An extractable unit is a rhythmic pattern of speech.

II. SEGMENTATION: ANALYSIS OF ACQUIRED UNITS INTO SMALLER ONES SG:SEGMENT. Attempt to Segment utterances you hear or utterances you have already Extracted into smaller linguistic units. 231

Internal cues for Segmentation SG:INTONATION.* Segment internationally salient places.

utterances

at

SG:RHYTHM.* Segment utterances at rhythmically salient places. SG:STRESS. Segment off a stressed syllable of an Extracted unit and store it separately. SG:END.* Segment off the last syllable of an Extracted unit and store it separately. SG:BEGIN.* Segment off the first syllable of an Extracted unit and store it separately. SG:REPETITION. Segment off sub-units that are repeated (in terms of segmentals or rhythm or intonation) within an Extracted unit and store them separately.

External cues for Segmentation SG:MATCH I. When the initial or final portion of an Extracted unit phonologically matches the whole of another Extracted unit, the remainder of the larger unit is a candidate for storage as a new unit. SG:MATCH2. More generally, when two Extracted units share ANY phonologically similar portion, the shared portion can be Segmented and stored as a unit, and so can the residue.

232

III. SEGMENTATION OF MORPHO-SYNTACTIC FRAMES FR:FRAME. If two (or more) units, after Segmentation by any of the SG: heuristics, appear to share a common sub-unit, A, followed or preceded by alternative sub-units, B or C, etc., take note of this fact, namely that there is a pattern in which A can be followed (or preceded) by either B or C, etc. FR:SLOT. When you have a list of items, e.g. B, C, etc., that can co-occur with a given unit, A, notice properties common to the members of the list and assume that other items that have those properties can also occur in that slot. FR:SEGMENT. Use known frames as templates in attempting to Segment new utterances.

IV. OPS FOR FEEDBACK SG:CONVERGE. If several Segmentation heuristics result in the same cut or sub-unit, the result is a better one than if only one Segmentation heuristic could have achieved it. SG:FREQUENCY. If a particular sub-unit resulting from a Segmentation occurs frequently, especially over a short span of time, it is better than one which occurs less frequently. SG:MEANING. If a clear meaning can be associated with a particular sub-unit resulting from a Segmentation, 233

then the cut is better than one that does not result in a unit with a clear meaning. SG:MONITOR.* Monitor focal aspects of your own speech and compare them with what you hear and remember of the speech of others. If a novel utterance based on a presumed Segmentation is not acceptable (not understood, not heeded, or somehow sounds funny) then such negative feedback casts suspicion on the Segmentation. SG:REANALYZE. When a Segmentation fares poorly with respect to these evaluation heuristics, reanalyze the original unit. 1

The term “to bootstrap” is used in a number of fields to refer to the activity of getting a system started with minimal resources—a process of pulling oneself up by the bootstraps. 2

See Peters (1977) for discussion of a child who had a natural preference for longer “fuzzier” chunks rather than shorter “clearer” ones. 3

Some of the OPs presented here are based on a set of OPs suggested by Slobin (1981); others are based on my own work (Peters, 1980; 1983). All of the OPs based on Slobin (1981) will be flagged with footnote number 3*. Since my goal in this chapter is to present a coherent, ordered set of OPs that would contribute to a theory of language perception, and since many of Slobin’s OPs were worded so as to reflect production rather than perception, I have taken the liberty of revising such 234

wording as I felt was necessary. I have divided the OPs into three sets dealing with successive aspects of Extraction and Segmentation. For ease of reference I have given the OPs in each of the three sets a common prefix: EX(tract), SG(ment), FR(ame). The Appendix contains a condensed list of all the OPs discussed in this chapter. 4

Ron Scollon, personal communication, reporting on his daughter Rachel. 5

I am indebted to the participants of the Slobin seminar for this observation. 6

Karuna Peters, personal communication.

7

This example was suggested to me by Dan Slobin.

8

The term “morpho-syntactic frame” is approximately equivalent to Braine’s (1976) “positional formula” and Wong Fillmore’s (1976) “sentence frame.” Braine and Wong Fillmore, however, are focusing on the productive aspects of early constructions, whereas I am focusing on the perception of structural patterns, and particularly on awareness of constraints on the order of constituents (syntax) and on the separation of content items from grammatical functors (morphology). The “limited scope” nature of the formulaic patterns that Braine discusses seems to me to be crucial: children do not immediately extract and use broadly general patterns. Rather, their early patterns are rather narrowly restricted, both semantically and syntactically (i.e. with respect to the range of constituents that can go in a particular slot). 235

These patterns are only slowly expanded and generalized, with possible difficulties arising (in the form of “groping patterns”) when two independent patterns with opposing constituents have been acquired. 9

[This position is spelled out in detail in this volume (Slobin, 1985).—Ed.] 10

The acquisition of Mohawk seems to proceed along these lines, according to Marianne Mithun (personal communication). 11

Here I specifically exclude children who fail to acquire language due to lack of exposure to language because of isolation: feral children, closet children, and certain deaf children. *See footnote 3.

236

14

Hungarian Language Acquisition as an Exemplification of a General Model of Grammatical Development

Brian MacWhinney Carnegie-Mellon University

Contents Descriptive Sketch of Hungarian Phonology Morphology Morphophonology Ordering Agreement

237

Sources of Data The Overall Course of Development The Competition Model Competition in the Lexicon Applications of the Model to the Development of Phonology Rote, Analogy, and Combination (Operating Principles 1, 2, and 3) Patterns (Operating Principle 4) Competition (Operating Principle 5) Rule Acquisition (Operating Principles 10 and 15) Item Acquisition (Operating Principle 11) Amalgam Analysis (Operating Principle 12) Merger (Operating Principle 14) Applications of the Model to the Acquisition of Syntax Rote, Analogy, and Combination (Operating Principles 1, 2, and 3) Patterns (Operating Principle 4) Competition (Operating Principle 5) Rule Acquisition (Operating Principles 10 and 15)

238

Item Acquisition (Operating Principle 11) Amalgam Analysis (Operating Principle 12) Applications of the Model to the Acquisition of Lexical Semantics Rote, Analogy, and Combination (Operating Principles 1, 2, and 3) Patterns (Operating Principle 4) Competition (Operating Principle 5) Rule Acquisition (Operating Principles 10 and 15) Item Acquisition (Operating Principle 11) Amalgam Analysis of Morpheme Acquisition (Operating Principle 12) Inference (Operating Principle 13) Suggestions for Further Study Theoretical Implications

Hungarian is a member of the Ugric branch of the Ugro-Finnic language family. In the pre-Christian era the ancestors of the modern Hungarians inhabited an area between the Volga and Dnieper rivers in Central Asia. Subsequent migrations brought them into close contact with Turkic peoples in the area north of the Black Sea. Around 800 A.D., the Hungarians, who call themselves 239

Magyars, entered the Carpathian Basin, occupying most of the area of modern Hungary. They have remained in the Carpathian Basin up to the present date, maintaining close contact with speakers of Slavic, Germanic, Romance, and Turkic languages. Despite these close contacts and despite massive lexical borrowing, Hungarian maintains many of its original Ugro-Finnic characteristics. These include vowel harmony, pragmatically flexible word order, an elaborate set of “agglutinative” case suffixes, an extensive system of aspect markers, verb-object agreement, placement of the noun in the singular when there is a quantifier that is in the plural, a basic SOV word order, and a tendency toward incorporation of the object into the verb. As we see below, each of these characteristically Ugro-Finnic features of Hungarian provides interesting data that can be brought to bear upon hypotheses regarding language acquisition strategies. We also see that many of these data are directly relevant to the universal operating principles that have been suggested by Slobin (1973), MacWhinney (1978), and others.

1. DESCRIPTIVE SKETCH OF HUNGARIAN This sketch examines five major areas of grammar: phonology, morphology, morphophonology, ordering, and agreement. It uses as its basic reference the two-volume descriptive grammar assembled by the Hungarian Academy of Sciences (Tompa, 1970). Unfortunately, that grammar is only available in Hungarian. For those who do not read Hungarian, the only reference that is of any value at all is the textbook

240

Learn Hungarian by Bánhidi, Jókay, and Szabó (1965). This textbook provides a fairly good account of Hungarian morphology and phonology, but it is weak in the areas of morphophonology, ordering, and lexical structure.

1.1. Phonology There are at least three major types of phonological patterns: segmental patterns, phonotactic patterns, and morphophonological patterns. Here we briefly examine patterns of the first two types; the morphophonological patterns of Hungarian are considered in a later section. TABLE 14.1 Hungarian Consonants

241

1.1.1. Segmental Patterns The Hungarian sound system is summarized in Tables 14.1 to 14.3. The letters between slashes are International Phonetic Association symbols for each phoneme. The letters outside the slashes are the symbols of standard Hungarian orthography. Table 14.1 displays the system of 44 consonants; Table 14.2 summarizes the phonetic status of the 14 vowel phonemes (e and ë are dialect variants); and Table 14.3 displays the seven groupings relevant to the phonological processes of vowel harmony (explained below). Note that the phonetic basis of groups 6 and 7 is fairly opaque. TABLE 14.2 Hungarian Vowels

TABLE 14.3 Hungarian Vowel Harmony Groups Back Front

Front Rounded

1.

u

ü

2.

ú

ű

3.

o

4.

ó

e,ë

ö ő

242

5.

a

e

6.

á

é

7.

ja

i

1.1.2. Phonotactics Hungarian has several phonological rules that operate both within and between words. We will call such general phonological rules phonotactic rules. Phonotactic effects on vowels in Hungarian are quite different from those in Indo-European. Vowel sequences are never diphthongized except in some foreign words with /au/. However an epenthetic /j/ may be inserted into some vowel sequences. Thus fiú becomes fijú and fáért becomes fájért. This process is subject to stylistic, tempo, and dialect variation. Consonants show an allophonic accommodation of place to the following consonant or vowel. Thus /k/ is velar in kutya but palatal in kicsi and /n/ becomes labiodental /m/ in honfi. Voicing also assimilates regressively in consonant clusters. Thus kapd is pronounced kabd. This assimilation involves a series of allophonic alternations, since phonemes like /m/ that do not have a voiceless counterpart do not devoice. If a dental or palatal stop precedes a fricative, the stop and fricative merge into a long affricate that has the place of articulation of the affricate. Thus, hegycsucs becomes heccsucs and egyszer becomes etsser. Similarly, dental palatal stops followed by /j/ become long palatals. Thus hagyja becomes haggya and kutja becomes kuttya. Finally, long consonants shorten in clusters. Thus benn van becomes ben van and sarkkör becomes sarkör. 243

Vowel harmony operates both phonotactically and morphologically. Phonotactically, Hungarian words avoid the combination of any of the back vowels (u,ú,o,ó,a,á) with any of the front vowels (ö,ő,ü,ű, and e). However this pattern has numerous exceptions, particularly for recent loan words.

1.2. Morphology Hungarian morphemes can be divided into stems and affixes. In this section we review the most important types of affixes. The largest class of Hungarian affixes is the class of formative suffixes or kepzők. These are parallel to suffixes like -ate, -ness, -some, -al, and -ly in English. Like their English counterparts, most of these suffixes are fairly low in productivity. Nonetheless, as we will see in the section on child neologisms (particularly Table 5), there is a fair amount of evidence for some productivity for at least 25 formatives. There are also a few morphemes that are traditionally regarded as formatives, but which really act like productive inflections. These include the comparative suffix -bb, the superlative which uses the prefix leg-along with the suffix -bb, the hyperlative which adds the prefix leges-to the superlative, and the infinitive -ni. All of these “formatives” function much like inflections. There are 25 nominal case suffixes in Hungarian. However only about 15 of these can be freely combined with any noun in a fully inflectional way. These include the nine locatives (inessive, illative, elative, superessive, 244

sublative, and delative, adessive, ablative, allative), the dative, the instrumental, the accusative, the terminative, the causal-final and the translative-factitive. The sounds and meanings of these 18 case suffixes are as follows: 1. Inessive: stative position inside an enclosure, -ban, -ben. 2. Illative: moving toward toward a position inside an enclosure, -ba, -be. 3. Elative: moving away from a position inside an enclosure: -ból, -ből. 4. Superessive: stative position on a horizontal surface, -on, -en, -ön, -n. 5. Sublative: moving toward a position on a horizontal surface, -ra, -re. 6. Delative: moving away from a position on a horizontal surface, -ról, -ről. 7. Adessive: stative position at a point, -nál, -nél. 8. Ablative: moving toward a position at a point, -hoz, -hez, -höz. 9. Allative: moving away from a position at a point, -tól, -től. 10. Nominative: the unmarked form of the noun, the first mover (there is no passive in Hungarian).

245

11. Accusative: the direct object or element most affected by the action of the verb, -t, -ot, -et, -öt, -at. 12. Dative: the recipient or indirect object of the verb, -nak, -nek. 13. Instrumental/comitative: resembles English ‘with’, -val, -vel. 14. Terminative: movement up to an end -ig. 15. Causal-final: reason for an action, -ért. 16. Translative-factive: changing to a state, -vá, -vé. The other flectional suffixes in Hungarian are called “signs” or jelek. They include the plural, the possessives (marked on the noun possessed), and the verbal inflections. Plurality merges with the various possessive suffixes to yield a set of portmanteau forms. Thus, although -ok is the sign of the plural and -om is the sign of first person possession, the combination of plurality and first person singular possession yields not -okom but -aim. Also, when a possessive occurs with the accusative, the accusative may be optionally deleted. For both the possessive markers on the noun and the actor-agreement person markers on the verb, Hungarian has six basic persons—the same six as in English. In addition, when a first person singular acts on a second person there is a seventh type of verbal suffix (I - you). The personal suffixes have one set of six forms for

246

transitive verbs with definitive objects: -m, -d, -ja, -juk, -játok, and -ják and another set for intransitive verbs or transitive verbs with indefinite objects: -k, -sz, zero, -unk, -tok, and -nak. There are four basic inflected tense/ aspect/mood forms of the verb: present, past, conditional, and imperative, marked by zero, -t -na, and -j, respectively. The combination of the tenses with the person markers involves a series of irregularities suggesting that at least some of the combinations are portmanteaus. Nouns may be derived from stems that are themselves nouns or from other parts of speech such as verbs and adjectives. Whatever the source of the stem, the basic order of morphemes in the noun is as follows: 1. Hyperlative (leges-): must be used along with 2 and 5. 2. Superlative (leg-): must be used with 5. 3. Stem: a. Base (noun, verb, adjective, adverb) b. Compound c. Complex verb 4. Derivational formative: a. Denominative (-ven, -ék, -né, -ász, -onc -zat, -ista, -nok)

247

b. Non-productive nominalizer (-t, -aj, -ék, -alom, -ékony, -dalom, -mány, -vány, -dék, -lék, -tyú, -tyü, -óka) c. Productive nominalizer (-ó, -atlan, -as) 5. Comparative (-bb), numeric (-ad) 6. Adjectival nominalizer (-ik) 7. Substantivizer (-ság) 8. Adjectivalizer (-s) 9. Diminutives (various), adjectivalizer (-i) 10. New formatives (-féle, -beli, -szerű) 11. Plural, possessive, (6 different morphemes) 12. Signs of possessor (-é, -ei) 13. Case markers (26 different morphemes) The noun stem may appear by itself or with a derivational formative of position 4 above. The derivational suffixes noted under (4a) must attach directly to a simple verbal base which they nominalize. However, the derivational suffixes noted under (4c) can attach to a complex verb. Thus, a verb like vált-oz-tat-hat‘hange-PROG-POT’ (= ‘can be changed’) can add the nominalizer -atlan to form vált-oz-tat-hat-atlan (= ‘unchangeable”) which can be further inflected to

248

produce say, vált-oz-tat-hat-atlan-ság-om-nak (= ‘for my unchangeableness’). Complex verbs without nominal derivation may also take nominal inflections as in me-het-n-ék-em ‘go-POT-COND-lSG-my’ (= ‘my being able to go’) from the verb me-het-n-ék with the suffix -em ‘my’ added on. (See also nem-törödöm-ség ‘not-care-my-ness’ (= ‘my not caring’). Note also that even phrasal compounds with postpositions (see below) may serve as stems as in munkanélkül-i ‘job-less-characterizer’ (= ‘a jobless person’). The basic order of morphemes in the verb is as follows: 1. Separable prefix (igekötő—around 30) such as be-‘in’ or meg-‘completive’ 2. Stem: a. Base (noun, verb, adjective, adverb) b. Compound c. Complex noun 3. Derivational formative a. Deverbative (-ong, -g, -kod, -kol, -gál, -doz, -dokol, -an, -dogál, -int, -ámlik, -ul) b. Non-productive verbalizer (-ll-asz, -izál, -iroz, -fikál) c. Productive verbalizer (-ol, -oz, -ít, -ul)

249

4. Causative (-at, -tat) 5. Frequentative (-gat) 6. Passive (-ódik, -kódik -ozik, -kozik) 7. Potential (-hat) 8. Tense, mode, participles 9. Person (7 different morphemes) 10. Interrogative (-e) As in the case of nouns, the stem may appear by itself or with a derivational suffix of position 3. The derivational suffixes listed in (3a) must attach directly to a simple verb base. The derivational suffixes listed in (3b) attach to a simple noun base which they verbalize, and the derivational suffixes listed in (3c) can attach to a simple base or a complex noun. Thus, the denominative adjective nagy-ság-os ‘big-ness-ish’ (= ‘royal’) can add -oz to form nagy-ság-os-oz ‘big-ness-ish-act’ (= ‘to act like royalty’).

1.3. Morphophonology Hungarian has a large set of morphophonological rules which alter morphemes when they appear in combination. These rules are of two basic types: free and bound. Free phonological rules serve to modify the shape of a phonological feature or cluster of features in some general phonological environment. Bound

250

phonological rules or morphophonological rules modify the shape of phonological features on the basis of the presence of specific lexical items. Formal presentations of the most important morphophonological rules are given in MacWhinney (1978, pp. 21–26). Hungarian has five morphophonological rules that are so productive that they could almost be stated as free phonological rules. These five rules are: FRONTING HARMONY, ROUNDING HARMONY,

LINKING

VOWEL

INSERTION,

FINAL-A

The first two rules make suffix vowels agree with the fronting and rounding of the last vowel of the stem. Thus ház-ben becomes házban and kör-ok becomes körök. Whereas the effect of FRONTING HARMONY extends from the stem onward through any number of suffixes, the effect of ROUNDING HARMONY extends to only the first or second suffix after the stem. For patterns 1 to 5 of Table 14.3 these rules fully determine the shape of the suffix vowel. However, patterns 6 and 7 contain further irregularities. The third highly productive rule, LINKING VOWEL INSERTION, is a complex but highly general rule that breaks up clusters by inserting a vowel from vowel harmony pattern 3 (see MacWhinney, 1978, p. 24). For example, it takes ablak-t and converts it to ablakot, while leaving rizs-t unchanged. The two remaining highly productive rules are FINAL-A LENGHTENING and FINAL-ELENGTHENING. These rules convert words like pipa-k to pipdk and words like csésze-k to csészék by making both /a/ and /e/ lengthen before any suffix. LENGTHENING, and FINAL-ELENGTHENING.

251

Most of the rules of Hungarian phonology are clearly bound to specific lexical items. These rules do not apply generally and must be restricted to a particular set of morphemes in the lexicon. Bound rules can help determine the selection between the various allosegments of the allomorphs of a given morpheme. For example, INTERNAL VOWEL SHORTENING chooses between the /á/ and the /a/ and between zero and /a/ in the allomorphs maddr and madara of the morpheme ‘bird.’ Note that, although we may talk about selecting between allomorphs, the actual activational process is selecting between alternative segments or competing allosegments. Madár is the “citation form,” i.e. the form that appears in the nominative when no suffixes are attached. It is also the form that is used when the suffix begins with a consonant. When the suffix that follows madárlmadara has allomorphs which begin with a vowel, the rules choose the “oblique form” madara as in madara-m ‘bird + my’. This contrast between a citation from and an oblique form is quite general for Hungarian bases. Other rules that choose between oblique and citation forms include INTERNAL VOWEL DELETION by which tükör-ök becomes tükrök, V-INSERTION by which ló-t becomes lovat, J-INSERTION by which kabát-a becomes kabátja, V-ASSIMILATION by which házval becomes házzal, and VOWEL HARMONY for classes 6 and 7 by which olvas-i becomes olvassa. These rules are discussed in MacWhinney (1978, pp. 22-28). Other selections not discussed there include: 1. The choice between allomorphs with either -j or length. This applies to the imperative and three of the

252

present definite suffixes. It makes olvas-ja become olvassa and olvas-játok become olvassátok. 2. The choice between allomorphs ending in -t and those ending in -ss, -ccs, or -ssz before the imperative. This rule makes lát-j become láss and bomlaszt-j become bomlassz. Actually, this pattern utilizes three separate but parallel rules. 3. The choice of a linking vowel of pattern 5 as opposed to pattern 3 when the suffix is separated from the stem by at least one other inflectional suffix. This pattern makes mond-t-om become mond-t-am. 4. The choice between the second person singular verb suffixes -ol and -sz. The former is used with sibilant final verb and the latter is used elsewhere. 5. The choice between a series of verbs ending in either -sz or -ud, as in aludom which becomes alszom. (This group also uses V-INSERTION and INTERNAL VOWEL DELETION.) 6. A series of nonproductive rules selecting between allomorphs for a handful of highly irregular words. For example the choice between the stem allomorphs falu and falus ‘village’ is rule-governed, but the rule only applies to this noun. Another small class of nouns deletes its final vowel as in varj-a for varjú-a. Others change the shape of the vowel as in meze-je for mező-je.

253

7. The assimilation of the final z of the definite article az, ez to the consonant of the following suffix as in at-tól for az-tól. 8. The use of the az form of the definite article before nouns beginning with a vowel and the a form before nouns beginning with a consonant. The assignment of lexical stress in Hungarian is extremely simple and regular. Stress always falls on the first syllable of the word. Secondary stress peaks may also occur on the subsequent syllables numbered 3, 5, 7, 9, 11, etc. However, this post-initial iambic stress pattern is far less perceptible than the initial syllable primary stress pattern. Sentences also have characteristic stress patterns. Declaratives show a steady decline in pitch throughout the clause. Questions have a rise on the penultimate syllable followed by a drop on the final syllable. However, questions marked by the particle -e use declarative intonation. In any clause with a finite verb, the item before the verb receives the main sentential stress, particularly when it is pragmatically focused.

1.4. Ordering In this section the word “ordering” is used to refer to the system of rules governing the ordering of morphemes into words and sentences. The ordering of morphemes in Hungarian is governed by two types of rules: free rules and bound rules (MacWhinney, 1982). Bound rules are bound to particular morphemes in the lexicon whereas

254

free rules order morphemes on the basis of the roles they play in a given clause. The placement of affixes about a stem can be adequately controlled by a series of rules bound to each affix. These rules must specify two types of information. First, if the item is a prefix, the rule must specify that it precedes the stem. Similarly, if the item is a suffix, the rule must state that it follows the stem. In a language like Hungarian where there are often several suffixes after a stem, some of them derivational and some inflectional, it is also necessary to specify the order in which these suffixes are attached to the stem when more than one suffix appears. This can be done by associating a strength parameter with each affix. When several suffixes are lexicalized, they then enter into competition with one another (Bates & MacWhinney, 1982) and the ones with the highest strength index end up closest to the stem. For example, the plural always appears before the inessive in Hungarian. By giving the plural a higher strength index than the inessive, the child will produce the correct form ház-ak-ban ‘house-PL-INESS’ rather than the incorrectly ordered form ház-ban-ak ‘house-INESS-PL’. There are, of course, alternative ways of controlling affix order. For example, some affixes may be positioned in relation to other affixes as well as in relation to the stem. However, the placement of affixes by bound rules with strength indices offers a general solution to the task of affix ordering. An interesting problem is presented by the fact that case markers appear as suffixes after nouns but as prefixes 255

before personal pronouns. Thus, ‘to Emery’ is Imré-nek Imre-DAT’ while ‘to you’ is nek-ed ‘DAT-2SG’. One way of describing this alternation is to say that the meaning ‘DAT’ maps onto two different morphemes nek- and -nek. The former has the allomorphs nek- and nak- and the bound positional pattern: nek- + nucleus. The latter has the allomorphs -nek and -nak and the bound positional pattern: nucleus + -nek. Children seem to have particular difficulty acquiring these alternative morphemes, as we will see later. Within the noun phrase there is the following sequence of possible elements: 1. deictic determiner 2. postposition (same as in 15) 3. possessor (this may be a complex noun phrase) 4. dative marker on possessor (optional) 5. article (definite or indefinite) 6. locative modifier 7. complex phrase 8. quantity modifier (‘four’, ‘many’, etc.) 9. evaluative modifier (‘nice’, ‘useful’) 10. size modifier

256

11. color modifier 12. substance, provenance modifier 13. function modifier (this position may involve compounding) 14. head noun 15. postposition An example using each of these 15 positions is: ez alatta jó barátomnak az itteni sokszor használt három röeg nagy fekete gumi eső kabátja alatt ‘this under the good friend+my+DAT this the here many+times used three old big black rubber raincoats under’ (= ‘under these three often-used big old black rubber raincoats of my good friend’). Similar patterns are used to order strings of adverbs. In addition to these patterns for the placement of the operators on the noun, there are a series of patterns governing other minor sentence elements. These include: 1. FAMILY NAME + CHRISTIAN NAME: This is the opposite of the English pattern. Example: Bártók Béla (= ‘Béla Bártók’). 2. NEGATIVE + NEGATED: This applies to the negation of a single constituent. Example.ez nem alma ‘this not apple’ (= ‘this isn’t an apple’).

257

3. NEGATIVE + PERSON: Clausal negation places the negative before the verb that carries the person-number markers. Example: nem megy-ek ‘not go+1PS’ (= ‘I’m not going’). 4. FOCUS + PERSON-MARKED: The main information focus occurs before the verb that is inflected for person. (The word expressing sentential negation always receives primary focus.) Examples: én csirkét ettem T chicken+ACC ate+ 1PS’ (= I ate chicken’) neki mennie kell ‘DAT+3SG go+3SG must’ (= ‘he must go’) and el akar menni ‘away wants go + INF’ (= ‘he wants to go away’) where csirkét, mennie, and el are focused. 5. TOPIC + COMMENT: The topic precedes the comment. (But see rules 6 and 10.) Example: János # labdát rugott ‘John ball+ACC kick+PAST’ (= ‘John kicked the ball’) where János is the topic and labdát rugott is the comment. 6. COMMENT + AFTERTHOUGHT: Afterthoughts (Hyman, 1975) follow the comment. Example: Elment a János ‘away+went the John’ (= ‘John left’). 7. # + INTERACTIONAL: Interaction markers like ‘please’, exclamations, and vocatives are the first elements in a sentence. Example: Marcsi, gyere ide ‘Mary, come here’. 8. DIRECTIONALITY + PERSON-MARKED: The separable verbal prefix occurs before the verb that is inflected for person unless there is some other primary focus

258

positioned by rule 4. Example: Megettem a bogyót ‘Up+ate+lSG the berry+ACC (= T ate up the berry’). 9. IMPERATIVE + DIRECTIONALITY: The separable verbal prefix follows the imperative. 10. INTERROGATIVE + COMMENT: The interrogative precedes the comment but follows the topic (i.e., rule 5 takes precedence over rule 10). Example Eva hova ment? ‘Eva, where went?’ 11. ACTION + DEFINITE OBJECT: The definite object follows the verb, unless it is focused and placed before the verb by rule 4. Example: Jdnos megrugta a labdát ‘John PERF+kick+PAST the ball+ACC Indefinite objects, on the other hand are always considered to be focused and are ordered before the verb by rule 4 above. 12. MAIN + SUBORDINATE: The complements of modal verbs (which are marked for person) follow them. However focused complement verbs precede according to rule 4. 13. ACTION/PROCESS + x: Elements not ordered by the above strong rules follow the verb. 14. CONJOINER + CLAUSE: This is as in English. 15. TOPIC + ANTITHESIS: Some conjunctions like ‘however’ follow the topic, whereas in English they begin the clause.

259

16. RELATIVE English.

PRONOUN + RELATIVE CLAUSE:

This is as in

17. ACTION /PROCESS + VOLNA: The conditional acts like a verbal postposition. Relativized clauses always have a relative pronoun which almost always begins the clause in nonpoetic discourse. The relative is inflected to display its role in the subordinate clause. In many cases the relative ties in to a deictic head in the main cluase which is inflected to mark the role of the subordinate clause in the main clause. These two pronouns may assume any independent role as in azzal ehetek csak amit a kezembe adtál that+INSTR eat+POT+ 1SG just what +ACC the hand+lSG+INESS give+PAST+2SG. (= ‘I’ll be able to eat just whatever you put in my hand’). The deictic element is usually a deictic pronoun but may be a personal pronoun. In a few cases the head may be highly anaphoric and, hence, ellipsed. If there is a full noun, it may serve as the head with or without a deictic determiner. Because role marking occurs on two elements— one in the main clause and one in the relative clause—the positioning of the relative clause is remarkably free. Thus, the English sentence The dog pushes the cat that kisses the lion can also appear in Hungarian as The cat that kisses the lion, that one pushes the dog and The dog that the cat pushes kisses the lion can also appear as The cat by pushed dog kisses the lion. Many other orderings of each of these sentences are also possible.

260

Conjunction and complementation are often double-marked with an antecedent deictic pronoun in the main clause and a conjunction in the coordinate or subordinate clause. Thus, Hungarian has a large number of main clause-subordinate clause conjunction pairs such as addig—ameddig (= ‘from then—until then’), ahdny—annyi (= ‘so many—that many’), ott—ahol (= ‘there— where’), akkor—amikor (+ ‘then—when’), etc. There are at least 40 possible pairs of this type (Tompa, 1970, pp. 343, 354, 411). They give rise to sentences like Then I’m going to the store, when you come home or So many eggs I’ve got, as so many chickens. Here again, word order can be remarkably free. The main or nonconditionalized clause may occur either first or second. The conjunction of the main clause may be deleted or not. The only restriction is that the conjunction in the subordinate clause must always be present. Complementation follows similar rules. Verbs describing speech acts and mental acts usually take both the deictic accusative azt in the main clause and the relativizer hogy at the beginning of the complement. Thus, Hungarians say That I think, that he is going, for I think that he is going. The deictic pronoun can be omitted in some cases. In the case of factive or stative verbs like seem, appear, etc., az “it” is replaced by úgy ‘so’ or deleted altogether.

1.5. Agreement Hungarian has an exceptionally rich system of agreement marking. The ten most important patterns are: 261

1. Agreement of the noun with the number of the modifier so that a quantity modifier takes a singluar noun. (Quantity (Entity))- - ->((Quantity) (Singular (Entity))). Thus, Hungarians say ‘many tree’ rather than ‘many trees’. 2. Agreement of the verb with the number of the noun, so that a quantity modifier takes a singluar verb: ((Actor (Quantity (Entity))) (Action))- - - > ((Actor (Quantity (Entity)))(Singular (Action))). Thus, Hungarians say ‘many tree goes’ rather than ‘many trees go’. 3. Agreement of the verb with the number of the actor, so that a plural marker takes a plural verb. ((Actor (Number (Entity))) (Action))- - - > ((Actor (Number (Entity))) (Number (Action))). This is the same as in English. 4. Agreement of the topic with the number of the nominal comment in copulatives. ((Topic (Deixis ((Number)(Entity)))- - - >

(Entity)))((Comment

262

((Topic (Deixis (Number ((Number) (Entity)))).

(Entity))))(Comment

Thus, as we do in English, Hungarians say ‘those, doctors’ (= ‘those are doctors’) rather than ‘this, doctors’. 5. Agreement of the verb with the definiteness of the object. ((Object (Definite (Entity))) (Action))- - - > ((Object (Definite (Entity))) (Def (Action)). In the sentence meg-ett-em a hús-t ‘PERF-ate-lSG the meat-ACC’ the -em on megettem is in the definite conjugation because the direct object húst has a definite article. 6. Agreement of the negative with the imperative mode on the verb: (Neg (Imp (Act)))- - ->((Neg (Imp)) (Imp (Act))). Thus, imperatives like mutasd ‘show’ are negated by the negative imperative ne rather than the standard negative nem. 7. Agreement of the verb with the case marking of an associated locative phrase occurs whenever the locative is directional:

263

((Direction (Topology (Place))) (Act))- - - > ((Direction (Topology (Place))) (Direction (Topology (Act)))). Thus, Hungarians say ‘ I in+went the house+in’. 8. Agreement of the infinitive verb in the obligatory mode with the person and number of the actor: ((Actor (Person-Number (Entity))) (Oblig (Act)))- - - > ((Actor (Person-Number (Entity))) (Oblig (Person-Number (Act)))). Thus Hungarians say ‘ DAT-3SG must go-3SG.POSS’. 9. Agreement of the demonstrative adjective with the number and case of the noun: (Deixis (Role (Number (Entity))))- - - > ((Role (Number (Deixis))) (Role (Number (Entity))). Thus, Hungarians say az-ok-kal a fá-k-kal ‘that PL-COM the tree-PL-COM’ (= ‘with those trees’) 10. When the clause is negated, all major indefinite constituents must also be negated: (Neg ((Role (Indef (Entity))) (X) (Act)))- - -&CT (Neg (Role (Indef (Entity)))) (X) (Neg (Act)). Thus, Hungarians say ‘No one is not going nowhere to do nothing for no one’.

264

2. SOURCES OF DATA An extensive bibliography of Hungrian research up to 1972 can be found in MacWhinney (1974, pp. 803–812) and Slobin (1972, pp. 130–40). That bibliography lists 96 titles; in addition there are at least 30 further publications on the topic that have appeared since 1972. The Hungarian child language literature includes diary studies, phonetic analyses, vocabulary counts, reviews of the literature, studies of motherese, psycholinguistic investigations, sociolinguistic comparisons, research on the relation of cognition and action to language, and theoretical position papers. However, most of the basic data to be discussed in this present chapter are diary data on errors, neologisms, and the order of emergence of grammatical devices. Fortunately, most of these data are of relatively high quality. In fact, the work of Balassa, Kenyeres, and Meggyes compares favorably with the best diary literature in any language. Although some of the other sources of diary data are less complete and less sophisticated, the data themselves seem to be accurate and reliable. Wherever reference is made to the diary literature, the reader who is interested in examining the actual data should consult pages 220–605 in MacWhinney (1974). The present chapter focuses on data on the acquisition of the grammatical structure of Hungarian. Because almost all of the available data relate exclusively to the acquisition of expressive language, this chapter has almost nothing to say about the acquisition of receptive language. Furthermore, data on phonological 265

development are not considered except as they relate to those phonotactic processes that impinge on morphophonological development. In the area of morphophonology, the relevant data are summarized in my monograph on the acquisition of morphophonology (MacWhinney, 1978). The major findings of that monograph are summarized in the section below on “spellout.” In the section below on “ordering,” the analysis of early combinatorial patterns offered in MacWhinney (1975b, 1976) is extended to later, more complex sentence productions, following the analytic framework I proposed in MacWhinney (1982). Finally, the section below on lexical retrieval deals with the use of semantic structures in lexical retrieval. That section is based on data that can be found in chapters 5 and 7 of my dissertation (MacWhinney, 1974, pp. 338–457 and 509–605).

3. THE OVERALL COURSE OF DEVELOPMENT In this section I provide a brief informal characterization of the way in which Hungarian children pass through the major stages of language development. The reader should bear in mind that all of what I have to say here pertains exclusively to the acquisition of language by middle-class children in Budapest. On the most general level, Hungarian children learn language very much like children the world over. During the first year, they cry, babble, and coo. I have not been able to detect any significant differences between the babbling of Hungarian children and those of American children.

266

However, there are clear individual differences between Hungarian children, with some children babbling more than others, some earlier than others, and some more expressively than others. Some children show a sharp demarcation between the period of babbling and the period of the first words; others do not. Hungarian children are given a great deal of encouragement by their parents and by others to express themselves through vocalization and gesture early on. Adults express a fair amount of interest in these vocalizations of very young children, often overwhelming the small child with ebullience and attention. There is a set of typical things that the young language learner is expected to know. As in other countries, these include the child’s name, his age, the names of his siblings and so on. In addition, children are expected to know the sounds produced by the various farm animals, the numbers up to ten, the names of the colors, and the names of various foods. Hungarian peasant culture sanctioned a wide variety of baby-talk forms (see MacWhinney, 1974 for a summary), many of which have survived in the current largely urban/ suburban society. However, systematic, across-the-board baby-talk modifications of the phonology are now rarer than in the past. Children speak their first word at around 12 months. The earliest reported words include the Hungarian equivalents of words like ‘here’, ‘there’, ‘I’, ‘want’, ‘come’, ‘please’, ‘mother’, ‘father’, and so on. As has been noted for other languages, early words generally refer to family members, common animals, body parts, 267

general locations, household items, and so forth. During the period from 12 to 18 months children continue to add to their lexical inventory without producing many two-word utterances. Although Simonyi reports a two-word utterance at 9 months (ka pi, or katona pipál ‘soldier smokes’), most children produce no word combinations before 18 months. Because Hungarian is an agglutinative language, and because word order is so variable, the acquisition of individual words assumes a particularly central role in the child’s development of communicative competence. In fact, the clearest indicator of the level of development of a Hungarian child is not so much the length of the sentences s/he can produce as the morphological complexity of the words contained in those sentences. The model that will be developed in the body of this chapter holds that all grammatical patterns develop out of what are originally unanalyzed single lexical items. However, the child’s learning mechanism sets limits not only on the size and shape of unanalyzed lexical items, but also on the ways in which these unitary items can be subjected to subsequent analysis. Consider first that many of the words acquired during the “pre syntactic” period contain inflections. Thus, on a purely descriptive level, the claim of Stern and Stern (1907) that syntax or ordering precedes morphology is clearly wrong for Hungarian. However, it is difficult to demonstrate any productivity for these early inflections and I have argued (MacWhinney, 1975b) that many of these early forms are not morphological combinations, but rather “amalgams” or unanalyzed units. The exact 268

shape of early inflected forms supports this analysis. Thus, verbs are often learned in the second person singular imperative, body parts in the third person singular possessive, foods in the accusative, tools in the instrumental, and so on. I have often had it pointed out to me, by both linguists and nonlinguists alike, that even in their earliest words, Hungarian children almost never make errors in vowel harmony. The vowel harmony rules are fairly complex and one might be inclined to view this avoidance of error as a fairly remarkable accomplishment. Although the data currently available do not prove this conclusively, it appears that vowel harmony patterns are used both accurately and productively by 2;6. Errors in the most general vowel harmony rules seldom occur past 2;6 and the errors that occur in 3-year-olds are not errors in classifying the harmony type of the stem, but overgeneralizations of vowel harmony to create new suffix allomorphs. In general, children have no problem at all with the agglutinative nature of the Hungarian language. They are quite content to add strings of suffixes to stems. The problems that arise are not with the attachment of these suffixes, but with the irregularities governing the selection of allomorphic variants. At this earliest time, we find the largest numbers of gross semantic extensions and violations of part-of-speech category. This is also the time when phonological substitutions are rampant. In both phonology and semantics, children vary markedly in the precision of 269

their forms. Some children use a small set of words rigidly and accurately. Other children have such a mushy phonology that they can barely be understood. One of my subjects used the sound /i/ at 1;10 for ír ‘write’, én ‘me’, itt ‘here’ and inni ‘drink’. Another subject (Andi at 2;0) would on occasion speak long strings with errors of all types from which I could often decipher less than 60% of the material. Children enter the two-word period around age 1;6. Word-order errors are quite rare, even at the youngest ages. Affix-order errors seldom occur and errors in the ordering of major constituents are extremely difficult to detect because Hungarian allows so much flexibility in pragmatic ordering. I have reported (MacWhinney 1975a) that between 85% and 100% of the first utterances produced by Hungarian children can be generated by a set of simple rules specific to individual lexical items. Following Braine (1976), these were called POSITIONAL PATTERNS. For example, young children are quite systematic in ordering the adjective before the noun or suffixes after stems. In each case, there is an item which can be viewed as a “operator” being ordered with respect to another item which functions as the “nucleus.” Shortly after the appearance of these first pivot-like constructions, the child also begins to produce utterances using positional patterns that are not specific to single lexical items. The most important of these patterns are those which order the topic before the comment and the focused item before the verb. Of these, the latter pattern seems to come in first. However, it appears in a very 270

particular form in sentences in which the verb itself is initialized. This leads to the interesting consequence that children who are learning a language which is essentially verb-final use verb-initial (VSO, VOS, VO, VS) ordering in their earliest productions. Similar tendencies to verb-initialization appear in other languages in which the verb is not usually initial, such as German and Japanese. This early use of verb-initialization in non-verb-initial languages can be viewed as indicating a universal tendency in early child grammar toward fronting of the most active element in the sentence. The most important suffixes enter between 1 ;8 and 2;8. (The data on the order of emergence of these suffixes are given in Tables 14.9 and 14.10.) Hungarian provides an interesting test case for theories like that of Brown (1973) which attempt to predict the order of acquisition of grammatical morphemes on the basis of semantic complexity. In Hungarian the formal complexity of the various grammatical markers is held constant by the fact that nearly all of the major markers are suffixes. Moreover, all of these suffixes are subject to the same basic set of rules of vowel harmony and many are subject to the rules of linking-vowel insertion. This situation contrasts sharply with the one in English where some grammatical markers are articles, some suffixes, some auxiliaries, and some discontinuous morphemes. In order to disentangle the roles of formal complexity and semantic complexity in determining the order of acquisition of grammatical morphemes, an examination of the data on Hungarian acquisition could prove quite useful. My own interpretation of Tables 14.9 and 14.10 is that the primary determinant of the order of 271

acquisition is not semantic complexity, but semantic reliability and applicability. For example, I see no reason to believe that the conditional is more complex semantically than the infinitive. However, the infinitive enters long before the conditional. On the other hand, the infinitive in Hungarian is much more frequent than the conditional. In general, it seems to me that applicability is the primary determinant of the order apparent in Tables 14.9 and 14.10. Unfortunately, a more precise evaluation of this claim would require more detailed data on frequency in both the child’s speech and the speech directed to the child. However, when the new word frequency dictionary of the Hungarian language appears, some more reasonable estimates of this relation can be made. Another way of studying the emergence of grammatical functions is through experimental study of children’s elicited productions. A study by MacWhinney and Bates (1978) examined the use of six grammatical devices in expressing relative givenness and newness. The devices were: the definite article, the indefinite article, pronominalization, contrastive stress, ellipsis, and initialization. Child speakers of English, Hungarian, and Italian from ages 3 to 6 were asked to describe several series of pictures in which elements increased or decreased in givenness and newness. Both the Hungarians and the Italians made more use of initialization of normally non-initial elements than the Americans. However, contrary to the claims of functionalist sentence perspective linguistics (Firbas, 1964), initialization did not show any clear relation to givenness. Rather, it showed a weak relation to increased 272

newness. Most interestingly, initialization increased with age in Hungarian, indicating the acquisition of an ability to use alternative word orders. In all three languages, ellipsis was used to mark increased givenness. However, between ages 3 and 5, there was a general drop in the use of ellipsis in all three languages. This drop had a different shape in each of the languages. In Hungarian, 3-year-olds used particularly high levels of subject ellipsis, reflecting a tendency in colloquial Hungarian. By age 5, the Hungarians were still using more subject ellipsis than the Americans, but the size of this difference had decreased. There was very little use of pronouns by the Hungarians at any age. When pronouns were used, they tended to express increased newness or contrastivity, rather than increased givenness as in English. This contrastive use of pronouns is a basic difference between the English and Hungarian that seems to be present even at age 3. Finally, there was evidence that use of the definite article increases markedly between ages 3 and 5 in Hungarian. This rise is steeper than the rise in English during this period. This seems to reflect the fact that the definite article in Hungarian is used in a variety of contexts each of which must be acquired separately. I analyze (MacWhinney, 1984) these various contexts as polysemes of the definite article. Although definite articles are used to express increased givenness, this use alone cannot explain the extent of the increase of definite article use in Hungarian children. Rather, it appears that the definite article is also being used to express exophora, cataphora, metaphora, set operation, and 273

inference, as well as the basic function of anaphora or givenness. It may be that some devices serve functions that cannot be adequately studied in the context of static picture descriptions. Recently, I have collected descriptions of two simple films from Hungarian 3-year-olds, 5-year-olds, 10-year-olds, and adults. In these descriptions, the youngest Hungarians made much more extensive use of VSO and VOS word orders than the older subjects. One way of looking at the use of alternative word orders in Hungarian is to view them as by-products of the demands of the process of sentence production (Bock, 1982). In the case of the movie descriptions, where actions like “chasing” and “hitting” were repeated across scenes, it may have been easiest for the children to begin their descriptions with the verb. However this kind of explanation of variable word order use must be supported by evidence that a given order is somehow easier for a specific stimulus than some other order. Such evidence is not yet available. One other area of grammatical function that deserves study in its own right is the contrast between the definite and indefinite conjugations of the verb. This contrast involves a fairly complex set of interactions between semantic and syntactic patterns. Intransitive verbs are always indefinite, as are verbs whose objects are indefinite. However, for young children, the unmarked form of transitive verbs is not the indefinite, but the definite. This is because children spend most of their time talking about the here and now in which objects are usually definite. However, some verbs are more likely to 274

take indefinite objects than others. For example, when talking about eating, we often refer to the thing being eaten generically or indefinitely. However, Hungarian adds a further twist to this contrast by conjugating verbs like eszik ‘eat’ in the irregular -ikes conjugation. In this conjugation the definite replaces the indefinite in the first person singular. Further complexity is added by the fact that the second and first person pronouns are considered to be indefinite when functioning as objects. Possessed objects, on the other hand, are always definite. Despite these various complexities, and despite the frequency with which speakers must make a choice between the definite and indefinite conjugations, Hungarian children make few errors in choice of a conjugation after the age of around 3;6. Exactly how they learn to control this system is a question that has not yet been given adequate attention. Some of the last acquisitions of the Hungarian child are in the areas of morphophonology, case-marking, derivation, and lexical acquisition. In morphophonology, the most persistent errors are in the application of those morphophonological rules that apply to a small set of items. In section 6 these errors are discussed as violations of rules that are bound to specific lexical items. For example, an error such as tetője for teteje ‘my roof’ involves a rule that applies to only a handful of stems. Such errors are usually among the last in Hungarian acquisition. In case-marking, the child tends to confuse markings such as ‘from inside’ and ‘from on top of’ or ‘to the top of’ and ‘to the side of’. In many cases the selection of the correct case-marking depends on the specific cases governed by specific verbs. 275

Governance errors of this type are listed in Table 14.4 of section 7. They generally disappear by age 6. Other late errors involve item-specific limitations on the semantic range of the bases to which derivational affixes are attached. Errors of this type are discussed extensively in section 8. Late lexical acquisitions include temporal adverbs, certain coordinating conjunctions, and some of the postpositions. Data from one child on late acquisitions of this type are given in Table 14.7 in section 8. Because ordering is pragmatically variable, few of the child’s productions are clear ordering errors (compare Berman, 1985). However, incorrect agreement and improper use of conjunctions and negatives persist until around 6;0. In the data from movie descriptions that I have recently collected, the emergence of complex means of expression between the ages of 5 and 10 was particularly impressive. The use of relative clauses, tense shifting, and backgrounding is quite extensive in the 10-year-old sample. Hungarian 10-year-olds also made more extensive use of causative verbs and other types of verbal conflations than English-speaking 10-year-olds. Although the major grammatical devices all make their appearance before age 5, there is a continual growth in children’s ability to express shades of meaning up to at least age 10. A large proportion of city children enter nursery school at around 1 ;6 and then enter preschool at around 3;6. Both at home and in school, children are exposed to a rich array of highly structured verbal activities. It is my

276

general impression on the basis of 14 months of observation in Hungarian nursery schools and preschools that Hungarian children develop the basic lexical items at least as soon as or perhaps sooner than American children. Even at the youngest ages, most children show an active interest in verbal engagements with both adults and children. The nursery school curriculum places particular emphasis on learning through group participation in songs, games, and rhymes. Two unifying themes of the nursery school curriculum developed by the Ministries of Health and Education are that young children should be encouraged to develop their musical abilities through choral singing and their verbal fantasies through story-telling and story-learning. There are certain clear changes in the use of language in the nursery school between the ages of 1;6 and 2;6. In the second year, children use language mostly in its directive function. They attempt to request objects and protest against mistreatment. In the third year, other functions of language begin to blossom out. Children joke with one another, making funny sounds and pretending to babble. They use language to control coparticipation in games with toys and imaginary objects. Some children engage in extensive soliloquies, whispering to themselves as they play. By 3;0 a few children begin to tell short stories about trips to the zoo, the Danube, the country, and so on. Young children differ widely in the extent to which they use elaborate techniques of argumentation. Before 3;0, rhetorical structure has very little in the way of hierarchical control. However, between 3;0 and 6;0, there are major 277

advances in this area and most children learn to hold their own in argument.

4. THE COMPETITION MODEL In 1978, I presented a model of morphophonological acquisition (MacWhinney, 1978) called the “dialectic model.” That model relied heavily on both the basic logic and the specific proposals that Dan Slobin presented in his 1973 article on universals in language acquisition (Slobin, 1973). The current elaboration of the model also makes extensive use of the work that Slobin has done in the interim and which is reported upon in this volume (Slobin, 1985). Whereas the 1978 version of the dialectic model dealt only with morphophonology, the present version deals with grammar in general. In addition, the present version treats grammatical processing as a fully dynamic process based on interactive competition between items and patterns (Anderson, 1983, MacWhinney, Bates, & Kliegl, in press; Thibadeau, Just, & Carpenter, 1982). This processing model, called the “competition model,” is now integrated as a fundamental conponent of the overall model. Because of the increased role of competition in this second version, we now call the entire model the “competition model.” Note that the central role of the dialectic remains unchanged. The dialectic of error monitoring component of the model is fairly simple. It begins by adopting three basic terms from dialectic philosophy. These are the notions of a thesis, an antithesis, and a synthesis. A THESIS is

278

something that is produced by the child, while an ANTITHESIS is some piece of information that does not match the thesis. For example, a child may say the word *wifes and moments later remember that s/he should have said wives. In this case the thesis would be the initial form *wifes and the antithesis would be the second form wives. When a child encounters such an opposition between a thesis and an antithesis, s/he attempts to construct a SYNTHESIS to resolve the opposition. In the *wifes — wives example, the child attempts to decide under what conditions an /f/ can be altered to a /v/. The pattern governing this alteration is a new synthesis. In general, development can be viewed as the continuous construction of new syntheses to resolve disequilibria between theses and antitheses. Each of the three basic structures (i.e. thesis, antithesis, synthesis) can be associated with one of the basic processes of the model: APPLICATION, MONITORING, AND ACQUISITION. A thesis can be generated during either speaking (expressive application) or listening (receptive application). In expressive application, the speaker has an INTENTION which s/he converts into an EXPRESSION. In receptive application, the listener has an AUDITION which he converts into a RECEPTION. The expression is the thesis of expressive application and the reception is the thesis of receptive application. The next step in processing is the monitoring of the thesis. Monitoring checks the accuracy and adequacy of the thesis and provides the system with information on the extent to which its goals have been attained. In the current implementation of the model, the goals are: 279

1. to be fully “expressive” by converting all underlying semantic intentions into linguistic forms; 2. to be fully “receptive” by developing an understanding of input utterances that fully parses their content; 3. to be formally “accurate” by using linguistic devices in the same way that adults use these devices Monitoring that checks for the first goal is called EXPRESSIVE TALLYING; monitoring that checks for the second goal is called RECEPTIVE TALLYING; and monitoring that checks for the third goal can be either EXPRESSIVE CRITICISM or RECEPTIVE CRITICISM. When performing tallying the system is monitoring for impasse. When performing criticism, the system is monitoring for error. In some cases, monitoring may detect an antithesis to the thesis formed by application. When this occurs, we can say that an “impasse” or “error” has occurred and the system is in disequilibrium. Equilibrium can be restored by a process of acquisition which searches for a new synthesis to resolve the conflict between the thesis and the antithesis. This newly formed synthesis then may become a new thesis when it is used subsequently in application. Because every synthesis can eventually become a thesis, there is a continuing resolution of opposites which results in the cycle given below:

280

The central claim contained in this diagram is that, for a given piece of data, processing must follow the order that is indicated by the three arrows: application must follow acquisition; monitoring must follow application; and acquisition must follow monitoring. This basic hypothesis can be coupled with the fairly commonplace observation that sometimes the child is engaged in reception (listening) and that sometimes s/he is engaged in expression (talking). This means that there are actually two dialectic cycles:

These two loops represent the basic metaprinciple of the dialectic: METAPRINCIPLE: In both reception and expression, language processing operates in terms of the dialectic cycle of application, monitoring, and acquisition. The notion of a dialectic can be elaborated into a full model of language acquisition. Aspects of this

281

elaboration are discussed in MacWhinney (1978, 1982). For our present purposes, it is sufficient to examine the competition model in terms of the set of Operating Principles listed below. These Operating Principles lead to a series of predicted behavioral consequences for each of the three processes to be analyzed—spellout, ordering, and retrieval. This paper does not completely review the predictions of the model: the four Operating Principles which deal with monitoring will not be discussed since there are few data in the Hungarian literature that are relevant to their operation. Although much of the data supporting the model come from languages other than Hungarian, in this review we will focus on the evidence that comes from the Hungarian literature. As noted earlier, the data come from two major sources. The first is a compilation and summary of the findings of the Hungarian diary literature (MacWhinney, 1974, pp. 219–608). The second is a set of experiments in elicited word formation (MacWhinney, 1978). These two sources will be referred to as “the diary literature” and “MacWhinney (1978).”

BASIC OPERATING PRINCIPLES OF THE COMPETITION1 MODEL 1. ROTE: Speakers can use full retrieval of a lexical item as one possible way of activating items in both expression and reception.

282

2. ANALOGY: Speakers can use construction of an item on the basis of patterns implicit in the lexicon as one possible way of activating items in both expression and reception. 3. COMBINATION: Speakers use combination of lexical items as one possible way of activating items in both expression and reception. 4. PATTERNS: The units juxtaposed by combination may be subjected to the operation of three types of patterns: predispositions, free rules, and bound rules. 5. COMPETITION: Patterns vary in strength. When several patterns apply to the same data structure(s) they compete in terms of their strength. Patterns may either inhibit or facilitate other patterns. When all data sources are in agreement, decisions are maximally quick and confident. When data sources do not converge, decisions are slower. 6. EXPRESSIVE TALLYING: In expression children can “listen” to their own output. By doing this, they can check to see if what they have said fails to include a part of what they wanted to say. 7. RECEPTIVE TALLYING: Children can check to see if they do not understand a part of a string. 8. EXPRESSIVE CRITICISM: When children “listen” to their own output, the representation of a combinatorial form can facilitate retrieval of a weak rote item.

283

Children can then check to see if there is a discrepancy between the weak rote receptive form and the combinatorial expressive forms. 9. RECEPTIVE CRITICISM: In reception, children sometimes note a discrepancy between the sound of the correct forms which children near and they attempt to match these forms. 10. STRENGTHENING: Every time a rule or item applies successfully, it gains in strength. Every time a rule or item applies unsuccessfully, it loses strength. Losses in strength from incorrect application are greater than increments obtained during activation. Thus, rule reliability is more important than rule applicability. 11. AMALGAM ACQUISITION: Children acquire new lexical items by making a new association between a sound and a meaning. The clearer the representation of both the sound and the meaning, the earlier the acquisition. • Clarity is enhanced when clusters are presented by themselves or against a background of known information. • Phonological clustering is determined by intensity, pitch, and juncture. Recent segments and stressed segments are stored most clearly. • Semantic clustering is determined by propositional relatedness.

284

12. AMALGAM ANALYSIS: The child first attempts to analyze words into clear continuous morphemes. Analysis occurs primarily during parsing in receptive application. 13. INFERENCE: During parsing, the child can infer aspects of the semantics of new items on the basis of the words with which they are concatenated by ordering patterns. 14. MERGER: When two sound strings are synonymous but not homophonous, the mismatch can be resolved by weak allomorphy, or strong allomorphy. Weak allomorphy merges the two allomorphs by setting up an archisegment. Strong allomorphy establishes the allomorphs as an archifeature and merges the poly-semes. Strong polysemy establishes separate morphemes. Weak allomorphy is preferred to strong allomorphy. Similarly, when two meaning clusters are represented by homophonous sound strings but are not synonymous, the mismatch can be resolved by weak polysemy or strong polysemy. Weak polysemy establishes separate morphemes. Weak polysemy is preferred to strong polysemy. 15. RULE FORMATION: When the form produced by application contains an error and no rule has been applied, a new rule is hypothesized on both the free and bound levels. Rules must be formulated in terms of already available features and units.

285

5. COMPETITION IN THE LEXICON The competition between rote, analogy, combination, and patterns in the lexicon is governed by a set of principles derived from the study of interactive activation models of lexical processing (McClelland & Rumelhart, 1981) and principles of pattern-matching in production systems (Anderson, 1983). Thinking of both items and patterns as “rules,” such models state the following principles for activation: 1. Strength of rules: The strength of rules reflects the frequency and recency of their successful firing. Stronger rules receive more activation. To illustrate this, note that common irregulars like went tend to resist overregularization more than less common irregulars like sent. 2. Specificity: The matcher rewards rules for having features matched. Thus, in word recognition, cat is better than bat as a match to /k/-/ae/-/t/, since it matches three segments and bat only matches two. A special case of this is the superiority of portmanteau forms to their analytic counterparts. In French, both du and de + le compete for the masculine partitive. However, because du is more specific, it gets more activation. 3. Accuracy: The matcher penalizes rules for having too many features, i.e. for having features that are not active in working memory. Thus, when matching to /k/-/ ae/-/t/, the item bat will be penalized for the failure to match a /b/.

286

4. Data refractoriness: The pattern matcher attempts to assign each active element in working memory to a single rule. This means that if a particular elements matches more than one rule there will be an inhibitory relationship set up among these elements. This is what prevents multiple competing rules from applying to the same goal. As McClelland and Rumelhart (1981) note, this inhibitory relationship makes “the rich get richer and the poor get poorer” in that good guesses are supported and poor guesses eliminated. 5. Top-down support: If there is considerable activation of a particular action element, it will support the production patterns that led to it. As we will see, this can lead to the phenonemon of haplology. These principles govern pattern-matching and activation in the lexicon. In the current formulation, the lexicon is understood to contain three types of mappings: 1. Phonology: morpheme-to-segment mappings (e.g. the morpheme “dog” has the segments /d/-/aw/-/g/) 2. Semantics: morpheme-to-meaning mappings (e.g. the morpheme “dog” has the meanings [+animal], [+object], [+warm-blooded], [+furry], etc.) 3. Syntax: morpheme-to-syntactic-pattern mappings (e.g. the morpheme “the” precedes nouns, but can be separated from them by other material) Note that each of these systems is being given a highly lexicalist definition. Our phonology is lexicalist, as is our 287

syntax and semantics. MacWhinney and Sokolov (in press) explain the rationale behind this lexicalist perspective and give details on the representations. In this chapter we will consider Hungarian data on the acquisition of phonology, syntax, and semantics. As we examine each system, we will consider how the 16 Operating Principles of the dialectic model help us to understand the data on Hungarian acquisition, as well as relevant data on the acquisition of other languages.

6. APPLICATIONS OF THE MODEL TO THE DEVELOPMENT OF PHONOLOGY First, we will examine the data currently available on the acquisition of the Hungarian phonology. This system takes a set of morphemes that have been activated by the speaker and maps them onto a set of segments or sound units. Note that this conception of phonology is highly lexicalist and includes all of morphophonology within psychology. Apart from direct morpheme-to-segment mappings, segment activation is also governed by a series of patterns called MORPHOPHONOLOGICAL RULES. Morphophonological rules govern the shape of affixes and stems when they co-occur in words. As an example of a morphophonological rule, consider the pattern in English which alters the final /f/ of wife to a /v/ in the word wives. This change is not necessitated by the phonology of English—a plural such as fifes is perfectly good phonologically. Rather it is an allomorphic or morphophonological fact that we say wives rather than wifes. In the competition model, the morpheme “wife”

288

activates both /f/ and /v/ after the /al/. However, the morphophonological rule for retrogressive voicing assimilation adds additional activation to the /v/ and this segment dominates over the /f/ in the final output. Because English is essentially an analytic language, it presents few very good examples of clearly productive morphophonological alternations. In Hungarian, however, morphophonological processes are very numerous and productive. Moreover, unlike Turkish (see Aksu-Koç, & Slobin, 1975), the irregularity and complexity of Hungarian morphophonological processes present the child with a major acquisitional challenge.

6.1. Rote, Analogy, and Combination (Operating Principles 1, 2, and 3) Operating Principles 1,2, and 3 hold that children use rote, analogy, and combination as alternative processing strategies in both expression and reception.

6.1.1. Rote (Operating Principle 1) The general consequence of the use of rote (Operating Principle 1) is as follows. CONSEQUENCE PHONOLOGY 1A: Some affixes are used by some speakers in only a limited number of combinations, are not applied to new stems, and are not subject, at first, to morphophonological regularization. The familiar developmental pattern of (1) correct use of a few common forms, (2) overgeneralized use, and (3)

289

correct use of all forms (Ervin, 1964) is also well-documented in the Hungarian literature. For example, I found (MacWhinney, 1974, p. 653) that my subject Zoli at age 1;8 controlled 16 different verbs in the past tense, all of them correctly. At this age, usage was correct despite the complexity of the rules needed to form these words by combination. However, around age 2;0, errors in past tense formation started to increase, indicating that Zoli had ceased relying on rote and had begun to produce past tenses by combination. This developmental shift between rote and combination can therefore be viewed as one important source of the U-shaped learning curves (Strauss, 1981) so commonly observed in language development. I have shown (MacWhinney, 1975b; 1978, p. 80) that Hungarian children make fewer morphophonological errors on common irregular words than on rarer irregular words and that there are fewer errors on regular real words than on nonce words. Both of these effects indicate that children rely on rote to produce at least some inflected forms. Frequent forms can become well-memorized because they are heard so often. (Compare consequence spellout 10a.) Presumably, the strong receptive (auditory) image for these words serves to support the accuracy of the expressive (articulatory) form. There can be no such advantage for nonce words which, by definition, children have never heard before. The diary literature shows that, even before they have learned to put two words together, children use words containing unanalyzed affixes. Similar observations have been made for English by Brown, Cazden, and Bellugi 290

(1968, p. 41) who note for example that some children use can’t before they use can. These observations may be stated in a general form: CONSEQUENCE PHONOLOGY 1B: Stems may be acquired in an inflected form before they are acquired in their citation form, when the inflected form expresses a high frequency meaning (although usually the reverse pattern holds). In English, one could hold that words like can’t and don’t are not really inflected forms at all. The Hungarian examples are far less ambiguous, since it is clear that the relevant forms display bona fide inflections. For example, Kenyeres (1926) reports that his daughter Eva used the inflected Hungarian word kenyeret ‘bread +accusative’ at 16 months, even before she had produced her first two-word sentence. In fact, early inflected words often contain case markings typical for the noun. Thus, in early vocabularies, foods are often in the accusative, tools are often in the instrumental, body parts are often in the possessive, and locations are often in the locative (MacWhinney, 1974, p. 346). Similarly, activity verbs are often in the imperative, whereas state or process descriptions are often in the third person singular (compare Clancy, 1985). In MacWhinney (1975b, 1978), these rote pairings of stems with affixes are called amalgams.

6.1.2. Analogy (Operating Principle 2) Operating Principle 2 holds that the second major way of activating items in spellout is through analogy. Analogy

291

is defined as a process which creates new forms not through the combination of pieces but by extending patterns that are implicit in already existing rote lexical items. In MacWhinney (1978) this process was said to operate on the basis of a single exemplar. However, within the current framework of an interactive competition model, it makes more sense to think of analogy as gathering information simultaneously from all forms that resemble the target (Stemberger, 1982). It would be a mistake to try to draw too sharp line between analogy and combination. Rather, analogy is best understood as the initial phase in the process of rule formation. If an analogy is successful, it can soon become formalized as a rule (Operating Principle 15). If it is unsuccessful, it is not reinforced. In the interactive competition model, forms that serve as the bases of an analogy may be primed in several ways. Phonologically, they are primed or activated by their resemblance to the target. Semantically, they are primed if they contain components of the intended meaning. Experimentally, they can be activated by use of a priming technique developed by MacWhinney (1974) for children and Ohala (1974) for adults. Using this technique, subjects are asked first for the plural of, say, scarf. Presumably, they respond with scarves. They are then asked to form the plural of the nonce word narf. If they use scarf or words like it as the basis of the analogy, they should respond with narves as the plural. According to this line of reasoning, if analogy is in fact used to produce inflected words, the following consequence should hold:

292

CONSEQUENCE PHONOLOGY 2A: For some speakers on some forms, morphophonological priming produces increased numbers of morphophonologically analogous forms. Using the priming technique, I found (MacWhinney, 1975b, 1978) that, in Hungarian, analogy played a very minor role in morphological formations. In the period between 3;0 and 5;1, there was no significant effect of priming on analogical formations. In fact, in my 1975 study, many apparent analogies such as narves actually occurred directly after the child had produced a non-analogic real plural like scarfs. In my 1978 study, however, there was some evidence for occasional use of analogy by both the 2;6–2;9 group and the 6;8–7;5 group (ages which had not been included in the 1975 study), but analogy appeared to be used in two very different ways at these two ages. Two-year-olds relied on analogy because they had not yet acquired a reliable set of morphophonological patterns. Older children relied on analogy in those particular cases where rules had marginal productivity and the phonological shape of forms provided partial cues to the application of the rule (Köpcke & Zubin, in press). The strongest evidence for the operation of analogy occurs when a form is produced for which there is only one possible analog. This is the classical “proportional analogy” that is so often discussed in historical linguistics. If one assumes that proportional analogy requires a STRICT rhyme between the basis and the target, then the following consequence should hold: 293

CONSEQUENCE SPELLOUT 2B; Occasionally speakers produce forms for which there could be only one possible morphophonological analog. For example, one might expect that, on occasion, children would produce forms like draught, cf. correct drank on the basis of the analogy drink: draught:: think : thought. Only a few such single-possible-basis analogies have been reported in the child language literature. One is the form /fowt/ for fought produced on analogy with write/wrote that Bybee (1978, p. 42) observed in her son at age 5. Another is the form /tUø/ for tooth on the analogy /tUø/:: /teø/ :: /fUt/ : /fet/ noted by Ann Peters (personal communication) in her son at age 6;0. However, such single-possible-basis analogies may be more common than hitherto suspected. In South Serbian, the only analog for višel (past of vide ‘see’) was ide: išel (‘to go’). The fact that such a weakly supported analogy can become institutionalized suggests that many speakers may have produced the form višel independently.

6.1.3. Combination (Operating Principle 3) The third major way of activating items in spellout is combination. Operating Principle 3 holds that one way of producing expressive forms is by combination. Combination competes directly with rote and analogy in the context of lexicalization in the competition model. In morphophonology, combination works to build up a word out of two or more pieces. Early combination is blind to the rules of morphophonology—it allows

294

morphemes to combine without regard to their particular allomorphic shape. CONSEQUENCE PHONOLOGY 3A: Morphophonological errors include all possible combinations of allomorphs, even ones that cannot be derived from analogy.2 In practice, it is very hard to establish that a given error cannot be a result of extended analogy. However, if analogy is limited to proportional analogy, there are certain errors that cannot be attributed to analogical processing. I report a few such clear cases in MacWhinney (1974, 1978). The most common type involved a combination of a frequent stem morph like pingvin ‘penguin’ with a frequent stem allomorph like -k ‘plural’ to produce *pingvin-k ‘penguin-PL’ for the correct pingvin-ek. Since there is no real Hungarian plural that ends in -nk, this form could not have been produced by proportional morphophonological analogy. Similarly, at age 2;2,0, my subject Zoli produced *ló-unk ‘horse-ours’ as a combination of 1ó and unk, although word-internal sequences of /óu/ almost never occur in Hungarian. An even clearer case is the error *lov-n ‘horse-SUPER’ in which the allomorph lov ‘horse’ is juxtaposed to an allomorph of the superessive -n without a linking vowel. In this case, the final -vn is an error both allomorphically and phonotactically. Here, proportional analogy seems to be even more strongly precluded. However, I believe that, if analogy is formulated in a sufficiently dynamic fashion, it may be possible to generate errors like these. Such a formulation would allow analogy to extract implicit patterns that extend over large numbers of lexical items. It seems to me that 295

it is impossible to find particular errors that could not be generated by an extended analogy of this type. However, it remains to be seen whether extended analogy can also account for major developmental shifts in the types of forms produced by spellout.

6.2. Patterns (Operating Principle 4) As soon as the child begins to combine morphemes, s/he begins to acquire the rules that alter their shape. Operating Principle 4 holds that, when operating combinatorially, the child may use any of three types of patterns: predispositions, free patterns, and bound patterns. Let us examine the consequences of the use of each of these three types of patterns.

6.2.1. Predispositions The most primitive and general type of pattern is the phonotactic predisposition (Ingram, 1979). Phonological predispositions are conceived of as universal processes which favor certain types of assimilations and simplifications in the articulatory output. For example, the predisposition for vowel harmony would reduce /daedi/ to /daedae/. Stampe (1969) argues that such simplifications are unlearned consequences of the way we control our vocal apparatus. Unfortunately, in many cases, it is hard to separate the action of a phonotactic predisposition from that of an allomorphic rule. In fact, it is often the case that predispositions support learned rules.

296

CONSEQUENCE PHONOLOGY 4A: Forms for which the effects of a phonotactic predisposition summate with those of a learned allomorphic rule will appear earlier than forms using the same rule to which no predisposition applies. In Hungarian it is difficult to distinguish the predisposition to vowel harmony from the highly applicable fronting harmony rule. I found (MacWhinney, 1978) that at 2;6 children tended to block rounding harmony for the nonce form önyv and produce önyvek. The high percentage of önyvek productions were explained in terms of a predisposition to block rounding harmony after certain clusters. This analysis was based on the Academy grammar (Tompa, 1970) which states that words like könyv taking unrounded linking vowels (as in könyv-ek) form a natural phonological class in which there is a final consonant cluster that begins with a nasal or a liquid. However, a closer examination of the Hungarian lexicon suggests that no such class exists. Therefore, my explanation of the high percentages of use of önyvek by even 3-year-olds may have to be reconsidered. It may be the case (Matthew Rispoli, personal communication) that önyvek is being produced on analogy with könyvek. These two accounts could conceivably be tested by using nonce stems like rüvnty that have no close analog in the lexicon. If önyvek is actually being formed by analogy with könyvek, we would expect Hungarian children to produce önyvek as the plural of önyv much more than r üntyek as the plural of rünty.

297

6.2.2. Free patterns The second type of allomorphic pattern is the free pattern. These rules apply generally whenever a given set of phonological features appears in the correct phonological environment. If such rules exist, the following consequences should hold: CONSEQUENCE PHONOLOGY 4B: At least some learned rules apply obligatorily in a given phonological context without regard to lexical or allomorphic facts. This type of rule can be typified by the Hungarian rule of final vowel lengthening that was discussed above. This rule alters stem final /a/ to /a:/ and /e/ to /e:/ before suffixes. Because there is no word-internal tendency to lengthen prefinal /a/ (malac is a good Hungarian word), this pattern cannot be simply a predisposition. However, like the pattern of fronting harmony, this pattern is acquired well before age 2;0. In fact, I reported (MacWhinney, 1978) a correct generalization of final vowel lengthening to a nonce stem at age 1;8,6. This remarkably early control of an allomorphic rule is perhaps the earliest demonstration available in the child language literature of the productivity of a morphophonological rule. In effect, the child has learned this rule not as a (bound) morphophonological rule, but as a (free) phonological rule. Another example of morphological rule with very high productivity is Hungarian fronting harmony. This rule applies to morphological combinations in order to

298

convert all suffix vowels into back vowels when the stem has back vowels and into front vowels when the stem has front vowels. The rule applies to most every combinatorial morphological formation in the language and has only a few exceptions. The major exceptions are a few suffixes such as -nek ‘first person singular conditional’, -ik ‘third person plural definite’, and -i ‘third person singular definite’ which have only one front vowel allomorph. However, in testimony to the extreme productivity of fronting harmony and in support of Consequence spellout 4b, we find reports of errors such as *-n ák, for -nék, *-jek for -ik, and *-je for -i in which the “missing” harmonizing allomorphs have been “created” in line with harmony patterns 6 and 7. In fact, there is even one report of the extension of vowel harmony to the postposition is ‘also’ to form os as in *mama os ‘mama also’ for the correct form mama is.

6.2.3. Bound rules The third type of pattern is the bound rule. Morphophonological rules are generally bound rules. Bound rules are limited in generality to precisely those morphemes that illustrate a given alternation. For example, the wife-wives alternation in English must be controlled by a rule that applies only to words that exhibit the /f/—/v/ alternation since it cannot apply to fife or sheriff. Because English has such a limited set of affixes, bound rules perform little work in our language. However, in Hungarian, the same alternation may occur with up to 26 case suffixes. Therefore, bound rules can function as important and useful predictors of allomorphic form. In order for a bound rule to apply to a

299

new stem, the alternative segments of the allomorphs of that stem must be acquired. This fact lead to the following consequence: CONSEQUENCE PHONOLOGY 4C: Bound rules only apply to new morphemes once the allosegments activated by that rule have been acquired. Experiment 2 in MacWhinney (1978) examined the application of the internal vowel deletion selection to nonce stems. This rule takes bokor-ok and converts it to bokrok. To use this rule, the subject must know that bokor has the allomorph bokr. He can then produce bokrok, bokrot, bokros, bokrod, and so on. Children were taught two allomorphs of each new nonce stem and then were asked to use these allomorphs in a new case. Five-year-olds showed some ability to apply selections in a productive way. However, the results of the study were partly equivocal and further research on this topic is needed (see Bybee & Pardo, 1980). The principle of top-down support in interactive activation leads the pattern matcher to accept rote output that contains a match to an affix as if it actually contained that affix. CONSEQUENCE PHONOLOGY 4D: Partial regularity will facilitate morpheme recognition and the production of derived forms. Bybee and Slobin (1982) have found support for this consequence in regard to English past tense formations. 300

Menn and MacWhinney (in preparation) discuss this process in some detail, reviewing further evidence in support of this consequence.

