Determinants of Grammatical Variation in English 9783110900019, 9783110176476

Table of contents :
Introduction
Frequency as a determinant in grammatical variation and change
Phonological determinants of grammatical variation in English: Chomsky’s worst possible case
Post-verbal constituent ordering in English
Grammatical variation in English: A question of ‘structure vs. function’?
Why are zero-marked phrases close to their heads?
Cognitive complexity and horror aequi as factors determining the use of interrogative clause linkers in English
Support for more-support
The role of extractions and horror aequi in the evolution of -ing- complements in Modern English
Gerundial complements after begin and start: Grammatical and sociolinguistic factors, and how they work against each other
Is there semantics in all syntax? The case of accusative and infinitive constructions vs. that-clauses
Aspects of iconicity and economy in the choice between the s-genitive and the of-genitive in English
Constructional semantics as a limit to grammatical alternation: The two genitives of English
Principles of grammaticalization and linguistic reality
Varieties of English from a cross-linguistic perspective: Intensifiers and reflexives
Non-standard English and typological principles: The case of negation
‘Every place has a different toll’: Determinants of grammatical variation in cross-variety perspective
Author index
Subject index

Citation preview

Determinants of Grammatical Variation in English

W DE G

Topics in English Linguistics 43

Editors

Bernd Kortmann Elizabeth Closs Traugott

Mouton de Gruyter Berlin · New York

Determinants of Grammatical Variation in English Edited by

Günter Rohdenburg Britta Mondorf

Mouton de Gruyter Berlin · New York 2003

Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter GmbH & Co. KG, Berlin.

© Printed on acid-free paper which falls within the guidelines of the ANSI to ensure permanence and durability.

ISBN 3-11-017647-5 Bibliographic

information published by Die Deutsche

Bibliothek

Die Deutsche Bibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data is available in the Internet at .

© Copyright 2003 by Walter de Gruyter GmbH & Co. KG, 10785 Berlin All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Cover design: Christopher Schneider, Berlin. Printed in Germany.

Contents Introduction

1

Frequency as a determinant in grammatical variation and change Manfred Krug

7

Phonological determinants of grammatical variation in English: Chomsky's worst possible case Julia Schlüter Post-verbal constituent ordering in English Thomas Wasow and Jennifer Arnold Grammatical variation in English: A question of'structure vs. function'? Stefan Th. Gries Why are zero-marked phrases close to their heads? John A. Hawkins Cognitive complexity and horror aequi as factors determining the use of interrogative clause linkers in English Günter Rohdenburg Support for more-support Britta Mondorf

69

119

155

175

205

251

The role of extractions and horror aequi in the evolution of -ingcomplements in Modern English Uwe Vosberg

305

Gerundial complements after begin and start Grammatical and sociolinguistic factors, and how they work against each other Christian Mair

329

Is there semantics in all syntax? The case of accusative and infinitive constructions vs. ί/ζαί-clauses Dirk Noel

347

vi

Contents

Aspects of iconicity and economy in the choice between the s-genitive and the o/-genitive in English 379 Anette Rosenbach Constructional semantics as a limit to grammatical alternation: The two genitives of English Anatol Stefanowitsch Principles of grammaticalization and linguistic reality Olga CM. Fischer Varieties of English from a cross-linguistic perspective: Intensifiers and reflexives Peter Siemund

413

445

479

Non-standard English and typological principles: The case of negation 507 Lieselotte Anderwald 'Eveiy place has a different toll': Determinants of grammatical variation in cross-variety perspective

531

Sali A. Tagliamonte Author index

555

Subject index

561

Introduction With the advent of different kinds of functionalism there has been a growing readiness to incorporate variability into grammatical theory-building, and an increasingly large number of (potentially) universal factors conditioning grammatical variation have come to be recognized. Even frameworks that go back to originally generativist approaches, such as Optimality Theory, have started to incorporate linguistic variation into their theoretical tenets. However, while Optimality Theory is quite appropriate in modelling a small set of hierarchically ranked constraints, it still has a number of serious limitations which are implied or made explicit in several contributions to the present volume. In the past, some functional approaches have predominantly centred on universal and pervasive semantic determinants like the iconically motivated distance principle (Haiman 1983; Givon 1984-1990; Dixon 1991) and still others have considered semantic factors exclusively (Wierzbicka 1988, 1991; Ferris 1993 and to a certain extent also Langacker 1987). By contrast, this book - while paying due attention to iconicity and other general semantic tendencies - is predominantly concerned with the exploration of major extra-semantic and largely neglected factors determining grammatical variation in both present-day English and earlier forms of English (including non-standard varieties). The sixteen contributions selected for this volume are all based on solid empirical research. Together, they cover a broad range of traditional and hitherto neglected variation phenomena. While some of these are frequently mentioned in the literature, they still remain to be fully accounted for. Most of the contributors subscribe to the view that different determinants may either reinforce or be in conflict with each other. Accordingly, several papers provide systematic and sophisticated descriptions of the constraint interactions observed. Despite the fact that virtually all papers invoke more than one factor determining grammatical variation, an attempt has been made to group them thematically. Since all of the contributions are preceded by an abstract, we will give only a very brief indication of the main issues in each individual paper. The first contribution by Krug explores the role played by frequency in grammatical variation and change. The author disentangles the domains promoting and speeding up processes of change from those delaying and retarding them. On the basis of his own research devoted, in particular, to contraction and cliticization phenomena, he critically assesses the value of

2

Introduction

two currently competing frequency concepts, String Frequency and Transitional Probability. The paper by Schlüter dealing with phonological influences on the morphology and syntax of present-day English and Middle English possibly represents the most provocative contribution to this volume. The author analyzes two universal phonological constraints promoting the alternation of vowels and consonants on the one hand and of stressed and unstressed syllables on the other. Her concluding remarks relate the empirical findings to the framework of Optimality Theory and critically assess the theory's merits and shortcomings. While differing in scope, the state of the art paper by Wasow and Arnold and the contribution by Gries are both concerned to show that postverbal constituent ordering is sensitive to a variety of constraints including the syntactic complexity of the phrases and the discourse status of the information expressed. Both papers clearly demonstrate that the effects of factors like these may be mutually reinforcing to a striking degree. In addition, Wasow and Arnold draw attention to a tendency motivated by the distance principle: Semantic connectedness between verb + prepositional phrase favours adjacency. The four following papers (Hawkins, Rohdenburg, Mondorf, Vosberg) revolve largely round the concepts of processing efficiency and processing complexity. Hawkins proposes a theory of adjacency, according to which phrases are close to their heads in proportion to the number of existing dependency relations. While some linguists would regard this correlation as iconically motivated, Hawkins explains it in terms of processing efficiency. On the basis of these findings, he then offers a convincing account of a set of correlations subsumed under Rohdenburg's complexity principle: Explicitly marked phrases are preferred over zero-marked counterparts in cognitively complex environments. Rohdenburg's paper provides a survey of the evolution and present behaviour of prepositional and verbal links introducing dependent interrogative clauses. The variation phenomena and changes observed are accounted for in terms of the complexity principle, which correlates the processing load of grammatical units with their relative degrees of explicitness (and which in important respects goes beyond Hawkins' efficiencybased theory) and the horror aequi principle, which describes the avoidance of (semantically unmotivated) identity effects. Mondorf s contribution, which focuses on the rivalry between synthetic and analytic comparatives of selected monosyllabic and bisyllabic adjectives (e.g. prouder vs. more proud), attempts to extend the basic idea

Introduction

3

contained in the complexity principle to all major linguistic levels: The more explicit variant (referred to as more-support) is argued to facilitate processing in a broad range of phonological, morphological, syntactic, semantic and pragmatic environments. In addition, her analysis discusses the role played by iconicity and invokes various discourse factors as well as phonological identity effects. Vosberg charts the evolution of infinitival and gerundial complements with selected governing expressions over the last few centuries. His most important finding is that various (difficult-to-process) types of extraction have always tended to favour infinitives over gerunds. In addition, it is shown that the distribution of the two alternatives is sensitive to horror aequi, with infinitival and gerundial complements being largely avoided after immediately preceding ίο-infinitives and -ing forms, respectively. The literature abounds in semantic hypotheses concerning the contrast between infinitival and gerundial complements where both are possible. In his paper devoted exclusively to begin and start, Mair finds that the distribution of the two complement types is influenced only by relatively weak and fuzzy semantic tendencies. In addition, he demonstrates that American English is more advanced in the establishment of -ing complements and identifies two horror aequi effects: the tendency to avoid two adjacent -ing forms or two adjacent marked infinitives. The next paper by Noel analyzes the alternation between finite complements and what used to be referred to as object raising structures. Vehemently rejecting any previous attempts to account for their differential behaviour in semantic terms, he succeeds in identifying two discourse factors motivating the distribution of the constructional alternatives: Unlike their finite counterparts, the verbs used in object raising structures (referred to as believe-type infinitive complements) show a distinct preference for contextually given objects and subordinate (i.e. backgrounded) clauses. The rivalry between the s-genitive and the of-genitive is the topic of two papers (Rosenbach, Stefanowitsch), which employ different research strategies and (possibly because of that) obtain surprisingly different results. Rosenbach's paper, which is based on elicitation experiments involving speakers of British and American English, concentrates on the semantic areas compatible with both options. Observing that American English is clearly more advanced in the extension of the s-genitive she finds that in both varieties the distribution of the alternatives is largely iconically motivated. Thus, (more highly) given elements tend to precede less accessible ones. By contrast, Stefanowitsch, whose findings are based on a corpus of American English, argues that each of the two grammatical

4

Introduction

alternatives has its own constructional semantics which typically defines a distributional potential complementary to that of its grammatical rival. The fact that the two genitives do overlap in their ranges of application is ingeniously accounted for in terms of a subsidiary contributory factor, the inherent semantics of the lexical items involved. Based on previous empirical research of her own, Fischer's paper provides a much needed critical assessment of the major principles of grammaticalization theory. Specifically, she calls into question the principle of unidirectionality and stresses the roles played by iconicity and syntactic persistence (the influence of synchronic language structure). With the last three contributions (Siemund, Anderwald, Tagliamonte) we move away from Standard English into the fields of dialectology and sociolinguistics. The papers by Siemund and Anderwald represent a recently established research paradigm in which grammatical analyses of non-standard varieties are guided by reference to general typological principles. In his discussion of intensifiers and reflexives, Siemund shows amongst many other things that the distribution of self-forms in nonstandard varieties of English is fully accounted for by various grammatical hierarchies such as the case hierarchy. In a similar vein, Anderwald demonstrates that the enormous post-war increase of non-standard uses like he don't or we wasn't is ultimately due to the operation of various typological principles including, in particular, the (iconically motivated) tendency to reduce morphosyntactic marking in negative environments. Tagliamonte analyzes the distribution of stative possessive have and its more recent rivals have got and got in three non-standard varieties which are more or less isolated from mainstream British English. While the three varieties can be seen to represent different evolutionary stages, they are sensitive to the same kinds of internal constraints (e.g. negation, use of abstract or concrete objects), thus again showing the importance of syntactic persistence. Finally, we wish to thank the series editors and the contributors for their fruitful cooperation in preparing this volume. We would also like to acknowledge the technical assistance provided by a number of students, in particular Eva Berlage and Andreas Mankel. References Dixon, Robert M.W. 1991 A New Approach to English Grammar, Oxford University Press.

on Semantic

Principles.

Oxford:

Introduction

5

Ferris, Connor 1993 The Meaning of Syntax: A Study in the Adjectives of English. London: Longman. Givön, Talmy 1984-1990 Syntax: A Functional-Typological Introduction. Volumes 1 and 2. Amsterdam/Philadelphia: Benjamins. Haiman, John 1983 Iconic and economic motivation. Language 59: 781-819. Langacker, Ronald W. 1987 Foundations of Cognitive Grammar. Volume 1, Theoretical Prerequisites. Stanford, Ca.: Stanford University Press. Wierzbicka, Anna 1988 The Semantics of Grammar. Amsterdam/Philadelphia: Benjamins. 1991 Semantic rules know no exceptions. Studies in Language 15: 371-398.

Frequency as a determinant in grammatical variation and change* Manfred Krug

Abstract This paper initially surveys a number of venerable frequency-related concepts such as analogy, economy, irregularity and predictability. In a second step, data from ongoing and historical variation are presented that help to explain the progressive and conservative effects of frequency in the formation of grammatical constraints. Examples include the spread of the 3rd Ps. Sg. -s inflection at the expense of -th during Early Modern English; HAVE and BE cliticization onto pronouns in present-day spoken English (British National Corpus, Bank of English Corpus, London-Lund Corpus); and negation strategies in Late Middle English (Helsinki Corpus). Finally, two frequency factors are discussed that figure prominently in current research: String Frequency and Transitional Probability. From the analysis of present-day variation between full and enclitic forms of NOT co-occurring with auxiliaries (e.g. would not ~ wouldn Ί) it emerges that, rather than being rivals, the two factors can be profitably combined.

1. Introduction During the last decade the role of frequency in the description of language and language change has received considerable attention, in particular from linguists working from a functionalist or usage-based perspective (see e.g. the papers in Bybee and Hopper 2001). But despite the fact that the turn of the millennium is the age of large computer corpora, present-day linguists are not the first to realize the significance of frequency as a determinant in grammatical variation and change. Many fundamental insights go back to the late 19th and the early 20th century. These will be surveyed in Chapter 2 of this paper. Interestingly, high discourse frequency has two apparently diametrically opposed effects, one being innovative, the other conservative. Chapters 2 and 3 will present a host of examples from phonology, morphology and syntax that help to disentangle the different forces and reconcile the apparent conflict between the two. Subsequently, I will discuss and exemplify two factors which have been proposed more recently but which build on the insights gained over the last

8

Manfred Krug

few centuries. These are String Frequency (SF) and Transitional Probability (TP). I shall test these two factors, which are sometimes described as rival factors, against new data, viz. the fusion of auxiliary verbs and NOT (e.g. would not ~ wouldn't). It will be seen that the two measures are in fact not rivals but highly related, both mathematically and conceptually. Nevertheless, each of them seems to produce the best results in different linguistic contexts. Finally I will suggest that the two measures of String Frequency and Transitional Probability can be profitably combined.

2. Time-honoured concepts: Irregularity, analogy, economy and predictability A number of frequency-related concepts have been proposed over the last two centuries, most of which are highly interrelated. The correlation between high frequency and irregularity (e.g. English BE, German SEIN) has been known at least since the beginning of the 19th centuiy. Witness Jacob Grimm (1822: 851): „Auxiliaria, d.h. verba, welche sehr häufig gebraucht werden und statt ihrer lebendigen bedeutung abstracte begriffe annehmen, tragen gewöhnlich solche Unregelmäßigkeiten an sich." ['Auxiliaries, i.e. verbs which are used very frequently and which take on abstract notions instead of their vivid meanings, usually display such irregularities.']. Nübling (2000) is a book-length treatment of this phenomenon in the Germanic languages, treating in great detail ten high-frequency verbs in ten Germanic languages. Inverse evidence for the correlation between high frequency and irregularity comes from analogy. In analogical levelling, infrequent items are affected first, whereas frequent items resist change - a fact already noted by Hermann Paul at the end of the 19th century (['1880] 1975: Chapter 5). In the Germanic languages, long-term analogical developments that are still in progress can be found for both nouns and verbs. Through the process of analogy, for instance, unproductive patterns such as -en plurals or formerly strong (and in other ways irregular) verbs are regularized. Consider the following examples of an infrequent verb (1) and a verb of intermediate frequency (2), both of which are regularizing, as is indicated by the co-existence of productive and unproductive participles. Then contrast a highly frequent form (3), which shows no sign of incipient regularization by analogy: (1) (2)

hove-heaved burnt ~ burned

Frequency as a determinant in grammatical variation and change

(3)

9

gone ~ *goed

Hooper (1976) and Bybee (1985: 119-120) demonstrate that the differences displayed by the items above are not anecdotal or accidental but can be integrated into a systematic, frequency-driven account. However, it need not necessarily be only one productive pattern that coexists with several unproductive patterns at a given point in time, even though this is the present-day situation for English inflections. Consider the case of plurals: Unproductive ~{e)n forms (like oxen), zero-marked plurals (such as sheep, fish) and i-mutated forms (e.g. mice, geese, men) exist alongside the productive ~{e)s plurals (e.g. cats). In principle, however, it is perfectly possible for patterns of different degrees of productivity to coexist. This is, in fact, typically the case in word-formation. The basic tendency applying to such competing forces is outlined by Cassidy and Ringler (1971: 37, emphasis added): "In any language, the more frequent structural patterns or those having a larger number of members constantly exert pressure upon the less frequent to conform by analogy, and thus to reduce irregularity". The italicized passage hints at potential empirical problems when it comes to quantitative verification, since frequency can be measured along different parameters. Possible criteria are: 1) Absolute numbers of occurrences of a certain form, e.g. the incidence of irregular past forms was or were compared to (the sum of all) individual instances of regular past tense ~(e)d forms. 2) The number of verbs that belong to a certain paradigm, e.g. felt, smelt etc.; shown, blown, mown, flown and the like. Decisions as to what exactly forms a paradigm will not always be straightforward. Does, for instance, the group of -n participles constitute one major, rather heterogeneous, class (including, e.g., blown, gone, done, frozen, seen, given) or are potential subclasses to be treated individually? Some related problems are discussed by Paul (1975: Chapter 5) on a theoretical basis. A step forward in tackling the empirical problems is Bybee (forthcoming), who distinguishes two kinds of frequency: token frequency and type frequency. She defines token frequency as "the frequency of occurrence of a unit, usually a word or morpheme, in running text." This absolute token frequency is often normalized (or standardized) by relating it to a certain amount of text, which gives the incidence of an item. For convenience and comparability, token frequency will therefore often be given per 10,000 words or per million words of running text. For example,

10

Manfred Krug

the token frequency of left (past tense form only) in the written LancasterOslo-Bergen Corpus is 153 times per million words, and compares to a token frequency of 17 occurrences per million for preferred (figures from Johansson and Hofland 1989 I: 211, 271). One can also compare token frequencies of constructions, both within a given corpus and across different corpora. Auxiliary (rather than spatial verb) BE GOING TO, for instance, has a token frequency of roughly 4 per thousand in the (informal) speech collected for the American Santa Barbara Corpus (which currently contains some 60,000 words). Thus it is roughly twice as frequent in this American corpus as in the spoken component of the British National Corpus (which contains some 10m words).1 Type frequency, on the other hand, is defined by Bybee (forthcoming) as "the dictionary frequency of a particular pattern, e.g. a stress pattern, an affix, etc." The highest type frequency for expressing plural with English nouns, for instance, is found for the suffix -(e)s, as in bushes or computers, which can be found with thousands of nouns. The type frequencies of -en plurals is three (with only one straightforward case, viz. oxen; the remaining two - children and brethren - involve additional morphological and/or semantic changes), that of pluralization by i-mutation seven {feet, geese, lice, men, mice, teeth, women). In addition, there are dozens of zero plurals, depending on how one counts them, plus some Latin and Greek plural formations whose details need not concern us here (cf. Quirk et al. 1985: 306-310). Given this preponderance of -(e)s types, it is obvious that every neologism will follow this particular pattern. Further empirical studies addressing the question of how exactly to count types and tokens are Bybee (1998), Bybee and Scheibman (1999) and Corbett et al. (2001). Even though the effects of type frequency and token frequency in language change are not fully known, it is widely acknowledged that type frequency is relevant for the productivity of a pattern, i.e. the probability for neologisms to follow a given pattern (Bybee 2001b: 119). What is fully clear, however, is that the ideal basis for frequency counts is (spontaneous) spoken discourse, especially if one wants to explore the motivations for frequency-driven diachronic processes of change such as analogy or some types of sound change. Definitions of analogy have been notoriously vague in the past (cf. e.g. Campbell 1998: 89). Building on work by Trudgill (1973, partly reprinted in Chambers and Trudgill 1998), I have modelled analogy in language change mathematically by making use of the notion of gravitation (Krug 2000: Chapter 5.7). According to this model, analogical pressures in language are comparable to gravitational forces. This means that the forces of

Frequency as a determinant in grammatical

variation and change

11

attraction between physical masses operate in similar ways as those between conceptually related linguistic items. More exactly, these gravitational forces are, in linguistic terms, the analogical pressures that a given item (or group of items) exerts on other items. As a consequence highly frequent items or patterns are assumed to have a strong influence on changes affecting similar items. The relative magnitudes of mutual influences within a given group of items (e.g. verb forms or word formation processes) can then be determined statistically in order to delineate likely paths of development.2 Such a model can incorporate different levels of productivity since frequency (be it type or token frequency) enters the formula as the equivalent for mass. And mass is the principal positive parameter for determining and influencing gravitational forces. Zipf s (1929, 1949) studies of the now well-known economy principles have corroborated some important interdependencies. First, there exists a correlation between high frequency and the age of words. Second, these two in their turn correlate inversely with structural simplicity (i.e. little morphological and phonological bulk). Put plainly, frequent words tend to be old and short. In support of these principles one could adduce frequency rankings of any sufficiently large corpus which contains a variety of text types. Two frequency lists given in Renouf (1992: 303-304) for the spoken and written components of the then called Birmingham Corpus (later renamed Bank of English Corpus, for details see the list of corpora preceding the references) may serve as an illustration. Among the top ten words in either corpus are the following nine: the, of, and, to, a, in, that, it, I. The two words that figure among the top ten items in only one corpus each are was and you. None of these eleven words is morphologically complex, none consists of more than three phonemes; and some of them - such as a < an(e), I < ic(h) have become even shorter since the earliest English records. The first ten words are no exceptions. Based on Renouf s frequency lists, Table 1 presents a classification of high-frequency items according to their number of syllables. Table 1. Mono-, di- and trisyllabic words among the top 150 words in the spoken and written Birmingham Corpus (based on Renouf 1992: 3 0 3 - 3 0 4 ) Top 50 words Number of syllables

51-100

101-150

1

2

3

1

2

3

1

2

3

Written component (18m words)

50

0

0

40

10

0

38

11

1

Spoken component (1.3m words)

46

4

0

45

5

0

41

8

1

12

Manfred

Krug

Approximately 130 of the top 150 words are monosyllabic, irrespective of the medium (spoken or written). Furthermore, the number of monosyllabic words shrinks progressively from the first fifty to the next fifty and so on. Conversely, the number of disyllabic items increases in both corpora with decreasing discourse frequency. Finally, there is only one trisyllabic item in each component (another and actually in the written and spoken component, respectively).3 Zipf s account of the correlation between the three factors frequency, shortness and old age in synchronic cuts of a language, then, is uncontroversial.4 The more interesting question is diachronic in nature. How do such correlations come about? On the face of it, the diachronic path seems obvious from the synchronic facts, too: Other things being equal, the older a word gets, the shorter and the more frequent it should become. Consider a paradigm case of grammaticalization: The English indefinite article a - as in many other languages - developed from the numeral an(e) 'one'. However, a drastic increase in discourse frequency over time is the exception rather than the rule. Many old words remain long and infrequent. This is in fact the default situation, since most lexical items do not grammaticalize. What is less clear, furthermore, are the relationships between frequency, old age and morphological simplicity. What are the causes and what the consequences? These questions require a good deal more research. One commonly held belief is that older items first erode semantically and so can be used in more contexts. This process of generalization is assumed to lead to frequency increases. Unfortunately, however, we might be facing a chicken-and-egg situation with regard to generalization and frequency increases. It is equally plausible that mutually reinforcing forces are at work, since frequency gains seem to contribute to loss in lexical, specific meaning and thus to the appropriation of abstract, more grammatical meaning. While the issue of causes and consequences cannot be settled at the present stage, the appropriation of abstract, grammatical meanings during generalization suggests that a fourth concept is intertwined with the Zipfian concepts of frequency, age and shortness: grammatical status. As for empirical evidence, Table 2 (again based on the frequency lists in Renouf 1992: 303-304) reveals that the most frequent words are not only generally old and short; the vast majority of high-frequency items belong to the relatively closed grammatical classes of pronouns, determiners, prepositions, conjunctions and auxiliary verbs (more exactly, primary and modal verbs). Among the top 50 items in the two corpora, a mere seven (out of 100) items do not belong to any of the grammatical classes, and some of these (e.g. yes, well) could be argued to be at least closed-class items.