6.3. Competition (Operating Principle 5) In phonological processing, both segments and morphemes compete for slots in the output. When allomorphs do not compete for the same position in a word, the child must learn that only one allomorph of a set may be used at a time. This means that when one allomorph is selected, the others must be inhibited. While the child is learning to do this, errors of overmarking are quite frequent. Some of these overmarkings may involve placement of a combinatorial suffix onto a correctly inflected rote form, as in shoeses. However, errors such as footses point more clearly to allomorph competition, since both plural suffixes are probably being used productively. CONSEQUENCE PHONOLOGY 5A: When children already know one allomorph of a morpheme and then acquire another, for a brief time they may use the new allomorph along with the old to produce overmarkings. This will occur primarily when the two allomorphs compete for at least partially different syllable-structure positions. This type of error is illustrated in English by forms like * shoeses and *hopeded. Hungarian errors such as *pingvin-k-ek and *lo-k-ak (MacWhinney, 1975b) also provide good evidence for this consequence. In these errors pingvin ‘penguin’ and ló ‘horse’ are the bases. The forms *pingvink and *lók are incorrect plurals and the 301

forms *pingvinkek and *lókak are overmarked plurals. Since the single plurals are also errors, rote is probably excluded and it thus appears that two plural allomorphs have been combinatorially attached to the bases. Note that the allomorphs -k and -ok are not in headlong competition, since -k is targeted for the syllable coda and -ok is targeted for positioning as a separate syllable. There are reports of 42 such reduplications in the Hungarian literature (MacWhinney, 1974, pp. 348 –352, see also MacWhinney, 1976, pp. 400–401 and Aksu-Koç. & Slobin, 1985). These reports include even more complex reduplications such as *rámomra ‘to+me+me+to’ in which both the sublative (-ra) and the first person singular (-om) are reduplicated. The current formulation of the process of competition views affixes as attempting to open up a slot vis-à-vis some stem. Generally, the affix which the speaker is attempting to order is related to the stem s/he is trying to produce. However, it may happen that more than one affix is active at a given time. In such cases, deciding which affix goes with which stem can become a problem. CONSEQUENCE PHONOLOGY 5B: When several semantically-unrelated affixes become activated at the same time, they may be attached to the wrong stems. In Hungarian, transitive verbs have case frames that can activate two, three, and even four case suffixes. Occasionally, children attach the wrong suffix to the wrong noun. Similar errors can also be observed in adults (see Stemberger, 1982, for English data). In her 302

daughter at age 2, Meggyes (1971, p. 50) reports several suffix-anticipatory substitutions of the form: NOUN A—SUFFIX A + NOUN B—SUFFIX B - - -> NOUN A—SUFFIX B + NOUN B—SUFFIX B. Meixner (1971) reports one suffix metathesis of the shape: NOUN A—SUFFIX A + NOUN B—SUFFIX B - - -> NOUN A—SUFFIX B+NOUN B—SUFFIX A. The suffixes in these errors are case markers. Such errors could be given a purely phonological interpretation. By this account, suffix metathesis would be just like Napa Valley - -> Napey Valla. Stemberger (1982) argues against the phonological interpretation of such errors, pointing out that adult English transpositions seldom involve whole syllables, whereas affix transpositions often do. Meggyes reports several other sentences indicating that the various affixes activated by a verb may be competing against each other for suffix slots. In these sentences, a surface case form seems to have come from a noun that never reached the surface. Thus, építem a Sömpikét ‘build+I the Sömpike +ACC Sömpike is Meggyes’ daughter’s nickname was used when the required for was építem a (Sömpikéenek (a házát) ‘build+I the Sömpike+for (the house+ ACC)’ It may be that each verb automatically activates a set of case suffixes. If a given noun role is not lexicalized, its suffix may become attached to another noun. In Japanese, Clancy (1985) reports similar errors in which adpositions are reversed about nouns. Errors of this type appear to be no more frequent in child language than in adult language and may well be interpreted as processing errors. However, it is not clear how one can really distinguish performance errors from competence errors. In the interactive activation model, no strict

303

separation is being made between competence errors and performance errors. Activation is the primary determinant of both correct production and erroneous production and activation may be a function of either long-term learning in the child or the current state of the system. Competition may also arise between forms that are spelled out by rote and forms that are spelled out by combination. In processing terms, these two modes of activation are viewed as operating in parallel. Both may achieve some activation of output phonological segments. However, the one that achieves the strongest activation of output segments is the one that “wins” the competition. In general, strong rote forms should dominate in this competition, because they are highly automatized ways of fully spelling out a meaning. If rote forms are not fully strong and automatized, combination emerges as a viable alternative. Thus, highly frequent words should be produced by rote, whereas less frequent words are more likely to be produced by combination. CONSEQUENCE PHONOLOGY 5C: At a given age, common irregular forms are subject to a smaller percentage of morphophonological overregularizations than infrequent irregular forms. Overall, there are fewer errors on common words than on rare words. As was noted earlier, I have shown (MacWhinney, 1975b; 1978, p. 80) that this consequence holds for Hungarian. Thus, overgeneralizations of the regular plural to frequent words like Ió ‘horse’ (i.e. *lók ‘horse+PL’ for lovak) are proprotionally less frequent 304

than overgeneralizations of the regular plural to infrequent words like darú ‘crane’ (i.e. *darúk for darvak). This consequence has also been supported for English (Graves & Koziol, 1971), Arabic (Omar, 1974), and German (Walter, 1975; MacWhinney, 1978). Usually, combination and analogy generate the same output. In such cases, when they lead to similar output, they work in concert to increase the activation of forms. However, competition can occasionally emerge between the forms spelled out by combination and those spelled out by analogy. In such cases, combination dominates quite strongly over analogy whenever there are strong rules governing allomorph selection. However, when the rules are not strong, analogy is as good a bet as combination. CONSEQUENCE PHONOLOGY 5D: When the rules governing a certain type of combination are opaque and full of exceptions, children and adults make comparatively more use of analogy. This consequence is supported by data from Hungarian (MacWhinney, 1978, p. 34), German (MacWhinney 1978, p. 69) and Russian (Zakharova, 1958). In Hungarian, I found that 2-year-olds were particularly susceptible to the priming techniques described above for narf and scarf. However, 6-year-olds continued to be sensitive to primes for minor rules.

305

6.4. Rule Acquisition (Operating Principles 10 and 15) The tenth Operating Principle of the dialectic model holds that every time an item or pattern applies successfully (i.e. does not lead to an error), it gains in strength. Note that “success” and “error” are defined here entirely by the operation of the process of monitoring. When monitoring detects no mismatch between the child’s system and data taken from the outside, then application is judged to be successful. Any rule that has applied without the detection of an error is strengthened. Rules that are highly APPLICABLE will soon become strong rules. However, rules must also be entirely RELIABLE. If rules lead to mismatches that are detected by monitoring, they will be significantly weakened. In general, this means that the child will attempt to acquire the most applicable rule that is also correct. When a rule or form is newly learned, its applicability leads to some early successes. These successes lead the child to apply it wherever possible. Thus, goed is used along with jumped and wanted by English-speaking children and ouvrié is used instead of ouverte by French children (Clark, 1985). However, after a while, incorrectness catches up with applicability and these overgeneral rules are reined back. The basic Darwinian3 principle expressed in Operating Principle 10 has a series of important consequences. The first consequence is identical to consequence spellout 5c above.

306

CONSEQUENCE PHONOLOGY 10A: Common irregular forms are subject to a smaller percentage of morphophonological overregularizations than infrequent irregular forms. Of course, common forms occur more often than uncommon forms and therefore have more errors overall. But, as Kuczaj (1978) has shown for English past tense marking, the percentage of errors declines for more common forms before it declines for less common forms. In Hungarian, I found (MacWhinney, 1975b) better performance on the plural of the common noun Ió ‘horse’ than on the rarer noun daru ‘crane’. Presumably the child learns the irregular plural lovak ‘horses’ so well that it almost always wins out over the regularized combinatorial form *lok. In the case of darvak ‘cranes’ the child has only a weak rote form and combination ends up producing *daruk. Similar results are reported by Berman (1985) for Hebrew. The second major consequence of Operating Principle 10 relates to the relative strengths of the different allomorphs of a given morpheme. CONSEQUENCE PHONOLOGY 10B: The first productive uses and the first over-generalizations of an affix will make use of the allomorph that is most applicable (across types). Here a distinction is being made between type frequency and token frequency. The English past tense alternation found in sing/sang is high in token frequency, because the verbs sing and ring are high frequency items. 307

However, the pattern applies to few types, since it only operates on about seven verbs in English. Note that Operating Principle 10 only strengthens the allomorph of an affix after combination has applied. When a form is produced by rote, the affix is not applied as a separate item and, hence, cannot be strengthened. This means that strengthening of weak alternative segments of allomorphs will occur only when forms are produced by combination rather than by rote. Since high token frequency items are likely to be produced by rote, the affix allomorphs used in their formation will not always be used. This leads to the surprising consequence that affixes should be more likely to be used for forms whose frequency is somewhat lower. Thus the first productive allomorph should be the one that combines with a large variety of stem types whose individual token frequency is not too high. Reflecting this, Consequence phonology 10b defines applicability in terms of types. However, it is important to note that the basic definition of applicability is more general. If a morphophonological rule applies in 20% of the words produced by a speaker, its applicability is twice that of rule that applies in 10% of the cases. However, in order to estimate the likelihood of a rule having applied, we must also be able to estimate the likelihood that the form was produced by rote. In those cases where rote applied, the rule itself was not applied and applicability is lower. As Clark (1985) notes, Guillaume’s (1927) data on the overgeneralization of French verbal suffixes indicate that the more applicable first conjugation suffixes dominate over the more frequent (in tokens) second and third conjugation suffixes. These data indicate that the number 308

of the types with which an allomorph combines is probably more important than the simple frequency of its potential appearances in determining allomorph strength. This interpretation has also received experimental support in a recent miniature linguistic system study by Lederberg and Maratsos (1980). In most cases the allomorph that is most frequent in terms of the types in which in appears is also the most frequent in terms of the tokens in which it appears. In such cases, support for Consequence phonology 10b is fairly straightforward. Such support has come from Arabic, English, French, German, Latvian, Russian, and Spanish. Since the writing of the review of these data in MacWhinney (1978), additional support for this consequence has been reported for Quiche (Pye, 1979), Dutch (-en suffix, Snow, Smith, & Hoefnagel-Hohle, 1980), German (verb stems and plural and adjectival affixes, Mills, 1985), Hebrew (plural and feminine suffixes, Berman, 1985, Levy, 1979), and Polish (Smoczyńska, 1985). For example, Smoczyńska notes that, in Polish, the earliest genitive is -a, the first nonfeminine prepositional is -u, and the earliest first person present is -m. In each case, the earliest allomorph is also the most frequent. In French, Clark (1985) finds that both que and qui function interchangeably as relative pronouns. To the child, que and qui may at first appear to be allomorphs. Since qui is the most frequent of the two, it is the one first overgeneralized. Of course, qui and que are not allomorphs, but different morphemes each using different ordering frames. However, the child may not realize this at first. 309

The Hungarian data also support Consequence phonology 10b. First, consider a small group of Hungarian suffixes -m, -t, -k, -s, -d, and -n which insert either /o/ or /a/ when following back vowel stems. Thus, ház + m becomes házam ‘house + my’, but hug + m becomes húgom Iittle:sister + my’. Most stems take /o/, but a significant number of high frequency, older words take instead /a/. Both /o/ and /a/ are low-vowels in regard to harmony (see Table 3) and selection between the two is not a vowel-harmony issue. The diary data show that errors such as házom are far more frequent than errors such as *húgam. In other words, applicable allomorphs such as -om and -ok are more likely to be overgeneralized than infrequent allomorphs like -am and -ak. Second, consider the allative suffix (-hoz, -hez, -höz) which presents another example of the importance of allomorph frequency. The only back vowel allative allomorph is -hoz, whereas front vowel words can take either -hez or -höz depending on rounding harmony. Because back vowel words are about as numerous as front vowel words, adults use -hoz about as frequently as -hez and -höz combined. In fact, I found (MacWhinney, 1978) that, in the earlier periods, many children used -hoz as their only allative allomorph. Here again frequency in the adult language seems to be mirrored in child usage. On the other hand, in cases where two allomorphs are of roughly equal frequency, both are overgeneralizations of front vowel allomorphs (*bokor-ben ‘bush-in’ for bokor-ban) are about as frequent as overgeneralizations of back vowel allomorphs (*szék-on ‘chair-SUPER’ for szék-en).

310

In a few cases, the early overextension of a strong allomorph may become extremely marked. For example, Clark (1985) cites a case in which a Spanish 2-year-old used the feminine article la almost exclusively, even when masculine el was required. In such cases, it may be that the child has simply not yet acquired el as a free form. Once el is acquired, the disproportionate use of la should taper off. Consequences phonology 10a and 10b have also received support of a somewhat different and interesting type. Bybee (1980), Bybee and Brewer (1980), and Zager (1980) have examined in some detail the effects of frequency and arkedness on leveling between and within paradigms in language change. Their results indicate that Consequences phonology 10a and phonology 10b may hold not only for child language but also for language change. There are at least four additional consequences of Operating Principle 10. These further consequences result from the interaction of the two parts of Operating Principle 10 when considered in relation to the framework of Operating Principles 1, 2, 3, and 4. The first part of Principle 10 holds that rules will gain in strength when they are highly APPLICABLE across form types. The second part holds that rules will lose strength if they are not also RELIABLE in their application. Therefore, the strongest rules are those that maximize both applicability and reliability. However, reliability cannot be sacrificed for applicability. All rules must be essentially reliable, and the problem is to find 311

the most applicable rules that are also reliable (MacWhinney, Pléh, & Bates, in press). The principle operates so that more applicable rules are acquired before less applicable ones. CONSEQUENCE PHONOLOGY 10C: Rules are acquired in order of applicability. This consequence has been supported by data from Finnish and German (MacWhinney, 1978, p. 82) as well as by my analysis (MacWhinney, 1978) of the order of acquisition of 15 Hungarian rules. In that study, the relative applicability of each of the 15 rules was calculated by determining the number of words (types) to which it might apply. The reliability of each rule was found by examining the relative proportion of words (types) to which it applied correctly. The free rules were all higher in applicability than the bound rules, and most of them were acquired before the bound rules. In general, applicability correlated with order of emergence at 75. However, this correlation was not perfect. When reliability was also considered, the correlation rose to 1.00. These results show quite clearly that both applicability and reliability are important determinants of the order of acquisition of morphophonological rules. Berman (1985) cites an interesting example of the results of Consequence phonology 10c for Hebrew. Words like simla ‘dress’ undergo two alternations in forming plurals like smalot. One alternation changes final /a/ to /ot/. This alternation is extremely general and is acquired early as a modification. However, the change of /iC/ to /Ca/ is far 312

less general and learned much later and with a fair amount of difficulty as a bound rule. Some bound rules apply to only a handful of items or perhaps only one item. The extremely late acquisition of these patterns in Hungarian supports Consequence spellout 10d. For example, at age 8;2, Eva Kenyeres produced the error *szavat for szót ‘word+ACC’. This error is an overgeneralization for the rule which chooses between allomorphs like szó- and szav- or hó- and hav-. The only exception to this selection is szót which must be learned by rote. The fact that the error *szavat is not reported earlier, indicates that the bound rule probably is learned quite late. Another bound rule which still causes problems in school-age children is the rule that changes /ö/ to /ej/ in the possessive. There are only four stems that could use such a selection. Similarly, the various rules selecting between the allomorphs ev- and ett-, falus- and falu-, ul and -an, -ott and -on, and szár- and száraz- are extremely limited in applicability and errors in their use continue until at least age 4 (MacWhinney, 1974, pp. 391–397). As was noted above, children will eventually abandon any rule that is not correct. However, as Slobin (1973) has noted in the discussion of his Operating Principle F (“avoid exceptions”), it may take them a while to realize that some overly general formulation can occasionally lead to errors. The fact that reliability exerts its force only in the long run after at least a few errors have occurred leads to this consequence:

313

CONSEQUENCE PHONOLOGY 10D: Rules that apply quite generally, but which are actually bound rules, are acquired initially in an overgeneral form. Consequence phonology 10d is well-supported for Hungarian (MacWhinney, 1978, pp. 36, 37, and 42). Thus, INTERNAL VOWEL DELETION, INTERNAL VOWEL LENGTHENING, and V-ASSIMILATION are all acquired initially in an overgeneralized shape. Formulations of bound morphophonological rules as free rules are also reported by Levy (1979) for Hebrew, where final /a/ is taken to be a general predictor of selection of /ot/ as a plural. Note that in all these cases there remains some tendency to treat the rule as free even after the initial period of over-generalization is checked. Operating Principle 10 feeds into Operating Principle 15. Operating Principle 15 governs initial formulation of rules on the basis of errors detected by monitoring or on the basis of the construction of an analogy. If the analogy is to be stored or if the error is to be eliminated, a rule must be articulated. Operating Principle 15 holds that the child will formulate the rule in both a bound and a free version. Consequences spellout 1 If held that bound rules may be initially applied as free rules. However, this does not mean that the child has not also formulated the pattern as a weaker) bound rule. Since both types of rules continue to be present, the weakening of one simply leads to “growing room” for the other. CONSEQUENCE PHONOLOGY 15A: When free rules are eliminated as incorrect, bound rules remain in force.

314

Thus, when Hungarian children stop overgeneralizing internal vowel deletion as in *motrok for motorok, they do not then fail to correctly form bokrok instead of *bokorok (MacWhinney, 1978). Although children try to formulate rules in a maximally general form, there are definite limits on the kinds of hypotheses they tend to entertain. In particular, children must first formulate rules in terms of units that are already available. Thus, it makes sense to formulate a rule for, say, front vowels or for dental stops. It makes far less sense to formulate a rule in which, say, dental clicks, lateral fricatives, and front rounded nasal vowels are treated as a single class. CONSEQUENCE PHONOLOGY 15B: Modifications are first formulated in terms of characterizable features in a simple way by phonetic parameters. I found (MacWhinney, 1978, p. 34) that there was a gap of almost 4 years between the times of attainment of similar levels of control on the two parts of the supposedly unitary rule of final vowel lengthening. Children acquired the /a/ to /a:/ alteration by around 2;6 but still had trouble on /e/ to e: at 6;8 to 7;5. It is clear that these two alterations are in no sense a single rule. The fact that these superficially parallel processes are not acquired in parallel is not too surprising when one considers the phonetic facts: /a/ is low, back, short, and lip-spread; /a:/ is middle, back, long, and somewhat tense; /e/ is low, middle, mid-to-front, short; and /e/ is mid-to-high, front, long, and very tense. The ratio /a/ : /a:/ :: /e/ : /e:/ is very abstract indeed and there is no 315

evidence that preschoolers ever treat /a/ and /e/ as members of a single class. Thus, it appears that children follow some commonsense principle of phonological concreteness, defining morphophonological rules in terms of phonetic parameters.

6.5. Item Acquisition (Operating Principle 11) The Hungarian literature is also rich in information on the determinants of the order of acquisition of lexical items. Operating Principle 11 holds that the child’s first conventional lexical items are formed by the association of a single intonationally-delimited phonological package with a package of semantic intentions the child wants to express. This is to say that the first rote items acquired by a child involve a minimum of either phonological or semantic analysis. The child associates an intonationally prominent and coherent piece of the speech stream with a salient and coherent subset of the semantics of the ongoing situational framework. One consequence of this principle is as follows: CONSEQUENCE PHONOLOGY 11A: Affixes are present in a morphophonological correct form before being used productively. In English, Brown (1973) found that grammatical morphemes are used in a semantically accurate fashion several weeks or months before their use becomes fully productive. I report (MacWhinney, 1974) a similar pattern of acquisition for affixes in Hungarian. Thus, at age 2;2, my subject ZoIi used a large number

316

of correct past tense verbs without showing evidence of productive use of the past tense. This kind of usage can be attributed to rote learning of words which are later analyzed as stem-plus-affix combinations. Within the dialectic mode, such combinations are called AMALGAMS. Newport and Meier (1985) have presented a convincing analysis of an entirely similar phenomenon (which they call FROZEN FORMS) in the acquisition of sign language. The important role amalgams seem to play in sign language is significant evidence for the universal importance of Operating Principles 1, 11, and 12. A second consequence of Operating Principle 11 (part b) is that early amalgams are defined as intonational units. CONSEQUENCE adpositions.

PHONOLOGY

11B: Affixes enter before

This is clearly true for Hungarian. Virtually all of the major suffixes are learned by age 3;0. At the same time, most of the postpositions are acquired after that time. However, it should be noted that the postpositions are used far less frequently than the suffixes in the adult language and that their meanings seem to be more complex. A further consequence of the intonational definition of early words is noted in part (a) of Operating Principle 11 where it is claimed that strings which occur by themselves as uninflected citation forms are likely to be picked up as intonational units.

317

CONSEQUENCE PHONOLOGY 11c: The first productive uses of stems and the first morphophonological overgeneralizations of stems usually make use of the uninflected citation allomorph, if the language has such forms. In the Hungarian diary literature, we find that overgeneralizations of citation allomorphs exceed overgeneralizations of oblique allomorphs by about 4 to 1. Thus, errors like *ló-k ‘horse-PL’ for lov-ak which use the citation form Ió are very common, whereas errors like *lova-nak ‘horse + DAT’ for ló-nak which use the oblique allomorph lova are quite rare. Consequence phonology lic has also received support from other languages. In German (Mills, 1985), children use the citation form of the adjective to form the comparative, as in hocher for höher. This is comparable to the English error gooder. Newport and Meier (1985) cite similar errors for ASL. In Hebrew, Berman and Levy (Berman, 1985) report early extensions of the singular citation form for nouns. Verbs, which do not have uninflected citation forms, show no such pattern. In Estonian, (Marilyn Vihman, personal communication) the first form of the verb is often the second person singular imperative which is also the citation form. In Polish, Smoczyńska (1985) reports forms such as chlepa and koteka which show over-generalization of citation allomorphs of stems. She also cites Wojtowicz as claiming that overgeneralization of citation allomorphs are “much more frequent than overgeneralizations of oblique allomorphs.” In Mandarin, the tone sandhi patterns of dipping tone words change in accordance with the tone of the following word (see MacWhinney, 318

1978 for a more detailed description). Li and Thompson (1976) have found that such dipping tone words first enter with the full dipping tone that is used when the word appears alone. This seems to be good evidence for the salience of citation forms. When the language does not provide the child with a frequent uninflected citation form, the child may attempt to use some other form as basic. Thus, in Japanese (Clancy, 1985) children may use the adjective with its present sense suffix as basic. In Romance 1PS and 3PS forms are often overgeneralized (Clark, 1985; Bybee & Brewer, 1980). The exact shape of the phonological packaging mentioned in Operating Principle 11 (part b) can be defined somewhat more precisely. Operating Principle 11 holds that children record the speech stream in terms of packages demarcated by intensity/pitch contours and pause junctures (Brown, Cazden, & Bellugi, 1968, p. 51). One consequence of this principle is: CONSEQUENCE PHONOLOGY 11D: Early imitations preserve accented syllables and, as much as possible, those unaccented syllables which are not separated from the accented syllable by pauses or junctures. In Hungarian, the first syllable of the words receives the main stress. In imitations, children tend to preserve the first syllable of a word or word string (Viktor, 1917). In Hebrew, where most words have final stress, initial syllables are often omitted in imitation (Berman, 1985). In Mohawk, Feurer (1980) found that, in long strings of affixes with penultimate stress, often only the accented syllable is preserved. 319

However, as Slobin (1973, p. 191) has noted in his Operating Principle A (“pay attention to the ends of words”), recency is also an important factor in syllable retention. In view of this, Operating Principle 11 (part b) is worded in a way that allows for a role for all three factors: stress, juncture, and recency. Thus, in addition to Consequence spellout 11d, the following consequence is allowed: CONSEQUENCE PHONOLOGY 11E: Final segments tend to be preserved more than earlier segments. Thus, Viktor (1917) found that, in three-syllable words, children would delete the middle syllable. Evidently, they preserve the first syllable for its stress and the third for its recency. However, if only one syllable is preserved, it is more likely to be the first. In general, it appears that early acquisition of the suffixational morphology of Hungarian seems to be facilitated by the aspect of Operating Principle 11. Consequence spellout l1e is also supported by data from several other languages. In Hebrew, the verbal prefixes, the definite prefix, and the prepositions are all omitted in early speech (Berman, 1985). On the other hand, the verbal suffixes of Japanese which are both word-final and often utterance-final are learned quite early (Clancy, 1985), as are the various pragmatic particles which appear in sentence-final position. Although most Hungarian amalgams are characterized intonationally as a syllable with main stress and the set of syllables that occur between it and the next main stress, there are a few cases in which pretonic syllables 320

are joined to the word without any juncture. In particular, the definite article (a, az) functions intonationally like an unstressed prefix. For example, children may pick up phrases such as az ebéd ‘the meal’ as rote items. Thus, along with stress and recency, juncture seems to be a factor in amalgam segmentation.

6.6. Amalgam Analysis (Operating Principle 12) Once the child has acquired a set of amalgams, s/he faces the task of analyzing those amalgams into their pieces. As Slobin (1973) has noted in his Operating Principle B (“the phonological forms of words can be systematically modified”), the very act of analysis indicates that the child is not bound to viewing the word as the final unit of meaning. In performing this analysis the child relies on Operating Principle 12. This principle, like Slobin’s Operating Principle D (“avoid interruption or rearrangement of linguistic units”), holds that the child first attempts to analyze words by breaking them up into perceptually clear continuous morphemes. This principle has at least six consequences. The first two are as follows: CONSEQUENCE PHONOLOGY 12A: The child is relatively slow in acquiring metathesis patterns. CONSEQUENCE PHONOLOGY 12B: The child is relatively slow in acquiring infixes. Hungarian provides no real test of these two consequences because it has so few productive cases of infixation or metathesis. However, one can point to the 321

fact that most Hungarian affixes are learned by age 3;0 as evidence for the general case of acquiring affixes and stems that are easily segmented. In Hebrew (Berman, 1985), productive control of the metathesis in simla—smalot is fairly late. The third consequence of Operating Principle 12 is as follows: CONSEQUENCE PHONOLOGY 12C: Occasionally, the child uses oblique stem allomorphs without affixes. There are a number of errors in the diary literature supporting this consequence. For example, children may use *köv for kő ‘stone’. The form *köv may be derived from an analysis of, say, követ (‘stone+ACC) into köv and et.4 The fourth consequence of Operating Principle 12 is as follows: CONSEQUENCE PHONOLOGY 12D: The acquisition of new bound lexical items by the analysis of amalgams relies on extraction of the high frequency allomorph and treatment of the residue as a new lexical item. This consequence is supported by the Hungarian diary literature. Stems are often extracted by removal of the most frequent allomorph of an affix. For example, the most frequent allomorph of the accusative is -t. When presented with the amalgam narancsot ‘orange+ACC, children recognize -t as the accusative and treat *narancso as the stem. In this case, however, the actual 322

citation stem is narancs ‘orange’. For further discussion of errors resulting from the use of frequent affix allomorphs in amalgam analysis see MacWhinney (1976) and Smoczyńska’s data (1985) on the analysis of poszlismy. Berman (1985) cites the error kadimanit ‘forward’ in Hebrew which can only be based on the extraction of a suffix *-anit from the word axoranit ‘backwards’ by the analogy axor ‘behind’ : axoranit ‘backwards’ :: kadima ‘forward’ : *kadimanit ‘forward’. This example shows how the extraction of a new morpheme can rely on a single exemplar. Another consequence of Operating Principle 12 is that the child may occasionally attempt to analyze new itesm which are morphologically indivisible. CONSEQUENCE PHONOLOGY 12E: Occasionally the child analyzes a unitary lexical item. Overanalysis usually occurs when a child tries to analyze a new unknown word. An English example of overanalysis would be analyzing carburetor as: car + buretor. A child making such an analysis might ask whether a truck has a buretor too. The majority of the reported Hungarian overanalyses (MacWhinney, 1974, pp. 398–399) divide the amalgam into a meaningful stem (car) and a meaningless residue (buretor). Sometimes the child is more successful and finds two meaningful stems as in the analysis of millio ‘million’ into mily ‘how’ and jó ‘good’. Other common errors of analysis in Hungarian involve attempts to decompose portmanteau morphemes like maguk ‘us’ (MacWhinney, 1974, p.

323

353) and tied ‘yours’ (MacWhinney, 1974, pp. 394–397). In maguk, children recognize the 3PL ending -uk and in tied they recognize the 2SG ending -ed. Although such analyses are incorrect, they seem superficially plausible. There are very few examples of overanalysis of affixes, although in Polish, Smoczyńska reports that children sometimes analyze the first person plural -śmy into -ś and -my. This is because the first person singular -my is a fairly common unit. A final consequence of both Operating Principle 12 and Operating Principle 11 is as follows: CONSEQUENCE PHONOLOGY 12F: Children will overgeneralize non-zero allomorphs even when zero allomorphs are of much greater applicability. This child does this because the zero allomorph is not actually a lexical item. Rather, it indicates the absence of a device for marking a function. Thus, when the non-zero allomorph enters, it is the first full allomorph for the function and is overgeneralized according to Consequence spellout 11b. The case of the third person singular indefinite in Hungarian illustrates this consequence. Nearly all verbs in the language take a zero-marking in the third person singular indefinite. However, a group of perhaps ten common verbs and other less common verbs end in -ik in this person. Despite the vastly greater applicability of the zero allomorphs, a large number of overgeneralizations of -ik are reported in the literature. Similarly, in the Polish genitive plural, -ow replaces the more frequent zero allomorph (Smoczyńska, 1985). 324

6.7. Merger (Operating Principle 14) In some cases amalgam analysis yields a new form which is synonymous but not homonymous with an old form. This may also occur in amalgam acquisition, particularly in bilingual environments. When this situation arises, Operating Principle 14 holds that there are two ways of resolving the synonymy. The first solution involves generalization rather than discrimination. It adds the new form to the old forms by simply changing the weights in a cluster concept or prototype. For example, in Hungarian, the plural allomorphs are -ak, -ek, -ok, -ök, and -k. This cluster of allomorphs may be represented by a vowel archisegment plus a /k/, as in -Vk. We will call such a situation weak allomorphy, since the allomorphs are not distinct morphemes. The second solution to the problem involves the introduction of a discrimination by adding a new morpheme. Strong allomorphy occurs when allomorphs are simply listed and not merged. Operating Principle 14 holds that weak allomorphy is preferred to strong allomorphy. As a consequence, we have: CONSEQUENCE PHONOLOGY 14A: When allomorphs have a large set of common features and when the roles governing allomorphy are opaque, the first productive allomorph is a phonologically central form. Thus, Mills (1985) reports that some German children use de as their first definite article. This form is phonologically central to the set of article allomorphs

325

(der, die, dem, des, den, das) although it is not a real allomorph itself. If a child uses expressive criticism (Operating Principle 6) and receptive criticism (Operating Principle 9) with a loose phonological filter, an item such as de is likely to be judged as acceptable more often than any alternative form. This possibility is also supported by data from Kunene (1979) on the acquisition of Si-Swati. In the case of Hungarian, applicability and phonological centrality are often confounded. I have shown (MacWhinney, 1975b, 1978) that the very first morphophonological errors in Hungarian involve attachment of a monoconsonantal suffix to the stem without a linking vowel. For example, in *pingvink, the child attaches the -k allomorph of the plural directly to the stem. As noted in Consequence phonology 11b, this allomorph is the most applicable of the five plural allomorphs. At the same time, it is also the common denominator of the five allomorphs of the plural (-ok, -ak, -ek, -ök, and -k). When allomorphs share little common phonological material, Operating Principle 14 holds that they should be acquired as separate morphemes. This leads to the following consequence: CONSEQUENCE PHONOLOGY 14B: When allomorphs share little common phonological material, they should be acquired and activated as separate morphemes and the alternation they represent should not be generalized to any other forms. Strong allomorphy is widespread in compound bilingualism, since one set of meanings often has two 326

forms (morphemes)—one in one language and one in the other. In the monolingual case, a new morpheme is established whenever two synonymous forms have different positional patterns. For example, the Hungarian nominal instrumental suffix -val must be a different morpheme from the pronominal instrumental prefix vel-. In certain cases, weak allomorphy and strong allomorphy may combine. For example, the Hungarian second person singular present indefinite has two sets of allomorphs: -ol, -el, -öl and -asz, -esz, -sz. It appears that these sets of allomorphs are best represented as two separate morphemes. There are no reported child overgeneralizations of the 1/sz alternation to any other forms.

7. APPLICATIONS OF THE MODEL TO THE ACQUISITION OF SYNTAX The dialectic model can also be used to account for the child’s learning of syntax. In the lexicalist model being used in this paper (MacWhinney & Sokolov, in press), syntax is defined quite narrowly as the system of rules governing the ordering of morphemes into the strings of surface structure. Here, we will consider the impact of Operating Principles 1,2, 3, 4, 5, 10, 11,12, and 13 of the dialectic model on the acquisition of Hungarian ordering.

327

7.1. Rote, Analogy, and Combination (Operating Principles 1, 2, and 3) 7.7.1. Rote (Operating Principle 1) According to Operating Principle 1, there are three ways the child can control the ordering of morphemes. First, s/he can learn a string of morphemes by rote. In terms of ordering, rote is revealed only by the occasional morpheme order errors it engenders. Such errors involve attachment of an affix to an unanalyzed amalgam. CONSEQUENCE SYNTAX 1A: When the child produces affix order errors, they will most often involve failure to separate an affix from a stem with which it frequently co-occurs. English illustrations of this type of error include I picked up my socks up and you pick up it (Menyuk, 1969). Smoczyńska (1985) cites a case of this type in Polish where children fail to separate the person suffix -m from the verb when attaching the suffix -by. In Hungarian, errors of this type usually involve the ordering of affixes in a word. Thus, föl-kel-ök ‘up-get-1SG’ often appears as *kel-j-föl-ök ‘get-IMP-up-1SG’ with the affix föl placed in the wrong position. Here the common sequence kel-j föl ‘get-IMP up’ is an unanalyzed amalgam to which the first person singular suffix -ök is attached. Similarly, in the form *kalctp-om-ka ‘hat-my-DIM’ for the correct kalap-ocská-m ‘hat-DIM-my’ the diminutive -ka (also -ocska) is incorrectly ordered after the first person possessive -om. Here, kalap-om ‘hat-my’ seems to function as an amalgam to which the diminutive is then added. 328

I have argued (MacWhinney, 1982) on the basis of English data that children may also learn strings of words by rote, although there seem to be limits on the size and number of such strings. The following consequence derives from that analysis: CONSEQUENCE SYNTAX 1B: Some words are used by some speakers in only a limited number of combinations, and are not subject at first to incorrect placements. For example, Bever (1970) and Richards (1979) report that 3-year-olds’ ordering of adjectives before the noun in English is usually correct, that it worsens around age 5, and then improves again in grade school. It may be that 3-year-olds are using adjective strings like great big or nice little which are learned by rote. For a variety of further evidence along this line consult MacWhinney (1982). Evidence of this type of U-shaped learning function has not been reported for Hungarian, probably because in Hungarian the existence of a multiplicity of word-order alteratives makes ordering errors less likely.

7.1.2. Analogy (Operating Principle 2) The second way in which the child can order a string of morphs is by analogy to some other string. This other string can be in either long-term memory or short-term memory. As I have noted (MacWhinney, 1982), there is very little evidence in favor of the idea that children can retrieve long strings of words from their long-term memories. Instead, the kinds of long strings that children tend to learn by rote are songs and poems rather than

329

standard sentence templates. On the other hand, it is clear from the research on sentence imitation that children can store long strings of words in short-term memory. Moreover, it appears that they can rely on Operating Principle 1 to use these strings as ways of producing new strings. CONSEQUENCE SYNTAX 2A: Children occasionally produce sentences or parts of sentences on analogy with preceding sentences spoken by themselves or by others. For example, Clark (1977) reports the following sequence: Adult: We’re all very mucky. Child: I all very mucky too. Here, one could argue that the first sentence serves as the basis for an analogy which takes part of the previous sentence and varies the rest. In Hungarian, this type of discourse analogy can often be detected by errors in verb inflection (MacWhinney, 1974, pp. 526, 527, 564, 584, 585). For example, if an adult says kérsz téát ‘Do you want tea?’, the child may answer kérsz ‘you want’ rather than kérek I want’. Such sequences are extremely common (see also Mills, 1985). Note that, in discourse analogy, the phrase that is borrowed from the previous sentence is not necessarily an item that the child has stored by rote in long-term memory. Rather, the sentence is built around an item in short-term memory. Moreover, the analogy involves not just ordering, but also semantics, since use of the earlier item in the current

330

sentence is based on an analogy between aspects of the earlier intention and aspects of the current situation.

7.1.3. Combination (Operating Principle 3) MacWhinney (1982) notes that, by combining the mechanisms of rote and analogy one can construct plausible accounts for the generation of virtually any phrase in a language. All that is required is that the system store a large number of templates for phrases and that it have a way of extracting similarities across templates which can be applied to new instances. Although it seems reasonable to imagine that small phrasal units may be stored as templates, it is much harder to believe that we store whole clauses and complex sentences as rote units. To the degree that rote and analogy fail to provide a sufficient account of sentence production, they must be supplemented by some mechanism that permits the combination of units. Operating Principle 3 holds that one way in which expressive forms may be produced is by combination. One obvious consequence of this principle is that not all forms are produced by rote. When forms are produced by combination, errors in the process of combination will lead to errors in the order of the morphemes. This means that the following consequences should hold: CONSEQUENCE SYNTAX 3A: Children and adults should make some errors in the ordering of morphemes in words.

331

The Hungarian literature reports only a few incorrect orderings of both affixes and words. In the next section, we will see why incorrect ordering is so quickly controlled.

7.2. Patterns (Operating Principle 4) Operating Principle 4 holds that combinations are shaped by the use of three types of patterns: predispositions, free rules, and bound rules. Let us consider the consequences of the use of each of these three ways of ordering lexical items.

7.2.1. Predispositions In the area of word order, predispositions operate to favor the placement of certain items before others. These predispositions can be viewed as universal preferences for the lexicalization of some items before others. There are at least three such predispositions that are relevant to early word combinations in Hungarian: informativeness, agency, and relatedness. Each of these three predispositions has received widespread attention in the international literature. However, as I have concluded from a review of this research (MacWhinney, 1982), the highly anecdotal, non-experimental nature of this literature has led to a certain confusion regarding the exact shape and relative importance of these predispositions. The first predisposition to be considered is the hypothesized tendency of the child to order the most informative (newest, least given, least predictable, etc.)

332

element first. This predisposition should lead to the following consequence: CONSEQUENCE SYNTAX 4A: In early combinations, children will tend to order the newest or most informative element first. In fact, Dezső (1970), MacWhinney (1975a), and Meggyes (1971) all report that there is a short period, early in Hungarian acquisition, when children tend to initialize and stress the verb, despite the fact that the most common word orders in Hungarian are SOV and SVO, and despite the fact that Hungarian sentences often begin with an unstressed element that is given or topical. Dezső (1970) argues that in such sentences the verb is expressing new or highly informative information. Presumably, in the situations confronting children, the objects are often highly given things like familiar playthings, common foods, or close family members. What is new and exciting are the activities in which these well-known objects engage. Verb initialization in early sentences has also been observed for French (Lightbrown, 1977), Italian (Bates, 1976; Fava & Tirondola, 1977), German (Park, 1974), and English (Braine, 1963b), none of which are verb-initial languages. This same predisposition may also be associated with the observation that in other SOV languages like Japanese (Clancy, 1985) and Turkish (Slobin & Aksu-Koç, 1985) children are so quick to pick up the use of right-dislocations of highly given material. The second predisposition we will consider is the hypothesized tendency for the child to order agents 333

before non-agents. Thus, in a string of nouns, the first noun would normally be the agent. CONSEQUENCE SYNTAX 4B: In early combinations, children tend to order the agent before the patient. The Hungarian literature suggests that such a predisposition is, at best, quite weak, since early sentences include Patient-Agent order as well as Agent-Patient orders. The same is true of Turkish (Slobin, personal communication). The third predisposition to be considered is the hypothesized tendency, also known as Behaghel’s Law, to place together words that are semantically related. CONSEQUENCE SYNTAX 4C: Children have no problem placing together items that are related in semantic structure, However, when semantic relations are not clear or when items are related to whole clauses, errors are more frequent. There are almost no reported errors in the placement of items according to consistent rules when there is a simple semantic relation between single items. However, items such as the negative nem, the interrogative -e, and the conditional volna, which modify the whole clause, are often placed in incorrect positions. Similarly, conjunctions such as mielőtt ‘before’ may be placed on the wrong clause (MacWhinney, 1974, p. 502), but errors of this latter type may be conceptual (Clark, 1971).

334

Altogether, the impact of predispositions on syntax seems to be confined to the earliest stages and weakly learned patterns. The main determinants of rule strength and competition are not predispositions, but the factors of reliability and applicability discussed earlier. In any case, it is crucial to remember, that, in the competition model, predispositions compete on an equal footing with rote, analogy, and combinatorial rules. Thus, the main empirical question relates to the relative strength of predispositions in this competition.

7.2.2. Free rules The second major type of ordering pattern is the free rule. Free ordering rules order morphemes on the basis of their role in the clause, rather than their specific lexical identity. Because they are not lexically bound, we call these rules lexically-free rules, or simply free rules. For example, morphemes connected by the modifier-modified relation are ordered by a ordering pattern which places the modifier before the modified. This rule orders big before dog in big dog and red before balloon in red balloon. The possible scope of such rules is discussed in detail in MacWhinney (1982). Free rules can be used to describe, for example, the dominance of SVO ordering in Indo-European (Clark, 1985; Mills, 1985) and SOV order in Japanese (Clancy, 1985). Many word order patterns in Hungarian could be viewed as either free or bound. For example, the MODIFIER + MODIFIED pattern could be viewed as actually a set of patterns such as BIG + X and RED + X bound to words like BIG and RED. Since either the free or the bound solution could work and since there are few errors to demonstrate

335

the presence of one pattern or the other, it is hard to exclude either interpretation. Moreover, if one allows bound rules to be activated by analogy, it may be possible to account for morpheme ordering without relying on free rules at all. Perhaps the best evidence for the reality of free rules in child grammars is the presence of strong developmental shifts in the ordering of the major constituents. If word order were governed simply by a set of lexically-bound rules, across-the-board developmental changes in word order would not be expected. At the same time, free rules should show their presence by applying automatically to new operators, as soon as they are required. CONSEQUENCE SYNTAX 4D: Children seem to demonstrate control of free rules by (1) making across-the-board changes in word order and (2) applying the pattern automatically to new operators. The Hungarian research is relevant to only the first part of this consequence. Meggyes (1971) reports that, from 1;9 to 2;0, her daughter used mostly VS ordering. During the period from 2;0 to 2;2 this ordering gave way to the SV ordering that is more common in the adult language. I observed (MacWhinney, 1974) a period during which an early predisposition for fronting of informative material gave way to the free pattern of ACTOR + ACTION. Hungarian has another set of word order patterns that cannot be controlled by bound rules, unless these are accompanied by some strong analogical mechanism. The 336

two most important of these are the rules that are used to order the topic and the focus. The first of these is the TOPIC + COMMENT PATTERN. This pattern orders the topical element at the beginning of the utterance and then follows it, after a slight pause, with the comment. Each sentence may have only one such topic. Often the topic is the actor (there is no need in Hungarian to distinguish between the “actor” and the “subject”). However, any noun phrase may be topicalized. This free pattern is quite consistent and children make few errors in its application. The second major pragmatic pattern is the FOCUS + ACTION/PROCESS pattern. This pattern orders the item of highest focus directly before the verb stem. Because of the complexity of this pattern, we will postpone its discussion until we also consider Operating Principle 5.

7.2.3. Bound rules The third major type of ordering pattern is the bound rule. These rules are bound to a particular lexical item or set of lexical items. I have referred (MacWhinney, 1982) to these rules as ITEM-BASED PATTERNS. Such patterns order a specific morpheme (the operator) either before or after some other morpheme or phrase (the nucleus) with which the operator is related semantically. For example, the English morpheme big always occurs before the morpheme which it modifies. Therefore, the item-based positional pattern for big is: big + X, where X is the nucleus and the plus indicates that the item on the left precedes the item on the right.

337

The psychological reality of an item-based pattern may be demonstrated in at least two ways. First, its productivity may be demonstrated by application to a new item. Thus, if a child learns the new word zebra and says there’s a big zebra, we can argue that big has been ordered by use of an item-based pattern. Second, the productivity of a pattern may be demonstrated by consistent placement of the operator. Thus, if a child places this in front of the noun to which it is related 9 times out of 10, we can argue that there is statistically significant evidence for a positional pattern (Braine, 1976). Third, if we find a development in the intonational integration between items, we can take this as some evidence for the presence of an item-based pattern with an associated intonational contour. These three types of evidence may be summarized in this way: CONSEQUENCE SYNTAX 4E: Soon after learning a new operator, children demonstrate control of an item-based pattern for that operator by (1) productive positioning of that operator with new nuclei, (2) consistent correct positioning of the operator, and (3) intonational integration of the operator with its nuclei. A study of over 11,000 early utterances from two Hungarian children (MacWhinney, 1975a) provided strong support for the second part of this consequence. There was only one clear cut case of a failure to use a relevant itembased pattern (MacWhinney, 1974, p. 474). Moreover, between 85 and 100 percent of the utterances produced by these two children could be analyzed in terms of a set of a few dozen item-based patterns, such as ott + X 338

‘there + X’ as in ott kutyus ‘there doggie’. In the case of affix ordering, the support for consequence ordering 4e is equally emphatic. Errors in affix ordering almost never occur and most of those that do occur can be attributed to amalgam usage (see consequence ordering la above). The third part of consequence ordering 4e is supported by Fónagy’s (1972) study of intonation in early word combinations. In that study, Fónagy found that combinations begin as successive single word utterances, but that over time, the final contour on the first word and the pause between the items tends to disappear until a single smooth intonational contour emerges. One common type of bound rule is the governance rule. In both Indo-European and Finno-Ugric, verbs govern the cases of their object in a fairly idiosyncratic fashion. For example, we can say fight with John, fight John, and hit John, but not *hit with John. It may be possible to formulate some of these patterns as free rules. However, many instances of governance seem to be lexically bound. As in the case of agreement, the verb and the case marker may often be noncontiguous in surface structure. Because the units first acquired by the child are amalgams (Operating Principle 11), the child will be slow to pick up longer-range dependencies. CONSEQUENCE SYNTAX 4F: Children will make errors in governance even when the rules are consistent. Mills (1985), Smoczyńska (1985), and Clancy (1985) report a variety of governance errors for German, Polish, and Japanese. Table 4 lists the governance errors that 339

have been reported for Hungarian (MacWhinney, 1974, pp. 545–548, 575–580). These errors are basically of two types. The first type involves confusion within the set of locative cases. Thus, the child says ‘into the leg’ instead of ‘onto the leg’ or ‘from the chair’ instead of ‘into the chair’. These errors either use the wrong direction for the movement or treat the reference point as the wrong topological type (enclosure for point, surface for enclosure, point for surface, etc.). The other type of error involves confusions between the three principle non-locative cases: the accusative, the dative, and the instrumental. These errors indicate that children are working on two very general goals. One is to lexicalize a locative suffix and the other is to lexicalize a suffix for the basic case relations. It might be the case that the fact that all of these devices are marked by suffixes increases their confusability. For example, it might be that, if the language marks the accusative with a suffix and the instrumental with a prefix, fewer confusions would arise. Governance errors in Hungarian are at least as prevalent as Table 14.4 suggests. They are certainly a characteristic aspect of the speech of preschoolers and they continue into the early grade-school years. It seems that, when children are not sure of the specific case governed by a specific verb, they attempt to generate a case by general semantic strategies such as “select a reasonable locative for directional movement.’’ However, the idiosyncrasies of the language make such strategies less than optimal and children must eventually learn governance verb by verb.

340

TABLE 14.4 Governance Errors

7.3. Competition (Operating Principle 5) The ordering of several operators about a single nucleus must be governed by a series of precedence constraints between patterns, as noted in Operating Principle 5. In effect, the child must learn which patterns will win out when two or more patterns come into competition (see MacWhinney, 1982, for details). The basic principle governing the resolution of competition is that the system attempts to maximize the overall fulfillment of patterns. Rules vary in strength and 341

the satisfaction of strong rules is more important than the satisfaction of weak rules. However, in a fully developed, well-organized system, some highly specific rules are set up to “buy off” or “suppress” their more analytic counterparts. Until this system is learned, we expect the following consequence to hold: CONSEQUENCE SYNTAX 5A: When two or more morphs have positional patterns that compete for a given position, at first the child makes frequent errors in their ordering. In these errors, the most strongly activated morpheme is placed closer to the stem. An example of this consequence in Hungarian is the competition between patterns for preverbal positioning. We have already noted in the discussion of consequence ordering 4d above that the focal element has the primary claim on this position. Focus itself is determined by a complex series of factors including contrastivity, negation, indefiniteness, and informativeness. Moreover, and to make matters even more complicated, when there is no focused item, the separable verbal prefix is placed before the verb stem. Otherwise, it is postposed. However, in the imperative, the separable prefix is obligatorily postposed and with certain modals it precedes both the modal and the stem in the order SEPARABLE PREFIX + MODAL + STEM. Because the vectors on the category of focus are so complex, and because there is so much competition for preverbal position, children make many errors in selection of the correct preverbal element. These errors (MacWhinney, 1974, pp. 466–471) include: (1) failure to prepose the perfective separable prefix when no other item qualifies for focus 342

(*Éva fürdik meg? ‘Éva bathes up?’ for Éva megfürdik ‘Éva up+bathes’); (2) failure to postpose the separable prefix in the imperative (megnézzük ‘up+see+we’ for nézzük meg ‘see+we up’); and (3)failure to postpose the separable prefix when there is a focused pre-verbal element (*nem megeszem ‘not up+eat+I’ for nem eszem meg ‘not eat+I up’). The child must learn to integrate the various free rules here with the bound rules for each of the separable prefixes. The large number of errors that have been reported for this system testifies to the difficulty facing the child in achieving this integration. I should perhaps add that it is my impression that these errors do not reflect the tendency of some children to prepose the separable prefix and for other children to postpose it. Rather, the children I have observed all make errors of each of the three types. However, the exact distribution of these errors warrants closer scrutiny. An example from French cited by Clark (1985) indicates how a conflict between affix positions may arise developmentally. The object clitic occurs very frequently in the position before the verb, as in je le mange I it eat’. When indirect object clitics are added to this structure, they are appended in front of the direct object clitic. This results from the greater strength of the DIRECT OBJECT + VERB pattern. For first and second person indirect object pronouns, this leads to no error. For third person indirect object clitics, however, the correct order is DIRECT OBJECT + INDIRECT OBJECT, but children use the opposite pattern and make errors. In Polish, the conditional -by competes with the person suffixes for postverbal positioning. Because the person suffixes 343

occur more frequently, they are often incorrectly allowed to win out in this competition. This leads to errors like *pisai-em-by (= ‘write-I-would’) or *by pisat-em (= ‘would write-I’) for the correct form pisai-by-m (= ‘write-would-I’) (Smoczyńska, 1985). In both the Polish and French examples, the child first solves these competitions by placing the frequently occurring morpheme closer to the stem. Operating Principle 10 holds that this is due to the greater strength of the positional pattern for the more common morpheme. Note, however, that many of these same phenomena are also predicted by Consequence syntax la which states that “when the child produces affix order errors, they will most often involve failure to separate an affix from a stem with which it frequently co-occurs.” The system of competition also has to deal with the competition between positions for a morpheme. Thus, the counterpart of syntax 5a is syntax 5b which holds that: CONSEQUENCE SYNTAX 5B: When two positions compete for a given morpheme, the child produces incorrect orders and overmarkings using that morph in both positions. Smoczyńska (1985) cites three errors of this type: 1. Incorrect and redundant placement of -m on both the verb and the conditional marker -by as in Nie pojeździł-em by-m ‘not ride-I would-I (= ‘I wouldn’t ride’) for the correct Nie pojeździł-by-m ‘not ride-would-I’. 344

2. Use of by as both a suffix and an enclitic. Thus we have errors like A moja mamusia też BY miała-BY ładne włoski ‘and my mommy also have-would pretty hair’ (= ‘and my mommy would also have pretty hair’). 3. Redundant marking of the first person plural of the past by -śmy on the verb when the pronoun my ‘we’ is available. Thus, children may say my-śmy poszli-śmy ‘we-first/plural went-first/plural’ (= ‘we went’) for correct my-śmy poszli ‘we-first/plural went’ (= ‘we went’). In English, comparable overmarkings such as I picked up my socks up are quite common (MacWhinney, 1982). A particularly difficult system of this type is the system of adjective declension in German (Mills, 1985). Overmarkings of endings such as -er ‘DEF.ART:MASC.SG.NOM’ lead to forms like *meiner guter Papa ‘my-DEF.ART:MASC.SG.NOM good-DEF.ART:MASC.SG.NOM Papa’ for mein guter Papa ‘my good-DEF.ART:MASC.SG.NOM Papa’. Clark (1985) also cites French errors in which postverbal positioning of direct object clitics in affirmative imperatives competes with preverbal positioning in negative imperatives.

7.4. Rule Acquisition (Operating Principles 10 and 15) The tenth Operating Principle of the dialectic model holds that every time a rule applies successfully, it gains in strength. Conversely, whenever a rule applies unsuccessfully, it loses strength. There are a series of 345

consequences of this principle of survival of the fittest for ordering. The first consequence relates to the role of predispositions. CONSEQUENCE SYNTAX 10A: It takes time for children to learn to block the application of predispositions toward certain incorrect word orders. Some of the earliest sentences produced by the Hungarian child provide support for this consequence. As we noted earlier, in their first sentences, children tend to place the most informative element first. This often leads to the placement of the verb in initial position even when the adult language would use another order. However, as Meggyes (1971) has noted, because of the pressure of the highly productive TOPIC + COMMENT pattern, children soon overcome this predisposition. The second consequence of Operating Principle 10 relates to the order of acquisition of ordering rules: CONSEQUENCE SYNTAX 10B: The first ordering patterns that are acquired are those which apply correctly to the largest number of combinations produced by the child This consequence is supported by two types of data. The first type of data is that there are virtually no errors at all in the use of bound rules for affixes. As in Turkish (Aksu-Koç & Slobin, 1985), the order of affixes after the stem in Hungarian can be stated in terms of rules that have no exceptions. Also as in Turkish, such rules are acquired very early with virtually no errors. The second type of evidence for syntax 10b is that order of 346

applicability also determines the order of acquisition for alternative placements of certain words. For example, the first productive rule for verbal separable prefix placement is the one which places the verbal separable prefix before the verb. Then the rule for verbal separable prefix postposing is learned and finally the rule for pre-modal placement emerges. This developmental order reflects the relative applicability of these three rule types. Similarly, in German (Mills, 1985), the rule placing nicht ‘not’ after the tensed verb is the first negative placement rule acquired and is also the most applicable. In Indo-European, the preverbal placement of the actor is a highly applicable rule which is learned early on. In Hungarian, the placement of the topic before the comment and the focus before the verb are similarly high in applicability. They are also learned quite early. The third major consequence of Operating Principle 10 is that children will attempt to formulate rules in the most general form possible. Operating Principle 15 holds that error monitoring leads to the formation of both free and bound rules. However, free rules are higher in applicability and will grow faster in strength, if they do not lead to errors. A consequence of this tendency of free rules to grow quickly in strength is as follows: CONSEQUENCE SYNTAX 10C: When the language makes use of bound rules that apply quite generally, the child overgeneralizes these rules. For example, the rule of separable prefix preposing is often overgeneralized on the basis of the semantic feature of [+ modality] to include the preposing of the 347

conditional postposition volna (MacWhinney, 1974, pp. 475, 478). Consequence ordering 10c is also widely supported for other languages. I review (MacWhinney, 1982) the literature showing how children attempt to interpret the complements of verbs like ask, promise, and tell in terms of a free rule. Doing so leads to errors in interpreting sentences with promise. Thus, in Grover promised Oscar to go children say that it is Oscar who will go. Clark (1985) reviews similar results for these verbs in French and Spanish.

7.5. Item Acquisition (Operating Principle 11) Although ordering involves the acquisition of rules rather than items, there are still some interesting consequences of Operating Principles 11, 12, and 13 for the development of ordering. Operating Principle 11, which holds that early lexical items may be amalgams, leads to this consequence (compare with Consequence phonology 11a): CONSEQUENCE SYNTAX 11A: Bound morphemes are ordered correctly even before they are used productively. However, free morphemes may be used productively before they are ordered correctly. Thus, at age 2;2, my subject Zoli used a large number of correctly placed past tense suffixes without showing any evidence of productive use of the past tense. He showed a similar pattern of correct placement preceding productivity across a whole array of affixes. In each case, Zoli appeared to be acquiring affixes initially as parts of amalgams. On the other hand, we find errors 348

such as labda Jancsi ‘ball Johnnie’ for Jancsi labda(ja) ‘Johnnie ball(POS)’ in which there is clearly a mistake in the order of free morphemes (MacWhinney, 1975a). Another consequence of Operating Principle 11 is closely related to the first: CONSEQUENCE SYNTAX 11B: The first ordering rules that the child will learn are those that govern the ordering of items in intonational units. As was noted above, the Hungarian data support this consequence in that the rules controlling affix ordering are acquired very early and are almost never violated.

7.6. Amalgam Analysis (Operating Principle 12) The twelfth Operating Principle holds that the child will first attempt to analyze amalgams by breaking them up into continuous morphemes. Thus, the child should acquire continuous morphemes before discontinuous morphs (Consequence phonology 12a). A further result of this strategy is that: CONSEQUENCE SYNTAX 12A: Patterns that place the component pieces of discontinuous morphemes into linear order are acquired late. The Hungarian data provide no clear data relating to this consequence, partly because there are so few items in Hungarian that are clearly discontinous morphemes rather than agreement markers. 349

8. APPLICATIONS OF THE MODEL TO THE ACQUISITION OF LEXICAL SEMANTICS The dialectic model can also be applied to the study of the acquisition of the patterns that use semantics to achieve activation of lexical items in retrieval. During retrieval, the speaker uses the mappings from semantic features and clusters onto morphemes to decide how to apply inflections and derivations in ways that are sanctioned by the semantic principles of the language:

8.1. Rote, Analogy, and Combination (Operating Principles 1, 2, and 3) There are three ways in which a speaker may take a meaning and convert it into a set of morphemes. These ways are rote, analogy, and combination.

8.1.1. Rote (Operating Principle 1) If a meaning can be directly mapped onto a single morpheme, then lexical retrieval has occurred by rote. Such rote forms often consist of several morphemes as in the case of idioms and other phrases. They also include inflected words for which the following holds. CONSEQUENCE SEMANTICS 1A: Meanings associated with affixes appear in amalgams before they are used combinatorially. This is to say that the child uses a few plural forms before he develops a general plural. Errors like *rices (for rice grains) do not appear at this time. This

350

consequence is the semantic flip-side of consequence spellout lb which was discussed earlier. Please refer to that discussion for evidence on Consequence retrieval la.

8.1.2. Analogy (Operating Principle 2) Analogy in retrieval would lead to the production of an item in which the product would have a polysemic structure like the basis. If analogy is being used to produce lexical structures, the following consequence should hold: CONSEQUENCE SEMANTICS 2A: Occasionally, speakers produce forms for which there could be only one possible semantic analog. This consequence is supported by a handful of examples from many adult languages. For example, in English the first form of the sit-in, be-in, teach-in class was sit-in. When teach-in was introduced as the second form, only sit-in was available as an analog. Further examples of this type of common in the literature on language change (Stern, 1931). In child language, if one looks at words in their discourse context, one finds occasional good support for Consequence retrieval 2a. Thus, Trencseny (MacWhinney, 1974, p. 381) reports a conversation in which he told his daughter Piroska to eat the little pieces of meat on her plate egy-enként ‘one-DISTRIBUTIVE’ (‘one at a time’). Piroska replies that one can also eat them *sok-enként ‘many-DISTRIBUTIVE’ (‘many at a time’). The only plausible basis for *sokenként seems to be egyenként. Note also that the analogy is not fully rule governed since the stems to which -nként can be attached

351

are always (+ delimited) in meaning and the use of -nként with sok ‘many’ violates this restriction. Certain other examples of Consequence semantics 2a are somewhat less convincing. For example, Vertes (MacWhinney, 1974, p. 384) reports that, at age 5;0, Laszlo used rabkulcs (rab ‘prisoner’ + kulcs ‘key’) for the correct form tolvajkulcs (tolvaj ‘thief + kulcs ‘key’) to express what corresponds to the English term skeleton key. The choice of rab in rabkulcs seems to be a result of an analogy with some weakly stored form of tolvajkulcs. Essentially, the idea is that if a key can be a thief, it might also be in some analogous sense a prisoner. Parallel examples that appear to be analogies are *borjufelho ‘sheep cloud’ for bárányfelhő ‘lamb cloud’ (MacWhinney, 1974, p. 383) and *körmös néni ‘nailish lady’ for manikürös kisasszony ‘manicurey girl’ (MacWhinney, 1974, p. 182). Lexical analogy plays a clear and important role in language change. In the child language literature it is often assumed that neologisms are produced by a process of proportional analogy. For example, given a neologism like cutter for “scissors,” one might argue that cutter is being produced on analogy with mower where the formative -er is taken to indicate for cutting just what it indicates for mowing. Unfortunately, few studies have directly investigated the operation of analogy as a psychological process rather than as a general descriptive mechanism. In principle, it would be possible to conduct such studies within the context of the nonce-probe task developed by Berko (1958) and modified by MacWhinney (1978). 352

If analogy is understood to include not just proportional analogy using single exemplars, but also extended analogy, then its descriptive and generative power is vastly increased. Lacking the necessary data on the frequency of the error types, many if not all of the examples of combinatorial derivations that will be discussed below could also be viewed as forms produced by extended analogy in retrieval.

8.1.3. Combination (Operating Principle 3) Operating Principle 3 holds that one way a child can express a meaning is by combining morphemes. Parallel to Consequence phonology 3a above we then have this consequence: CONSEQUENCE SEMANTICS 3A: Children form words and phrases from any semantically feasible combination of morphemes. The notion of semantic feasibility is the same as the one discussed under consequence spellout 3a above. This consequence is supported by the large array of neologisms that have been reported in the Hungarian child language literature. The clearest support for consequence semantics 3a comes from the nearly 300 reported neologisms by Hungarian children which violate no rule of any part of the grammar. These neologisms demonstrate productive uses of the 25 formative suffixes listed in Table 14.5. Of course, there are also a few “formatives” like the infinitive -ni that are so productive in adult Hungarian that their use is almost

353

never erroneous. In terms of the polysemy they illustrate, such formatives might be better classified as inflections. There are at least four reasons why these more than 300 derivational neologisms and others like them are not fully acceptable: 1. they may use a stem which is either nonproductive or idiosyncratic; 2. they may use an affix which is in arbitrary competition with another affix; 3. they may produce a result which is already present as a non-analyzed stem, or 4. they may fill a lexical gap which is of little importance to the adult world. TABLE 14.5 Productive Formative Suffixes Suffix

Derivational function

English Earliest approximation use

Suffixes making verbs from verbs: 1. -an, -en

momentaneous

suddenly

2;3

2. -ad, -ed

inchoative

become

4;9

3. -kod, -köd

iterative

frequently

—

4. -kál, -kél

frequentative

keep

3;6

5. -gat, -get

frequentative (highly productive)

keep

2;6

354

6. -dogél, -dogál

frequentative

keep

3;6

7. -ódik, -ődik

middle voice

—

3;1

8. -ít

causative

make

2;0

9. -tat, -tet

causative (highly productive)

make

—

causative

make

3;6

10. -aszt, eszt

SUFFIXES MAKING VERBS FROM NOUNS OR ADJECTIVES: 11. -oz, -ez, -öz

general denominative

-ize

2;1

12. -ít

general denominative

-ize

2;2

13. -ol, -el, -ö1, general denominative -I

-ize

1;6

14. -kodik, -kedik, -ködik

agential denominative

-ize

3;0

15. -ul, -ül

de-adjectival

become

3;0

16. -odik, -edik, de-adjectival -ödik

become

2;5

SUFFIXES MAKING NOUNS FROM VERBS: 17. -ó, -ő

general adverbative

-er

1;8

18. -oda, -ede, -ode

locative deverbative

-ery

—

19. -at, -et

resultative (English -ant)

-ant

6;0

20. -ság, -ség

abstract deverbative

-ness

4;8

21. -es, -as

resultative

-ing

5;6

-ist

5;0

SUFFIXES MAKING NOUNS FROM NOUNS: 22. -os, -es, -ös, profession -as, -s SUFFIXES MAKING ADJECTIVES: 23. -as, -es, -s

denominative

-y

6;0

24. -só, -ső

de-adverbial

-y

5;0

355

25. -os, -ős

deverbative

-y

1; 11

These four reasons for neologism unacceptability are cases of what Clark and Clark (1979) call pre-emption. In each case the child is forming words by combination in a semantically plausible way. However, lexical facts override semantic plausibility just as they override morphophonological plausibility. Just as the existence of went overrides the plausibility of goed, so the existence of scissors overrides the plausibility of cutters. For each of the four types of preempted neologisms there are different reasons why a lexical item already exists that precludes the combinatorial form. Let us consider each of the four types and examples of these reasons in greater detail. 1. There are only three reported errors in which children use a non-productive or idiosyncratic stem in an otherwise correct neologism. In the form kancsalít ‘to make act like a pitcher’ (MacWhinney, 1974, p. 364), the stem kancsal ‘act like a pitcher’ is itself a neologism. Here an adult would simply say leönt ‘pour out’. The neologism összecsicered ‘stick together’ (MacWhinney, 1974, p. 363) uses a stem csicer which is the child’s own onomatopoeic creation symbolizing sticking. Finally, the neologism vakaros ‘scratchy’ (MacWhinney, 1974, p. 379) is based on a misuse of vakar ‘scratch’, since the child really wanted to say viszketős ‘itchy’. 2. Neologisms can also arise when two affixes mean the same thing or nearly the same thing. English examples (from Stemberger, 1982) include the use of decidal for

356

decision, philosophist for philosopher, and disparence for disparity. In such cases the various stem and affix combinations will have to be learned by rote. Hungarian seems to be particularly rich in such errors. Table 14.6 lists the neologisms of this type that have been reported. In each case, the standard form and the neologism are so close that one suspects that the child may have been familiar with the standard form but encountered some difficulty in retrieving it. However, having retrieved the stem of the standard form, another affix then moved into the slot vacated by the standard affix. This interpretation seems plausible and interesting, but would have to be tested against new data. TABLE 14.6 Neologisms Stemming from Wrong Suffix Choice Child Form

Gloss

Adult Form

Gloss

öltöztetés

dress-cause-ing öltözet

dress

törös

breaky

törekény

breakable

lefolytat

drain out

lefolyat

drain out

mozdít

make move

mozgít

make move

röpdögél

fly about

repked

fly about

szállkál

land about

szálldogál

land about

peregít

turn

perdít

turn

ebédez

dine

ebédel

dine

kövekel

stone

kövez

cobble

besoroz

line up

besorol

line up

megsimál

smoothen

megsimít

smoothen

rámnehezekedik get hard for me rám nehezedik get hard for me bekormít

get dirty

bekormoz

357

get dirty

elkormít

get dirty

elkormoz

get dirty

megkarcít

scratch

megkarcol

scratch

aranyít

make golden

aranyoz

make golden

lovagoz

ride horse

lovagol

ride horse

maradás

remains

maradék

remains

ugrat

jump

ugrás

jump

rakat

stack

rakás

stack

bukat

jump

bukás

jump

koesmás

inn-keeper

kocsmáros

inn-keeper

3. The third and most common reason why child neologisms may not be acceptable is the existence of some alternative form that expresses in one morpheme the same meaning as the child tries to express by the combination of morphemes. For example, in English one of the alternative meanings or polysemes of the word youth is equivalent to the compound *youngness. Thus, youth preempts *youngness. In the Hungarian literature there are many neologisms of this type; they are presented in Table 14.7. In the majority of the cases, Hungarian pre-empts the productive derivation in ways that are quite parallel to English. In some other cases, given at the end of the Table, there is no English parallel to the Hungarian pre-emption. 4. The fourth reason why child neologisms are not accepted is perhaps the weakest reason of all. In a number of cases children make up perfectly good words that simply do not correspond to anything that needs to be permanently coded in the lexicon. Of course, adults do this too (Clark & Clark, 1979) when they talk about jet coaching and Americanness. In particular, Hungarian 358

encourages neologistic use of the two demonstrative verbalizers -oz and -ol and the suffix -kodik ‘act like’. Thus, we find forms like malacozik (malac- ‘pig’ + -ozik ‘act like’) and ridegenkedik (ride gen ‘stern’ + -kedik ‘act like’). Some of these forms that were reported as neologisms earlier in this century are now established in the language. These include montíroz (MacWhinney, 1974, p. 366) ‘to blacksmith’, homokoz (MacWhinney, 1974, p. 367) ‘play in the sand’, szemezi+k (MacWhinney, 1974, p. 367) ‘play with eyes’, and the compound pénztartó ‘money holder’ (MacWhinney, 1974, p. 382).5 An area in which Hungarian displays particular neologistic richness is the area of compound formation. The five most productive compounding patterns in Hungarian each lead to several reported child neologisms. In the descriptions below, the numbers in parentheses indicate the pages in MacWhinney (1974) where the errors are reported in full. 1. ENTITY x + ACTION Y + NOMINALIZER: an entity Z which Y’s (a, the) X, as in the English compound potholder ‘a thing which holds pots’. Hungarian child neologisms of this type include fájgyógyító ‘pain-curer’, cf. correct gyógyszer ‘medicine’, (MacWhinney, 1974, p. 381),pénztartó, ‘money-holder’ cf. correct ridikül ‘purse’, villanyégető ‘electricity-burner’, cf. correct lámpa ‘lamp’, falterítő ‘wall-coverer’, cf. correct falvédó ‘wall protector’, libahúzó ‘goose- puller’, cf. correct játék, and utcanéző, ‘street-looker’, cf. correct erkély ‘balcony’ (MacWhinney, 1974, p. 382). In these compounds the Hungarian suffix -ó which is equivalent 359

to English -er imposes a transitive reading on the verb and codes the noun as an object. TABLE 14.7 Neologisms Stemming from Existence of Non-inflected Form Child Form

Gloss

Adult Form

pillogó

blinker

szempilla

eyelash

haragudó

angry-er

szemöldők

eyebrow

füstölők

smoker

kémény

chimney

író

writer

ceruza

pencil

kötő

binder

zsinor

cord

pucó

cleaner

rongy

rag

kukucsoló

peek-a-boo-er

sztereoszkóp stereoscope

bekenő

in-spreader

orrkenőcs

nose salve

dolgozó

worker

kalapács

hammer

vágó

cutter

olió

scissors

kapáló

hoer

kapa

hoc

rajzoláló

sketcher

ceruza

pencil

nyomó

presser

festék

paint

ülő

sitter

szék

chair

körüljáró

around walker

palló

boardwalk

hallat

heard

hang

sound

parancsolatos orderer

katona

soldier

fogas

toother

fogorvos

dentist

étvágyas

hungery

éhes

hungry

iszos

drinky

szomjás

thirsty

lökős

pushy

biliárd

billiard ball

csurizós

slinger

csúzli

slingshot

360

Gloss

villanyégető

lightburner

lámpa

lamp

röpülőlabda

flying ball

léggömb

balloon

kargyürü

wrist ring

karktöbő

bracelet

zeneskatulya

music box

verkli

organ

utcanéző

street looker

erkély

balcony

levizesül

water down

megázik

soak

kirongyosít

raggen

kirojtoz

wrinkle

rámagasít

highen

ráemel

raise

hidegít

colden

lehüt

cool

lehéjázik

off rindize

meghamoz

peel

tanultat

cause to learn

tanít

teach

megdögleszt

cause to die

megöl

kill

sebzik

woundize

vérzik

bleed

elromlít

breakate

elront

break

leszálltat

cause to come down

leernel

lower

festékez

paintize

fest

paint

ekéz

ploughize

szant

plough

2. ENTITY X + ENTITY Y: a Y of the type such that X possesses it. Child neologisms here include cipőliszt ‘shoe flour’, cf. correct hintőpor ‘talc’ (MacWhinney, 1974, p. 381), kargyürü ‘arm ring’, cf. correct karkötő ‘bracelet’ (MacWhinney, 1974, p. 383), utaóház ‘traveler house’, cf. correct palyaudvar ‘railway station’, rabkulcs ‘prisoner key’, cf. correct tolvajkulcs ‘thief key’, szemtükör ‘eye mirror’, cf. correct szemüveg ‘eyeglasses’ (MacWhinney, 1974, p. 384), and kottatér ‘note ground’, cf. correct zenepavilon ‘music pavilion’ (MacWhinney, 1974, p. 473).

361

3. ENTITY X + ENTITY Y: a Y that typically deals with X. Child neologisms here include jegybácsi ‘ticket man’, cf. correct kalauz ‘conductor’, motorbácsi ‘motor man’, cf. correct motorkerékpáros ‘motorcyclist’, and szemszerész ‘eye repairman’, cf. correct látszerész ‘optician’ (MacWhinney, 1974, p. 382). 4. ENTITY X + ENTITY Y: a Y that can be characterized as an X. Child neologisms here include csontember ‘bone man’, cf. correct csontváz ‘skeleton’, bárány felhő ‘sheep cloud’, cf. correct borjúfelhő ‘lamb cloud’ (= ‘cloud that resembles a lamb’), virágcsalán ‘flower nettle’, (= ‘nettle that looks like a flower’) (MacWhinney, 1974, p. 383), vonatkép ‘train picture’ (= ‘picture of a train’), and zeneskatulya ‘music box’, cf. correct verkli ‘music box’ (MacWhinney, 1974, p. 384). 5. MODIFIER X + ENTITY Y: an entity Y that has the characteristic X. Child neologisms include pírosebéd ‘red meal’ (= ‘tomato dish’), szárazhíd ‘dry bridge’ (= ‘bridge over railway tracks’), and repülőlabda ‘flying ball’, cf. correct léggömb ‘balloon’ (MacWhinney, 1974, p. 382). Apart from these neologisms, errors in the use of compounding patterns are of two basic types. Some compounds are produced in violation of these feature-based patterns. Thus, *lentágy ‘down bed’ (= ‘downstairs bed’) and fentágy ‘up bed’ (= ‘upstairs bed’) (MacWhinney, 1974, p. 382) use adverbs rather than the adjectives required by pattern 5. The forms *tejivó ‘milk drinker’ (= ‘pitcher for milk’) and *tejiszó ‘milk drinker’ (= ‘udder’) (MacWhinney, 1974, p. 473) use pattern 1 362

incorrectly to form nominalizations expressing instrumentality. The error *mosófog ‘washer tooth’, cf. correct fogmosó ‘tooth washer’, which incorrectly places the verb before the object (MacWhinney, 1974, p. 472). Füstölhaz ‘smokes house’ (= ‘chimneyed house’) follows pattern 5, but should have the derivational suffix -ő on füstöl to produce füstölőház ‘smoking house’. The second type of error is produced by very young children who try to separate nominals from almost any descriptive combination of words (see consequence retrieval la). Thus, we find szél-fúj ‘wind-blows’ (= ‘a gust of wind’), fúj-a-szél ‘blows-the-wind’ (= ‘a gust of wind’), kézbefog táska ‘hand-in-grab satchel’, cf. correct ridikül ‘handbag’, nem béka ‘not frog’ (= ‘the part of the violin without the frog’), semmi szél ‘no wind’, and mindig be ‘always in’ (= spiral) (pp. 472–473). The comparison between the large number of compound neologisms and errors reported for Hungarian and the paucity of similar reports for Romance (Clark, 1985) indicates that compounds are not being produced haphazardly in Hungarian. Nor are they being formed in alliance with some universal predisposition. Rather, they are being generated combinatorially by the application of systematic patterns.

8.2. Patterns (Operating Principle 4) Operating Principle 4 holds that the units juxtaposed by combination may be subjected to activation from three types of patterns: predispositions, free rules, and bound rules. 363

8.2.1. Predispositions Many polysemic patterns are probably common across language. For example, children have little trouble learning that the word for any object can also be construed to refer to the corresponding toy (see Kooij, 1971). On a higher level, the word for a literary work like Hamlet may be interpreted as referring to a book or to the literary creation found in the book. If patterns such as these are truly universal, it might be reasonable to view them as predispositions. Such predispositions would alter the interpretation of a sememe in a standard context. For example, when playing with toys, all words can be understood to refer to the corresponding toy replicas.