Frequency as a determinant in grammatical variation and change

13

Table 2. High-frequency words (the top 50 from Renouf 1992: 303-304) according to word-class: The interrelation between frequency and grammatical status

Written Spoken A)

prona)

det

14 8

5 4

prep conj 7 7

5 5

prim vb 8 6

mod vb 1 2

other grcal 1 2

ambig grcal 6 7

ambig grcal-lex 2 3

lexical 1 6

= pronouns; DET = determiners; CONJ = conjunctions; PRIM VB = primary verbs; MOD VB = modal verbs; other grcal = grammatical items that do not belong to any of the aforementioned classes (e.g. not, very)\ ambig grcal = ambiguous between at least two grammatical word classes (e.g. that, as, /o); ambig grcal-lex = ambiguous between grammatical and lexical classes (e.g. there, like, no)

PRON

Interestingly, this correlation between high frequency and grammatical status is not restricted to general word-frequency lists. Just as it applies across different word classes, so it applies within the more specific domain of verbs. Among the top thirty verbs of English, the proportion of auxiliary verbs is significantly higher than among less frequent verbs (Krug 2000: 24). In other words, the most frequent verbs are also most likely to be grammatical. More generally, therefore, we can state that there is a correlation between high frequency and grammatical status. To return to the question of causes and consequences, Zipf (1935: 29) himself assumes that "high frequency is the cause of small magnitude" but Haspelmath (1999: 192) has already noted that he "did not explain how frequency shrinks linguistic units." Part of the answer lies in the fact that in the diffusion of sound change frequently used words are often affected first (Bybee 2001b: 11). This insight is not entirely new. In his critique of the Neogrammarian position that sound changes are exceptionless, Schuchardt (1972 [1885]: 24-26), for instance, assumed there to be such a correlation, as well as one between frequency and irregularity (in a similar vein, Aitchison 2001: 86; Labov 1994: 506-507, 530): Die grössere oder geringere Häufigkeit im Gebrauche der einzelnen Wörter welche ja bei den Analogiebildungen eine so hervorragende Rolle spielt, ist auch für ihre lautliche Umgestaltung von hoher Wichtigkeit, nicht innerhalb kleinerer, wohl aber innerhalb bedeutender Differenzen. Sehr selten gebrauchte Wörter bleiben zurück, sehr häufig gebrauchte eilen voran; von beiden Seiten also bilden sich Ausnahmen von den Lautgesetzen. Es ist schon eine sehr alte Erfahrung dass in allen Sprachen gerade die allergewöhnlichsten Wörter, von denen man doch am Ersten Gehorsam gegen die Lautgesetze erwarten sollte, am Meisten Neigung zeigen sich von ihnen zu emancipiren, ... ['The greater or lesser frequency in the use of individual words that plays such a prominent role in analogical formation is also of great importance for their phonetic transformation, not within rather small differences, but within significant ones. Rarely-used words drag behind; very frequently used ones hurry

14

Manfred Krug

ahead. Exceptions to the sound laws are formed in both groups. It is an old experience that in all languages the most common words show the greatest inclination to emancipate themselves from the sound laws. (And it is just from these words that one would expect obedience to the sound laws.)' Translation from Vennemann and Wilbur (1972: 58)]

Stock examples in the older literature are Romance and Germanic words meaning 'go'. More recent work (e.g. Zwicky 1972; Hooper 1976; Bybee 2001b: 11,41; Aitchison 2001: 86) commonly invokes the absence or presence of schwa in the penultimate syllable of words like every, factory, nursery, slavery or cursory. As Hooper (1976) has shown, deletion of schwa is most likely in the most frequent word in this set, viz. every. In sound change, therefore, frequent items appear to be prone to change. That this statement needs qualification will be seen shortly. It is in particular though not exclusively (Labov 1994: 433, 506-507, 530) - leniting historical changes that favour frequent items (e.g. Fidelholtz 1975; Bybee 1998; Pierrehumbert 2001). Connected with this economical principle concerning the leniting of frequent items is the final time-honoured frequency-related notion to be discussed here: predictability (sometimes equated with probability). Accordingly, words that are highly predictable from the context are shortened. The link between frequency and predictability is easily established, of course, because only what (co-)occurs often can be predictable. Such contextual predictability - and its interaction with degrees of informativeness, processing efforts on the part of the hearer and least effort tendencies on the part of the speaker - was already conjectured to play a role by Schuchardt (1972 [1885]: 27), too: Von Bequemlichkeit ist überall die Rede wo die Ursachen des Lautwandels in Erwägung gezogen werden; was ist nun natürlicher als dass man es sich da am Ersten bequem macht, wo in der Ueberhäufigkeit der stärkste Antrieb dazu liegt und die Gefahr des Missverständnisses am geringsten ist? ['There is always talk about the principle of least effort whenever the causes of sound change are being debated. What is more natural than making things easier whenever over-frequency provides the strongest impulse for this and wherever the danger of misunderstanding is least?' Translation by Vennemann and Wilbur (1972: 59)]

The same argument is repeated in Jespersen (1922: 267-268). In more recent studies, Fowler and Housum (1987) and Jurafsky et al. (1998, 2001) have found strong empirical support for the interdependence between contextual predictability and weakening. As will be seen in section 3.2.3, in particular Jurafsky et al. have developed the concept of predictability further, both mathematically and theoretically.

Frequency as a determinant in grammatical variation and change

15

2.1. Recent refinements 2.1.1. Entrenchment Fundamental to cognitivist thinking on the role of frequency is Bybee's (1985: 117) statement that every mention of a word leaves a trace in the mental lexicon: If we metaphorically suppose that a word can be written into the [mental] lexicon, then each time a word in processing is mapped onto its lexical representation it is as though the representation was traced over again, etching it with deeper and darker lines each time. Each time a word is heard and produced it leaves a slight trace on the lexicon, it increases its lexical strength.

Within the framework of cognitive grammar, Langacker (1987) has introduced the notion of entrenchment for modelling the strength of mental representation. As the following quotation illustrates, the strong point of this concept is that it can account for both the emergence and the extinction of a grammatical structure: Every use of a structure has a positive impact on its degree of entrenchment, whereas extended periods of disuse have a negative impact.5 With repeated use, a novel structure becomes progressively entrenched, to the point of becoming a unit; ... (Langacker 1987: 59).

Recently it was discovered that frequency and entrenchment have significant consequences for the debate over fixedness and age-thresholds in the grammars of individual speakers. Croft (2000: 57-58) and Aitchison (2001: 202-204) cite evidence for the position that adult grammars (and lexicons) are not fixed but can change through shifting frequency conventions among different social groups. Inverse evidence for the role of frequency is provided by the fact that adults can lose part of their nativespeaker ability when they are no longer exposed to their mother tongue, as in the case of emigrants. Conversely, "nativelike ability can be regained, at least to some extent, if the speaker is placed in the native speech community again" (Croft 2000: 58). In a similar vein, a long-cherished belief in the field of second language acquisition has recently been qualified. While conceding that the ability to acquire a foreign language does slow down during adolescence, Croft (2000: 57) notes that "there is no sudden cutoff' for the possibility of foreign-born speakers to acquire nativelike proficiency.6

16

Manfred Krug

2.2. Resolving the conflict between conservative and progressive effects of frequency: Some empirical evidence 2.2.1. BE An important issue is how to reconcile the conservative and progressive effects of frequency (cf. Bybee 1998). The case of BE serves well to illustrate both effects. To begin with, in all languages that have copular verbs, it is the most frequent verb. In English (but also in other languages) several of the well-known principles apply: First, it is highly irregular. In fact, with eight different forms - be, am, are, is, was, were, been, being — BE is the most irregular English verb and its paradigm possesses twice as many different forms as a regular, weak verb. Second, as Zipf would have predicted, BE is old and contains little phonological bulk, no matter which of its many verb forms we look at. Furthermore, it shows no tendency whatsoever towards regularization in standard English: *beed (for PAST or PERFECT) and *bees (for 3SG PRES) are equally ungrammatical. This fact is usually seen as evidence for the hypothesis that high-frequency items are stored and accessed individually in the mental lexicon, unlike those that are stored with other items following a similar pattern (e.g. phonologically conditioned past-tense affixes or smaller groups like keep-kept, weep-wept, etc.). Furthermore, shorter variants of individual word forms coexist, some of which are reflected in the spelling, e.g. 's, 'm, 're, while others such as [WDZ ~ wz] aren't. Such variants are interesting to the historically and theoretically minded linguist in that they almost always reflect diachronic weakening and assimilation processes which help us understand the evolution of cliticization and morphophonemic alternation. Compare the different allomorphs of the third person singular of BE: (4) (5)

[iz] grammatical word > clitic > inflectional affix

In a related fashion, the above frequency-driven development illustrates the path from syntax (is) via morphology to morphophonemics ([s, z]) in Givon's well-known cyclic wave from discourse pragmatics to zero (1979: 208-209, see [7] below). This evolutionary functionalist approach has been called the discourse-shapes-grammar tradition. It can safely be considered a frequency perspective since it subscribes to the position that through repeated use discourse strategies sediment as syntax, which in its turn becomes morphologized over time through the same mechanism of repetition. Ultimately, continued erosion can lead to the loss of overt grammatical information. (7)

Discourse —> Syntax —» Morphology -» Morphophonemics —> Zero

Usually the same cycle starts again in a given functional domain. The emergence of a new grammatical morpheme (or layer), however, does not set in only after an older related morpheme is lost but typically happens while the older morpheme is traversing the earlier stages of the path

18

Manfred Krug

outlined in (7). This ever-recurring process, found in almost any area of grammar of any living language and leading to the layering of newer and younger variants (Hopper 1991), has been labelled an evolutionary spiral (rather than a cycle) by van der Gabelentz (1891). If we look at some specific high-frequency grammatical items - both constructions and individual words like will ~ 71, want to ~ wanna, have ~ 've, going to ~ gonna or you [ju:] ~ [ja] - it turns out that the coexistence of older (and longer) and younger (and shorter) forms is a typical concomitant of discourse frequency. This phenomenon has been called multiple representation (Bybee 2001b), which is meant to indicate that frequent items often develop more than one relatively stable representation and more than one overt phonological form. These forms can be free variants. Often, however, such variants develop morphosyntactic idiosyncrasies: (8) Will/ *'ll he be at home? (9) Are you going to/gonna talk to her? (10) Are you going to/ *gonna the cinema? Roughly, then, the progressive and conservative effects of frequency relate to different levels of linguistic organization: In phonology, high frequency is a progressive force that causes the lenition and erosion of overt phonetic substance and thus ultimately the emergence of shorter, relatively stable variants on the path from syntactic constructions and grammatical words to clitics and zero. Here, frequency seems to be a Zipfian, speaker-based principle of least effort that leads to automatic reduction. In morphology, by contrast, frequency turns out to be normally a conservative force inhibiting, for instance, regularizing tendencies such as analogical levelling. However, qualification and refinement is necessary because exceptions to this distribution of tasks do exist. Just think of exceptions to the Great Vowel Shift (such as you or could), where quite conceivably the retention of the vowel quality - i.e. a conservative trait could be due to high entrenchment.9 But despite the existence of lexical approaches to sound change (e.g. Ogura 1987; see Labov 1994: Chapter 17 for a critique), the exact workings of frequency in phonological change are not sufficiently understood. Certainly entrenchment alone cannot explain why, for instance, the Great Vowel Shift regularly applied to other highfrequency items like I, he or she. Studies focusing on the effects of frequency in historical and ongoing phonological change are therefore likely to yield important insights.

Frequency as a determinant in grammatical variation and change

19

2.2.2. Third person singular simple present tense: The change from -th to -s during Early Modern English An example of frequency effects in morphology and phonology is the historical development of the 3rd person inflection for the simple present tense, i.e. the spread of the originally northern -s morpheme at the expense of -th. During Early Modern English (1500-1700), the older dominant and southern form (e.g. desireth) was gradually superseded by the -s form. It is an interesting if well known fact that the -th forms of DO and HAVE, i.e. doth and hath, survived far longer than those of other verbs (e.g. Barber 1993: 185, 191; Kytö 1993). A number of factors have been shown to have played a role in promoting the spread of -s across all verbs. These include phonotactic, stylistic, discourse and social factors such as region, class and sex (see for instance Stein 1987; Kytö 1993; Nevalainen 1996b). Occasionally, frequency is mentioned as a determinant too (Holmqvist 1922: 186— 188; Kytö 1993: 121-123; and in particular Ogura and Wang 1994). Perhaps typological facts provide an additional clue to why it was the -s morpheme that has ultimately become the standard present-day inflection and indeed the sole surviving form. The (inter-)dental sounds [0] and especially the voiced [ö] in final position are rare in the languages of the world and thus marked sounds. When marked and unmarked forms compete on the same linguistic ground, it is more likely for the unmarked form to oust the marked one than for the reverse development to occur. To the extent that unmarked forms are crosslinguistically frequent (and marked forms crosslinguistically rare), markedness could be interpreted as a typological frequency factor. It is not fully clear, though, whether markedness per se will always constitute a sufficient explanation for change. If this were the case, then analogical pressures would constantly be at work across languages - an assumption which may seem far-fetched, but it is not alien to research in linguistic areas or Sprachbünde. On the other hand, typological markedness may often also be a reflex of a more fundamental force, for instance, the speaker's and hearer's economy (as probably in the case of marked and unmarked phonemes, on which see the following paragraphs; or in universal word order preferences, on which see Hawkins, this volume). While the details of the spread of -s in English are complex, it is frequency of occurrence which is probably the single most powerful factor. This at least is strongly suggested by the figures for Early Modern English verbs in the Helsinki Corpus given in Ogura and Wang (1994: 133-138). They point out that "the change started slowly from a handful of high-

20

Manfred Krug

frequency words [and that] once the infrequent verbs got started, they changed more quickly than the frequent verbs (1994: 122)." Rather than attempting to explain the frequency effects, however, relevant studies do not usually extend beyond stating the correlation between high frequency and the retention of the older prestige form during the later stages of Early Modern English, when most verbs had been regularized to virtually the present situation. From a theoretical point of view, three considerations need to be stressed. First, regional factors apart, it is variation within high-frequency words that initiated the change. This is not untypical, as was seen in the previous section. Second, it seems reasonable to assume that it was the high entrenchment of the forms hath and doth which retarded the analogical levelling of these two verbs. It is no coincidence therefore that SAY, the third most frequent verb after HAVE and DO in Ogura and Wang (1994: 133), displays a similarly conservative behaviour. A third factor seems to have been largely overlooked in previous studies, probably because these attempted to account for the change either morphology-internally or in regional terms, rather than look at the potential interplay between phonology and morphology. The additional factor I am proposing here can best be accounted for within the framework of natural phonology, as it can be viewed as a phonetically conditioned change motivated by greater ease of articulation. It is a well known fact that [Θ] and [ö] are marked (and rare) sounds in the world's languages, not only in final position (Maddieson 1984: 45). One reason for this crosslinguistic rarity is probably that the tongue position of the interdental fricatives is almost maximally front10 and hence their production involves a high degree of motor activity: movement of tongue and lower jaw. Such a high production cost would certainly not qualify as user-optimal in Haspelmath's (1999) account. It seems therefore reasonable to assume that more languages had at some point (inter-)dental fricatives but that over time many have been substituted by phon(em)es requiring less articulatory effort. Varieties of English in fact can serve to illustrate this point. Consider some variants and substitutes commonly found in vernacular speech: -

-

r/z-stopping [t, d] reduces motor activity drastically. No movement of the lower jaw is involved and, being dental or alveolar, the tip of the tongue is more central than in interdental position. Alveolar fricatives [s, z] are much like the alveolar stops. In fact, they can be regarded as further lenited than stops since they are typical lenitions of homorganic stops (Campbell 1998: 41).

Frequency as a determinant in grammatical variation and change

-

21

Labiodental fricatives [f, v] require only jaw or lip movement, while the tongue does not participate. Glottal stops, which involve no tongue movement, and ί/ζ-deletion are found mainly as full or partial substitutes for the definite article.

It should be noted that most of these variants are not restricted to vernaculars (or even second language varieties). While it is true that dialects and English-based Creoles are further advanced in this respect since they use the variants more systematically, it is also true that at least some of the above variants are common features in the allegro speech of standard speakers, in particular when high-frequency items occur in phonetically favourable environments such as get the teacher to talk to your father, where the th of the article is much more likely to be stopped, globalized or omitted than that in father. If substitution of marked phones indeed played a role in the spread of -(e)s at the expense of -(e)th, we still need to know why it was the alveolar fricatives that spread, and not any of the other variants commonly found for [Θ, Ö]. What springs to mind, an account in terms of maximal similarity with the erstwhile dominant inflectional morphs, does not yield any clear favourites: Each of the variant pairs11 [s, z], [t, d] and [f, v] share two features with [θ, δ], either place of articulation and the voicing contrast or manner and the voicing contrast. Crucially, however, there existed dialectal variation so that, unlike word forms ending in stops or labiodental fricatives, -(e)s forms did not have to be invented ex nihilo. An obvious functional factor prevented [t, d] from being selected, since these constituted already the productive pattern for past tense and present perfect forms. Merger of present and past would have rendered English verb inflection dysfunctional. This leaves us essentially with a choice between [s, z] and [f, v] (see Croft 2000 on the principles at work in the selection of linguistic variables during language change). Here, a frequency-based explanation may provide an important clue. English plurals and possessives already were largely regularized when the spread of the verbal inflection ~(e)s occurred. Therefore, since all productive nominal inflections ended in [s] or [z], these two sounds enjoyed a much higher frequency than any other word-final sound. This is likely to have caused analogical pressures on -th inflections. The spread of -(e)s, then, appeals to the interplay between analogy, iconicity and functional factors: a) analogy, because [s, z] were extremely frequent in word-final position; b) iconicity, because [s, z] were (and still are) almost iconic of inflectional suffixes in English since [t, d] were not

22

Manfred Krug

eligible due to c) functional factors, which prevented the merger of simple past and third person present inflection. It must be admitted, however, that the account offered here leaves unexplained why it was not the zero suffix that spread. Considering the existing present-day variation (e.g. in East Anglia or African-American Vernacular English) and general principles of language change (see Givon's cyclic wave quoted on p. 17 above), loss of the third person inflection is in fact not an unlikely development in the long-term future. If the above account is correct, then the historical change from -(e)th to -(e)s in the third-person inflection can be attributed to frequency-driven motivations in various guises. Such an account is furthermore related to the first point mentioned earlier on: It is high-frequency items that are most likely to be affected by leniting phonological changes, starting as they do by developing phonological variants (see above). This explanation, finally, also reconciles the seemingly peculiar fact that the change started with high-frequency items but went to completion last with exactly the same verbs. One might wonder why the most frequent verb, BE, is not among those verbs that are conservative with regard to -th retention, but this fact can be explained on historical frequency grounds. While -th and -s forms such as beth, bes, beoth and byeth did exist in Middle English, it seems certain that due to the continued preponderance of the suppletive forms is/ys (which had been dominant at least since Old English times), the regular -th and -s forms of BE played only a negligible role in Early Modern English. On the other hand, the forms hath and saith, which are conservative with regard to their inflectional morphology, also display the progressive effects of frequency, since both exhibit phonetic erosion.12 Regular Old and Middle English verbs suffixed -eth/-es in the third person singular indicative of the simple present tense (the ME schwa in the suffix used to be a full vowel in early Old English). HAVE and SAY, however, neither retain their original stem-final (semi-)consonant nor does the full regular inflection (including the vowel) surface in Early Modern English. Old English had both a contracted and a full variant of HAVE (haefö, haefad) and SAY (sae^Ö, ssejaÖ). In Early Modern English, however, it is only variants that have undergone vowel and/or consonant elision that survive: hath ~ has and saith ~ says. Again, therefore, both verbs are progressive as far as leniting phonetic processes are concerned, but conservative morphologically. The following section will present further evidence of frequencydriven effects in Early Modern and present-day English variation and change.

Frequency as a determinant in grammatical variation and change

23

3. Extending ZipPs principles to word dyads and constructions 3.1. String Frequency It seems an obvious step to test whether Zipfian principles also apply to word sequences and contiguous constructions. If so, this would suggest that, just like single-word items, strings of immediately adjacent items also undergo erosion diachronically and are stored separately as chunked units in the mental lexicon (Bybee 1998, 2001b). Krug (1998) provides a detailed outline of (and empirical support for) the concept of String Frequency. In essence, it is a principle which captures the observable tendency that, ceteris paribus, the more frequent a given string of words is (hence the term String Frequency) in spoken discourse, the higher becomes the probability that its component items undergo univerbation (or coalescence, contraction, fusion).13 The 'other things' that need to be controlled include, in particular, formal factors, i.e. syntactic and phonological properties, but also discourse-semantic factors (such as informational content and recoverability) which are likely to affect processing. For two linguistic items X and Y, String Frequency is therefore defined as the absolute frequency of the dyad (or string, sequence) XY. Normalization is of course possible, indeed indicated for a comparison across different data bases. To that end, one divides String Frequency by the number of words contained in the corpus. This calculation produces the relative frequency (or incidence) of the string. 3.1.1.

HAVE cliticization

To illustrate the effects of String Frequency, two graphs are given below that are based on two spoken corpora of British English: the London-Lund Corpus (hereafter LLC; c. 0.5m words) and the spoken component of the Bank of English Corpus (BEC; at the time of consultation c. 8.5m words). The most general tendency displayed by the graphs is that a String Frequency effect indeed applies to both sets of data: the higher the frequency of relevant strings (e.g. I have + I've), the higher is the actual contraction ratio.14 For instance, the contraction ratio for the most frequent string [I HAVE = I have + I've] is approximately 80% while that for the less frequent string [THEY HAVE = they have + they've] is roughly 65%. This tendency applies across the board, and results in the strictly increasing graphs (with one exception in each corpus).

24

Manfred

Krug

String Frequency: Occurrences per million words Figure 1. String Frequency and probability of contraction in BEC and LLC: Pronouns with full and clitic HAVE

A number of observations attest to the validity of the data. First, the frequency order of strings is identical in both corpora. From less to more frequent hosts to full or clitic HAVE the order is: here-how-where-therewho-they-we-you-I (see also Figure 2). These hosts were chosen for comparison because distracting syntactic and phonological factors can be largely controlled. All of them are pronouns and end in sonorants (in standard British speech the phonological homogeneity is even greater since all end in vowels). In addition to preserving an identical order, the normalized frequencies for each string (e.g. I have + I've) are also similar in the two corpora. Since the low-frequency strings are difficult to interpret on the basis of Figure 1, where many items fall into the first, very steep part of the curves, Figure 2 presents the figures in a frequency-ranked form so as to facilitate the reading of actual contraction ratios for a given pronoun in the two corpora.

Frequency as a determinant in grammatical variation and change

25

Hosts to HAVE: String Frequency ranking Figure 2. String Frequency in BEC and LLC: Pronouns and have/'ve

Interestingly, a diachronic factor can be isolated from the data as well. The LLC tapes were recorded largely in the 1960s and 1970s, whereas the BEC data are mostly from the late 1980s and 1990s. Hence there is a time-depth between the two corpora of some 25 years. Since the BEC contraction ratios are almost always higher than those in the LLC, the erosion of HAVE in the relevant contexts appears to have progressed further over the past generation. Logistic regression finds two roughly parallel lines with a maximum likelihood factor of 4/3, which means that the probability of contraction has increased by approximately one third. This finding is in line with the general principles that are known to be at work according to a discourse-frequency and typological approach to phonological erosion (see [7] in section 2.2.1 above). Quite possibly, one can abstract from the descriptive tendencies at hand and interpret the curves in Figure 1 as suggesting that the frequency with which two words cooccur in a sequence can be exploited for predicting the

26

Manfred Krug

likelihood of their coalescence so that contraction ratios can actually be read as probabilities. A ratio of 90% would thus correspond to a probability of ρ = 0.9. This is more than a simple recalculation, since it implies that if we know the frequency of a given string (say, 500 occurrences per million words being the value on the abscissa), we should be able to predict the contraction ratio of the string simply by identifying the relevant probability value (say, ρ = 0.65) on the curve - always provided of course that other things (such as the phonological conditions at the word boundary) are comparable. 3.1.2. Emerging English modals In another domain of fusion in English, String Frequency even seems to have a bearing on the grammaticalization of new (or emerging) English auxiliaries (cf. Krug 2000). Figure 3 gives the apparent-time distributions for those emerging modals whose contracted (i.e. progressive) forms and full (i.e. conservative) forms are transcribed in the British National Corpus: GOING TO, GOT TO, WANT TO.

60+

45-59 • gonna

35-44 II gotta

25-34

15-24

1-14 Age groups

• wanna

Figure 3. A study in apparent time: Full and contracted forms in the spoken BNC TO, GOT TO, WANT TO)

(GOING

Frequency as a determinant in grammatical variation and change

27

For a comparison, the bar chart plots the percentages of the innovative forms (i.e. gonna, gotta, wanna) for the respective age groups. It is based on the relevant subcorpus of the BNC, that is, speakers tagged for age (approximately 6m words).15 In this spoken component of the British National Corpus the order of frequency for the three constructions is as follows: (11)

going to > got to > want to16

This frequency order of the three constructions corresponds exactly to that of their contraction ratios, as is shown in Figure 3. For each individual age group, the order of contraction ratios is the same as that for frequency, i.e. the one given in (11). Interestingly but not unexpectedly, the contraction ratio for all three items increases from the older to the younger cohorts. Despite the necessary caution that has to be applied (cf. Labov 1994: Chapters 3-4), three such parallel apparent-time distributions point to change in real time, which is towards higher rates of univerbation. Most importantly, however, both frequency and fusion in their turn seem to correlate with degrees of grammaticalization. GOING TO is not only the construction that most often coalesces in this set, it is also the one most drastically reduced. Consider some prototypical phonetic (British English) realizations for the full and contracted forms (cf. the relevant entries in Windsor-Lewis 1972; Wells 1990; Jones 1997): (12) (13) (14)

['gaüiqtü] [' gDtu] [ ' w D n t u ]

vs. vs. vs.