8.2.2. Free Rules Patterns that alter meanings in an across-the-board fashion are free rules. If such rules exist in expressive retrieval, the following consequence should hold: CONSEQUENCE SEMANTICS 4A: At least some learned rules apply obligatorily in a given phonological context without regard to lexical or allomorphic facts. The neologisms discussed in the previous section can be viewed as involving the application of free rules. For example, in a compound like money-holder for ‘purse’, the noun money is treated as the object of the verb holding. Rather than marking this fact by use of the accusative, the lexicalization of the nominalizer -ó along with a verb allows one of the polysemes of money to assume the object role. Or in the neologism malacozik 364

‘to act like a pig’ the use of the -ozik verbalizer allows malac ‘pig’ to assume an agentive role. The neologisms we have discussed can be divided into two types. In the first type, there is some resemblance between the child’s neologism and the corresponding adult form. In non-compound derivations, the child preserves the adult stem and uses the wrong affix. In the compound derivations, half of the compound is right and half is wrong. In such cases, it is extremely hard to exclude the possibility that the correct form was at least partially available to the child. Neologisms of this type seem to result from either analogy or competition between affixes in retrieval. In the second type of neologism, there is no morphemic correspondence between the child’s neologism and the adult form which pre-empts it. For example, the neologism utcanéző ‘street-looker’, cf. correct erkély ‘balcony’ has no match to the adult form. In such cases, it seems far less likely that analogy was operative. Rather, the form appears to have been produced by combining utca ‘street’ with néz ‘see’ and ő ‘nominalizer’. The other major type of free rule is the agreement rule. These rules may either activate semantic properties of one cluster onto another cluster or they may inhibit marking on one cluster on the basis of aspects of other clusters. Agreement of the noun with the number of the modifier requires that a quantity modifier takes a singular noun:

365

(Quantity (Entity)) - -> (Quantity) (Singular (Entity))) In other words, the presence of a quantity term must suppress activation of the plural sememe. There are 14 failures to apply this rule, leading to errors like *sok fák ‘many trees’ for the correct sok fa ‘many tree’ (MacWhinney, 1974, p. 523). A similar suppression involves agreement of the verb with the number of the actor such that a quantity modifier takes a singular verb. (The term “actor” is being used here to represent not only actors but also subjects of stative verbs and copulas.): ((Actor (Quantity (Entity))) (Action)) - -> ((Actor (Quantity (Entity))) (Singular (Action))) Errors such as *mért vannak ennyi lyuk ‘why are so+many hole’ for the correct mért van ennyi lyuk ‘why is so+many hole’ and sok fák van ‘many trees is’ for the correct sok fa van ‘many tree is’ show that rule 2 can be applied separately from rule 4 (agent—verb number agreement). Four errors of this type have been reported (MacWhinney, 1974, p. 524). The other side of this coin is that the child must also learn to activate agreement marking when needed. Learning of plural activation leads to a roughly equal number of errors as learning of plural suppression. Activation of plural agreement occurs to mark subject-verb agreement, so that a plurally marked noun takes a plural verb:

366

((Actor (Number (Entity))) (Action)) - -> ((Actor (Number (Entity))) (Number (Action))) There are 11 reported errors due to failure to apply this rule (MacWhinney, 1974, p. 525–527). An example is *megy a tehenek ‘go+SG the cattle’ for the correct form mennek a tehenek ‘go++ the cattle’. There is also a parallel rule for agreement in person. However, since pronominal subjects are usually deleted, errors in the use of this rule may often go unnoticed. The plural must also be activated to mark agreement of the deictic topic with the number of the comment: ((Topic (Deixis (Entity))) (Comment (Number (Entity))))- -> ((Topic (Deixis (Number (Entity)))) (Comment (Number (Entity)))) There are six errors in which the child fails to pluralize the topic to agree with the comment (MacWhinney, 1974, P. 528). An example is az galambok ‘that doves’ for the correct azok galombok ‘those doves’. The types of agreement in Hungarian we have discussed so far operate solely on number. Hungarian also has a pattern of agreement of the verb with the definiteness of the object. ((Object (Definite (Entity))) (Action)))- -> ((Object (Definite (Entity))) (Def (Action))) There are 53 reported errors in the use of this rule (MacWhinney, 1974, p.529– 535). However, this 367

number is certainly an underestimate of the true relative trequency of this error. Judging from the fuller samples presented by Meixner (1971) and MacWhinney (1974), this rule probably causes children more difficulty than any other single rule in Hungarian grammar. One should remember that every transitive sentence in Hungarian requires a decision regarding verb-object agreement and that the acquisition of definiteness as a category seems to confront the child with basic conceptual problems (Karmiloff-Smith, 1979; Mar-atsos, 1976). Given this, the high level of errors is not too surprising. Furthermore, note that in sentences with SVO word order, the verb may be produced before the object is fully lexicalized or before its definiteness is determined. No such problem occurs for agent-action (i.e., subject-verb) agreement since the agent is usually lexicalized before the verb is produced.

8.2.3. Bound rules In retrieval, bound rules serve to promote the candidacy of a given morpheme by activating one of the features that activates it or by inhibiting one of the features it does not contain. Unlike free rules, these rules are bound not to general features of semantics but to specific morphemes. This means that they cannot operate for a morpheme until its polysemic alternatives have been acquired: CONSEQUENCE SEMANTICS 4b: Bound retrieval rules will apply to new morphemes as soon as the relevant polysemes are acquired.

368

The clearest support for this consequence comes from the scores of semantically appropriate neologisms using the verbal formatives -oz and -ol. These formatives place the noun stem in one of the five polysemic roles: (1) actor, (2) instrument, (3) result, (4) location, or (5) object transferred. The neologisms reported in MacWhinney (1974, pp. 366–371) indicate that children soon learn to use -oz and -ol appropriately in each of these alternative ways. For example, children can produce neologisms meaning things like ‘to act like a blacksmith’, ‘to use soap bubbles’, ‘to make a spot’, ‘to act like at school’, and ‘to make into a bride’. Although bound rules are highly limited in their applicability, they are quite reliable in their application. Therefore, the errors in the use of bound polysemic rules should be quite limited: CONSEQUENCE SEMANTICS 4C: Errors in the shape of the context of bound polysemic rules (as distinguished from reformulations as modifications) are quite rare. In fact, the only reported error in the shape of a bound polysemic rule is the neologism szépekedik ‘pretty + inchoative’ in which a non-process adjective is used as the basis for the selection of the inchoative polyseme of -kedik (MacWhinney, 1974, p. 372). However, this is probably a deeper semantic error rather than an polysemic error. That is, the child seems to be viewing ‘pretty’ as a process rather than a state.

369

8.3. Competition (Operating Principle 5) When we were discussing competition in phonology, we noted that children often end up using two allomorphs of a given morpheme. When discussing syntax, we found that the competition between morphemes for slots could lead to misorderings. In semantics, sememes compete for the opportunity to express underlying semantic intentions. Here, competition may lead to a couple of possible consequences. One is the child may choose some analytic lexicalization because he has not learned how to make the weaker rote lexicalization suppress the analytic lexicalization. CONSEQUENCE SEMANTICS 5A: When a given meaning could conceivably map onto two different lexicalizations, at first the child makes errors in selecting the correct alternative. There are at least four competitions of this type reported in the Hungarian literature. Each has several reported errors. 1. The suffix -ai expresses third person singular possession of plural objects. This suffix competes semantically with the combination of the plural suffix -ek and the third person singular possessive -je. Thus, errors like *tete-jé-k ‘roof-POSS-PL’ (MacWhinney, 1974, p. 353) compete with correct forms like tetei. Similarly, -aim ‘1SG.POSS-PL’ can be viewed as an alternative to -ak ‘PL’ and -am ‘1SG.POSS’. The error *többi-ek-em ‘others-PL-1SG.POSS’ (MacWhinney, 1974, p. 517), cf.

370

correct több-eim ‘others-PL: 1SG.POSS’ is the result. Also -ei ‘plural possession of singular possessor’ competes with -e ‘sign of possessor’ and -k ‘PL’ and we find kiék ‘whose-POSS-PL’ for kiéi ‘whose-POSS.PL’ (MacWhinney, 1974, p. 517). 2. The portmanteau negatives nincs(en), sent, sincs(en), and sincsenek express the same semantic content as the expressive analytic forms nem van ‘not is’, *nem is ‘not also’, *nem is van ‘not also is’ and *nem is vannak ‘not also are’. Errors involve use of the analytic forms instead of the portmanteau form or use of both at the same time. Clancy (1985) reports similar errors for negative portmanteaus in Japanese. In Romance, portmanteaus such as du and au are analyzed into *de le and *á le. Clark (1985) notes that French and Romanian children use analytic forms such as de moi and á elle rather than single forms like mon and lui in the early school years. However, unlike the Hungarian analytic forms, these forms are often acceptable in adult French and Romanian. 3. The dative pronoun stem nek- takes the ending -ik in the third person singular rather than the usual -ük. However, children often produce *nekük ‘DAT+3PL’ for nekik ‘DAT+3PL’ (MacWhinney, 1974, p. 398). 4. The normal second person singular suffix is -d. In the imperative this -d suppresses use of imperative marker -j in the second person singular. However, in some child errors both are present. Thus, *ad-j-ad ‘give-IMP-2SG’ appears instead of ad-d ‘give-2SG.IMP’ (MacWhinney, 1974, p. 398). 371

A second consequence of Operating Principle 5 is that, when a meaning could be expressed in two different ways, the child activates the morphemes for each alternative and ends up with semantic redundancy. CONSEQUENCE RETRIEVAL 5B: When a meaning could be expressed in two different ways, the child may produce redundancies. I have reported (MacWhinney, 1974, pp. 354, 538–544) 22 errors in which there is some type of semantic redundancy. For example, the word holnap ‘tomorrow’ codes location in time. In Hungarian, as in English, one says ‘on Thursday’, but not ‘on tomorrow’. Thus, *holnapon ‘tomorrow+SUPER’ is a case of suffix redundancy. Similarly, vagy ‘you are’ is the 2PS copular verb. Addition of the 2PS suffix to produce *vagyol is a redundancy which indicates that the semantic analysis of vagy is incomplete. Redundancy may also occur between two stems as in *sem nem nincsen ‘or not it+isn’t’. Hungarian permits multiple negation, but each negative must attach to its own constituent. The form *sem nem nincsen is a triple redundancy. These Hungarian redundancies are similar to a type of error in retrieval that is fairly widely reported in French (Clark, 1985, Karmiloff-Smith, 1979). For example, Clark notes that, in French, possession may be marked by the possessive adjective (mon) by a possessive nominal (le mien), or by a propositional phrase (á moi). Until the child learns to assign different functions to these

372

devices, errors such as mon mien de chapeau á moi may occur. In such errors, the forms the child is retrieving are redundantly expressing the same semantic content in slightly different ways. The child must learn that retrieval of one of these forms should inhibit retrieval of the others.

8.4. Rule Acquisition (Operating Principles 10 and 15) When a morpheme has more than one polyseme, the most frequently used alternative should be the strongest. CONSEQUENCE SEMANTICS 10A: The first productive uses and the first semantic overgeneralizations of an affix make use of the most frequent polyseme. Thus, the neologisms based on the nominalizer -ó are chiefly extensions of the instrumental and agentive polysemes, rather than the locative polyseme. Similarly, most of the neologisms using -oz and -ol are agentive and instrumental. Consequence semantics 10a also applies to the polysemes of case markers. For example, in Japanese, the strongest polyseme of the postposition o is that of the direct object. However, for verbs of nonintentional perception, ga marks the object. Children err in such cases by using o to mark the object (Clancy, 1985). Similarly, the strongest polyseme of ga is its use as an agent marker; its uses as a marker of the recipient or a marker of the object of perception verbs are acquired somewhat later. If some of the polysemes of a sememe 373

are not known to the child, he may use circumlocutions to express the same meaning that can be found in a simple term. For some errors of this type in French, see Clark (1985). Although we do not have enough data to state the order of acquisition of polysemic rules in Hungarian in detail, there is enough evidence to support the following consequence, parallel to Consequence semantics 10a above: CONSEQUENCE SEMANTICS 10B: Bound polysemic rules that apply fairly generally will be acquired initially as free rules. In support of this, we note that the selection between the polysemic meanings of the denominative verbalizers -oz and -ol is occasionally overgeneralized to the verbalizer ít. The errors are vödörít ‘to use a bucket’ and elmáskárit ‘to use a mask’ (MacWhinney, 1974, p. 372). Since ít does not have an instrumental polyseme, these forms are errors. Children have assumed that, since two verbalizers have instrumental readings, all verbalizers might have such readings. The selection between a neutral and an affirmative response reading of the verbal separable prefix is also occasionally generalized as a free rule. Thus, the answer to the question should we not gather them together? should be but yes, together. However, the child responded with but yes, gather (MacWhinney,

374

1974, p. 481). Here, the child seems to have assumed that since verbal prefixes can be used in this way to answer questions, verbs can be used this way too. Bowerman (1975) and Lord (1979) have investigated in great depth the overextension of the English pattern of zero-derivation of causative transitives from intransitives as in transitive open from intransitive open. In the present model, this pattern can be seen as a rule which activates one of the semantic features of which a morpheme is composed. At first, the pattern overgeneralizes to suppress rote causatives like drop and bring. Eventually, the pattern must be acquired as a bound rule. Bowerman (1982) notes that her daughter Christy had eight verbs which might have served as the basis for this rule. She also notes that, even once the pattern is reined in, it continues to show some productivity (Bowerman, 1982). Moreover, adult errors of this type are far from uncommon (Stemberger, 1982). In Hungarian, transitives may not be formed by zero-derivation from intransitives. Nevertheless, early in the third year, there are seven reported errors in which intransitives are used for the corresponding lexical transitives (elbúj ‘hide’, gurul ‘roll’, alszik ‘sleep’, and nött ‘grew’, (MacWhinney, 1974, p. 546); ül ‘sit’, kel ‘rise’ , feküd ‘lie’, MacWhinney, 1974, p. 547). There are also two errors in which intransitives are conflated into transitives (‘reach it out’, MacWhinney, 1974, p. 546; ‘telegraph someone’, MacWhinney, 1974, p. 547). Although Hungarian does not form transitives from intransitives by zero-derivation, it does use transitives without direct objects. Just as in English, Hungarians can say ‘John eats’ without specifying what he is eating. This 375

use of some verbs both with and without objects may induce children to form a polysemic rule that allows any verb to be used with an object. In Hungarian, such a rule would be more short-lived than in English. In the competition model, many cases of semantic extension can be explained in terms of the formulation of overly general polysemic rules. However, most cases simply involve errors during competition between morphemes for activation. If two morphemes share 14 features in common and differ on only one, any error in the activation of this one feature can lead to the activation of the wrong morpheme. The competition model offers a very powerful account for the detailed patterns of overgeneralization and undergeneralization found in semantic development. When a child takes a form to be more plurifunctional than it really is, this should be because he has not yet acquired the necessary competitors or because the competitors are still weak. For example, the child could use the singular for both plural and singular reference (see Bowerman, this volume). This is not because singulars are coded as plurals, but because no strong competitors to the single have yet been learned. Similarly, the child will use “synthetic” forms that code too much material simply because these forms have not yet been analyzed and no competitors are higher in specificity and accuracy. Unfortunately, a full discussion of the issue of semantic extension lies outside the scope of this chapter. In particular, this chapter passes over data on the overextensions

376

of stems in Hungarian. For the most part, such overextensions are parallel to those so thoroughly documented for Indo-European by Clark (1973).

8.5. Item Acquisition (Operating Principle 11) Operating Principle 11 holds that “children cluster concepts in terms of their semantic relatedness.” The most likely cluster, in most cases, is the one corresponding to a simple object or action. Thus, in many cases, children think they are learning simple nouns or verbs even though additional material is attached. Failing to detect the presence of this additional material, children may then use stems with inflections when only bare stems are required. Alternatively, children may actually add new material that contradicts material present in the unanalyzed rote form. CONSEQUENCE SEMANTICS 11A: Errors in the association of a set of semantic features with a phonological string in early lexical items are revealed by contradictions and superfluities. I have cited five errors (MacWhinney, 1974, pp. 353, 548) in which the child takes an item with insufficient semantic detail and adds on a suffix which contradicts an affix already in the word. Thus, *szá-d-am-ba ‘mouth-yours-my-in’ is used for szá-m-ba ‘mouth-my-in’. Evidently the child assumed that sza-d ‘mouth-my’ actually referred to just ‘mouth’. In 36 other errors (MacWhinney, 1974, pp. 346–348, 476–478, 592–593), Hungarian children use stems with 377

superfluous suffixes attached. The reports from Kenyeres and Balassa indicate that, for a period, children produce the relevant errors with some consistency. Each stem is used with the suffix it “deserves.” Tools have superfluous instrumentals attached (‘saw-INSTR’ is used for ‘saw’); body parts and clothes have possessives (‘eye-3SG.POSS’ is used for ‘eye’); foods have accusatives (‘bread-ACC is used for ‘bread’); and locations have locatives (‘park-INESS’ is used for ‘park’). Even verb phrases may be picked up as under-analyzed amalgams. Thus, Balassa’s son used ‘open the door’ and ‘close the door’ as if they meant ‘open’ and ‘close’. In addition, we should note that the redundancies and reduplications discussed under Consequence phonology 5c could also be viewed as resulting from improper attachment of sound to meaning in amalgam acquisition. However, it seems to me that we make fewer tenuous assumptions by viewing reduplications and redundancies as products of allomorph competition. On the other hand, competition does not seem to be a reasonable account for superfluities and contradictions, since these errors point to some defect in the semantic encoding of lexical items. According to Operating Principle 11, clarity of phonological and semantic segmentation is the chief determinant of the initial acquisition of free morphemes. Although clarity of semantic segmentation is extremely hard to define, intuitively, there seem to be at least some words that involve complex, hard-to-segment relations between abstract concepts. Items expressing such meanings should be acquired fairly late. 378

CONSEQUENCE SEMANTICS 11B: Items expressing meanings that extend across stretches of cognitive material will be acquired late. Kenyeres (1928) made detailed observations regarding his daughter Eva’s lexical development up to age 8;6. Table 14.8 lists 34 lexical items that were learned between 3;6 and 8;6. These late acquisitions come from two semantic fields. The first involves the representation of complex temporal relations. These include terms such as ‘tomorrow’, ‘afternoon’, and ‘in the evening’. As Cromer (1976) has noted, the lateness of the acquisition of such terms is not a result of their low frequency, but their cognitive complexity. While clearly complex cognitively (Cromer, 1976), these terms do not present the child with segmentation problems. However, segmentation problems do seem to arise in the learning of the other words in Table 14.8. For example, in order to learn the use of ‘nonetheless,’ the child needs to be able to relate one event to another potentially disenabling event. Use of ‘anyway’ is similar but also suggests that the limitation is being specifically or purposefully ignored. In order to use ‘either … or’, the child must be able to see two events as alternatives. Temporals such as ‘during’ and ‘while’ require that one full action be used as background to another. Words like ‘to be sure’ and ‘indeed’ require certain acknowledgments of views of the listener and the ways they fit in with the claims of the speaker.

379

8.6. Amalgam Analysis and Morpheme Acquisition (Operating Principle 12) The issue of the order of emergence of inflections has been a central question in recent child language research. Table 14.9 from MacWhinney (1976) summarizes the order of emergence of the 29 earliest inflections in Hungarian. Suffixes of the first group generally emerge before those of the second group, and so on. However, within a group, the order of emergence is indeterminate. Table 14.10 summarizes the order of acquisition of the first inflections from my subject Zoli between 1;5.6 and 2;2.3, using the scoring system of Cazden (1968). The six samples involved contain 4 to 8 hours of free-speech data apiece. There is a reasonably close correspondence between Tables 14.9 and 14.10. The first group of suffixes emerges in periods III and IV. The second group is distributed across periods II to V, with a concentration in IV and V. The third group is evenly distributed across periods IV, V, and VI. Most of the fourth group and one suffix in the third group have not yet emerged at period VI. The differences between Tables 14.9 and 14.10 could reflect peculiarities in Zoli’s individual development. They could also reflect differences between Cazden’s criterion and the criteria of the diary studies. For example, it is not clear how to define obligatory contexts for the diminutive. Nor are the contexts of the 3PS possessive always clear. Even more problematic is the occurrence of acquisition without productivity. Zoli acquired the past tense in period II, but there was no evidence of productivity until period VI.

380

However, the accusative was acquired in period III and demonstrated productivity in that same sample. TABLE 14.8 Late Acquisitions of Eva Kenyeres WordTranslation

Age

ugyanolyan just like

3;6

nappal

in the day

3;6

éjjel

in the night

3;6

helyett

instead of

4;0

felé

toward

4;0

délelött

forenoon

4;1

délután

afternoon

4;1

tégnap

yesterday

4;1

ma

today

4;1

holnap

tomorrow

4;1

holnap után day after tomorrow

4;6

tegnap elött day before yesterday

4;8

addig … amig

as long as

4;11

mihelyt

as soon as

5;0

óta

since

5;0

régóta

since a long time ago

5;0

meddig?

until when?

5;1

mire

by that time that

5;1

amint

as soon as

5;2

aközben

during

5;4

381

múlva

past

5;7

túl

beyond

6;0

kivül

besides

6;0

ugyan

anyway

6;4

söt

indeed

6;5

azonban

however

6;5

ellen

against

6;6

miatt

because of

6;6

iránt

toward

6;6

taján

in the vicinity of

6;6

hosszatt

long (time)

6;6

mind … mind

both … and

6;7

hát

to be sure

6;8

mivel

inasmuch as

6;8

-hoz képest in comparison with

7;10

mióta

since

8;1

miközben

while

8;3

mialatt

while

8;6

miután

after

8;6

ellenben

on the contrary

8;6

akár … akár

either … or

8;6

TABLE 14.9 Generalized Order of Emergence of Early Inflections First group Accusative

Plural

382

Diminutive

Allative ‘to’

Illative (Inessive) ‘to’

Second

Dative

Illative ‘to’

Instrumental

Group

Past Tense

1PS Indefinite

Infinitive

Third group

Sublative ‘onto’

Superessive ‘on’

1PS Possessive

3PS Possessive Sign of Possession 1PP Indefinite 3PP Indefinite 1PS Definite Fourth group

1PP Definite

Elative ‘from in’

Adessive ‘toward’ Causal-Factitive

Ablative ‘from’ 3PS Definite

Conditional

3PP Definite

Assuming that Tables 14.9 and 14.10 can be taken as the basic pattern of Hungarian development, we are led to ask several further questions. First, why do the first locatives express ‘motion towards’? Both MacWhinney (1974) and Meggyes (1971) found that ‘position at’ was coded by locative deictics and that ‘movement from’ was seldom mentioned before 2;2. A second question is why do indefinite suffixes tend to enter before definite suffixes? It should be noted that the indefinite is used for all intransitives, whereas transitives may be definite or indefinite. What is the role of this division? Thirdly, role suffixes like the dative, accusative, and instrumental appear quite early. Is this due to their pragmatic importance or are they somehow functionally basic? TABLE 14.10 Emergence of Inflections in Zoli from 1;5.2 to 2;2.3

383

Period

Number of utterances

New acquisitions

I 1;5,3–1;5,5

51

II 1;6,29–1;6,30

228

III 1;8,6–1;8,8

2675

Instrumental, Diminutive, Allative

IV 1;10,0–1;10,6

1911

Sign of Possession, Plural, Accusative 1PS Poss., Illative, 2PS Def. Imper., 1PP Indef., 3PP Def. Imper., Infinitive

V 2;0,0–2;0,5

835

Dative, 2PS Poss., 3PS Poss., IPS Def., 1PS Indef., Def. Article

VI 2;2,0–2;2,3

1826

— Past Tense

2PS Indef., 3PS Def., 1PP Def., 3PP Def., 3PP Indef., Sublative

Taken by itself, the principle of relative ease of analysis started in Operating Principle 12 cannot fully account for the order of acquisition of grammatical morphemes. On the one hand, Operating Principle 11 holds that markers will only be acquired if they are picked up as parts of amalgams. Thus, using both Brown’s data for English and new data for Quiche Mayan, Pye (1979, 1980, p. 57) has found that perceptual saliency is a better predictor of marker acquisition than either frequency or semantic complexity. In Hungarian, nearly all of the relevant markers are monosyllabic suffixes. Therefore, perceptual salience is similar across markers. It may be that, in Hungarian, the order of acquisition of grammatical markers is determined by the factors of reliability and applicability (Operating Principle 10).

384

8.7. Inference (Operating Principle 13) Operating Principle 13 holds that, during parsing, the child can infer aspects of the semantics of new items on the basis of the words with which they are concatenated by ordering patterns. For example, given the phrase my fungo, we know that fungo should refer to an object. Alternatively, if we wish to use formal rather than functional terms, we may say that fungo is a noun. However, many frames are ambiguous in regard to the semantic/formal class of the items they allow. For such frames, we expect children to produce erroneous assignments to semantic class. CONSEQUENCE SEMANTICS 13A: Wherever the language has ambiguous co occurrence frames, children make errors in assigning words to semantic class. For example, English-speaking children Often analyze the word behave into be + have. Since be has a frame that takes a following stative adjective, children assume that have is a stative adjective and produce errors like I’m being have (Peters, 1980, 1985). In Hungarian, I have reported (MacWhinney, 1974, pp. 549–552) 36 errors in assignment of items to a part of speech. The nature of these errors is particularly interesting. Let us take as an example the child’s use of savanyú ‘sour’ to refer to szőlő ‘grape’. One possible source of this error could be the positional pattern EZ + ENTITY ‘this + entity’ which at an early period is often used in the “naming game.” From this positional pattern, the child infers that savanyú is an entity. Later, when

385

s/he has learned the form EZ + MODIFIER, he will avoid such erroneous inferences. However, in the meantime, he may use savanyú for a short period as a noun. Similarly, possessive pronouns, numerals, and modifiers all occur in comment position. However, only modifiers occur prenominally. If a possessive pronoun or numeral is learned from a sentence in which it is the comment, the child may conclude that it is a modifier. As a result, s/he may erroneously place it in prenominal position. There are several errors of this type (MacWhinney, 1974, pp. 553–553). Examples include *enyém virág ‘mine flowers’, cf. correct virág-om ‘flower-my’, *kettő bárány, cf. correct két bár-ány) ‘two sheep’, and Mártikáé almája ‘Mártika’s apple+POSS’, cf. correct Mártika almája ‘Mártika apple+POSS’. Other errors that may be due to ambiguity in acquisitional frames include leül ‘down+y’ (= ‘down’) (MacWhinney, 1974, p. 520), egy homokot ‘a sand-+ACC (MacWhinney, 1974, p. 518), leig ‘down+until’ (MacWhinney, 1974, p. 521), belülje ‘insidey+its’ (MacWhinney, 1974, p. 521), and piszokok ‘dirt-+PL’ (MacWhinney, 1974, p. 544). There are four errors which resulted from failure to code certain adjectives as non-comparable: csepebb ‘weer’, túlabb ‘beyonder’, elégebb ‘enougher’ (MacWhinney, 1974, p. 511), and csomóbb ‘buncher’ (MacWhinney, 1974, p. 519). There are also two errors which arise from treating hamár ‘soon’ as an adjective (MacWhinney, 1974, p. 520). The use of the reflexive with inanimates as in *‘the package opened itself or *‘the stone threw itself (MacWhinney, 1974, p. 364) may derive from analogy with acceptable structures like ‘the door closed itself. 386

Similarly, the error *sokenként ‘many at a time’ that was discussed above (MacWhinney, 1974, p. 512) can be based on analogy with egyenként ‘one at a time’. Errors using the nominal form of case suffixes with pronouns are extremely common. Thus, in adult Hungarian, ‘with Imre’ is Imrével Imre+COM’ but ‘with you’ is veled ‘COM+2SG’. However, children take a pronominal base of the possessive such as ti and then add a nominal suffix, producing *tivel ‘2SG+COM’. The feature [+ pronominal] or [+deictic] fails to make the correct separation between the contexts of -vel and vel-, since the pronouns az and ez ‘this’ and ‘that’ take the nominal suffix as in avval and evvel. Moreover, the third person pronouns maga and maguk also act like nouns in magával and magukkal. It is clear then that the distinction between pronouns that take the prefix and ones that take the suffix is semantically opaque. Moreover, the availability of both personal stems (like ti-) and personal suffixes (like -d) forces the child to specify the context of these forms by bound rules. Inference of semantic markers from positional patterns can also work to control arbitrary formal classes such as conjugation and declension. In Hungarian, the only productive morphological marking for an arbitrary formal class is for the group of verbs that take -ik in the third person singular present indefinite indicative. These -ikes verbs also show five further irregularities in their indefinite conjugation. 1. -am, -em instead of -ek in the first person singular conditional. 387

2. -m instead of -k in the first person singular present. 3. -m instead of -k in the first person singular imperative. 4. -al instead of zero in the second person singular imperative, and 5. ék instead of on in the third person singular imperative. Because of these correlated properties, this group of verb stems can be reasonably viewed as a conjugational paradigmatic class (Maratsos & Chalkley, 1980). Inference allows the child to construct feature markers which can be entered into the lexical encoding for the verbs of the -ikes paradigm. There seem to be a variety of semantic and phonological features which may also give support to assignment of a verb to the -ikes conjugation, but this has not yet been adequately examined in Hungarian. However, there is evidence in German (MacWhinney, 1978; Köpcke and Zubin, 1982) that gender selection, a similar paradigmatic phenomenon, is far from arbitrary and that phonological and semantic cues summate activation to select the feature of gender which is then attached to the noun. However, these predicators are not fully determinate and there are exceptions. Therefore control is best when gender is computed for each item separately. Given this, we expect the following consequence:

388

CONSEQUENCE SEMANTICS 13B: Children will make many errors when morphological marking is governed by formal class membership. In fact, errors in the use of the -ikes verbs continue into adolescence. However, because some dialects fail to utilize the -ikes conjugation, this late acquisition may be viewed, at least in part, as a result of sociolinguistic variation.

9. SUGGESTIONS FOR FURTHER STUDY If one compares our knowledge of the acquisition of Hungarian with our knowledge of the acquisition of English, it becomes clear that there are at lest two major gaps in the Hungarian research. First, there exists almost no literature on the development of comprehension strategies in Hungarian children. Work by MacWhinney, Pléh, and Bates (in press) on the development of sentence comprehension strategies begins to close this gap. However, it is clear that work on the resolution of polysemy in noun phrases, prepositional phrases, adverbial phrases, and derivational formations will also be needed. A second major gap in the Hungarian literature is in the area of the acquisition of conversational functioning. There are very few studies in Hungarian of the development of responses to questions, elaborations of narrative structures, modes of argumentation, control of politeness, attempts to elicit information, monitoring of communication error and success, mother-child interaction, or verbal play. This is not to say that there is 389

no literature on these topics. However it is very difficult to find much in this literature that reflects the distinctly Hungarian nature of the ways that Hungarian children come to control these abilities. At the same time it is intuitively clear that conversational functioning in Hungarian uses forms and rules that are quite different from those of English. In addition to these two major gaps, the Hungarian language provides us with structures that deserve further intensive research. In particular, further research on the child’s acquisition of the definite-indefinite contrast in the conjugation of the verb could prove quite interesting. In the area of morphophonology, there is a need for experimental studies using techniques like those of MacWhinney (1978) in which the nonce form is, however, not a stem but an affix. This would provide a useful way of assessing the true productivity and scope of a number of phonological rules whose description is presently incomplete. Finally, there is a need for studies of the productivity and semantic form of derivational affixes in school age children. Here there are a number of developments that may not be completed until late childhood.

10. THEORETICAL IMPLICATIONS This review has summarized the basic facts that can be found in literature on Hungarian language acquistion. The two important exclusions have been data on phonological development and studies of the acquisition of sociolinguistic competence. For the three basic areas

390

of the grammar, the competition model has provided a useful framework for both describing and explaining the central facts about Hungarian development. Moreover, many of the data we have gathered from other languages can also be satisfactorily explained by the model. Of course, it is clear that the model is far from complete. Nevertheless, it seems to me that real progress can be made within a framework of this type toward the important goal of understanding the acquisition of the world’s languages by the world’s children.

ACKNOWLEDGMENT Preparation of this review was supported by a grant from the National Institute of Mental Health MH 31160-02. I would like to express my appreciation to Ruth Berman, Dan Bullock, Joan Bybee, Eve Clark, László Dezső, Suzanne Gendreau, Lise Menn, Ann Peters, Matthew Rispoli, Ron Schaeffer, and Joe Stemberger for their excellent detailed comments on the earlier versions of this chapter. My thanks also to Dan Slobin for his extensive and perceptive editorial assistance and continuing intellectual support. Csaba Pléh, and Gusztáv Habermann, of the ELTE Psychology Department have worked closely with me in my experimental studies in Hungarian. Laszló Dezső, György Szépe, Zita Réger, and Zsolt Lengyel of the Linguistics Institute of the Hungarian Academy of Sciences have furthered my understanding of the structure of Hungarian and provided crucial assistance for data collection. Erzsébet Polónyi and Katalin Forrai allowed me to work in the nursery school of the National Nursery School

391

Methodological Institute. Mrs. József Szvitil, Mrs. Tibor Patzek, Edit Géra, and Erzsébet Várga also deserve my thanks for allowing me to work in their schools. Thanks to the many children who talked and played with me in their schools and to my good friends the Névai’s for making it all much more fun. And, finally, my appreciation to my grandparents, Dezső and Celia Nanasy, for awakening my interest in Hungarian language and culture.

REFERENCES Anderson, J. A.C.T. Cambridge, MA: Harvard University Press, in press. Bánhidi, Z., Jókay, Z., & Szabó, D. Learn Hungarian. Budapest: Tankönyvkiadó, 1965. Bates, E. Language and context: Studies in the acquisition of pragmatics. New York: Academic Press, 1976. Bates, E., & MacWhinney, B. Functionalist approaches to grammar. In E. Wanner & L. R. Gleitman (Eds.), Language acquisition: The state of the art. New York: Cambridge University Press, 1982. Berko, J. The child’s learning of English morphology. Word, 1958, 14, 150–177. Berman, R. A. The acquisition of Hebrew. In D. I. Slobin (Ed.), The crosslinguistic study of language

392

acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Bever, T. J. The cognitive basis for linguistic structures. In J. R. Hayes (Ed.), Cognition and the development of language. New York: Wiley, 1970. Bock, K. Toward a cognitive psychology of syntax: Information processing contributions to sentence formulation. Psychological Review, 1982, 89, 1–47. Bowerman, M. F. Cross-linguistic similarities at two stages of syntactic development. In E. H. Lenneberg & E. Lenneberg (Eds.), Foundations of language development: A multidisciplinary approach. New York: Academic Press, 1975. Bowerman, M. Reorganizational processes in lexical and syntactic development. In E. Wanner & L. R. Gleitman (Eds.), Language acquisition: The state of the art. New York: Cambridge University Press, 1982. Braine, M. D. S. The ontogeny of English structure: The first phase. Language, 1963, 39, 1–13. Braine, M. D. S. Children’s first word combinations. Monographs of the Society for Research in Child Development, 1976, 41, Whole No. 1. Brown, R. A first language: The early stages. Cambridge, MA: Harvard University Press, 1973.

393

Brown, R., Cazden, C., & Bellugi, U. The child’s grammar from I to III. In J. P. Hill (Ed.), Minnesota symposia on child development. Minneapolis: University of Minnesota Press, 1968. Bybee, J. Child language and morphophonemic change. Linguistics, 1978, 17, 21–50. Bybee, J. Morphophonemic change from inside and outside the paradigm. Lingua, 1980, 50, 45–59. Bybee, J., & Brewer, M. Explanation in morphophonemics: Changes in Provençal and Spanish preterite forms. Lingua, 1980, 52, 271–312. Bybee, J., & Pardo, E. On the lexical and morphological conditioning of alterations: A nonce-probe experiment with Spanish verbs. Linguistics, 1981, 19, 937–968. Bybee, J., & Slobin, D. I. Rules and schemas in the development and use of the English past tense. Language, 1982, 58, 265–289. Cazden, C. B. The acquisition of noun and verb inflections. Child Development, 1968, 39, 433–438. Clancy, P. The acquisition of Japanese. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Clark, E. On the acquisition of the meaning of before and after. Journal of Verbal Learning and Verbal Behavior, 1971, 10, 266–275.

394

Clark, E. V. What’s in a word? On the child’s acquisition of semantics in his first language. In T. E. Moore (Ed.), Cognitive development and the acquisition of language. New York: Academic Press, 1973. Clark, E. V. The acquisition of Romance, with special reference to French. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1), Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Clark, E. V., & Clark, H. H. When nouns surface as verbs. Language, 1979, 55, 767–811. Clark, R. What’s the use of imitation? Journal of Child Language, 1977, 4, 341–358. Cromer, R. Developmental strategies for language. In V. Hamilton & M. Vernon (Eds.), The development of cognitive processes. New York: Academic Press, 1976. Dezső, László. A gyermeknyelv mondattanának elvi-modszertani kerdései. Általános nyelvészeti tanulmányok, 1970, 7, 77–99. Ervin, S. M. Imitation and structural change in children’s language. In E. H. Lenneberg (Ed.), New directions in the study of language. Cambridge, MA: MIT Press, 1964. Fava, E., & Tirondola, G. Syntactic and pragmatic regularities in Italian child discourse: Grammatical

395

relations and word order. 1977. manuscript: Instituto di Glottologia, Italy.

Unpublished

Feurer, H. Morphological development in Mohawk. Papers and Reports on Child Language Development (Dept. of Linguistics, Stanford University), 1980, 18, 25–42. Firbas, J. On defining the theme in functional sentence. Travaux Linguistique de Prague, 1964, 1, 267–280. Fónagy, I. A propos de la genése de la phrase enfantine. Lingua, 1972, 30, 31–71. Graves, M., & Koziol, S. Noun plural development in primary grade children. Child Development, 1971, 42, 1165–1173. Guillaume, P. Les débuts de la phrase dans le langage de l’enfant. Journal de Psychologie, 1927, 24, 1–25. (Translated in C. A. Ferguson & D. I. Slobin. Studies of child language development. New York: Holt, Rinehart, & Winston, 1973.) Hyman, L. On the change from SOV to SVO: Evidence from Niger-Congo. In C. Li (Ed.), Word order and word order change. Austin: Texas University Press, 1975. Ingram, D. Phonological patterns in the speech of young children. In P. Fletcher & M. Garman (Eds.), Language acquisition: Studies in first language development. New York: Cambridge University Press, 1979.

396

Karmiloff-Smith, A. Afunctional approach to child language: A study of determiners and reference. New York: Cambridge University Press, 1979. Kenyeres, E. A gyermek első szavai és a szófajok föllépese. Budapest: Kisdednevelés Kiadása, 1926. Kooij, J. G. Ambiguity in natural language: An investigation of certain problems in its linguistic description. Amsterdam: North-Holland Publishing Co., 1971. Köpcke, K., & Zubin, D. Die cognitive organisation der Genuszuweisung den einsilbigen Nomen der deutschen Gegenwartssprache. Zeilscrift für germanistische Linguistik, 11, 166–182. Kuczaj, S. Children’s judgments of grammatical and ungrammatical irregular past-tense verbs. Child Development, 1978,49, 319-326. Kunene, E. The acquisition of SiSwati as a first language: A morphological study with special reference to noun prefixes, noun classes, and some agreement markers. Unpublished doctoral dissertation, University of California, Los Angeles, 1979. Lederberg, A., & Maratsos, M. Generalization of a linguistic rule by preschool children: An experimental study. Southwestern SRCD Meeting, Lawrence Kansas, 1980.

397

Lightbrown, P. M. Consistency and variation in the acquisition of French: A study of first and second language development. Unpublished doctoral dissertation, Columbia University, 1977. Li, C., & Thompson, S. The acquisition of tone in Mandarin-speaking children. Journal of Child Language, 1976, 4, 185–199. Lord, C. “Don’t you fall me down”: Children’s generalizations regarding cause and transitivity. Papers and Reports on Child Language Development (Dept. of Linguistics, Stanford University), 1979, 17, 81–89. MacWhinney, B. How Hungarian children learn to speak. Unpublished doctoral dissertation, University of California, Berkeley, 1974. MacWhinney, B. Pragmatics patterns in child syntax. Papers and Reports on Child Language Development (Dept. of Linguistics, Stanford University), 1975, 10, 153–165. (a) MacWhinney, B. Rules, rote, and analogy in morphological formations by Hungarian children. Journal of Child Language, 1975, 2, 65–77. (b) MacWhinney, B. Hungarian research on the acquisition of morphology and syntax. Journal of Child Language, 1976, 3, 397–410.

398

MacWhinney, B. The acquisition of morphophonology. Monographs of the Society for Research in Child Development, 1978, 43, Whole no. 1. MacWhinney, B. Basic syntactic processes. In S. Kuczaj (Ed.), Language acquisition: Syntax and semantics. Hillsdale, NJ: Lawrence Erlbaum Associates, 1982. MacWhinney, B. Grammatical devices for sharing points. In R. Schiefelbusch (Ed.), Communicative competence: Acquisition and intervention. Baltimore, MD: University Park Press, 1984. MacWhinney, B., & Bates, E. Sentential devices for conveying givenness and newness: A cross-cultural developmental study. Journal of Verbal Learning and Verbal Behavior, 1978, 17, 539– 558. MacWhinney, B., Bates, E., & Kliegl, R. The impact of cue validity on sentence interpretation in English, German, and Italian. Journal of Verbal Learning and Verbal Behavior, in press. MacWhinney, B., Pléh, Cs., & Bates, E. The development of sentence interpretation in Hungarian. Cognitive Psychology, in press. MacWhinney, B., & Sokolov, J. The learning of syntax. In B. MacWhinney (Ed.), Mechanisms of language acquisition. Hillsdale, NJ: Lawrence Erlbaum Associates, in press.

399

Maratsos, M. P. The use of definite and indefinite reference in young children. Cambridge: Cambridge University Press, 1976. Maratsos, M., & Chalkley, M. The internal language of children’s syntax: The ontogenesis and representation of syntactic categories. In K. Nelson (Ed.), Children’s language. New York: Gardner Press, 1980. Meggyes, K. Egy kétéves gyermek nyelvi rendszere. Nyelvtudományi Értekezések, 1971, Vol. 73. Meixner, I. Gyermeknyelvi hibák. Doctoral dissertation, Gyógypedagógiai föiskola, Budapest, 1971. Menyuk, P. Sentences children use. Cambridge, MA: MIT Press, 1969. Newport, E. L., & Meier, R. P. The acquisition of American Sign Language. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Ohala, J. Phonetic explanation in phonology. In A. Bruck et al. (Eds.), Papers from the parassession on natural phonology. Chicago: Chicago Linguistic Society, 1974. Omar, M. The acquisition of Egyptian Arabic as a native language. The Hague: Mouton, 1974.

400

Park, T.-Z. A study of German language development. 1974. Psychological Institute, Bern: Unpublished manuscript. Peters, A. The units of language acquisition. In Working Papers in Linguistics. Hawaii: University of Hawaii at Manoa, 1980. Peters, A. M. Language segmentation: Operating principles for the perception and analysis of language. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. II). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Pye, C. L. The acquisition of grammatical morphemes in Quiche Mayan. Unpublished doctoral dissertation, University of Pittsburgh, 1979. Richards, M. Adjective ordering in the language of young children: An experimental investigation. Journal of Child Language, 1979, 6, 253–278. Slobin, D. I. Leopold’s bibliography of child language: Revised and augmented edition. Bloomington: Indiana University Press, 1972. Slobin, D. I. Cognitive prerequisites for the development of grammar. In C. A. Ferguson & D. I. Slobin (Eds.), Studies of child language development. New York: Holt, Rinehart, & Winston, 1973.

401

Slobin, D. I. Crosslinguistic evidence for the Language-Making Capacity. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. II). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Smoczyńska, M. The acquisition of Polish. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. I). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Snow, C. E., Smith, N. S. H., & Hoefnagel-Hohle, M. The acquisition of some Dutch morphological rules. Journal of Child Language, 1980, 7, 539–553. Stampe, D. The acquisition of phonetic representation. In Papers from the Fifth Regional Meetings. Chicago: Chicago Linguistic Society, 1969. Stemberger, J. The lexicon in a model of language production. Unpublished doctoral dissertation, University of California, San Diego, 1982. Stern, G. Meaning and change of meaning. Göteborg: Göteborgs Högskolas Ärskrift, 1931. Stern, C, & Stem, W. Die Kindersprache. Leipzig: Barth, 1907. Strauss, S. U-shaped learning curves. Hillsdale, NJ: Lawrence Erlbaum Associates, 1981.

402

Thibadeau, R., Just, M., & Carpenter, P. P. A model of the time course and content of reading. Cognitive Science, 1982, 6, 157–203. Tompa, J. A mai magyar nyelv rendszere. Budapest: Akadémiai Kiadó, 1970. Viktor, G. A gyermek nyelve: A gyermeknyelv irodalmának ismertetése fökent nyelvészeti szempontból. Nagyvárad, 1917. Waddington, C. H. The strategy of the genes. New York: Macmillan, 1957. Walter, S. Zur Entwicklung morphologischer Strukturen bei Kinder. Heidelberg: Diplomarbeit, 1975. Zager, D. The protean changes of pronoun markers on Eskimo verbs. In J. Kreiman & A. Ojeda (Eds.), Papers from the parasession of pronouns and anaphora. Chicago: Chicago Linguistic Society, 1980. Zakharova, A. V. Usvoenie doškol’nikami padežnyx form. Doklady Akademii Pedagogičeskix Nauk RSFSR, 1958, 2, 81–84. (Translated in C. A. Ferguson & D. I. Slobin (Eds.), Studies of child language development. New York: Holt, Rinehart, & Winston, 1973.) 1

Technical terms used in formulations of Operating Principles are explained at relevant points in the text. 2

The notion of “possible combinations” of allomorphs requires some explanation. It presumes that the child

403

knows how meanings are linked in semantic structure. For example, s/he presumably knows that plurality is a fact about the noun and not the verb. If his language encourages him to also mark plurality on the verb or on any other sentence element in addition to the noun, then the child must specifically learn to alter his semantic structure in this way. Such learning of agreement patterns is treated in the dialectic model as a case of the acquisition of free-rules of polysemy. In regards to consequence spellout 3a, the basic point is that, given his knowledge of the principles of possible semantic combinations, the child is still free to make a variety of morphophonological errors. This does not mean that the child cannot also make errors in combining morphemes that do not “go together” semantically. However, such errors are polysemic errors and not morphophonological errors. 3

Until we achieve a fuller understanding of the neuronal basis of item and rule strength, it will not be clear whether the principle involve here is Darwinian or Lamarckian in nature. If neurons adapt in the direction of their use, the principle seems to be Lamarckian. If potential connections drawn out of some very large set are weeded out by failure, the principle would appear to be Darwinian. In neither case is the notion of strengthening Skinnerian. 4

Note that *köve + -t is a less reasonable analysis of követ, since final e would change to é before the allomorph -t. The absence of undersegmented allomorphs ending in e or a can be taken as evidence that the child monitors his segmentations by sending them 404

back through his system of productive combinations. In this case the rules would yield *kövét which would not match követ and thus analysis of követ into köv plus -et would be blocked. 5

Other essentially error-free neologisms include: bibizget ‘go ouchy’, találgat ‘find about’, and kihoppan ‘go whoops out’ (MacWhinney, 1974, p. 363), lehoppáztat ‘go whoops down’ (MacWhinney, 1974, p. 365), 52 forms in -ez (MacWhinney, 1974, pp. 366–370), vizslálkodik ‘act like a vizslá’, jókálkodik ‘act good’, elvilágoskodik ‘lighten’, lilul ‘get lilac’ (MacWhinney, 1974, p. 372), 18 forms in -ol (MacWhinney, 1974, pp. 373–375), koptalan ‘unworn’ (MacWhinney, 1974, p. 510), enyveló ‘gluer’ (MacWhinney, 1974, p. 377), and vásárság ‘fairness’ (MacWhinney, 1974, p. 378.)

405

15

Capacity

Crosslinguistic Evidence for the Language-Making

Dan I. Slobin University of California Berkeley

Contents Operating Principles for the Construction of Language Basic Child Grammar Filters for Primary Perception and Storage of Input Attention to Speech Entering and Tagging Information in Storage The Pattern Makers

406

Semantic Space and Grammatical Morphemes Content Words, Routines, and Functors Grammaticizable Notions Prototypical Scenes Perspective-Taking in Scenes Relevance and Proximity in Semantic Space Summary: Semantic Space and Functors The Persistence of Basic Notions Grouping Information in Storage General Problem-Solving Strategies Review Strategies Interim Production Strategies Strategies for Grammatical Organization of Stored Information Organizing Information in Storage: Linguistic Units Organizing Information in Storage: Form-Function Mapping Organizing Information in Storage: Position of Elements Conclusion

407

In one way or another, every modern approach to language acquisition deals with the fact that language is constructed anew by each child, making use of innate capacities of some sort, in interaction with experiences of the physical and social worlds. The purpose of our detailed explorations of the acquisition of fifteen different languages in this book has been to describe children’s growing use of speech and sign in order to more precisely characterize linguistic aspects of experience and behavior in development.1 It is only by detailed examination of patterns of children’s verbal interaction with others that we can form a picture of the child’s activity in constructing language. By observing repetitions of such patterns across individual children and languages we can begin to form hypotheses about the underlying capacities that may be responsible for language acquisition in general. The authors of each of the chapters have proposed various sorts of processes that may account for observed developmental patterns in particular languages. These processes include strategies of perception and production, constraints on hypotheses, preference for forming certain types of rules, guided interaction with caretakers, and cognitive pacesetting of language development. All of these proposals, taken together, constitute a broad and detailed attempt to describe the means used by the child to construct language. In this chapter I try to weave together these many strands of data and theory into a general description of the mental equipment responsible for the child’s linguistic achievements. In order to avoid the theoretical baggage that Chomsky’s LAD (“Language Acquisition Device”) has to carry 408

about with “him,” and to avoid the burdens of sexist terminology required by reference to LAD and/or Ervin-Tripp’s LAS (“Language Acquisition System”), I shall attempt to make use of our crosslinguistic findings to discuss what we currently know about “LMC”—the LANGUAGE-MAKING CAPACITY of the child. Rather than “pre-tune” LMC to a particular current theory of abstract syntax, I prefer to work backward from acquisition data to propose systems of knowledge and information processing that seem to be prerequisite for the sorts of data that we encounter crosslinguistically. Clearly, LMC must begin life with some initial procedures for perceiving, storing, and analyzing linguistic experience, and for making use of capacities and accumulated knowledge for producing and interpreting utterances. I believe that we do not know enough yet about LMC to be very clear about the extent to which it is specifically tuned to the acquisition of language as opposed to other cognitive systems, or the degree to which LMC is specified at birth— prior to experience with the world of people and things, and prior to interaction with other developing cognitive systems. These issues are full of controversy precisely because we know so little about LMC and comparable capacities for the acquisition of other forms of structured knowledge and behavior. The only way that we can ever gain more clarity about issues of innateness and task-specificity is to obtain considerably more detailed descriptions and theoretical accounts of the course of development of language and of other systems. In spite of the many pages of this book, and many other publications, it is evident that we have only the most preliminary understanding of LMC, and it is difficult to find comparable treatments of other aspects 409

of development. In this chapter, therefore, I try to pull together what is suggested by current crosslinguistic comparison in regard to the nature of LMC, leaving it to future scholars to find a place for this capacity in a broader theory of the mind and its development. I will restrict myself primarily to the data presented in this series attempting to lay out the present limited state of our knowledge of LMC.

OPERATING PRINCIPLES FOR THE CONSTRUCTION OF LANGUAGE The task, then, is to propose a set of procedures for the construction of language. I have used the term “Operating Principle” (OP) to denote the “procedures” or “strategies” employed by LMC (Slobin, 1971, 1973, 1982). OPs, whatever their ultimate origin, are necessary prerequisites for the perception, analysis, and use of language in ways that will lead to the mastery of any particular input language. The postulation of OPs constitutes a psycholinguistic attempt to respond to the challenge of Chomsky’s claim that “knowledge of grammatical structure cannot arise by application of step-by-step inductive operations … of any sort that have yet been developed within linguistics, psychology, or philosophy” (1965, p. 58). The goal of this chapter is to propose a set of OPs based on the crosslinguistic evidence currently available, trying to sort out the ways in which knowledge of grammatical structure is given in advance and the ways in which it is constructed in the course of linguistic, cognitive, and social experience.

410

OPs function in the processing of linguistic input and in the organization and reorganization of stored linguistic material. (They are not to be confused with short-term “strategies” that have been postulated to account for the ways in which children deal with tasks posed in psycholinguistic experiments, although the ways in which speech is perceived and interpreted clearly contribute to the ways in which language comes to be organized. We are also not concerned with strategies for producing individual utterances, although our data for postulating OPs come almost entirely from children’s speech productions.) Our concern here is to search for principles that a child may use in beginning to construct a GRAMMAR—that is, a system of combinatorial principles according to which particular linear placements of meaningful words and grammatical morphemes results in utterances with regularly predictable propositional meanings and pragmatic force. It will be our conclusion that the application of a basic set of OPs to any particular input language results in a universally specifiable “Basic Child Grammar” which reflects an underlying ideal form of human language. Briefly, as suggested by the nature of early child speech, such a grammar consists of the mapping of a collection of basic notions onto certain types of fixed arrangements of clearly identifiable, acoustically tangible, and relatively invariant phonological forms. However, we will have to cover a good deal of crosslinguistic ground in order to approach this conclusion and fill in some of the details. To summarize the terms introduced thus far: The Language-Making Capacity (LMC) is described in terms 411

of a collection of Operating Principles (OPs) which, in their initial form, exist prior to the child’s experience with language. In the course of applying OPs to perceived speech and associated perceptions of objects and events, LMC constructs a preliminary Basic Child Grammar that guides the production of meaningful, structured utterances. Our evidence comes from children’s speech, but we will assume that Basic Child Grammar corresponds to the internal organization and storage of linguistic structures in relation to meanings. Given the paucity of theory and data in regard to mechanisms of comprehension, the role of Basic Child Grammar in comprehension must remain an open question. Beyond Basic Child Grammar, as the child gathers more information about the peculiarities of the particular input language, new OPs are brought into play, and the growing body of linguistic data influences the shape of grammar beyond the basic phase. As a result, children move from a UNIVERSAL grammar to the divergent grammars of individual languages. At all points in development there are interactions between OPs, as well as interactions between specific OPs and the level of grammar construction achieved by LMC. Basic Child Grammar, then, is an idealization of the beginning phases of structured sound-meaning correspondences, as reflected in children’s speech across languages. The available evidence is many steps removed from the conclusion. Primarily, we have data of child SPEECH, with limited information about the referential and communicative contexts in which the speech occurred. Within such data we can discern regularities which 412

correspond to the input language (“error-free” acquisition) and regularities which deviate from the input (“errors”). In addition, we have data on the course of development (precocious and delayed or protracted development and reorganizations in inferred structures). Our evidence, then, consists of (1) the sorts of systems that are easily accessible to the child, (2) the ways in which the child alters less accessible systems, and (3) the sorts of systems which pose particular problems in acquisition. On the basis of such unavoidably inadequate data we make guesses about the child’s internal representation of language. Using such evidence, we postulate two major types of OPs: (1) those which convert speech input into stored data which the child will be able to use in constructing language—the PERCEPTUAL AND STORAGE FILTERS, and (2) those which are used to organize stored data into linguistic systems—the PATTERN MAKERS. The first type of OPs, the filters for the perception and primary storage of speech input, are explored at length by Ann Peters in this volume, and will be substantially taken as given in the present chapter. These principles are responsible for the “primary linguistic data” which are available to the child. They determine which segments of the speech stream will be of sufficient perceptual salience to be noticed and stored, and they establish the basic “filing systems” for such speech segments in storage. This chapter, however, primarily explores OPs of the second sort—those which systematize stored data into morpheme classes and patterns of morpheme placement in meaningful utterances. Our main concern, then, is with the initial organization of morphology and 413

syntax—because we have the richest body of crosslinguistic data in this regard, and because early morphosyntactic structures provide a window to the child’s predilections to express particular meanings in particular ways.

BASIC CHILD GRAMMAR The central claim of this chapter is that LMC constructs similar early grammars from all input languages. The surface forms generated by these grammars will, of course, vary, since the materials provided by the input languages vary. What is constant are the basic notions that first receive grammatical expression, along with early constraints on the positioning of grammatical elements and the ways in which they relate to syntactic expression. When we know enough about the OPs with which LMC is equipped, we should be able to predict the early phases of grammar in any new child language that we encounter. Why should we expect to find a universal Basic Child Grammar, instead of a collection of individual acquisition stories in which the child successively approximates the form of the input language, filling in more and more details with age? Doesn’t child speech simply progress from “telegrams” to fluent prose? The data of this series clearly show that this is an inadequate picture. Early child speech represents a more interesting schematization of the input than a telegram, because some grammatical elements ARE present, and their range of application is often rather different from that of the

414

input language. We find, repeatedly, that some kinds of structures are easily ASSIMILATED—suggesting that they are close to a set of initial linguistic schemas, while in other cases the child ACCOMMODATES in specifiable ways to kinds of structures that cannot be assimilated early on. I have intentionally phrased this statement in terms of “assimilation” and “accommodation” as opposed to “errors” and “error-free” acquisition. As Ochs and Givón have quite appropriately pointed out in their chapters, we unnecessarily limit ourselves to an adult, normative sense of language by characterizing child speech as “in error.” In addition, we would do well to bear in mind that language acquisition takes place, by and large, remarkably free of error. What is learned readily is just as significant theoretically as what is deformed in acquisition. Yet, we have little else to go on besides attempts to compare what children say to what they hear, and error analysis forms the substance of most of our crosslinguistic developmental data. Given the goal of defining Basic Child Grammar, we will have to attend to all available evidence of how learners shape language. We must make use, then, of ANY systematic patterns of analysis, re-analysis, limitation and extension of rules, and the like— for all such patterns are suggestive of characteristics of LMC. Our data are, admittedly, scanty, and we must attend to every hint to the possible linguistic idealizations attempted by children. Thus, in the theoretical discussion that follows, I abstract across individual children and individual languages, making a broad first pass at describing Basic Child Grammar and the mechanisms of LMC that may be responsible for children’s PREFERENCES to construct language in 415

particular ways, knowing full well that such abstracted and generalized preferences cannot account in detail for the acquisition patterns of particular, individual children.2 Underlying the entire endeavor is a characterization of the child’s endowment for discerning and relating meanings and linguistic forms: The child’s fundamental task is to construct mappings between meaning and form. In the most general terms, it has often been stated that such mappings tend to be ONE-TO-ONE. However, having said this, we must ask what constitutes “one meaning” and “one form” for the child. It turns out that the answer to this question is not at all evident, because Basic Child Grammar (and ALL grammars) allow for COMBINATIONS or CONFLATIONS of meanings, and because “form” can be analyzed on a number of levels. For example, Turkish children (Aksu-Koç & Slobin3) easily acquire the accusative inflection, ‘DEFINITE direct object’. Grammatically, the inflection is a single, monosyllabic morpheme; but morphophonologically it consists of four allomorphs, adjusted to the preceding stem according to vowel harmony. Yet Turkish children cannot be said to have shown a propensity for many-to-many mapping in this early acquisition, because the notion “definite direct object” is a UNITARY notion from the child’s point of view, and because the vowel-harmony allomorphs are “natural,” contextually conditioned variants of one underlying form. Thus the principle of one-to-one mapping is neither meant to be proven nor disproven by this example. Rather, taking the principle as given, it guides us in discovering what constitutes “one meaning” and “one form” from the 416

child’s point of view at this phase of development. (Later, of course, the child will have to learn what constitutes “one form” and “one meaning” from the adult point of view at every point in the grammar of the language.) Similar examples can be offered in the syntactic realm. A single form can be extended beyond its initial function, indicating an underlying unity to a range of functions. For example, the Japanese genitive particle no is extended from constructions in which one noun modifies another (N no N) to constructions of modification in general, including *ADJ no N and *REL.CL no N, indicating that “modification” may constitute a general, or unitary syntactic function for Japanese children (Clancy). On the level of syntactic form, we can infer that a clause with a finite verb constitutes a unitary form that can occur in varying environments, since children learning Indo-European and Semitic languages have little difficulty in acquiring means for conjoining and subordinating such clauses, while children learning Turkish have difficulty in identifying constructions with nonfinite verbs (participles especially) as instances of clauses. Thus “one form—one function” on the clausal level requires a complex definition of “clause” from the point of view of the child. At every point in development, LMC is presented with a multiplicity of linguistic forms and functions. An orderly grammar can be constructed only if LMC is prepared to sort and classify both units of meaning and units of 417

sound, and relate them to one another. An account of acquisition must describe the means used for such analysis. To state the Leitmotiv of this chapter succinctly: Our analytic task is to define two sorts of SIMILARITY SPACE with which LMC is equipped: (1) a SEMANTIC SPACE in which core notions and clusters of related notions can be identified, and (2) a FORMAL SPACE in which allomorphs and construction types can be located and related to one other. Furthermore, within each of these similarity spaces, entities are arrayed in an ACCESSIBILITY HIERARCHY according to which some notions and forms are likely to emerge earlier in development than others. OPs function within and across these spaces to construct language. Let us begin with the most basic OPs for perception, segmentation, and storage of speech data, building up to issues of semantic mapping and formal structures.

FILTERS FOR PRIMARY PERCEPTION AND STORAGE OF INPUT ATTENTION TO SPEECH On the most basic level, accessibility of linguistic material can be defined in terms of PERCEPTIBILITY. That is to say, the only linguistic material that can figure in language-making are stretches of speech that attract the child’s ATTENTION to a sufficient degree to be noticed and held in memory. The first of the OPs I proposed in

418

1973 is prototypical of principles of this sort: Operating Principle A: “Pay attention to the ends of words.”4 That OP was derived from a widespread crosslinguistic finding that post-verbal and post-nominal locative markers were acquired earlier than pre-verbal and pre-nominal locative markers, holding semantic content constant.5 Most of the chapters in this series report special salience for final syllables and clause-final particles. However, they also report salience for stressed syllables and initial syllables (though not for prepositions or preposed articles or verbal particles). Overall, children have difficulty with grammatical morphemes that are less readily identifiable as distinct acoustic entities—morphemes that are bound, contracted, asyllabic, unstressed, and varying in form in different environments. It is the contribution of Ann Peters’ chapter on language segmentation to propose detailed “Operating Principles for the Perception and Analysis of Language,” in which the old Operating Principle A finds its place among a network of OPs that filter and analyze speech input into the units and frames that provide the substance for grammatical structures.6 We will take her principles as given, presenting only a condensed summary here. These are OPs that play a role in providing LMC with a preliminary body of data for linguistic analysis. The child must EXTRACT, SEGMENT, and STORE input in ways that allow for language-making. In the most general terms, the child must orient to particular types of sounds and sound sequences:

419

OP (ATTENTION):SOUNDS. Store any perceptually salient stretches of speech. An OP of this generality, of course, requires specification of “perceptually salient.” Peters suggests that, at first, the child will orient to whole utterances (i.e. stretches of speech bounded by silence) and particular intonational and rhythmic patterns of speech (her OPs for “Extraction,” labeled SILENCE, SUPRASEGMENTAL, TUNE, and RHYTHM). These OPs will “extract” sound sequences from the input, providing the child with “amalgams” (Mac-Whinney) and “formulas” (Wong Fillmore, 1976)—that is, units that are often longer than individual words or word-stems—along with some separate words. Much work remains to be done before we can specify the parameters governing unit extraction. Eventually, following OPs such as those of Peters, LMC comes to operate with word-size units. Much of our attention in this chapter, however, is devoted to units that are smaller than words, as discussed below. In addition to unit extraction, all pattern recognition requires a basic ability to take note of familiarity and unfamiliarity. This is only possible if the organism keeps track of the frequency of patterns in experience, with automatic capacities to strengthen the traces of repeated experience and to more readily retrieve frequent and recent information. Without such capacities, of course, the many segments and patterns resulting from OPs such as those proposed in this chapter would lead nowhere. So we need something like Peters’ SG:FREQUENCY, which can be generalized as: 420

OP (STORAGE): FREQUENCY. Keep track of the frequency of occurrence of every unit and pattern that you store. Within perceptually salient speech units, certain types of segments stand out in short-term memory and begin to be stored as separate elements in relation to the speech material with which they co-occur. Such elements are isolated by Peters’ OPs for “Segmentation” (INTONATION, RHYTHM, STRESS, END, BEGIN, REPETITION). The most pertinent for our purposes in this chapter are those OPs that direct children’s attention to acoustic material that may eventually figure in the isolation of grammatical morphemes. Perceptual/ attentional OPs provide speech units that will become candidates for functors, following a pattern-making OP to be presented later, OP:FUNCTORS.7 Crosslinguistic data suggest that final, stressed, and initial syllables are perceptually salient, as reflected in the following OPs. OP (ATTENTION): END OF UNIT. Pay attention to the last syllable of an extracted speech unit. Store it separately and also in relation to the unit with which it occurs. (Based on Peters’ SG:END.) OP (ATTENTION): STRESS. Pay attention to stressed syllables in extracted speech units. Store such syllables separately and also in relation to the units with which they occur. (Based on Peters’ SG:STRESS.) OP (ATTENTION): BEGINNING OF UNIT. Pay attention to the first syllable of an extracted speech 421

unit. Store it separately and also in relation to the unit with which it occurs. (Based on Peters’ SG:BEGIN.) (In a related vein, MacWhinney suggests determinants of perceptual salience in his Operating Principle for “amalgam acquisition”: “Phonological clustering is determined by intensity, pitch, and juncture. Recent segments and stressed segments are stored most clearly.” And in his Operating Principle for “amalgam analysis” he suggests initial segmentation and storage of contiguous units: “The child first attempts to analyze words into clear continuous morphemes.”)

ENTERING AND TAGGING INFORMATION IN STORAGE It is not enough, of course, to simply extract and segment bits of speech input and store them away. Material must be stored in ways that will be useful for linguistic analysis. On a preliminary basis, the system must be preset to tag, amalgamate, and count speech data in ways related to extraction, segmentation, and eventual higher-order processing. Peters suggests OPs for comparison (EX:COMPARE) and storage (EX:STORE) of extracted units, which we can combine as a general OP for creation of units in storage: OP (STORAGE): UNITS. Determine whether a newly extracted stretch of speech seems to be the same as or different from anything you have already stored. If it is different, store it separately; if it is the

422

same, take note of this sameness by increasing its frequency count by one. For grammar building to be possible, more than units must be stored. As Maratsos has convincingly argued (Maratsos, 1982; Maratsos & Chalkley, 1980), children must keep track of co-occurrences of content words and grammatical morphemes in order to construct grammatical classes. He suggests, for example (1982, pp. 258–263), that German gender classes are induced from storage of nouns with differing patterns of co-occurring articles, and that the English verb class is induced from co-occurrences of particular lexical items with verbal inflections and auxiliaries. Mills’ German data show rapid acquisition of article-noun pairings, as do Clark’s Romance language data. Berman reports that Hebrew verbs are learned with their appropriate prepositions. And Ochs notes that Samoan clitic pronouns are always correctly placed before the verb. Even though they are infrequent, this is the only position in which they occur in Samoan. It appears, then, that if a form has fixed placement, the position may be easily acquired even if it is relatively infrequent. Facts such as these are reflected in Peters’ OPs for “segmentation of morphosyntactic frames” (FR:FRAME and FR:SLOT). We can generalize this prerequisite for the construction of grammar as: OP (STORAGE): CO-OCCURRENCE. For every segmented unit within an extracted speech string, note its co-occurrence with any preceding or following unit and store sequences of co-occurring

423

units, maintaining their serial order in the speech string. In this brief section on “filters for primary perception and storage of input” we have been concerned with fairly automatic and “low-level” processes of noticing and storing acoustic material. Our central concern, however, is with the use of acoustic material to “make sense.” LMC must systematically relate sounds to meanings. Accordingly, the collection of OPs must be enriched by attention to meaning as well as sound.

THE PATTERN MAKERS The heart of this chapter deals with OPs for constructing PATTERNS. In a sense, the chapter can be seen as an elaboration of the old Operating Principles E, F, and G (Slobin, 1973): E: “Underlying semantic relations should be marked overtly and clearly.” F: “Avoid exceptions.” G: “The use of grammatical markers should make semantic sense.” Essentially, an elaboration of Operating Principle E requires a specification of both “semantic relations” and “overt and clear marking.” When these terms are filled in by crosslinguistic developmental-analysis, it is no longer necessary to speak of “avoiding exceptions”; rather, what emerges is an early and consistent version of Basic Child Grammar in which some (but not all) types of exceptions are absent. Furthermore, we shall see that “semantic sense,” though a primordial motivation for grammatical marking, is not the only organizing principle available to LMC, since means of formal, nonsemantic organization

424

are also available early on. We will begin, however, with semantic bases for the discovery and mapping of the first grammatical morphemes, later filling in evidence for nonsemantic bases as well. Our first task is to characterize the semantic notions that are available to LMC for the construction of Basic Child Grammar, in conjunction with patterns of speech sounds of particular types. It is appropriate to set out at the start one of the most basic prerequisites for language—namely, that utterances are produced to some discernible human purpose. The child attempts to relate input to some amalgam of semantic and pragmatic meanings, which we will come to define as locations in a “Semantic Space.” In the most general terms, children attend to and store MEANINGFUL utterances. MacWhinney phrases this prerequisite as part of his Operating Principle for “amalgam acquisition”: “Children acquire new lexical items by making a new association between a sound and a meaning. The clearer the representation of both the sound and the meaning, the earlier the acquisition.” Peters proposes an OP for extraction that she calls “MEANING.” We will assume that the child, of necessity, begins with some prelinguistic inklings of meaning and attempts to “map” them onto salient sounds, setting up the framework for a dictionary: OP (MAPPING): DICTIONARY. Pay attention to sound sequences that have a readily identifiable meaning and store them in a Dictionary, along with a representation of the context in terms of available semantic and pragmatic Notions in Semantic Space.8 425

Having introduced attention to meaning as part of LMC, we must now begin to characterize the ways in which sound sequences are PATTERNED in relation to meaning.

SEMANTIC SPACE AND GRAMMATICAL MORPHEMES Content Words, Routines, and Functors To begin with—beyond primary storage of simple linguistic units and co-occurring patterns, along with frequency counts and some mapping onto basic meanings—stored material must be organized in ways that will facilitate linguistic analysis. That is to say, once LMC has some units and patterns in storage, tagged in preliminary ways for language acquisition, the material must be grouped and systematized. Units and patterns must be searched and compared, and new units and classes of units must be established. Although we have no direct evidence for the basic strategies for grouping information in linguistic storage, some OPs can be proposed as at least minimal prerequisites.

Grouping Information in Storage Peters is much concerned to provide for the establishment of larger, recurring segments—a necessity to move the child beyond the storage of a huge amount of small bits of speech. She proposes strategies for the matching and comparison of segments (SG:MATCH1, SG:MATCH2), which can be generalized as:

426

OP (STORAGE): UNIT FORMATION. If you discover that two extracted units share a phonologically similar portion, segment and store both the shared portion and the residue as separate units. Try to find meanings for both units. Up to this point, we have been equipping LMC with very basic means of segmenting speech and keeping track of units and pairs of units. The last OP, however, presents LMC with a preliminary means of storing word-stems and affixes. At this point, we cannot easily go on with a discussion of “grouping information in storage” without paying attention to meaning as well as form. LMC is involved in making patterns on several levels at once, grouping and organizing semantic information and linguistic structures. However, unlike LMC, WE must discuss levels of operation separately at first, in order to interrelate them later. Before we can proceed further in exploring LMC’s capacity to group and organize information in storage, it is necessary to equip LMC with a crucial distinction for all further analysis—the distinction between lexical items that make reference, CONTENT WORDS, and grammatical morphemes, FUNCTORS. This distinction arises as a consequence of OPs for segmentation and mapping, operating within Semantic Space.

Content Words and Routines Our first Operating Principle for mapping, OP:DICTIONARY, instructs LMC to “pay attention to utterances that have a readily identifiable meaning.”

427

Everything we know about the beginnings of child language shows that the first meanings are relatively unanalyzed and tied to particular narrow communicative routines and concrete references. This must be our starting point, and we will assume that certain perceptual and sensorimotor experiences are highly salient to the infant and susceptible of symbolic representation in terms of verbal acoustic events that occur in conjunction with such experiences. The first speech sequences that are extracted, segmented, and placed in storage are mapped onto the event characteristics that are most salient to child cognition and social interaction. This fact can be presented as a more specific OP for mapping: OP (MAPPING): CONTENT WORDS AND ROUTINES. Try to map extracted speech units onto representations of objects and events—the core referential meanings and pragmatic functions associated with typical activities and interactions. Store units with their meanings. This OP will remain in force throughout acquisition. In its early uses, before the discovery of grammatical morphemes, and before the elaboration of means for analyzing object and event schemas, it reflects LMC’s initial assumption that all words are content words, holophrases, or interactional routines. Later, it will continually seek to interpret newly-extracted speech segments as content words. Much more could be said about this OP, as a large literature deals with the meanings of first words. It is also evident that child speech is involved in a range of pragmatic functions—requesting, commenting, greeting, refusing, 428

and so on. Given the state of the art and the data reviewed in our chapters, we will have little to say about the mapping of such functions, but it is clear that LMC has access to a “Pragmatic Function Space” in conjunction with Semantic Space. For our present purposes, though, it is sufficient to establish the primacy of content words and routines, however defined, as a background for the acquisition of grammatical morphemes, which is our central question.

Operators Routines are often carried out by single-word utterances that “operate” on situations in fairly global ways, noting disappearance or cessation (e.g. allgone, byebye), requesting or predicting recurrence (e.g. again, more), and so forth. Lists and classifications of such forms are widespread in the child language literature. Such operators also occur in conjunction with content words, first as amalgams, such as more milk and later as productive limited-scope formulae (Braine, 1976), such as more read. The child analyzes such two-word amalgams into two words, following OP:UNIT FORMATION. One of these words can be mapped onto representations of objects or events, becoming a content word (eventually noun or verb), while the other can be mapped onto the routine in which the content word figures (e.g. noting disappearance, requesting recurrence). Thus a conjoint operation of OP:UNIT FORMATION and OP:CONTENT WORDS AND ROUTINES can break two-word amalgams into a referential term and a relational term—given some built-in cognitive predisposition to

429

conceive of situations in terms of objects and operations upon objects. (Gopnik [1984], for example, has shown that the use of gone emerges in contexts where children make objects disappear and re-appear, showing a close relationship between development of the object concept itself and children’s actions and expectations with regards to objects.) Such a conceptual distinction (between external events and one’s relation TO those events) is a prerequisite for the discovery and use of relational terms, including grammatical morphemes. Such elements are often bound morphemes, thus not occurring as single words or as halves of two-word amalgams. In such instances, OP:UNIT FORMATION provides a content word and an uninterpreted segment that has not occurred as a separate word in a routine. In analyzing such amalgams, it is crucial that LMC have means for assigning interpretations to both root and affix. The key would be to assign a referential meaning to the root and to search for a relational meaning for the affix. This key is provided by a central Operating Principle for the mapping of functors, which builds upon the distinction between referential and relational meanings.

Functors Eventually, given the operation of OPs for attention to salient segments, in conjunction with OPs for storage (OP:CO-OCCURRENCE, OP:UNIT FORMATION) and mapping (OP:CONTENT WORDS AND ROUTINES), LMC will encounter leftover segments in storage that have not been mapped as content words.

430

This is the opening wedge for the discovery of grammatical morphemes. All human languages depend on combinations of reference-making and relational morphemes. Sapir set forth this distinction with his usual clarity in Language (1921/1949, p. 93): What, then, are the absolutely essential concepts in speech, the concepts that must be expressed if language is to be a satisfactory means of communication? Clearly, we must have, first of all, a large stock of basic or radical concepts, the concrete wherewithal of speech. We must have objects, actions, qualities to talk about, and these must have their corresponding symbols in independent words or in radical elements…. And, secondly, such relational concepts must be expressed as moor the concrete concepts to each other and construct a definite, fundamental form of proposition. In this fundamental form there must be no doubt as to the nature of the relations that obtain between the concrete concepts. We must know what concrete concept is directly or indirectly related to what other, and how. Sapir went on to note that relational concepts are “normally expressed by affixing non-radical elements to radical elements or by position” (p. 101). The uninterpreted speech segments that LMC finds in storage, after mapping content words and routines, are likely to be such elements, and we must equip LMC with an OP for mapping those elements onto relational concepts. Let us call the linguistic segments FUNCTORS and the relational concepts GRAMMATICIZABLE NOTIONS, 431

leaving it to our crosslinguistic survey to fill in these terms with content as we go along. This OP has several levels of operation, depending on the level of grammatical knowledge achieved by LMC. At the first level, it will map such uninterpreted segments onto the most available nonreferential notions: OP (MAPPING): FUNCTORS. If a speech segment remains uninterpreted after the establishment of content words and routines, try to map it onto an accessible grammaticizable Notion that is relevant to the meaning of adjacent referential units in the situation in which the speech segments occur. If you succeed, store such a nonreferential relational unit (“functor”) with its meaning and its placement in relation to associated linguistic units and their meanings.9 Later we will specify this OP in somewhat more detail, filling in the meanings of “relevant” and “situation.” For now, the crucial phrase is “accessible GRAMMATICIZABLE Notion.” I propose that we must assume that LMC distinguishes between the two primordial types of linguistic concepts defined by Sapir. The child conceives of events in terms of relations between concrete entities, and is, in some sense, “set” to discover symbols for both concrete and relational concepts. In order to speak any PARTICULAR language, certain types of relations must be grammatically marked (“grammaticized”). Crosslinguistic evidence of the earliest mappings of functors in child speech can reveal universal points of departure for language-particular marking of grammatically-relevant Notions. As more functors are 432

interpreted and related to one another, OP: FUNCTORS will have a broader base of intersecting Notions to use in refining the mapping of functors in the language, and eventually language-specific patterns may well have a Whorfian effect on the organization of Semantic Space. To begin with, however, the first grammaticizable Notions must be provided by LMC.