['ggna] [' gDts] ['wons]

In other words, the correlation suggests that, other things being equal, the higher the frequency of a construction and the higher the rate of fusion, the higher will be its degree of grammaticalization. This is to some extent to be expected given two of Lehmann's (1995: 164) paradigmatic parameters of grammaticalization: integrity and paradigmatic variability. The former predicts that through the process of attrition a potentially polysyllabic, weakly grammaticalized item turns into a more grammaticalized, oligo- or monosegmental item. Paradigmatic variability refers to the fact that during the process of grammaticalization, an item becomes more obligatory for the expression of a certain grammatical function. This, of course, is concomitant with an increase in frequency. On the basis of the present data, therefore, the two parameters of integrity and paradigmatic variability - and thus

28

Manfred Krug

their respective processes, viz. attrition and obligatorification - are closely intertwined factors of grammaticalization. For the above quantitative considerations not to be circular, however, it is important that there should also be qualitative indicators which support the position that the congruence of String Frequency order and fusion rates (GOING TO > GOT TO > WANT TO) indeed reflects degrees of grammaticalization. Such evidence is not difficult to find. Subject independence, for example, one criterion for auxiliarihood proposed by Quirk et al. (1985: 1 3 7 ) , is not fully realized with WANT TO: (15) It's going to/got to rain. vs. *It wants to rain. (16) This paper is going to/ ?has got to/ ? ?wants to show... Behaviour regarding voice neutrality points in the same direction. While propositions regularly preserve logical equivalence in active and corresponding passive sentences of BE GOING TO and HAVE GOT TO when the following main verb is transitive, such variation is not normally found with WANT τ ο :

(17) Someone's going to/got to do it ~It's going to/got to be done. (18) Someone wants to do it ~ ??It wants to be done. As far as meanings are concerned, a similar case can be made: ['gons] or [ gana] are quite detached from go [gau], not only phonologically, but also semantically.17 Traces of 'movement', the lexical source meaning of the auxiliary construction, are rather opaque in gonna (cf. Hopper and Traugott 1993: Chapters 1, 4). Similarly, the original possessive semantics of HAVE GOT is not transparent for the present-day user of modal gotta. For wanna the argument is somewhat more complex. While no trace of the original 'lack' meaning is present, it still expresses volition, like its full verb relative that takes noun phrase complements. Elsewhere (Krug 1998) I have offered further evidence of the workings of String Frequency in coalescence phenomena in English. Generally, it turns out that Zipfian frequency effects indeed surface in word-boundary fusion in all the relevant studies carried out so far (see also next section). Further, morphological aspects of entrenchment that are due to frequency seem to affect even larger phrases and indeed clauses, as will be seen in the next section. Some such instances can be subsumed under the notion of idiomaticization. But this should not lead us to assume that we are dealing with purely lexical phenomena. Indeed, much if not most of what will be

Frequency as a determinant in grammatical variation and change

29

discussed in the remainder of this paper is grammatical in nature and not unexpectedly so. After all, recent research in the functional tradition has firmly established that repeated use leads to the emergence of grammar (see Bybee et al. 1994; papers in Bybee and Hopper 2001), a view which is already encapsulated in DuBois' (1985: 363) statement that "grammars code best what speakers do most." 3.1.3. Evidence from historical variation Subject-Verb order in interrogatives and NOT negation provides evidence for the preservation of older syntax that is almost certainly due to high entrenchment brought about by String Frequency. Compare: (19) He should not/shouldn 't go. (20) Will you go?

vs. vs.

He does not read. Does he read?

In present-day English NOT negation and VS order are distinctive criteria for modal verbs and primary verbs (for simplicity subsumed under auxiliary verbs in what follows). Both features, I believe, have their origin in the high discourse frequency of relevant strings such as will you, should not, am not (and their older forms, of course) in Middle English, i.e. before 1R

the grammaticalization of DO support. This assumption has also been voiced in Krug (2000) and Bybee and Thompson (2000), but neither of these studies presents quantitative evidence. On the face of it, the validity of this hypothesis seems almost self-evident, since BE, DO and HAVE are the three most frequent English verbs, and the modals form the largest homogeneous group among the top thirty verbs, at least in English (Biber et al. 1999: Chapters 5, 6; Krug 2000: Chapter 2). As will be seen, with the exception of DO, the situation in Late Middle English was not drastically different. In order to test the influence of String Frequency, I have investigated the linguistic forms immediately preceding those variants of NOT that occur with a frequency of greater than one in the Helsinki Corpus of Late Middle English (1420-1500; c. 0.2m words). These are not, nott, nat, nought, noght, noghte, noujt and nojt, which together occur 1,335 times. More than half of these instances of NOT are preceded by auxiliaries, and the figure can be raised further if one includes the marginal modals such as DARE and NEED. 19 In order to provide a rough estimate also for construction types other than the auxiliaries, a random sample of one hundred NOT examples (excluding those which occur after punctuation marks) was

30

Manfred

Krug

analyzed. This yields the distribution given in Table 3, which tallies well with the overall auxiliary figures, thus indicating that the random sample was not skewed: Table 3. Items preceding NOT in Late Middle English (Helsinki Corpus, c. 0.2m words) Modals

Other Verbs

Sum Verbs

Noun Phrases

Conjunctions

Prepositions

Sum NonVerbs

5%

2%

(17%)

cd

«j -α £S

•Π

c Η υ

π

"

£Ρ «3

36%

4%

19%

ο 3^

ίa

40% A)

Primary Verbs

1420-1500,

w

tg >

κΙ

5% 25%

α §ο

ο

Ω

§ε

? £ ιζ;

=3

IX,

1% 18%

(83%)

8% 2%

WILL, WOULD, SHALL, SHOULD, CAN, COULD, MAY, MIGHT, MUST ( c a p i t a l i z a t i o n

indicates

the inclusion of orthographic variants and inflected forms such as shalt). B)

DARE, NEED, OUGHT

Table 3 shows that 40% of the items preceding NOT are modal verbs. Since an additional 2 5 % are constituted by the primary verbs BE, HAVE and DO, auxiliary verbs (i.e. those verbs that have retained NOT negation till the ΛΛ present day) together account for roughly two-thirds of all occurrences in the sample. Two-thirds is no small proportion, in particular if one takes into account that the remaining share of non-auxiliary words include non-verbal items (17% in the random sample), most noteworthy among which are pronouns (at 8%) and conjunctions (5% in the sample). These commonly occur in structures that have the following constituent order: Constituent (e.g. object/adverbial) — auxiliary — pronoun — not Some relevant examples are given below ( 2 1 - 2 5 ) . These show that the proportion of auxiliary negation is indeed even higher if discontinuous structures are included in the count. While such longer strings as wyll I not can be covered under a wider definition of String Frequency, I have not included them here since the results are unambiguous even under a narrow definition.

Frequency as a determinant in grammatical variation and change

31

Auxiliary constructions (21) And he saith he had an other sone whos name was philemelus / and more haue I not red of the noble Iason / but... (Helsinki Corpus ME, Caxpro) (22)

With suche a shrewe wyll I not melle. (Helsinki Corpus ME, Ludus)

(23) For at my biddyng woldepou not be/And therfore my woo wyte Y thee; (Helsinki Corpus ME, York) Main verb

BE

and

HAVE

(24) Pat is for to seyn, pat tyme be good and able, and not to ouer-hote, ne to ouer-colde, ne not in the hundyn dayes, ne on other dayes pat am forbeden. (Helsinki Corpus ME, Reynes) (25) oper informacion of hem haue I not at pis tyme. (Helsinki Corpus ME, Capser) With a share of two percent of the auxiliary negations in over thousand instances of NOT (or 1% of all constructions in the random sample), immediately adjacent DO was still rare in Late Middle English. But this is not a surprise finding. In declaratives - the vast majority of the structures investigated here - the spread of DO negation really started only in the 16th century, as Ellegärd (1953: 162) has shown. According to his figures, less than 10% of negative declaratives took DO support before 1500. In sum, the more surprising result is the extremely high proportion of auxiliary verbs in the corpus, which suggests that while it is true that in principle all verbs (including lexical verbs) could be negated with NOT in Late Middle English, it was in actual fact overwhelmingly auxiliaries which did occur before NOT. Only one in three instances of NOT is preceded by a non-auxiliary item. The String Frequency-based entrenchment hypothesis as an explanation for a grammatical constraint on present-day English auxiliary verbs is thus strongly supported. It appears further that a similar tendency holds in Middle English for what is today called operator inversion, i.e. the contrast between auxiliary and main verbs exemplified in (20) above. Since I do not have hard statistical evidence for this claim, the quantitative details of this aspect of String Frequency certainly deserve fuller treatment. Overall, however, the conclusion that it was high String Frequency (leading to high entrenchment) which made these verbs immune to the changes that affected the less frequent lexical verbs, which now regularly take operator DO (NOT) - seems

32

Manfred Krug

uncontroversial (cf. also Bybee and Thompson 2000). This would even seem plausible if in the extant written documents interrogatives with [AUXILIARY — PRONOUN] strings were found to be relatively infrequent. Entrenchment is primarily a matter of (and created by) spoken rather than written discourse, and questions can be assumed to enjoy a far greater currency in face-to-face interaction than in the average Middle English written text, which is certainly not replete with everyday exchanges. In summary, two of the auxiliaries' most salient properties are not accidental constraints but can be motivated, at least partially, by String Frequency. Now that such frequency effects have been substantiated quantitatively for Middle English, it seems only logical to look at the further development in Modern English. Let us therefore continue with a brief discussion of the changing negation and interrogative strategies in Early and Late Modern English. By circa 1700 the use of DO support was regularized to virtually the present distribution, that is, obligatory for lexical verbs in questions and under negation (Ellegärd 1953: 162). As Barber points out, however, ... verbs that resist the use of do in negative sentences include care, come, doubt, know, mistake, speak, and trow: even in the late seventeenth century it is common to find expressions like Ί know not', 'if I mistake not'. Verbs that resist the do construction in questions include come, dare, do, have, hear, mean, need, say, and think: in the late seventeenth century we still find expressions like 'Say you so?', 'What think you?' (1997: 196; emphasis added)

Barber's examples demonstrate very clearly that what happened to the modals and primary verbs is not an arbitrary development that stops at a certain frequency cut-off point. Other frequent verbs display the same tendency and resisted regularization, probably for the same reason: entrenchment of the sequences [VERB — PRONOUN ...?] and [VERB — NOT]. Barber orders the verbs resisting DO in alphabetical order (care, come, doubt, know, mistake, speak, trow; and come, dare, do, have, hear, mean, need, say, think). The typical examples he adduces, however, enjoyed particularly high currency in Early Modern English and include the topfrequency verbs say, think and know.21 In present-day English, these three are among the five most frequent lexical verbs in both spoken and various types of written English (Biber et al. 1999: 373-375; Krug 2001: 310), and there is evidence that the basic situation was not fundamentally different historically for these verbs either.22 Verbs of thinking, saying and knowing as well as most of the other verbs that Barber mentions belong to semantic classes that a language, or rather its users, simply cannot do without.

Frequency as a determinant in grammatical variation and change

33

As was indicated above, entrenchment is likely to play a role in the fixing of larger phrases and clauses where an older syntactic pattern has been preserved until the present day. One might argue though, that here we are dealing with rather more fossilized or idiomatized language. Consider sequences like the following (emphasis added): (26) How goes it? Have you sold anything? (BNC ANF 1049) (27) (28)

Well, I'll go on my own if need be. (BNC KPV 3703) "Give the Germans five Deutschmarks and they will save it", said John Major as Chancellor of the Exchequer, "but give the British five pounds and they will borrow £25 and spend it. " (BNC BPH 1203)

Example (26) is another instance of retained inversion in interrogatives with lexical verbs, even though one might add that GO in some constructions resembles an auxiliaiy from a semantic point of view (e.g. in the GOING TO future; or in Go and see your tutor). If need be 'if necessary', as in (27), is the retention of subjunctives in conditional clauses, which have been declining drastically since Early Modern English. (27), finally, is an example of Subject-Verb inversion after a clause-initial constituent (or a full reported clause), giving a largely obsolete word order: Thus, what unites all three examples is that they are instances of entrenchment. However, they display different degrees of productivity. While (26) and (27) are fossilized fixed expressions, (28) is somewhat more productive since it can still be found with many types of subject (full nouns and pronouns), in different tenses (e.g. says he) and with different verbs of thinking and saying (e.g. thought she\ see Markus 1999: 325-329 for statistical detail). Thus, while entrenchment is operative in all three instances, only (28) reflects a truly syntactic constraint on (or rather, variant of) present-day English that is due to String Frequency. 3.2. Transitional Probability In recent usage-based accounts (e.g. Bybee and Scheibman 1999; Bush 1999, 2001; Bybee 2001a, b) the notion of Transitional Probability (TP) has figured prominently alongside String Frequency (SF). For the sequence of two words X and Y, Transitional Probability is defined statistically as the raw frequency of the instances when X is followed immediately by Y

34

Manfred Krug

(notice that this part of the definition is String Frequency), which is then divided by the raw frequency of X. Mathematically this can be expressed by the following formula: (29) Formula: TP =

Frequency of the dyad XY —M Frequency of X

Put plainly, Transitional Probability gives the percentage of a word sequence (e.g. would you) relative to all occurrences of the first word in that sequence (here: would). Alternatively, percentages can be expressed as probabilities (hence the term), so that Transitional Probability gives the likelihood for a certain word to follow another. 3.2.1. Word-boundary palatalization (Bush 1999, 2001)23 Bush (2001), which is based on his 1999 MA thesis and builds on work by Saffran et al. (1996), has tested the effects of String Frequency and Transitional Probability in word-boundary palatalization. Relevant examples include such pairs as don 'tyou and cat you; or would you and good you. Bush observes that the first dyad in each of these pairs is far more likely to undergo palatalization than the second. The most likely pronunciations of the (former) word boundaries are thus [-tj-] in don't you\ [-d3~] in would you; [-tj-] in cat you; and [-dj-] in good you. Bush finds clear evidence that frequency is the principal factor conditioning higher palatalization ratios. As to whether String Frequency or Transitional Probability is the more accurate frequency factor for this phenomenon, his results are unambiguous. Despite various statistical manoeuvres that improve the performance of Transitional Probability, String Frequency provides a superior description of palatalization in his set of data (cf. Bybee 2001b: 165). Since furthermore Bush's critique of String Frequency is flawed in various ways, some aspects of it need to be addressed here. The following reanalysis of his empirical data, however, should not distract us from the fact that he does provide a number of valuable theoretical points of a general nature. Bush (p. 256) summarizes his study as follows: "word-boundary palatalization is more likely between two words if those words occur together with high frequency. ... [Sjuch a predisposition towards word-boundary palatalization is indicative of a cognitively-motivated chunking phenomenon which causes frequently-used sequences of lexical material to acquire lexical

Frequency as a determinant in grammatical

variation and change

35

storage as single, agglutinated mental representations (cf. Boyland 1996; Bybee and Scheibman 1999)." This single most important empirical result and its theoretical implications are undisputed. It should be pointed out, perhaps, that this result is equally compatible with the predictions of String Frequency and Transitional Probability. To clarify which of the measures is superior in the present case, we now have to turn to a reconsideration of the actual data. In order to test the effects of String Frequency, Bush posits an arbitrary cut-off point of five occurrences that differentiates high-frequency from lowfrequency sequences. Table 4 provides his main quantitative results, including the Chi-square values and their corresponding significance levels.24 String Frequency correctly predicts that high-frequency items palatalize, i.e. undergo coalescence, significantly more often (more exactly: two out of three times) than low-frequency items (which palatalize less than one in three times). As for statistical significance, the resultant Chi-square value (27.5) suggests that the interdependence between String Frequency and word-boundary palatalization is highly significant (p < 0.001). The most frequent type did you alone accounts for 46 out of the total 86 instances of palatalization among the high-frequency items. In order to test whether this sequence skews the results for String Frequency, Bush carries out a Chisquared test under exclusion of this sequence, but the result remains unchanged. This control statistic produces the same high degree of significance ( p < 0.001). Λ ί

Table 4. String Frequency as a predictor in word-boundary palatalization of final -N or -/d/ and initial /j/- (based on Bush 2001: 260-261) High-SF dyads (SF > 5)

Low-SF dyads (SF < 5)

Palatalization (%)

86 (68.8%)

23 (30.7%)

N o palatalization (%)

39(31.2%)

52 (69.3%)

Sum (100%)

125

75 2

Chi-square value

χ =27.5

Significance level

ρ 5%) with those produced by the one chosen by Bush. Table 5. Transitional Probability (TP) as a predictor in word-boundary palatalization of final -IM or -/d/ and initial /j/- (cf. Bush 2001: 264) (Bush) Scenario 1 Cut-off TP > 0.0444

Scenario 2 Cut-off TP > 0.05

High-TP dyads

Low-TP dyads

High-TP dyads

Low-TP dyads

Palatalization [%]

97 [59.5%]

12 [32.4%]

92 [59.7%]

17 [37%]

No palatalization [%]

66 [40.5%]

25 [67.6%]

62 [40.3%]

29 [63%]

37

154

46

Sum [100%]

163 2

2

Chi-square value

χ =8.9

χ =7.4

Significance level

ρ 3

30O

^ tN Μ tN

tN tN

Γos

ox tN OS SO Γxp

Μ f). Ν

Ο

t-

S

t

^t „ tN OS Μ Ο tN CN

ox

a

ox ^ OS

so -if

•5 ox

"2 tN fN"

2

Ν=· 00

oo so

öS

^

O Tf

o

SO OS Os 1-1 Ο OS

f ox m Ό 1-1 Ο •>3· TT

tN Ρ a

1 c

g % ~.2 •3«

Έ

s

u. a -S 2P PL, [t, . S W 95 t: C/5

.. 13

u -α CL, Η

•5

•SP ε

tN ε £ BO Ο Ξ PP, by 1 word

by 2 - 4 words

by 5 - 6

by 7+

60% (58) 40% (38)

86% (108) 14% (17)

94% (31) 6% (2)

99% (68) 1%(1)

Consider an example of the relevant alternation:

178

John A. Hawkins

(3) a. The man

VP[waited

1

b. The man vp [waited 1

PP1

[for his son] 2 3 4

PP2

PP2

[in the cold but not unpleasant wind]] 5

[in the cold but not unpleasant wind] 2 3 4 5 6 7 8

PP i

[for his so«]] 9

PP2 consists of seven words here, PP] of three, making a difference of four words. The data of (3) show that there is an 86% preference for short before long (i.e. for [V PPi PP2]) in sequences such as this, rising to 94% for 5-6 word differentials, and to 99% for 7+ word differentials.

2.2. A description and possible explanation for the performance preferences One way of accounting for the preference for (3 a) is in terms of the efficiency with which syntactic structure can be parsed (Hawkins 1994, 1999a, 2001). The immediate constituents (ICs) of a phrase can typically be recognized on the basis of less than all the words dominated by that phrase, and some orderings reduce the number of words needed to recognize these ICs compared with others. The result is faster and more efficient constituent structure recognition. Thus, the three ICs, V, PPi and PP2 can be recognized on the basis of five words in (3 a), compared with nine in (3 b), assuming that head categories such as Ρ project to mother nodes like PP and enable the parser to immediately construct them when their heads are encountered in the on-line parse (cf. Hawkins 1994; Jackendoff 1977; Pollard and Sag 1994). Following the procedure of Hawkins (1994: chapter 3) the greater efficiency of (3a) can be captured by dividing the number of ICs by the number of words required to recognize those ICs within the VP Constituent Recognition Domain and expressing the result as a percentage. A Constituent Recognition Domain is defined in (4): (4)

Constituent Recognition Domain (CRD) The CRD for a phrasal mother node Μ consists of all non-terminal and terminal nodes dominated by Μ on the path from the terminal node that constructs the first IC on the left to the terminal node that constructs the last IC on the right.

Why are zero-marked phrases close to their heads?

179

The VP CRD in (3a) has an IC-to-word ratio of 3/5, or 60%. The same domain in (3b) has a ratio of 3/9, or 33%. The higher ratio of (3a) captures its greater efficiency. A principle of Early Immediate Constituents (EIC) is defined in Hawkins (1994: 69-83) as follows: (5)

Early Immediate Constituents (EIC) The human parser prefers linear orders that minimize CRDs (by maximizing their IC-to-word [or IC-to-nonIC] ratios), in proportion to the minimization difference between competing orders.

EIC predicts that the greater the preference for structure A over a competing A' (measured in terms of higher IC-to-word ratios, i.e. smaller CRDs), the more frequent and easy to process the A structure will be in performance, compared with any B/B' pair with less degree of preference. And preferred structures will be those that occur in general in performance. Both predictions are correct for the data of Table 1. A possible explanation for this EIC preference is that small CRDs reduce simultaneous processing demands in working memory. If, in (3a), the same constituency information can be derived from a five-word viewing window rather than from nine words, then phrase structure recognition can be accomplished sooner, there will be fewer additional (phonological, morphological, syntactic and semantic) decisions that need to be made simultaneously with this one, and less demands will be made on working memory. I have argued in Hawkins (1990, 1994, 1999a) that we should view working memory as efficiency-based rather than as capacityconstrained (as in models such as Just and Carpenter 1992). Many of the preferences to be discussed in the present context do not exceed proposed capacity limits in either their head-adjacent or their non-adjacent orders, and there are basic orderings in some languages that will regularly exceed proposed limits.

2.3. Adjacency preferences in grammars Grammatical conventions across languages reveal the same preference for small and efficient CRDs (cf. Hawkins 1990, 1994, 1998, 2001). The relative quantities of languages reflect the preferences, as do hierarchies of cooccurring grammatical properties. To cite just one example, in the Greenbergian (1966) word order correlations adjacency of lexical heads is

180

John A. Hawkins

massively preferred, as shown in the following data (from Dryer's 1992 language sample): (6) a.

c.

vp [V pp [P

NP]] = 161 (41.4%) IC-to-word: 2/2 = 100%

b. [[NP P]pp V ] w = 204 (52.4%) IC-to-word: 2/2 = 100%

ypfV [NP P]pp] = 18 (4.6%) IC-to-word: 2/4 = 50%

d. [pp[P NP] V] vp = 6 (1.5%) IC-to-word: 2/4 = 50%

Assume: V - 1 word; Ρ = 1; NP = 2 EIC-preferred (6a) + (b) = 365/389 (94%) Verb-initial order within VP correlates with prepositions in PP (6a) rather than with postpositions (6c), while verb-final order correlates with postpositions (6b) rather than prepositions (6d). The adjacency of V and Ρ in (6a) and (6b) guarantees the shortest possible domain for the recognition of the two ICs of VP (V and PP): two words suffice, hence 2/2 = 100% IC-toword ratios. In the non-adjacent domains of the (c) and (d) orders, ratios are significantly lower and exemplifying languages significantly fewer. The preferred (a) and (b) structures collectively account for 94% of all languages. It is patterns like these that have motivated the head-initial (VO) and head-final (OV) parameters in language typology. The two types are mirror images of one another and EIC offers an explanation for why both (a) and (b) should be highly productive: they are optimal and equally efficient strategies for phrase structure recognition, whereas the mixed types (c) and (d) are much less efficient and much less productive.

3. A general hypothesis for adjacency If relative weight preferences are to be accounted for in terms of reduced CRDs, then a clear generalization suggests itself that links these preferences with the processing domains for other syntactic and semantic relations. We can hypothesize that all syntactic relations, of combination or dependency, defined on two (or more) categories should prefer these categories to occur in the smallest possible surface structure domains that need to be accessed for the processing of these relations and dependencies. This is captured in (7):

Why are zero-marked phrases close to their heads?

(7)

181

Minimize Domains The human processor prefers to minimize the connected sequences of linguistic forms and their conventionally associated syntactic and semantic properties in which relations of combination and/or dependency are processed. The degree of this preference is proportional to the minimization difference between competing sequences.

A relation of combination is defined for present purposes as: two categories A and Β are in a relation of combination iff they occur within the same syntactic mother phrase or maximal projection (Jackendoff 1977; Chomsky 1981). Dependency is defined as: two categories A and Β are in a relation of dependency iff syntactic or semantic properties of Β require access to A for their assignment when this relation is parsed. Notice that this definition for combination is given in purely grammatical terms, whereas the definition for dependency is a processing one, in terms of the required access by the parser to one category while parsing another (see section 4 below for further details and Hawkins in preparation). I believe that this is the most adequate definition for dependency, and it enables me to capture the adjacency effects from performance and grammars that are discussed here. 3.1 Multiple adjacency preferences in performance Evidence for a general Domain Minimization principle along the lines of (7) comes from the same structures that motivate EIC. The overall preference for the adjacency of a phrase to the verb appears to be proportional to the number of combinatorial and/or dependency relations that are contracted between that phrase and the verb whose surface processing domains can be minimized. The Heavy Noun Phrase Shift conversion of [V NP PP] to [V PP NP], for example, is motivated not just by relative weight, but also by the existence of a lexical-semantic dependency between V and PP whereby the meaning of one cannot be processed without access to the other, e.g. take into account X. Wasow (1997) calls such examples opaque collocations and they had a 60% shifting ratio in his corpus, compared with only 15% shifting for non-collocational sequences like take to the library X. Hence when V-PP adjacency is preferred both syntactically (minimize phrase structure processing domains) and semantically (minimize domains of lexical-semantic interdependency), then there is significantly more shifting.