Grammaticizable Notions Several years ago, Melissa Bowerman (1976, p. 110) posed an important question: “We must determine how, out of all the cognitive discriminations a child is potentially capable of making at a given time, some begin to get connected to language and hence to take on semantic significance while others do not.” For example (Slobin, 1979, pp. 90–92; 1982, pp. 131–136), an English-speaking child must learn to grammatically mark number but not gender; a Hebrew-speaking child must learn to mark both; and a Chinese child neither. And no child has to learn to grammatically mark color, or speed, or degree of physical effort—because these are not grammaticizable notions at all. As Schlesinger has put it (1982, p. 70): We are capable of organizing the world around us in innumerable ways, perceiving any number of relations between objects, attributes, states, actions, etc. We are also capable, in principle, of talking about any one of these perceived aspects. But many of them are not linguistically relevant; that is, the grammar provides no rules for tying them to surface structures of sentences.

433

Leonard Talmy (1978, 1983, 1985) has made a clear distinction between the unbounded range of concepts that can be expressed by “open-class lexical elements” (content words, i.e. the stems of nouns, verbs, and adjectives) and the peculiarly language-specific and limited range of concepts that are expressed by the languages of the world as “closed-class grammatical forms” (functors, grammatical categories, and syntactic structures of phrases and clauses). His conclusions come from linguistic evidence, and we will seek supporting evidence in our child data. In regard to the conceptual bases of grammatical forms, he notes (1983, p. 227): They represent only certain categories, such as space, time (hence, also form, location, and motion), perspective-point, distribution of attention, force, causation, knowledge state, reality status, and the current speech event, to name some main ones. And, importantly, they are not free to express just anything within these conceptual domains, but are limited to quite particular aspects and combinations of aspects, ones that can be thought to constitute the “structure” of those domains. For example (Talmy, 1978, p. 2), a locative adposition such as the English through specifies (1) some motion, (2) along a line, (3) located in a medium, as in I walked through the water / the woods. But such terms, according to Talmy, never specify the kind of substance making up the medium (e.g. water vs. trees), the sensorimotor characteristics of the motion (e.g. swimming vs. walking), the shape of the path (e.g. straight line vs. zigzag), or the rate of motion (e.g. walking vs. running). 434

Note that all of these concepts CAN be encoded by content words, as in the parenthetical examples. But grammatical elements seem to be restricted to the sorts of notions that Talmy describes as “quasi-topological” and “relativistic rather than absolutely fixed quantities.” In Talmy’s (1985) detailed survey of grammatical patterns we begin to see Sapir’s “relational concepts” in cognitive perspective. I have suggested elsewhere (Slobin, 1980, 1981) that such notions must constitute a privileged set for the child, and that they are embodied in the child’s conceptions of “prototypical events” that are mapped onto the first grammatical forms universally. Talmy’s proposals come from comparative grammar, and the proposals of Bowerman, Schlesinger, and Slobin are reasoned arguments based on a small body of child data and inferences about children’s cognitive capacities. The larger range of developmental crosslinguistic data made available by the contributors to this volume gives us the opportunity to expand the list of candidates for grammaticizable notions available to LMC. We can glean evidence for such notions by looking at examples of semantic limitations, extensions, and reinterpretations of grammatical forms in child speech. When functors are first acquired, they seem to map more readily onto a universal set of basic notions than onto the particular categories of the parental language. Later in development, of course, the language-specific use of particular functors will train the child to conceive of grammaticizable notions in conformity with the speech community, as Bowerman has often pointed out (e.g. 1981). At first, however, there is considerable evidence 435

that children discover principles of grammatical marking according to their own categories—categories that are not yet tuned to the distinctions that are grammaticized in the parental language. Having reviewed crosslinguistic evidence for such categories in Basic Child Grammar, we will be in a better position to evaluate the possibility that they constitute a privileged set of grammaticizable notions. I do not wish to recapitulate the dozens of proposals in the literature for the semantic categories that may be expressed in early grammatical combinations. Rather, I want to try to map out several areas of Semantic Space that provide the organizing points for grammatical constructions. An examination of several clusters of salient notions, as revealed by children’s early uses of functors, will demonstrate some of the first patterns of nonreferential meanings available to OP:FUNCTORS. As a first example, let us explore early markings of linguistic transitivity, following my earlier discussion of “prototypical events” (Slobin, 1981), and developing the notion of “Prototypical Scenes” that provide the conceptual material for the mapping of functors.

Prototypical Scenes The Manipulative Activity Scene Linguistic transitivity appears among the first notions marked by grammatical morphemes in the languages under consideration. There are early accusative inflections in Hungarian, Polish, and Turkish, an early direct object marker in Hebrew, and early ergative

436

inflections in Kaluli. Even before accusative inflections are productive, their use in unanalyzed amalgams shows a basic attention to a nominative-accusative distinction. For example, MacWhinney’s (1973) Hungarian subject, Zoli, did not use accusative forms for naming objects, and used them 44% of the time to refer to objects that he wanted to have or to build. And even in noninflectional languages like English, there is some evidence of attention to agent-patient relations in early fixed word-order schemas (Braine, 1976). The information about Zoli is informative: accusative nouns indicated OBJECTS THAT THE CHILD WANTED TO HAVE OR ACT UPON. In Other words, the grammatical marking first appears in conjunction with a particular SCENE, in the sense introduced by Fillmore (1977, p. 84): Meanings are relativized to scenes. I am using the word SCENE in this discussion in a technical sense that includes its familiar visual meaning, but much more as well. I mean by it any coherent individuatable perception, memory, experience, action, or object. In Basic Child Grammar, the first Scenes to receive grammatical marking are “prototypical,” in that they regularly occur as part of frequent and salient activities and perceptions, and thereby become the organizing points for later elaboration of the use of functors. Gee and Savasir (Gee & Savasir, 1985; Gee [Gerhardt], 1983; Savasir & Gee, 1982) discuss such recurrent scenes as “activity-types,” focusing on the activities that children engage in while using particular linguistic forms. De Lemos (1981) uses the term “interactional format,” 437

attending to the role of adult-child dialogue in constituting activity-types and their associated linguistic expressions. I will use the more neutral term, “Scene,” to include the complex of perception, action, and interaction that constitutes the meanings of linguistic forms. Let us characterize the prototypical Scene reflected in linguistic marking of transitivity as the MANIPULATIVE ACTIVITY SCENE. Manipulative activities involve a cluster of interrelated notions, including: the concepts representing the physical objects themselves, along with sensorimotor concepts of physical agency involving the hands and perecptual-cognitive concepts of change of state and change of location, along with some overarching notions of efficacy and causality, embedded in interactional formats of requesting, giving, and taking. There is no basis for proposing any particular analysis of this complex of notions into cognitive categories or semantic features. More plausibly, the notions cohering in the Manipulative Scene form a Gestalt-like conceptual “prototype” such as the “experiential gestalt” of Lakoff and Johnson (1980). The Manipulative Activity Scene corresponds to their “prototypical direct manipulation”—the experiential gestalt of a basic causal event in which an agent carries out a physical and perceptible change of state in a patient by means of direct body contact or with an instrument under the agent’s control10 I propose that OP:FUNCTORS begins by mapping acoustically salient, uninterpreted speech segments onto Scenes as a whole, or onto focused elements of Scenes. 438

Using such preliminary mappings as an anchor, Scenes can gradually be analyzed into those particular Notions which are grammaticized in the parental language. The process is clearly a “bootstrapping operation” (Carey, 1978; Pinker, 1982), in which the acquisition of new functors and syntactic forms motivates further analysis of Scenes along lines set down by earlier analyses, eventually resulting in the mapping patterns of particular languages. Basic Child Grammar, however, tends to grammaticize whole Scenes and their most salient components; and since Scenes represent prototypical activity types and conceptions, rather than language-specific categories, the first functors to appear in child speech, universally, should relate to the same complexes of Notions, regardless of the particular surface forms extracted from the input language. Accusative and Ergative Inflections. This phenomenon is displayed exceptionally clearly if we compare acquisition of an accusative and an ergative language. In an inflectional accusative language, transitivity is marked on the direct object—the noun that expresses the patient in simple active sentences, while in an ergative language the critical inflection appears on the subject noun, expressing the agent. Data on the acquisition of Russian (Gvozdev, 1949) and Kaluli (Schieffelin) show the same UNDERextension of grammatical marking of transitivity, although the nature of the surface marking is so different between these two types of language. Gvozdev noted that in his son’s acquisition of Russian, the accusative inflection was apparently first limited to the direct objects of verbs involving direct, physical action on things—such as ‘give’, ‘carry’, ‘put’, and 439

‘throw’. That is, the accusative inflection marked the Manipulative Scene rather than the grammatical direct object. In Kaluli, the ergative inflection first appears only on the subjects of verbs such as ‘give’, ‘grab’, ‘take’, and ‘hit’, and it tends to be omitted in sentences with verbs such as ‘say’, ‘call-out’, and ‘see’. Again, we find a functor limited to expressions of the Manipulative Scene. Schieffelin reports additional early restrictions on children’s use of the ergative marker, suggesting a special conception of manipulation. She finds that the marker is used earlier, and with greater consistency, in utterances with past-tense verbs, as opposed to present or future. This tendency appears first with the verb ‘give’, and later with a collection of highly transitive verbs, including ‘take’, ‘hit’, ‘bite’, ‘light (a fire)’, ‘cut’, ‘eat’, ‘cook’, and ‘touch’. Note that all of these verbs have a clear end-state, and children may consider that manipulation involves actual achievement of the relative conclusion. Thus, at first, an event has to be manipulative AND ACTUALLY REALIZED in order to receive grammatical marking as an instance of the Manipulative Scene. There is also suggestive evidence that NEGATED verbs tend to be accompanied by grammatically unmarked agents. Again, an unrealized or uncompleted event appears to lie outside of the scope of early grammaticization of manipulation. Kaluli also provides evidence of a type of “non-occurring error.” Children do not OVERextend the ergative inflection to the subjects of intransitive verbs— however active their meaning—such as ‘run’ and ‘jump’. 440

Thus they are not grammaticizing the notion of actor in general, but are only grammatically marking manipulative activities.11 At this point, then, one can only speak of a Basic Child Grammar in which a salient, nonreferential speech segment is used to mark a Scene of a certain type in Semantic Space. The fact that a grammatical morpheme appears as an accusative inflection in one language and an ergative inflection in the other does not yet represent the child’s discovery of the accusative or the ergative case. The particular functors appear because they are associated with a prototypical Scene, and because they are picked up as perceptually salient by OPs for extraction and segmentation. Highlights of Scenes. Each prototypical Scene has “highlights”—the conceptual components that stand out in attention. The Highlights of the Manipulative Scene are the agent (initially, the child acting through his or her own hands) and the physical object affected by the child’s manipulation. The Slavic-speaking child will come to notice that functors marking the Manipulative Scene are attached to the encoding of one Highlight, namely, the content words encoding the physical objects that function as patients in that Scene; while the Kaluli child will relate the corresponding corresponding functors to the other Highlight, namely words denoting the agents of manipulative activities. As additional OPs go to work to build word classes and paradigms, the semantic relations inherent in the prototypical Manipulative Scene will be generalized to less prototypical Scenes. Eventually, casemarking will be 441

extended to relatively desemanticized grammatical roles in general, along lines suggested, for example, by Schlesinger’s (1982) model of “semantic assimilation,” in which less prototypical uses of grammatical forms are assimilated by means of partial similarity, metaphoric extension, and the like. For now, however, our task is to find the semantic relational notions that figure in EARLY uses of functors. At this point, we have isolated an early relational notion of direct causality, mapped onto noun suffixes and word-order patterns in “limited-scope formulae” (as described by Braine, 1976). Through examination of several other prototypical Scenes, we can begin to discern the level of generality of such notions by noting the characteristics of Scenes that are mapped onto functors in early child speech. Again, our evidence comes from the peculiarities of children’s uses of grammatical morphemes extracted from the input language.

The Basic Figure-Ground Scene It is well established that object names play an important part in early vocabularies. In the prototypical Manipulative Scene, object names are related to marking of transitivity, constituting one of the Highlights of that Scene. Objects are also Highlights of closely related Object Placement and Object Transfer Scenes. In these Scenes, the LOCATION or DESTINATION of an object is also a Highlight. It is useful to think of these two Highlights in locative Scenes as “figure” and “ground,” drawing upon the terminology of Gestalt psychology to characterize another prototypical Scene, the “Figure-Ground Scene.” Leonard Talmy (1983) uses

442

these classical terms to analyze the ways in which “language structures space.” In his definitions (p. 232): The Figure is a MOVING or conceptually moveABLE object whose site, path, or orientation is conceived as a variable the particular value of which is the salient issue. The Ground is a reference object (itself having a stationary setting within a reference frame) with respect to which the Figure’s site, path, or orientation receives characterization. Talmy’s notion of Ground combines Fillmore’s (1968) cases of “Locative,” “Source,” “Path,” and “Goal” in terms of “the crucial spatial factor they have in common,” i.e. “their function as reference object for a figural element” (1983, p. 233). There is suggestive evidence from child speech that this “crucial spatial factor” constitutes a primary grammaticizable Notion in Basic Child Grammar. Children seem to be interested in the Ground itself as a Highlight, regardless of whether it is the endpoint of a trajectory or a static background, regardless of whether it is a place or a person, and regardless of a number of special relations between Figure and Ground (such as the distinction between location and possession). Several different patterns of child speech suggest that Ground, in Talmy’s sense, functions as a general relational Notion for the mapping of functors. Static-Dynamic Neutralizations. One aspect of Talmy’s definition of Ground is the neutralization of static locations and the goal or end-point of the trajectory of a 443

Figure. In languages which provide distinct case inflections for locative state and locative goal, a typical child error consists in confusion of the two forms, generally with the stative form used for both functions. Such errors are found in German in regard to dative (static) and accusative (directional) casemarking on articles; in Polish and other Slavic languages in regard to nominal suffixes of the accusative and other cases (typically prepositional); and in Turkish in regard to the locative and dative-directional suffixes on nouns and deictics. These patterns suggest a primary orientation to the Figure-Ground relation, with only secondary orientation to the manner in which the relationship is temporally manifested (enduring versus coming-into-being). (Furthermore, these two locative modes are apparently more salient than a Figure’s departure from a Ground, since it seems to be a universal that expressions of “Source,” or moving away from a Ground, always develop later than expressions of “Locative State” and “Goal.”)12 Conflation of Location and Possession. Broadly conceived, possession is a locative state in which the Ground is an animate being and the Figure-Ground relation is of an enduring or socially-sanctioned nature. We find that children often do not attend to these peculiar social and conventional definitions of Ground, marking only the Figure-Ground relation. One type of evidence comes from extension of a locative functor to mark possession as well, where the language provides distinct means of expression. For example, Mills reports errors of German 3-year-olds such as die grossmama ZU den affe ‘the grandma TO the monkey’ (= ‘the monkey’s 444

grandma’), in which a locative preposition is extended to possession. Other languages provide a single marker for both location and possession. If both meanings emerge at the same point in development, one can infer that the two meanings have semantic affinities from the child’s point of view. For example, Clark reports the simultaneous emergence in a French 1½-year-old of the preposition à for both functions: Nini à bouche ‘in Nini’s mouth’ and chaise à Pierre ‘Pierre’s chair’. Goal as Place or Person. Additional evidence for the generality of Ground as a Basic Notion comes from conflations of location and possession as a goal of directed movement or object transfer. Many languages do not distinguish locations as goals from recipients as goals (e.g. English go TO papa, give it TO papa, Spanish vete A papa, daselo A papa, Turkish baba-YA git, baba-YA ver). Hungarian, however, does grammaticize such a distinction, using the allative case for goals of movement and the dative case for recipients of object transfer. MacWhinney reports child errors of using the dative for both meanings: papa-nek ‘papa-DAT’, used correctly for recipient (‘give to papa’), and incorrectly for goal (‘go to papa’) in place of papa-hoz ‘papa-ALLAT’.13 Locative Relations between Figure and Ground. In addition to tendencies to neutralize potential Ground distinctions between location and destination, and between goal and recipient, Semantic Space provides a basic set of locative relations arranged in an accessibility 445

hierarchy determined by cognitive development. All crosslinguistic acquisition data point to an initial salience of topological notions of containment, support, and contiguity, with later development of projective and Euclidean notions of spatial orientation to particular features of objects or angles of regard (Parisi & Antinucci, 1970). In linguistic terms, this means that the first locative pre- or postpositions encode Notions of ‘in’, ‘on’, ‘under’, and ‘beside’. (Furthermore, as Herskovits, 1982, suggests, the earliest uses of these terms should encode the most prototypical spatial relations for each Notion—for example, prototypical containment involves a smaller object located within and supported by a larger object, with physical contiguity between the two. Similarly, Johnston (1984) has shown that the first uses of behind in English refer only to a smaller object totally hidden from view by a larger object.) Johnston and Slobin (1979) have found a common order of development of locative Notions in English, Italian, Serbo-Croatian, and Turkish, and to the extent that relevant data are available in other languages, the sequence seems to be supported crosslinguistically. The sequence is extra-linguistically determined, though it interacts with aspects of linguistic complexity in particular languages. The basic development order seems to be the following (where “F/B” refers to objects that have an inherent front and back): ‘in’/‘on’ < ‘under’ < ‘beside’ < ‘behind + F/B’ < ‘in front + F/B’ < ‘between’ < ‘behind – F/B’ < in front – F/B’.

446

Basic Child Grammar, then, will map locative functors onto prototypical examples of Figure-Ground relations, beginning with the most accessible relations. The starting point is probably a neutral term meaning ‘located at’, embracing location, possession, goal, and recipient. Further development is motivated by conceptual growth, in interaction with the linguistic forms and meanings of the collection of locative functors provided by the input language.

Perspective-Taking in Scenes So far we have characterized Scenes as having Highlights, such as Agent and Patient, Figure and Ground. Highlights function to foreground particular Notions for mapping onto functors. Another foregrounding factor comes from the ability to view Scenes from different perspectives. Whenever a Scene has a temporal dimension it can be viewed from the perspective of different points in time. The ability to take perspectives on Scenes underlies the acquisition of such grammatical forms as passives, participles, and tense-aspect markers. Even at the one-word stage a collection of salient temporal perspectives is evident, as shown by the presence in all languages of early terms marking completion and repetition of events (e.g. terms meaning ‘all gone’, ‘all done’, ‘more’, ‘again’, etc.).

Grammaticization of Results A particularly salient Perspective focuses on the RESULTS of events; this Perspective provides an early mapping point for salient speech segments associated

447

with content words referring to actions. In all languages for which there are relevant data, whenever there is an acoustically salient past-tense or perfect marking on the verb, its first use by the child seems to be to comment on an immediately completed event that results in a visible change of state of some object. The past tense or perfect thus appears first on verbs like ‘fall’, ‘drop’, ‘break’, and ‘spill’ in the languages of our sample, as well as a broader sample including languages as diverse as Hindi (Varma, 1979), Mandarin (Erbaugh, 1978), Finnish (Toivainen, 1980), and Armenian (Geodakjan, 1976). De Lemos (1981) provides an insightful analysis of the emergence of the perfective in Portuguese between 1;6 and 2;1, showing its early role in adult-child dialogues to call attention to completion of actions and non-canonical states of objects. Note that this Perspective—the Result Perspective—applies to the Manipulative Scene. Earlier, we took a neutral perspective on that Scene, finding that a functor could be mapped onto one of two Highlights, the Agent (ergative) or Patient (accusative), following the location of salient speech segments in the input language. (However, we also found that in Kaluli, grammatical marking of the Manipulative Scene first emerges in past-tense utterances—that is, in regard to manipulations that have actually achieved results. Thus, early on, the Result Perspective may be especially relevant to Manipulative Scenes.) In taking the Result Perspective on the Manipulative Scene, the Change-of-State is highlighted, and it is natural to find that children map VERB functors onto this Perspective. The other Highlight should be the the 448

that undergoes the change of state encoded by the verb. Often this entity surfaces as the subject of an intransitive or middle-voice verb, and receives no special grammatical marking. However, in some situations early mapping of grammatical morphemes indicates children’s special attention to the affected entity as well as to the process. Good evidence comes from Antinucci and Miller’s (1976) study of Italian child language. In the standard language, the past tense of transitive verbs is expressed by a neutral (MASC:SG) form of the past participle, regardless of the number or gender of either subject or object (e.g. levato ‘removed’ or preso ‘took’ in the examples below). But Italian 2-year-olds make the participle agree with the OBJECT, saying things like Chi gli ha levate le gambe? ‘Who took off the legs?’, with plural agreement between levate ‘took:off+PL’ and le gambe ‘the+PL leg+PL’, instead of Chi gli ha levato le gambe; or Ho presa la campana ‘I took+FEM:SG the+FEM:SG bell+FEM:SG’, said by a boy, instead of Ho preso la campana. These children seem to be explicitly marking the past tense as a description of an affected object, since the agreement marking clearly indicates a Perspective on the patient rather than the agent. OBJECT

Evidence of restrictions of grammatical forms in other languages also points to early attempts to encode events without focusing on the agent. In Turkish the passive is agentless, and is marked by a single, syllabic particle on the verb. It is acquired early, and according to Savasir (1983) it is first used for a particular perspective in Manipulative Scenes—a Negative Result Perspective. He has noted early passives in situations in which a 449

child, having failed to bring about a desired result, focuses on the object of manipulation with a negative, third-person passive. For example, a child of 2;3 tries to open a door, fails, and says aç-il-mi-yor ‘open-PASS-NEG-PRES’ (= ‘it isn’t being opened’), thereby shifting attention from her own action to the resisting object. Often these early third-person negative passives immediately follow first-person actives with the same verb: ‘I’ll open it. It isn’t begin opened.’ Presumably, children hear similar sequences in the speech of others, where shifts in perspective are associated with the passive morpheme. (It may also be significant that the negative morpheme shifts stress to the preceding syllable, and, as a result, the passive receives additional emphasis with negation.) In terms of OP:FUNCTORS, then, a salient speech segment is mapped onto a Notion tied to Perspective on a prototypical Scene. Berman notes that in Hebrew the Result perspective provides the first context in which children use the passive-type intransitive verb patterns (without agent). In English, as well, the passive appears fairly early to encode the Result Perspective without mention of agent, using the got+ PARTICIPLE construction, as in It got broken. Horgan (1978) notes that such early passives tend to have inanimate subjects (i.e. patients) and verbs that are also used as statives. Here we are dealing with a construction type, rather than a grammatical morpheme, but the same Perspective seems to be at play. Herring (1981) provides suggestive evidence that English-speaking children may attempt to grammaticize the Result 450

Perspective. For example, she notes that Roger Brown’s Adam, at 3;0, said things like I broken that thing and I can let it spilled. Such forms may be attempts to mark the resultant state of an object on the verb, as in Antinucci and Miller’s Italian examples. In both languages, Basic Child Grammar seems to mark the same Notion, regardless of the conventions of the parental language. Berman (personal communication) discusses children’s simplification of the system of Hebrew verb patterns into those used to mark the Manipulative Activity Scene (causative transitive patterns) versus the Result Perspective (intransitive, middle voice patterns). Verbs which are highly result-focused, such as ‘break’, often emerge first in the intransitive, middle voice pattern, as, e.g. nišbar kadur ‘got:broke ball (=‘the ball got broken’), and only later in the transitive pattern (hu šavar kadur ‘he broke:TRANS ball’). On the other hand, verbs that are prototypically manipulative may appear in the transitive pattern even when expressing the Result Perspective; e.g. ze *zarak lasal ‘it threw: TRANS to (the) basket’ in place of nizrak ‘was thrown/got thrown’. These observations suggest that both the Manipulative Activity Scene and the Result Perspective emerge early, and are used as contrasting grammatical organizing points. Further, it is of interest to find that particular verbs may gravitate to one of these poles, suggesting a basic division between “Agentive” and Result Perspectives.

451

Grammaticization of Process, Result, and Aspect Basic Child Grammar orients to two major temporal Perspectives, which we can characterize as Result (punctual, completive) versus Process (nonpunctual, noncompletive, ongoing). This distinction is marked early on by the perfective-imperfective forms of verbs in Slavic languages, by the present (-Iyor)—past (-dl) forms in Turkish, by the progressive-past forms in English (-ing vs. -ed) and Japanese (-te iru vs. -ta), etc. In early transcripts of child speech it is difficult to decide if the child is marking a progressive-nonprogressive or an imperfective-perfective distinction, although the parental language may orient itself to only one of these major distinctions. I would suggest that the two temporal Perspectives of Basic Child Grammar are neutral and superordinate to these language-specific categories. That is to say, ‘process’ and ‘result’ figure as semantic distinctions in tense/aspect systems of individual languages, but they do so in conjunction with other distinctions, such as ‘habitual’, ‘iterative’, ‘continuous’, ‘past’, ‘currently relevant’, and others. The Result Perspective can develop, for example, into a perfective, a perfect, or a preterite; the Process Perspective can develop into an imperfective, a progressive, or an iterative. In each language the child must learn the ways in which Result and Process interact with other distinctions to determine the use of verb forms in the language. For example, English-speaking children must learn that their language

452

grammatically marks past events only as in progress or not, without regard to completion. Herring (1981) reports that Adam, at 2;7, said Cowboy did fighting me, apparently trying to mark both process (-ing) and endpoint (did); but English will require him to choose, eventually, between Cowboy fought me and Cowboy was fighting me, without grammatical marking of completion. A Polish child, on the other hand, will have to attend to completion (perfective aspect) but not the ongoing progress of a past event. Again, we find that Semantic Space provides Basic Child Grammar with a level of organization that serves as an opening wedge to the acquisition of language-specific grammatical distinctions, without at first biasing the child to any particular language.

Pragmatic Perspective In addition to taking temporal Perspectives on Scenes, little children can also take PRAGMATIC Perspectives on the roles played by utterances in prototypical interactions. That is, they can take different perspectives in regard to the social effect of an utterance. In languages where such Perspectives are conveyed by salient speech segments, as Japanese sentence-final particles, we have evidence of basic pragmatic Perspectives which are accessible to LMC. Clancy reports early emergence and pragmatically correct use of several such particles by age 1;11. It is clear from her summary that these forms emerge as part of several prototypical interactions, embodying particular Perspectives:

453

Yo is used when the child is encountering resistance or lack of mutuality, and feels he can impose his information or will on the addressee; ne is used when the child is in rapport with the addressee, agreeing with him or expecting his confirmation or approval. The emotional content of no is less fixed than yo and ne; typically no seems to be fairly neutral in affect, occurring in the ordinary give and take of information shared in the speech context (p. 435). Because pragmatic Perspectives are so often expressed by prosody, word-order shifts, and special vocabulary in the languages of our survey, we have little evidence of their role in guiding OP:FUNCTORS to the discovery of the meanings of affixes and particles. Similar particles appear early in child speech in Turkish and German. Schieffelin (1979, p. 272) reports similar dimensions underlying the choice of pronouns in Kaluli. She notes that in order for her New Guinea 2-year-olds to have arrived at their appropriate usage of these forms, the child must be: able to minimally assess the situation, to understand what is being done around her or him or what is being asked of her or him. The child has to recognize when it is correct to assert that she will do something herself or that she, not someone else, will do the task. The child must learn which forms can be used to initiate something in the discourse and which forms must be used to respond to something in the discourse. What has gone on in the prior discourse and nonverbal context must be taken into account.

454

Recent work at Berkeley and Nijmegen points to the role of such pragmatic Perspectives in the choice of functors. For example, Gee (Gee & Savasir, 1985; Gee, 1983) has found that English-speaking 3-year-olds use gonna and will to mark two types of futurity, depending on the “activity-type” in which the child is engaged. Gonna is used for activities of “Planning,” in which children refer to anticipated end-states of activities and to events and entities removed from current play contexts; while will is used in activities of “Undertaking,” in which children commit themselves to an immediate course of action in an interpersonal framework. These Perspectives, like those described for Japanese and Kaluli, show that children use linguistic means to focus on the expected effects of their actions upon others and on the temporal course of realization of their intentions. Deutsch and Budwig (1983) have made use of related distinctions to account for 2-year-olds’ marking of possession in English. For example, they have found that Adam tended to use my or mine when taking a “Volitional” Perspective on the Manipulative Scene, requesting or demanding objects, or claiming to maintain possession of objects; while he used his name as possessor (e.g. Adam truck) when taking a “Constative” Perspective, indicating possession that was not in question. With continuing research it should be possible to define a collection of basic pragmatic Perspectives such as Planning, Undertaking, Willing, and Stating. It is clear that some of these Perspectives are closely bound up with grammatical devices of interrogation, conditionality, and modality.

455

Further evidence for a pragmatic perspective of Focus is presented later in the chapter, in relation to the widespread finding that pragmatically appropriate variations in word order appear early in all languages which allow for such variations (OP:VARIABLE WORD ORDER). Clearly, prototypical Scenes have pragmatic components, and a full understanding of grammatical development requires attention to both referential-propositional and social-conversational meanings of utterances. However, since our data deal primarily with nonprag-matic aspects of grammar, we will go on to characterize other properties of Semantic Space.

Relevance and Proximity in Semantic Space So far we have described Semantic Space as a representational area in which we find certain Notions that serve to relate Highlights within Scenes, and other Notions that serve to relate the speaker’s Perspective to a Scene or one of its Highlights in an interactive context. In addition, Notions are located near one another in Semantic Space to the extent that they are RELEVANT to one another.14 Relevance is both an intuitive and an empirical dimension, and we will seek evidence for it at various points in our crosslinguistic exploration. At this point, in regard to the emergence of grammatical morphemes, we can ask (1) whether some semantic properties show affinities or distance from particular Scenes, and (2) whether those properties that are relevant to a Scene show affinities to a particular Highlight in that Scene.

456

Relevance of Semantic Properties to Scenes Early uses of functors in the Figure-Ground Scene (case inflections, locative particles and adpositions) map onto the Figure-Ground relation per se, without attending to particular manifestations of that relation. Children’s similar treatments of location and possession, and of various locative goals and recipients, suggest that ANIMACY may not be a Notion that children readily attend to in establishing mappings for functors—or at least the functors we have been considering. Additional evidence comes from the role of animacy in the Manipulative Scene. The prototypical patient is inanimate, and so the first accusatives always appear on inanimate nouns. Some languages, however, have distinct accusative forms for animate and inanimate objects, and it appears that this distinction is not readily acquired when animate nouns begin to figure as patients in child speech. Smoczyńska reports that the distinction in Polish, where two forms exist in the masculine singular, is not always adhered to in adult speech, and that children go beyond adult usage in tending to use one accusative inflection for both animate and inanimate nouns of this type. In Russian, as well, Gvozdev (1949) reports that the animacy distinction in the masculine singular is a relatively late acquisition, and that Zhenya at first used one accusative form for both animate and inanimate nouns.15 The Polish and Russian data could be simply interpreted as evidence that the accusative inflection maps onto a unitary Notion for children, and cannot be subdivided on semantic grounds. However, Russian provides

457

counterevidence. During the period when Zhenya did not differentially mark animate and inaminate accusatives, he DID distinguish whole and partial objects, differentially marking, for example, ‘give apple’—a single, whole object, versus ‘give milk’ or ‘give money’—part of a larger object. Morphological errors indicated that the partitive was used productively at this time. One can speculate that it is “relevant” to the Manipulative Scene whether one grasps or acts upon a whole object or part of a mass (cf. Brown’s 1957 evidence for children’s comprehension of the count noun—mass noun distinction in English), but not relevant whether the object is classified as animate or inanimate. Indeed, many inanimate objects in the child’s world are grammatically classified as animate, such as stuffed animals and dolls, and the Manipulative Scene is the same whether one ‘gives teddy bear’ or ‘gives ball’. In a similar vein, the Figure-Ground Scene is essentially the same whether one ‘goes to the toychest’ or ‘goes to papa’, and whether one ‘puts something in the toy-chest’ or ‘gives something to papa’. That is to say, animacy does not seem to be semantically relevant to Figures in the Manipulative Scene or to Grounds in the Figure-Ground Scene. More generally, Notions in Semantic Space have relative “affinities” or “proximities” to one another in terms of their participation in Scenes. Animacy is relatively “distant” from Notions that are mapped onto accusative and locative functors because it is not central to the prototypical Scenes that co-occur with use of these functors. As such, it will not be considered as a possible dimension in defining the use of these functors (and, 458

later, in building morphological paradigms). At this point it is useful to return to Sapir’s distinction between concrete and relational concepts, trying to account for patterns of adherence between particular instances of the two types of concepts. We have seen that Sapir’s universal linguistic distinction arises, ontogenetically, from the two OPs for mapping that we have been considering thus far. The linguistic expressions associated with Scenes break down into the content words denoting objects and actions, and residual material—functors—that are mapped onto relational Notions most RELEVANT to those Scenes. It appears, from the examples considered above, that animacy is not highly relevant to the meanings that children mark grammatically in the Manipulative and Figure-Ground Scenes, and that the part-whole distinction is relevant to the Manipulative Scene. (It remains to be determined whether animacy is, overall, simply not a highly ACCESSIBLE Notion for grammaticization, or whether it is just not relevant to these particular Scenes.) It is important to note that in introducing the criterion of relevance we have made a distinction between cognition in general and SEMANTICS, or that portion of cognition which, in Schlesinger’s sense, is required for constructing and using language. Little children are clearly cognizant of animacy as an important aspect of the world: they know the difference between stuffed dogs and living dogs, and perceive and treat them differently. However, in terms of Semantic Space, animacy has particular “locations” relative to other Notions. OP:FUNCTORS instructs LMC to “try to relate … functors … to adjacent referential units.” The 459

dimensions of relation will be determined by relevance to Scenes, rather than the total range of concepts applicable to objects and events. Given a Semantic Space structured along the lines sketched out above, LMC will not readily assign the feature of animacy to functors that are adjacent to words denoting entities that are acted upon and entities that function as Grounds of location and movement.

Relevance and the Location of Functors Relevance plays several roles in LMC. We have just noted that not everything that a child may know about an event is relevant for the mapping of functors. In addition, if a Notion is relevant, it may not be relevant to all of the lexical items that encode a Scene. We have found that transitivity morphemes are located in relation to names of agents and patients, that result morphemes are located in relation to words describing changes of state, and so on. OP:FUNCTORS instructs LMC to map a functor “onto an accessible grammaticizable Notion that is relevant to the meaning of an adjacent referential unit.” Continuing crosslinguistic research is needed to more fully specify the relations between functors and content words that LMC defines as relevant in this sense. For example, we have found that Basic Child Grammar marks some aspects of the Manipulative Scene on verbs, as in early Italian object agreement and Turkish passives. The Notion of Patient-of-Manipulation “interpenetrates” with Change-of-State, and is thus accessible to verb marking such as agreement and passive. By contrast, in Hungarian the verb must agree with the DEFINITENESS of the object, and Mac-Whinney reports that “this rule

460

probably causes children more difficulty than any other single rule in Hungarian grammar.” We know that children do mark definiteness elsewhere, particularly on articles associated with nouns (as in English, German, and Romance languages), in the Hebrew noun prefix, and on noun suffixes (as in Turkish). It would appear, then, that the verb is not a highly relevant location for such a Notion. We cannot conclude, however, that Notions encoded by functors are rigidly “pre-set” to occur in relation to particular parts of speech. What is important is the relevance of the Notion to the content word. Let us return to the Manipulative Scene. Recall that in Russian child language an early distinction is made inflectionally between whole and partial objects. A similar distinction is made in Finnish, and children younger than 2;0 (Toivainen, 1980) distinguish between objects as whole entities, using the accusative suffix, and partial objects, using the partitive suffix. However, in Finnish the same suffixes are used to distinguish completed from noncompleted processes. Toivainen reports that, at the same age, the inflections are also used to indicate the two temporal perspectives, as in saying ‘book+PARTITIVE’ while looking for a book, versus ‘book+ACC when a book was put into a bag. In his summary, partitives are “used frequently to denote an incomplete action” (p. 129) while accusatives are used to refer to “the completion of the process depicted” (p. 135). We can conclude that Notions of partial and whole objects, and “partial and whole” temporal events, are close to one another in Semantic Space. In face, one can easily think of Scenes in which acting on a partial object 461

coincides with a noncompleted activity, and in which completing an activity also involves acting on a whole object. The overlap is, of course, not total, but it is enough to enable us to use the English words “partial” and “whole” to refer to both objects and events. It is thus not surprising that LMC should find it easy to map aspectual Notions of completion and noncompletion onto functors associated with nouns as well as verbs.

Summary: Semantic Space and Functors It will take a great deal more crosslinguistic data to map out the dimensions of Semantic Space available to LMC. More detail will emerge later in this chapter, when we consider other OPs involved in unit formation and semantic organization. We will also see that Semantic Space is not static, but is continually reorganized with development, influenced both by conceptual development and by the semantic structures of the parental language. Our first pass has been aimed at laying out the framework within which OPs operate, and to characterize some of the most basic Notions that are grammaticized. To recapitulate: The utterances that children hear are filtered through OPs that leave a residue of perceptually salient segments. The situations in which utterances occur are represented as Scenes, in which salient objects, object relations, and activities stand out as Highlights. Perspectives can be taken on Scenes, and, as a consequence, some elements are foregrounded. Extracted speech segments are first mapped onto concrete representations of objects and activities. Remaining uninterpreted segments are mapped 462

onto less concrete, relational Notions associated with objects and activities, according to the relevance of these Notions to the Scenes with which they are associated.

The Persistence of Basic Notions I have suggested that some grammaticizable Notions serve as the primordial building blocks of Basic Child Grammar, giving a similar cast to early child speech everywhere. However, in asides, and in Slobin (1977), I have also noted that some of these Notions continue to function in adult use of language, underlying adult errors, language change, and language universals. The fundamental lineaments of Semantic Space remain as a framework, and can be discerned at other points in development beyond the beginnings. For example, Bowerman (1981) has noted errors later in development that are reminiscent of the early irrelevance of animacy to the Figure-Ground Scene. She reports confusions, in English, of put and give, such as: Christy, 3;4: Eva, 2;7:

You put the pink one to me. (request for mother to give her the pink one of two cups) Give some ice here, Mommy. Put some ice in here, Mommy. (pointing to ice crusher)

As Bowerman points out, “some languages do not formally mark the distinction between animate and inanimate goal that is observed in give vs. put.” She suggests “that children’s recurrent substitution errors may arise from deep-seated cognitive predispositions

463

towards perceiving certain kinds of similarities among events or relationships, regardless of whether these similiarities are formally recognized in the language being learned” (p. 17). These predispositions are inherent in the structure of Semantic Space, and are available to speakers for reorganizations of their lexical and grammatical systems. (See OP:SEMANTIC REORGANIZATION, below.) It is not yet clear, however, what types of form-function interactions trigger early or late recourse to particular levels of semantic organization. At this point, having laid out a beginning framework for grammatical morphemes, we can turn to the OPs that work to convert Basic Child Grammar into the grammars of individual languages—finding, at every turn, universal and particular aspects of the child’s construal of language.

GROUPING INFORMATION IN STORAGE We have now equipped LMC with a preliminary set of procedures for perceptually isolating content words and functors, and assigning preliminary meanings to functors. Eventually, however, much more subtle and language-specific morphological and syntactic analyses will have to be achieved by the child. These achievements require, first of all, more systematic grouping of stored linguistic material, in conjunction with strategies for reviewing and revising interim solutions. In this section we lay the framework for category formation; in the next two sections, REVIEW

464

STRATEGIES and INTERIM PRODUCTION STRATEGIES, we equip LMC with some general problem-solving procedures. Finally, having laid this groundwork, we can examine the crosslinguistic data in detail to propose a set of more specific OPs for grammar construction (ORGANIZING INFORMATION IN STORAGE: LINGUISTIC UNITS; FORM-FUNCTION MAPPING; POSITION OF ELEMENTS). To begin with, LMC must have means for creating word classes in storage and establishing higher-order patterns of co-occurrence of linguistic elements. This problem has been dealt with insightfully by Maratsos and Chalkley (1980; Maratsos, 1982), who have proposed that children form word classes, to some extent, on the basis of “co-predicting” grammatical operations that are applied to individual words. For example, in English, the words that occur with such functors as -ing, -s, -ed, and others, come to form a word class (“verbs”). At the same time, however, certain semantic criteria also play a role in establishing grammatical uses of word classes (e.g., the passive reliably applies to verbs “denoting actional or experiential relations,” and children must be able to make note of such characteristics as well as purely distributional facts). Maratsos (1982, p. 247) has recently stated clearly that both distributional and semantic facts must play a role in children’s formation of word classes: Formal categorization (given by co-predictive uses of terms) and semantic denotations … interact in complex fashions to determine the applicability of particular grammatical uses. Thus a claim that what makes formal 465

classes formal is definition by appearance in a set of grammatical uses is not a claim that semantic denotations of the terms are necessarily irrelevant. In fact, both kinds of analysis are clearly part of grammatical analysis. It is the not the task of this chapter to recapitulate or refine the efforts of Maratsos and Chalkley. Rather, we will take something like their position as a prerequisite for more specific OPs, as formulated in the following OP for the establishment of word classes in storage: OP (STORAGE): WORD CLASSES. Store together as a class all words (phonological speech unit and meaning) that co-occur with a given functor. Store together as a class words that co-occur with the same groups of functors across utterances. Try to systematize word classes on semantic grounds, forming prototypes and looking for common features. Such an OP cannot be substantiated or disconfirmed by crosslinguistic evidence. Rather, its usefulness lies in the fact that it provides a basis for other OPs that require the existence of word classes. (The provision for “prototype”-like semantics underlying word classes has received considerable support, for example, from the work of Bowerman, Clark, and de Villiers, as discussed in this collection and elsewhere.) Grammar-building also requires the establishment of functor classes, participating eventually in the construction of paradigms. For example, a child learning an inflectional language must note that a variety of 466

affixes occur in conjunction with nouns, and must try to assign distinct functions to the various affixes, eventually arriving at paradigms for case, number, etc. This process is explored in detail later in the chapter, in the presentation of OP: MORPHOLOGICAL PARADIGMS. For the present, we must provide LMC with the means for creating classes of grammatical morphemes, along with the well-established guiding principle of “one-to-one mapping.” The creation of word classes and the creation of functor classes operate “in tandem,” as expressed by the following OP: OP (STORAGE): FUNCTOR CLASSES. Store together all functors that co-occur with members of an established word class, and try to map each functor onto a distinct Notion. Once LMC has begun to establish morpheme classes (content word and functor classes), it can begin to keep track of sequences of classes, providing stored material for phrases and, eventually, sentence patterns. Everything we know about child language points to the conclusion that children keep track of sequential orders of elements: order tends to be preserved in imitation; amalgams and formulas perserve order; errors in order of bound morphemes are almost nonexistent; word-order errors are rare in fixed word-order languages like English; etc. Indeed, order is so essential to human language that an organism unequipped to notice and store sequential information could hardly acquire such systems. In addition, the orders that are significant in language are orders of CLASSES of elements, and, accordingly, LMC must be instructed to use class 467

membership of elements in setting up sequential information in storage. In addition to classes of content words and functors, it seems unavoidable to postulate organization in terms of higher-order entities as well. We will assume that LMC stores together sequences in elements that co-occur in the expression of basic Notions (phrases), and that LMC is equipped with definitions of “proposition types” in terms of some sort of presumably innate predicate calculus, allowing for the creation of linguistic clauses. (It would appear to be necessary, at the minimum, that LMC take note of verb valence—that is, the number of arguments taken by a verb, along with some specification of the type of arguments, probably at first in terms of basic semantic Notions such as those discussed above. It is beyond the scope of this chapter to speculate further on such issues; suffice it here to claim that some definition of “phrase,” “clause,” and “proposition” must be inherent to LMC. These terms will be essential to the formulation of OPs for placement of grammatical morphemes, discussed below.) These presuppositions provide for two more basic OPs, completing the prerequisites for the form of storage of linguistic material: OP (STORAGE): PHRASES. Store together sequences of word classes and functor classes that co-occur in the expression of a particular Notion or that co-occur with content words belonging to the same class. OP (STORAGE): CLAUSES. Store together ordered sequences of word classes and functor classes that

468

co-occur in the expression of a particular proposition type, along with a designation of the proposition type. We have now characterized all of the basic units that LMC requires for linguistic pattern-making: semantic Notions and proposition types, in conjunction with content words, functors, phrases, and clauses. What remains is the task of building particular languages out of acoustic and conceptual material stored in these terms. Note that these four OPs for Storage require and build upon one another. Word classes are not set up until functors have been established, and functor classes require the establishment of some word classes. Phrases and clauses cannot be set up until some classes of content words and grammatical morphemes are available. And so forth. It is thus necessary for LMC to have means for monitoring the outputs of particular OPs and coordinating the application of OPs throughout the course of acquisition. Accordingly, we must first provide LMC with very GENERAL procedures of operation—the Review Strategies and Interim Production Strategies—and finally we will be in a position to propose an array of task-specific strategies, guided by the general OPs presented below.

469

GENERAL PROBLEM-SOLVING STRATEGIES Review Strategies The Language-Making Capacity, like all learning devices, must constantly review its operations, monitoring input and success, and revising hypotheses and solutions accordingly. In their general form, the review strategies presented here are not specific to the task of language acquisition. Although they are phrased as OPs of LMC, it is clear that LMC must draw such principles from a store of cognitive procedures that are prerequisite to all of mental development, learning, and problem-solving throughout life.

Strengthening LMC is, by its very nature, continually involved in systematizing stored linguistic material, using and re-using all available and applicable OPs, across and within subsystems of the language. As a consequence, children come to develop an overall sense of the general form and consistent patterns of the parental language, moving from universal to more and more language-specific means of expression. In the process, the functioning of OPs becomes “specialized” to the particular language, in accordance with repeated attempts to solve linguistic problems in particular ways. In the broadest terms, “nothing succeeds like success”:

470

OP (REVIEW): STRENGTHENING. Whenever an attempted solution succeeds, apply the same strategies to similar problems.16 The “success” of a solution will be defined in terms of the goals set forth in specific OPs, to be introduced below. As a result of continued application of particular OPs to a particular language, some OPs will come to be favored, with the consequence that structures consistent with overall typological features of the language will become more and more accessible to the child. Following the format used in the rest of this chapter, supporting evidence for OPs will be presented in paragraphs introduced by the symbol §. The data are drawn from chapters in this collection, unless indicated otherwise. In order to keep this chapter within readable bounds, I have limited presentation of such evidence to some of the most pertinent examples. § One sequence of OP:STRENGTHENING is that the more pervasive a morphological category is in a language, the more readily will it be learned. For example, gender must be marked in almost every utterance in Hebrew, where it controls subject-verb agreement (including the form of imperatives), determiner-noun agreement (numerals, demonstratives, adjectives), and plural-marking. Furthermore, the markings show considerable similarity across part-of-speech environments. It is not surprising that gender should be acquired early as a GENERAL grammatical category in Hebrew. As Berman notes:

471

Here … we may seek an explanation in terms of the notion of a “formal imperative” in the sense of knowledge dictated by a pervasive structural property of the language being learned. Because so-called “gender” is the basic way of categorizing nouns in the grammar of Hebrew, and because all nouns fall within this binary system, the distinctions are crucially ingrained in the language, and they have far-reaching formal consequences—for agreement between nouns and their associated verbs, adjectives, and demonstratives … The child is therefore impelled to encode them very early on … By contrast, in Romance languages gender figures marginally, only controlling agreement within the nounphrase (gender-number marking on determiners, adjectives, and nouns). Clark reports typical early errors of gender-marking and agreement in French and Spanish, perhaps due to the fact that children have to pay much less attention to gender OVERALL in these languages. In Polish, where gender intersects in fairly regular fashion with casemarking, Smoczyńska reports early acquisition of gender-based casemarking. However such acquisition is delayed in German—where gender clearly controls casemarking in far fewer instances than Polish, and in Russian—where widespread homophony in case paradigms makes it difficult to isolate gender as a determining category.17 We can conclude that the more frequently a child succeeds in using grammatical gender to solve a linguistic problem in the domain of OP:MORPHOLOGICAL PARADIGMS (see below), the more likely will gender be considered a relevant distinction in solving further 472

morphological problems. Thus OP:STRENGTHENING can strengthen the use of a particular grammatical cateogry within a language. § It would be impossible to make any progress at all in acquiring a Semitic language like Hebrew or Arabic without arriving at the basic realization that lexical meanings can be expressed by consonant frames, with intercalated vowels carrying grammatical meanings (e.g. kotev ‘writes’, katav ‘wrote’, katuv ‘written’, ktav ‘handwriting’, etc.). Hebrew- and Arabic-speaking (Badry, 1983; Omar, 1973) children easily come to this realization, no doubt because of OP:STRENGTHENING (in conjunction with requisite perceptual and analytic capacities). By contrast, English-speaking children are late to realize the role of vowel alternation in consonant frames in their language, where such patterns are used only for a small number of irregular verb past tense and noun plural forms (e.g. ride/rode, slide/slid, mouse / mice), and with very little regularity within such forms.18 Furthermore, after vowel alternation becomes established as a standard morphological pattern in Hebrew, it can resist competition from other devices, like prefixation, which may appear to be more salient perceptually. For example, most verbs have a regular vowel alternation for tense, with present o-e and past a-a, as in kotev/katav ‘writes/wrote’, omed/amad ‘stand/ stood’, bone/bana ‘builds/built’, and ose/asa ‘do/did’. However, several types of verbs mark present tense by a form of prefixed m-, as in mekabel/kibel ‘get/got’ and matxil/hitxil ‘start/started’. Children overextend the m473

prefix to verbs whose form does not allow for the o-e/a-a alternation, and which do not distinguish present from past tense forms, producing present-tense errors like *me-sim (=sam ‘put:PRES/put:PAST’) and *mi-kanes (=nixnas ‘enter:PRES/enter:PAST’). In so doing, children are following PART of OP:EXTENSION (see below), which instructs LMC to try to mark a Notion (in this instance, present tense) wherever it can apply. However, OP:EXTENSION also instructs LMC to try to use the SAME linguistic means to mark the Notion in every instance. Israeli children do not go this far: they use m- where vowel alternation is not possible (given the forms of roots like sam and kanes, which do not allow for the o-e/a-a alternation), but they do NOT overextend m- to verbs where vowel alternation is possible. Thus they have two means of present-tense marking—vowel alternation and mprefixing. Because OP:STRENTHENING is responsible for a firm reliance on vowel alternation as the basic morphological device of the language, the full application of OP:EXTENSION is held in check, and m- prefixing is limited in application. (This is one of many examples of the joint operation or interaction of OPs). Repeated problem solutions along similar lines reinforce the Semitic-speaking child’s sense of learning a language of a particular TYPE, thus strengthening and maintaining the typologically consistent characteristics of the language (cf. Antinucci, Duranti, & Gebert, 1979). OP:STRENGTHENING instructs LMC to “apply the SAME strategies to SIMILAR problems.” In the examples above we have treated “same strategies”

474

in terms of morphological criteria used in the solution of problems in the domain of OP:MORPHOLOGICAL PARADIGMS. In the course of exploring additional OPs in detail, below, the notion of “same strategies” will become clearer. For now, it is important to note, as well, that children’s application of particular strategies to tasks of linguistic analysis can also reveal something of how children define “similar problems.” For example, children’s extensions of the use of grammatical morphemes suggest that they can define “similar problems” in terms of definitions of GRAMMATICAL CLASSES, giving reality to the categories established by OP:WORD CLASSES. § Data from several languages suggest that children define a general function of noun modification, treating diverse parts of speech as if they were members of a uniform modifier class. In German, Mills reports the extension of adjective endings to other prenominal modifiers, such as numerals and the quantifier all ‘all’ (e.g. die *zwei-en äpfel ‘the two-ADJ apples’, die *all-en manner ‘the all-ADJ men’ (= ‘all the men’)). She suggests: “Both numbers and all appear in certain contexts to behave like adjectives, so that this regularization is quite likely as well as indicating the strong tendency to establish patterns in the system.” In Japanese, the genitive particle no is extended from NOUN no NOUN constructions, indicating possession, to *ADJ no NOUN (a frequent error) and to *REL.CL no NOUN (at least in one child). As Clancy suggests: “It seems that these children have recognized the single grammatical category of nominal modifiers, and see the functional similarity between the two patterns of noun 475

modification which have been developing, one with nouns and one with adjectives.” Smoczyńska reports similar errors in Polish, where children try to collapse genitival possessors and adjectives into one category for purposes of case inflection. All of these examples indicate that the grammatical treatment of noun modification constitutes a set of “similar problems” for children, and that solutions arrived at for one subset in this problem space will be extended to others, in accordance with OP:STRENGTHENING.19

Monitoring Virtually every model of acquisition requires that the learner attend to discrepancies between expectations and outcomes. Eve Clark (1982, p. 172), for example, has discussed the “metacognitive” activities involved in language acquisition: Children need to line up elements of the system used in comprehension (whatever stored representations of words, phrases, or even whole utterances are addressed in memory when the child tries to understand some input utterances) with elements of the system used in production (the stored representations of forms retrieved from memory when the child tries to produce utterances to express his own intentions). One major component required to coordinate these two processes, then, is a mismatch detector…. The mechanism needed to detect such mismatches, I will argue, is provided by the general capacity to monitor, check, and, when necessary, repair the system being acquired—in this instance, language.

476

She offers extensive support for such metacognitive activities, using data of children’s speech repairs. In our terms, a basic developmental drive in LMC is provided by constant monitoring of input in terms of OPs. MacWhinney has elaborated this point into a collection of OPs for “tallying” and “criticism.”20 We can summarize these important observations as a general procedural OP for monitoring, checking, and internal repairing: OP (REVIEW): MONITORING. Compare utterances you hear with forms that you would produce in the same situation. Store mismatches and attempt to accommodate your grammar to unassimilated input forms by applying relevant OPs to the area of mismatch. If you find that a given OP is responsible for the mismatch, attempt to change the level of operation of that OP, or cease to use it with regard to the area of mismatch. In addition to on-line monitoring, LMC continues to work at solutions, checking analyses against current and inherent criteria for adequate grammatical description. Such continued internal processing can be characterized by the following OP: OP (REVIEW): PERSISTENCE. Whenever an OP fails to achieve a solution, and whenever a previous solution is found to be inadequate, return to the task from time to time until a solution is achieved.

477

It is important to note that criteria for solutions do not remain fixed, since analyses of some parts of the language can call for reanalyses elsewhere. The fact that LMC approaches language as a SYSTEM is inherent in the formulation of many of the OPs. This point is made repeatedly by the contributors to this collection, and has been stated with particular cogency by Smoczyńska: Any subsystem is acquired as part of the whole system of a language and its acquisition cannot be analyzed in isolation. If the structure and function of forms under acquisition are coherent with those of related parts of the system, and if the same criteria are maintained across subsystems, these forms are much easier to acquire than in cases where contradictions and incoherences are present…. The child not only discovers that the language he hears is rule-governed but also that the rules are interrelated to form a system. OP:STRENGTHENING reinforces the systematicity of linguistic solutions, and this factor plays a significant role in the functioning of OP:MONITORING and OP:PERSISTENCE.

Developmental Changes In addition, language-external development engenders continuing re-analyses of previous solutions. One important developmental influence comes from increasing processing capacity. We will not explore such effects in this chapter, but will assume that the capacities of the filters to notice and store primary speech data increase in scope and complexity, thereby providing new

478

sorts of data that cannot be immediately assimilated by existing procedures. Of more relevance to the issues considered here are aspects of cognitive development that influence the organization of Semantic Space. We have already discussed one example of such changes in regard to the accessibility of locative relations determined by conceptual growth (see “Locative Relations between Figure and Ground,” above). In simple terms, new notions seek new forms of expression. More deeply, however—as Melissa Bowerman has repeatedly pointed out—conceptual growth can bring about far-reaching REORGANIZATIONS in children’s linguistic systems. Relevant data are presented in sections on “Reorganizations in Development” in many of the chapters in this collection. Emerging conceptual capacities can interact with language-specific categories, allowing particular languages to influence the organization of Semantic Space beyond the categories of Basic Child Grammar, as, for example, in the Turkish distinction between witnessed and non-witnessed events, the Polish distinction between virile (masculine plural) and nonvirile (other plural) genders, and many others. The emergent cognitive and language-specific interactions in reorganization of Semantic Space are captured in the following OP: OP (REVIEW): SEMANTIC REORGANIZATION. Continually reexamine Semantic Space as your cognitive structures develop and as you discover the semantic categories used in your parental language. Analyze Notions into related subordinate Notions and 479

group Notions into new superordinate patterns; make semantic extensions of Notions and assimilate Notions to one another. Apply new analyses to existing semantic paradigms and definitions of functors and grammatical operations. This OP works in conjunction with constant dictionary building. Following OP:DICTIONARY, morphemes are stored “in terms of available semantic and pragmatic Notions in Semantic Space.” With reorganization of Semantic Space, the Dictionary must be reorganized accordingly, continually guided by all of the Review Strategies: OP (REVIEW): DICTIONARY REORGANIZATION. Review stored linguistic forms and their meanings, systematizing your Dictionary according to available OPs and the principles of linguistic organization that have proven useful in dealing with your language.

Range of Application of Procedures As OPs are applied to new problems, there is evidence that LMC appears to be conservative in the solutions it initially attempts. Maratsos (1983) has recently surveyed such evidence, concluding that: A recurring finding of the last years is that children often make highly specific analyses of combinations, and apply possible generalizations cautiously, rather than rapidly making highly general ones which are productively extended immediately.

480

Such conservatism is reflected in a final general procedure: OP (REVIEW): LIMITED FUNCTIONS. At first apply a solution to the smallest motivated category and do not extend it without evidence. LMC’s definition of “motivated category” will be based on the existing level of analysis of a linguistic domain, as given by OPs for grammatical organization (see below). “Evidence” must always be defined in terms of what is currently available to the mismatch detector of OP:MONITORING. For example, before the discovery of grammatical gender in a particular language, the smallest MOTIVATED category for pluralization may be simply NOUNS, and a single plural marker may be applied to all nouns. Thus in Hebrew the first stage of plural marking is characterized by application of the masculine ending to all nouns. At the next stage, gender comes to be determined solely on the basis of a particular noun-stem ending (feminine -a versus everything else), and the appropriate gender-based plural marker is applied only to nouns of that form, with the masculine ending applied—appropriately and inappropriately—to all other nouns. Thus “the smallest motivated category” for the masculine ending actually includes fewer nouns at stage two than at stage one, but OP:LIMITED FUNCTIONS applies at both stages. What is important is the basis of “motivation,” and it is in this regard that children are conservative, extending a rule only so far as the current level of analysis will allow. In the Hebrew example (as discussed in more detail in regard to 481

OP:MORPHOLOGICAL PARADIGMS, below) the level of analysis arrived at in stage two precludes recognition of other types of noun-stems as feminine, and therefore they cannot constitute evidence for a broader application of the feminine plural. Evidence changes as analyses change—guided by the OPs discussed above—but OP:LIMITED FUNCTIONS may operate throughout acquisition as new problems are encountered. With this brief summary of Review Strategies we are one step away from OPs for grammar construction (UNITS, MAPPING, and POSITION). However, we must first make an “aside” to consider some typical aspects of children’s speech production that may NOT reflect their level of grammatical organization. After a discussion of “Interim Production Strategies,” we will devote the remainder of the chapter to speech data that best reveal LMC’s grammatical propensities.

Interim Production Strategies At the outset of the chapter we noted that our primary data consist of patterns of child SPEECH, from which we seek to infer underlying linguistic organization. However, not all aspects of speech production may reflect such organization. There is no straightforward way to exclude from our analysis the effects of performance pressures on speech behavior except to try to rely on speech patterns that are reliably and frequently exhibited in regard to particular types of linguistic structures. However, I would like to exclude two types of production data from the crosslinguistic search for 482

grammar-building Operating Principles: (1) patterns that seem to be rote-learned or relatively unanalyzed, and (2) patterns that may be more indicative of short-term processing constraints than of a deeper level of grammatical analysis. Accordingly, I introduce three OPs at this point that function as “interim production strategies.” In a sense, these OPs function like the Filters, holding linguistic material in storage or in attention, allowing it to be worked on by other OPs. Unlike the Filters, however, they are OPs that govern production rather than perception. Like the Review Strategies, they are general procedures involved in problem-solving tasks. The evidence for Interim Production Strategies comes from the fact that some linguistic forms are used in rather specific ways in speech before they seem to have been fully systematized in the grammar. The appearance of such forms suggests that the child is attempting to stabilize the use of frequent forms and to maximally mark forms that are in the process of being mastered.

Memorization Roger Brown noted in 1973 that some functors were present in Stage 1 “telegraphic” speech in a number of languages, even if they were not used meaningfully. He suggested that frequency and perceptual salience could allow for the memorization of some forms (p. 88): If functor x has high frequency and high perceptual salience than whether conditioned or not, and whatever

483

its semantic role, it will occur in Stage I but only in prefabricated routines. Clifton Pye (1979, 1980) has gone on to consider the role of perceptual saliency in the acquisition of grammatical morphemes in English and Quiche Mayan, and has concluded that this factor is the major determinant of the point of acquisition of functors, rather than frequency or meaning. In our crosslinguistic data, as well, there are numerous reports of precocious but apparently uncomprehending use of salient grammatical morphemes. § Clancy reports early use of postnominal o and, frequently, of verb-final -te. She suggests that children may realize that a certain particle frequently occurs after a verb or a noun, before having fully ascertained its function. For some children, particle placement may even be restricted to a given sentence position. For example, ga generally marks the first noun as subject of a sentence, though it also occurs in other positions and in other functions. Some children go through a phase of simply using ga to mark the first noun in a sentence. Clancy suggests: Although children typically assume that the use of grammatical markers should make semantic sense (cf. Slobin, 1973), when the input does not permit the formulation of a semantic hypothesis, they will turn to other solutions. The errors which occur with ga in Japanese seem to indicate that some children make a syntactic, or positional hypothesis, namely, that ga follows the first nominal argument in a sentence. 484

§ Aksu-Koç and Slobin note that Turkish children sometimes insert meaningless syllables between the verb stem and the final person-number affixes, suggesting that the child attempts to retain some rhythmic picture of complex verbs, incomprehendingly inserting morphemes that sound like the passive and causative particles that appear in such positions. Facts such as these suggest an Interim Production Strategy based on attention to both form and position of perceptually salient elements: OP (PRODUCTION): UNINTERPRETED FORMS. If a speech element is frequent and perceptually salient, but has no obvious semantic or pragmatic function, use it in its salient form and position until you discover its function; otherwise do not use it. In addition to such uninterpreted forms, ANY form that can be rote memorized will appear in children’s speech. Stated as an OP: OP (PRODUCTION): ROTE. If a form appears frequently enough for you to have memorized it and made its use automatic, continue to use it in that manner while you are systematizing the use of similar but less automatized forms. Often such forms are used meaningfully, as in the early correct uses of irregular forms reported in all languages. (This accounts for the initial peak in “U-shaped” curves.) Even after regular principles have been acquired, they can sometimes be overridden by rote. MacWhinney has 485

taken “rote” to be a major point of departure in his “Dialectic Model,” competing with formations based on “analogy” and “combination.” And it is the continuing possibility of rote forms that has motivated child language investigators—at least since Berko (1958) and Bogoyavlenskiy (1957)—to use nonce items to assess children’s control of grammatical forms. In our framework, we must be aware of the continuing possibility that rote memorization may obscure or override the child’s insights into the structure of the parental language.

Maximal Substance It has been frequently reported—throughout the child language literature, and in this volume as well—that children often use more explicit forms in their speech than adults do. In many instances this reflects a highly analytic approach to language, and we discuss this possibility below in regard to OP:ANALYTIC FORM, according to which children prefer analytic to synthetic constructions. However, it is of interest to note that OVER explicit marking often occurs in the EARLY phases of productive control of a form or construction. This suggests that the phenomenon may also be a sort of “language-learning aid”—constituting an Interim Production Strategy rather than reflecting a deep-seated aspect of Basic Child Grammar. The issue cannot be easily resolved, since we have no way of knowing if the child considers overmarked or redundant forms to be an obligatory or systematic part of the grammar. However, at this point, let us consider some examples of such forms in terms of a production strategy:

486

OP (PRODUCTION): MAXIMAL SUBSTANCE. While you are mastering the linguistic expression of a Notion, mark that Notion with as much acoustic substance as possible, with maximal phonological separation of the form in question from adjacent speech units.21 This OP is based on a general problem-solving procedure that Annette Karmiloff-Smith (1979, 1981) has insightfully explored in several domains, including language acquisition: “Newly discovered distinctions must be marked tangibly externally before they can be integrated into an internal plurifunctional system (1979, p. 97).” In the realm of language, for example, she reports French-speaking children who make distinctions such as une voiture to mean ‘a car’ and une de voiture to mean ‘one car’; who use forms like toutes les miennes de voitures ‘all the:PL mine:PL of cars’ in place of mes voitures ‘my:PL cars’; and similar overly explicit forms. She argues that such slightly ungrammatical forms emerge when the child becomes aware of separate meaning elements, needing “to mark each function separately for a time in order to render the distinction tangible” (1979, p. 95). Similar phenomena appear in nonlinguistic tasks, suggesting the existence of a general “metaprocedural” level in problem-solving (1979, p. 115): The “stepping up” to a metaprocedural level appears to take place each time the child has a handle on his currently functioning system. This enables the child to treat the system, or parts thereof, as units functioning in their own right. Thus, each time a representational tool 487

functions well procedurally, the child somehow takes it apart to analyze its implicit components and thereby the representational tool becomes part of the problem space itself. The metaprocedural level corresponds to the “trough” in U-shaped learning curves—the point at which the child is most actively involved in linguistic analysis and rule formation. Thus evidence for OP:MAXIMAL SUBSTANCE is useful in revealing the grammatical categories that children attend to at various phases in acquisition. This issue constitutes the topic of a treatise in itself; at this point we can only consider a few examples of explicit and/or redundant marking. One consequence of OP:MAXIMAL SUBSTANCE is that children will prefer free morphemes over bound or contracted morphemes. At first, the types of grammatical morphemes present in a child’s speech will be simply perceptually determined; and, as a consequence, both free and bound morphemes may be present, depending on their “perceptual packaging.” Bound and contracted morphemes are present in amalgams and routines (such as English won’t, and early nouns with case suffixes in languages like Polish, Hungarian, or Turkish); and perceptually salient free morphemes are also present (such as German and Polish auxiliaries and Japanese particles). Evidence for OP:MAXIMAL SUBSTANCE comes at the point when a child begins to control both the free and the bound or contracted form of a given morpheme and tends to use only the free form, or, where possible, to use both in a single utterance.

488

§ Perhaps the best-known example of this principle is Bellugi’s (1967) finding in regard to English auxiliaries: “We notice that the children produce the non-contracted full form of will in their declarative sentences most of the time, even though in the speech that they hear, the contracted form ‘ll is the one that is produced by the mothers” (p. 106). Bellugi noted that this phenomenon occurred at the point in development at which children were using auxiliaries in declarative sentences as well as questions, echo statements, and emphatic inversions. This suggests that they were operating on a “metaprocedural level,” tangibly marking the auxiliaries for themselves. In this regard it is of interest to note Bellugi’s comment that “the auxiliary verbs in the child’s speech during this period are somehow very precise. There is no evidence of slurring pronunciation, and little contraction of forms” (p. 100). § Bellugi’s interesting observation of precise pronunciation of auxiliaries suggests that OP:MAXIMAL SUBSTANCE may operate to clearly mark functors in general. For example, Rūķe-Draviņa (1963) noted that in the Latvian speech of her son, newly acquired conjunctions and other connecting words were stressed, even if they were unstressed in adult speech. This principle may apply to bound functors as well. It has been noted for Russian-speaking children that “clarity and accuracy of pronunciation appear first of all in the inflections. At the same time the word stem continues to sound inarticulate” (Levina, quoted by Leont’yev, 1965, p. 101).

489

Redundant marking is frequently observed in situations where a Notion can be marked in several different ways, or in several different positions across constructions. While such redundancy may often reflect simple confusions, detailed examination of particular cases may point to a period of regular use of an Interim Production Strategy. In most cases we do not have sufficient data on individual children to decide this issue. However, frequent reports of redundant marking, such as the following, at least suggest the possibility of “metaprocedural” production. §Karmiloff-Smith’s French examples, such as toutes les miennes de voitures cited above, reflect attempts to “spread” marking of a semantic notion across an utterance, using an array of available means in the language. Clark notes similar French examples, such as *mon mien de chapeau à moi ‘my mine of hat to me’ (= ‘my hat’), in which three forms of first-person possessive marking appear in a single utterance. The same type of overmarking has been noted in English (Ivimey, 1975), where children have been observed to mark plural redundantly in possessives on the Berko test, producing forms with the numeral two such as two nizzes hats, and even forms with two and double-marking of posession, as two of the nizzes’ hats. These findings, like those of Karmiloff-Smith, are produced by children beyond preschool age under the pressure of testing—which suggests, again, that such forms represent the child’s attempt to tangibly hold onto distinctions for purposes of an immediate communicative or cognitive task.

490

Overmarking may also reflect the child’s confusion about where to express a notion in a particular utterance, if the language provides several different forms determined by structural criteria that may not yet be clear to the child. § In Hungarian, the conditional is marked by a postposition, volna in the past tense, and by a verb suffix, -na in the present. MacWhinney (1973, p. 520) notes two types of errors in conditional marking. Although he presents a small number of examples, it appears that younger children tend to use volna for both tenses. This is consistent with a preference for free forms, determined by the part of OP:MAX-IMAL SUBSTANCE that instructs LMC to strive for “maximal phonological separation of the form in question from adjacent speech units.” Older children, however, make errors of using both conditional markers in the present, as reflected in overmarked forms such as VERB-na volna in place of the appropriate VERB-na or the earlier error, VERB volna. § Similar errors are reported by Bellugi (1967) for later periods in the development of Adam and Sarah, in which a negated auxiliary occurs together with a negative pronoun, determiner, or adverb: No one didn’t took it. / No one’s not going to do what I’m doing. / He never won’t scare me. Bellugi concludes (p. 143): “It is not difficult to imagine a rule which would account for these sentences: something like apply negative wherever possible in single propositions.” At this point, again, it is difficult to decide between “rule” and “production strategy” in accounting for such forms, since we have no 491

access to the child’s grammar in respect to these constructions beyond their production. § An observation by Kuczaj (1978, p. 324) also suggests that redundant marking—in this case of the English past tense—is a late error that may reflect OP:MAXIMAL SUBSTANCE. He notes that 3- and 4-year-old children are rarely semantically redundant, at least insofar as irregular past-tense verbs are concerned. Older children, however, are as likely or even more likely to accept past + ed forms as they are base + ed forms, suggesting that the older children are more concerned with clearly marking a syntactic form than with avoiding semantic redundancy. Here, then, is support for Slobin’s (1973) hypothesis that children are predisposed to try to mark semantic notions clearly, although in this case the children have acquired a great deal of their native language by the time the errors caused by this predisposition become common. § In some instances, a particular inflection or particle has different privileges of occurrence in different constructions. This seems to pose children with considerable difficulty, leading to redundant marking during the process of acquisition. Several examples are offered in Polish. Person-marking is normally suffixed to the past participle of the verb, and children first learn this position. However, in conditionals the person-marker belongs on the conditional particle, by, rather than on the verb. Smoczyńska notes frequent double-marking: VERB+PERSON by+PERSON. The

492

conditional marker can also occur in two alternate positions, as an enclitic and postverbally. Again, Smoczyńska notes redundant marking (in both Polish and Russian), such as: a moja mamusia też BY miała-BY ładne włoski ‘and my mommy also WOULD have-WOULD pretty hair’. Another consequence of OP:MAXIMAL SUBSTANCE is an under-use of ellipsis where it is allowed by the language. Early on, material is lacking in child utterances because it has not been adequately acquired. There are many reports of deletion of recently acquired elements under time pressure or under the demands of producing complex utterances. Such omissions do not constitute communicative uses of ellipsis, but rather simple deletion of elements due to processing constraints. However, once the elements of a system have been mastered, ellipsis can be used communicatively, especially to background presupposed information. Although the data are limited, it seems that after children have learned to clearly mark each notion, they seem to find it difficult to make such pragmatic deletions. § In languages that clearly mark the verb for person, the use of a pronoun with a verb is specialized for functions such as introducing or re-introducing a topic, contrastive focus, and so forth. Suggestive data from Italian, Polish, and Turkish indicate that children in the older preschool period over-use pronouns, perhaps relying on the pronoun to clearly and distinctly mark a semantic role.

493

§ Clancy notes that Japanese is a language which extensively uses ellipsis to omit presupposed information, but that children are slow to make use of such patterns: The adult language presents Japanese children with a model which supports their early tendency to omit presupposed information, but the limits on this process must be acquired gradually. It is especially interesting that older Japanese children may under-use ellipsis, and produce too much information from an adult point of view. Ito (personal communication) reports that 3 and 4 year old Japanese children sometimes fail to use ellipsis where appropriate, creating redundant sentences. Smoczyńska (personal communication) has noted a similar tendency in Polish children beyond the early stages of language acquisition. The development of ellipsis in Japanese has not yet been investigated; it may show a U-shaped curve, with ellipsis being used less adequately by older children before it is used appropriately by adult standards. There is an interesting parallel here with Slobin’s claim (1973) that when a child first controls the full form of a linguistic entity which can undergo contraction or deletion, for a while only the full form will be used (p. 425).

STRATEGIES FOR GRAMMATICAL ORGANIZATION OF STORED INFORMATION Thus far we have equipped LMC with procedures for perceiving and storing speech input and carrying out

494

rudimentary sorting and classification of stored material, along with a set of general problem-solving strategies. What remains to be presented are procedures for ELABORATING and REFINING patterns in stored linguistic material. LMC must have means for cross-classifying and systematizing units in storage, and for adjusting grammatical systems when monitoring reveals mismatches between stored forms and input forms. The ultimate result is a grammar of the input language. Our crosslinguistic data allow us to explore the establishment of: (1) morphological classes and paradigms, (2) rules for more detailed form-meaning relations, and (3) rules for the positioning of morphemes in utterances. The following three sections deal with these issues of organizing information in storage. First we consider the child’s developing grasp of LINGUISTIC UNITS, exploring issues of allomorphy of word-stems and grammatical morphemes and the construction of paradigms. The following section, FORM-FUNCTION MAPPING, deals with ways in which meanings of grammatical forms are analyzed, extended, and limited in the course of development. In the final section, POSITION OF ELEMENTS, early syntactic preferences are defined by OPs for functor placement and word order.

Organizing Information in Storage: Linguistic Units The chapters in this collection, along with numerous other studies of child language, provide repeated evidence of children’s attempts to adhere to one-to-one mappings between semantic entities and speech forms. 495

In an earlier comparison (Slobin, 1977) of diachronic tendencies in child language, historical language change, and formation of pidgin and creole languages I proposed a basic ground rule to “be clear”—that is, a constraint “that the surface structures of Language must not be too different in form and organization from the semantic structures which underlie them” (p. 186). The data reviewed in the present chapter reflect pervasive tendencies towards transparency and iconicity (Slobin, 1985). However, these are tendencies, and not absolutes—as we see in children’s acceptance of certain sorts of semantic conflation and certain sorts of allomorphy. The tendency has been phrased most sharply by Naro (1978) as an idealized strategy for pidginization that he has called “the factorization principle”: “Express each invariant, separately intuited element of meaning by at least one phonologically separate, invariant stress-bearing form … [and] avoid excessive accumulation of separately intuited elements of meaning in single surface units.” This ideal is often approached by children, but even at very early periods of grammatical development we find that some principled bases of allomorphy ARE available to LMC. Just as we have found nodes of related meanings in a semantic similarity space, we find nodes of related phonological forms in a formal similarity space. The dimensions of this space can be revealed by examining both one-to-one and many-to-one form-function relations in child speech. The ways in which LMC deals with variant forms are defined by OPs for the creation and organization of linguistic units: OP:WORD FORMS, OP:PHONOLOGICAL CONDITIONING OF ALLOMORPHY, OP:MOR496

PHOLOGICAL PARADIGMS, OP:CONNECTIVES, and OP:CANONICAL CLAUSE FORM. Preliminary grounding principles have provided the child with a Dictionary in which regularly occurring sound sequences are stored with their meanings. OP:UNITS and OP:UNIT FORMATION operate within a phonological similarity space to store together units that are “same” or “phonologically similar.” However, we have no clear definition of the dimensions according to which children hear some sound sequences as instances of the same morpheme, and it is beyond the bounds of this chapter to attempt to provide such a definition. Rather, we will assume that children identify morphemes in ways that are not too different from our own, and we will ask how they deal with variations in the shapes of words. Our evidence, unfortunately, comes from production. There is nothing in our crosslinguistic review about how children perceive varying forms of words, but clearly they must recognize a much wider range of forms than they produce. For example, some Hebrew nouns require a stem change with plural suffixation, as simla/smalot ‘dress/dresses’; children, however, prefer to add to an invariant stem, producing the plural *simlot. Yet we would not want to claim that they fail to comprehend the adult plural, or fail to hear it as related to the singular form. Our question, rather, will be one of determining the situations in which children seem to accept or reject alternate morpheme forms for use in their own speech. In this section we will look at word-stems, and in the next section, affix allomorphy—having already equipped the child with a

497

distinction between content words (word-stems) and functors. The OPs proposed here are different from the OPs considered thus far in that they allow for failure to find solutions and for interim solutions. Each solution is operated upon by OP:PERSISTENCE and OP:STRENGTHENING, and is sensitive to the existing state of knowledge and level of organization of the language being acquired. In regard to establishing word forms, children are motivated by detecting alternate forms, and they try to find principled bases for distinguishing forms on phonological or semantic grounds. Only if they fail, do they adhere to a single form. These procedures are summarized in an OP that has several levels of operation:

Word Forms OP (UNITS): WORD FORMS. If you discover more than one form of a word or word-stem in storage, or if monitoring reveals a mismatch between your word form and that in the input, try to find a phonological or semantic basis for distinguishing the forms: (a) Phonologically attempt to change your word form in the given environment, following a hierarchy of possible adjustments of word forms. At first try to maintain the consonant frame and syllable structure (number of syllables, stress placement). (b) Try to find distinct meanings for words or word-stems that occur in varying forms, checking for relevant Notions. (c) If you cannot find a principled basis for

498

differentiating the forms of a word or word-stem, pick one form as basic and use it in all environments. (a) Phonological Adjustment of Word Forms. Part (a) of OP:WORD FORMS appeals to a “hierarchy of possible adjustments of word forms.” There is evidence, for example, that the consonantal-rhythmic structure of a word provides a frame within which vowel changes (and probably also tonal changes) can be tolerated, whereas deformation of the consonantal-rhythmic structure is at first resisted. Hebrew-speaking children seem to easily recognize and accept a wide range of patterned vowel alternations, as discussed earlier, while having more difficulty with vowel deletion and stress shift, as in the singular/plural pairs: simlá/* simlót in place of smalót, béged/*bégedim in place of bgadím, and other examples offered by Berman. Bybee and Slobin (1982) found that English-speaking children more readily learn irregular past-tense forms that maintain a consonant frame (e.g. break/broke, give/gave), as opposed to those which are open syllables (e.g. blow/blew, see/saw). Apparently word forms sharing a consonantal-rhythmic frame are relatively close to one another in a phonological similarity space.22 Stem changes that are phonologically conditioned by affixes suggest a hierarchy of acceptable alternation. For example, Turkish children apparently have no difficulty in devoicing final stops before vocalic affixes, as in kitap/kitab-l ‘book’/’book-ACC’. However, Israeli children have difficulties with stop-spirant alternation, making errors such as *li-ktov INF-write’ (= ‘to write’) for li-xtov, maintaining the initial stop of the trigram root 499

k-t-v (as in katav-ti ‘write-1SG:PAST’ = ‘I wrote’); or, conversely, taking the spirantized version as basic, as in *fizar-ti ‘scatter-1SG:PAST’ for pizar-ti, along with le-fazer ‘INF-scatter’. The voiced-unvoiced alternation may be more accessible (more “natural”) than the stop-spirant alternation. MacWhinney, following work in natural phonology, suggests that the hierarchy of possible adjustments of word forms may be based on universal predispositions: The most primitive and general type of pattern is the phonotactic predisposition (Ingram, 1979). Phonological predispositions are conceived of as universal processes which favor certain types of assimilations and simplifications in the articulatory output. For example, the predisposition for vowel harmony would reduce /dædi/ to /dædæ/. Stampe (1969) argues that such simplifications are unlearned consequences of the way we control our vocal apparatus. The languages in our sample offer few instances of the operation of phonotactic predispositions in the acquisition of alternate forms of word-stems; we will find this principle to play a role in phonological conditioning of affix allomorphy as well. § There is one remarkable example of phonological adjustment of word stems as defined by a following SYNTACTIC, rather than phonological environment. In Hungarian, lengthening of a final vowel before a suffix is acquired before 2;0. In fact, MacWhinney (1978) reported a correct generalization of final vowel 500

lengthening to a nonce stem at age 1;8. One of his longitudinal subjects, Zoli, overextended final vowel lengthening to words preceding the word is ‘also’ between the ages of 1 ;6 and 1;10. Apparently the child had formed a general rule of vowel lengthening before a class of functors, including ‘also’, which is functor-like, but not a suffix in Hungarian. Here there is no phonetic or articulatory “congruence” between vowel lengthening and functor affixation. Apparently, changing a stem by lengthening the final vowel does not seriously deform the stem. In a formal similarity space that defines degrees of deformation of items, pairs of words differing only in vowel length would seem to be closer than pairs of words differing, for example, in stress or number of syllables. And in addition, on a hierarchy of nonsemantic grounds for changes in word forms, both phonological and morphosyntactic grounds can be found early on as factors conditioning stem allomorphy. Thus the one-to-one principle must already be modified to some extent. For the most part, though, our data show functioning at level (c)—use of one form as basic. However, there is one important suggestion of a level (b) solution, showing a domain in which a semantic distinction is salient enough to motivate one-to-one mapping—namely, the domain of ASPECT. (b) Semantic Adjustment of Word Forms. In the Slavic languages (Czech, (Pačesová, 1979) Polish, Russian, Serbo-Croatian), where verb-stems change for aspect, children readily acquire alternate verb forms to express the basic distinction between perfective and 501

imperfective. For example, Smoczyńska notes Polish child errors which reflect an attempt to establish pairs of perfective and imperfective verb forms, such as pomylić się [PFV] / *pomylić się [IPFV] in place of pomylić się / mylić się ‘to make a mistake’. Gvozdev (1961, p. 426) reviews a wide range of aspectual neologisms in Russian child speech, and concludes: “These childish neologisms present numerous examples pointing to the fact that imperfective and perfective verbs are joined in pairs: where a verb of one aspect is present, a verb of the other aspect emerges as well.” Joan Bybee (1985a,b) has suggested that the more RELEVANT a semantic category is to the meaning of the verb, the more likely is it to fuse with the verb phonologically. She notes (1985a, p. 15) “that aspect is the category that is most directly and exclusively relevant to the verb.” On the basis of a systematic sample of the world’s languages, she has found that “aspect conditions changes in the verb stem more frequently than any other inflectional category” (1985, p. 27). Accordingly, if there is one category that children accept as conditioning verb-stem alternation, it should be aspect. As we see below, other semantic categories, such as tense and person, do not readily provide the child with principled means of verb-stem alternation, suggesting that semantic Notions are available in a HIERARCHY OF RELEVANCE to lexical items. (c) Failure to Adjust Word Forms. Situations in which children persist in using a single word stem (or confuse alternate forms) reveal lower levels on hierarchies of semantic and phonological conditioning of word forms. On the semantic plane, we find neutralizations of tense and person distinctions: 502

§ TENSE. Where verb-stems change for tense, as in Romance, Germanic, and Slavic languages, children tend to use one form for all tenses, as in: French *prendu for pris ‘took’, from the stem prend-; German *gedenkt for gedacht ‘thought’, from the stem denk-; Polish *pierzył for prał ‘washed’, from PRES pierze rather than INF prác. Note that the child DOES mark tense by affixation on verbs or by the use of auxiliaries; but tense is probably not RELEVANT enough to the meaning of the verb itself to condition systematic change of the verb-stem as well. § PERSON. Children readily acquire person/number affixes on verbs, but where verb-stems change for person, as in Romance, Germanic, and Slavic languages, children tend to use one form for all persons, as in: Spanish *tieno for tengo ‘have:lSG’, retaining tien-, stem of 2-3SG; German *habt for hat ‘have:3SG’, retaining hab-, stem of 3SG, 1-3PL, and INF; Russian *vidu for vižu ‘see:lSG’, retaining vid-, stem of other persons and INF. Following Bybee, distinctions of person do not seem relevant enough to verb meaning to condition the phonological form of the stem. (An event described by a verb changes character if it is seen as perfective or imperfective, but not if it is carried out at one point in time or another, or by a participant described as first, second, or third person in a discourse.) On the phonological plane, it is hard to generalize from available developmental data, though it is evident that many kinds of stem changes pose difficulties to children. We have noted above that children tend to preserve a consonantal-rhythmic frame, thus finding it difficult to 503

acquire vowel deletions and accompanying stress changes. There are also reported tendencies to retain the normal stress pattern of the root or citation form, rather than shift stress to an affix. This is shown, for example, in stress maintenance in early Hebrew plurals, which require stress shift to the plural suffix (e.g. šaón/*šaónim for šeoním ‘clocks’); in Russian past tense forms which shift stress for the feminine suffix (e.g. dal/*dála for dalá ‘give:PAST:MASC’/‘give:PAST:FEM’; and elsewhere). (Gvozdev finds early acquisition of gender suffixes on verb past tense forms, but prolonged difficulty in acquiring the appropriate root alternations and stress shifts. He notes (1961, p. 419): “All of the difficulties are to be found exclusively in the stems.”) Hungarian presents considerable problems for extraction of both roots and suffixes, with difficulties in acquiring changes in basic word forms such as vowel insertion or deletion, vowel shortening, and nonstandard vowel harmony. (Standard vowel harmony and vowel lengthening seem to be easily accessible, however.) As MacWhinney comments, “morphophonolo-gical processes are very numerous and productive. However, … the irregularity and complexity of Hungarian morphophonological processes present the child with a major acquisitional challenge.” It is beyond the scope of this chapter to specify in detail how children ultimately solve such challenges; however, some additional evidence is provided by examination of the acquisition of phonological conditioning of choice of affix.

504

Phonological Conditioning of Allomorphy As in the case of word-stem allomorphy, some kinds of functor allomorphy are accessible to LMC. Where allomorphy is phonologically conditioned, we have further evidence for predispositions, or natural dimensions of phonological similarity in a Formal Similarity Space. Phonologically conditioned variants of affixes provide obvious, though usually neglected evidence of early many-to-one form-function correspondences. For example, the English-speaking child who controls voicing assimilation easily produces two different plural suffixes, /s/ and /z/, though we usually attribute one-to-one mapping to the child who adds a final sibilant to dogs and cats. Voicing assimilation is not simply an automatic consequence of the use of the articulatory apparatus, since children can also say things like dogsled and that zebra, thus showing an awareness of word boundaries and the operation of a morphophonological rule within a word. We tend to disregard such accomplishments as “low-level” adjustments, but it is important to note that these facts limit the generality of claims for semantic bases of early grammar, as expressed, for example, in my 1973 Operating Principle G, “The use of grammatical markers should make semantic sense.” It will be necessary to modify a proposed consequence of that principle, Universal Gl, which stated: “When selection of an appropriate inflection among a group of inflections performing the same semantic function is determined by arbitrary formal criteria (e.g. phonological shape of stem, number of syllables in stem, arbitrary gender of stem), the child initially tends to use a single form in all

505

environments, ignoring formal selection restrictions” (Slobin, 1973, p. 206). What is needed is a procedure for adjusting affixes to stems on phonological grounds, making use of morpheme classes established by OP:WORD CLASSES and OP:FUNCTOR CLASSES, along with an accessibility hierarchy of phonological adjustments of adjacent forms: OP (UNITS): PHONOLOGICAL CONDITIONING OF ALLOMORPHY. If you discover diverse forms of a functor expressing a given Notion with regard to words of a particular class, compare these forms with the forms of the other functors and/or the word-stems with which they occur. (a) If you find phonological similarities between co-occurring forms, regularly adjust the form of the functor to fit its environment (following your hierarchy of predispositions for phonological conditioning of allomorphy). (b) If you fail, use a single form of the functor where possible and/or omit the functor. Note that the default outcome (b) allows either for use of a single form (one-to-one mapping) or omission (telegraphic speech). Both of these default outcomes are widely attested in the literature. In any individual instance, one outcome or the other will be determined by a complex of factors including perceptual salience, frequency, diversity of forms (and consequent availability of a single clear default form), etc.

506

Evidence for outcome (a) is limited in our sample. It is clear from English /-s ~ -z/ (N.PL, V.3SG, POSS) and /-d~-t/ (PAST) alternations that voicing assimilation is a strong predisposition. Similar examples of early voicing assimilation appear in Turkish locative inflections (e.g. ev-DE ‘house-LOC’ / bardak-TA ‘glass-LOC’), and in other languages. The Turkish example also shows the effect of vowel harmony on affixation (-de ~ -ta). In both Turkish and Hungarian, vowel harmony is easily acquired—both for adjustment of affix to stem and for affixes to preceding affixes (as in Turkish göz-ler-im ‘eye-PL-1SG: POSS’ = ‘my eyes’ / kuş-lar-im ‘my birds’; and even longer strings, e.g. göz-ler-im-de ‘eye-PL-lSG:POSS-LOC’ = ‘in my eyes’). MacWhinney reports: “I have often had it pointed out to me, by both linguists and non-linguists alike, that even in their earliest words, Hungarian children almost never make errors in vowel harmony.” With detailed study of the acquisition of more types of languages it will be possible to fill out the dimensions of phonological similarity that play a role in the functioning of this OP. However, phonological conditioning of allomorphy is not a central issue in the specification of LMC. Of greater interest are children’s uses of more “cognitive” categories for classifying apparently synonymous functors, as evidenced by the development and elaboration of morphological paradigms.

Morphological Paradigms Children can learn to use variant forms of a functor if they can find a principled basis to distinguish use of the forms. It is helpful to represent such solutions in terms of

507

PARADIGMS in which functor allomorphs are cross-classified on grammatically relevant dimensions. A familiar instance is provided by case-inflectional paradigms, where, for example, choice of the appropriate allomorph for the accusative can be determined by semantic criteria such as animacy and number, by the phonological shape of the citation form, or by desemanticized categories such as formal gender. The capacity to create paradigms is central to LMC. Crosslinguistic study of children’s paradigm construction reveals semantic and formal dimensions used by LMC to sort and classify linguistic material. Pinker and Lebeaux (1981) have pursued this issue in some detail, proposing that language learners sort affixes into matrices of grammatical features, beginning with quite simple rote paradigms for individual words, such as: PAST broke PRESENT break Sets of related affixes participate in the construction of broader paradigms. They propose that “grammatical information … serves as an indexing system, under which particular affixes are appended”; and go on to suggest that “the child is seen as hypothesizing a dimension and a level in the matrix in which the affix is put.” Pinker and Lebeaux add the additional constraint that “a matrix only allows a single entry (affix) per slot.” In our terms, the dimensions hypothesized by the child 508

are accessible Notions or formal principles (to be defined below), and the one-entry-per-cell principle is a consequence of OP:FUNCTOR CLASSES, which instructs LMC to “try to map each functor onto a distinct Notion.” This principle of one-to-one mapping is also built into OP:MORPHOLOGICAL PARADIGMS. The OP is complex, and we will consider it paragraph-by-paragraph. The point of departure, like that of the previous OP, is the child’s discovery of diverse forms to express a given function. We assume that such a discovery impels LMC to assign each form to a uniquely defined cell in a paradigm, thus implicitly following a one-to-one principle. We assume, further, that the most accessible solutions are phonological conditioning (following OP:PHONOLOGICAL CONDITIONING) and semantic conditioning of allomorphy, followed by various attempts at formal solutions. OP (UNITS): MORPHOLOGICAL PARADIGMS. If you find more than one functor expressing a given Notion relative to a particular word class, and choice of functor cannot be determined by phonological conditioning: (a) Try to find semantic grounds for subdividing the Notion expressed by the functors, and map each new Notion onto one of the functors. (a) Paradigms Established on Semantic Grounds. Semantic grounds are readily available to LMC for assigning functions to grammatical morphemes, as discussed at length in regard to OP: FUNCTORS. 509

Children quickly learn to pick verbal affixes expressing notions of aspect, tense, person, and number; and nominal affixes expressing notions of number and an array of basic cases (accusative, ergative, locative, possessive, instrumental, etc.). However, OP:MORPHOLOGICAL PARADIGMS is concerned with the problem of SUBDIVIDING a Notion that is already expressed by a functor. It seems, from the data at hand, that semantically-based subdivisions pose numerous difficulties. We have seen that Slavic-speaking children find it difficult to subdivide the masculine accusative case into separate forms for animate and inanimate direct objects. Additionally, Polish children find it exceptionally difficult to subdivide case inflection in the plural on grounds of “virility” (masculine plural versus all other plurals). The critical issue, again, seems to be one of the mutual relevance of intersecting semantic Notions. We have noted that children at first have difficulty in subdividing the Notion of ground into static-point versus end-point-of-trajectory, or ground-as-location versus ground-as-person (possessor, recipient). These subdivisions, however, are arrived at earlier than subdivisions of the accusative on grounds of animacy or virility. Presumably this is because distinctions such as static-dynamic, location-possessor, and goal-recipient ARE of some relevance to locative Notions. Another example of early semantically-based subdivision of a case hierarchy is acquisition of accusative-partitive distinctions. As discussed in regard to the Manipulative Activity Scene, the distinction between whole and partial object is highly relevant to grammatical marking of 510

direct object, and children are quick to acquire distinct direct object functors for these two Notions in Slavic and Finnish. I suspect that detailed study would reveal a hierarchy of intersecting Notions, according to which some types of semantically-based paradigms would be established earlier than others. § Another example, in the realm of verb paradigms, is suggestive of the directions that such study could take. Stages in the acquisition of the Turkish past tense (Aksu-Koç, in press; Aksu-Koç & Slobin) present good evidence of an accessibility hierarchy for paradigm construction. There are two past-tense morphemes, a “witnessed” form for reporting direct experience, and a “non-witnessed” form for information gained by inference or hearsay. At first children use only the direct experience form (the default option for OP:MORPHOLOGICAL PARADIGMS). At the next stage, it is fairly easy for them to set up a preliminary subdivision of the past-tense paradigm on semantic grounds, using one form to report directly-experienced events and the other for non-witnessed events as inferred from physical evidence. Evidentiality on perceptual-experiential grounds is apparently an accessible Notion early on, and is relevant to verbs. However, it takes children much longer to include hearsay among the semantic criteria for use of the non-witnessed form, and they continue to report hearsay as direct experience. Finally, in accordance with OP:SEMANTIC REORGANIZATION, they re-allocate hearsay to the non-witnessed cell in the paradigm of verb inflections. Here we have a nice example of successive

511

refinements in the semantic criteria governing the choice of an affix. (b) Paradigms Established on Phonological Grounds. In most instances, affix allomorphy is not based on simple phonological conditioning or on semantic subdivision of a larger functor class. The child is faced with an array of seemingly arbitrary, formal problems, such as the choice of case inflections on the basis of grammatical gender of nouns in German and Slavic languages, or the alternate forms of the English comparative on the basis of the number of syllables in the stem (e.g. prettier, but more beautiful rather than *beautifuler). Children cope with many formal problems surprisingly well, attending to the shape of citation forms (b), and, later, to the environments in which citation forms occur (c). As MacWhinney has pointed out, “uninflected citation forms are likely to be picked up as intonational units.” He presents evidence from Hungarian, English, German, Hebrew, Estonian, Polish, and Mandarin that children are sensitive to citation forms such as singular nouns in the nominative or unmarked case and verbs in bare imperative forms, using such forms as basic stems for affixation. The salience of citation forms figures in the nonsemantic solutions provided by OP:MORPHOLOGICAL PARADIGMS: OP:MORPHOLOGICAL PARADIGMS (b). If you cannot find semantic grounds for choice of functor, check the citation forms of the associated words or stems and try to differentiate them on systematic phonological grounds. If you succeed, set up a

512

paradigm in which choice of functor is conditioned by the phonological shape of the citation form. § The best example of option (b) in our data is Smoczyńska’s comparison of the acquisition of the Polish and Russian paradigms of case inflections on nouns. In both languages the choice of case suffix is based on the formal gender of the noun, as indicated by the phonological shape of the citation form (nominative singular), with no consistent semantic support. Masculine nouns typically have a zero ending (stem-final consonant); feminine nouns end in -a; neuter nouns in -o/-e. Polish children acquire this distinction before the age of 2, and when case suffixes emerge the appropriate allomorph is selected according to the gender of the noun. Russian children, however, take a very long time to sort out the various case allomorphs, and at first use a single salient form for each case, ignoring gender. The difference is attributable to the transparency and predictability of the system. In Russian, unstressed -o is pronounced the same as unstressed -a, thus neutralizing the distinction between the citation forms of many neuter and feminine nouns. In addition, many diminutive forms end in -a and are declined like feminine nouns, even if they are affixed to masculine or neuter nouns. Thus OP:MORPHOLOGICAL PARADIGMS (b) is blocked, since the child will not succeed in choosing case allomorphs on the basis of differentiation of citation forms of nouns on systematic phonological sounds. Consequently the default option (d) is taken at first—namely the “inflectional imperialism” (Slobin,

513

1966a,b) of using a single affix for each case. Smoczyńska concludes: Early acquisition of gender in Polish shows that the child is able to learn rules which are based on totally arbitrary criteria, provided these criteria are consistent and clear enough to be discovered; and that selecting one salient (or phonetically unique) ending is only a strategy that the Russian children apply when faced with the impossibility of discovering adult criteria (p. 646). § Similar reliance on nonsemantic criteria is shown by the apparently easy acquisition of appropriate masculine and feminine articles in Romance languages. Clark notes that French-speaking children are able to rely on the phonological shape of the noun to pick the appropriate article by age 3. On the basis of a detailed series of experiments on gender-marking in French, Karmiloff-Smith concludes that “as soon as articles were used consistently, the child constructed a very powerful, implicit system of phonological rules, based on the consistency … of phonological changes in word endings. Classification by gender was clearly based on suffixes” (1979, p. 167). It is of interest that the phonological strategy took precedence over the semantic cue of natural gender. For example, in one experiment children were presented with two obviously female “Martians,” introduced with the nonsense words deux bicrons ‘two “bicrons” ‘, with the typically masculine ending-on. Three-year-olds followed the phonological, rather than the semantic cue, referring to one of these creatures as UN bicron ‘a:MASC’ or LE bicron ‘the:MASC’, thus giving formal masculine gender to bicron on the basis of 514

its ending. At the same time, they used the feminine pronoun elle to refer to the creature, indicating full awareness of its sex. Here we have a very interesting case of competition between formal (phonological) and semantic bases of allomorphy, and it would seem that a clear phonological cue is a more reliable basis for article selection, while sex is a reliable basis for choice of referential pronoun. Perhaps natural gender is low on a hierarchy of accessible grammaticizable Notions, allowing a clear phonological distinction to take precedence. Thus, in this instance, option (a) of OP:MORPHOLOGICAL PARADIGMS fails at first because of the low salience of gender in Semantic Space (at least in regard to its relevance to the functions expressed by articles). § Similar evidence comes from Spanish and Hebrew, where it is clear that children easily establish binary morphological categories using formal criteria. Clark reports that Spanish-speaking children are quick to divide nouns into two classes on the basis of their endings, even adjusting inconsistencies in noun forms to bring them into agreement with article choice. For example, nouns that take the indefinite article un generally end in -o, and those taking una generally end in -a. Children make this binary division clear in errors such as *un papelo for un papel, and *una flora for una flor. In Hebrew, the feminine plural is generally suffixed to nouns ending in stressed -a and unstressed -et and -at, while other nouns take the masculine plural. Children often impose a simple division, suffixing the feminine plural to all nouns that end in -a (or sometimes all nouns ending in a vowel), using the masculine plural for all 515

other nouns. These solutions show that children can deal with a formal, nonsemantic basis of allomorph selection, using a highly salient division of words on a single criterion, such as the final sound of the word. (The Romance and Hebrew divisions are binary and the Polish is ternary; we have no evidence of systematic divisions of paradigms into more than three categories. Acquisition of systems with more divisions such as Bantu noun categories, would be informative in this regard.) § Levy (1983) has recently carried out a detailed examination of the acquisition of semantic and grammatical uses of gender morphology in English, German, French, Polish, Russian, and Hebrew, concluding that formal criteria are more accessible than semantic criteria in this domain: Phonological properties that determine inflectional patterns are mastered to an impressive degree during the early stages, that is, between 2 and 3 years of age. The children work out the formal-distributional patterns and do especially well when those are systematic (e.g. Polish, Hebrew)…. On the other hand, children aged 2–3 seem not to take advantage of whatever understanding they possess regarding cognitive gender distinctions in working out the intricacies of the linguistic system. Their inability to see the linguistic applicability of gender notions is probably another aspect of the lack of cognitive clarity and cognitive salience of gender at this young age.

516

(c) Categories Established on Co-occurrence Grounds. Some problems of paradigm construction cannot be solved by reference to either the meaning or the phonology of the citation form.23 In some instances, citation forms can be categorized on the basis of their regular co-occurrence with other forms. A prime example is the German case inflectional system, which is marked on articles, rather than on nouns. Most nouns do not have predictable gender on phonological grounds, but gender is always clearly given by an accompanying definite article, der ‘MASC’ / das ‘NEUT’ / die ‘FEM’, and other gender cues given by indefinite articles and adjectives. Similarly, the gender of many French nouns can be determined only by the form of the article. The evidence is that such cues are less salient that word-endings. The German case-inflectional system is acquired considerably later than Slavic systems, and French-speaking children use phonology as a more reliable gender cue than the form of the article. It would seem that it is easier to access the citation form itself than to access a co-occurring morpheme such as an article, in order to determine the class membership of a lexical item. Accordingly, the third option of OP: MORPHOLOGICAL PARADIGMS takes account of processing constraints: OP:MORPHOLOGICAL PARADIGMS (c). If you do not succeed in setting up a paradigm based on the phonology of the citation form, and if your procedural capacities allow you to check the immediate environments of citation forms, try to differentiate the functors on the basis of elements that systematically co-occur with the citation forms, and 517

set up a paradigm in which choice of functor is conditioned by factors that regularly co-occur with the citation form. (d) Failure to Establish Paradigm. Finally, as an interim default solution, we find various sorts of “inflectional imperialism” and one-to-one mapping without adjustment on formal grounds. Note that the default for this OP is not simple omission, since the OP only goes into effect at the point at which the child already has established at least one functor to express a given Notion. This functor is presumably the most salient on grounds of perceptibility and applicability, and should predominate in speech until OP:PERSISTENCE, in reapplying OP:MORPHOLOGICAL PARADIGMS, results in a new solution (in conjunction with requisite developments elsewhere in the overall mental system). OP:MORPHOLOGICAL PARADIGMS (d). If you fail, use only the most salient and applicable functor to express the given Notion in the given position. Examples of option (d) are widespread and familiar. They are interesting in that they reveal particular Notions that are especially salient for grammaticization. We find early unique marking of Notions such as: § NUMBER: one plural suffix for all nouns, regardless of gender, in Hebrew and Russian. § CASE: one suffix for each case in Russian, regardless of gender (e.g. universal use of FEM:ACC -u, MASC/ NEUT:INSTR -om); one accusative suffix in Polish and 518

Russian, regardless of animacy; no distinct case marking in Polish for “virile” nouns (masculine plural). § DEFINITENESS: one article in German for all three genders (e.g. de). § TENSE: in languages with gender-conditioned tense forms (Polish, Russian, Hebrew), expression of appropriate tense with one form only, ignoring sex of speaker or addressee. In Turkish, use of one past-tense form regardless of source of evidence.

Clause-Level Units Connectives. As a consequence of OP:PHRASES and OP:CLAUSES, sequences of classes of morphemes are stored. The existence of higher-order units allows for the discovery of connectives. Once the grammar is organized on the clausal level, and inter-propositional relations are semantically conceptualized, an OP for mapping comes into play to interpret hitherto unanalyzed clause-level functors. This OP—parallel to OP:FUNCTORS on the word level—attempts to interpret salient nonreferential speech segments in relation to clauses rather than content words. We must assume that LMC has means for relating clauses to one another as conjoined or subordinated in various ways. With development, Semantic Space is organized in terms of temporal and causal sequence, coordination, simultaneity, conditionality, and so forth. At this point, connectives can stand out as figures against the ground of relatable propositions:

519

OP (MAPPING): CONNECTIVES. Once you have established means of linguistic expression for whole propositions (clauses), if an uninterpreted functor cannot be mapped onto an accessible Notion, and it occurs in an utterance with two clauses, try to assign it a function that relates the two clauses. If you succeed, store the functor and a definition of its interclausal function and placement. Canonical Clause Forms. It often happens, however, that conjoined or subordinated clauses lose some of the distinctive form that they have as free-standing, main clauses. Just as LMC prefers to keep word forms constant across environments, a similar OP is directed at maintaining invariant clause forms. OP (UNITS): CANONICAL CLAUSE FORM. If a clause has to be reduced, rearranged, or otherwise deformed when not functioning as a canonical main clause, attempt to use or approximate the full or canonical form of the clause. There are many examples in our data of the operation of this principle. § Where clause boundaries are clearly marked by particles, and the internal form of clauses is not seriously altered, multiclausal structures are easily acquired—as, for example, various types of Indo-European relative clauses and complement clauses. Thus, English-speaking 2-year olds easily produce forms such as I have a dog THAT went away and Let me show you WHERE Peter lives. Polish 2-year-olds produce relative clauses with 520

the undeclinable connective co, as in: A gdzie te nacyńka CO je wyjmowałem? ‘And where are those dishes THAT I was washing them?’ Smoczyńska also reports very early conditionals, with clear marking of both clauses, as in the following example at 2;0: JAK ty byś chciała, TO bym ci dała ‘IF you wanted, THEN I would give you’. § By contrast, Aksu-Koç and Slobin report late acquisition of relative clauses, complement clauses, and some kinds of conjunctions in Turkish. They argue that children have difficulty with clauses that are expressed as deverbal nominalizations, losing the finite verb and normal case inflections of canonical main clauses. Turkish children prefer separate clauses for causal conjunction and modification, keeping close to canonical forms for each clause. Connectives in the form of verbal suffixes (“converbs”) are learned earlier than those mat convert verbs into nouns—presumably because a clause with an identifiable verb is closer to the canonical form than is a clause with a nominalized verb that takes noun suffixes. Other sorts of evidence also suggest that children try to keep embedded clauses as much like main clauses as possible. The data repeatedly show attempts to express all of the nominal arguments in embedded clauses, even though the parental language allows for or requires deletion. As in the discussions of OP:MAXIMAL SUBSTANCE and OP:EXTENSION (below), overt marking of all sentence participants is an early and persistent characteristic of child language.

521

§ In Polish, noncoreferential complement clauses are embedded with the complementizer żeby and a finite verb, while coeferential clauses are embedded without żeby and with the infinitive. Children make errors of extending the former pattern, marking the embedded clause explicitly in the first person, as in: *Ja chce, żeby miałem kotka ‘I want that I:have cat’. § In Hebrew, children of late preschool age consistently tend to copy the subject pronoun in relative clauses, though this is not the norm: ha yeled še HU nafal ba mayim ‘the boy that HE fell in the water’; ha iša še HI ra’ ata et ha naxaš ‘the woman that SHE saw OBJ the snake’. § Children also try to maintain normal casemarking on sentences when they function as embedded clauses. In some Turkish relative clauses and complement constructions, for example, the subject of the embedded clause appears in the genitive case, as “symbolic possessor” of the action attributed to it by a nominalized verb. A frequent error is to retain the nominative form of such nouns in embedded clauses, reflecting their underlying subject role. (This is also a consequence of OP: FUNCTOR CLASSES, which instructs LMC to “try to map each functor onto a distinct Notion.” In this instance, nominative and genitive cases maintain their separate functions to encode subject and possessor.) *

*

*

We have now equipped the child with procedures for sorting and classifying lexical and grammatical 522

morphemes, and maintaining canonical clause forms. In the course of development, the meanings of linguistic elements are refined and rules are elaborated for their placement in sentences. We turn next to OPs for refinement of form-function mapping.

Organizing Information in Storage: Form-function Mapping As LMC groups information in storage in more ways—establishing units, classes of units, paradigms, and clause forms—form-function mappings are reviewed and systematized following an important set of OPs that adjust the grammatical means of expressing Notions. The child tries to mark Notions consistently, in form and position, reconstructing forms that are discovered to be plurifunctional and analyzing forms that conflate distinct Notions.

Overextension OP:LIMITED FUNCTIONS instructs LMC to be conservative, extending formal means of expression only so far as the current level of analysis allows: “At first apply a solution to the smallest motivated category and do not extend it without evidence.” As evidence is gathered, however, conservatism is replaced by a drive towards inclusiveness and generality. LMC notes that a given Notion is marked by a particular functor or by a “configuration” of elements (combination of morphemes), and extends its solution to other instances. Where the extension matches the input language, acquisition is appropriate, and is generally not singled

523

out as a phenomenon in the literature. But frequently the child’s solution is more general than the input would warrant, and “overextensions” provide important evidence of principles that guide grammar construction. Thus although the following OP is phrased in terms of extension, all of the relevant data are, in fact, instances of OVERextension. OP (MAPPING): EXTENSION. If you have discovered the linguistic means to mark a Notion in relation to a word class or configuration, try to mark the Notion on every member of the word class or every instance of the configuration, and try to use the same linguistic means to mark the Notion. Replacement of Zero Forms. OP:EXTENSION, builds upon, but goes beyond the 1973 Universal E2: “There is a preference not to mark a category by zero. If a category is sometimes marked by zero and sometimes by some overt phonological form, the latter will, at some stage, also replace the zero” (Slobin, 1973, p. 202). MacWhinney (1985) proposes a similar consequence of his Operating Principles for “Amalgam Acquisition” and “Amalgam Analysis”: “Children will overgeneralize non-zero allomorphs even when zero allomorphs are of much greater applicability” (Consequence spellout 12f). Both of these formulations apply to a situation in which the child has already acquired some formal marking of a Notion (an “overt phonological form” or a “non-zero allomorph”), and the input language has gaps in which the Notion is sometimes not formally marked. The OP thus operates only in relation to the child’s ongoing 524

paradigm construction. (Note that children do not invent means for marking citation forms, such as nominative forms of nouns. Such forms are not formally marked in the languages of our sample, and thus do not constitute instances in which a non-zero form can be extended.) § In Polish the genitive case in the plural is marked by zero for feminine and neuter nouns, but by a suffix, -ów, for masculine nouns. Smoczyńska reports a persistent error of use of -ów to mark plural genitive in all three cases. Recall that Polish children quickly master the overall case-gender paradigm, using distinct gender suffixes on nouns for each grammatical case. Following OP:STRENGTHENING, they expect the genitive to be marked by a suffix in the plural as well; finding a gap, they overextend the only available suffix, the masculine-ów. It is important to note LMC’s attention to the paradigm as a SYSTEM, with the expectation that each cell be filled in a consistent manner. § MacWhinney reports a similar phenomenon in Hungarian, showing that an infrequent morpheme can be overextended if the alternative is zero-marking. Most verbs in the language are unmarked in the third person singular indefinite. However, a small group of about ten frequent verbs, and several others, have an overt marking, -ik. Although this form is of limited applicability, it is widely overextended to other verbs. § An interesting example from Hebrew acquisition was presented earlier, in regard to OP:STRENGTHENING. Some verbs are not marked for present tense, and have a 525

type of root that does not permit the normal triconsonantal frame needed for the usual vowel alternation used to mark this tense. Children overextend a present tense prefix, m-, from another class of verbs, thus following OP:EXTEN-SION in trying to mark present tense on every verb. But this OP is held in check by a general sense of the centrality of vowel alternation as a morphological pattern in Hebrew, built up by OP:STRENTHENING, and children do not “try to use the same linguistic means to mark the Notion.” In this instance, prefixing is used to replace zero-marking, but is not overextended to verbs that can be marked by vowel alternation. A Notion must be frequently enough marked in a paradigm to attract children’s attention to the possibility of formal marking of the Notion. It is probably also the case that the higher a type of grammatical marking is on a “universal hierarchy,” the less evidence is needed to trigger its application in development. In addition, OP:EXTENSION is also constrained by the probability and relevance of marking a particular Notion in regard to a particular part of speech. § In English, third-person marking is limited to the final -s in present-tense singular, and is not extended to third-person singular in other tenses or to third-person plural. Person-marking is rare in English verb paradigms; additionally, third person tends to be unmarked in other languages. Both of these factors could play a role in constraining the application of OP:EXTENSION.

526

§ In German, the accusative remains unmarked for a considerable period in children’s acquisition of articles. Accusative is a salient notion, and is marked earlier on personal pronouns, where unique, and unanalyzed forms are available for most persons. But in the paradigms of case-inflected articles, there is a unique accusative form only for the masculine singular. Nominative (citation) and accusative forms are identical for neuter, feminine, and plural forms of articles, demonstratives, and possessives. In this situation, the operation of OP:EXTENSION is probably delayed by the difficulty in discovering the one available nonzero allomorph—masculine singular accusative. (When overgeneralizations DO occur, this form is used.) Extension in Violation of Formal Constraints. Even though German-speaking children acquire accusative forms of pronouns, these are separate morphemes, and there is no way to extend casemarking from pronouns to articles (e.g. pronouns: ich/mich ‘I/me’, er/ihn ‘he/him’, wir/uns ‘we/us’; nominative/accusative articles: der/den MASC, das/das NEUT, die/die FEM and PLURAL).However, extensions CAN go beyond form-class constraints if the grammatical marker can be analyzed and freely combined. Such overextensions reveal the child’s sense of functions that should receive grammatical marking in the language. § In Japanese, nouns and nominal adjectives are negated with ja nai, adjectives with -ku nai, and verbs with -nai. In the first stages of acquiring negatives, children use the single form nai to negate these various parts of speech. Thus the central grammaticizable Notion is negation, and 527

the form of the negative element is not controlled by the form class of the negated item. Similarly, the genitive particle no is used only in constructions where one noun modifies another (N no N), but children extend this marker to other sorts of modification as well, making errors of *ADJ no N and *REL.CL no N. Again, a general function—modification—is marked in similar fashion across constructions, ignoring form-class constraints. § The general function of modification is also shown in extensions of a bound morpheme—the German adjective ending. Mills notes extensions to other sorts of prenominal modifiers, including numerals (Gibt mir die *zweien (=zwei) äpfel ‘give me the twoADJ apples’) and the quantifier all ‘all’ (die *allen männer (=all die männer) ‘all the men’). She notes: “Both numbers and all appear in certain contexts to behave like adjectives, so that this regularization is quite likely as well as indicating the strong tendency to establish patterns in the system.” Children often ignore the larger syntactic context in which a Notion appears, preferring to use the same local marking across constructions. § In Kaluli the ergative marker only appears on subjects in the focused OSV order, and not in the neutral SOV. However, children mark the subject as ergative in both orders, following OP:EXTENSION to mark a Notion in every instance.

528

§ In Hebrew, children often use the direct object particle, et, on postverbal patients in dative and passive constructions, even though such nouns should function grammatically as subjects. Berman suggests that children overgeneralize the rule of marking the definite direct object, attending to the semantic role of the noun as patient rather than its surface syntactic role. She reports errors such as: nafal lo et ha mitriya ‘fell:MASC to:him OBJ the umbrella:FEM’ for nafla lo ha mitriya ‘fell:FEM to:him the umbrella:FEM’ (= ‘his umbrella fell / he dropped his umbrella’), and ba sefer lo mesupar et ze ‘in:the book not tell:PASS OBJ that’ (‘=in the book, that’s not mentioned’). § In Japanese, the direct object is generally marked by the particle o, but objects of various transitive stative verbs and adjectives are marked with ga, which usually is used for subjects. Clancy reports errors such as the use of o, rather than ga, to mark objects of verbs of involuntary perception such as mieru ‘appear/be visible’. As she notes: “In other languages, these nominals are treated like direct objects, and sometimes Japanese children do so as well.” Like the Hebrew extensions of et, a direct object particle is extended to marking of direct objects or patients in general, ignoring particular structural constraints. Semantic Constraints on Extension. OP:EXTENSION is phrased in terms of marking of a “Notion”—our general term for nodes in Semantic Space. Some sorts of extensions provide a picture of the child’s definition of the range of a Notion.

529

§ In German, the possessive suffix -s is limited to proper nouns, and first emerges in this function (e.g. in papa-s kragen ‘in papa’s collar’). Children extend it to nouns in general, making errors with both animate and inanimate possessors (e.g. *männer-s wagen ‘men’s car’ and *eisenbahn-s wohnung ‘train’s house’). Like errors reported above, a general notion—possession—is extended. In this instance, the extension is not only syntactic (marking of all nouns rather than only proper nouns), but semantic (ignoring animacy as a constraint on the form of possessive marking). These errors may be due to a general low level of animacy as a relevant criterion for grammatical marking, as discussed above in regard to Semantic Space and functors. § We have noted above that third-person marking is not extended in English from the present-tense singular to other third persons. Important factors seemed to be the unmarked character of the third person, along with the rarity of person-marking in English. In languages like Portuguese, however, person-marking is much more widespread in verb paradigms, and person-marking is extended across tenses. Bybee (1979), analyzing acquisition data of Simões and Stoel-Gammon (1979), discusses an interesting example of distinguishing first and third persons. The child had acquired distinct first and third singular inflections in the present and preterite tenses; these persons do not have distinct forms in the imperfect, but the child extended the first singular marker -o from the present tense to imperfect verbs, attaching it to the third-singular stem of the imperfective (e.g. *queril-o/ *queril-a ‘wanted-lSG/wanted-3SG’ for queria). Note that person-markers are applied ACROSS 530

tenses, while stems are kept constant WITHIN tenses in this child’s extension. These data suggest that extensions follow relations of semantic relevance: in this instance, tense is relevant to the verb itself (as indicated by the verb-stem), while person is not relevant to tense, and is generalized from one tense to another. Extension of Configurational Expression. In addition to extension of the function of grammatical morphemes, the function of a configuration of morphemes can also be extended. The best example is the widespread use of the sentence frame to indicate the valence of the verb, rather than to mark the verb itself as being intransitive, transitive, or causative. The most familiar examples come from the data of Bowerman (e.g. 1974, 1982) and Lord (1979), in which English-speaking children replace lexicalized forms with analytic constructions (e.g., I’m gonna fall (=drop) this on her; It can hear (=be heard) now; so it can’t put (=get) on my nose). In English there is a model for such uses, since the language provides valence-neutral verbs like open (The door opened / I opened the door). However, the use of a verb-plus-participants frame to define valence is a widespread abstraction, and is even reported in languages where there is no potential model in the input, such as Hungarian and Hebrew. For example, Hebrew has regular means of forming causative verbs, but Berman reports child forms such as ima *oxelet oti ‘mother eats me’ in place of ma’axila ‘feeds’, formed from the same root, ?-x-l). Use of a valence-neutral verb in sentence-frames with varying number of participants is reported for child speech in English, French, Polish, Portuguese, Hebrew, Hungarian, and Turkish. It reflects 531

an extension of a very basic means of expressing propositional Notions. (In addition, where such forms replace inflected or lexical expressions of valence, the extensions are co-determined by OP:ANALYTIC FORM, as discussed below.)

Affix-Checking There is an interesting constraint on OP:EXTENSION that sometimes appears in child language and in historical language change. We have reviewed general tendencies to ADD affixes to stems or to CHANGE stems to map the meanings of functors onto content words. But in a subset of instances, the word may look as if it has already been grammatically marked, and additional marking will not be added. For example, Bybee and Slobin (1982) and others (e.g. Kuczaj, 1978; Slobin, 1971b) have noted that English-speaking children, when overregularizing the past tense, tend NOT to add an -ed to verbs that already end in a dental, such as hurt, put, and hit. That is to say, *hurted, *putted, and *hitted are rare errors, in comparison with errors such as *growed and *runned. In addition, it is easier for children to learn irregulars with vowel change and final dental, such as slept, than verbs with vowel change alone, such as saw. Bybee and Slobin propose that children create a SCHEMA of the past tense as ending in t or d, and that they orient to the form of a verb in relation to this schema, rather than add a suffix to modify its meaning. § Similar examples come from late acquisition of the syllabic plural /-iz/ in English, which is added to stems that already sound like plurals. German-speaking 532

children sometimes omit the adjective ending -er on the word ander ‘other’, avoiding ander-er—presumably because ander appears to already have the adjective ending. They also tend to omit plural suffixes on words that already appear to end in plurals (-er, -e, -s), using words such as hammer ‘hammer’, pfeife ‘pipe’, and glas ‘glass’ as plurals (MacWhinney, 1978). The Hungarian accusative -ek is often omitted from words whose stems end in -k (Réger, 1975, cited in Menn & MacWhinney, 1983). In general, the consequence of matching the overall form of a word to a schema is that affixation or internal change is not automatic. Rather, it “competes” with other tendencies. Menn and MacWhinney (1983) have discussed this particular constraint, “the repeated morph constraint,” at length, noting that the languages of the world—like language learners—frequently avoid repetition of phoneme strings across morphs. They propose a procedure of “affix-checking,” which “looks to see if the marker in question was already present when the form was retrieved from the lexicon. If the checking process finds the marker ‘already there’, the rule is blocked and the form is allowed to pass unmodified.” This procedure is expressed in the following OP for mapping, which constitutes a general early constraint on the form of affixing and paradigm construction: OP (MAPPING): AFFIX-CHECKING. Do not add an affix to a word or word-stem that appears to contain that affix in the relevant position.

533

Eventually this OP is overriden, as affixation becomes a more and more applicable solution, in accord with OP:STRENGTHENING. Later in development, affix-checking seems to be countermanded, resulting in overmarked forms like *duckses and *breakded, and the emergence of forms like *hitted. Similar overmarking is reported for other languages, and, as discussed above, is also a consequence of OP:MAXIMAL SUBSTANCE. Affix-checking is probably countermanded during the period in which the child is consolidating a period of active work on on the affixal marking of a Notion.

Unifunctionality During early periods of marking and extending the marking of individual Notions, children seem to deal with each form-function mapping as a fairly separate problem. With time, however, LMC carries out higher-level systematization, comparing the means of marking particular Notions and comparing the functions carried out by particular linguistic forms. As a consequence of OP:SEMANTIC REORGANIZATION, clusters of Notions in Semantic Space are analyzed; and as a consequence of OP:DICTIONARY ORGANIZATION, entries are enriched and systematized in storage. The resulting organization is guided by OPs for clarity of mapping, with the consequence that, for a time, children avoid pluri-functional morphemes (OP:UNIFUNCTIONALITY) and synthetic forms (OP:ANALYTIC FORM).

534

Annette Karmiloff-Smith (1979) has devoted detailed attention to this issue, and has argued “that language development involves passing gradually from a series of juxtaposed unifunctional markers and processing procedures, to the intralinguistic organization of plurifunctional systems of options for modulating meaning” (p. 19). In her analysis, children begin by treating morphemes as “a series of unifunctional homonyms,” unaware that the same word has several different meanings. In a second phase, children realize such separate meanings, and attempt to mark them distinctly, using redundant, overanalyzed, or semi-grammatical forms. For example, French-speaking children distinguish the uses of un/une ‘a/one’ as indefinite article and numeral by differentiating une voiture ‘a car’ from une de voiture ‘one car’. Similarly, the plurifunctionality of même ‘same’ is resolved into two expressions, la même X to mean ‘the same one’ and la même de X to mean ‘same kind’. Morphemes that conflate several meanings are analyzed into their constituents: the first-person plural possessive mes is replaced by the analytic toutes les miennes de ‘all:PL the:PL my:PL of, thus separately marking totality, plurality, and possession by unifunctional words. We can summarize Karmiloff-Smith’s rich findings in an OP that guides LMC toward unifunctional terms at the point in development when plurifunctionality is realized: OP (MAPPING): UNIFUNCTIONALITY. If you discover that a linguistic form expresses two closely related but distinguishable Notions, use available 535

means in your language to distinctly mark the two Notions. Note that this OP goes into effect only for Notions that are “closely related but distinguishable,” and at the point when intralinguistic systematization reveals that they are expressed in the same manner. Thus many early instances of overextension are not affected by OP:UNIFUNCTIONALITY, since they precede the requisite analysis. For example, the German-speaking child who uses the same possessive form for animates and inanimates is not using a plurifunctional form, because these Notions are not yet distinguishable; rather, they are only closely related, which makes them amenable to OP:EXTENSION, but not OP:UNIFUNCTIONALITY. Again, one-to-one mapping depends on the child’s definitions of “one form” and “one meaning.” In the preceding section, LINGUISTIC UNITS, we have seen various ways in which LMC interprets “one form.” At this point, we encounter various ways in which LMC interprets “one meaning.” At both of Karmiloff-Smith’s levels—early unifunctional homonyms and later explicit marking of differentiated meanings—plurifunctionality is absent from the child’s point of view. As discussed in regard to OP:MAXIMAL SUBSTANCE, one cannot decide whether these phases represent deep-seated constraints on the form of early grammars, or whether they reflect “metaprocedural” attempts to mark distinctions clearly at the point at which they may be prone to confusion. In the final phase (which is quite late in Karmiloff-Smith’s study: age 8–12) children accept 536

plurifunctional morphemes, and the examples of redundant marking and overly explicit, ungrammatical forms fade away. In her terms (1979, p. 226): One marker simultaneously carries the burden of indicating several functions and children are thus economical in their utterances…. In general, the over 8 year old seems concerned to avoid ‘overdetermining’, i.e. he makes a clear distinction between what must be stated and what can be left unsaid with respect to clarity of referential information, an important Gricean concept. This conclusion is in accord with the discussion of children’s under-use of ellipsis when following OP:MAXIMAL SUBSTANCE, and it will appear again in regard to OP:ANALYTIC FORM, below. It is apparent that the means of grammatical expression are organized in consort with the child’s conception of discourse, along with growing automatization of the processes of accessing and producing linguistic forms in ongoing speech. § Several other examples of OP:UNIFUNCTIONALITY appear in our data. Interestingly, they all come from French, a language with an exceptionally high degree of homonymity and plurifunctionality. Clark reports an idiosyncratic case of a child making novel use of a morphological device—contraction—to distinguish between possessive and partitive uses of the preposition de. This child used contracted de + ART for partitive and uncontracted de + ART for possession: PARTITIVE—J’ai du pain ‘I have some bread’ and Y’a 537

*da neige ‘There is some snow’ versus POSSESSIVE—le chapeau *de le monsieur ‘the man’s hat’ and le Soulier de la dame ‘the woman’s shoe’. § It seems to be a general phenomenon that when a new form enters, it segregates a semantic field and takes over part of it. Two stages in the development of French connectives point to the difficulty of plurifunctionality for the child. At first, que ‘that’ is used both as a complementizer and a relativizer, When qui ‘who’ emerges, it takes over as the sole relativizer, leaving que to mark complement clauses. Note that this solution (not the adult solution) limits the plurifunctionality of both que and qui, assigning a distinct syntactic function to each.

Analytic Form Avoidance of plurifunctionality for purposes of clarity goes hand-in-hand with avoidance of synthetic forms in favor of more analytic expressions. We have already encountered examples of preferences for analyticity in regard to configurational definitions of verb valence and analysis of the French possessive. There are numerous examples of analysis and overanalysis in our data, supporting another OP for transparency of mapping: OP (MAPPING): ANALYTIC FORM. If you discover that a complex Notion can be expressed by a single, unitary form (synthetic expression) or by a combination of several separate forms (analytic expression), prefer the analytic expression.

538

This OP can only go into operation when the child realizes that a given Notion is complex—that is, at the point when some internal semantic analysis has been carried out. (This point has been repeatedly made by Bowerman in regard to her many examples of analyzed forms; e.g. 1974, 1981, 1982.) The applicability of OP: ANALYTIC FORM also depends on the analytic options made available by the parental language, though, as we have seen, children can forge analytic means beyond those provided by the language. A number of examples of children’s analytic forms are presented below. Each one reveals important dimensions of children’s definitions of components of meaning. § POSSESSION. There seems to be a tendency to separate the fact of possession from other characteristics of the possessor and possessed. Clark reports a common tendency in the Romance languages to use analytic PREPOSITION + PRONOUN constructions (e.g. French de moi ‘of me’) rather than forms which conflate the fact of possession with person of possessor and gender and number of possessed (e.g. Mon/ma/mes ‘my:MASC/ FEM/PL’). She notes: “The prepositional constructions for possession appear to be more analytic than the possessive adjectives like mon in that they mark explicitly the relation between object-possessed and possession.” In German, Mills reports a preference for analytic possessives for all nouns, proper and common, animate and inanimate, using von ‘from/of or *zu ‘to’. § VERBS OF MOTION AND DIRECTION. Of special interest in light of Talmy’s (1985) crosslinguistic study of verbs of motion is the tendency in Spanish, Italian, and 539

Hebrew for children to reanalyze verbs which conflate motion and direction into separate expression of these two notions, while at the same time conflating motion and manner. For example, whereas in English we say run down, the normal Spanish equivalent is bajar corriendo ‘descend runningly’. Spanish-speaking children, however, often produce expressions similar to the English analysis, such as *correr abajo ‘run down’. Perhaps there is some iconic basis (Slobin, 1983) for a relative semantic proximity of notions of movement and manner, allowing for conflations such as run, walk, crawl, swim, fly with direction of movement as a sort of general operator on such verbs (e.g. up, down, along, through). Note that manner of movement cannot be separated from the fact of movement, while directionality is an independent dimension. Children’s analyses of such expressions indicate their use of semantic Notions such as these. § NEGATION, TENSE, AND EXISTENCE. We will have more to say about the placement of negative morphemes in regard to OP:OPERATORS, which engenders clause-external placement of negation. Concomitantly, it is evident that conflation of negation with particular other Notions is dispreferred. Japanese children make errors of using nai, the nonpast form of negation, across tenses. This may reflect a sense that a single morpheme should not conflate two Notions as distinct as negation and tense. Use of nai with verbs inflected for the past may represent an analytic option of marking tense and negation separately.

540

§ Hungarian children also show a preference for separate marking of the negative, where the language allows for portmanteau expressions with the copula and the ‘additive’. MacWhinney reports: “The portmanteau negatives nincs(en), sent, sincs(en), and sincsenek express the same semantic content as the expressive analytic forms nem van ‘not is’, *nem is ‘not also’, *nem is van ‘not also is’, and *nem is vannak ‘not also are’. Errors involve use of the analytic forms instead of the portmanteau form or use of both at the same time.” These errors also suggest that existence and negation are separate concepts for the child. In Japanese, where the existential has distinct affirmative and negative forms (aru ‘there is’ and nai ‘there is not’), children sometimes negate the affirmative form (*aru-nai for nai ‘there is none’). Children’s errors of overanalysis in the face of rote-learnable fused forms point to the exceptional strength of OP:ANALYTIC FORMS at particular stages in acquisition. Several examples discussed above reflect children’s analysis of forms which are available as rote forms. Analysis is apparently carried out when (or whenever?) an analytic solution is available. For example, there is no way that English- or German-speaking children could analyze casemarked pronouns, since there is no pattern that would allow for segmentation and recombination (e.g. I/me, he/him, she/ her, we/us, they/them, and the comparable more complex German system). In Hebrew, however, oblique pronouns fuse in unpredictable ways following prepositions, but the language provides separate prepositions, pronouns, and preposition + noun combinations. Thus children can 541

replace rote-learnable fused forms such as alav ‘on him’ and bišvilenu ‘for us’ with available separate forms *al hu ‘on he’ and *bišvil anaxnu ‘for we’.

Organizing Information in Storage: Position of Elements Thus far we have focused primarily on issues of unit formation and semantic mapping. LMC also needs OPs for the solution of morphosyntactic problems. Stored linguistic material is continually scanned for the relative positioning of elements. OPs for position of elements regularize placement of linguistic units, and are responsible for early syntactic preferences on the levels of word, phrase, and clause. On the most basic level, morphemes will be kept in order in words, phrases, and clauses. Beyond that, sequences of grammatical morphemes will be ordered in terms of semantic relevance, and operators will tend to be placed according to their scope of operation.

Intraword Morpheme Order Our current data strongly support the 1973 Universal C1: “The standard order of functor morphemes in the input language is preserved in child speech.” This tendency is especially pronounced with regard to the bound morphemes that constitute words, and can be rephrased as the following OP for position of elements: OP (POSITION): INTRAWORD MORPHEME ORDER. Keep the order of morphemes in a word

542

constant across the various environments in which that word can occur. § In agglutinative languages like Hungarian and Turkish, errors in ordering of affixes are almost nonexistent from the very beginning of acquisition. Menn and MacWhinney (1983) propose a general “position-checking mechanism.” They point out, following Maratsos and Kuczaj (1978), that semantically plausible combinations do not occur if they violate the “positional frames” of affixes. For example, the English form *breakinged would be a plausible past progressive (e.g. *He was breakinged it), but such errors have never been reported. Nor are they reported for agglutinative languages, where strings of syllabic morphemes may facilitate misordering. For example, Turkish children always say el-ler-im ‘hand-PL-1SG:P0SS’ (=‘my hands’), and never *el-im-ler ‘hand-lSG:POSS-PL’. (OP:RELEVANCE may also play a role here, as discussed below.) § The rhythmic-intonational contour of the word as a whole entity is crucial to OP:INTRAWORD MORPHEME ORDER. In French and Polish, for example, children have difficulty in placing clitics, which appear in varying orders, and are less bound to particular words, whereas words with affixes seem to form perceptual units.

Phrasal Morpheme Order Fixed order of elements within phrases also seems to be accessible to LMC, though we have less evidence in this

543

regard. Few errors are reported crosslinguistically in the order of elements in nounphrases and verbphrases, suggesting the following OP: OP (POSITION): PHRASAL MORPHEME ORDER. Keep the order of morphemes in a phrase constant across the various environments in which that phrase can occur. § The de Villiers report that English errors of auxiliary order “seem surprisingly rare.” Possible non-occurring errors are double auxiliaries (*must will go), wrong affixation (*he have going, *he will eaten it), tense affixation on main verb (*he does ate it), and misordering (*he have been might going there). Thus ordering seems as accessible in the English auxiliary system as ordering of inflectional morphemes in other languages. Perhaps the ordering of functors is generally easy (while positioning them in relation to other elements may be difficult, as discussed below).

Word Order We have already seen that OP:PHRASES and OP:CLAUSES result in the storage of sequences of classes of morphemes, along with designations of phrase and clause types, and that clauses are treated as units following OP:CANONICAL CLAUSE FORM. This is not the place to lay out a taxonomy of clause types. In the absence of any detailed crosslinguistic application of a grammatical model to acquisition, we can only assume that a universally applicable model exists. The terms of such a model, whatever its ultimate form, will classify

544

clauses according to the number and type of nominal participants in relation to particular types of predicates, and will provide definitions of phrases that can function as constituents in syntactic structures. We must assume that LMC has procedures for determining the realizations of phrase and clause types in the parental language. (See Pinker, 1984; Wexler & Culicover, 1980; and papers in Baker & McCarthy, 1981, and Bresnan, 1983 for examples of early attempts to apply current syntactic models to theories of acquisition.) Whatever the appropriate designation of sequences may be, there must be OPs for mapping which assign meanings to sequences, and OPs for position which order elements in clauses. The languages of the world differ in the functions played by word order. Crosslinguistic developmental evidence suggests that children are quick to realize that variations in word order can serve pragmatic functions of deployment of attention in utterances (perspective-taking, focus on new information, backgrounding of given information, etc.), while they seem to be reluctant to allow changes in word order to signal changes in meaning or to be based on purely syntactic constraints with no communicative function. Early on (e.g. Slobin & Bever, 1982), children determine that basic grammatical relations are marked in the parental language either by fixed word order (as in English) or by case inflections (as in Slavic languages and Turkish). Such information is noted by OP:CLAUSES, which stores pairings of proposition types and ordered sequences of word and functor classes. Given a basic grasp of clause types and their syntactic 545

expression in the language, LMC is guided by two major OPs for word order, OP:VARIABLE WORD ORDER and OP:FIXED WORD ORDER. These OPs interact in complex fashion, in response to the functions of word order in the parental languages, and in interaction with other OPs for the position of elements presented in this section. Variable Word Order. Whenever OP: CLAUSES results in storage of more than one word-order pattern for a given clause type, LMC seeks a basis for such variation: OP (MAPPING): VARIABLE WORD ORDER. If you find that a clause type occurs in more than one word order, attempt to find a distinct function for each order. Candidate functions for variable word order appear to be arrayed in an accessibility hierarchy, with pragmatic functions highly accessible, and changes in sentence modality (e.g. interrogation) and mood (e.g. passive) less accessible. All of the languages in our sample that allow pragmatic variations in word order show a developmental tendency summarized by MacWhinney: In early combinations, children will tend to order the newest or most informative element first (Consequence ordering 4a). The converse of this principle is that highly presupposed elements will be postposed by children. Data are available on both initialization and postposing in early speech. 546

§ MacWhinney discusses this tendency in terms of verb initialization, noting many early Hungarian examples in which children initialize and stress the verb, even though the most frequent input orders are SOV and SVO, often beginning with an unstressed topic. He notes similar early verb initialization in French, Italian, German, and even English, as well as right-dislocation of given material in SOV languages like Japanese and Turkish. He suggests: “Presumably, in the situations confronting children, the objects are often highly given things like familiar playthings, common foods, or close family members. What is new and exciting are the activities in which these well-known objects engage.” § Clancy discusses similar phenomena in Japanese as examples of “the postposing of sentence constituents which are highly presupposed in the speech context.” She cites numerous examples from early child speech in which children postpose referents which have just been mentioned or which are present in the context. § Aksu-Koç and Slobin also note early control of the Turkish norm of placing new information before the verb and presupposed or predictable information after the verb, along with highlighting of the verb by verb initialization and postposing of all additional material. They note: “In reading through our own transcripts of Turkish child speech we have been struck by the extreme rarity of contextually inappropriate word orders, reinforcing the impression that pragmatic variation in word order is a precocious acquisition.”

547

§ Schieffelin reports that in Kaluli, children make pragmatically appropriate use of word order even before they use grammatical casemarking correctly. She notes that even in two-word utterances, children encode agent-verb when the agent is in focus, and object-verb when the object is in focus. In longer utterances, SOV is used for neutral reports and OSV is used to highlight the agent. § Although Smoczyńska does not assess Polish children’s competence in pragmatic word order variation, cursory examination of data from Polish, Russian, and Serbo-Croatian suggests that Slavic-speaking children, as well, make appropriate early use of varying word orders (though perhaps, in some instances, AFTER acquisition of case inflections; cf. Radulović, 1975; Slobin & Bever, 1982). In sharp contrast to the numerous examples of flexible use of word order reviewed above, English- and German-speaking children have difficulty in acquiring question inversion and patient-verb-agent orders in passive constructions. It would appear that such functions are low on an accessibility hierarchy in regard to word order variation. Data presented below suggest that, aside from pragmatic variation, children prefer to maintain canonical clause order (OP:FIXED WORD ORDER) and place clause-level operators outside the clause (OP:OPERATORS). Fixed Word Order. We have noted that children learn the roles of both variable and fixed word order in the input language. Unless there are clear morphological 548

indications of semantic relations, there is a strong tendency to rely on word order, especially for indication of agent and patient. This is obviously the case in “fixed word-order” languages like English, and may also be evidenced in early stages of the acquisition of languages with unclear or inconsistent inflectional morphology, such as the Slavic languages. An OP for fixed word order expresses this widespread tendency. It applies wherever OP:VARIABLE WORD ORDER cannot apply or fails to find motivation for alternate orders. OP (POSITION): FIXED WORD ORDER. If you have determined that word order expresses basic semantic relations in your language, keep the order of morphemes in a clause constant. § In languages with inflectional marking of the accusative, there is suggestive evidence that noncanonical position of object nouns may be avoided until they can be inflectionally marked. The direct object particle, o, is infrequent in Japanese speech, and is acquired late. Clancy reports an instance of adherence to standard OV order, with postposed objects appearing only after acquisition of o. In Serbo-Croatian, Radulović (1975) reports adherence to subject-object order before the acquisition of accusative inflections, and the use of fixed order even with inflections for a period of at least four months after the acquisition of the accusative. § In languages with fixed word order, like English, the use of variable word order by children is marginal. The

549

de Villiers report rare word-order errors in Stage I English speech. § Changes in word order that do not serve pragmatic functions pose great difficulty. French-speaking children try to keep their language a strict word-order language in those instances in which changes of word order are not tied to a clear communicative or semantic function. Standard word order is SVO with nouns as objects, while pronominal objects require SOV and use of a clitic pronoun. Clark reports errors of use of tonic pronouns instead of clitics and retaining canonical SVO order. She also reports use of standard VO order in both affirmative and negative imperatives, while the input language requires OV order for negatives. Pronoun order errors are frequent in double clitic ordering of direct and indirect objects, where children use a standard IO-DO order for indirect objects of all persons, avoiding the adult DO-IO order for third person indirect objects. Lacking inflectional cues for marking nouns as subject, direct object, or indirect object, children prefer a uniform positional definition. The interacting factors that determine positioning in French—nominal versus pronominal arguments, affirmative versus negative imperatives—are not interpreted as relevant to the expression of basic semantic relations in clauses, and, accordingly, fixed word order takes precedence.

Morpheme Placement OP:FIXED WORD ORDER is phrased in terms of the overall, ordered configuration of words in a clause. However, many problems of ordering can be conceived

550

of in terms of the position in a construction where a particular Notion is expressed. Children have particular difficulties with moveable elements, such as clitics, which don’t have a single fixed position. Many instances of fixed word order result from an OP directed towards standard placement of grammatical markers: OP (POSITION): MORPHEME PLACEMENT. Mark a Notion in the same place in the various constructions in which it can occur. If you discover that a particular class of words or functors occurs in different positions in different constructions, try to find a principled basis to differentiate the constructions. If you fail to define distinct construction types, use the same position across constructions. § One area of application of OP:MORPHEME PLACEMENT is the treatment of DISCONTINUOUS MORPHEMES. It is a widespread finding that the child, at first, is content to mark a Notion only once, and in only one place (the perceptually most salient). Thus the first form of the English progressive is the verbal inflection -ing with no preverbal auxiliary; the German past participle first appears without the preceding auxiliary or prefix; the first form of French negation is the pas of ne … pas; and so on. § If most items of a semantic class occur in a given position, this position will attract others, even if they don’t occur in that position in the parental language. In Hebrew, the numeral ‘one’ follows the noun, while other numerals precede the noun. Berman reports errors of 551

placing ‘one’ before nouns. MacWhinney reports that even a postposition can be drawn into a preposed location if it belongs to a class which is preposed. Hungarian children treat the conditional postposition volna like the modal prefixes, placing it before the verb. In Polish, children come to expect person-number to be marked on the verb, and have difficulty marking person-number on the conditional particle by and the conditional connective żeby. In Japanese, where the position of the negative marker varies with verb tense, children prefer a standard positioning of V+TNS+NEG rather than V+NEG+ ka+TNS in the past. (Some of these examples are co-determined by OP:RELEVANCE and OP:OPERATORS, as discussed below.) § Clitics that don’t have fixed position pose great acquisition problems in Polish and French, indicating that, at first, fixed position is preferred. We have noted problems in French of placement of full and clitic pronouns. Problems also arise in determining the environments for the different possessive forms (e.g. mon, à moi, le mien). Smoczyńska reports prolonged difficulty in Polish with the moveable 1PL enclitic -śmy, resulting in overanalysis, deletion, or overmarking. Overmarking is commonly a later development, representing a stage in which several positions have been learned, and competition between them has not been worked out (as influenced also by OP:MAXIMAL SUBSTANCE). § Data from German acquisition, reported by Mills, suggest that children can go through several stages of fixed position rules. At first children have a stable 552

verb-final rule, probably based on adult input with frequent final infinitives in imperatives and modal constructions. This verb position is quickly replaced by a verb-second rule, suggesting ease of positional learning and reorganization. Later, verb-final placement in subordinate clauses is easily acquired (perhaps continuing the earlier verb-final schema for main clauses). A general verb-final placement in subordinate clauses is probably responsible for late acquisition of positioning of the finite verb in subordinate clauses with two verbal elements, resulting in the error of placing the content verb at the end. At each point in the child’s reorganization, OP:MORPHEME PLACEMENT seems to work towards fixed placement.

Semantic Relevance in Morpheme Placement We have made use of Joan Bybee’s (1985,a,b) term, RELEVANCE, in discussing affinities of Notions in Semantic Space. In the section on “Relevance and the Location of Functors” evidence has been presented for children’s preferences to mark Notions and particular parts of speech. Relevance also plays a role in morpheme ordering, and children sometimes change the placement of functors in order to adhere to the relevance principle. This principle has a long history in linguistics and psycholinguistics, with many current formulations elaborating, in one way or another, “Behagel’s First Law,” which can be paraphrased as: WHAT

BELONGS TOGETHER MENTALLY IS PLACED CLOSE

TOGETHER SYNTACTICALLY.

(“Das oberste Gesetz ist

553

dieses, daß das geistig eng Zusammegehörige auch eng zusammengestellt wird” [Behagel, 1932, p. 4].) This “law” underlies data on relevance, as well as data on the placement of operators, reviewed in the next section. Givón (1985b), in a similar vein, has proposed a general PROXIMITY/SCOPE PRINCIPLE, combining a PROXIMITY PRINCIPLE and a SCOPE/RELEVANCE PRINCIPLE: PROXIMITY PRINCIPLE: The closer together are two concepts semantically or functionally, the more likely they are to be put adjacent to each other lexically, morphotactically or syntactically. SCOPE/RELEVANCE PRINCIPLE: The more RELEVANT the operator is to the operand, and the more SPECIFIC and EXCLUSIVE it is to the operand, the closer to the operand it will be placed. PROXIMITY/SCOPE PRINCIPLE: Items that are SEMANTICALLY CLOSER to each other or FUNCTIONALLY MORE RELEVANT to each other will be coded closer together in the word or in the clause. Bybee (1985a) has explored such principles in depth with respect to verbal inflections. On the basis of detailed analysis of a sample of the world’s languages, she has shown a regular crosslinguistic “diagrammatic” relation between meanings and their expression: Verbal inflections differ with respect to the extent to which they are RELEVANT to the verb, that is, the extent to which their meanings DIRECTLY AFFECT THE 554

LEXICAL CONTENT OF THE VERB STEM.

The different

degrees of relevance of verbal categories that can be inflectional is reflected diagrammatically in three ways: (1) The more relevant a category is to the verb, the more likely it is to occur in a synthetic or bound construction with the verb; (2) The more relevant a morphological category is to the verb, the closer its marker will occur with respect to the verb stem; (3) The more relevant a morphological category is to the verb, the greater will be the morpho-phonological fusion of that category with the stem (pp. 11–12). We have noted that children do not have great difficulty in acquiring affixes on verbs marking tense/aspect and person/number, while they have more difficulty in learning to mark verbs for gender and for definiteness of the direct object. We have also noted that children accept stem changes for aspect, but tend to maintain a single stem for all persons and tenses. These child language findings are consistent with Bybee’s crosslinguistic summary of “adult” languages. At this point, we can turn to questions of morpheme ordering, proposing an OP for semantic relevance in ordering; OP (POSITION): RELEVANCE. If two or more functors apply to a content word, try to place them so that the more relevant the meaning of a functor is to the meaning of the content word, the closer it is placed to the content word. If you find that a Notion is marked in several places, at first mark it only in the position closest to the relevant content word. 555

§ In Hungarian and Polish, as discussed in more detail with regard to OP:OPERATORS, below, children tend to re-order person and conditional markers to maintain the order VERB-PERSON-CONDITIONAL. Note that person is more relevant to the verb, while conditionality operates on the level of the proposition. Japanese children try to keep tense closer to the verb than negation, re-ordering in the past to maintain a standard, and relevant order of VERB-TENSE-NEGATIVE. § Where number is marked redundantly, children often limit its marking to the noun. In French, for example, Clark reports errors of plural subjects with singular verbs. In Serbo-Croatian, Radulović (1975) reports early marking of case, gender, and number only on the noun, and not on accompanying adjectives or numerals. (Casemarking of numerals poses great difficulties to children learning Slavic languages. The function of counting or ennumeration is clearly distinct from the marking of grammatical relations.) Conversely, in languages like Turkish, where number is not marked redundantly, children do not make errors of extending the plural morpheme to verbs or adjectives if there is already a plural noun present. Thus “negative evidence” also suggests that the noun is the most relevant locus for number inflection. § In Kaluli NOUN + DEICTIC constructions, case is marked only on the deictic element, but children often mark both the noun and the deictic, resulting in double marking of locative, dative, ergative, and genitive cases. The noun comes first, and case is more relevant to nouns than to deictics. In this instance, relevance can “pull” a 556

marker away from a less relevant position (while other factors, such as overall attention to inflections in a verb-final language, can keep the marker in its original position as well, resulting in overmarking).