182

John A. Hawkins

Hawkins (1999a) examined the lexical-semantic dependencies between intransitive verbs and PPs in his corpus using various semantic measures and quantified the contributions of syntactic weight and of semantics to relative ordering. One test, the Verb Entailment Test, asked: does [V PPi PP2] entail V alone, or does V have a meaning dependent on either PPj or PP2. E.g., the man waitedfor his son in the rain entails the man waited, but the man counted on his son in old age does not entail the man counted. A second test, the Pro-Verb Entailment Test, asked: can V be replaced by some general Pro-verb, or does one of the PPs require that V for its interpretation. Thus, the girl played on the playground entails the girl did something on the playground, but the girl depended on her mother does not entail the girl did something on her mother. When there was a dependency (or occasionally a greater dependency) between V and one of the PPs by these tests, then 73% had the dependent PP (PPd) adjacent to V (155/211), i.e. their processing domains were minimal. Recall that 82% had a shorter PP before a longer PP in (3), i.e. minimal CRDs. For PPs that were both shorter and semantically dependent (PPid), the adjacency to V was almost perfect at 96% (103/107), which was (statistically) significantly higher than for each factor alone, with only 4% (4/107) being non-adjacent. Hence the more syntactic and semantic processing relations whose domains are minimized in a given order, the greater the preference appears to be for that order. The full data for this dependency by weight interaction are set out in (8): (8)

PPd > PPi by 5+

2^4

PPd = PP( ϊ

PPj > PPd by \

[VPP d PPj] 10% (3) 32% (6) 76% (19) 83% (24) 92% (24) [VPPiPP d ] 90% (28) 68% (13) 24% (6) 17% (5) 8% (2)

2^4 96% (49) 4% (2)

5+ 100% (30) 0% (0)

3.2. Multiple adjacency preferences in grammars Grammatical conventions reveal the same preferences for structures in proportion to the number of combinatorial and dependency relations whose surface domains are minimized, as I argued in Hawkins (2001). Consider the relative ordering of complements and adjuncts in relation to their heads. Complements prefer adjacency and are generated in a position adjacent to their heads in the basic orders of numerous phrases in English and other languages, cf. e.g. Jackendoff (1977) and Pollard and Sag (1987). Tomlin's

Why are zero-marked phrases close to their heads?

183

(1986) Verb Object Bonding principle provides further cross-linguistic support for this since he points to languages in which it is impossible or dispreferred for adjuncts to intervene between a verbal head and its subcategorized direct object complement. Why should complements prefer adjacency? The basic reason I would offer is that there are more combinatorial and dependency relations between heads and complements than between heads and adjuncts. For example, complements are listed in a lexical co-occurrence frame that is defined by, and activated in on-line processing by, a specific head, such as a verb (cf. Hawkins 1995; MacDonald et al. 1994). Adjuncts are not so listed and they occur in a wide variety of phrases with which they are semantically compatible (cf. Pollard and Sag 1994). Processing the lexical co-occurrence frame of a verb therefore favors a minimal domain linking the verb to its lexically specified complements or arguments. In addition there are more productive relations of semantic and syntactic interdependency between heads and complements than between heads and adjuncts. I.e. there are more cases in which the meaning or grammar of one of these categories requires access to the other for its assignment. Processing these interdependencies favors minimal domains for the same reason that verbs and PPs classified as opaque collocations by Wasow (1997) or as dependent by the entailment tests of Hawkins (1999a) (cf. section 3.1) prefer adjacency. A direct object NP, for example, receives a theta-role from a transitive verb, typically a subtype of Dowty's (1991) Proto-Patient theta-role, depending on the particular verb. Adjuncts, by contrast, do not receive theta-roles, cf. Grimshaw (1990). A direct object is also syntactically required by a transitive verb for grammaticality, whereas adjuncts are not syntactically required sisters. Conversely, transitive verbs regularly undergo function category range reduction (cf. Keenan 1979) by the direct object NP. Compare the different senses of run in combination with different objects: run the race; run the water.; run the advertisement. The verb is lexically dependent on the object for its precise interpretation in these examples, but it is not dependent on an accompanying adjunct such as in the morning (in e.g. run the race in the morning). Similarly intransitive verbs are frequently dependent on PP complements for their interpretation (as in count [on Λφ. Complements therefore involve the processing of more syntactic and semantic relations with the head than adjuncts and this results in multiple processing preferences for adjacency. Adjuncts do not receive theta-roles, are not grammatically required by their heads, and their heads are not

184

John A. Hawkins

lexically dependent on them. More processing domains can be shortened when heads and complements are adjacent.

3.3. Adjacency hypothesis The data considered so far, from performance and grammars, suggest the hypothesis in (9): (9)

Adjacency to Heads Given a phrase {Η, {Χ, Υ}}, Η a head category and X and Y phrases that are potentially adjacent to H, then the more combinatorial and dependency relations that link Η and X, and the greater the minimization preference in each processing domain when X is adjacent to H, the greater will be the preference for their adjacency over that of Η and Y.

When X contracts more relations with Η than Y does, there will be more processing operations whose domains can be minimized when Η and X are adjacent. Phrase structure recognition, lexical-semantic dependencies, theta-role assignments, etc. will be processable in smaller and more efficient domains that will involve less simultaneous processing and less demands on working memory. It is hypothesized here that additional processing operations will exert a cumulative effect on adjacency and that each preference will also apply according to its degree, reflecting the degree of relative weight for example. The adjacency hypothesis in (9) is in the spirit of Behaghel's (1932: 4) and Givon's (1991) proximity principle, which Behaghel defined as "das geistig eng Zusammengehörige [wird] auch eng zusammengestellt", i.e. what belongs together mentally is placed close together. Givon proposed that "Entities that are closer together functionally, conceptually, or cognitively will be placed close together at the code level, i.e. temporally or spatially". The formulation in (9) is an attempt to define this general idea within the context of an explicit set of grammatical and parsing assumptions (cf. Hawkins 1994, 1991, 2001). In the process it^extends the proximity preference to the processing of syntactic as well as semantic relations, and it makes predictions for variation data in which there are competing preferences for proximity, as in (8). It can also explain the data that motivate Rohdenburg's Law, to which I now return.

Why are zero-marked phrases close to their heads?

185

4. Dependency relations and dependent categories with and without zero Why are zero-marked phrases close to their heads? The general answer I wish to give is that zero-marked phrases are more dependent on their heads for the assignment of syntactic and semantic properties than explicitlymarked phrases, and they are also shorter by 1 word. Greater dependency plus EIC's short-before-long preference therefore result in a tighter adjacency to the head, for the same reason that semantically dependent and shorter PPs prefer adjacency to the verb in the data of Table 1 and (8). In order to justify this answer I need to consider the notion of dependency in more detail and the role of zero versus explicit marking in relation to it. 4.1. Property assignments in dependency relations Many syntactic properties can either be formally marked, through explicit words or morphemes, or they may not be marked, whereupon their existence has to be assigned through a dependency relation or inference in processing (see Hawkins in preparation). A complementizer or relativizer in English gives formal expression to the subordination property, zero complementizers and relativizers do not. A relative pronoun or resumptive pronoun marks co-indexation, a zero relativizer or gap does not. Agreement morphology on adjectives can make explicit the attachment to the relevant head noun, no agreement does not. Case marking on NPs can formally signal properties in a verb's argument structure by assigning NPs to their appropriate nominative, accusative or dative cases, and to the theta-roles that are associated with each. Zero marking in all these examples results in more minimal domains and in greater adjacency across languages, and this can be accounted for by saying that there are more dependencies between categories A and Β when certain syntactic properties are not explicitly marked. Assigning a given property to category Β now requires access to category A in parsing, instead of just consulting the more explicitly marked and more informative Β itself. For example, languages without adjective agreement, such as English, are dependent on a right-adjacent (or proximate) noun for NP construction and for attachment of the adjective to it. Languages with adjective agreement or with other similar formal devices can assign a dominating NP online, independently of the noun, by Agreement Projection (Hawkins 1994:

186

John A. Hawkins

366-373, in preparation), and they contain explicit information about syntactic attachment that is lacking in the zero-marked adjectives of English. An NP whose case morphology renders it unambiguously accusative or dative and whose case is recognizable without accessing the verb or other clues in the syntactic environment permits the relevant syntactic and semantic properties to be assigned independently of the verb. But a bare NP is crucially dependent on access to the verb for these same property assignments, hence there are more dependency relations between them, and hence tighter adjacency by (9). In the theory of dependency proposed in Hawkins (in preparation) and assumed here I take the perspective of a parser receiving terminal elements one by one in a parse string. When the parser receives the first two words of an English sentence, e.g. the boy, it can recognize the categories determiner + noun, it can attach them to a mother noun phrase (Hawkins 1993, 1994), it can assign lexical-semantic content to boy, and a uniqueness semantics to the (Hawkins 1978, 1991), but it cannot yet assign a case or a theta-role. If the immediately following word is the verb ran, the theta-role agent can be assigned to the boy, in addition to nominative case. If the next word is fell, a patient theta-role is assigned, in addition to nominative. If the parse sequence is the boy I like, then patient and accusative case are assigned. The NP the boy is therefore neutral or zero-specified with respect to case and theta-role in these examples and these properties can only be assigned to it by accessing other categories. I shall say that the boy "depends on" a following intransitive verb for case and theta-role assignment. Similarly the key depends on the following transitive VP opened the door for assignment of the instrument role and for nominative case (Fillmore 1968), and this tent depends on a following sleeps four in this tent sleeps four for locative and nominative assignments (Rohdenburg 1974; Hawkins 1986). These NPs are zero-specified for theta-role and case and these properties are assigned by a dependency relation on a following V or VP in the absence of explicit signalling in the NP itself. The pronouns he vs. him, on the other hand, are explicitly marked for case, but the nominative is still zero-specified for theta-role {he ran involves an agent, he fell a patient). Conversely, the verbs in these examples can be said to be dependent on the choice of NPs for selection of their appropriate lexical co-occurrence frame (Hawkins 1995, in preparation) and for selection of their appropriate meaning from the often large set of dictionary entries with respect to which a verb is ambiguous or polysemous. Run is syntactically ambiguous in English between intransitive and transitive uses (the boy ran/the boy ran

Why are zero-marked phrases close to their heads?

187

the race), and it is semantically ambiguous or polysemous between a whole range of interpretations, as I mentioned above, depending on the choice of subject for intransitive verbs (the boy ran/the water ran/the stocking ran/the advertisement ran) or of object for transitives (the boy ran the race/ran the water/ran the advertisement), cf. Keenan (1979). I shall say that run depends on the relevant NPs for selection of its syntactic co-occurrence frame and meaning from the total set that is listed in its lexical entry. The verb open likewise has several syntactic co-occurrence frames (John opened the door with the key/the key opened the door/the door opened, cf. Fillmore 1968) and several meanings as well, and it depends on its accompanying NPs and PPs for disambiguation and polysemy reduction. 4.2. Defining dependency I can now give a more detailed definition for dependency. (10) A word or phrase A depends on another Β iff the parsing of A requires access to Β when assigning syntactic or semantic properties to A with respect to which A is zero-specified or ambiguously or polysemously specified. A property that is assigned via a dependency relation will be called a dependency assignment. Such assignments are made within domains of terminal elements in surface structure and their associated syntactic and semantic properties. I shall understand a dependency domain to be the smallest connected sequence of forms and properties that are sufficient for the parsing of the dependency relation in question. The same notion of domain has been invoked for combinatorial relations as well (cf. the definition of a Constituent Recognition Domain in [4] above). Since combination was defined above as "two categories A and Β are in a relation of combination iff they occur within the same syntactic mother phrase or maximal projection" it will be clear that some relations between a given A and Β will involve both, and I shall refer to these as combinatorial dependencies. The verb-preposition sequence count on is an example. The verb count combines with the PP constructed by on and is also dependent on it for its precise meaning assignment (cf. section 3.1 and Hawkins 1999a). Dependencies that hold between categories that are not in the same mother phrase will not be combinatorial dependencies. Examples are coindexation dependencies between antecedent and anaphor in Johnt

188

John A. Hawkins

chastised himself ι and between filler and gap in the student, that I taught O, (Hawkins 1999b). This parsing approach to dependency captures the original intuition underlying dependency theories (cf. Tesniere 1959 and Hays 1964 for early proposals) in a way that is now more empirical and easier to test. We can examine individual words or phrases and ask whether a given property can be assigned to them based on their explicit forms and corresponding lexical entries and grammar, independently of their syntactic or semantic context, or whether a particular property assignment does depend crucially on some other word or phrase that must be accessed within a given domain, as in the examples I have given. The result is a rather different set of predictions from those that are made by purely grammatical approaches, and the success or otherwise of these predictions will provide evidence for or against this definition. For example, contrary to Tesniere (1959) dependency relations between a given A and Β can now be either symmetrical or asymmetrical, and this has profound consequences for the explanation of symmetries and asymmetries across languages (Hawkins in preparation). Dependencies can also be impacted by surface morpho-syntactic differences, such as the presence or absence of case marking. In the boy ran there is a nominative case dependency on the finite verb, but no such dependency in he ran since he is intrinsically nominative and the parser does not need to access the finite verb in order to assign it (even though this is a property listed in the lexical co-occurrence frame for run). Verb-final languages with explicit case marking, such as Japanese (Kuno 1973) and Kannada (Bhat 1991), are not dependent on a following verb for case assignment and for the associated theta-role assignments, by this logic, since the parser can assign these properties on-line prior to the verb in whose lexical co-occurrence frame they are actually listed. The verb, on the other hand, will be dependent on preceding case-marked NPs for selection of its appropriate syntactic cooccurrence frame and for semantic disambiguation and polysemy reduction. Languages with flexible syntactic categories and regular ambiguities even between noun and verb, such as the Polynesian language Tongan (Broschart 1997), make use of nominal and verbal particles and of other contextual clues for category disambiguation and I shall say that the ambiguous predicates are dependent on such particles for category assignment. A huge number of predicates in English are category-ambiguous and are dependent on (often immediately preceding) closed class words or other contextual indicators for category assignment (a run vs. to run; a play vs. to play·, etc). Languages with adjective agreement, as I have mentioned,

Why are zero-marked phrases close to their heads?