Scope of Operators Relevance also functions on the clause level. Just as operators on nouns and verbs are attracted to those parts of speech on grounds of relevance, operators on clauses are attracted to the clause as a whole, or to the verb—as the most central, defining element of the clause. Frequent extraclausal positioning of operators that function on a whole clause provides evidence both for the clause as a unit and for sensitivity to the scope of operators. Current findings in regard to placement of operators were anticipated in the 1973 Universal D3: “There is a tendency to preserve the structure of a sentence as a closed entity, reflected in a development from sentence-external placement of various linguistic forms to their movement within the sentence.” It is now evident that these “various linguistic forms” are clause-level operators. MacWhinney has noted that: Children have no problem placing together items that are related in semantic structure. However, when semantic relations are not clear or when items are related to whole clauses, errors are more frequent. (Consequence ordering 5c) Such errors, however, seem to be motivated by notions of scope. Our final OP guides LMC in operator placement:

557

OP (POSITION): OPERATORS. If a functor operates on a whole structure (phrase or clause), try to place it external to that structure, leaving the structure itself unchanged. The bulk of evidence for OP: OPERATORS comes from pervasive difficulties with negation of various sorts, with additional evidence in regard to marking of interrogation and conditionality. Negation. In a variety of ways, children indicate in their restructuring of parental languages that the scope of negation should be the proposition, as indicated by the verb or the clause as a whole, rather than any particular nonverbal lexical item within a clause. Evidence comes from errors in regard to negative placement and the role of negation in conditioning various sorts of reordering and allomorphy. NEGATIVE PLACEMENT. Children seem to prefer a separate, rather than a bound morpheme for clausal negation (cf. OP:ANALYTIC FORM), often moving the negative operator outside of the verb complex or clause. § In English, early negative forms are attached to primitive sentences (No do this), later moving within the sentence (I no do this, and, with auxiliary modal development, I can’t do this) (Bellugi, 1967; Klima & Bellugi, 1966; Menyuk, 1969; Snyder, 1914). At later stages, children overuse not as an uncontracted morpheme with the copula, although contracted forms (isn’t, wasn’t, etc.) predominate

558

in parental speech. The contracted forms can’t and don’t tend to be preserved, however, suggesting that negation combines naturally with notions of ability and prohibition, while expressed separately for clausal negation. This may be because the proper scope of negation in such instances is ‘ability’ or ‘prohibition’. (However, as Itzchak Schlesinger has pointed out to me, it may be easier for English-speaking children to analyze isn’t and wasn’t into two morphemes, while treating can’t and don’t as unanalyzed amalgams.) § Smoczyńska reports similar negative development for one child in Polish, with early sentence-external negation (e.g. Nie Basia śpi ‘not Basia sleeps’ for Basia ne śpi), later replaced by sentences which keep negative and verb together, but still with sentence-initial negation (Nie śpi Basia). Other children have early sentence-final negation, such as mamusia kopać będzie nie ‘mommy dig will not’. § In Turkish, verbal predicates are negated by insertion of a negative particle after the verb-stem, before modal, tense, and person suffixes, while nonverbal predicates are negated by separate nonexistential or negative copula morphemes. Some children go through a phase of using one of the nonverbal negators with verbs, resulting in a clause-final separate negative morpheme in place of verb-internal inflectional negation (e.g. koy-du-m yok ‘put-PAST-lSG NEG:EXIS’ in place of koy-ma-di-m ‘put-NEG-PAST-1SG’). These children (up to about age 2½) control strings of agglutinated particles, so the problem is not one of agglutination, but rather choice

559

and positioning of a negative morpheme (Aksu-Koç & Slobin). § In Japanese, where the negative marker is inserted between the verb-stem and the past-tense marker, similar errors of negative placement are widespread in the past tense. A frequent error is the use of the negative non-past form nai as an unanalyzed suffix (e.g. deki-ta-nai ‘can-PAST-NEG’ = ‘I couldn’t do it’, in place of deki-na-katta ‘can-NEG-PAST’). Clancy cites many errors of this sort in regard to negative past forms of verbs and adjectives, concluding: “Apparently, children prefer to increase the salience of the negative marker by placing it in word-final position. This makes sense, since the main point of a statement in the negative past is the negation, not the tense.” § French-speaking children may also prefer not to break up the verb complex by inserting a negative element. Clark reports errors such as *Je lui ai fait rien ‘I to:him have done nothing’ for Je lui ai rien fait. EFFECTS OF NEGATION ON OTHER ELEMENTS. Children not only prefer that an operator like negation have a fixed location, but also that it not affect the form or placement of other elements in a sentence. Again, this suggests that negation is an operator whose scope is the clause as a whole. § In Hungarian, focused material is placed in preverbal position, with the consequence that a separable verbal prefix is obligatorily postposed. Negation is treated as a focused element; however children often place the 560

negative morpheme before a prefixed verb, without postposing the prefix (e.g. *nem megeszem ‘not up:eat:1SG’ for nem eszem meg ‘not eat:1SG up’). For children who control postposing for focus, it could be the case that they simply do not conceive of negation as an instance of focus. Or, in any case, children may have a predisposition that negation (or proposition-scope operators in general) should not require movement of other sentence elements. (Similar evidence is presented in regard to conditionals and interrogatives, below.) § As discussed above in regard to word order, in French imperatives, object pronouns follow the verb in the affirmative but precede it in the negative. A frequent regularization is to maintain the affirmative order with negation, as in: *Cache-le pas ‘hide-it not’ for (Ne) le cache pas and *Fais-le pas tomber ‘make-it not fall’ for (Ne) le fais pas tomber. Clark reports such errors for children as old as 9, and suggests that such forms may also appear in adult speech to children. § In Polish and Russian, the direct object of a negated verb is placed in the genitive, rather than the accusative case. Children find this a difficult distinction to acquire, and use the accusative for negated direct objects. § English requires that inderminates be changed to indefinites in negated sentences, but Bellugi (1967) found that her subjects used a single indeterminate form for affirmatives and negatives (e.g. I want some cookie and *Mama didn’t have some cookie). Smoczyńska reports errors of a somewhat related nature for Polish, 561

where the form of the adverb is not adjusted for negation (e.g. A Jesus to zawsze nie gryzie ‘But Jesus always not bites’, where nigdy ‘never’ should be used instead of zawsze ‘always’). § A particular type of English error suggests that children try to keep negative marking in close association with the negated proposition. There is a subset of mental verbs with optional not-transportation (think, want, anticipate, expect, believe). Children treat think as an ordinary complement-taking verb, maintaining negation in the embedded sentence (e.g., I think it’s not full up to the top), although parental speech always has the negative in the embedding sentence (e.g., / don’t think the man wants to taste it). § It also appears that, for children, negating a state of affairs, or treating it as irrealis, should not affect the internal description of that state of affairs. For example, in Russian a clause embedded under ne nado ‘not necessary’ (=‘one mustn’t’) is obligatorily imperfective, even if it refers to a single event considered as a totality (i.e., the normal use of the perfective). Gvozdev reports errors of using the perfective (e.g., *zakryt’ [PFV] ne nado ‘one mustn’t cover’ cin place of zakryvaf ’ [IPFV]). He notes (1961, p. 425): “Such use of the perfective aspect is completely justifiable and is more consistent than the way in which this issue is dealt with in the language.” Interrogation. Question-marking, like negation, poses children with problems of position and ordering of elements. Where languages provide a special intonation 562

contour for yes-no questions, comprehension and use of this device emerge very early in development (e.g. English, German, Japanese)—perhaps because it marks the utterance as a whole, rather than a particular lexical item. Where both intonational and morphosyntactic means are available, intonation seems to develop earlier. Clancy reports that in Japanese, where the most frequently used sentence-final markers in questions also occur in declarative sentences, at least one child relied upon intonation earlier than lexical-grammatical cues to interrogation. Question inversion in English is late to develop, in comparison with rising intonation for yes-no questions. In Finnish (Bowerman, 1973), where the only means of marking yes-no questions is morphological, such questions apparently do not appear in early child speech. Acquisition data from English and Hungarian suggest that interrogation is treated as an operator of propositional scope: § In English (Bellugi-Klima, 1968), children acquire inversion for yes-no questions before wh-questions—that is, at the same stage in which a child uses inverted forms like Can he ride in a truck?, s/he also uses noninverted forms like What he can ride in? Note that both forms mark the utterance initially as a question, either by the noncanonical position of the auxiliary or by a question-word. § In Hungarian, a question particle, -e, is attached to the predicate (main verb or predicate complement). MacWhinney (1973, p. 464) reports errors based on the apparent problem that this particle “relates to the whole 563

proposition, rather than just one noun phrase or verbal element. The child has learned that suffixes are attached to the elements in which they stand in semantic relation. But, in the case of -e, the child doesn’t know what element that should be.” One child regularly placed the particle on the last element of the sentence, whether or not it was a main verb or predicate complement, while another child regularly placed the particle on the first element. In addition, the particle is restricted to affirmative yes-no questions, but some children treat it as a general polar interrogative marker, placing it on negatives as well; e.g., nem veri meg-e ‘not beat PFV-INT’ in place of merveri-e ‘PFV:beat-INT (=‘Will he beat it’); ebbe nincs tea-e ‘in:this isn’t tea-INT’ in place of ebbe van-e tea ‘in:this is-INT tea’ (=Ts there tea in this?’). Conditionality. In languages where the conditional is marked by a special particle, children have problems similar to those of negation and interrogation (though available data are limited). Again, there is a tendency towards separate and verb-external marking of an operator. § In Hungarian the conditional is marked in the present tense by a bound particle on the verb, -ne, in the order VERB+ne+PERSON; and in the past by a postposition, volna. MacWhinney (1973, pp. 593–594) reports early errors of marking present conditional by VERB+PERSON volna, such as beszélek volna ‘speak:lSG COND’ in place of beszélnék ‘speak:COND:lSG’. Such errors are reminiscent of Hungarian and Japanese errors in negation, reported 564

above, in which a clause-level operator seems to resist incorporation into the verb, and is preferentially marked as a separate element rather than a bound morpheme. § The Hungarian errors also suggest a tendency to keep person-number-tense marking close to the verb. This tendency is clearly shown in Polish, where the parental language requires person-number marking on the conditional particle, by, and the conditional connective, żeby. Children, however, prefer to maintain person-number marking on the verb, resulting in a postposed or separate conditional marker; e.g. in place of pisał-by-m ‘wrote-COND-1SG’ or Ja by-m pisał T COND-1SG wrote’, children say pisał-em-by ‘wrote1SG-COND’ or Ja by pisał-em T COND wrote-1SG’. (Note that these errors also reflect a reluctance to attach a person suffix to an element that is not a verb, following OP:RELEVANCE.)

CONCLUSION The conclusion of my original paper on Operating Principles (Slobin, 1971a, pp. 369–370) can well be placed at the end of this chapter almost 15 years later. What has changed is the emphasis—from general cognitive prerequisites to those that seem more adapted to the task of language acquisition in particular. To a great extent, the current work is an elaboration of what I referred to as the child’s “inherent definition of the general structure and function of language.” Let me quote the 1971 conclusion in its entirety, and conclude in 1985 by commenting on it.

565

What has been sketched out on the preceding pages is only an outline of what some day may evolve into a model of the order of acquisition of linguistic structures. It has several major components, all of which must be elaborated. The first component, I have argued, is the development of semantic intentions, stemming from general cognitive development. The child, equipped with an inherent definition of the general structure and function of language, goes about finding means for the expression of those intentions by actively attempting to understand speech. That is to say, he must have preliminary internal structures for the assimilation of both linguistic and non-linguistic input. He scans linguistic input to discover meaning, guided by certain ideas about language, by general cognitive-perceptual strategies, and by processing limitations imposed by the constraints of operative memory. As in all of cognitive development, this acquisition process involves the assimilation of information to existing structures, and the accommodation of those structures to new input. The speech perception strategies engender the formation of rules for speech production. Inner linguistic structures change with age as computation and storage space increase, as increasing understanding of linguistic intentions leads the child into realms of new formal complexity, and as internal structures are interrelated and reorganized in accordance with general principles of cognitive organization. All of these factors are cognitive prerequisites for the development of grammar. While we can disagree about the extent to which this process of developing grammars requires a richly detailed innate language faculty, there can be no doubt that the process requires a richly structured and active child mind. 566

The current characterization of Semantic Space allows the child not only “to scan linguistic input to DISCOVER meaning,” but also to scan linguistic input for the means of expressing highly accessible, prelinguistic meanings. The “preliminary internal structures for the assimilation of both linguistic and non-linguistic input” are not totally separate from one another, and, as a consequence, Basic Child Grammar has recognizable form and content universally. Semantic intentions stem not only from “general cognitive development” but from particular ways of thinking and interacting which, in the human species, are most intimately related to our means of expression in structured language. General cognitive mechanisms are most apparent in the use of Review Strategies and Interim Production Strategies. Structured systems grow and interact continually, and the application of particular Operating Principles depends on a complex of existing linguistic and extralinguistic knowledge, processing constraints, the structure of social interaction, and the structure of the language being acquired. There are strongly inbuilt tendencies to analyze speech into certain types of units, and to systematize and interrelate and cross-classify these units in highly specific ways, as suggested by the collection of Operating Principles for Organizing Information in Storage. Detailed examination of the ways in which children acquire a variety of different types of languages leads to the inescapable conclusion that the structure of human language is highly adapted to the structures of human perception, thought, and action—and that the capacity to 567

construct a human language in earliest childhood must, ultimately, be part of the genetic capacity of our species. The Operating Principles presented here represent a preliminary and partial attempt, on the part of a psycholinguist, to sketch out that capacity—the human Language-Making Capacity.

ACKNOWLEDGMENTS Many people have contributed to the development of the ideas presented in this chapter— more colleagues, students, and friends than can be acknowledged in a brief note. All of the contributors to this volume were of immeasurable help, both for the data that they painstakingly gathered and summarized, and for hours of good discussion. I am especially grateful to a group of careful critics who suggested alternatives and reorganizations, sharing both their puzzlements and their excitement about what I was trying to do: Ruth Berman, Melissa Bowerman, Nancy Budwig, Joan Bybee, Susan Carey, Pat Clancy, Eve Clark, Chuck Fillmore, Julie Gerhardt, Annette Karmiloff-Smith, Paul Kay, Elena Lieven, Itzchak Schlesinger, Magdalena Smoczyńska, Ayşegül Talay, Dick Weist. Each of them will recognize where I understood their suggestions, as well as where I failed to accommodate to them. The long process of systematizing data and preparing manuscripts was eased by the good help of Jane Edwards, Tanya Renner, Jeff Sokolov, and Laurie Wagner. Congenial working environments were provided by the Summer Linguistic Institute at the University of New Mexico, the Max-Planck-Institut für Psycholinguistik in Nijmegen,

568

and the Institute of Human Learning and Department of Psychology of the University of California at Berkeley. Research contributing to this chapter was supported by the National Science Foundation (Grant BNS 80-09340: “Crosslinguistic Study of Language Acquisition”) and by the Sloan Foundation Program in Cognitive Science, University of California, Berkeley. To all of these, my thanks. And, above all, to Ayşegül Talay, who has shown me that Yeats was wrong when he said: “The intellect of man is forced to choose - Perfection of the life, or of the work …”

REFERENCE Aksu-Koç, A. A. The acquisition of past reference in Turkish. In D. I. Slobin & K. Zimmer (Eds.), Studies in Turkish linguistics. Amsterdam: John Benjamins, in press. Aksu-Koç, A. A., & Slobin, D. I. The acquisition of Turkish. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Ammon, M. S., & Slobin, D. I. A cross-linguistic study of the processing of causative sentences. Cognition, 1979, 7, 3–17. Antinucci, F., Duranti, A., & Gebert, L. Relative clause structure, relative clause perception, and the change from SOV to SVO. Cognition, 1979, 7, 145–176.

569

Antinucci, F.,& Miller, R. How children talk about what happened. Journal of Child Language, 1976, 3, 167–189. Badry, F. Acquisition of lexical derivational rules in Moroccan Arabic: Implications for the development of Standard Arabic as a second language through literacy. Unpublished doctoral dissertation, University of California, Berkeley, 1983. Baker, C. L., & McCarthy, J. J. The logical problem of language acquisition. Cambridge, MA: MIT Press, 1981. Behagel, O. Deutsche Syntax: Eine geschichtliche Darstellung. Vol. 4: Wortstellung. Periodenbau. Heidelberg: Carl Winters Universitätsbuchhandlung, 1932. Bellugi, U. The acquisition of negation. Unpublished doctoral dissertation, Harvard University, 1967. Bellugi-Klima, U. Linguistic mechanisms underlying child speech. In E. M. Zale (Ed.), Proceedings of the conference on language and language behavior. New York: Appleton-Century-Crofts, 1968. Berko, J. The child’s learning of English morphology. Word, 1958, 14, 150–177. Berman, R. The acquisition of Hebrew. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum, 1985.

570

Bickerton, D. Roots of language. Ann Arbor, MI: Karoma Publishers, 1981. Bogoyavlenskiy, D. N. Psixologija usvoenija orfografii. Moscow: Izd-vo Akademii Pedagogičeskix Nauk RSFSR, 1957. [Partial translation: Bogoyavlenskiy, D. N. The acquisition of Russian inflections. In C. A. Ferguson & D. I. Slobin (Eds.), Studies of child language development. New York: Holt, Rinehart & Winston, 1973.] Bowerman, M. Learning the structure of causative verbs: A study in the relationship of cognitive, semantic and syntactic development. Papers and Reports on Child Language Development (Department of Linguistics, Stanford University), 1974, 8, 142–178. Bowerman, M. Semantic factors in the acquisition of rules for word use and sentence construction. In D. M. Morehead & A. E. Morehead (Eds.), Normal and deficient child language. Baltimore: University Park Press, 1976. Bowerman, M. Beyond communicative adequacy: From piecemeal knowledge to an integrated system in the child’s acquisition of language. Papers and Reports on Child Language Development (Department of Linguistics, Stanford University), 1981, 20, 1–24. Bowerman, M. Starting to talk worse: Clues to language acquisition from children’s late speech errors. In S. Strauss (Ed.), U-shaped behavioral growth. New York: Academic Press, 1982. 571

Braine, M. D. S. Children’s first word combinations. Monographs of the Society for Research in Child Development, 1976, 41 (Serial No. 164). Bresnan, J. (Ed.) The mental representation of grammatical relations. Cambridge, MA: MIT Press, 1983. Brown, R. Linguistic determinism and the part of speech. Journal of Abnormal and Social Psychology, 1957, 55 1–5. Brown, R. A first language: The early stages. Cambridge, MA: Harvard University Press, 1973. Bybee Hooper, J. Child morphology and morphophonemic change. Linguistics, 1979, 17, 21–50. Bybee, J. L. Diagrammatic iconicity in stem-inflection relations. In J. Haiman (Ed.), Iconicity in syntax. Amsterdam: John Benjamins, 1985. (a) Bybee, J. L. Morphology: A study of the relation between meaning and form. Amsterdam: John Benjamins, 1985. (b) Bybee, J. L., & Slobin, D. I. Rules and schemas in the development and use of the English past tense. Language, 1982, 58, 265–289. Carey, S. The child as word learner. In M. Halle, J. Bresnan, & G. A. Miller (Eds.), Linguistic theory and psychological reality. Cambridge, MA: MIT Press, 1978.

572

Chomsky, N. Aspects of the theory of syntax. Cambridge, MA: MIT Press, 1965. Clancy, P. M. The acquisition of Japanese. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Clark, E. V. Language change during language acquisition. In M. E. Lanb & A. L. Brown (Eds.), Advances in child development, Vol. 2. Hillsdale, NJ: Lawrence Erlbaum Associates, 1982. Clark, E. V. The acquisition of Romance, with special reference to French. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. de Lemos, C. Interactional processes in the child’s construction of language. In W. Deutsch (Ed.), The child’s construction of language. London: Academic Press, 1981. Deutsch, W., & Budwig, N. Form and function in the development of possessives. Papers and Reports on Child Language Development (Stanford University, Department of Linguistics), 1983, 22, 36–42. de Villiers, J. G., & de Villiers, P. A. The acquisition of English. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985.

573

Erbaugh, M. Acquisition of temporal and aspectual distinctions in Mandarin. Papers and Reports on Child Language (Department of Linguistics, Stanford University), 1978, 15, 30–37. Fillmore, C.J. The case for case. In E. Bach & R. T. Harms (Eds.), Universals in linguistic theory. New York: Holt, Rinehart & Winston, 1968. Fillmore, C. J. Topics in lexical semantics. In R. W. Cole (Ed.), Current issues in linguistic theory. Bloomington: Indiana University Press, 1977. Gee, J., & Savasir, I. On the use of will and gonna: Towards a description of activity-types for child language. Discourse Processes, 1985,8, 143–175. Geodakjan, I. M. Expression of the category of time in the speech of a young child. In J. Průcha (Ed.), Soviet studies in language and language behavior. Amsterdam: North-Holland, 1976. Gerhardt [Gee], J. B. Tout se tient: Towards an analysis of activity-types to explicate the interrelation between modality and future reference in child discourse. Unpublished doctoral dissertation, University of California, Berkeley, 1983. Givón, T. Function, structure, and language acquisition. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 2). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. (a)

574

Givón, T. Iconicity, isomorphism, and non-arbitrary coding in syntax. In J. Haiman (Ed.), Iconicity in syntax. Amsterdam: John Benjamins, 1985. (b) Gopnik, A. The acquisition of gone and the development of the object concept. Journal of Child Language, in press. Gvozdev, A. N. Formirovanie u rebenka grammatičeskogo stroja russkogojazyka. Moscow: Izd-vo Akademii Pedagogičeskix Nauk RSFSR, 1949. [Reprinted in A. N. Gvozdev, Voprosy izučenija detskoj reči. Moscow: Izd-vo Akademii Pedagogiůeskix Nauk RSFSR, 1961, pp. 149–467.] Herring, S. Tense vs. aspect and focus of attention in the development of temporal reference. Unpublished paper, University of California, Department of Linguistics, 1981. Herskovits, A. Space and the prepositions in English: Regularities and irregularities in a complex domain. Unpublished doctoral dissertation, Stanford University, 1982. Hopper, P. J., & Thompson, S. Transitivity in grammar and discourse. Language, 1980, 56, 251–299. Horgan, D. The development of the full passive. Journal of Child Language, 1978, 5, 65–80. Ingram, D. Phonological patterns in the speech of young children. In P. Fletcher & M. Garman (Eds.), Language

575

acquisition: Studies in first language development. Cambridge: Cambridge University Press, 1979. Ivimey, G. P. The development of English morphology: An acquisition model. Language and Speech, 1975, 18, 120–144. Johnston, J. R. Acquisition of locative meanings: behind and in front of. Journal of Child Language, 1984, 11, 407–422. Johnston, J. R. Cognitive prerequisites: The evidence from children learning English. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 2). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Johnston, J. R., & Slobin, D. I. The development of locative expressions in English, Italian, Serbo-Croatian and Turkish. Journal of Child Language, 1979, 6, 529–545. Karmiloff-Smith, A. A functional approach to child language. Cambridge: Cambridge University Press, 1979. Klima, E. S., & Bellugi, U. Syntactic regularities in the speech of children. In J. Lyons & R. J. Wales (Eds), Psycholinguistics papers. Edinburgh: Edinburgh University Press, 1966. Kuczaj, S. A., II Children’s judgments of grammatical and ungrammatical irregular past-tense verbs. Child Development, 1978, 49, 319–326.

576

Lakoff, G., & Johnson, M. Metaphors we live by. Chicago: University of Chicago Press, 1980. Leont’yev, A. A. Slovo v rečevoj dejatel’nosti. Moscow: Nauka, 1965. Levy, Y. It’s frogs all the way down. Cognition, 1983, 15, 75–93. Lord, C. “Don’t you fall me down”: Children’s generalizations regarding cause and transitivity. Papers and Reports on Child Language Development (Department of Linguistics, Stanford University), 1979, 17, 81–89. Lord, C. The development of object markers in serial verb languages. In P. J. Hopper & S. A. Thompson (Eds.), Syntax and semantics, Vol. 15: Studies in transitivity. New York: Academic Press, 1982. MacWhinney, B. J. How Hungarian children learn to speak. Unpublished doctoral dissertation, University of California, Berkeley, 1973. MacWhinney, B. The acquisition of morphophonology. Monographs of the Society for Research in Child Development, 1978, 43, Whole No. 1. MacWhinney, B. Hungarian language acquisition as an exemplification of a general model of grammatical development. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 2). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985.

577

Maratsos, M. The child’s construction of grammatical categories. In E. Wanner & L. R. Gleitman (Eds.), Language acquisition: The state of the art. Cambridge: Cambridge University Press, 1982. Maratsos, M. Some current issues in the study of the acquisition of grammar. In P. H. Mussen (Ed.), Handbook of child psychology (Vol. 3), 4th ed. New York: Wiley, 1983. Maratsos, M., & Chalkley, M. A. The internal language of children’s syntax. In K. Nelson (Ed.), Children’s language, Vol. 2. New York: Gardner Press, 1980. Maratsos, M., & Kuczaj, S. A., II Against the transformationalist account: A simpler analysis of auxiliary overmarking. Journal of Child Language, 1978, 5, 337–345. Menn, L., & MacWhinney, B. The repeated morph constraint: Towards an explanation. Unpublished paper, Aphasia Research Center, Boston University School of Medicine and Psychology Department, Carnegie-Mellon University, 1983. Menyuk, P. Sentences children use. Cambridge, MA: MIT Press, 1969. Mills, A. E. The acquisition of German. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985.

578

Naro, A. A study of the origins of pidginization. Language, 1978, 54, 314–347. Ochs, E. Variation and error: A sociolinguistic approach to language acquisition in Samoa. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Omar, M. The acquisition of Egyptian Arabic as a native language. The Hague: Mouton, 1973. Parisi, D., & Antinucci, F. Lexical competence. In G. B. Flores d’Arcais & W. J. M. Levelt (Eds.), Advances in psycholinguistics. Amsterdam: North-Holland, 1970. Pačesová, J. Řeč v raném dětstvi. Brno: Univerzita J. E. Purkyně, 1979. Peters, A. M. Language segmentation: Operating principles for the perception and analysis of language. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 2). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Pinker, S. A theory of the acquisition of lexical interpretive grammars. In J. Bresnan (Ed.), The mental representation of grammatical relations. Cambridge, MA: MIT Press, 1982. Pinker, S. Language learnability and language development. Cambridge, MA: Harvard University Press, 1984.

579

Pinker, S., & Lebeaux, D. A learnability-theoretic approach to children’s language. Unpublished manuscript, 1981. Pye, C. L. The acquisition of grammatical morphemes in Quiche Mayan. Unpublished doctoral dissertation, University of Pittsburgh, 1979. Pye, C. The acquisition of person markers in Quiche Mayan. Papers and Reports on Child Language Development (Department of Linguistics, Stanford University), 1980, 19, 53–59. Radulović, L. Acquisition of language: Studies of Dubrovnik children. Unpublished doctoral dissertation. University of California, Berkeley, 1975. Réger, Z. Közös törvényszerűségek az anyanyelvi-elsajátítás és a gyermekkori idegen nyelvi-elsa-játítás folyamatában. Magyar Nyelvőr, 1975, 99, 343–350. Rüķe-Draviņa, V. Sprachentwicklung bei Kleinkindern. I. Syntax: Beitrag auf der Grundlage lettischen Sprachmaterials. Lund: Slaviska Institutionen vid Lunds Universitet, 1963. Sapir, E. Language: An introduction to the study of speech. New York: Harcourt Brace, 1921. [Harcourt, Brace & Co. Harvest Book, 1949.]

580

Savasir, I. How many futures? Unpublished M.A. dissertation, University of California, Department of Psychology, 1983. Savasir, I., & Gee, J. The functional equivalents of the middle voice in child language. Proceedings of the Berkeley Linguistic Society, 1982. Schieffelin, B. B. How Kaluli children learn what to say, what to do, and how to feel: An ethnographic study of the development of communicative competence. Unpublished doctoral dissertation, Columbia University, 1979. [To be published by Cambridge University Press.] Schieffelin, B. B. The acquisition of Kaluli. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Schlesinger, I. M. Steps to language: Toward a theory of native language acquisition. Hillsdale, NJ: Lawrence Erlbaum Associates, 1982. Simōes, M. C. P., & Stoel-Gammon, C. The acquisition of inflections in Portuguese: A study of the development of person markers on verbs. Journal of Child Language, 1979, 6, 53–67. Slobin, D. I. Soviet psycholinguistics. In N. O’Connor (Ed.), Present-day Russian psychology: A symposium by seven authors. Oxford: Pergamon, 1966. (a)

581

Slobin, D. I. The acquisition of Russian as a native language. In F. Smith & G. A. Miller (Eds.), The genesis of language: A psycholinguistic approach. Cambridge, MA: MIT. Press, 1966. (b) Slobin, D. I. Developmental psycholinguistics. In W. O. Dingwall (Ed.), A survey of linguistic science. College Park, MD: University of Maryland Linguistics Program, 1971. (a) Slobin, D. I. On the learning of morphological rules: A reply to Palermo and Eberhart. In D. I. Slobin (Ed.), The ontogenesis of grammar: A theoretical symposium. New York: Academic Press, 1971. (b) Slobin, D. I. Cognitive prerequisites for the development of grammar. In C. A. Ferguson & D. I. Slobin (Eds.), Studies of child language development. New York: Holt, Rinehart & Winston, 1973. Slobin, D. I. Language change in childhood and in history. In J. Macnamara (Ed.), Language learning and thought. New York: Academic Press, 1977. Slobin, D. I. Psycholinguistics, 2nd ed. Glenview, IL: Scott, Foresman & Co., 1979. Slobin, D. I. The repeated path between transparency and opacity in language. In U. Bellugi & M. Studdert-Kennedy (Eds.), Signed and spoken language:

582

Biological constraints on linguistic form. Weinheim (German Federal Republic): Verlag Chemie, 1980. Slobin, D. I. The origins of grammatical encoding of events. In W. Deutsch (Ed.), The child’s construction of language. London: Academic Press, 1981. Slobin, D.I. Universal and particular in the acquisition of language. In E. Wanner & L. R. Gleitman (Eds.), Language acquisition: The state of the art. Cambridge: Cambridge University Press, 1982. Slobin, D. I. The child as a linguistic icon-maker. In J. Haiman (Ed.), Iconicity in syntax. Amsterdam: John Benjamins, 1985. Slobin, D. I., & Bever, T. G. Children use canonical sentence schemas: A crosslinguistic study of word order and inflections. Cognition, 1982, 12, 229–265. Smoczyńska, M. The acquisition of Polish. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Snyder, A. D. Notes on the talk of a two-and-a-half-year-old boy. Pedagogical Seminary, 1914, 21, 412–424. Stampe, D. The acquisition of phonetic representation. In Papers from the Fifth Regional Meetings. Chicago: Chicago Linguistic Society, 1969.

583

Talmy, L. Relation of grammar to cognition. In D. Waltz (Ed.), Proceedings of TINLAP-2 (Theoretical Issues in Natural Language Processing). Champaign, IL: Coordinated Science Laboratory, University of Illinois, 1978. Talmy, L. How language structures space. In H. Pick & L. Acredolo (Eds.), Spatial orientation: Theory, research, and application. New York: Plenum Press, 1983. Talmy, L. Lexicalization patterns: Semantic structure in lexical form. In T. Shopen (Ed.), Language typology and syntactic description, Vol. 3: Grammatical categories and the lexicon. Cambridge: Cambridge University Press, 1985. Toivainen, J. Inflectional affixes used by Finnish-speaking children aged 1–3 years. Helsinki: Suomalaisen Kirjallisuuden Seura, 1980. Varma, T. L. Stage I speech of a Hindi-speaking child. Journal of Child Language, 1979, 6, 167–173. Wexler, K., & Culicover, P. Formal principles of language acquisition. Cambridge, MA: MIT Press, 1980. Wong Fillmore, L. The second time around: Cognitive and social strategies in second language acquisition. Unpublished doctoral dissertation, Stanford University, 1976.

584

1

The languages included in this collection of acquisition summaries are ASL, English, French, German, Hebrew, Hungarian, Italian, Japanese, Kaluli, Polish, Portuguese, Rumanian, Samoan, Spanish, and Turkish. At some points I have added data on Finnish, Russian, Serbo-Croatian, Czech, and Quiche. Acquisition of ASL is not addressed in this chapter, as the summary of ASL was not yet available at the time of writing. It will be of interest, therefore, to determine the extent to which the approach presented here is also applicable to the acquisition of language in another modality. 2

I suspect that many of the principles of Basic Child Grammar will also appear in the development of Creoles and in processes of historical language change, along lines suggested, for example, by Bickerton (1981) and in earlier works of my own (Slobin, 1977, 1980). 3

references with no citation norefer2 to papers in this collection, and appear at the end of the chapter with the date 1985. 4

For convenience of expression, OPs are phrased as imperatives to LMC. This phrasing, however, should not suggest any conscious metalinguistic activity, nor does it require a “little linguist” hard at work at solving problem sets. Each OP can be readily re-stated in a passive or third-person format to lessen the impression of conscious agency; e.g., “Attention will be paid to …”;or”The child will pay attention to …”; or “The final syllable of an intonational unit will attract attention”; etc.

585

5

Judith Johnston, in this volume, provides an insightful critique of the problem of holding semantic content constant. For our present purposes, keeping her important caveats in mind, we will operate on the assumption of universality for at least the INITIAL semantic notions that receive formal expression in Basic Child Grammar, seeking evidence for such notions. 6

Scattered through my 1973 paper, “Cognitive prerequisites for the development of grammar,” are other principles that have implications for perception and the sorts of material that will be stored by the child. Operating Principle C, “Pay attention to the order of words and morphemes,” claims— quite correctly in regard to current evidence—that some aspects of input will be stored in the order received. However, as discussed in the present chapter, current crosslinguistic data reveal language-particular differences in regard to the types of word and morpheme orders that are salient to the child. It is not always the case, as was stated in Universal C3, that: “Sentences deviating from standard word order will be interpreted at early stages of development as if they were examples of standard word order.” We now know that children develop an early sensitivity to both semantic and pragmatic roles of word order in the input language, and that “Universal” C3 must be limited to languages that rely heavily on word order to express semantic relations (Slobin, 1981, 1982; Slobin & Bever, 1982). Indeed, an important finding emanating from crosslinguistic study is the fact that children develop a general notion of the TYPE of language that they are learning—in part through the successive strengthening or weakening of particular OPs 586

on the basis of their usefulness in deciphering the input language. (See OP:STRENGTHENING, below.) Operating Principle D, “Avoid interruption or rearrangement of linguistic units,” applies primarily to production. (As such, it should be phrased in positive terms as a tendency to keep related material together in speech, rather than as an “avoidance” of structures that are not uttered. Descendants of this OP appear in the present chapter as more specific OPs for positioning of grammatical elements in speech.) On the perceptual side, however, a consequence of Operating Principle D has received ample crosslinguistic support: Universal D4: “The greater the separation between related parts of a sentence, the greater the tendency that the sentence will not be adequately processed …” Relevant data from relative clause comprehension in English, French, Japanese, and Turkish are discussed in this volume. These data also support Universal E5: “It is easier to understand a complex sentence in which optionally deletable material appears in its full form.” This proposed universal was a consequence of Operating Principle E: “Underlying semantic relations should be marked overtly and clearly”—an OP which is replaced by a more detailed set of OPs in the present chapter. Stated more broadly, the languages considered here differ considerably in the extent to which they provide “local cues” (Ammon & Slobin, 1979; Slobin, 1982) to the segmentation of sentences into clauses and to the identification of the grammatical roles of constituents in sentence processing. In terms of the perceptual filter, it is likely that elements such as case inflections, verb inflections, pre- or postpositions, and conjoining and 587

subordinating particles provide major orienting points for the perception of structure. 7

OPs are norefer2red to by key-words in capitals, omitting the parenthetical indication of an OP’s functional domain as stated in the initial definition of an OP. The functional domains of OPs are: ATTENTION, STORAGE, MAPPING, REVIEW, PRODUCTION, UNITS, and POSITION. A full listing of OPs and their definitions is given in the Appendix to this chapter. 8

Terms that are given special meanings in this chapter are capitalized. 9

As a consequence of the OPs considered thus far, we should expect to find grammatical particles and affixes in early child speech if they are perceptually salient and expressive of basic notions. This prediction was made by Brown (1973, p. 88) on the basis of crosslinguistic data available to him at the time, and it is amply supported by reports in this volume of early noun and/or verb inflections in Polish, Hungarian, Turkish, Japanese, and Hebrew. In addition to salience of certain sorts of affixes, it is apparent from acquisition of Semitic languages that children are also sensitive to the role of vowel patterns within consonant frames, as discussed below in regard to OP:WORD FORMS, allowing intercalated vowels also to stand out perceptually and serve as candidates for functors. Patterns of tonal morphemes may also have a similar status in the acquisition of languages like Chinese.

588

10

Linguistically, such concepts are encoded in transitive clauses. Hopper and Thompson (1980) have characterized a “highly transitive clause type” as a universal conceptual core of grammatical marking of transitivity. In this core one finds the grammaticization of Scenes in which a human agent behaves “actively, volitionally, and totally to a definite or norefer2ential object.” Thus a basic child activity type also seems to be an organizing point for linguistic expression in the languages of the world. Lord (1982) has observed historical parallels in which a verb meaning ‘take’ in serial verb constructions has been reanalyzed as a preposition to mark patient nouns in Mandarin and in Benue-Kwa languages of West Africa. Furthermore, she finds that this pattern of diachronic innovation is manifested most strongly in clauses which are affirmative and perfective (cf. discussion of Kaluli acquisition below). 11

The acquisition of another ergative language, the Mayan language Quiche, presents a similar picture. In that language, ergativity is marked on the verb by a choice of ergative or absolutive person markers. As in Kaluli, children do not extend the ergative markers from transitive to intransitive constructions (Pye, 1979, 1980). Ochs suggests that Samoan children, in their use of word-order regularities, also show a sensitivity to agents as differentiated from actors of intransitive verbs. She finds that “young children reserve the location immediately following the verb for absolutive constituents—transitive patients and intransitive major arguments—but exclude ergative constituents—agents—from this position. (In this way, 589

they treat patients and intransitive arguments as a single category, distinct from agents.)” 12

Note that languages often neutralize the static-dynamic locative distinction. For example, in English one says I poured the milk IN the glass and The milk is IN the glass, allowing the distinction between an active verb and the copula to differentiate in as goal and in as location. (Departures from a state, however, are encoded distinctly, e.g., I poured the milk OUT OF the glass.) 13

Melissa Bowerman (personal communication) has pointed out that semantic errors in the use of grammatical forms must be considered in relation to linguistic aspects of the forms themselves. For example, location-possession confusions do not occur in early English child speech, where a clear word-order difference is available (e.g. mommy ball ‘mommy’s ball’ versus ball table ‘the ball is on the table’). In regard to mapping, then, my claim is that certain conflations of meaning are likely in those situations in which the language either (1) does not present highly differentiated forms (e.g. the alternation between two suffixes in Hungarian is probably less salient to LMC than the two word-order patterns of English), or (2) one of the forms is more accessible to LMC in terms of OPs for formal acquisition (e.g. the German prepositional construction is probably easier to acquire than the genitival constructions required for possessive marking). In such situations, one form will embrace several meanings if those meanings are close to one another in Semantic Space.

590

14

MacWhinney uses the term “propositional relatedness” for a similar notion in his Operating Principle for “amalgam acquisition.” My use of the term “relevance” is due to Joan Bybee (1985a,b), who has introduced it as a technical term to characterize the degree of “interpenetration of semantic frames.” She has used this notion to account for the finding that some grammatical categories are more likely than others to be expressed inflectionally and to engender morphological fusion with the stem to which they are affixed. Using such criteria, for example, aspect is “more relevant” to the meaning of a verb than tense, and mood and person are “less relevant” than either aspect or tense. These notions are explored in more detail below in the presentation of children’s positioning of grammatical morphemes, as guided by OP:RELEVANCE. 15

In Polish it is the animate form that is overextended to all masculine accusatives, while in Russian it is the inanimate. However, for the present argument it is only significant that one form is used for both. The uncertainty of adult usage in Polish provides additional evidence of the difficulty in attending to animacy in choice of accusative form. 16

MacWhinney proposes a similar OP: “10. STRENGTHENING. Every time a rule or item applies successfully, it gains in strength. Every time a rule or item applies unsuccessfully, it loses strength. Losses in strength from incorrect application are greater than increments obtained during activation.” I have phrased the OP in terms of “attempted solutions,” rather than “rules” or “items,” in order to allow for strengthening of 591

particular OPs or parts of OPs in application to particular input languages. I have no evidence of the relative effects of successful and unsuccessful attempts; therefore the OP simply allows for strengthening through success, with the automatic consequence that unsuccessful attempts will result in relatively less reliance on the strategies involved. 17

Russian data, originally coming from Gvozdev (1949), are summarized in part in Slobin (1966a, 1966b). 18

Late errors of overgeneralized vowel-change patterns in English suggest that even in such a marginal domain children DO eventually attempt Semitic-type generalizations, as in ride/red, fight/fit, and sting/stang (reported for third-graders by Bybee & Slobin, 1982). 19

Again, it is important to point out that OP:STRENGTHENING must function in conjunction with other applicable OPs. For example, a solution can only be extended to similar problems if such extensions are possible within the morphological structures of the language. Kaluli children do not extend case suffixes from noun to pronouns, probably because pronouns are not suitable environments for case suffixes, given patterns of usage stored in terms of OP:CO-OCCURRENCE. That is, since such forms never occur on pronouns, children do not attempt a possible extension. Note that the examples of a noun modification given above are all consistent with OP:CO-OCCURRENCE. In German, the numeral ein ‘one’, and quantifiers like viel ‘many’, DO take adjective inflections. In Japanese, no occurs as a unbound particle 592

before nouns. And in Polish some case suffixes can be placed on possessors. OP:STRENGTHENING requires that a solution be successful SOMEWHERE before it is extended. 20

The relevant OPs from MacWhinney’s chapter are: “6. EXPRESSIVE TALLYING: In expression children can ‘listen’ to their own output. By doing this, they can check to see if what they have said fails to include a part of what they wanted to say. 7. RECEPTIVE TALLYING: Children can check to see if they do not understand part of a string. 8. EXPRESSIVE CRITICISM: When children ‘listen’ to their own output, the representation of a combinatorial form can facilitate retrieval of a weak rote item. Children can then check to see if there is a discrepancy between the weak rote receptive form and the combinatorial expressive forms. 9. RECEPTIVE CRITICISM: In reception, children sometimes note a discrepancy between the sound of their own incorrect rote expressive forms and the correct forms to which they are repeatedly exposed.” 21

This OP leads to the 1973 Universal E4 as a consequence: “When a child first controls a full form of a linguistic entity which can undergo contraction or deletion, contractions or deletions of such entities tend to be absent.” 22

If this is the case, the definition of “speech segment” in OP:FUNCTORS should include vowel patterns in consonant frames which have acquired norefer2ential meaning. That is to say, for example, that once a Hebrew-speaking child has identified k-t-v in some 593

embodiment having to do with ‘writing’ (e.g. katav ‘he wrote’), the occurrence of other vowel patterns in that frame should attract attention as “uninterpreted speech segments” (e.g. kotev ‘he writes’), leading to attempts to map the vowel patterns -a-a- and -o-e- onto accessible grammaticizable Notions (e.g. tense/aspect). Later, following OP:FUNCTOR CLASSES, such vowel patterns will be systematized into classes of grammatical morphemes, along with affixes, particles, etc. 23

Some categories must be simply rote-learned, such as the list of verbs that do not take past-tense suffixing in English, or the arbitrary allocation of Russian nouns ending in palatalized consonants to either masculine or feminine gender. Rote-learning of exceptions or unsystematic uses is of little relevance to the cognitive issues underlying an exposition of LMC. We will assume that OP:ROTE provides LMC with a vast capacity to store and use forms that resist systematization.

594

Appendix: Operating Principles for the Construction of Language

FILTERS FOR PRIMARY PERCEPTION AND STORAGE OF INPUT ATTENTION TO SPEECH OP (ATTENTION): SOUNDS. Store any perceptually salient stretches of speech.

595

OP (ATTENTION): END OF UNIT. Pay attention to the last syllable of an extracted speech unit. Store it separately and also in relation to the unit with which it occurs. OP (ATTENTION): STRESS. Pay attention to stressed syllables in extracted speech units. Store such syllables separately and also in relation to the units with which they occur. OP (ATTENTION): BEGINNING OF UNIT. Pay attention to the first syallable of an extracted speech unit. Store it separately and also in relation to the unit with which it occurs.

Entering and Tagging Information in Storage OP (STORAGE): FREQUENCY. Keep track of the frequency of occurrence of every unit and pattern that you store. OP (STORAGE): UNITS. Determine whether a newly extracted stretch of speech seems to be the same as or different from anything you have already stored. If it is different, store it separately; if it is the same, take note of this sameness by increasing its frequency count by one. OP (STORAGE): CO-OCCURRENCE. For every segmented unit within an extracted speech string, note its co-occurrence with any preceding or following unit and store sequences of co occurring units, maintaining their serial order in the speech string. 596

OP (STORAGE): UNIT FORMATION. If you discover that two extracted units share a phonologically similar portion, segment and store both the shared portion and the residue as separate units. Try to find meanings for both units.

THE PATTERN MAKERS Grouping Information in Storage OP (STORAGE): WORD CLASSES. Store together as a class all words (phonological speech unit and meaning) that co-occur with a given functor. Store together as a class words that co-occur with the same groups of functors across utterances. Try to systematize word classes on semantic grounds, forming prototypes and looking for common features. OP (STORAGE): FUNCTOR CLASSES. Store together all functors that co-occur with members of an established word class, and try to map each functor onto a distinct Notion. OP (STORAGE): PHRASES. Store together sequences of word classes and functor classes that co-occur in the expression of a particular Notion or that co-occur with content words belonging to the same class. OP (STORAGE): CLAUSES. Store together ordered sequences of word classes and functor classes that co-occur in the expression of a particular proposition type, along with a designation of the proposition type.

597

Strategies for Grammatical Organization of Stored Information Organizing Information in Storage: Linguistic Units OP (UNITS): WORD FORMS. If you discover more than one form of a word or word-stem in storage, or if monitoring reveals a mismatch between your word form and that in the input, try to find a phonological or semantic basis for distinguishing the forms: (a) Phonologically attempt to change your word form in the given environment, following a hierarchy of possible adjustments of word forms. At first try to maintain the consonant frame and syllable structure (number of syllables, stress placement). (b) Try to find distinct meanings for words or word-stems that occur in varying forms, checking for relevant Notions. (c) If you cannot find a principled basis for differentiating the forms of a word or word-stem, pick one form as basic and use it in all environments. OP (UNITS): PHONOLOGICAL CONDITIONING OF ALLOMORPHY. If you discover diverse forms of a functor expressing a given Notion with regard to words of a particular class, compare these forms with the forms of the other functors and/or the word-stems with which they occur.

598

(a) If you find phonological similarities between co-occurring forms, regularly adjust the form of the functor to fit its environment (following your hierarchy of predispositions for phonological conditioning of allomorphy). (b) If you fail, use a single form of the functor where possible and/or omit the functor. OP (UNITS): MORPHOLOGICAL PARADIGMS. If you find more than one functor expressing a given Notion relative to a particular word class, and choice of functor cannot be determined by phonological conditioning: (a) Try to find semantic grounds for subdividing the Notion expressed by the functors, and map each new Notion onto one of the functors. (b) If you cannot find semantic grounds for choice of functor, check the citation forms of the associated words or stems and try to differentiate them on systematic phonological grounds. If you succeed, set up a paradigm in which choice of functor is conditioned by the phonological shape of the citation form. (c) If you do not succeed in setting up a paradigm based on the phonology of the citation form, and if your procedural capacities allow you to check the immediate environments of citation forms, try to differentiate the functors on the basis of elements that systematically co-occur with the citation forms, and set up a paradigm

599

in which choice of functor is conditioned by factors that regularly co-occur with the citation form. (d) If you fail, use only the most salient and applicable functor to express the given Notion in the given position. OP (UNITS): CANONICAL CLAUSE FORM. If a clause has to be reduced, rearranged, or otherwise deformed when not functioning as a canonical main clause, attempt to use or approximate the full or canonical form of the clause.

Organizing Information in Storage: Form-Function Mapping OP (MAPPING): DICTIONARY. Pay attention to sound sequences that have a readily identifiable meaning and store them in a Dictionary, along with a representation of the context in terms of available semantic and pragmatic Notions in Semantic Space. OP (MAPPING): CONTENT WORDS AND ROUTINES. Try to map extracted speech units onto representations of objects and events—the core norefer2ential meanings and pragmatic functions associated with typical activities and interactions. Store units with their meanings. OP (MAPPING): FUNCTORS. If a speech segment remains uninterpreted after the establishment of content words and routines, try to map it onto an accessible grammaticizable Notion that is relevant to the meaning of adjacent norefer2ential units in the situation in which 600

the speech segments occur. If you succeed, store such a nonnorefer2ential relational unit (“functor”) with its meaning and its placement in relation to associated linguistic units and their meanings. OP (MAPPING): CONNECTIVES. Once you have established means of linguistic expression for whole propositions (clauses), if an uninterpreted functor cannot be mapped onto an accessible Notion, and it occurs in an utterance with two clauses, try to assign it a function that relates the two clauses. If you succeed, store the functor and a definition of its interclausal function and placement. OP (MAPPING): VARIABLE WORD ORDER. If you find that a clause type occurs in more than one word order, attempt to find a distinct function for each order. OP (MAPPING): EXTENSION. If you have discovered the linguistic means to mark a Notion in relation to a word class or configuration, try to mark the Notion on every member of the word class or every instance of the configuration, and try to use the same linguistic means to mark the Notion. OP (MAPPING): AFFIX-CHECKING. Do not add an affix to a word or wordstem that appears to contain that affix in the relevant position. OP (MAPPING): UNIFUNCTIONALITY. If you discover that a linguistic form expresses two closely related but distinguishable Notions, use available means in your language to distinctly mark the two Notions. 601

OP (MAPPING): ANALYTIC FORM. If you discover that a complex Notion can be expressed by a single, unitary form (synthetic expression) or by a combination of several separate forms (analytic expression), pnorefer2 the analytic expression.

Organizing Information in Storage: Position of Elements OP (POSITION): INTRAWORD MORPHEME ORDER. Keep the order of morphemes in a word constant across the various environments in which that word can occur. OP (POSITION): PHRASAL MORPHEME ORDER. Keep the order of morphemes in a phrase constant across the various environments in which that phrase can occur. OP (POSITION): FIXED WORD ORDER. If you have determined that word order expresses basic semantic relations in your language, keep the order of morphemes in a clause constant. OP (POSITION): MORPHEME PLACEMENT. Mark a Notion in the same place in the various constructions in which it can occur. If you discover that a particular class of words or functors occurs in different positions in different constructions, try to find a principled basis to differentiate the constructions. If you fail to define distinct construction types, use the same position across constructions.

602

OP (POSITION): RELEVANCE. If two or more functors apply to a content word, try to place them so that the more relevant the meaning of a functor is to the meaning of the content word, the closer it is placed to the content word. If you find that a Notion is marked in several places, at first mark it only in the position closest to the relevant content word. OP (POSITION): OPERATORS. If a functor operates on a whole structure (phrase or clause), try to place it external to that structure, leaving the structure itself unchanged.

GENERAL PROBLEM-SOLVING STRATEGIES Review Strategies OP (REVIEW): STRENGTHENING. Whenever an attempted solution succeeds, apply the same strategies to similar problems. OP (REVIEW): MONITORING. Compare utterances you hear with forms that you would produce in the same situation. Store mismatches and attempt to accommodate your grammar to unassimilated input forms by applying relevant OPs to the area of mismatch. If you find that a given OP is responsible for the mismatch, attempt to change the level of operation of that OP, or cease to use it with regard to the area of mismatch.

603

OP (REVIEW): PERSISTENCE. Whenever an OP fails to achieve a solution, and whenever a previous solution is found to be inadequate, return to the task from time to time until a solution is achieved. OP (REVIEW): SEMANTIC REORGANIZATION. Continually reexamine Semantic Space as your cognitive structures develop and as you discover the semantic categories used in your parental language. Analyze Notions into related subordinate Notions and group Notions into new superordinate patterns; make semantic extensions of Notions and assimilate Notions to one another. Apply new analyses to existing semantic paradigms and definitions of functors and grammatical operations. OP (REVIEW): DICTIONARY REORGANIZATION. Review stored linguistic forms and their meanings, systematizing your Dictionary according to available OPs and the principles of linguistic organization that have proven useful in dealing with your language. OP (REVIEW): LIMITED FUNCTIONS. At first apply a solution to the smallest motivated category and do not extend it without evidence.

Interim Production Strategies OP (PRODUCTION): UNINTERPRETED FORMS. If a speech element is frequent and perceptually salient, but has no obvious semantic or pragmatic function, use it in its salient form and position until you discover its function; otherwise do not use it. 604

OP (PRODUCTION): ROTE. If a form appears frequently enough for you to have memorized it and made its use automatic, continue to use it in that manner while you are systematizing the use of similar but less automatized forms. OP (PRODUCTION): MAXIMAL SUBSTANCE. While you are mastering the linguistic expression of a Notion, mark that Notion with as much acoustic substance as possible, with maximal phonological separation of the form in question from adjacent speech units.

605

16

What Shapes Children’s Grammars?

Melissa Bowerman Max-Planck Institute for Psycholinguistics Nijmegen, The Netherlands and University of Kansas

Contents The Processing Approach in Historical Perspective Some Methodological Problems The Time Lag Between Intention and Mastery Errors in Spontaneous Speech Testing Proposed Operating Principles Dealing with Findings Counter to Predictions

606

Choosing Among Alternative Explanations From a List of Operating Principles to a Theory of Language Acquisition Basic Child Grammar: Is Only One Way of Structuring Semantic Space “Basic”? Unbiased “Opening Wedge” Language-Specific Learning in Hierarchically Organized Domains

Often

Impossible:

Unbiased “Opening Wedge” Often Possible: Overlapping Semantic Categories Finding the Balance Between Flexibility and Constraint in Starting Semantic Space Priorities for Future Crosslinguistic Research

Crosslinguistic comparisons have been used systematically in the study of the human capacity for language acquisition for only a short time, less than 20 years. But research has been intensive in this period. The present volumes are a landmark in the continuing development of this essential research tool in part because they bring together a wealth of information about the acquisition of various languages, most of it quite recent. Beyond this, however, they are important because they represent the state-of-the-art in the vigorous program of crosslinguistic research and theorizing launched over 10 years ago by Dan Slobin in 607

his seminal 1973 article, “Cognitive prerequisites for the development of grammar.” It is a good time to take stock of how far we have come with this approach, what its strengths and weaknesses are, and where we should go with it next. This chapter is devoted to these questions. The discussion is organized as follows. First I try to place Slobin’s approach in a historical perspective, since to appreciate the conceptual and methodological impact of “Cognitive prerequisites …” on research of the last decade it is necessary to recall what preceded it. In the next three sections I critically evaluate several important aspects of the approach as it is represented in these volumes and especially in Slobin’s integrative chapter, Crosslinguistic evidence for the Language Making Capacity”: its methods of analysis, its overall coherence as a theory of language acquisition, and the viability of the characterizations of early child language to which it has given rise. I end the chapter with some brief suggestions for further crosslinguistic research. A recurrent theme throughout these discussions is to find the right balance between the contribution of the child and the contribution of the linguistic input to the acquisition of language. In particular, I explore the problem of determining to what extent children’s early grammars are shaped by universal, inherent preferences for certain ways of structuring meaning and mapping it into linguistic form, and to what extent by experience with the semantic and structural properties of the language being learned.

608

1. THE PROCESSING APPROACH IN HISTORICAL PERSPECTIVE In the years immediately leading up to the 1960s, investigators typically viewed child language through a filter provided by adult language. Children’s progress was described in terms of deficiencies or improvements with respect to the categories and rules of the full-fledged system. This changed in the early ’60s when researchers, inspired in part by Chomsky’s claim that language acquisition is a dynamic process of formulating abstract rules, began work aimed at characterizing child language in its own terms. Empirical investigations into children’s rule systems were embedded in a broader theoretical concern: the nature of the human capacity for language acquisition. Chomsky (1959, 1965) had drawn special attention to this problem with his compelling arguments that the behaviorist approach, accepted by many scholars at that time, was grossly inadequate, and by his strong counterproposal that children come to the language learning task equipped with inborn knowledge of both substantive and formal linguistic universals. If children indeed have innate knowledge that guides and constrains their formulation of linguistic rules, it was reasonable to expect similarities in language acquisition across children learning different languages. Researchers therefore began to explore whether the regularities that had been documented primarily in the speech of children learning English would prove to be general

609

(e.g. Bowerman, 1973; Brown, 1973; Slobin, 1970). It was hoped that crosslinguistic commonalities, where found, would reveal general characteristics of the child’s language acquisition capacity. Although this approach to crosslinguistic research was methodologically sound, it had an important limitation. Testing whether certain properties are common to the speech of all children required hypotheses about potential universal of form, content, or function. Three such hypotheses of the 1960s were “telegraphic speech,” the “pivot grammar,” and “adherence to rigid word order.” If crosslinguistic testing disconfirmed universality—as it in fact has done for these once-promising characterizations—it was unclear where to look next. Meanwhile, data that were language-specific—hence not relevant in any obvious way to the testing of universal patterns—went unexploited, since it was uncertain with which data from other languages they could be compared to yield interesting generalizations. A crucial contribution of Slobin’s “Cognitive prerequisites … ” was to show a way beyond this sticking point. The fundamental insight was a new conception of what could serve as a basis for drawing inferences about the language acquisition capacity. Slobin proposed that, in addition to looking for universal stages of acquisition, researchers could use crosslinguistic data to determine the relative difficulty for children of the formal linguistic devices that languages employ to express meaning, such as prefixes, suffixes, and word order. Patterns of relative difficulty 610

could then provide clues to how children approach the linguistic input in constructing a grammar. Slobin phrased his preliminary conclusions about how children proceed in the form of “Operating Principles”: children’s self-instructions for perceiving and producing speech under short-term processing constraints and for organizing and storing linguistic rules. This notion had been foreshadowed much earlier by Slobin in a 1966 article. The discussion there concerned whether children should be credited with innate knowledge of categories and possible rules of grammar, as Chomsky (1965), McNeill (1966) and others had advocated. Slobin proposed that children may start out equipped not with “knowledge” but with a set of procedures for analyzing linguistic input. Language universals, in this view, would reflect the interaction of the analytical procedures with the input, but would not in themselves be innately represented. This approach (also suggested by Fodor, 1966) became known as the “process” approach to language universals. The orientation of Chomsky and McNeill, which emphasized substantive and formal universals, was termed the “content” approach. Although a sharp distinction between process and content is no longer maintained (some researchers argue that these are merely notational variants of each other), modern descendents of the two approaches can still be identified. The work discussed in these volumes clearly carries on the “process” tradition, while work by scholars such as Goodluck (1981), Otsu (1981), Roeper (1982), and others (e.g. in Tavakolian, 1981) elaborates the “content” approach. An 611

important difference between the two orientations is in the methods typically used; these in turn reflect differences in underlying assumptions and in the primary method of reasoning. “Content”-oriented researchers usually proceed deductively, starting from hypothesized universal constraints on the form of grammars, making predictions about how language–learning children should behave if these constraints are innate, and devising experiments to test these predictions. Comprehension is the performance modality usually studied, and research has only rarely involved crosslinguistic comparisons. “Process”-minded researchers, in contrast, have typically been either skeptical about whether there are a priori constraints on the child’s development of grammar or agnostic about what such constraints might be like. They have therefore proceeded inductively, examining the details of children’s spontaneous speech at successive phases of development and trying to infer general principles that could give rise to the observed patterns. Although a researcher may work on only one language, crosslinguistic comparisons are considered essential to allow general procedures of the language-learning child to be disentangled from those that result from exposure to a language with particular structural properties (or transmitted in a particular social milieu, etc.). Experimental work is sometimes done, but the purpose is not usually to test predictions but rather to gather more controlled information on how children learning different languages proceed in a given structural domain (e.g. Johnston & Slobin, 1979), or to decide between 612

alternative explanations for children’s behavior by comparing children learning languages that differ structurally in key ways (e.g. Hakuta, 1981). One of the most important consequences of Slobin’s Operating Principles (OP) approach has been its stimulating effect on researchers around the world. The approach has both inspired further data collection and focussed attention on certain ways to interpret the data. When the immediate goal of crosslinguistic comparison is seen as to determine the relative difficulty for children of different formal devices for expressing meaning, the usable data base is much broader than when the goal is seen as to establish universal “stages” of development. Every piece of evidence may contain a clue; none is too small or too language-specific to take its place in the larger pattern of explanation. In response to new information, further theoretical work, and suggestions from contributors to these volumes, Slobin (1985) has modified and extended the Operating Principles model. It has become more comprehensive both in the number of factors it considers relevant to acquisition and in the variety of linguistic structures whose acquisition it seeks to explain. There is also now an important new hypothesis about the initial outcome of the interaction between Operating Principles and linguistic input: the notion of “Basic Child Grammar.”1 We can anticipate further expansion and reformulation, of course, but the approach has by now developed clear enough outlines to allow some serious evaluation.

613

In what follows, I focus on those aspects of the approach that I believe present the toughest challenges and will need the closest attention in the coming years if the OP approach is to realize its promise. I begin with some remarks about methodology, and then turn in more detail to questions about testing proposed OPs, about how the OP approach hangs together as a whole, and about the “Basic Child Grammar” hypothesis.

2. SOME METHODOLOGICAL PROBLEMS 2.1. The Time Lag Between Intention and Mastery Within the OP framework, as it was conceived by Slobin in 1973, it is essential to be able to establish the relative acquisitional difficulty of various linguistic devices, since this is the primary basis on which inferences about the language acquisition capacity are drawn. Children’s spontaneous speech offers a rich variety of clues, but it is no small task to bring this uncontrolled data source into some kind of order. In “Cognitive prerequisites …” Slobin proposed an ingenious new technique for dealing with this problem. It depended on a “very strong developmental psycholinguistic universal … [that] the rate and order of development of the semantic notions expressed by language are fairly constant across children learning different languages, regardless of the formal means of expression employed” (Slobin, 1973, p. 187). If this universal is true, observed Slobin, and if the onset of the

614

child’s intention to communicate given meanings can be identified, then the time lag from intention to mastery of the conventional form can be taken as an index to how “complex”—i.e. how difficult on formal linguistic grounds those forms are for children. The relative complexity of different forms (e.g. prefixes vs. suffixes) can then be established by comparing how long it takes children learning different languages to move from first intention to communicate a meaning to mastery of the forms used to express that meaning. 2.1.1. Problems in Measuring the Time Lag. Despite the intuitive appeal of this technique, its primary contribution in the long run may turn out to have been conceptual rather than methodological: It has forced researchers to distinguish clearly between meanings—and the cognitive prerequisites for these meanings— and their somewhat arbitrary formal linguistic packaging. This clarification has led to progress in many domains. But applying the method in the way Slobin initially envisioned has proved difficult. First, identifying the time of “first intention” to express a given meaning is often difficult or impossible (see Johnston, 1985). Second, establishing the time of mastery of a form is no easier. As Clark (1985) points out, measures of mastery based on children’s production sometimes differ from those based on comprehension. Which measure should we then use? A further problem that affects our assessment of both intentions and the mastery of forms is the possibility of reorganization in the child’s linguistic system. As recent research 615

emphasizes (e.g. Bowerman, 1982b, 1982c; Karmiloff-Smith, 1979a, 1979b), children’s initial appropriate uses of a form are often based on relatively superficial knowledge; “acquisition” beyond this point may be a drawn-out process in which the child discovers successively deeper levels of structure and regularity at the levels of both form and meaning. Where in the span of weeks or months beyond the first appearance of a form should the point of “acquisition” be set? Given these difficulties in setting the lower and upper boundaries of the lag between intention and mastery, it is not surprising that the relative difficulty of linguistic forms has in practice often been estimated simply on the basis of the relative ages at which children learning different languages begin to use the conventional forms to express a given meaning. This substitute might seem harmless. However, it has an important drawback: It provides no check on the assumption that given communicative intentions arise in all children at about the same age regardless of their local language. If in fact properties of the language being learned systematically influence the time at which given intentions emerge (a possibility also recognized by Slobin, 1982), time of acquisition per se cannot provide an unambiguous guide to linguistic complexity.2 This problem, especially when coupled with the common practice of estimating “average time of acquisition” for a whole population on the basis of data from a small number of children, makes “time of acquisition” information difficult to interpret. 2.1.2. Confounded Determinants of Time of Acquisition. As we have noted, the logic of the OP approach to 616

assessing relative linguistic complexity depends critically on the assumption that meanings arise in the child independently of knowledge of the forms with which to express them. A further crucial assumption is that the time of acquisition of a form is influenced by only two factors: the difficulty of the meaning it expresses and its formal complexity for children. Given these assumptions, we can reason that if meaning is held constant and time of acquisition varies across languages, then difficulty of form must be the determining factor.3 The problem, of course, is that cognitive and linguistic complexity are not the only two factors that can influence time of acquisition, as research of recent years has made increasingly clear. For example, in “The Language-Making Capacity,” Slobin (1985) discusses evidence for the role of two additional factors: the relative pragmatic usefulness for the child of different structures (see also Eisenberg, 1981) and differences in frequency of modeling.4 The list may well become longer. Although recognizing additional influences on time of acquisition is not objectionable on general theoretical grounds, it destroys the logic of the OP procedure for determining formal linguistic complexity. Unless we can figure out how to isolate the contribution of each determinant to the time at which a linguistic ability is acquired, then relative time of acquisition can serve as a guide neither to relative formal complexity nor to relative cognitive complexity.

617

2.2. Errors in Spontaneous Speech Because of the difficulties of measuring and interpreting the time needed for acquiring a form, a second source of evidence for relative complexity has gradually increased in importance: errors in children’s spontaneous speech. In fact, as Slobin points out, “error analysis forms the substance of most of our crosslinguistic developmental data” (1985, p. 206). The analysis of errors is accepted as an essential research technique by almost all researchers in child language, but it is sometimes criticized by those in neighboring fields (e.g. see Givón, 1985). Certain criticisms, I think, reflect a misunderstanding of the purpose of error analysis rather than real problems. On the other hand, there are some serious difficulties in analyzing children’s errors that are rarely discussed. Let me try to distinguish between these. 2.2.1. Prescriptivism and Other Nonproblems. The most common charge against the study of children’s errors is that it implies a prescriptive attitude: Adult speech is taken as “correct” and child speech is interpreted only in terms of its “deficiencies” with respect to this standard. This is seen as a failure to recognize or respect the integrity of the child’s own linguistic system, in which departures from adult grammaticality may not be errors at all. This complaint reflects a fundamental misconception. The motivation behind virtually all contemporary analyses of children’s speech errors is precisely to understand the structure and functioning of the child’s own system. Children’s

618

failures to meet adult standards are of no inherent interest; their value is rather that they are one of the most compelling sources of evidence we have about the workings of the system we are trying to understand. This is because, by virtue of their discrepancy from the input, they implicate children’s own efforts to process and organize what they have heard. When children’s utterances are unremarkable by the norms of their speech community (regardless of whether the language being learned is a “standard” or “substandard” dialect), it is hard to infer the knowledge that underlies them. First, superficially correct usage can be compatible either with adultlike knowledge or with a more constrained or superficial understanding (e.g. a highly context-bound grasp of word meaning, or an unanalyzed or only partially analyzed representation of strings that for adults have complex internal structure). Second, even when children have demonstrably analyzed strings of a certain type (i.e. control all the morphemes in them as independent elements), more than one description of their structural knowledge is still almost always possible (just as in the case of adult speech). We cannot determine which description has psychological reality at a given stage of development—i.e. captures the categories and rules that are functional in the child’s linguistic system—solely by looking at conventionally acceptable output. Errors, however, if carefully observed and fully exploited, can provide important clues.5 Givón (1985) warns that “conceiving of children’s output as ‘error’ may lead to de-emphasizing what is systematic and universal about [their language], in favor 619

of over-emphasizing what is idiosyncratic, language-specific or irregular” (p. 748). Although this might in principle seem a danger, in fact, the opposite is nearer the truth: Researchers have typically used error patterns within particular languages to infer universal predispositions concerning either the formal structure of language, its semantic/conceptual underpinnings, or both (e.g. see Berman, 1980; Bowerman, 1982b, 1982c, 1983a; Clark, 1977, 1981; Karmiloff-Smith, 1978, 1979a; MacWhinney, 1978, 1985; and Slobin, 1973, 1985, to name only a few; see also footnote 19). 2.2.2. Interpreting Children’s Errors. What then are the more serious difficulties in using error data? One requirement is to become as sensitive to errors of omission as we are to errors of commission (see Brown, Cazden, & Bellugi, 1968, on these terms). Errors of commission are departures from adult norms that, like foots and goed, are explicitly deviant. Errors of ommission, in contrast, are restrictions in the distributional range of a form. Interesting examples discussed by Slobin (1985) include the initial limitation of markers of subject and direct object to highly transitive agentive events. Errors of commission are much easier to spot than errors of omission, and they have received far more attention in the literature. A child’s apparent failure to use a form over its full range is not only unobtrusive but also hard to verify: The “missing” uses may simply be infrequent and therefore not sampled, rather than absent entirely. Nevertheless, information about restricted usage, if carefully evaluated, can contribute just as much as outright errors to our

620

inferences procedures.

about

children’s

language-learning

It is just as essential to evaluate suspected errors of commission carefully. As Ochs (1985) and Smoczyńska (1985) point out, forms that our grammar books indicate are unacceptable in language X may in fact be common in the everyday speech children hear. If so, no special reference to children’s own processing or organizational dispositions is required to explain them. Especially when dealing with languages or dialects not our own, we have to study the actual input, not our idea of it. Even when we are confident that certain utterances really are deviant by the norms of the child’s speech community, lack of information almost always hampers us in trying to interpret them. First, have errors of this type been documented repeatedly in the speech of an individual child and in other children learning the same language, or have they been observed only once or twice in all? Sometimes elaborate arguments are made on the basis of a couple of utterances. Almost anything can happen once or twice. For errors, just as for conventional utterances, we cannot base inferences on single exemplars; we need to know about patterns of usage, both within and across children. Second, once it has been established that errors of a certain type occur repeatedly, we must ask how frequent they are relative to correct usage (see also Smoczyńska, 1985). When children are rarely or never correct in their use of a certain form (word, sentence pattern, etc.), and if their errors are consistent, it is reasonable to attribute the 621

errors to characteristics of the grammatical rules the speaker has formulated. However, some types of errors, although recurrent, are infrequent relative to correct usage: Most of the time the child uses the forms involved correctly. Such “occasional” errors should not be taken as clues to the speaker’s grammar; they are usually better accounted for by reference to performance factors (although the lines along which occasional errors are made can reveal important principles of underlying organization; we will come back to this later). Third, is the error pattern observable from the child’s first uses of the affected forms, or does it set in only after a period of correct usage? (See Bowerman, 1978, 1982a, 1982b, 1982c on the phenomenon of “late” errors.) Fourth, are there other forms in the child’s repertoire that may compete with or otherwise interact with a target form, causing errors? And fifth, if an explanation for an error type has been hypothesized, do other forms in the child’s speech that are in principle vulnerable to the same source of trouble also show errors? Answers to all these questions can be essential for deciding among alternative explanations for the errors under study, as will become clear from examples to be discussed. And how we interpret the errors, in turn, has far-reaching consequences for determining what the child brings to the language acquisition task. These topics are pursued in the following sections.

622

3. TESTING PROPOSED OPERATING PRINCIPLES One of the most salient features of the OP approach is its capacity to assimilate and make sense of new data of a variety of types. This is an important source of its appeal, but it is also a danger. One drawback is that because of its “tolerance,” the model tends to discourage testing: Researchers over the last years have often been content with pointing out “hits”—ways in which their data accord with one or another operating principle—and have failed to search as assiduously for “misses.” More seriously, even when researchers do adopt a hypothesis testing attitude—and contributors to this volume were explicitly encouraged to do so6—the most basic claims of the model often turn out to be difficult to test. This is because many, perhaps even most, types of counterevidence can be readily dealt with either by reinterpreting the data to fit existing OPs, by simply expanding the OPs, or by adding in new ones, leaving the basic framework undisturbed. This gives rise to a perplexing state of affairs. Naturally we want a theory that can account for all observed outcomes. And the suggested amendments and additions to the theory proposed in Slobin (1985) are generally quite plausible. But the overall effect is that the theory becomes increasingly enclosed and self-protecting, and it becomes harder to test whether observed outcomes really confirm the processes hypothesized to underlie them or simply are not incompatible with them.

623

A related problem is that of redundancy. As more OPs are added, both to account for counterinstances and to handle phenomena not previously dealt with, the OPs begin to encroach on each other’s territories. As the system gets heavier, it sometimes becomes possible to account for a single phenomenon in more than one way. Again, this is often plausible: Certain outcomes may indeed be multiply determined. But this is probably not always so, and we need to become sensitive to redundancy and to develop principled methods for deciding among alternative explanations and for weeding out those that, however attractive, are superfluous. The examples with which I illustrate these problems revolve around one of the most difficult, frustrating, and fascinating issues in the study of child language, the role of meaning in the child’s construction of grammar.

3.1. Dealing with Findings Counter to Predictions One of the central tenets of the OP approach since its 1973 inception is that children initially strive for a clear (one-to-one) mapping between units of meaning and units of form. The OPs in Slobin (1973) designed to reflect this tendency were E: “Underlying semantic relations should be marked overtly and clearly,” and G: “The use of grammatical markers should make semantic sense.” These two are missing from the current formulation of the OP approach (Slobin, 1985), but others designed to capture the same propensity have grown up in their place. It is with respect to children’s assumed preference for clear mapping that the OP 624

approach has become particularly well defended against disconfirmation. 3.1.1. The Use of Grammatical Markers Should Make Semantic Sense. In her chapter on the acquisition of Japanese, Clancy (1985) notes that young children initially add case particles after nouns in a semantically and pragmatically unmotivated way. She observes that this goes counter to the OP that “The use of grammatical markers should make semantic sense,” and suggests that children may sometimes override this OP in their efforts to create a canonical surface structure with its characteristic intonation pattern. On the basis of this and other similar evidence, Slobin proposes a new OP called UNINTERPRETED FORMS: “If a speech element is frequent and perceptually salient, but has no obvious semantic or pragmatic function, use it in its salient form and position until you discover its function; otherwise, do not use it” (1985, p. 246). This OP functions alongside OP:FUNCTORS (a descendent of “The use of grammatical markers should make semantic sense”), which specifies (in brief) that the child tries to map uninterpretable speech segments (after extraction of content words) onto grammaticizable “Notions” (meanings). These two OPs together form a closed circle: If the child uses a grammatical marker with a consistent meaning he is following OP:FUNCTORS, and if he uses it without meaning or does not use it at all, he is following OP: UNINTERPRETED FORMS. The substantive claim made earlier in “The use of grammatical markers should 625

make semantic sense” has become greatly attenuated. However, it survives at least in the implication that children will first search for a semantic or pragmatic basis for grammatical markers, and only switch to OP: UNINTERPRETED FORMS if they fail. This sequence is also proposed, in more explicit form, in OP: MORPHOLOGICAL PARADIGMS. Do children in fact always first try to find semantic or pragmatic solutions and only use certain items meaninglessly if they cannot do so? There seems to be no independent evidence for this.7 Without such evidence we can just as well imagine an inverted OP that says “Use any (noncontent) forms that you have noticed (because they are salient, frequent, etc.) freely in their usual positions until it occurs to you that they have an associated meaning; after that, use them only in conjunction with this meaning.” When we can so easily restate an OP as its converse, we must question whether it captures children’s grammar-constructing procedures accurately. 3.1.2. Plurifunctionality. Another test of the assumption that children initially strive for a clear mapping between forms and meanings comes with plurifunctionality. If interpreted strictly, the one-to-one mapping hypothesis predicts (among other things) that children should not use forms that are plurifunctional—that express different meanings on different occasions. In fact, we know that they often do. Does this mean the hypothesis is wrong? Not necessarily. There are at least two ways to

626

reinterpret counterinstances to be compatible with one-to-one mapping. First, as Slobin points out, what for adults are two distinct meanings may for children be undifferentiated, hence, only “one meaning.” He suggests that if we accept the principle of one-to-one mapping as given, it “guides us in discovering what constitutes ‘one meaning’ and ‘one form’ from the child’s point of view” (1985, p. 207). Second, as Karmiloff-Smith (1978, 1979a) has argued, children who in general adhere to one-to-one mapping could nevertheless use the same form for what they perceive as two distinct meanings as long as they do not yet recognize that the form is “the same” in these disparate uses.8 In this case, when they do realize that one form is serving two functions, they should begin to mark these meanings with distinct, often ungrammatical forms. Karmiloff-Smith gives several examples of this phenomenon, and Slobin, guided in part by her arguments, proposes OP:UNIFUNCTIONALITY: “If you discover that a linguistic form expresses two closely related but distinguishable Notions, use available means in your language to distinctly mark the two Notions.” There is nothing inherently implausible about either of these reinterpretations of children’s behavior to make it compatible with the principle of one-to-one mapping, but the argumentation is circular. We assume that children do not like purifunctionality. When they do use forms plurifunctionally, however, we do not count it as

627

evidence against this assumption. It just means that they do not yet realize that they are using one form for two meanings. If and when they do discover this, they will try to mark the meanings distinctly. Clearly we need to break out of this circle by finding methods independent of how children use language for determining whether they differentiate certain meanings and whether they recognize that certain forms are “the same.” This will not be easy, unfortunately, since the relevant children are extremely young, the meanings are usually subtle, and both the meanings and the forms, because they often belong to the “grammaticized” rather than the “content” portion of language, are particularly inaccessible to conscious reflection. If we widen the data base for evaluating OP:UNIFUNCTIONALITY, we find other reasons for worry: on the one hand the absence of errors that should occur if the OP is really functioning, and on the other hand the presence of errors that should not occur. With respect to nonoccurrence of predicted errors, it is important to recognize that every language is riddled with forms that are plurifunctional, at least by crosslinguistic standards. That is, they apply indiscriminately across meaning distinctions that other languages require speakers to make. If children were really concerned with eliminating plurifunctionality, we could expect many more errors than have in fact been reported.9 Of course, even though errors are not reported, they may still occur. But when we set the tiny handful of error types of which we are aware against the 628

vast number of structures in which children could be predicted to make errors according to OP:UNIFUNCTIONALITY, we must wonder why there are so few. What about errors that would not be expected to occur? OP:UNIFUNCTIONALITY is designed to account for “late” errors in children’s speech that, in effect, give separate marking to two closely related notions that previously were (correctly) expressed by the same form. Another genre of “late” errors is essentially the opposite: after a long period of using certain forms separately and accurately for distinct but related meanings, children begin to substitute these forms for each other occasionally in violation of the dividing line between their meanings (Bowerman, 1978, 1982c, 1983a). Examples in English include mutual substitutions of make and let (active versus permissive causation, respectively; e.g. MAKE me watch TV(=let; begging for permission) and I don’t want to go to bed yet. Don’t LET me go to bed (=make)); of put and give (cause-change-of-location to inanimate versus animate goal, respectively; e.g. We’re PUTTING our things to you ( =giving) and GIVE some ice in here (=put; into ice-crusher)); and of put and make (cause-change-of-location versus cause-change-of-state, respectively; e.g. I PUT it brown (=made; after child colors skunk’s stripe brown), MAKE them back up (=PUT: request to have dolls that had fallen off table put back). Errors like these are made at about the same age and with roughly the same frequency as errors that have been 629

used in support of OP:UNIFUNCTIONALITY (both are infrequent, relative to correct usage). It is unclear why they should occur if children prefer unifunctionality and try to eliminate plurifunctionality, since they result in the blurring of meaning distinctions that previously were observed. A possible explanation is Slobin’s (1985) suggestion that “Basic Notions” sometimes “persist”: he uses the put/give confusions to illustrate the continuation of an earlier tendency to collapse the distinction between possessors and locative goals. One problem with this explanation is that it does not account for why the errors do not occur from the child’s first uses of the forms involved. But more important for present purposes is the methodological problem it raises. Taken together, OP:UNIFUNCTIONALITY and “persistence” define a closed system: errors showing the “splitting” of semantic notions constitute evidence for the striving for one-to-one mapping, but comparable errors showing the “lumping” of notions do not constitute counterevidence because they can be explained in some other way. I return later to the possible causes of “splitting” and “lumping” errors.

3.2. Choosing Among Alternative Explanations When every outcome can be explained but the explanatory principles buttress each other and so resist disconfirmation, it is time to look hard at whether the principles are the right ones or whether others would not do the job better. Efforts to home in on the right explanations become even more essential when there is more than one way to interpret given findings. Alternative explanations can in some cases be found 630

within the set of OPs proposed by Slobin (1985), and the possibilities expand when we consider other researchers’ efforts to account for the same data, e.g. the set of OPs offered by MacWhinney (1985). Although the problem of multiple explanations is endemic to social science research, it is particularly serious for the OP approach, as it is developed by Slobin (1985), for the following reason. A pervasive feature of the approach, as noted earlier, is that children’s errors are interpreted wherever possible in terms of the learner’s predispositions or preferences for how meanings should be mapped into linguistic forms. Slobin’s hypotheses about these “mapping preferences,” as I will call them, are interesting and generally plausible. However, because they postulate an intrinsic mode of organization, independent of children’s experience with any actual language, they require particularly strong justification. In my view, the need for OPs specifying inherent mapping preferences has not yet been satisfactorily demonstrated because we have not yet thoroughly explored alternative, more everyday interpretations of the phenomena that such OPs are designed to explain—in particular, explanations that focus on children’s experience with the structural properties of the language being learned. Forcing ourselves to pose alternatives to our favorite explanations is tedious, and we often lack the data to do careful evaluations. However, in making the effort we can pinpoint problem spots where data of a certain type are needed, and perhaps guide future research to come up with the critical information. Whatever conclusions

631

we then reach about the nature of the language acquisition capacity will be correspondingly stronger. In the previous section I discussed some difficulties with OP:UNIFUNCTIONALITY, which specifies that children try to mark separate meanings with separate forms. Let us now look closely at certain other OPs that are concerned with how meaning is expressed, paying special attention to alternative explanations of the phenomena on which they are based. 3.2.1. Explanations that invoke “mapping preferences’’. Consider the following error types: 1. The use of inflections or derivational markers on words that already encode the desired meaning, e.g. feets, ated (Kuczaj, 1978), to smoothen (=to smooth), to unopen (=open), to untake off (=take off) (Bowerman, 1981, 1984); in Turkish, the use of the causative derivational suffix on verbs that are already inherently causative (Aksu–Koç & Slobin, 1985). 2. Constructions containing two or more bound or free forms that express essentially the same meaning but that in adult speech cannot be combined in a single sentence, e.g. Mon mien de chapeau à moi ‘my mine of hat to me’ (=my hat) (Clark, 1985), He never won’t scare me (Bellugi, 1967). 3. Constructions that, although grammatical, are more analytic than adults would use in the same speech contexts, e.g. make die or make dead for kill, make my shoe come on my foot for put my shoe on my foot 632

(Bowerman, 1982a, 1982c), I will for I’ll (Bellugi, 1967). 4. The separate expression of functors that, when their meanings are juxtaposed, should be expressed with fused, portmanteau forms, e.g. de le ‘of the’ for du in French (Clark, 1985), and al hu ‘on he’ for alav ‘on him’ in Hebrew (Berman, 1985). 5. Constructions in which a functor is misordered or mentioned twice. The Polish conditional offers particularly nice examples of these (Smocyńska, 1985). In the adult system, the personal ending is attached to the conditional particle by, which can either follow the verb, precede the verb as an enclitic, or constitute part of the connective żeby ‘in order that’. Children make errors of the following types: a. Personal ending attached to verb instead of BY: pisał-(e)m-BY ‘wrote-lSG-WOULD’ or ja BY pisał-em ‘I WOULD wrote-1SG’ for pisał-BY-m ‘wrote-WOULD-1SG’ or ja BY-m pisał ‘I WOULD-1SG wrote’ (I would write). b. Personal ending redundantly marked on both verb and BY: pisa?-(e)m-BY-m ‘wrote-1SG-WOULD-1SG’. c. Conditional particle redundantly placed both before and after verb: a moja mamusia tez BY mia#x0142;a-BY iadne w#x0142;oski ‘and my mommy also WOULD have-WOULD pretty hair’.

633

Slobin (1985) explains errors types 1–4 and 5b,c by reference to children’s striving for the explicit marking of meaning. The phenomenon can be covered by either OP:ANALYTIC FORM (“If you discover that a complex Notion can be expressed by a single unitary form (synthetic expression) or by a combination of several separate forms (analytic expression), prefer the analytic expression”) or by OP:MAXIMAL SUBSTANCE (“While you are mastering the linguistic expression of a Notion, mark that Notion with as much acoustic substance as possible … “). Slobin posits both of these because, as he notes, there is ambiguity about whether redundant and overly explicit marking reflect how children try to construct their grammars (as in OP:ANALYTIC FORM) or is simply an “interim production strategy” that children employ in the early phases of getting productive control over a form or construction (as in OP:MAXIMAL SUBSTANCE). For error type 5a and related constructions, appeal is made to a different mapping predisposition, the notion of relevance, which was inspired by Bybee’s (1983, 1985) finding that there is strong crosslinguistic consistency in the relative ordering with respect to the verb stem of markers for aspect, tense, person, etc. The hypothesis is that children have inherent preferences for which kinds of meaning ‘ ‘belong together’’ and should be expressed by forms in close proximity to each other. Two OPs are designed to capture this, and either one can account for error type 5a. OP:RELEVANCE states that “If two or more functors apply to a content word, try to place them so that the more relevant the meaning of a functor is to the meaning of the content word, the closer it is placed to 634

the content word….” This OP is invoked to explain why Polish children often attach the personal ending to the verb rather than to the conditional particle: Markers of person are more relevant to verbs. OP:OPERATORS states that “If a functor operates on a whole structure (phrase or clause), try to place it external to that structure, leaving the structure itself unchanged.” This one accounts for the same error in terms of children’s preferences for handling the conditional particle rather than the personal ending: as a clause-level operator, it should not disturb the rest of the clause (e.g. by receiving the personal ending, which in all nonconditional sentences goes on the verb), and it should ideally be clause-external. Of course, these two “relevance” principles do not conflict but in fact could work together to co-determine this particular error type. 3.2.2. Competition. All of these explanations are plausible, and they have a certain intuitive appeal to cognitively minded researchers. But the overall effect is somewhat piecemeal; for example, error types that seem to be closely related (e.g. 5a versus 5b and c) are sometimes explained with completely different principles. A more integrated proposal to account for errors of types 1–5 invokes a different kind of processing principle: competition among different methods for expressing the same or closely related meanings (Bowerman, 1978, 1981, 1984; MacWhinney, 1978, and especially his detailed treatment in this volume). Competition is seen as a process that takes place at the moment of speech, when the speaker’s intention to 635

express a certain meaning activates more than one linguistic device associated with this meaning. The conflict is usually resolved by implicit pre-speech editing, but when it is not, errors occur.10 The type of error depends both on what has been competing with what and on the relative activation strengths of the rival forms. Sometimes two or more linguistic devices are inappropriately combined within a single word (error type 1 above, e.g. plurality is double-marked in feets) or within a single sentence (error type 2, e.g. negation is double-marked in He never won’t scare me). Second, analytic variants may win out in contexts where adults prefer synthetic variants or vice versa (error type 3, e.g. I will … for I’ll … ). Third, when two forms cannot combine with each other but must be replaced by a synthetic form (e.g. du for de le ‘of the’ in French), the two forms may nevertheless each have such high activation strength as the preferred method of expressing their particular meaning that they are inappropriately selected together (error type 4). Competition accounts for errors 5a–c in terms of the activation of more than one possible position for a given morpheme (MacWhinney, 1985). In 5c the conditional particle is redundantly expressed in two separate positions. Both positions are activated because both are acceptable in Polish; errors that result from selecting them simultaneously are analogous to common errors of English speakers like Pick up your socks up. In 5a the personal ending is inappropriately attached to the verb rather than to the conditional particle, and in 5b it occurs on both the verb and the particle. Both positions are activated 636

because, although in conditional sentences the ending should go on the particle, in all nonconditional sentences it goes on the verb. Notice that, in contrast to Slobin’s account of error 5a with “relevance” OPs, the competition account makes no assumption that Polish children have an inherent preference, independent of experience with the structure of Polish, to put personal endings on verbs rather than elsewhere, or to keep conditional markers from interacting with the rest of the clause. To researchers interested in the role of meaning in language acquisition, competition might at first appear to be a mechanical approach that “explains away” children’s errors as uninteresting or irrelevant to the development of a theory of language acquisition without offering anything constructive in return. Fortunately, this is not true. It is true that competition is a very general cognitive principle, applicable to many behaviors in addition to language production (e.g. see Norman, 1981). However, when it is invoked in explaining phenomena in any particular behavioral domain, competition offers an unparalleled opportunity to discover underlying structuring principles that are difficult to get at in other ways. The important clues lie in what competes with what. Competing forms have been activated by the same underlying intention, whether the actor’s purpose is linguistic expression or some other goal. By observing which linguistic forms compete and interfere with each other at successive points in children’s development, and inferring the semantic and grammatical bases for this competition, we can achieve a deeper understanding of how meaning is structured for young children, how the 637

organization of meaning changes and develops over time, and how children develop an implicit sense of the way the various parts of the linguistic system are interrelated. 3.2.3. “Mapping preferences” versus “competition”: Which account to preference Neither Slobin nor MacWhinney regards “mapping preferences” and “competition” as mutually exclusive explanations. MacWhinney allows for the role of “relevance”-like predispositions, along with competition, in promoting certain ordering errors. Similarly, Slobin suggests that certain ordering and over-marking errors reflecting OP:RELEVANCE, OP:OPERATORS, or OP:MAXIMAL SUBSTANCE are co-determined by OP:MORPHEME PLACEMENT: “Mark a notion in the same place in the various constructions in which it can occur….” This OP is a version of the “competition” principle according to which the position associated with a certain morpheme (or class of morphemes) in sentences of one type can have high enough activation strength to attract that morpheme even in sentences where another position is required. Slobin and MacWhinney do not take up the problem of whether both accounts are really necessary, and, if so, which one applies where. In evaluating the two approaches, it seems to me that “competition” has the following points in its favor: 1. It is more conservative in what it attributes to the prelinguistic child. According to “mapping preference” OPs, children have certain predispositions, independent 638

of experience with a particular language, for relating meanings to forms in certain ways, and they draw on these predispositions in constructing their grammar. The competition account, in contrast, does not assume that the child has any particular prelinguistic expectations about form-meaning relationships. Instead, it looks to children’s experience with the structures of the language they are learning for help in explaining errors of the types we have been considering. Any model of language acquisition must deal with the fact that children learn alternative ways to express similar meanings, and account for how they sort out which devices to use on which occasions. The competition account simply exploits this process in accounting for children’s errors. I have argued earlier that, as a general procedure, we should avoid crediting children with inherent preferences in form-meaning mapping unless we are forced to. As long as explanations based on general learning principles in interaction with the properties of the input seem to be sufficient to explain the data, the burden of proof is on the advocates of inherent preferences to show that this more powerful explanation is really needed. 2. The competition interpretation is more parsimonious, accounting with a single principle for data that require several “mapping preference” OPs. The best model in the end may not be the most parsimonious one, but, just as we should not postulate intrinsic linguistic preferences unless the phenomena they are designed to handle cannot be satisfactorily explained with other principles that our model will clearly need anyway, we should not favor a less parsimonious model over a more parsimonious one unless it provides a better account of the available 639

evidence. Competition is not only the more parsimonious explanation but at present it seems to account for several error types that are not currently covered by “mapping preference” OPs and that sometimes in fact violate mapping preference predictions. These include: a. The substitution errors described earlier as counterinstances to OP:UNIFUNCTIONALITY, E.G. put for give or make for let (see Bowerman, 1978, 1984, and MacWhinney, 1985, for an interpretation of these in terms of competition between semantically related forms). b. Errors in functor ordering where an appeal to “relevance” is difficult because two functors with closely related meanings show symmetrical rather than directional misordering with respect to their “closeness” to the nucleus: e.g. both hook out for unhook (a necklace) and untake for take out (stitches in a sewing project) (Bowerman, 1981). c. Errors in which a synthetic lexical form is used where its analytic counterpart is needed (the converse of error type 3 above), e.g. It BRINGS your wishes true (=makes your wishes come true; said of a “magic” pebble); water BLOOMED these flowers (=made these flowers bloom) (Bowerman, 1981, 1982a). Notice that such errors run counter to OP:ANALYTIC FORM, just as substitutions like put for give run counter to OP:UNIFUNCTIONALITY. 3. The competition account is more comfortable with the low frequency of many key error types in children’s 640

speech, relative to correct usage. Prevailing correct usage or alternation between correct and incorrect forms is something of an embarrassment to the “mapping preferences” approach. If children strive for explicit marking in accordance with OPs UNIFUNCTIONALITY and ANALYTIC FORM, why do they not use unifunctional and analytic forms routinely? (See also Maratsos, 1979, who criticizes Slobin’s (1973) earlier Operating Principle G, “Semantic relations should be marked overtly and clearly,” on this basis.) Since the competition account does not ascribe errors to children’s mapping preferences, infrequency of errors does not matter; the proportions of incorrect to correct versions of a construction would be explained purely by reference to the relative activation strengths of the competing alternatives. 4. The competition approach can deal with the absence in child speech of many errors that the “mapping preference” approach predicts. It was noted earlier that languages are peppered with plurifunctional forms where errors reflecting OP:UNIFUNCTIONALITY would be predicted but do not seem to occur. Similarly, there are countless opportunities for OP.ANALYTIC FORM to go into action, but in most cases children appear to be content with synthetic forms (see Bowerman, 1982c). This noncommission of errors is puzzling for the “mapping preference” approach, but unproblematic for the competition account, as long as it can plausibly be argued that only the correct form should get activated in connection with the relevant meaning or at least that it has much higher activation strength than more analytical competitors. 641

Despite these immediate advantages for the competition account, it would be premature to conclude that children’s “mapping preferences” play no role in errors of the sort we have been considering. At least the following points should be considered: 1. Crediting children with notions of “relevance”—ideas about what meanings “belong together” and could be expressed by the same form or by forms in close proximity to each other—may seem gratuitously innatist if competition can account for the same error data. But if children have no intrinsic notions of relevance, we are left with a mystery: Why, as Bybee (1983, 1985) has shown, is there so much crosslinguistic consistency in the placement of various grammatical markers relative to content words? Perhaps, then, children do have notions of relevance after all. The problem is that there is as yet no strong evidence for this. Earlier, we saw that when Polish children attach the personal ending to the verb rather than to the conditional marker where it belongs, there is no need to invoke notions of relevance. This is the normal position for the ending in nonconditional sentences, so Polish children could be expected to favor it on general grounds of consistency and frequency (which contribute to greater activation strength), even if they have no a priori notions about relevance at all. The other examples given by Slobin (1985) in support of the two “relevance” OPs, RELEVANCE and OPERATORS, are not compelling for the same reason: The positions in which the morphemes incorrectly occur in accordance with “relevance” are positions that in other common 642

construction types are appropriate for these morphemes (or, in some cases, their contextually-determined variants). Additionally, in the case of operators, the variant the language uses in the “more relevant” clause-external position tends to be freestanding and acoustically more salient than its bound, sentence-internal counterpart, so it might be acquired first simply because it is more easily picked out of the speech stream rather than because its position is favored on semantic grounds.11 To strengthen OP:RELEVANCE and OP:OPERATORS, we need additional evidence. There are two kinds that would help, and they may both eventually be forthcoming: (a) evidence that children misorder forms in accordance with “relevance” even when there are no alternative construction patterns in the language in which the forms are positioned in the same way as in the errors, and (b) evidence that when both a “relevant” and a “nonrelevant” position for a morpheme are modeled in the input, children prefer the “relevant” position even when the language uses it in less frequent or far fewer construction patterns than it uses the “nonrelevant” position. 2. Karmiloff-Smith (1979b) found that while doing an extended task that required devising a notational system for route directions, children sometimes spontaneously decomposed their own “synthetic” symbols. That is, they abandoned symbols that they had used earlier to convey two or more pieces of information simultaneously, replacing them with separate symbols that represented each meaning element more explicitly. Karmiloff-Smith 643

argues that these decompositions are related to language errors in which children replace a synthetic morpheme with a sequence of morphemes in which each meaning element is spelled out separately: She attributes both to the child’s efforts to mark separate meanings with separate forms until the meanings are thoroughly mastered. It is not clear how the competition account could be extended to cover cases of “decomposition” when there is no input system (whether language or something else) that offers the child alternative ways to express the same or similar meanings. In sum, the “mapping preference” account can interpret a variety of one-to-one mapping behaviors within a common framework, whereas the competition account apparently cannot (Karmiloff-Smith, personal communication). 3. Even if we restrict our attention to language, competition seems at best to provide only a partial explanation for errors that have been cited in support of one-to-one mapping preferences. In particular, it gives no account of (a) why certain forms compete in children’s speech and other do not, (b) why errors that seem to reflect competition often set in late in children’s speech, long after the forms involved are well established and used productively and flexibly (Bowerman, 1978), and (c) why, at the same time that children begin to produce distinct markings for the meanings encoded by certain plurifunctional forms, they may also start to make comprehension errors that suggest that they have come to associate the form narrowly with only one of its meanings (Karmiloff-Smith, 1979a, 1979b). 644

As noted earlier, one important basis for competition among forms is meaning: Forms that, from the speaker’s point of view, express the same or closely similar meanings will tend to be activated simultaneously and compete for selection. This means that the most satisfying explanation for many of children’s errors may ultimately be one that combines the “on-line” notion of competition with an account of how children organize meaning and how their meaning structures change over time. I defer further discussion of this possibility until a later section in which I consider children’s early structuring of semantic space in connection with Slobin’s (1985) “Basic Child Grammar” hypothesis. 3.2.4. Other Interpretations Based on Linguistic Experience. Not all errors discussed by Slobin as evidence for “mapping preference” OPs can be interpreted alternatively in terms of competition. However, other interpretations for these errors often need to be considered which—like competition—invoke the child’s experience with the distinct structure of the language being learned, and correspondingly minimize the need to postulate intrinsic preferences for how meaning should be mapped. Consider the following example. A characteristic error of children learning Hebrew is to insert a subject pronoun into relative clauses, as in ha yeled še HU nafel ba mayim ‘the boy that HE fell in the water’; ha iša še HI ra’ata et ha naxaš ‘the woman that SHE saw OBJ the snake.’ Slobin (1985) cites these examples in support of OP:CANONICAL CLAUSE FORM: “If a clause has to be reduced, rearranged, or 645

otherwise deformed when not functioning as a canonical main clause, attempt to use or approximate the full or canonical form of the clause.” This OP reflects the more general “mapping preference” theme that children prefer a clear marking of meaning: “As in the discussions of OP:MAXIMAL SUBSTANCE and OP:EXTENSION … , overt marking of all sentence participants is an early and persistent characteristic of child language” (Slobin, 1985, p. 265). In this case, children avoid deletion of the nominal argument of a verb in an embedded clause. If overt marking of all sentence participants is important to children, why do children learning English not make the same error, since—as the English glosses on the Hebrew sentences indicate—there is the same opportunity? And here a structural difference between Hebrew and English becomes relevant. In adult Hebrew, relative clauses often contain a resumptive pronoun (i.e. a pronoun that “copies” the relativized head into its underlying position in the relative clause). Such pronouns, which are required for oblique objects and optional for direct objects, are suggested by the following translation-equivalents: ‘Here’s the boy that I gave the candies to-HIM’; ‘Here are the candies that the boy took THEM’ (Berman, 1985 and personal communication). Resumptive pronouns do not occur in subject position, however. In English, resumptive pronouns are considered ungrammatical in any position in relative clauses (although they do occasionally occur with oblique objects). These structural differences—a regular pattern of resumptive pronouns in Hebrew relative clauses to 646

which subjects are an exception, versus no such pattern in English—coincides with making or not making the error of inserting a subject resumptive pronoun. This suggests that Hebrew-speaking children make the error not because of a language-independent preference to mark all nominal participants in a clause overtly (because then English-speaking children should make the same error), but because they try to extend a characteristic pattern of Hebrew to cases that are exceptions to it.12 This example provides a clear illustration of how accounting for errors requires careful study of the structure of the language being learned. Taken out of context, many errors might be interpreted as evidence for prelinguistic mapping preferences. Analyzed with reference to related construction patterns in the language, however, they often point instead to children’s close attention to the structural properties of the system to which they are exposed.

3.3 From a List of Operating Principles to a Theory of Language Acquisition In the preceding sections I have discussed some specific examples of problems in the OP approach as it now stands, like the freedom with which new OPs can be inserted to deal with potential counterevidence, the way simple rewording can sometimes convert the preferences or procedures stated in OPs to their opposite, the explanation of apparently related error types with completely independent

647

OPs, and the tolerance for alternative explanations of the same data. These difficulties all seem to reflect a larger problem, of which those who have worked within or been inspired by the OP approach are well aware: the lack of conceptual “glue” to bring the OPs into a compelling relationship with one another. The need for such glue in the development of a theory is insightfully discussed by Kaplan (1964). Kaplan defines a theory as, most fundamentally, a system of laws. Of the two basic types of such systems described in his book, one stands out as immediately applicable to the OP approach: A concatenated theory is one whose component laws enter into a network of relations so as to constitute an identifiable configuration or pattern. Most typically, they converge on some central point, each specifying one of the factors which plays a part in the phenomenon which the theory is to explain…. This is especially likely to be true of a theory consisting of tendency statements, which attain closure only in their joint application.(p. 298, emphasis in the original) In the OP approach, language acquisition is of course the phenomenon to be explained. The OPs are the component laws that converge on this phenomenon, and they must operate jointly to obtain closure. But the “network of relations,” “pattern,” or “configuration” is still missing. Kaplan is rightly emphatic about the need for this:

648

The laws are altered by being brought into systematic connection with one another…. The theory is not the aggregate of the new laws but their connectedness, as a bridge consists of girders only in that the girders are joined together in a particular way. The theory explains the laws, not as something over and above them, but by giving each the strength and purpose which derives from the others. (p. 297, emphasis added) Without connectedness, there are not enough constraints on what is possible. Every acquisitional phenomenon can in some way be accounted for, but the sense of compellingness or inevitability that would come from a more integrated system is still absent. Where could the needed integration come from? One step, as Clark (1980, 1985) has urged, may be to establish how hypothesized OPs must be ordered and weighted. This would allow us to specify when each OP comes into play (e.g. as children move sequentially through possible solutions to a problem, or depending on what they have already learned about their language), and to state which procedure will take precedence in cases of conflict between OPs.13 Will this be enough to weld the girders into a bridge? I think that further work will be needed. In particular, I suspect that at present the OP approach does not provide sufficiently for how children home in on a grammar with certain “deep” structural properties and not others. The OP approach has done a great deal to increase our awareness of the importance for language acquisition of surface structural details like whether relative clauses 649

look rather like main clauses placed after their head noun or are compacted into prenominal modifiers, and whether spatial location is expressed with prefixes or suffixes, prepositions or postpositions. But the OPs do not guide the child toward a grammar with more abstract universal syntactic constraints or parameters of the sort associated with the work of Chomsky (e.g. 1981) and his colleagues, nor are they intended to. Nativist theorists would certainly urge that what is missing from the OP approach is a theory of grammar: a conception of how surface variability is constrained by deeper syntactic principles, and an account of how children’s obedience to these principles guides their construction of a grammar for a particular language. I am reluctant to promote this solution too strongly because I am not yet convinced that we know which theory of grammar to install in the place of honor, and I think that at least some of the formal principles that nativist theorists have proposed crediting to the child’s innate endowment can be learned on the basis of experience (see Bowerman, 1983b). It is probably true, however, that before the OP approach can develop into a truly satisfying theory of language acquisition it will have to deal more directly with the problem of how purely formal constraints on grammatical structure are incorporated into children’s grammars. To summarize, in this section I have considered the problem of testing and systematizing proposed OPs. I first argued that, with the expansion of the OP model, it has become difficult to test OPs directly because they 650

often form closed systems that can interpret all observed outcomes. In addition, counterevidence can often be reinterpreted to be compatible with existing assumptions. A second problem is that multiple explanations for children’s language behaviors have tended to proliferate within the OP approach. We need to test alternative explanations by systematically drawing out their predictions and determining whether these are met. In some cases we find absence of errors that should be committed if a particular OP is correct, whereas in others we find errors that should not be committed. Only by submitting candidate OPs to rigorous crosslinguistic analysis can we determine which ones fit the contours of the data and which are superfluous. Finally, I stressed the need to integrate OPs into a coherent system. Woven through these methodological arguments has been a persistent substantive worry: that many of the OPs outlined in Slobin (1985) attribute to the child stronger inherent predispositions about the way meanings should map into linguistic forms than can be justified by the data currently available. In the following section we look more directly at the problem of how strongly language acquisition is directed by propensities that arise independently of experience with a particular language.

651

4. BASIC CHILD GRAMMAR: IS ONLY ONE WAY OF STRUCTURING SEMANTIC SPACE “BASIC”? Children who acquire language with Slobin’s (1985) set of Operating Principles may not yet have been accorded much help from inborn ideas about formal syntactic structure, but when it comes to the organization of meaning they are richly endowed. Piecing together evidence from many languages, including most centrally those discussed in these volumes, Slobin (1985) has formulated a claim of great theoretical importance. He argues that children approach the language acquisition task with a prestructured “semantic space” in which meanings and meaning clusters constitute a “privileged set of grammaticizable notions” (see p. 217) onto which functors and other grammatical constructions are initially mapped. The particular forms that get mapped vary from language to language, of course, but the basic meanings are constant, along with positioning constraints and certain other syntactic properties that result from the way OPs like RELEVANCE interact with the linguistic input, as I discussed in the preceding section. The outcome of this initial mapping process, according to Slobin, is a “universally specifiable ‘Basic Child Grammar’ which reflects an underlying ideal form of human language” (p. 204, emphasis added). The Basic Child Grammar (BCG) hypothesis is a bold and intriguing extension of what has long been a deep conviction of many child language scholars, that language-learning children do not simply passively 652

accept the structures their language offers them, but actively strive to organize and make sense of the linguistic input in their own way. Beyond this, the claim has particular current importance because it accords closely with arguments from other fields that there is a universal cognitive/semantic substratum for language, most notably Bickerton’s (1981) proposal, based on Creole studies, for an innate language “bioprogram.” Because of its theoretical importance and because it attributes strong built-in dispositions to the language-learning child, the BCG hypothesis demands the closest scrutiny. How we evaluate it depends partly on how literally we take its claims. Slobin himself is in fact rather cautious. Reminding us that the available data are still sparse, he proposes the BCG hypothesis as a “broad first pass … at the mechanisms of the L[anguage] M[aking] C[apacity] that may be responsible for children’s preferences to construct language in particular ways, knowing full well that such abstracted and generalized preferences cannot account in detail for the acquisition patterns of particular, individual children” (p. 206). While I recognize this qualification and appreciate the coherent theoretical perspective that such a “first pass” provides, I have chosen to interpret the notion of a universal basic child grammar rather strictly, for two reasons. First, a strict interpretation is called for by the word “grammar.” A “grammar” is not a loose collection of tendencies or preferences, to be sometimes observed and sometimes not, but a strong set of constraints. Although Slobin perhaps does not intend for the notion of a 653

universal child “grammar” to be interpreted so literally, the word carries an impression of rigor that will encourage readers, especially those outside the field of child language, to assume that children’s early grammars are much more uniform than they are. Second, the only way we can make progress in understanding the conceptual underpinnings of children’s early grammars is to test hypotheses about these underpinnings against data from individual children. Of course, hypotheses about “preferences” cannot be discarded immediately just because counter-instances are found. However, if it turns out that children frequently or consistently fail to behave as the hypotheses predict, we will have to conclude that the proposed preferences are weak or nonexistent.14 The problem to be solved is a classic “nature-nurture” dilemma: To determine the relative contribution to children’s early grammars of, on the one hand, inherent, universal tendencies to structure conceptual material in certain ways and, on the other, experience with an input system that shapes meanings in a language-specific fashion.15 In the following discussion I argue that the BCG hypothesis does contain a fundamental insight into early language development: that children’s starting semantic space is not a tabula rasa, passively awaiting the imprint of the language being learned before taking on structure. Rather, children are conceptually prepared for language learning: They can spontaneously categorize objects, events, situations, etc. for purposes of linguistic expression and, moreover, in doing so they use 654

meaning distinctions that are relevant for language—i.e. distinctions of the sort that often figure in the semantic systems of natural languages. However, I will urge that these “candidate meaning distinctions” are far less rigid than the BCG hypothesis predicts. Specifically, they do not define a single, privileged set of semantic notions that strongly attracts the grammatical forms of the input and molds the initial usage of these forms to a uniform pattern. Rather, they are better understood as a system of relatively accessible alternatives for structuring semantic space within any particular conceptual domain—i.e. as a set of salient meaning distinctions that children will tend to try out first. If, as I contend, the initial organization of semantic space is not fixed but flexible, there will be variation in the meanings children initially link to grammatical forms. One important factor that can influence the meanings children adopt is the semantic structure of the input language—i.e. the specific meaning categories associated with the grammatical forms of the language in the speech of fluent speakers. I argue that children are prepared from the beginning to accept linguistic guidance as to which distinctions—from among the set of distinctions that are salient to them—they should rely on in organizing particular domains of meaning. In consequence, there is no single, universally shared “Basic Child Grammar.” Children begin with grammars that are slanted toward the semantic structure of the input language, even if not yet in perfect accordance with it.16

655

Like Slobin, I base the conclusion that there is structure in starting semantic space on children’s errors—both errors of commission, in which a form is extended to meanings for which it has never been modeled in the input, and errors of omission, in which a form is used over a semantically restricted portion of its full adult range. Errors of both types are typically linguistically “sensible,” in that they reflect principles for categorizing meaning that play a specifiable role in the structure of one language or another. Sometimes errors reflect the child’s reliance on a principle of semantic categorization that is relevant for the meanings of certain forms in the language being learned, but not the forms to which the child has applied it. Alternatively, the semantic principle may have no clear function in the child’s language, but be important in the structure of another language, often even in connection with a form exactly parallel to the one with which the child has linked it.17 My conclusion that the structure of starting semantic space is more flexible than Slobin postulates in his Basic Child Grammar hypothesis is based on three types of evidence: 1. In some—and probably many or most—domains of meaning associated with grammatical forms, children adopt correct or near-correct meaning distinctions (i.e. those used by the local language and displayed in the input to the child) essentially from the start, even when these distinctions partition semantic space in strikingly different and language-specific ways.

656

2. When children do use forms in connection with nonmodeled meanings (i.e. meaning categories that are not displayed in the input), this deviation from adult norms is often not their first step. Errors that have been interpreted as reflecting a “basic” or “ideal” organization of semantic space sometimes emerge only after a period in which the forms involved are used flexibly and productively across correct and language-specific categories of meaning. Moreover, when “nonmodeled-meaning” errors arise relatively late, they often remain infrequent relative to correct use. They therefore should not be assumed to have direct implications for the properties of the grammar the child has formulated (see p. 309 above). 3. Children learning the same language vary to some extent in the meaning categories they first associate with grammatical forms. Some children first use a form in connection with a very narrow (underextended) category of meaning, whereas other children extend it across a broader category from the start. If even children with exposure to similar linguistic input make different form-meaning mappings, the initial organization of semantic space must be less rigid than the BCG hypothesis predicts. The following discussion of flexibility in starting semantic space cannot cover all the evidence exhaustively. I focus therefore on evidence for initial language-specificity as in (1) and (2), and refer the reader to Bowerman (1976, 1982b) for some examples concerning (3).

657

Before considering language-specificity in the initial mapping of grammatical forms, let us take a quick look at the BCG claim about the child’s starting point. Slobin proposes that children’s starting semantic notions are language-neutral: “Semantic Space provides Basic Child Grammar with a level of organization that serves as an opening wedge to the acquisition of language-specific grammatical distinctions, without at first biasing the child to any particular language” (1985, p. 228), emphasis added). As children progress, however, they are led to the diverse grammars of their individual languages through linguistic experience: “Later in development, of course, the language-specific use of particular functors will train the child to conceive of grammaticizable notions in conformity with the speech community. At first, however, there is considerable evidence that children discover principles of grammatical marking according to their own categories—categories that are not yet tuned to the distinctions that are grammaticized in the parental language” (1985, p. 218). Whether children can begin their mapping of a particular domain of semantic space in a universal, language-neutral way depends on whether an unbiased “opening wedge” is logically possible.18 In some domains it is difficult to imagine what such a wedge would be like. In others, a candidate wedge can be envisioned; the question is whether children indeed use it. I will look at some cases of each kind, using the distinction between domains that are or are not susceptible to an unbiased “opening wedge” as an

658

organizing framework within which evidence for both claims (1) and (2) above can be presented.

4.1. Unbiased “Opening Wedge” Often Impossible: Language-Specific Learning in Hierarchically Organized Domains Languages often differ in how finely they subdivide particular categories of meaning. For example, English expresses the first person plural notion “we” with a single pronoun, we. In many languages, ‘we’ is subdivided into two categories, either on the basis of whether ‘we’ includes the listener (we-inclusive versus we-exclusive), as in Tamil, or on the basis of whether it includes two people or more than two (we-2 versus we-more than 2), as in West Greenland Eskimo. In still other languages, e.g. Hawaiian, both these criteria are applied at once, resulting in a four-way contrast: we-2-inclusive, we-2-exclusive, we-more than 2-inclusive, we-more than 2-exclusive. Finally, some languages, e.g. Nogogu, go even further, honoring not only all these distinctions but also breaking down ‘more than 2’ into ‘3’ and ‘more than 3’ for a grand total of six categories, all marked with separate forms (see Ingram, 1978). As in this example, successive subdivisions in a semantic domain often form rough hierarchies or semantic trees, with some languages making no distinctions at all, or only one, others making an intermediate number, and still others making a great many. Other good examples of hierarchies involving the meaning categories associated with grammatical forms are discussed by Denny (1978, 1979).

659

When semantic distinctions are related hierarchically, there is often no language-neutral starting point for the child. No matter which level of distinction-making is compatible with the distinctions recognized in starting semantic space, it is likely that this level has already been chosen by one or more languages, so BCG would automatically be biased toward those languages in this corner of semantic space. Biasing can at least be minimized by assuming that the child’s preferred entry level in such cases is the most superordinate. This would allow children to proceed by gradually moving down the semantic hierarchy as required by their language, making successively finer subdivisions of meaning. If they start at the most superordinate level children will often collapse meaning distinctions that their language requires speakers to make. Slobin (1985) gives several examples of this phenomenon, including the tendency to conflate static locations and goals on the one hand, and locations and possessors on the other. The hypothesis that children start at the top of semantic hierarchies may seem plausible at first glance, since, as is well known, they often cannot or do not discriminate between meanings that the adult language distinguishes. However, if we consider the problem from a general perspective, it is clear that children cannot be expected always to enter a hierarchy at the most undifferentiated level found in any language of the world. For example, studies of the acquisition of word meaning show that children typically enter a taxonomic hierarchy of lexical items at a middle level, and proceed beyond this point not only by learning more differentiated, subordinate 660

words but also by learning more abstract, superordinate ones (e.g. Brown, 1958; Rosch et al., 1976). Unfortunately, however, if the notions the child will come to associate with grammatical markers are organized in starting semantic space at a level more finely differentiated than the uppermost, the idealization of an unbiased starting point is rather seriously violated: Children would start by making semantic distinctions that are observed only in certain languages and then have to erase them if it turns out that their local language is less finely differentiated in this semantic domain. In summary, no solution is ideal for the assumption that starting semantic space is not biased toward any particular language, but a start at the top of the hierarchy is preferable on theoretical grounds. If some children start at a lower level, crosslinguistic comparisons become crucial to establish whether this is at least the universally preferred entry point. If it turns out that making certain distinctions is associated with learning a language for which this is appropriate (or varies across children and cannot be linked to language), then a key tenet of Basic Child Grammar is disconfirmed. The available data are still quite inadequate for a thorough study of whether all children choose the same entry level for semantic domains that, seen in crosslinguistic perspective, form hierarchies of meaning distinctions. One problem is that children make no errors when they enter a hierarchy at the appropriate level for their language, so correct, language-specific learning tends to escape notice.19 A second problem is that when errors are reported that reflect a child’s 661

failure to distinguish two meanings that are differentiated by the language, it is seldom clear whether these errors represent the child’s standard usage or are a minor disturbance in what is basically a correct and language-specific pattern of usage. Despite these problems in performing a full-scale evaluation of children’s initial entry-levels into semantic hierarchies, there is evidence that children are capable of observing certain language-specific semantic distinctions from the start, and no evidence as yet that their counterparts learning languages that do not make these distinctions also attempt to observe them. I offer two examples. 4.1.1. IF and WHEN in Future Predictives. The first example concerns that section of semantic space mapped into complex “future predictive” sentences of low hypotheticality. These constructions refer to two future events that might well occur, with the occurrence of the second contingent on the occurrence of the first: e.g. If John comes home tonight, we’ll go out. A common pattern across languages is for the antecedent event to be mentioned in a subordinate clause introduced by a subordinating conjunction, and the consequent event to be expressed in the main clause. In formulating the subordinate clause, speakers of English must attend to a rather fine meaning distinction. If the antecedent event can be fully expected to take place, the conjunction of choice is when. If the antecedent is only a possibility,, however, the speaker should use if.20 The role of subjective certainty can be appreciated by comparing IF John comes home tonight we’ll go out (John may or may not come) with WHEN John comes home tonight we’ll go out (it is assumed that John will come). The certainty/ 662

uncertainty distinction is irrelevant for comparable constructions in many other languages: a single subordinating conjunction (e.g. wenn in German, als in Dutch) applies freely across the entire domain.21 Cross linguistic studies of the acquisition of complex sentences (Bowerman, in press; Clancy et al., 1976) show that when emerges before if in children learning English, and is used appropriately to express future antecedents whose occurrence can be expected (hereafter, “certain antecedents”). Children learning languages with the generalized if / when conjunction also first use this form to express certain antecedents. This restricted starting point might reflect a cognitive developmental sequence by which contingencies involving certain future events are easier to conceptualize than those involving uncertain ones (Clancy et al., 1976). Whatever the explanation, it is the next step that is critical for our purposes: how are uncertain future antecedents expressed when they come in (which typically happens toward the end of the third year)? Children learning languages with a single if/when form readily extend this form to cover uncertain antecedents as well as certain ones, and there is no evidence at present to suggest that they conceptualize the domain as involving two distinct meanings that just happen to share the same (plurifunctional) form.22 This might tempt us to conclude that the meanings are conflated in the starting organization of semantic space. However, evidence from English-speaking children forces us to reconsider. 663

In studying the division of labor between if and when in three children for whom detailed longitudinal data were available (Bowerman, in press),23 I found that the children were essentially perfect from the very beginning in their choice between these two forms: When was reserved for antecedent events that the child had excellent grounds for assuming would take place, like an anticipated instance of a recurrent daily event (When Daddy comes home …. When I go to bed … ), an event being planned and about to be executed (When I go outside … , … when you cook it), the completion of a bounded, ongoing event (When I’m through … ), and growing older (When I get bigger … , When this house gets very old …). If, in contrast, was used for antecedent events about which the child had no grounds for certainty, e.g. If you lose that black thing … if that thing loses, then you won’t have any key (child looking at plastic top of mother’s key), If we go out there we haf wear hats (spoken on a lazy rainy day), If I get my graham cracker in the water, it’ll get all soapy (child in tub), If somebody takes the newspaper I’ll be sad (after child has postponed bringing in the newspaper). The apparent ease with which English-speaking children incorporate the meaning distinction signaled by the choice between if and when into their grammars—and do not, for example, just overextend when initially or persistently confuse the two forms—makes it unlikely that the two notions are conflated in starting semantic space. On the other hand, the ease with which children learning languages that do not make the distinction extend a single form across both meanings suggests that 664

the conflation can readily be made. In short, both organizations are apparently accessible to children from the start of their acquisition of grammatical markers for this semantic domain. They can readily adopt whichever one is displayed in their language, at least assuming the meanings are clearly marked. 4.1.2. Spatial and temporal meanings. A second example of language-specific acquisition in a hierarchically organized semantic domain concerns spatial and temporal meanings. It has long been recognized that meanings of these two kinds have strong conceptual affinities. For instance, languages often extend words that are considered basically spatial to temporal meanings as well: e.g. in Hebrew and many other languages the words for in front of and behind also mean temporal before and after, respectively, and in English we can apply long both to objects extended in space (a long stick) and to events extended in time (a long concert). These and related phenomena have led many linguists to propose that spatial location is a basic organizing metaphor on which the temporal system of languages are built (e.g. Bennett, 1975; Traugott, 1978; see also Lyons, 1977, p. 718ff.). The pervasive relationship across languages between spatial and temporal expressions makes it plausible to hypothesize that concepts of space and concepts of time are close neighbors in semantic space. Children might therefore be expected to show an initial tendency to conflate parallel space-time notions (see Traugott, 1978, for this hypothesis). Subsequent language-specific experience would teach them that separate marking is 665

required, at least for certain meanings in certain languages. The conflation of notions of space and time would be evidenced by errors in which a single form is incorrectly extended across both spatial and temporal meanings, or in which parallel spatial and temporal forms, if the child knows both, get confused. In accord with this prediction, my longitudinal records of the language acquisition of my two daughters are sprinkled with errors in which a spatial form is substituted for a temporal one. For example: 1. E 3;9

Can I have any reading behind the dinner? (=after. To M, who is fixing dinner; a request to be read aloud to.)

2. C 7;6

I don’t remember behind those two. (=before. C has just been recalling her last two birthday parties; cannot remember any previous ones.)

3. E 3;9

(E telling that invited children failed to show up at a party): E: They didn’t come (even) at the back of the birthday. M: (confused): At the back of the birthday? What is that? E: When the birthday is over.

4. C 4;5

(C telling sequence of activities at a birthday party she has just been to): The balloons is on the other side, after I ate. But there might have been more on the first side. (“on the other side” = after (the eating); “on the first side” = first, before (the eating).)

666

5. C 7;2

Do we have room before we go to bed for another reading? (=time. M has been reading aloud in the evening; has just finished book.)

Equipped only with the information that these errors occur, we might well conclude that children’s initial organization of semantic space conflates notions of space and time. But one further piece of information is crucial: errors like these were never observed in the first period of the children’s use of the relevant forms. They appeared quite late, months and—for some words—even a year or more after both the spatial and temporal members of a pair (e.g. behind and after) were well established—i.e. used frequently, in varied contexts, and exclusively for either spatial or temporal meanings. Even during the period when the errors occurred, they were always very infrequent relative to correct usage (see Bowerman, 1982c, 1983a, for discussion). These findings make it impossible to maintain that Basic Child Grammar’s ideal or preferred organization of semantic space treats parallel spatial and temporal meanings as equivalent for purposes of grammatical marking. English-speaking children are quite willing to differentiate these meanings from the start. Initial language-specific learning, followed later by occasional but recurrent substitution errors, also characterizes English-speaking children’s approach to a variety of other semantic domains. For example, as mentioned earlier in the discussion of OP:UNIFUNCTIONALITY (p. 314), children make substitution errors involving put and give (locative

667

versus possessive goal), put and make (cause-change-of-location versus cause-change-of-state), and make and let (active versus permissive causation) only after an extended period of using these forms flexibly and correctly. In every case the errors reflect an alternative semantic organization that is linguistically well-motivated: in many languages a single form is conventionally applied to both meanings in constructions that are otherwise parallel to the English ones. If the child had happened to be learning such a language, her ability to collapse the two meanings that English distinguishes— i.e. to appreciate their equivalence at a higher level of abstraction—would have fed directly into her construction of grammar. We will come back to the question of why, once children have already learned the semantic organization required by their language, they still sometimes make errors suggestive of other possible organizations.

4.2. Unbiased “Opening Wedge” Often Possible: Overlapping Semantic Categories In the hierarchically organized semantic domains I have considered, there has been no plausible semantically neutral “opening wedge” for BCG to employ. The least biased starting point in such situations is the most superordinate, but the relatively few detailed analyses so far available of the development of the meanings associated with grammatical forms indicate that children may respect the more differentiated categories required by their language from their first productive use of the relevant forms. 668

Not all crosslinguistic differences in the semantic categories associated with grammatical forms can be described in terms of the number of discriminations made in a shared hierarchy of potential distinctions. In many grammatical systems languages make approximately the same number of basic cuts through semantic space, but they do so according to different criteria. When we compare the categories defined by these criteria crosslinguistically, we find that they overlap partially.24 In some cases of overlap it is possible to imagine how children could start out in an unbiased way and wait for their language to lead them into the system required by their language. Do they do so? I explore this question with three examples involving major systems of grammatical marking. 4.2.1. Subjects and Objects. The three most basic grammatical roles associated with the noun arguments of predicates are subject of a transitive verb (transitive subject), object of a transitive verb (object) and subject of an intransitive verb (intransitive subject). Some languages (e.g. Takelma, an American Indian language) mark nouns in all three roles distinctly. However, most languages reduce the three categories to two by marking nouns in two of the roles identically. Transitive subjects and objects are always distinguished in such systems. Where languages differ is in whether they treat intransitive subjects like transitive subjects or like objects. Languages that opt for the former solution, like English and Hungarian, are called “nominative-accusative” languages, while those that opt for the latter, like Eskimo and Hua (a language of Papua New Guinea), are called “ergative” (Dixon, 1979, 669

Haiman, 1979). These differences are shown in Fig. 16.1. The way in which languages handle the grammatical marking of subjects and objects is an instance of a more general pattern of crosslinguistic differences in categorization discussed insightfully by Andersen (1973) and Haiman (1978). When three (or presumably more) categories in a domain can be discriminated, languages have several options. Some may choose to make no linguistic distinctions at all in the domain, while others distinguish all three categories. However, many languages make only a two-way contrast. In this case they often agree that two of the original three categories constitute extremes that should be differentiated. Where they differ, however, is in how they treat an ambiguous “middle” category that shares properties with both extremes: Some languages choose to assimilate it to one extreme and some to the other. As Haiman remarks, “since both generalizations involve the suppression of significant differences, neither is more nor less arbitrary than the other” (1978, p. 582). The solutions can thus be regarded as equally “natural.” Domains in which two categories stand out crosslinguistically as saliently different, but one or more additional categories share properties with both of them and are therefore ambiguous, lend themselves well to either of two hypothetical language-neutral learning strategies. 1. According to the first strategy, children would start by distinguishing the two extreme categories (e.g. using 670

different markers for them, or marking one but not the other). This strategy would reflect the cognitive salience for children of the two extremes and the polar opposition between them. It would also be a safe beginning since every language will at least differentiate the extremes if it makes any distinctions in the domain at all. Children would wait to see how to mark the middle category, since they would not yet know with which extreme their language groups it, or, alternatively, whether it requires still a third treatment.

FIG. 16.1. Differences between nominative-accusative languages and ergative languages in the grammatical treatment of subjects and objects. Hatching indicates the two constituents that the language treats alike with respect to grammatical marking.

FIG. 16.2. (a) Strategy one: start by distinguishing the extreme categories; wait to see whether the middle category should be treated like one extreme or the other, (b) Strategy two: start by learning how to mark the middle category; wait to see to which extreme these

671

markers should extend. (Arrows show the direction in which markers would be extended.) Children following this strategy would at first associate markers of subject and object (e.g. case endings, word order patterns, control of verb agreement) only with transitive subjects and objects, leaving intransitive subjects unmarked. This means that if they are learning a nominative-accusative language they would underextend subject markers; conversely, if they are learning an ergative language they would underextend object markers. Eventually the markers from the appropriate extreme category would be extended to cover intransitive subjects. This hypothetical sequence is shown in Fig. 16.2a. 2. A second strategy, which intuitively seems less likely, would be to start with the ambiguous middle category, intransitive subjects in this case. Again, there would be underextension: children learning nominative-accusative languages would at first limit their use of subject markers to intransitive subjects, leaving transitive subjects unmarked; conversely, children learning ergative languages would limit their use of object markers to intransitive subjects, leaving objects unmarked. Later the treatment accorded to intransitive subjects would spread to the appropriate extreme—transitive subjects for speakers of nominativeaccusative languages and objects for speakers of ergative languages. At this point appropriate markers for the opposite extreme might also come in. This hypothetical sequence is illustrated in Fig. 16.2b.

672

Slobin (1985) indeed finds evidence for the initial underextension of markers for subjects and objects, and upon casual inspection it might appear that the marking pattern he identifies conforms to strategy one. However, a close look shows that this is not necessarily the case. According to Slobin, children’s initial marking of subjects and objects is limited to utterances expressing a universally salient type of situation that he terms the “Manipulative Activity Scene.” This scene is characterized by a cluster of interrelated properties like an agent intentionally and physically acting on an object, with the object often undergoing a change of state or location as a result of the action. The evidence for the importance of this scene in starting semantic space is that (1) children learning nominative-accusative languages tend to restrict their initial use of the accusative marker (i.e. the marker of objects) to direct objects that refer to objects acted upon (changed, etc.), instead of extending it to all direct objects, and (2) children learning ergative languages tend at first to limit their use of the ergative marker (i.e. the marker of transitive subjects) to subjects that refer to agents involved in manipulative activities, instead of extending it to all transitive subjects. These patterns of underextension do suggest the cognitive salience to children of scenes with a strongly transitive relationship between agent and object, but they do not in themselves constitute evidence for a beginning that is neutral between the nominative-accusative and ergative systems. They are indeed compatible with an unbiased beginning, since transitive subjects and intransitive objects, whether conceived of too narrowly 673

or appropriately broadly, are distinguished in all languages. But they are also compatible with a beginning that is biased in either a nominative-accusative or an ergative direction. The information that is critically needed to determine whether children begin in an unbiased way is what they initially do with intransitive subjects. If learners of nominative-accusative languages treat them like transitive subjects (e.g. by choosing nominative forms for both, positioning both in the same way relative to the verb, or allowing both to control verb agreement), then they have adopted a language-specific system from the start, even if they restrict their accusative marker to a subset of the direct objects for which it is appropriate. Conversely, if learners of ergative languages treat intransitive subjects like objects (e.g. by positioning or by using the absolutive marker for both), then they are also in tune with the structure of their local language, even if they limit their ergative marker to a subset of the transitive subjects for which it is appropriate. A thorough investigation of whether children treat intransitive subjects like transitive subjects, objects, or neither, is far beyond the scope of this paper. Nevertheless, even a cursory look reveals that many children display languagespecific patterns from the start. The evidence is clearest from studies of word order, and I will restrict myself to this.25 Braine (1976) has shown that children learning nominative-accusative languages often have a productive “actor-action” pattern in the two-word period. This pattern applies to words referring to both actors that initiate actions on objects (a subset of 674

transitive subjects) and actors that initiate intransitive actions like self-movement (a subset of intransitive subjects); it positions constituents of both types identically with respect to the verb. In English, for example, transitive and intransitive actors are routinely placed before the verb, e.g. Kendall break, Kendall bite (i.e. Kendall breaks/bites something), and Kimmy come, Mommy sleep. This positioning remains stable for actors when three-term S-V-O strings become common (although some children do show uncertainty over the placement of nonactor intransitive subjects that undergo change, as in thread break; Braine, 1976).26 Braine speculates that the agent category may differentiate out of the broader actor category when actors that are causes are distinguished from other objects that move (1976, p. 68). The identical treatment of transitive and intransitive actors in the early sentences of children learning nominative-accusative languages seems so natural to adults who speak such a language that it hardly occurs to us to remark on it. But it becomes striking by comparison with Schieffelin’s (1985) finding that children learning Kaluli, an ergative language, never extend the ergative marker (i.e. the marker of transitive subjects) from transitive actors to intransitive actors. Taken together, these findings strongly suggest that children attend from the start to whether their language regards actors who perform an intransitive action like ‘coming,’ ‘sleeping,’ or ‘sitting’ as similar to actors who perform a transitive action like ‘breaking,’ ‘biting,’ or ‘opening.’

675

Schieffelin’s chapter does not give enough information about Kaluli children’s handling of intransitive actors and other intransitive subjects to allow us to determine whether they treat these like objects, as ergative languages require, but the evidence from Samoan, another ergative language, is unequivocal: according to Ochs (1985), young Samoan children from the two-word stage on “reserve the location immediately following the verb for absolutive constituents—transitive patients [objects] and intransitive major arguments [including intransitive subjects]—but exclude ergative constituents—agents—from this position … In this way they treat patients and intransitive arguments as a single category, distinct from agents’’ (p. 831, emphasis altered). In summary, there is good evidence that children show sensitivity even in their earliest sentences to the classification their language imposes on the constituent “intransitive subject,” especially that subset of intransitive subjects comprising actors who perform actions described by verbs like come, sleep, cry, and sit. Children learning nominative-accusative languages typically position nouns in this role like transitive subjects, whereas children learning the only ergative language for which this information is available—Samoan—do not: Strikingly to the contrary, they position intransitive subjects like objects. The non-equivalence of intransitive subjects and transitive subjects for learners of ergative languages is also shown by their failure to extend the marker of transitive subjects to intransitive subjects.

676

4.2.2. Noun categories. Let us apply the line of reasoning built up in the last section to a second domain in which languages agree on which semantic end points should be distinguished but differ on what to do with an ambiguous middle category. Lucy (1981) has described some fundamental crosslinguistic differences in how languages subdivide common nouns with their systems of noun markers. English divides nouns into two major classes, termed “count” and “mass.” Count nouns, e.g. pig, can co-occur with the indefinite article a, plural -s (or its irregular equivalent), numerals, and certain other forms. Mass nouns, e.g. mud, cannot take the indefinite article or a plural marker (*a mud, *muds) and require a classifier (see footnote 17) to be able to co-occur with a numeral, e.g. two PIECES OF mud, not *two muds. For nouns with concrete referents, the count-mass distinction correlates, although imperfectly, with perceptual differences among entities: Bounded objects that are not too small are typically referred to with count nouns, e.g. dog, book, tree, whereas unbounded substances and collections of very small, undifferentiated objects are typically referred to with mass nouns, e.g. water, mud, rice. The nouns of Yucatec Maya follow quite different rules for grammatical marking. First, only animate nouns may receive a plural suffix. Second, all nouns require a classifier when they are accompanied by numerals (comparable to two PIECES OF mud), but animate nouns are once again distinguished from inanimate

677

nouns: they take the classifier túul, while all other nouns take p’éel. The count-mass and animate-inanimate distinctions make orthogonal cuts through semantic space. Simultaneous application of the two distinctions results in four noun categories, as shown in Fig. 16.3a: Type 1 [+ count, +animate], e.g. pig; Type 2 [+count, –animate], e.g. ball; Type 3 [–count, –animate], e.g. water; Type 4 [–count, +animate], e.g. poultry. Type 4 is not discussed by Lucy and I will eliminate it too for the sake of brevity. Figure 16.3b shows the differences between English and Yucatec in the grammatical handling of the remaining three categories, Types 1–3. Despite their differences, English and Yucatec are similar in one important respect, as Fig. 16.3b makes clear: Both distinguish grammatically between Type 1 nouns (pig) and Type 3 nouns (mud). Where they differ is in the handling of Type 2 nouns (ball). English focuses on the “countability” of the referents of these nouns and groups them grammatically with other countable objects like pig, glossing over their difference in animacy. Yucatec, in contrast, concentrates on the inanimacy of the referents and groups them with other inanimate entities like mud, ignoring their difference in countability. Lucy (1981) argues that the opposing solutions to the grammatical handling of Type 2 nouns adopted by English and Yucatec are widespread in the languages of the world. He concludes that these solutions are “not accidental, local phenomena but… the genuine products of different ways of organizing experience in the categories of language” (p. 12).27 678

FIG. 16.3. (a) Four classes of nouns defined by the simultaneous application of the [± count] distinction and the [± animate] distinction, (b) Differences between English and Yucatec Maya in the co-occurrence of nouns of three types with plural markers and numerals. Hatching indicates the two categories that the language treats alike with respect to grammatical marking. (Adapted from Lucy, 1981.) Children could approach the learning of noun markers in English and Yucatec in a way biased toward neither language with either of the two strategies described in the preceding section. Following strategy one, they would start by distinguishing grammatically between the two extreme categories, words for animate countable objects and words for inanimate substances. For example, English-speaking children would at first use the indefinite article and plural only with animate count nouns and wait for further evidence about whether this treatment is also appropriate for inanimate count nouns. Conversely, Yucatec-speaking children would at first use p’éel, the classifier for inanimate nouns, only with words for substances, extending it to words for countable inanimate entities like ball only after further linguistic experience.

679

Alternatively, following strategy two, children would begin by associating noun markers with the ambiguous “middle” category, words for inanimate countable objects. Children learning English, for example, would first use the indefinite article and the plural with members of this category, and children learning Yucatec would mark these nouns with the inanimate classifier p’éel. After further experience, they would extend the markers associated with the middle category to the appropriate extreme—animate count nouns for English speakers and inanimate mass nouns for Yucatec speakers. Unfortunately, we do not have acquisition data from Yucatec-speaking children. However, it is easy to determine whether English-speaking children adopt one or the other of these two “language-neutral” strategies. We need only check whether they initially limit their use of the indefinite article a and plural -s either only to animate count nouns (strategy 1) or only to inanimate count nouns (strategy 2). I know of no study that explicitly looks at this, but many studies have given detailed accounts of the acquisition of the indefinite article and the plural in connection with other questions (e.g. Brown, 1973; Gathercole, 1983; Gordon, 1982; Macnamara, 1982; Maratsos, 1976). Since nouns referring to both animate and inanimate objects are common in children’s speech from the one-word stage on, and since objects of both types have usually been included in testing comprehension or eliciting use of the markers, it is highly likely that if children initially used or understood a or the plural exclusively with either 680

animate or inanimate nouns, this would have been noticed. It has not been remarked upon, however, so we can be reasonably confident that English-speaking children treat dog, baby, ball, and cookie alike when it comes to articles and plural markers. In doing so, they pass over an excellent language-neutral “opening wedge” and crack into the English system of noun subdivision with categories that are right for English but wrong for Yucatec and many other languages.28 4.2.3. Temporality. A third domain to which the general approach I have been discussing can be applied, although somewhat less precisely, is temporality. Slobin (1985) proposes that the initial organization of semantic space in Basic Child Grammar is oriented toward two major temporal perspectives: Result (punctual, completive) versus Process (nonpunctual, noncompletive, ongoing). This distinction is marked early on by the perfective-imperfective form of verbs in Slavic languages, by the present (-Iyor)—past(-dl) forms in Turkish, by the progressive-past forms in English (-ing vs. -ed) and Japanese (-te iru vs. -to), etc. In early transcripts of child speech it is difficult to decide if the child is marking a progressive-nonprogressive or an imperfective-perfective distinction, although the parental language may orient itself to only one of these major distinctions. I would suggest that the two temporal Perspectives of Basic Child Grammar are neutral and superordinate to these language-specific categories. (p. 227)

681

Beyond this beginning point, children “must learn the ways in which Result and Process interact with other distinctions to determine the use of the verb forms in the language” (p. 227). For example, the English-speaking child must learn to mark events in the past differently depending on whether he views them as progressive or not (was fighting versus fought) but not on the basis of whether they were complete, whereas the child learning Polish must learn to attend to completion but not to ongoing progress. Notice that the developmental sequence Slobin proposes is similar to the one associated with the child’s adoption of “strategy one” in the examples considered earlier. According to this sequence, children first use grammatical markers to differentiate “extreme” categories of meaning; later they discover which semantic properties of these extremes are critical for their language, and they determine how these properties interact with other properties to define (in this case) a language-specific tense-aspect system. Slobin uses this example to illustrate the notion of an “unbiased opening wedge” in Basic Child Grammar; the image is indeed apt. There is no single ambiguous, middle category in the temporal system that children must learn to assimilate to one pole or the other. Most languages set up a variety of oppositions with their temporal markers, and many of the resulting categories are intermediate between Result and Process. Nevertheless, by simplifying a bit we can test whether children actually proceed in the hypothesized “strategy one” fashion. 682

According to the hypothesis, children first use past tense or perfective forms for verbs expressing accomplishments in the past (punctual, completive events with visible consequences like The milk spilled, I fell down, The car broke), and present, progressive, or imperfect forms for verbs expressing activities in the present (nonpunctual, noncompletive, ongoing events like John’s sleeping, The baby’s crying). An important question is what children do with the ambiguous category activities in the past (e.g. John slept yesterday, The baby cried all night). The hypothesis predicts that they do not at first mark verbs referring to such activities at all, since past activities do not conform to either of the poles that define the basic temporal opposition in starting semantic space, but rather share properties with both. Later, children learning English (for example) must relax the punctual, completive requirement for the use of -ed and extend this marker from the Result pole to past events in general, including activities. Conversely, children learning Polish must relax the present, ongoing requirement for the use of the imperfective form and extend it from the Process pole to noncompletive events in general, including past activities. This hypothesized sequence is shown in Fig. 16.4. The hypothesis that children do not initially mark past activities like cried with their past tense or imperfective forms is based on several studies (e.g. Antinucci & Miller, 1976; Bloom, Lifter, & Hafitz, 1980; Stephany, 1981). However, a careful recent study by Weist and his colleagues (1984; summarized in Smoczyńska, 1985) shows that the use of past tense or imperfective forms for past activities was often simply infrequent in the corpora 683

on which these studies were based, not entirely absent. These authors go on to provide compelling evidence that the development of Polish children does not fit the hypothesis. Their “single most important finding is that imperfective activity verb phrases with past-tense inflections were observed in all children starting from an early phase of tensed communication” (p. 353). Examples include the translation equivalents of (She) was swimming and (I) was eating. More generally, the data show that Polish children learn tense distinctions and the perfective-imperfective distinction simultaneously. There is no period in which they restrict their use of the imperfective to present activities or the perfective to past accomplishments. The lack of errors in their Polish subjects’ use of perfective and imperfective forms led Weist et al. to conclude that “the distinction between perfective and imperfective aspect appears to be primitive in child Polish” (p. 369). Comparing this finding with Aksu’s (1978) proposal that the initial aspectual split for children learning Turkish may be punctual versus durative, they make a proposal that accords completely with the view of the initial organization of semantic space that I have been urging throughout this discussion of the Basic Child Grammar hypothesis:

FIG. 16.4. Basic Child Grammar hypothesis about the development of verb markers in English and Polish,

684

including the way in which markers would be extended to verbs expressing past activities. Children can take different perspectives. They can view situations internally and externally. When a situation is viewed internally, features like incomplete, durative, and continuous are salient and when viewing the situation externally, the salient features are completed, punctual, and discontinuous. Depending on the language a child is learning, one or more of these feature oppositions will characterize the fundamental aspectual distinction. In Polish, the perfective versus imperfective distinction is transparent in the morphology and children can readily process the relevant affixes. (Weist et al., 1984, p. 370, emphasis added) In summary, children do not in the beginning inexorably conceptualize events in terms of the opposition between Result and Process, but are sensitive to—and capable of incorporating into their grammars—some of the specific semantic oppositions defined by the temporal system of their language, at least as long as these oppositions are clearly marked.

4.3. Finding the Balance between Flexibility and Constraint in Starting Semantic Space The various kinds of evidence I have presented suggest strongly that starting semantic space is flexible, not tightly structured. It does not limit the child to a fixed set of “Grammaticizable Notions” but rather makes available some alternative ways of distinguishing among objects, events, situations, and so on, and categorizing 685

them for purposes of linguistic expression. Equipped with these options and guided by clues from the linguistic input as to which criteria for classifying are important in the various grammatical subsystems of their language, children construct grammars that diverge semantically from the very beginning in the direction of the input language. In arguing against the notion of a “Basic Child Grammar” I have repeatedly stressed that there is diversity in children’s starting options. However, in rejecting the proposal that there is a single starting set of “Notions” I do not want to suggest that children are conceptually so flexible that all structure is provided by the input. Accordingly, I return in this section to the problem of constraint, with the goal of sketching out a hypothesis about starting semantic space that provides the latitude necessary to allow language-specific learning, at least within limits, but is still structured enough to account for the ways in which children depart from the semantic system displayed in the input. At least two important sources of constraint in starting semantic space can be envisioned. One I have mentioned earlier (see p. 336 and footnote 17): that the meaning distinctions children bring to bear on the learning of linguistic forms are linguistically sensible, in that they are the kinds of meaning distinctions that figure in the grammatical systems of natural languages. The second source of constraint, to be pursued in this section, involves differences in the relative accessibility of alternative categorization schemes.

686

4.3.1. Accessibility Hierarchies. Up to now, I have treated accessibility in all-or-none terms, arguing that if children can learn different systems for partitioning meaning from the beginning, starting semantic space must provide more than one option for categorizing a given conceptual domain. In fact, however, it is likely that the accessibility of alternative schemes is a matter of degree: Several options may in principle be available, in the sense that all of them can be used by young children at least under some conditions, but some may be more obvious or congenial to children than others. The idea that there is a hierarchy in the accessibility to children of various semantic notions is, of course, an important source of constraint in the “Basic Child Grammar” hypothesis. It might therefore appear that, having gone to much effort to show that children do not all start their construction of grammar in the same way, I am now returning to essentially the same hypothesis after all. This is not the case. The difference between the BCG hypothesis and what I am suggesting may be subtle, but it has important consequences for the conclusions we draw about how much the early stages of grammar acquisition are shaped by children’s own organizing predispositions as opposed to by their experience with the structure of a particular language. According to Slobin, alternative schemes for classifying meaning in a particular domain are not present from the beginning of the development of grammar, but only become available over time as the result of cognitive maturation. Those schemes that are present from the beginning—the initial set of “Grammaticizable 687

Notions”—are seen as the “most accessible”; they constitute the semantic basis for the “ideal form of human language” as it is reflected in Basic Child Grammar. Alternative, “less accessible” schemes emerge later; these consist, for example, of finer differentiations or higher-order groupings of the starting set of semantic categories. As these alternatives become available, they enable children gradually to modify their initial grammar in the direction of the specific semantic properties of the input language (see OP:REORGANIZATION, Slobin, 1985, p. 243.) The developmental unfolding of new ways of organizing meaning is indeed an important source of constraint, and must be recognized by any theory of language acquisition. However, our theory should also, I suggest, take account of differences in the relative accessibility to children of alternative schemes for partitioning meaning that are available from the beginning of acquiring the markers in particular grammatical subsystems. My proposal is that children can partition given conceptual domains in more than one way, but that, all else being equal, they may favor certain schemes (not necessarily just one per domain). This hypothesis preserves what is most attractive about the Basic Child Grammar hypothesis, the idea that some methods of categorizing a domain may be particularly likely or natural for children. However, “all else” is often not equal, since, among other things, children are exposed to different languages. The hypothesis thus allows, as the BCG hypothesis does not, for variation in the categorizational principles children initially rely on in learning grammatical forms. In particular, it allows for 688

the immediate influence on the child of the semantic structure of the input language. Differences in the accessibility of co-existing options for categorizing can perhaps best be thought of as variability both in the probability that children will “try out” these options when they are learning the grammatical forms of their language and in their persistence in trying to apply them. Highly accessible meaning distinctions are those that children are likely to entertain early in the course of working out what governs the distribution of forms in the input. Such distinctions should also be relatively robust, in the sense that children will often be able to identify patterns in the data that conform to them even when these patterns are marked somewhat inconsistently or with acoustically unsalient forms. In some domains there may be several equally likely and equally robust principles, and in others perhaps only one or two. Meaning distinctions of intermediate accessibility will have a lower priority and be less robust against noisy input, but if they are clearly and consistently marked in the language being learned, children will also be able to incorporate them into their grammars from the very beginning. Principles for classification that are not yet available because of the child’s cognitive immaturity will not be used at all in the early stages of acquisition. Children faced with a language that uses such a principle for categorizing a given conceptual domain—or that uses a principle of intermediate accessibility but does not mark the meanings defined by it clearly and consistently—may initially fall back on a more accessible principle, and make errors accordingly. 689

Invoking relative accessibility to explain the relative ease with which children learn alternative language-specific semantic systems is circular unless we can find some independent guideline to what is accessible (i.e. we first identify relative accessibility only on the basis of language acquisitional phenomena and then use it to explain these same phenomena). One intriguing possibility is that the relative accessibility for children of alternative schemes for partitioning meaning in a given conceptual domain is correlated with the frequency with which these schemes are instantiated in the languages of the world. Some categorization schemes turn up very frequently, others rarely. Given enough information, it is in principle possible to rank order alternative methods of categorizing a particular domain on the basis of their relative frequency across (unrelated) languages. It is plausible that relative frequency is correlated with “ease” or “naturalness” for the human mind; if so, it can serve as an index to the likelihood that particular bases for categorizing will arise spontaneously in children even in the absence of guiding linguistic input. Of course, even if relative frequency indeed reflects cognitive ease for humans, its power to predict what is easy for children is likely to be imperfect, since ways of categorizing that are easy and natural for adults may not yet be available to the very young child. However, predictions based on relative frequency can perhaps be adjusted and supplemented to some extent on the basis of our more general knowledge about the course of cognitive development. Unfortunately, tentative rank orderings of the frequency of various classification schemes are available at present 690

for very few semantic domains (for examples, see Ingram, 1978, on personal pronoun systems and Stassen, 1985, on comparatives). However, an important precedent for such analyses is found in studies of the relative frequency across languages of alternative word order patterns (e.g. Greenberg, 1966; Hawkins, 1982, 1983). These studies show that when several permutations of a word order pattern are possible, a few patterns account for the large majority of languages (e.g. SOV, SVO, VSO) while other patterns are rare or nonexistent (VOS, OVS, OSV). It is worth pointing out that, although it may be reasonable to call rare patterns “marked” or unusual in some way, there are no grounds for choosing just one of the remaining patterns as “most basic” (“unmarked,” “natural,” etc.). In the same way, information about the relative frequency of alternative schemes for classifying a semantic domain may suggest which schemes are cognitively “marked” or less natural, but they will not reveal any single “ideal” or “basic” system. 4.3.2. Predicting Errors on the Basis of Relative Accessibility. At last we are in a position to return to a question raised in an earlier section but postponed until the necessary background could be developed: is there a need for “mapping preference” OPs like UNIFUNCTIONALITY and ANALYTIC FORM in a theory of language acquisition? As discussed earlier, OP:UNIFUNCTIONALITY specifies that when a language uses the same form to encode different meanings in different contexts, children will try to eliminate this plurifunctionality by giving 691

each meaning a distinct marking. Similarly, OP:ANALYTIC FORM states that when a language uses a single form to encode more than one element of meaning simultaneously, children will try to spell out these meanings with separate morphemes. Some problems with these OPs, I argued, is that they predict that children should make many types of errors that have not been documented, and that children do make errors that they should not make if they are striving for unifunctionality and analyticity (for example, word substitutions that blur meaning distinctions previously observed, and the use of conflated, synthetic forms where analytic ones are needed). OPs that specify that children seek unifunctionality and analyticity are on shaky ground as long as we have no way of knowing when children will follow them and when they will not. Is there any principled way we can predict which forms in a language may give rise to errors, and what the direction of the errors will be? An approach to prediction is suggested by the hypothesis that, although human cognition is flexible and can categorize the same set of referents in alternative ways, some of these alternatives have a higher probability of being used than others. Specifically, we can predict that the likelihood and direction of errors in a given semantic domain is related to the position on the accessibility hierarchy of the categorization scheme employed by the child’ s language for that domain. If the language uses a scheme that is very common, we can hypothesize that it categorizes in a way congenial to children’s spontaneous ideas about what meaning 692

distinctions are important in the domain. Errors should therefore be infrequent or absent; if they do occur, they should reflect the transitory influence on performance of another categorization scheme that is also very common, not a rare one. In contrast, if the scheme is uncommon—hence, by hypothesis, further from children’s spontaneous categorizational preferences—we can predict errors in the direction of a more common scheme. For example, if the child’s language uses a single (plurifunctional) form for two meanings that language typically distinguish, we can expect the child at least occasionally to try to disambiguate. Conversely, if the child’s language distinguishes certain meanings that languages equally or more often do not differentiate, the child may overextend one form or make substitution errors that suggest that the distinction is easily overlooked. Sometimes the influence of a common scheme may show up in the very beginning, e.g. the child initially associates a certain grammatical marker with a semantic category that, although incorrect, is “easier” than the one required. In other cases the child learns the language-specific system from the start, but later makes occasional errors reflecting the passing influence of an alternative, highly accessible method of categorizing the domain. Late errors might be especially likely to reflect the developmental emergence of categorization schemes that are inaccessible to very young children because of their cognitive immaturity, but that are powerful organizing principles for older children and adults.

693

The psychological process that I assume would mediate many errors reflecting the influence of alternative, highly accessible categorization schemes is competition among semantically related words or construction patterns. As noted earlier (p. 322), the principle of competition alone is insufficient to account for children’s errors, since it does not tell us why certain forms regularly compete in a child’s speech while others apparently do not. However, if we combine competition with some hypotheses about how children conceptualize things at successive points in development—e.g. which referents strike them as similar; which meanings they conceive of as single units and which they break down into sub ele-ments—we may be able to specify the conditions under which competition takes place. For example, forms that distinguish between two meanings that are very often not differentiated in the world’s languages (i.e. are encoded with the same form) may compete and recurrently substitute for each other because their meanings are so closely related in human cognition that the intention to encode one of them will frequently activate the forms associated with both. In addition, perhaps, the intentions are not distinct enough to allow the form for one of them to be unerringly discriminated from the form for its sister intention in the stage of pre-speech editing. Conversely, forms that apply indiscriminately across meanings that languages commonly do distinguish are candidates for the process of differentiation that OP:UNIFUNCTIONALITY is designed to capture. In some cases this differentiation may be an incidental byproduct of competition: one of the meanings associated with a plurifunctional form is 694

also associated with another form, and both of these forms may be activated and often incorrectly combined when the child attempts to encode this meaning, but not when he encodes the other. In other cases children may more actively attempt to differentiate meanings that seem distinct to them by constructing idiosyncratic forms, as Karmiloff-Smith (1979a, 1979b) has argued. Finally, if the forms of a language draw meaning contrasts in a given domain in a very common way, errors of substitution (“lumping”) and differentiation (“splitting”) may be relatively rare. For example, forms that differentiate meanings that languages usually do distinguish—even though some languages collapse them—may rarely substitute for each other, either because the intention to encode one meaning has little power to activate forms associated with the others or because its meaning is distinct enough to allow the appropriate form to be readily singled out from any competing alternatives. With respect to synthetic versus analytic forms, we can hypothesize that forms that conflate meaning elements that languages rarely conflate will tend to compete with forms that express one or more of these meanings separately, resulting in substitutions or blends (as in the error genres illustrated in 1, 2, and 4 on p. 315). This is because the elements have—by hypothesis—a high degree of conceptual independence for speakers, and each has its own power to activate the forms associated with it in other construction patterns. Conversely, forms that conflate elements that very often are conflated will cause less difficulty: since the cluster of meaning elements is 695

apparently easily conceptualized as a single unit, only one form will tend to be activated. These various predictions are based on the hypothesis that certain conceptual dispositions or “preferences” for the organization of meaning arise in the child independently of the language being learned. Although the overt behaviors—i.e. whether or not errors are made, direction of errors—may vary depending on the semantic structure of the language, the underlying dispositions themselves are assumed to be the same across children. It is possible that, in addition to any such universal categorizational biases, preferences develop differently in different children in response to certain abstract properties of the input language. Crosslinguistic studies of syntax have shown that the various word order patterns of languages (e.g. the order of subject, verb, and object, of noun and modifier, and of noun and adposition) are not independent, but hang together in cognitively “harmonious” sets (Hawkins, 1982, 1983). Analogously, the particular semantic solutions languages have evolved for the partitioning of different conceptual domains may cluster together in cognitively coherent systems (see Stassen, 1985, for an interesting hypothesis along this line concerning comparative and temporal constructions). If such interconnections exist between the way different conceptual domains are partitioned for encoding in language (presumably reflecting very “deep” principles of cognitive organization), children could develop expectations about what meaning distinctions will be 696

important in one domain on the basis of what they have learned about other domains. This would facilitate rapid and error-free learning if the input language is harmonious. However, languages may be “disharmonious,” in the sense that the properties of one or more of their subsystems deviate from what would be expected, given the properties of the other subsystems. Children learning such a language might develop incorrect expectations that lead them to make errors; these errors would, in effect, bring a wayward subsystem into greater semantic harmony with the other subsystems. Such errors, if they occur, would probably not characterize the first stages of learning a discordant subsystem, but instead set in only later as learners developed a sense of how the various subsystems of their language hang together in a larger, semantically coherent pattern. 4.3.3. Identifying Categorizational Predispositions: Some Problems. In the preceding discussion I have concurred with Slobin that some of children’s language errors reflect the influence of categorizational principles that are especially “natural” or accessible to children. However, the effects of the learner’s pre-dispositions for the organization of meaning will be subtle and extremely difficult to identify with confidence. One important problem is simply that preferences for categorizing in certain ways will interact with, and often be obscured by, the formal complexity for children of the linguistic devices used to encode the relevant meanings. As Slobin (1973) has shown, some devices are easier for children than others, independent of the meanings they 697

express. This means that a semantic categorization scheme that is in principle somewhat less favored might nevertheless be adopted faster or with fewer errors than a more favored scheme, if the markers through which it is manifested in a particular language are formally easier for children. A second, even knottier problem is that when children use a linguistic form in connection with an inappropriate semantic category, either consistently or occasionally, the ultimate source of their error is often difficult to isolate: errors that at first seem to reflect a nonmodeled—hence, by inference, spontaneously generated—way of classifying can, on closer inspection, often be interpreted equally well as extensions of a semantic pattern conventionally associated with one or more forms that the child already knows or is learning. For example, consider English-speaking children’s use of spatial words like behind in contexts where temporal words like after are needed (see pp. 335– 336). These errors might arise spontaneously simply because it is “natural” for humans to use a spatial framework for conceptualizing temporal notions. However, children in fact receive considerable linguistic encouragement for this conceptualization, since many English words are conventionally applied to both spatial and temporal meanings: e.g. AT school, AT four o’clock; OVER the river, OVER the next few months; a SHORT boy, a SHORT nap. In learning these words in both their spatial and temporal senses, children are in effect continually invited to see an analogy between space and time. If they pick up on this invitation, it would not be surprising if 698

the (learned) analogy also began to affect the use of words for which it happens to be inappropriate (see Bowerman, 1982c, 1983a). As a second example, take the following error by a German child, cited by Mills (1985): Die Grossmama ZU den Affe ‘The grandmother TO the monkey’ (=‘The monkey’s grandmother’). Here a locative preposition, zu ‘to’, is extended to a possessive meaning; von ‘of would have been appropriate. Slobin (1985) interprets this error as stemming from—and, in turn, as evidence for— the conflation of location and possession in starting semantic space. This explanation might be correct. However, we should note that adult German in fact uses zu in many everyday contexts to mark a meaning better described as ‘belonging together with’ or ‘going with’ than as ‘locative goal’. In many such constructions zu and von are interchangeable: e.g. ‘the top ZU (VON) this bottle,’ ‘the cover ZU (VON) this book,’ ‘the hat ZU (VON) the boy’ (this last example would be appropriate in the context of assembling a jigsaw puzzle, for instance); ‘this belt PASST ZU (= goes with) this dress’; ‘this father GEHÖRT ZU (=belongs with) these children’ (e.g. pointing out pictures in a book). Since the error of using zu in place of possessive von occurred quite late in the speech of the child discussed by Mills (age 3;7), it is likely that he was already familiar with the ‘belonging together’ sense of zu. His error might therefore simply have reflected his (overly) productive application of a pattern displayed in his language rather than an inherent tendency to conflate possession and location.

699

The confounding of possible explanations seen in these two examples is pervasive, since principles of classification that children may prefer on a non-linguistic basis very often turn out to be functional somewhere in the input language, sometimes in connection with forms closely related to the ones on which children err and sometimes even in connection with specific usages of the same forms. Of course, in some of these cases linguistic patterns may simply reinforce children’s inherent preferences; i.e. both could work together to co-determine particular error types. However, it is a serious methodological problem to determine when this is the case and when the semantic categories displayed in the input are in fact critically responsible for suggesting to learners what elements of their experience should be regarded as similar for purposes of linguistic expression. Crosslinguistic comparisons are essential for resolving this confounding. For example, if errors reflecting the inappropriate conflation of time and space, possession and location, and so on, are made only by children learning languages in which these meanings are conventionally conflated in the use of certain common forms, we will have to give more credit to children’s experience with the linguistic input and less to the intrinsic structure of starting semantic space in explaining the salience to children of these categorization schemes. On the other hand, of course, if errors of these types occur regardless of whether the relevant semantic organization is displayed in the input language, then the hypothesis that children start language acquisition with a prestructured semantic space is correspondingly strengthened. 700

4.4. Priorities for Future Crosslinguistic Research The previous section has ended with a theme to which I have repeatedly returned in this chapter: To what extent do the patterns we observe the speech of young children reflect their inherent dispositions for organizing linguistic material in certain ways, as opposed to their experience with the structural properties of a specific input language? On the basis of the crosslinguistic research presented in these volumes and elsewhere, Slobin (1985) has concluded that children’s early grammars show the same basic structure everywhere. From this he infers that predispositions for particular ways of partitioning meaning and mapping it into linguistic form play a critical role in the early stages of language acquisition. Although I have agreed with Slobin that an optimal theory of language acquisition will need to take into account certain predispositions for the organization of meaning, I have argued that these predispositions are neither as strong nor as single-channeled as Slobin has proposed. For example, counter to the Basic Child Grammar hypothesis, children often show language-specific patterns from the outset of learning important grammatical subsystems. Further, although certain errors at first may seem to reflect organizing principles that have no model and hence must have arisen spontaneously in the child, closer examination often reveals plausible within-language sources for them: e.g. the child’s growing grasp of syntactic regularities or principles of semantic categorization displayed in the

701

linguistic input, or competition at the moment of speech between alternative forms the child has learned for expressing the same or similar meanings. It is striking that even after almost two decades of intensive cross linguistic research, possible explanations for many acquisitional phenomena are still so heavily confounded. This testifies not to the inutility of the crosslinguistic method as a technique for disentangling the roles of nature and nurture in language acquisition, but to the extraordinary complexity of the problem. Although many of the relatively global questions of an earlier era have been answered (e.g. do all children prefer fixed word order regardless of the rigidity of word order in the input language?), new and increasingly difficult sources of confounding keep coming into view. At present, the problem is particularly severe with respect to interpreting the causes and implications of errors in children’s spontaneous speech, and this is why I have devoted so much attention to this issue. Although the Operating Principles approach has not yet provided us with definitive answers, it has served as an invaluable framework within which critical issues can be conceptualized and investigated. As discussed in an earlier section, operating principles and other proposals within the Operating Principles approach have been developed inductively, primarily through comparison of the spontaneous speech of children learning different languages. This phase of inductive research has served several important functions. For example, it has allowed us to identify and explore a number of significant acquisitional phenomena that might well have been 702

missed if research efforts had been narrowly directed toward hypothesis-testing. Additionally, the development of a framework that can accommodate and give meaning to scattered bits of data from many languages has been enormously important in fostering among child language scholars—many of whom have worked in relative isolation in their countries—a shared frame of reference and a sense of joint participation in a cooperative scientific enterprise. Although we should not neglect inductive techniques in research of the coming years, we need now to become more rigorous, supplementing observation and inference with explicit hypothesis-testing. In particular, proposed operating principles and hypotheses about children’s categorizational predispositions need to be subjected to critical evaluation. It is not enough to establish that there are data that conform to the proposals. We need also to draw out the predictions of the proposals for the acquisition of specific languages and see how systematically they are met, whether apparent counterevidence can be explained in some reasonable way or should cause us to modify or reject a hypothesis, whether the phenomena cited in support of various proposals can be explained plausibly in other ways, and so on. Testing the hypotheses of the Operating Principles approach will require crosslinguistic comparisons to be more carefully preplanned than has typically been necessary before. The languages to be used in testing a particular proposal will need to be selected so that they differ critically with respect to rather fine structural 703

details. I would like to urge that the most pressing need in this respect is for studies that take crosslinguistic variation in semantic structure seriously. In spite of the general interest in meaning shown by our field in recent years, variation among languages in the makeup and scope of grammatically-relevant semantic categories has been persistently neglected in crosslinguistic studies of grammatical development. In general, the way in which languages organize meaning has not been regarded, as an integral part of their structure, equivalent in status to syntactic or morphological structure and comparable to these in its potential to influence rate of acquisition, likelihood of errors, and so on (Bowerman, 1985). For example, in the Operating Principles approach, grammatical variation is recognized as a critical independent variable, either to be controlled while the role of cognitive complexity is evaluated, or permitted so that its impact on acquisition can be studied (see p. 305 and footnote 3). In sharp contrast, variation in the semantic categories associated with the grammatical forms of languages has been virtually ignored. Thus, in studies designed to assess the effects on acquisition of crosslinguistic differences in grammatical form, “meaning” is typically controlled simply by choosing for comparison forms that are translative-equivalents across at least a portion of their usage range (e.g. in, on, and their counterparts in other languages). No attention is paid beyond this rough match to differences in the way these forms, together with other closely related forms, divide up the domain of meaning over which they operate into contrasting categories. Yet these differences may have important consequences for 704

language acquisition. Further—and still more important— semantic organization has not been treated as an independent variable to be carefully contrasted while grammatical form and general conceptual domain is controlled. The neglect of semantic variation in crosslinguistic studies of language acquisition reflects, I think, the difficulty our field has had in distinguishing meaning as it is structured by language from nonlinguistic, cognitive organization. In much recent theorizing about language acquisition, children are seen as possessing powerful cognitive capacities that enable them to organize and interpret their experiences independently of language. When language starts to come in, according to this view, it does not introduce new meanings, but simply allows children to express those meanings they have already formulated. The emphasis in this model is on the cognitive understanding that all children are assumed to share; correspondingly, there is little attention to crosslinguistic variation in meaning structure, or to the process by which children learn to the specific way in which their language categorizes elements of experience (see Bowerman, 1985; Schlesinger, 1977; and Slobin, 1979 for discussion). This model is not incorrect so much as it is seriously incomplete. Recognition of the role in language acquisition of conceptual development has been one of the important advances of our field in the last two decades, and I do not dispute that children’s acquisition of linguistic forms is supported by their general cognitive understanding of the relevant concepts. 705

Locative prepositions will not be learned until the child realizes that objects can be spatially related in various ways, and subordinate clauses with when, if, and so on will not come in until the child can conceive of contingent relations between events. Nevertheless, we must recognize that the child’s ability to understand and interpret spatial configurations, contingent relations, etc. on a nonlinguistic basis is not isomorphic with the ability to categorize these relations in the way required by the language being learned. We therefore cannot be satisified with a theory that stops with the observation that meanings in some sense precede the acquisition of the forms that encode them. We need to go beyond this to determine how children work out the principles of semantic categorization that are functional in their language. Crosslinguistic comparisons will be essential to this effort. Only by studying how children approach language systems that differ in their organization of what is, at a deep level, the “same” conceptual material can we begin to discover how language-learners construct a highly structured and language-specific meaning system from their nonlinguistic understanding of daily experience. I hope research of the coming years will pay more attention to this central and fascinating problem of language acquisition.

ACKNOWLEDGMENTS I would like to thank Eve Clark, Herbert Clark, Mary L. Foster, Julie Gerhardt, Susan Sugarman, and Lee Ann

706

Weeks for their helpful comments about both content and form in an earlier draft of this chapter, and Edith Sjoerdsma and Lee Ann Weeks for their help in preparing the manuscript. In developing the ideas expressed in the section entitled “Basic Child Grammar: Is Only One Way of Structuring Semantic Space ‘Basic’?” I have benefitted greatly from discussions over the years with Dedre Gentner, Dan Slobin, and Leonard Talmy. My interest in the problem of semantic categorization and the conviction that there are often alternative, equally “natural” methods for partitioning the same semantic domain I owe originally to the insights and teaching of Roger Brown. Of course, none of these colleagues necessarily agrees with the views I present here.

REFERENCES Aksu, A. A. Aspect and modality in the child’s acquisition of the Turkish past tense. Unpublished doctoral dissertation. University of California at Berkeley, 1978. Aksu-Koç, A. A., & Slobin, D. I. The acquisition of Turkish. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Andersen, E. S. Lexical universals of body part terminology. In J. H. Greenberg (Ed.), Universals of human language, Vol. 3: Word Structure. Stanford: Stanford University Press, 1978.

707

Andersen, H. Abductive and Language, 1973, 49, 765–793.

deductive

change.

Antinucci, F., & Miller, R. How children talk about what happened. Journal of Child Language, 1976, 3, 167–189. Bellugi, U. The acquisition of negation. Unpublished doctoral dissertation, Harvard University, 1967. Bennett, D. C. Spatial and temporal uses of English prepositions: An essay in stratificational semantics. London: Longman Group, 1975. Berman, R. A. Child language as evidence for grammatical description: Preschooler’s construal of transitivity in the verb system of Hebrew. Linguistics, 1980, 18, 677–701. Berman, R. The acquisition of Hebrew. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Bickerton, D. Roots of language. Ann Arbor, MI: Karoma, 1981. Bloom, L. Language development: Form and function in emerging grammars. Cambridge, MA: MIT Press, 1970. Bloom, L., Lifter, K., & Hafitz, J. Semantics of verbs and the development of verb inflection in child language. Language, 1980, 56, 386–412.

708

Bowerman, M. Early syntactic development: A cross linguistic study, with special reference to Finnish. Cambridge, Eng.: Cambridge University Press, 1973. Bowerman, M. Commentary on structure and variation in child language. In L. Bloom, P. Light-bown, & L. Hood (Eds.), Monographs of the Society for Research in Child Development, 1975, 40(2), serial No. 160. Bowerman, M. Semantic factors in the acquisition of rules for word use and sentence construction. In D. M. Morehead & A. E. Morehead (Eds.), Normal and deficient child language. Baltimore, MD: University Park Press, 1976. Bowerman, M. Systematizing semantic knowledge: Changes over time in the child’s organization of word meaning. Child Development, 1978, 49, 977–987. Bowerman, M. The structure and origin of semantic categories in the language-learning child. In M. L. Foster & S. H. Brandes (Eds.), Symbol as sense: New approaches to the analysis of meaning. New York: Academic Press, 1980. Bowerman, M. The child’s expression of meaning: Expanding relationships among lexicon, syntax, and morphology. In H. Winitz (Ed.), Native language and foreign language acquisition. New York: The New York Academy of Sciences, 1981. Bowerman, M. Evaluating competing linguistic models with language acquisition data: Implications of 709

developmental errors with causative verbs. Quaderni di Semantica, 1982, 3, 5–66. (a) Bowerman, M. Reorganizational processes in lexical and syntactic development. In E. Wanner & L. R. Gleitman (Eds.), Language acquisition: The state of the art. Cambridge, Eng.: Cambridge University Press, 1982. (b) Bowerman, M. Starting to talk worse: Clues to language acquisition from children’s late speech errors. In S. Strauss (Ed.), U-shaped behavioral growth. New York: Academic Press, 1982. (c) Bowerman, M. Hidden meanings: The role of covert conceptual structures in children’s development of language. In D. R. Rogers & J. A. Sloboda (Eds.), Acquisition of symbolic skills. New York & London: Plenum, 1983. (a) Bowerman, M. How do children avoid constructing an overly general grammar in the absence of feedback about what is not a sentence? Papers and Reports on Child Language Development. (Stanford University Dept. of Linguistics), 1983, 22. (b) Bowerman, M. Beyond communicative adequacy: From piecemeal knowledge to an integrated system in the child’s acquisition of language. In K. Nelson (Ed.), Children’s language, Vol. 5. Hillsdale, NJ: Lawrence Erlbaum Associates, 1985.

710

Bowerman, M. First steps in acquiring conditionals. In E. Traugott & C. A. Ferguson (Eds.), On conditionals. Cambridge, Eng.: Cambridge University Press, in press. Braine, M. D. S. Children’s first word combinations. Monographs of the Society for Research in Child Development. 1976, 41 (1), Serial no. 164. Brown, R. How shall a thing be called? Psychological Review, 1958, 65, 14–21. Brown, R. A first language: The early stages. Cambridge, MA: Harvard University Press, 1973. Brown, R., Cazden, C., & Bellugi, U. The child’s grammar from I to III. In J. P. Hill (Ed.), Minnesota Symposium on Child Development (Vol. 2). Minneapolis: University of Minnesota Press, 1968. Bybee, J. L. Diagrammatic iconicity in stem-inflection relations. Paper presented to Symposium on Iconicity in Grammar, Stanford University, 1983. (To be published in volume edited by J. Haiman; Amsterdam: John Benjamins.) Bybee, J. L. Morphology: A study of the relation between meaning and form. Amsterdam: John Benjamins, 1985. Chomsky, N. Review of Skinner (1957). Language, 1959, 35, 26–58.

711

Chomsky, N. Aspects of the theory of syntax. Cambridge, MA: MIT Press, 1965. Chomsky, N. Lectures on government and binding. Dordrecht: Foris, 1981. Clancy, P. The acquisition of Japanese. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Clancy, P., Jacobson, T., & Silva, M. The acquisition of conjunction: A cross-linguistic study. Papers and Reports on Child Language Development (Stanford University, Dept. of Linguistics), 1976, 12, 71–80. Clark, E. V. Universal categories: On the semantics of classifiers and children’s early word meanings. In A. Juilland (Ed.), Linguistic studies offered to Joseph Greenberg. Saratoga, CA: Anma Libri, 1977. Clark, E. V. Lexical innovations: How children learn to create new words. In W. Deutsch (Ed.), The child’s construction of language. London: Academic Press, 1981. Clark, E. V. The acquisition of Romance, with special reference to French. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985.

712

Comric, B. Conditionals: A typology. In E. Traugott & C. A. Ferguson (Eds.), On conditionals. Cambridge, Eng.: Cambridge University Press, in press. Denny, J. P. Locating the universals in lexical systems for spatial deixis. In D. Farkas, W. M. Jacobsen, & K. W. Todrys (Eds.), Papers from the parasession on the lexicon, Chicago Linguistic Society. Chicago: University of Chicago Press, 1978. Denny, J. P. The ‘extendedness’ variable in classifier semantics: Universal features and cultural variation. In M. Mathiot (Ed.), Ethnolinguistics: Boas, Sapir and Whorf revisited. The Hague: Mouton, 1979. deVilliers, P. A., & de Villiers, J. G. Form and function in the development of sentence negation. Papers and Reports on Child Language Development (Stanford University, Dept. of Linguistics), 1979, 17, 57–64. Dixon, R. Ergativity. Language, 1979, 55, 59–138. Eisenberg, A. R. The emergence of markers of current relevance. Papers and Reports on Child Language Development. (Stanford University, Dept. of Linguistics), 1981, 20, 44–51. Ervin, S. Imitation and structural change in children’s language. In E. H. Lenneberg (Ed.), New directions in the study of language. Cambridge, MA: MIT Press, 1964.

713

Fodor, J. A. How to learn to talk: Some simple ways. In F. Smith & G. A. Miller (Eds.), The genesis of language: A psycholinguistic approach. Cambridge, MA: MIT Press, 1966. Gathercole, V. The mass count distinction: Children’s use of morpho syntactic vs. semantic approaches. Papers and Reports on Child Language Development (Stanford University, Dept. of Linguistics), 1983, 22, 58–65. Givon, T. Function, structure, and language acquisition. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 2). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Goodluck, H. Children’s grammar of complement subject interpretation. In S. L. Tavakolian (Ed.), Language acquisition and linguistic theory. Cambridge, MA: MIT Press, 1981. Gordon, P. The acquisition of syntactic categories: The case of the count/mass distinction. Unpublished doctoral dissertation, MIT, 1982. Greenberg, J. Some universals of grammar with particular reference to the order of meaningful elements. In J. H. Greenberg (Ed.), Universals of Language. Cambridge, Mass.: MIT Press, 1966. Haiman, J. Conditionals are topics. Language, 1978, 54, 564–589.

714

Haiman, J. Hua: A Papuan language of New Guinea. In T. Shopen (Ed.), Languages and their status. Cambridge, MA: Winthrop Publishers, 1979. Hakuta, K. Grammatical description versus configurational arrangement in language acquisition: The case of relative clauses in Japanese. Cognition, 1981, 9, 197–236. Hawkins, J. Cross-category harmony, X bar and the predictions of markedness. Journal of Linguistics, 1982, 18, 1–35. Hawkins, J. Word order universals. New York: Academic Press, 1983. Ingram, D. Typology and universals of personal pronouns. In J. H. Greenberg (Ed.), Universals of human language, Vol. 3: Word Structure. Stanford, CA: Stanford University Press, 1978. Johnston, J.R. Cognitive prerequisites: The evidence from children learning English. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 2). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Johnston, J. R., & Slobin, D. I. The development of locative expressions in English, Italian, Serbo-Croatian and Turkish. Journal of Child Language, 1979, 6, 529–545. Kaplan, A. The conduct of inquiry: Methodology for behavioral science. San Francisco: Chandler, 1964.

715

Karmiloff-Smith, A. The interplay between syntax, semantics, and phonology in language acquisition processes. In R. N. Campbell & P. T. Smith (Eds.), Recent advances in the psychology of language. New York: Plenum Press, 1978. Karmiloff-Smith, A. Afunctional approach to child language. Cambridge, Eng.: Cambridge University Press, 1979. (a) Karmiloff-Smith, A. Micro- and maerodevelopmental changes in language acquisition and other representational systems. Cognitive Science, 1979, 3, 91–118. (b) Kuczaj, S. A. Children’s judgments of grammatical and ungrammatical irregular past tense forms. Child Development, 1978,49, 319–326. Laver, J. D. M. The detection and correction of slips of the tongue. In V. A. Fromkin (Ed.), Speech errors as linguistic evidence. The Hague: Mouton, 1973. Lucy, J. A. An empirical approach to the Whorfian question. Paper presented to the Northwestern University Psycholinguistics Colloquium. Evanston, II., 1981. Lyons, J. Semantics. (Vol. 2). Cambridge, Eng.: Cambridge University Press, 1977. Macnamara, J. Names for things. Cambridge, MA: MIT Press, 1982.

716

MacWhinney, B. The acquisition of morphophonology. Monographs of the Society for Research in Child Development, 1978, 43(1,2), Serial no. 174. MacWhinney, B. Hungarian language acquisition as an exemplification of a general model of grammatical development. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 2). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Maratsos, M. P. The use of definite and indefinite reference in young children: An experimental study in semantic acquisition. Cambridge, Eng.: Cambridge University Press, 1976. Maratsos, M. P. How to get from words to sentences. In D. Aaronson & R. Rieber (Eds.), Perspectives in psycholinguistics. Hillsdale, NJ: Lawrence Erlbaum Associates, 1979. McNeill, D. The creation of language by children. In J. Lyons & R. J. Wales (Eds.), Psycholinguistic Papers. Edinburgh: Edinburgh University Press, 1966. Mills, A. E. The acquisition of German. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Norman, D. A. Categorization of Psychological Review, 1981, 88, 1–15.

717

action

slips.

Ochs, E. Variation and error: A sociolinguistic approach to language acquisition in Samoa. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Otsu, Y. Universal grammar and syntactic development in children: Toward a theory of syntactic development. Unpublished doctoral dissertation, MIT, 1981. Roeper, T. The role of universals in the acquisition of gerunds. In E. Wanner & L. R. Gleitman (Eds.), Language acquisition: The state of the art. Cambridge, Eng.: Cambridge University Press, 1982. Rosch, E., Mervis, C., Gray, W., Johnston, D., & Boyes-Braem, P. Basic objects in natural categories. Cognitive Psychology, 1976, 8, 382–439. Schlesinger, I. M. The role of cognitive development and linguistic input in language acquisition. Journal of Child Language, 1977,4, 153–169. Slobin, D. I. Comments on McNeill’s ‘Developmental psycholinguistics’. In F. Smith & G. A. Miller (Eds.), The genesis of language: A psycholinguistic approach. Cambridge, MA: MIT Press, 1966. Slobin, D. I. Universals of grammatical development in children. In W. J. M. Levelt & G. B. Flores d’Arcais (Eds.), Advances in psycholinguistic research. Amsterdam: North-Holland, 1970.

718

Slobin, D. I. Cognitive prerequisites for the development of grammar. In C. A. Ferguson & D. I. Slobin (Eds.), Studies of child language development. New York: Holt, Rinehart & Winston, 1973. Slobin, D. I. The role of language in language acquisition. Unpublished paper, University of California, Berkeley, 1979. Slobin, D. I. Universal and particular in the acquisition of language. In E. Wanner & L. R. Gleitman (Eds.), Language acquisition: The state of the art. Cambridge, Eng.: Cambridge University Press, 1982. Slobin, D.I. Crosslinguistic evidence for the Language-Making Capacity. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 2). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Smoczyńska, M. The acquisition of Polish. In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1). Hillsdale, NJ: Lawrence Erlbaum Associates, 1985. Snow, C. E. Mother’s speech research: From input to interaction. In C. E. Snow & C. A. Ferguson (Eds.), Talking to children. Cambridge, Eng.: Cambridge University Press, 1977. Stassen, L. Comparative and conjunction: An essay in universal grammar. Oxford: Basil Blackwell, 1985.

719

Stemberger, J. P. Syntactic errors in speech. Journal of Psycholinguistic Research, 1982, 11, 313–345. Stephany, U. Verbal grammar in modern Greek early child language. In P. S. Dale & D. Ingram (Eds.), Child language: An international perspective. Baltimore, MD: University Park Press, 1981. Tavakolian, S. L. (Ed.). Language acquisition and linguistic theory. Cambridge, MA: MIT Press, 1981. Traugott, E. C. On the expression of spatio-temporal relations in language. In J. H. Greenberg (Ed.), Universals of human language. Vol. 3: Word structure. Stanford, CA: Stanford University Press, 1978. Watkins, L. Shape versus position: Classiftcatory verbs in North America. Paper presented at the Linguistic Society of America Annual Meeting, Philadelphia, 1976. Weist, R. M., Wysocka, H., Witkowska-Stadnik, K., Buczowska, E., & Konieczna, E. The defective tense hypothesis: On the emergence of tense and aspect in child Polish. Journal of Child Language, 1984, 11, 347–374. 1

This hypothesis credits the language-learning child not only with operating principles but also, as we discuss later, with the spontaneous tendency to classify the elements of experience into certain categories of meaning for purposes of linguistic expression. Through this proposal “content” has reentered the “process” approach. However, this type of content differs from 720

what researchers previously had in mind in discussing prestructure in the inherent capacity for language acquisition, in that it concerns semantic organization rather than syntactic form. 2

For instance, the relatively late acquisition of a linguistic device would be compatible with formal simplicity for the child as long as the underlying communicative intention also emerged relatively late, for whatever reason. 3

Conversely, of course, if formal difficulty is held constant across a set of forms expressing related meanings, variations in time of acquisition can be attributed to differences in the difficulty of the meanings. This reasoning has been used in studies of the acquisition of locative markers (Johnston & Slobin, 1979) and connectives (Clancy, Jacobsen, & Silva, 1976), among others. 4

The role of frequency is still controversial. However, even investigators who consider it relatively unimportant acknowledge that extremely frequent or infrequent modeling probably influences time of acquisition (e.g. Brown, 1973), and frequency also seems to play a significant role in determining which formal devices a child will learn first to express a given meaning when the language provides more than one option (Snow, 1977). 5

For example, with respect to the problem of unanalyzed forms, morphological overgeneralizations like foots, corned, and uncapture (=release) indicate that the child understands the internal structure of legitimate forms of 721

the same pattern (Bowerman, 1982b; Ervin, 1964). Conversely, errors like It’s tell, It’s have wheels, that a my book (Brown, 1973) and Lemme me do that. Show me me that (from my unpublished records) indicate that complex strings (it’ s, that a, let me, show me) are unanalyzed by the speaker. For further evidence on how children’s errors give clues to underlying categories and rules, see, e.g. Berman, 1980; Bowerman, 1982a, 1982b, 1982c, 1983a, 1983b; and Clark, 1981). 6

This challenge was taken particularly seriously by Smoczyńska, with a number of insightful and provocative results. 7

See Smoczyńska (1985) for a related argument that solutions based on meaning may enjoy no special privileges in initial child grammars. She proposes that consistency is what children look for, and a consistent relation between form and meaning is only one instance of this. 8

Notice that if this is true, then, contrary to Slobin’s proposal, we cannot safely take children’s plurifunctional uses of a form as a guide to what they regard as a single meaning. It might be that they in fact regard the meanings as distinct but have not yet realized that the form is the same. 9

Examples of common (American) English words that are clearly plurifunctional include to know, to mean, to live, and can. These words can apply across a semantic range that is divided and assigned to two words in a number of other languages. To know means both “to 722

know something,” e.g. a fact, as in “I know how old you are,” and “to be acquainted with,” as in “I know Mary” (Dutch weten versus kennen). To mean means both “to signify,” as in “This word means … ” and “to intend,” as in “I mean … ” (Dutch betekenen versus bedoelen). To live means both “to dwell,” as in “John lives in New York” and “to be alive, exist,” as in “How long do turtles live?” (Dutch wonen versus leven). Finally, can means both “be able to,” as in “I can swim,” and “have permission to” (=may), as in “Can I have a cookie?” (Dutch kunnen versus mogen). If Englishspeaking children strive to eliminate plurifunctionality, they should at some point try to disambiguate these meanings, but there is no evidence that they do; can, in fact, persists stubbornly with both meanings despite efforts by parents and teachers to persuade children to substitute may for one of its senses. 10

Competition is assumed to affect the speech of adults as well as children; see, e.g. Laver, 1973; Stemberger, 1982). An interesting difference between children and adults is that adults often self-correct their own competition errors and recognize them immediately as errors when asked to reflect on them; children, in contrast, correct at least certain kinds of errors very infrequently and may be unaware that they are errors (even though they may self-correct other kinds of errors) (Bowerman, 1978; see also Kuczaj, 1978, on children’s explicit acceptance of forms like ated). This might seem an argument against interpreting the errors of children in terms of competition. However, it is not necessary within the competition approach to assume that the speaker has firm ideas about which of the competing alternatives is 723

right, or knows that certain forms cannot be combined with each other; this may well be an independent kind of knowledge that develops with age. 11

For example, the clause-external no found in the speech of some children learning English may have its source in input sentences with anaphoric negatives like No, don’t cut two holes (child version: no (me) cut two holes) (see deVilliers and deVilliers, 1979, for discussion) and Kendall doesn’t want to take a bath, no . . oo . . oo! (child version: Kendall take bath no) (construction of this latter type were common in the speech of my subject Kendall and her parents; from my unpublished records). 12

This interpretation is adequately provided for within the OP system by OP:EXTENSION: “If you have discovered the linguistic means to mark a Notion in relation to a word class or configuration, try to mark the Notion on every member of the word class or every instance of the configuration, and try to use the same linguistic means to mark the notion.” This OP reflects a very general principle, not limited to language, such as “be consistent” or “avoid exceptions” (see also Smoczyńska, 1985). It does not say anything about what kinds of mappings children prefer, independently of the language they are learning, and it is therefore to be favored over OPs that do call on such preferences whenever both are applicable. 13

One important problem with working out the notion of ordered OPs is the following. In the OP approaches of both Slobin (1985), and MacWhinney (1985), it is 724

essential that children be able to distinguish between adequate and inadequate solutions to linguistic problems. When an adequate one is achieved the child is held to stop working on the problem and the strategi(es) used in finding it are strengthened; in contrast, when the solution is inadequate the child repeatedly comes back to the problem, applying, by hypothesis, sequentially ordered strategies until it is solved (OP:PERSISTENCE, Slobin, 1985). How does the child know whether his solution is adequate or whether he needs to try the next OP on the list or the next clause in the same OP? According to Slobin and MacWhinney, the child’s critical cue that a solution is inadequate is his detection, through monitoring, of mismatches, either (a) between what he says or would have said in particular contexts and what fluent speakers say, or (b) between the input and his ability to parse and make sense of it. Mismatches are surely important, but they can by no means be the only factor that spurs the child to do further analysis. In particular, further work clearly takes place in situations of the following two kinds, even in the absence of mismatches: 1. Children sometimes construct overly general rules that account for all acceptable utterances of a given pattern (e.g. passives, shifted datives) and err only in that they also generate unacceptable ones (e.g. she said me ‘no’ = she said ‘no’ to me). In these situations, children can predict and fully account for all exemplars from other speakers, and they receive little or no negative 725

feedback about their own productions. According to the “mismatch” account they should consider the problem solved. Nevertheless, they eventually identify words or semantic classes of words to which the rule does not apply, and adjust the rule accordingly (see Bowerman, 1983b, for discussion). What motivates their further work on the problem? 2. Children sometimes move from the correct, flexible, and adultlike use of certain construction patterns to incorrect, “mismatching” forms. This often seems to reflect the formulation of new, more abstract rules or relationships that link forms that were previously independent in the child’s grammar; however, a grasp of these relationships is not needed for either normal production or comprehension (Bowerman, 1982b, 1982c; Karmiloff-Smith, 1978, 1979b). Again, what motivates the child to look for a new “solution” to what is not in any clear sense a “problem”? An adequate theory of language acquisition must be able to account for cases like these where children find structure “because it is there,” and not because they cannot resolve “mismatches” otherwise. Any attempt to order OPs in terms of when they will apply must come to grips with this phenomenon. 14

Here the lessons of the pivot grammar should not be forgotten. The pivot grammar was the reigning model in the 1960s of how children develop syntax. Like “Basic Child Grammar,” the model was a composite, built up from the findings of several researchers who had studied different children. When the predictions of the model 726

were finally tested carefully against data from individual children, it turned out that no child observed its rules (Bowerman, 1973; Brown, 1973). The model’s failure to predict or describe what children’s utterances actually looked like, coupled with proposals for a “deeper” and more satisfying level of description (Bloom, 1970), quickly led to the model’s demise. 15

This problem is related to, but not identical with, the issue of whether children have inherent mapping preference, which was discussed in the previous section. There the question was not what children’s meanings are, but whether children have intrinsic ideas about how meanings–whatever these may consist of—should be encoded in language (e.g. one form for each separate meaning element?). In the present section I am concerned with whether there is prestructuring, independent of linguistic experience, of the child’s categories of meaning themselves. 16

In a footnote (no. 13), Slobin (1985) provides what is in fact a very important qualification on how deterministically BCG imposes its semantic structure on the input. Slobin suggests that children may give separate treatment to semantic notions that BCG is in general prepared to conflate if their language distinguishes these notions with linguistic devices that, on formal grounds, are equally easy for children to learn, especially if these devices are highly differentiated (e.g. involve opposite word order patterns). If the implications of this footnote were carefully worked through, I think the BCG hypothesis would have to be weakened so much to allow influence from the semantic structure of 727

the input language on the child’s initial organization of semantic space that the position reached would be very similar to the one I am advocating. 17

For example, Clark (1977) shows that the perceptual properties on the basis of which young children overextend words for objects (e.g. roundness, longness) are very similar to those that repeatedly turn up as important in the noun classifier systems of the languages of the world. (A noun classifier is a morpheme used obligatorily in constructions involving (depending on the language) counting, predicating motion or manipulation of, or locating objects, e.g. “Nine round-things gourds.”) Similarly, in Bowerman (1980) I proposed that one child’s use of night night in connection with the horizontal position of objects that are normally extended vertically (e.g. a Christmas tree) reflected a sensitivity to features of extendedness and spatial orientation that play a relatively small role in the semantic system of English (e.g. stand versus lie) but that are of central importance to the classifier systems of many languages (e.g. Denny, 1979; Watkins, 1976). I also suggested that overextensions such as kick applied to an action of throwing, sleeves to pant legs, and ankles to wrists reflect an equation between upper and lower extremities that, although not formally encoded in English, is routine in many other languages (e.g. the word for “finger” is often also used for “toe”; Andersen, 1978). See also p. 336 of this chapter and Bowerman (1978, 1983a) for further examples and discussion of how children’s incorrect form-meaning mappings often reflect linguistically sensible distinctions.

728

18

To forestall possible confusion, let me distinguish immediately between two ways in which we can speak of starting semantic space as being “biased.” If children have their own ideas, independent of linguistic input, about how to structure semantic space, then in one sense their initial organization of semantic space can be said to be “biased.” Slobin has made a number of arguments about biases or preferences of this sort. However, he also hypothesizes that these initial preferences are not biased toward any particular language; they can develop equally easily in a variety of directions. In this sense, then, they are neutral or “unbiased.” Since the following discussion is about whether children’s early grammars show language specificity, I will be using the terms “biased” and “unbiased” in this latter sense. 19

A drawback of reliance on error data, as Givón (1985) has pointed out, is that it draws attention to areas of grammar in which children have trouble at the expense of areas in which they learn quickly and well. Ironically, however, Givon sees this as leading to overemphasis on (among other things) the language-specific and neglect of the universal (see p. 308), whereas I am here suggesting the opposite: errors have often led to speculation about universals, with corresponding inattention to correct and language-specific learning. 20

In sentences that do not refer to specific events but to generic or habitual events, both when and if (as well as whenever) are often acceptable, e.g. When / if / whenever it rains I take my raincoat. See Comrie (in press) for a crosslinguistic analysis of how the speaker’s evaluation

729

of the “degree of hypotheticality” of an antecedent event influences the choice of markers. 21

Both German and Dutch have other conjunctions that do differentiate the meanings, but these are not colloquial, everyday forms comparable to English if and when; they occur in somewhat more formal speech and (especially) writing, and are learned later by children. Hypotheticality may also be indicated by the use of subjunctive (German) or conditional (German, Dutch) verb forms; however, these are also learned relatively late. 22

“Lack of evidence” admittedly does not provide as strong a foundation as I would like for the claim that the form is not plurifunctional for these children. Let me therefore fortify my argument with a small but revealing set of observations. If als and wenn are plurifunctional forms (beyond a certain age) for Dutch and German children, respectively, and if they seek ways to differentiate the meanings, we would predict that in learning a second language like English that provides distinct forms, they should find it natural and easy to map the two forms onto their correct functions. However, the behavior of K, a Dutch child I have observed, suggests that this is not the case. K began to learn English at age 4 in a natural setting. By age 5 she could readily produce complex future predictives of low hypotheticality. However, she made—and at age 6;8 continues to make— many errors in her choice between if and when, most typically extending if to subjectively certain antecedents as well as 730

uncertain ones, e.g.: IF I am bigger, I’m gonna do that too, what you’re doing. (watching older sister make a candle); What are we going to do IF we are through? (as family eats a meal); / wanta wear a T-shirt IF I’m home. (K riding in car on way home; is planning what she will do when she gets there). Such errors were nonexistent in the records of three monolingual English speaking children, as noted in the text. We can conclude from K’s errors that she did not spontaneously make a categorical distinction between “certain” and “uncertain” future antecedent events, and this in turn indicates that when she used the Dutch form als ‘if/when’ in complex future predictive sentences in Dutch, it was not plurifunctional for her. 23

I am grateful to Eve Clark for making data from one of these children available to me. 24

Explaining crosslinguistic differences in the partitioning of many, perhaps most, semantic domains requires reference to differences both in the number of cuts made and in the criteria used to make them. For example, there can be partially shared semantic hierarchies in which languages that agree on the first one or more basic cuts diverge at a certain level in the hierarchy, not because one goes on to make finer distinctions than the other but because both languages subdivide a given meaning category further on the basis of different criteria, or the same criteria applied in a different order (see Denny, 1978, 1979, for relevant discussion and examples). Alternatively, languages may differ in their first cuts, but languages that have adopted the same starting point may subdivide further according 731

to a shared hierarchy. Although the discussion here is couched in terms of first cuts that differ, I intend it to apply more generally to what children could do at any point in the division or subdivision of a semantic domain where languages partition meaning into partially overlapping categories. 25

Information about which constituents control verb agreement in children’s early sentences and about the distribution of case markers and contrastive pronoun forms (e.g. he versus him) is scattered and frequently inadequate to answer our question; e.g. authors often do not distinguish between transitive and intransitive subjects in discussing the emergence of these grammatical markers. Case marking errors are common at least in the speech of children learning nominative-accusative languages, e.g. see Clancy (1985) on Japanese and Mills (1985) on German. However, these errors seem to involve transitive subjects and objects at least as often as intransitive subjects. It is not until children are reliably distinguishing between these two roles that it makes sense to ask whether they associate intransitive subjects with one, the other, or neither. In general, the absence of reports that children learning nominative languages at first mark only transitive subjects or only intransitive subjects with the nominative (or other subject markers) is compatible with the developmental picture obtained by looking at word order patterns. 26

It may reasonably be argued that shared word order alone is an insufficient criterion for common category membership, for the following reason: since there are 732

few possible sentence positions in two-and three-word utterances, words that from the child’s point of view perform distinct semantic functions might nevertheless happen to share the same (plurifunctional) position (see Bowerman, 1975, p. 87ff). Fortunately, Braine (1976, pp. 59–61) gives careful attention to the problem of how we can determine whether two hypothetical order-based patterns—for example A (transitive actor-action, in this case) and B (intransitive actor-action)–are really two independent patterns or one. Briefly, Braine suggests examining how order of emergence varies among children: “If A appears before B in some children but B before A in others (or if A appears without B in one corpus and B without A in another), then we can infer that A and B are separate patterns that can be independently acquired.… If A and B appear simultaneously in all children (or if any corpus that has one always has the other), then the hypothesis that they are really one pattern is confirmed” (p. 59). By these criteria we must conclude that actor-action was a unified pattern for the children that Braine studied: children who produced patterned actor-action strings with intransitive actors also produced them with transitive actors, and vice versa. 27

The solutions discussed by Lucy can presumably also be elaborated by further subdivision on the basis of gender, shape (as in classifier languages), etc. 28

Working out the details of the system being acquired may take many years, as studies of the mass-count distinction in English have shown (Gathercole, 1983; Gordon, 1982). Nevertheless one of the basic parameters 733

that distinguishes the English system from other possible systems seems to be in place essentially from the beginning: that nouns naming countable objects can be treated alike grammatically, regardless of whether these objects are animate or inanimate.

734

Subject Index

The abbreviations given below indicale discussion of an index category with regard to a particular language. Entries without a language code refer to a general discussion of the category in question. Note that the grammatical sketches of individual languages are not included in the index. All references refer to discussions of acquisition or of general linguistic issues in the chapters.

ABBREVIATIONS FOR LANGUAGE CODES Ab Arabic Ar Armenian ASL American Sign Language Ba Bantu

735

Ch Cz Du Ek En Es Fi Fr Ge Ha He Hu H IE It Jp Ka La Mo No PI Pr Qu Ro Rs Sa SC Se Si SI Sp

Chinese Czech Dutch Eskimo English Estonian Finnish French German Hawaiian Hebrew Hungarian Hua Indo-European languages Italian Japanese Kaluli Latvian Mohawk Nogogo Polish Portuguese Quiche Mayan Romance languages Russian Samoan Serbo-Croatian Semitic SiSwati Slavic languages Spanish

736

Ta Tamil Tu Turkish Yu Yucatec Maya

A accusative, He 1123, 1147, Pl, Rs 1241 adjective, En 967f, 977, 981, 984, Ge 1196, 1224, 1226, Jp 1224 adverb, Pl 1241 affixation, formal properties, En, Ge, Hu 1226f agent, 1296f, Sl 1177 agglutinating morphology, 1058, En 1232, Fr 1057, Hu 1073, 1232, Tu 1232 agreement, En, Ge, Jp 1297, Ba, He, Sp 1058, Hu 1081f, 1138f, It 1182 allomorphy, 1208–1219, He, Si 1115f, Hu 1115f, 1180 ambiguity, 1014, 1017 amalgam, 1033, 1035, 1165, ASL 1110, Hu 1084, 1096, 1110, 1117, 1128 analogy, Hu 1097f, 1104, 1118, 1130

737

analytic construction, 1239, 1272, 1276–1278, 1307–1309, En 1226, Fr, Ge, He, Hu 1226, 1229–1231, It 1229–1231, PI, Pr 1226, Ro, Sp 1229–1231, Tu 1226 animacy, 1187, En 1189f, 1299–1301, Ge 1225, Yu 1299–1301 and casemarking, Pl, Rs 1186 article, Hu 1087 and gender, Fr, Ge, Sp 1217f aspect, En 1183, 1301–1303, Cz 1210, Fi 1188, Jp 1301–1303, PI 1184, 1210, Rs 1210, 1241, SC 1210, SI 1183, 1210, 1301–1303, Tu 1183, 1301–1303 habitual, En 966 imperfective, Pl 1302f perfective, En 965, 967, Ka, Pr 1181 and tense, 1181 auxiliary, En 966, 1204 order, En 1232 B basic child grammar, 1260, 1283f, 1301, 1304f

738

Benue-Kwa, 1176 C canonical form, 1220f, En, Fr, Jp, SC 1234f casemarking, 1219, Hu 1073, Ge, Jp, Ka 1238, Sa 1297f, Sl 1238, 1297f, Tu 1221 causative, En 966, 1143, 1270, 1292, Hu 1143 citation form, Es, Ge, Fr, He, Hu, Jp, Ma, Pl, Ma 1111 clitic, 1056, Fr 1126, 1236, Pl 1236 cognition, 1054, 1140, 1164, 1170, 1173–1190, 1198f cognitive bases of acquisition, 961–1004 structure, 986f, 989–992 comparative, 1306, En 980, Hu 1148f competition among related forms, 1308f, En, Fr, Pl 1273–1278 complement, Fr, 1229, Pl, Tu 1220f comprehension, En 964f concepts, of dimension, 981–983

739

and language, En 963–985, 991–993 of quantity, En 974–977 of space, En 969–972 conditional, En, Du, Ge, Fr, He, Pl, 1126, 1242f, 1271–1274, 1296f, 1298–1300, Hu 1205, 1242f conjunction, 1219f, En 966, Hu 1080, 1146, Tu 1220 copula, En 1148 count/mass distinction, En 1298–1301 creole, 1207 D dative, Hu 1123, 1147, 1180 definiteness, 1019, 1219, En, Ge, He, Ro, Tu 1188, Hu 1088, 1147, 1188 deixis, En 971 derivation, Hu 1073–1075, 1129-1144 diachrony, 1014–1016, 1019f, 1176, 1189, 1207 direct object, 1293–1299, En, He, Hu, Ka, PI, Tu 1174f, Rs 1176f, Ge 1223, He, Jp 1224f and definiteness, Jp, IE, Se, Tu 1162f 740

discontinuity, 1046f discourse, 1022 E ellipsis, 1012f, 1229, Hu 1087, It, Jp, PI, Tu 1206 ergative, En, Ek, H, Hu 1294–1298, Ka 1176f, 1224, Qu, Sa 1177 ethnology, Hu 1083–1089 experimentation, Hu 1086 F fillers, 1036, Tu 1058 first words, 1030, 1033, 1052f, En 984, Hu 1084 frequency, 1034, 1036, 1202, 1235, En, Ge, He, Ro, Sa, 1166f, Fr 1048 functor, 1171–1173, 1176ff, 1191f, 1219f, Hu 1210 fusional morphology, 1050, Fr 1057 gender, Fr 1199f, 1217f, Ge 1150, 1194, 1218, He 1194, 1199f, 1217, Pl 1194, 1216, Ro 1217, Rs 1194, 1616, Sp 1194, 1199f, 1217 genitive, Jp 1196, Pl 1196, 1222

741

H hypothetical, En, Du, Ge 1298f I imitation, 1053 imperative, Fr 1241 individual differences, 1032, 1053f, 1286 inflectional morphology, 1020f, 1047, 1058, En, 1205, Ge 1218, Hu 1205, La 1204, Pl 1205, 1216, Rs 1204, 1216 innateness, 1031, 1045, 1119, 1158ff, 1272, 1275ff, 1271, 1281–1285 input, 1025–1037, En, Ka, Sa 1051–1053, 1060–1062, and onset of communicative intentions, 1262 instrumental, Hu 1123, 1147 intonation, 1022 L lexicon, 994, En 984

742

locative, En 964f, 966–977, 1173, 1180, 1292, Ge 1178–1182, 1310f, Hu 1124, 1180, It 1180, Pl 1178–1182, SC 1180, Sl 1178–1182, Tu 1178–1182 and possessive, En 1270 M methodology, 983–985, 993, 1050–1054, 1261–1263, 1283f crosslinguistic, 962, 994–996, 1050f, 1258ff, 1311, 1313f error analysis, 1006f, 1056, 1059, 1162, 1263–1266, 1288f, 1292, 1307–1312, Du 1269f, En 1285f, Tu 1271–1279 ethnography, 1051–1053, 1060–1062 experimentation, 1055, Ba, En, Fi, He, Hu, Si, Su 1038ff morpheme, order, Ge 1127, Hu 1074, 1103, 1117–1129, Jp 1103, Tu 1127 morphophonology, 1057, Ar 1195, En 1195, 1209, 1212f, Fi 1061f, Fr 1211, Ge 1211, He 1209, 1211, Hu 1075–1077, 1095–1116, 1210, 1012f, Pl 1211, Pr 1061f, Rs 1211, Si 1061f, Sl 1211, Sp 1211, Tu 1061f, 1209, 1215 motion, He, It, Sp 1230

743

N negation, 1021, 1239–1241, 1277, En 1205, 1239f, Fr 1240, Hu 1230, 1240, Jp 1224, 1230, 1240, Ka 1177, PI, Tu, Rs 1240 neologism, En 965, Hu 1129–1144, Pl, Rs 1210 nominalization, Tu 1220 nominative, Ch, En, Es, Ge, He, Hu, PI 1216 number, 1219, En 973, Fr 1238, Ge 1226, He 1199f, SC, Tu 1238 numeral, He 1236 O operating principle, 986, 1029–1067, 1092f, 1157–1256 relationships among, 1279–1282 testing proposals about, 1213f, Jp 1266–1279 operator, 1170f placement, 1239, 1277, Pl 1272f order, morpheme, 1235–1238, En, Hu, Tu 1231f order, word (see pragmatics), 1164f, 1232–1235, En 964, 988, Fr 1241, Ge 1236, Hu 1070

744

overgeneralization, 1017, 1020, 1222–1227, 1280f, Ab 1104–1111, En 1105ff, 1107, 1202, Ge 1104–1111, 1202, He 1106, Hu 1104–1111, 1202, Jp 1202, La 1107, Pl 1202, Pr 1107, 1202, Rs, Sp 1107, 1202, Tu 1202 P paradigm, morphological, 1213–1219 partitive, En 1186f, Fi 1188, Fr 1229, Rs 1186f passive, 1012f, 1021, En 987–989, 1015, 1182, He 1182, 1224, Tu 1182 perfective, Hu 1125 person, En 1223, 1225, Pr 1225 and verb, Pl 1205 phonology, 1034ff, 1037–1042, 1209, Hu 1070–1072 pidgin, 1006, 1008, 1207 pivot structure, Hu 1085 plurifunctionality, 1227–1229, 1268–1270, 1307f, Du, En, Ge 1290 polysemy, Hu 1142

745

possession, En 1185, Fr 1179, 1229f, Ge 1179, 1125, 1230 possessive, En 1292, Ge 1310f postposition, Hu 1236 pragmatics, 1010–1016, 1018f, 1229, 1233f, En 988, 1184f, Ge 1184f, Hu 1120–1122, It 1206, Jp 1184f, 1206, Ka 1184f, Pl 1206, Tu 1184f, 1206 prefix, 1045, 1058 processing, 986, 1018f, 1047, 1054, 1094, 1150, 1259f pronoun, Fr 1217, It, Jp, Pl, Tu 1206 pronoun number, Ek, En, Ha, Ta, No, 1287 prosody, 1035f, 1038, 1045, 1055ff’, 1060f, 1112, 1123, En 1204, 1241, Fr 1204, Ge, Jp 1241, La, Rs 1204 prototype, En 984, He, Hu, Ka, Pl, Tu 1174–1182 Q quantifier, En 964, 973f question, En, Fi, Ge, Hu, Jp 1241f semantics, En 966 R

746

redundancy, En 1126, Fr, 1141, Ge 1126, Hu 1102f, 1126, 1141, Pl 1126, Tu 1103, 1141 relative clause, 1023, En 1220f, 1278f, Fr 1229, He 1220f, 1278f, Hu 1080, Pl, Tu 1220f relevance, 1272, Pl 1274ff rote learning, 1096, 1214, 1218, Jp, Tu 1201f S salience, cognitive, 1173–1190, 1218f, 1294, 1296 perceptual, 1034–1036, 1038ff, 1045, 1047f, 1148, 1164–1166, Ch 1172, He, Hu 1112, Jp 1112 1201f, Mo 1112, Se 1172, Tu 1201 semantic categorization, in children’s initial grammars, 1282–1314 and cognition, 1284–1286, 1304–1306, 1309, 1313, En 1287, 1293f, 1300f, Ha, No 1287, Ta, WGI 1287, 1293f, Yu 1300f spatial word, En 1291f, 1310, He 1291f subject, 1019, 1293–1299

747

subordination, 1298–1300

1011,

1021,

1023,

En,

Du,

Ge

suffix, 1045, 1058 synonomy, 1014, 1017 syntax, 1010–1015, 1021–1025, 1233, En 1241f synthetic construction, 1275f, 1308f T tense, 1219, Pr 1225 future, En 1185 morphology, Fr, Ge, He, Pl, Rs, Sl, Sp 1211, Jp, 1240 past, En 965, 967, 1205, 1226, 1302–1304, Ar, Ch, Fi Hi 1181, It 1182, Ka 1176, 1181, Pl 1302–1304, Tu 1215 timeword, En 966, 1291f, 1310, He 1291f topic, 1011–1013, 1019, 1234, Hu 1122, 1127 transitivity, 1293–1299, 1175ff, Qu, Sa 1177 two word utterance, 1170f typology, language, 1005–1014, 1050f, 1055, 1194, 1309

748

Ute, 1025 U universals, 1005–1027, 1309 V valence, En, Fr, He, Hu, Pl, Pr, Tu 1225f voice, He 1183 vowel harmony, 1057, Hu 1085, 1095–1116, 1212f, Tu 1212f W word order, Hu 1077–1081, 1085f, 1088, 1117–1129 pragmatics, En, Fr, Ge, Hu, It, Jp, Ka, Pl, Rs, SC, Tu 1234 Z zero morpheme, En 1015, 1222f. Pl, Ge, He, Hu 1222f, Sp 1015

749

Author Index

Numbers in italics indicate pages with complete bibliographic information. A Abrahamson, A., 967, 1004 Adams, M., 988, 990, 1003 Aksu, A. A., 1303, 1315 Aksu-Koç, A. A., 1036, 1061, 1062, 1162, 1215, 1245, 1271, 1315 Ames, L., 967, 997

750

Ammon, M. S., 1165, 1245 Ammon, P., 981, 1000 Andersen, E. S., 1285, 1315 Andersen, H., 1294, 1315 Anderson, J., 1089, 1094, 1152 Antinucci, F., 967, 997, 1180, 1182, 1195, 1245, 1248, 1302, 1315 Argoff, H., 1062, 1062 B Badry, F., 1195, 1245 Baker, C. L., 1233, 1245 Baldie, B., 987, 997 Baldwin, P., 964, 1001 Balfour, G., 964, 973, 1000 Bánhidi, Z., 1070, 1152 Barrie-Blackley, S., 964, 997 Bartlett, E., 965, 967, 978, 979, 980, 981, 984, 997, 998 Bateman, W., 967, 998 751

Bates, E., 991, 993, 998, 1078, 1086, 1089, 1108, 1120, 1150, 1152, 1154 Bebout, L., 964, 998 Behagel, O., 1237, 1245 Beilin, H., 973, 974, 987, 988, 989, 990, 991, 998 Bellugi, U., 987, 988, 1000, 1098, 1112, 1152, 1264, 1271, 1315, 1316, 1204, 1205, 1239, 1241, 1242, 1245, 1247 Benigni, L., 991, 993, 998 Bennett, D. C., 1291, 1315 Berko, J., 1131, 1152, 1202, 1245 Berman, R. A., 1088, 1105, 1107, 1108, 1111, 1112, 1113, 1114, 1152, 1209, 1245, 1264, 1271, 1279, 1315 Berndt, R., 967, 980, 998 Bever, T. J., 964, 974, 987, 988, 1002, 1117, 1152, 1165, 1233, 1234, 1249 Bickerton, D., 1162, 1245 Block, E., 968, 998 Bloom, L., 966, 967, 973, 998, 1001, 1036, 1062, 1302, 1315

752

Bock, K., 1087, 1152 1327 Bogoyavlenskiy, D. N., 1202, 1245 Bohn, W., 967, 998 Bonvillian, J., 968, 1002 Bonvillian, J., 996, 1002 Bowerman, M., 966, 993, 994, 998, 1143, 1152, 1172, 1174, 1189, 1225, 1242, 1245, 1259, 1262, 1264, 1265, 1270, 1271, 1273, 1275, 1278, 1281, 1282, 1284, 1285, 1286, 1290, 1292, 1297, 1310, 1314, 1315, 1316 Boyd, W., 967, 998 Boyes-Braem,P., 984, 1003, 1288, 1318 Braine, M. D. S., 996, 998, 1042, 1049, 1062, 1085, 1120, 1122, 1152, 1170, 1178, 1245, 1297, 1316 Brainerd, C.,986, 994, 998 Branston, M., 993, 1002 Bresnan, J., 1233, 1245 Bretherton, I., 991, 993, 998 Bruner, J. S., 1036, 1062 Brewer, W., 967, 978, 979, 999, 1107, 1112, 1152

753

Brown, A., 964, 999, 1000 Brown, R., 966, 973, 987, 988, 999, 1000, 1036, 1062, 1086, 1110, 1098, 1112, 1152, 1172, 1187, 1201, 1245, 1259, 1263, 1264, 1284, 1288, 1300, 1316 Brush, L., 973, 999 Buczowska, E., 1302, 1303, 1319 Budwig, N., 1185, 1246 Bullock, M., 964, 973, 975, 976, 979, 999 Burke, D., 986, 994, 1001 Bybee, J. L., 1098, 1102, 1107, 1112, 1152, 1185, 1195, 1210, 1211, 1225, 1226, 1237, 1246, 1212, 1316 C Camaioni, L., 991, 993, 998 Campbell, R., 964, 968, 1001, 1004 Caramazza, A., 967, 980, 998 Carey, S., 964, 965, 967, 971, 973, 978, 979, 999, 1002, 1176, 1246 Carpenter, P. P., 1089, 1155 Carskoddon, G., 968, 1002

754

Cazden, C., 1098, 1112, 1145, 1152, 1264, 1316 Chafe, W., 1011, 1026 Chalkley, M., 1150, 1154, 1167, 1190, 1247 Chapman, R., 964, 993, 999, 1002 Chomsky, N., 1007, 1026, 1258, 1259, 1281, 1316 Clancy, P., 1097, 1103, 1112, 1113, 1120, 1121, 1123, 1141, 1142, 1152, 1263, 1267, 1289, 1290, 1297, 1316 Clark, E. V., 964, 965, 967, 968, 970, 978, 999, 1105, 1106, 1107, 1112, 1118, 1120, 1121, 1125, 1126, 1128, 1132, 1134, 1137, 1141, 1142, 1144, 1146, 1152, 1196, 1262, 1264, 1271, 1280, 1285, 1316 Clark, H., 967, 978, 979, 981, 999, 1132, 1134, 1152 Clark, R., 1035, 1046, 1048, 1062 Cohen, W., 975, 1003 Coker, P., 964, 999 Comrie, B., 1289, 1316 Cook, N., 964, 968, 982, 999 Coots, J., 965, 967, 979, 980, 999 Corrigan, R., 996, 999

755

Court, S., 967, 999 Culicover, P., 1233, 1249 Craft, K., 970, 1000 Cromer, R., 964, 966, 968, 999, 1000, 1145, 1152 Cruse, P., 984, 1000 D Dansky, J., 993, 1001 deLemos, C., 1175, 1181, 1246 DeMarcellus, O., 987, 988, 989, 991, 1003 Denny, J. P., 1285, 1287, 1293, 1316 Denny, N., 982, 1000 Deutsch, W., 1185, 1246 deVilliers, J. G., 1277, 1316 deVilliers, P. A., 1277, 1316 Dewart, H., 989, 1000 Dezső, L., 1120, 1153 Dixon, R., 1294, 1316

756

Donaldson, M, 964, 973, 974, 978, 980, 1000 Dore, J., 962, 994 Duranti, A., 1019, 1026, 1195, 1245 E Ehri, L., 980, 981, 1000 Eilers, R., 965, 978, 979, 1000 Eisenberg, A. R., 1263, 1316 Ekmekçi, Ö. F., 1036, 1062 Ellington, J., 965, 978, 979, 1000 Emerson, H., 964, 1000 Erbaugh, M., 1181, 1246 Erreich, A., 962, 1000 Ervin, S. M., 1096, 1153, 1264, 1317 Ervin-Tripp, S., 968, 1000 Estes, K., 965, 975, 1000 F Fava, E., 1120, 1153

757

Feurer, H., 1112, 1153 Ferguson, C. A., 1008, 1009, 1026, 1030, 1051, 1064 Fillmore, C. J., 1175, 1178, 1246 Firbas, J., 1086, 1153 Fishbein, H., 970, 1000 Fisher, K., 982, 1003 Flavell, E., 970, 1002 Flavell, J., 970, 986, 994, 1000, 1002 Fónagy, I., 1123, 1153 Fraser, C., 987, 988, 1000 French, L., 964, 1000 G Gaer, E., 987, 991, 1000 Gale, H., 967, 1000 Gale M., 967, 1000 Gallistel, C., 975, 1000 Garcia, E., 1017, 1026

758

Garman, M, 965, 973, 974, 1004 Gathercole, V., 1300, 1301, 1317 Gebert, L., 1195, 1245 Gee, J., 1175, 1185, 1246, 1248 Gelman, R., 964, 970, 973, 975, 976, 979, 995, 999, 1000, 1003 Gentner, D., 994, 996, 1000 Geodakjan, I. M., 1181, 1246 Gerhardt (Gee), J. B., 1175, 1246 Gitterman, D., 979, 980, 981, 1001 Givón, T., 1008, 1011, 1012, 1012, 1019, 1022, 1024, 1026, 1237, 1246, 1263, 1264, 1288, 1317 Gleitman, H., 968, 1002 Gleitman, L., 968, 1002 Goodluck, H., 1259, 1317 Goodson, B., 987, 1001 Gopnik, A., 1171, 1246 Gordon, P., 1300, 1301, 1317

759

Grant, J., 967, 1001 Graves, M., 1104, 1153 Gray, W., 984, 1003, 1288, 1318 Greenberg, J., 1306, 1317 Greenfield, P., 981, 987, 1001 Grieve, R., 964, 1001 Griffith, J., 973, 974, 1001 Griffiths, P., 965, 973, 974, 1004 Guillaume, P., 1106, 1153 Gvozdev, A. N., 1176, 1186, 1210, 1212, 1241, 1246 H Hafitz, J., 967, 998, 1302, 1315 Hagen, J. W., 1039, 1063 Haiman, J., 1294, 1317 Hakuta, K., 1260, 1317 Harris, A., 989, 990, 1002 Harris, L., 967, 1001

760

Harwood, F., 988, 1001 Hatch, E. M., 1054, 1063 Hawkins, J., 1306, 1309, 1317 Hayes, L., 982, 1004 Hayhurst, H., 988, 991, 1001 Heath, S. B., 1030, 1051, 1052, 1053, 1063 Heber, M., 981, 1001 Herring, S., 1183, 1184, 1246 Herskovits, A., 1180, 1247 Hoefnagel–Hohle, M., 1107, 1155 Holmes, T., 966, 1001 Holland, V., 964, 973, 976, 1001 Hood, L., 966, 1001 Hoogenraad, R., 964, 993, 1001 Hopper, P. J., 1175, 1247 Horgan, D., 988, 1001, 1182, 1247 Hunt, M., 982, 1004

761

Huttenlocher, J., 986, 994, 1001 Hyman, L. M., 1045, 1063, 1079, 1153 I Ingram, D., 966, 1001, 1004, 1099, 1153, 1209, 1247, 1287, 1306, 1317 Inhelder, B., 969, 1001, 1002, 1003 Ivimey, G. P., 1204, 1247 J Jacobson, T., 1263, 1289, 1290, 1316 Jamison, W., 993, 1001 Johnson, D., 984, 1003, 1288, 1318 Johnson, H., 964, 1001 Johnson, M., 1175, 1247 Johnston, J., 967, 968 , 970 , 971 , 984 , 1001, 1180, 1247, 1260, 1262, 1263, 1317 Jókay, Z., 1070, 1152 Just, M., 1089, 1155 K

762

Kaplan, A., 1280, 1317 Karmiloff-Smith, A., 1139, 1141, 1153, 1203, 1217, 1227, 1228, 1230, 1247, 1262, 1264, 1268, 1277, 1278, 1281, 1308, 1317 Karttunen, L., 1024, 1026 Keiffer, K., 970, 1000 Demler, D., 982, 1004 Kenyeres, E., 1096, 1145, 1153 Kessel, F., 968, 998 Kinsbourne, M., 987, 988, 1002 Kintsch, W., 1039, 1063 Kliegl, R., 1089, 1154 Klima, E. S., 1239, 1247 Kohn, L., 964, 999 Konieczna, E., 1302, 1303, 1319 Kooij, J. G., 1137, 1153 Köpcke, K., 1098, 1150, 1153 Koziol, S., 1104, 1153

763

Kuczaj, S. A., 967, 1001, 1105, 1153, 1205, 1226, 1232, 1247, 1273, 1317 Kunene, E. C. L., 1035, 1040, 1061, 1063, 1116, 1153 L Lakoff, G., 1175, 1247 Laurendeau, M., 969, 1001 Laver, J. D. M., 1273, 1317 Layton, T., 980, 1002 Learned, J., 967, 997 Lederberg, A., 1106, 1153 Lee, L., 966, 1002 Lempers, J., 970, 1002 Lempert, H., 987, 988, 1002 Leont’yev, A. A., 1204, 1247 Levine, S., 971, 1002 Levy, Y., 1218, 1247 Lewis, S., 970, 1000 Li, C., 1019, 1026, 1027, 1112, 1153 764

Lifter, K., 967, 998, 1302, 1315 Lightbrown, P. M., 1120, 1153 Long, M., 1008, 1027 Lord, C., 1143, 1153 Lucy, J. A., 1298, 1299, 1300, 1317 Lyons, J., 1291, 1317 M Macnamara, J., 1300, 1317 MacWhinney, B., 1047, 1048, 1063, 1070, 1078, 1082, 1083, 1084, 1085, 1086, 1087, 1092, 1094, 1096, 1097, 1099, 1100, 1101, 1104, 1105, 1106, 1107, 1108, 1109, 1110, 1116, 1117, 1118, 1119, 1120, 1121, 1122, 1125, 1126, 1128, 1130, 1131, 1133, 1134, 1139, 1140, 1141, 1142, 1143, 1144, 1145, 1149, 1150, 1151, 1152, 1153, 1154, 1174, 1222, 1226, 1227, 1232, 1242, 1247, 1264, 1275, 1280, 1317, 1318

1076, 1089, 1102, 1112, 1123, 1136, 1147, 1205, 1268,

1077, 1091, 1103, 1114, 1124, 1138, 1148, 1210, 1273,

Maratsos, M., 967, 968, 1001, 1002 , 1106 , 1139, 1150, 1153, 1154, 1167, 1190, 1191, 1199, 1232, 1241, 1300, 1318 Masongkay , Z.., 970, 1002

765

McCarthy , J. J., 1233, 1245 McCluskey , K. , 970 , 1002 McIntyre , C., 970, 1002 McNeill , D., 1036, 1062, 1259, 1318 Meggyes, K., 1103, 1120, 1121, 1127, 1147, 1154 Mehler , J.., 974, 1002 Meier, R. P., 1111, 1154 Menyuk, P., 966, 1002, 1117, 1154, 1239, 1247 Mervis, C., 984, 1003, 1288, 1318 Miller, J., 993, 1002 Miller, R., 967, 997, 1182, 1245, 1302, 1315 Mills, A. E., 1297, 1310, 1318 Moore, T., 989, 990, 1002 Murphy, M., 964, 999 Murray, D., 993, 1001 N Naremore, R., 966, 967, 1002, 1004

766

Naro, A., 1207, 1248 Nelson, K., 964, 968, 981, 982, 986, 987, 993, 996, 1001, 1002, 1004 Newport, E., 968, 1002, 1045, 1063, 1111, 1154 Nickerson, N., 989, 1002 Norman, D., 986, 1002, 1274, 1318 Nussbaum, N., 966, 1002 O Ochs, E., 1019, 1026, 1030, 1051, 1052, 1063, 1265, 1298, 1318 Ohala, J., 1097, 1154 Oiler, D. K., 965, 978, 979, 1000 Olmar, M., 1195, 1248 Olson, D., 989, 1002 Olson, G. M., 986, 994, 1002 Omar, M., 1104, 1154 Osherson, D., 982, 1002 Otsu, Y., 1259, 1318

767

P Palermo, D., 964, 970, 973, 976, 1001, 1002, 1004 Pardo, E., 1102, 1152 Parisi, D., 1180, 1248 Park. T.-Z., 1120, 1154 Pelsma, R., 967, 1002 Peters, A. M., 1031, 1032, 1033, 1035, 1036, 1039, 1046, 1047, 1050, 1054, 1055, 1059, 1063 Phillips, J. R., 1030, 1063 Piaget, J., 969, 1002, 1003 Pinard, A., 969, 1001 Pinker, S., 1176, 1233, 1248 Pléh, Cs., 1108, 1150, 1154 Pye, C, 1039, 1040, 1042, 1063, 1107, 1148, 1154, 1177, 1201, 1248 R Radulović, L., 1234, 1235, 1238, 1248 Réger, Z., 1226, 1248

768

Reichle, J., 993, 1002 Ricciuti, H., 974, 982, 1003 Rice, M., 984, 993, 1003 Richards, M., 967, 1003, 1117, 1154 Roberts, R., 982, 1003 Robinett, F. M., 1061, 1064 Roeper, T., 1259, 1318 Rommetveit, R., 987, 993, 988, 991, 1003, 1004 Rosch, E., 984, 1003, 1288, 1318 Ryan, M, 965, 973, 1004 S Saltzman, E., 981, 987, 1001 Sapir, E., 1171, 1248 Savasir, I., 1175, 1185, 1246, 1248 Saxe, G., 974, 975, 993, 1003 Schieffelin, B. B., 1030, 1051, 1052, 1063, 1176, 1178, 1184, 1248

769

Schlesinger, I. M., 994, 996, 1003, 1173, 1248, 1314, 1318 Scholnick, E., 970, 1003 Segalowitz, S., 964, 998 Shantz, C., 970, 973, 974, 1001, 1003 Shatz, M., 970, 986, 1003 Shipstead, S., 970, 1000 Siegel, I., 973, 974 Siegel, L., 964, 975, 976, 977, 979, 1003 Silva, M., 1263, 1289, 1290, 1316 Simōes, M. C. P., 1225, 1248 Sims-Knight, J., 970, 1002 Sinclair, A., 987, 988, 989, 991, 1003 Sinclair, H., 987, 988, 989, 991, 996, 1003 Sinha, C., 970, 1004 Slobin, D. I., 962, 967, 968, 982, 984, 986, 987, 991, 1001, 1003, 1004, 1033, 1036, 1045, 1047, 1061, 1062, 1063, 1064, 1070, 1082, 1089, 1102, 1109, 1112, 1113, 1127, 1152, 1154, 1159, 1162, 1165, 1168, 1173, 1174,

770

1180, 1189, 1194, 1201, 1205, 1207, 1222, 1226, 1230, 1233, 1234, 1243, 1248, 1249, 1257, 1259, 1260, 1261, 1265, 1266, 1267, 1268, 1269, 1271, 1281, 1282, 1283, 1285, 1286, 1288, 1310, 1311, 1314, 1315, 1317, 1318

1212, 1245, 1262, 1272, 1296,

1275, 1246, 1263, 1278, 1301,

1216, 1247, 1264, 1280, 1305,

Smith, M., 966, 1004 Smith, N. S. H., 1107, 1155 Smoczyńska, M., 1105, 1111, 1114, 1115, 1117, 1123, 1126, 1155, 1265, 1268, 1271, 1272, 1279, 1302, 1318 Snow, C. E., 1030, 1051, 1064, 1107, 1155, 1263, 1318 Snyder, A. D., 1239, 1249 Sokolov, J., 1094, 1116, 1154 Stampe, D., 1100, 1155, 1210, 1249 Stanovich, K. E., 1039, 1063 Stassen, L., 1306, 1309, 1318 Stemberger, J., 1097, 1103, 1133, 1143, 1155, 1273, 1318 Stephany, U., 1302, 1318 Stem, C., 1036, 1064, 1084, 1155

771

Stern, G., 1130, 1155 Stern, W., 1036, 1064, 1084, 1155 Stick, S., 980, 1002 Stoel–Gammon, C., 1225, 1248 Stone, J., 967, 978, 979, 999 Strohner, H., 964, 1004 Strommen, E., 967, 1001 Sugarman, S., 970, 974, 982, 996, 1004 Szabó, D , 1070 1152 T Talmy, L., 1173, 1178, 1249 Tanz, C, 968, 1004 Tavakolian, S. L., 1259, 1319 Thibadeau, R., 1089, 1155 Thomas E. K., 1046, 1064 Thorns, J., 976, 1004 Thompson, S., 1019, 1026, 1027, 1112, 1153, 1175, 1247 772

Tirondola, G., 1120, 1153 Toivainen, J., 1181, 1188, 1249 Tolbert, M. K., 1036, 1064 Tompa, J., 1070, 1080, 1100, 1155 Townsend, D., 978, 979, 980, 1004 Traugott, E. C, 1291, 1319 Tremaine, R., 989, 1004 Turner, E., 987, 988, 991, 1004 Tyack, D., 966, 1004 V Valian, V., 962, 1000 Varma, T. L., 1181, 1249 Vaughn, B., 970, 1002 Velten, H. V., 1039, 1064 Vietz, P., 982, 1004 Vihman, M. M., 1054, 1064 Viktor, G., 1112, 1155

773

Voegelin, C. F., 1061, 1064 Volterra, V., 991, 993, 998 W Wales, R., 964, 965, 968, 973, 974, 978, 980, 1000, 1004 Walkerdine, V., 970, 1004 Walter, S., 1104, 1155 Wannemacher, J., 965, 973, 1004 Washington, D., 967, 1004 Watkins, L., 1285, 1319 Watson, J., 970, 982, 1003, 1004 Webb, P., 967, 1004 Weeks, T. E., 1040, 1064 Weiner, S., 964, 973, 974, 1004 Weir, R. H., 1040, 1060, 1064 Weist, R. M., 1302, 1303, 1319 Wellman, H., 986, 994, 1000 Wexler, K., 1233, 1249

774

White, G., 964, 998 Wilcox, S., 970, 1004 Windmiller, M., 970, 1004 Witkowska-Stadnik, K., 1302, 1303, 1319 Wittgenstein, L., 1008, 1027 Winzemer, J., 962, 1000 Wong Fillmore, L., 1042, 1043, 1046, 1054, 1064, 1165, 1249 Woodward, M., 982, 1004 Wysocka, H., 1302, 1303, 1319 Z Zager, D , 1107, 1155 Zakharova, A. V., 1104, 1155 Zubin, D., 1098, 1150, 1153

775