189

contain explicit information about sister attachment to a given noun and also about NP domination. Languages without such agreement are dependent on the noun itself for NP construction and attachment. More generally, languages with rich inflections (Latin) or agglutinative morphology (Turkish) will permit more processing and property assignments within words and phrases themselves, whereas isolating languages with impoverished morphologies (Chinese, Vietnamese, and to a considerable extent English) will involve more dependency assignments and they will exploit neighboring words to a greater extent. This has important consequences for adjacency. Before I pursue these consequences further, a brief word about the processing aspect of the definition of dependency given in (10). Notice that this definition is phrased in terms of parsing (i.e. comprehension), rather than production. There is a reason for this. The language producer uttering the boy fell can already assign a patient theta-role to the boy prior to the production of fell by accessing the lexical entry for this latter in the Formulator (Levelt 1989). And quite generally there are numerous forms and properties that the speaker has some inner representation of, in the Conceptualizer and the Formulator, prior to their articulation (Levelt 1989). But the hearer cannot assign the patient role to the boy without hearing the verb that determines the assignment. And the need for contextual clues in the assignment of all dependent properties is primarily a need for the hearer, to which, interestingly, speakers are highly sensitive when packaging their sentences for the hearer in performance. In fact, one can argue that property assignments in on-line sentence production mirror those in comprehension, resulting in many types of production data, such as those discussed in this paper, that are mostly beneficial for the hearer (cf. Hawkins 1998 for discussion). 5. Zero vs. formal marking of dependent categories in English 5.1. That/zero complementizers The complement clause with an explicit complementizer, in (1) above, which I repeat for convenience, is inherently recognizable as a sentential or S-dominated constituent, by virtue of its nominative subject he and its finite auxiliary had, both of which occur only in S-dominated clauses (see Hawkins 1994: 381-387 for discussion of the construction and parsing of sentential constituents by unique node constructors such as these):

190

John A. Hawkins

(1)

a. I realized [that he had done it] with sadness in my heart. b. I realized with sadness in my heart [that he had done it].

It is also recognizable as subordinate, by virtue of that (cf. Hawkins 1994: 387-393 on subordination indicators). Subordination can be defined for present purposes as: attach S to some non-S mother. This instruction is carried out in (1) by attaching s[that he had done it] to the matrix verb phrase. The zero-marked complement in (2), by contrast, he had done it, is inherently sentential by virtue of he and had, but it is vague or ambiguous with respect to main or subordinate clause status. (2)

a. I realized [0 he had done it] with sadness in my heart. b. ?I realized with sadness in my heart [0 he had done it].

Occurring in isolation he had done it is a well-formed main clause. In (2) it is a subordinate clause, attached to a non-S, namely the matrix VP. Subordination is now recognized not through an explicit signal within the clause itself, but by accessing the verb realized and by linking the clause to the sentential complement option within its lexical co-occurrence frame, thereby recognizing subordination rather than main clause status. The parsing of he had done it is therefore dependent on the verb for assignment of the subordination property. The verb in turn is dependent on the subject / and on the clausal complement for selection of the intended lexical cooccurrence from among other possibilities. These include e.g. I realized this fact, or NP-V-NP. Both the explicit and the zero-marked clauses require access to the verb for assignment of the additional property of direct objecthood (or accusative case) within the lexical co-occurrence frame for realize. There is no case marker or other indicator inherent in (that) he had done it to signal its (sentential) subject or object status. The verb is also just as dependent on the explicitly marked as on the zero-marked clause for selection of this cooccurrence item, as opposed to a noun phrase such as this fact. But the overall number of dependent versus independent properties assigned by reference to the head is greater for the zero forms. There are more dependency relations that have their surface processing domains minimized, therefore, when zero-marked complements are close to their heads. The following data (journalistic British English) are from Rohdenburg (1999: 102):

Why are zero-marked phrases close to their heads?

191

(11) a. Finite clause complements adjacent to V {realize) THAT= 63% (294) ZERO = 37% (172) b. Finite clause complements with intervening material between V and complement THAT= 97% (62) ZERO = 3% (2) Only 2/174 (1.1%) of the more dependent and shorter zero-marked complements are non-adjacent to the verb realize in these data and the remaining 98.9% are adjacent. This distribution is in accordance with the adjacency prediction of (9). The data here are also similar to those involving semantic dependencies and weight given in (8). Only 3.7% (4/107) of the semantically (more) dependent and shorter PPs were non-adjacent to V in these data. Rohdenburg (1999) gives some further figures that can be explained in terms of parsing efficiency. The majority of the zero-marked clauses in his data have subjects that are personal pronouns (127/174 = 73%), as in (2) above, whereas a minority of the /Aa/-clauses do (137/356 = 38%). Most personal pronouns in English are inherently case-marked for nominativeness and hence a pronoun such as he in (2a) can immediately construct a mother S over the pronominal NP by Grandmother Node Construction (Hawkins 1994: 361). This renders the onset of a new clause immediately recognizable, whereas a full NP is not inherently marked for case, and explicit casemarking therefore compensates for the zero marking of subordination and avoids the misassignment of many full NPs to the matrix VP as direct objects on-line {I realized this fact...). Pronouns are also short single-word items and even those that are not case-marked (e.g. you and it) will involve a minimal processing delay before the next unambiguous clause constructor is encountered, namely the finite verb (or auxiliary), cf. Hawkins (1994: 381-387). This approach therefore predicts the following preference hierarchy for subjects in leftmost position within zero-marked subordinate clauses such as (2): (12) Leftmost subject preference hierarchy for zero-marked subordination Case-marked pronoun > non-case-marked pronoun > short full NP > longer full NP This hierarchy should hold in adjacent environments and especially in any non-adjacent ones. It predicts that this distribution of leftmost elements

192

John A. Hawkins

should decline down the hierarchy in zero clauses and conversely that it should increase in corresponding i/wrf-clauses. Rohdenburg does not give figures for these items in non-adjacent versus adjacent environments (there are only 2 non-adjacent zero-clauses anyway), and he does not distinguish case-marked from non-case-marked pronouns. But he does distinguish personal pronouns (all of which are case-marked except for you and it) from full NPs, and within the latter he distinguishes those that are 1-2 words in length from those that are 3+ words in length. The relative distribution of items in (12) between zero and ίΑαί-clauses is as predicted: Table 2. Finite complements dependent on the verb realize zero subordinate clauses

that subordinate clauses

Personal pronoun subjects

48% (127)

52% (137)

1-2 word full NP subjects

26% (32)

74% (89)

3+ word full NP subjects

10% (15)

90% (130)

The 73% preference for personal pronoun subjects within the zero subordinate clauses (127/174), mentioned above, is therefore a preference for what are predominantly case-marked NPs. 5.2. WH, that/zero relativizers Zero-marked relative clauses in English, e.g. the Danes you taught, involve a clause (you taught) that requires more dependent processing on the head noun for property assignments than the explicitly marked who(m) you taught or that you taught. The zero clause is either ambiguous or incomplete and the noun must be accessed by the parser for recognition of subordination status, whereas subordination is signalled explicitly by who(m) and that. The head noun must also be accessed for a second reason in the zero relative, involving the parsing of the verb and its co-occurrence requirements. The transitive taught takes a subject, here you, and a direct object. The direct object is the position relativized on in this clause, and in the absence of an explicit relative pronoun or a resumptive pronoun (as in Hebrew structures such as the Danes that you taught them, cf. Hawkins 1999b), the parser must simply access the head Danes in order to assign the appropriate direct object to the verb teach. The verb is therefore dependent on the head noun for the parsing of its syntactic co-occurrence frame, in the theory proposed here (recall [10]). In structures that contain explicit pronominal elements, whether regular pronouns as in Hebrew or relative

Why are zero-marked phrases close to their heads?

193

pronouns as in English, this dependency can be realized within the relative clause itself, since both subject and object are contained within this clause, and the relative clause is therefore an independent processing unit for these verb co-occurrence dependencies. The pronouns provide local arguments and the clause-external head does not need to be accessed, for these dependencies at least. There are other dependencies between the relative clause and head noun that must be parsed, however, even when relative and resumptive pronouns are present. The referential index on the head Danest needs to be copied onto the bound pronoun (the Danes, that you taught them,) or relative pronoun (the Danes,· whom, you taught) or gap (the Danes, whom, you taught Oj) or subcategorizor (the Danes, that you taught,), depending on one's theory of how co-referential indexing works (cf. Chomsky 1981; Pollard and Sag 1994; Hawkins 1999b). The only important point here is that there is a form of identity between the head noun, external to the relative clause, and the position relativized on within the relative clause, and this identity can only be captured by accessing the head, regardless of the type of relative clause used and its degree of explicitness. A restrictive relative clause is also a type of adjunct to the head and its precise interpretation will vary with, and be dependent on, the head, again regardless of explicitness. For example, the clause that was flat is dependent on the head noun for its precise interpretation (compare the road that was flat with the beer that was flat). The clause that John ran is similarly dependent (compare the race that John ran with the water that John ran), cf. Keenan (1979). In addition, the head noun Danes undergoes a potential restriction in its referential range when a restrictive relative follows (but not when an appositive one does), and depends on the relative for assignment of this range (i.e. the Danes in question were the ones that you taught, not the ones that anyone else taught). Head and restrictive relative are therefore symmetrically dependent on one another with regard to these properties, and they also combine with one another within a common NP. But there are more dependencies linking head and zero relative than head and explicitly marked relative and I therefore predict tighter adjacency for zero (by [9]). The following data (spoken Standard British English) are from Quirk (1957) and involve restrictive relative clauses relativizing on non-subject positions (relativizations on subjects do not allow the zero option except in rare environments such as there was a man came to see you yesterday). Adjacency is favored over non-adjacency overall by a significant margin (548 to 62). This can be explained in terms of the number of combinatorial and dependency relations that link the relative

194

John A. Hawkins

clause and its head, even when there is explicit marking. What is significant is the distribution of WH and that to zero in the non-adjacent environments of (13b): (13) a. Restrictive (non-subject) relatives adjacent to the head (n = 548) WH = 28% (152) THAT= 32% (174) ZERO = 41% (222) b. Restrictive (non-subject) relatives with intervening material (n = 62) WH = 50% (31) THAT= 44% (27) ZERO = 6% (4) Only 4/222 (1.8%) of the more dependent and shorter zero relatives are non-adjacent to their head nouns. The remaining 98.2% prefer adjacency. (A very similar distribution for explicit and zero relative pronouns in adjacent and non-adjacent environments has been confirmed in the more recent quantitative study of Guy and Bayley [1995]). These non-adjacent relatives could be either NP-internal (with e.g. a PP intervening between the head noun and relative clause) or NP-external as a result of Extraposition from NP (as in the Danes came to stay (that) you taught), and Quirk's coding of the data does not distinguish between them. Lohse (2000) has accordingly examined the different non-adjacency types and has quantified the distribution of explicit which and that versus zero, using the Brown corpus. NP-external extraposition generally involves a greater distance between relative clause and head than NP-internal nonadjacency, and hence if there are more dependency relations whose processing domains can be minimized with zero, then the distribution of zero should be significantly higher in NP-internal non-adjacent environments than in NP-external ones, while the distribution of which/that should be higher in NP-external non-adjacent environments than in NP-internal ones. Lohse's data, given in (14), show that this is the case: (14) a. Non-adjacent NP-internal relatives WHICH/THA Τ = 72% (142) ZERO = 28% (54) b. Non-adjacent NP-external relatives WHICH/THAT =94% (17) ZERO = 6% (1) Non-adjacent relatives with zero have a 28% distribution in NP-internal position compared to 72% for explicit which/that, and only a 6% distribution compared to 94% for which/that in NP-external position. In addition, only 1/55 (1.8%) examples have the more dependent and shorter zeromarked relative in NP-external position at all.

Why are zero-marked phrases close to their heads?

195

5.3. Finite S complement/infinitival to- VP/infinitival zero-VP Rohdenburg (1999) gives data on the alternation between semantically equivalent finite and infinitival complements of the verb promise, and between infinitival complements of the verb help with and without infinitival to. These data enable us to test the adjacency prediction of (9) with respect to different degrees of (in)explicitness in clausal complements. When the subject of the finite complement of promise is identical to the matrix subject, as in You promised that you would go to the doctor, then English allows an infinitival complement instead, You promised to go to the doctor. In early work in generative grammar the infinitival was derived from the finite complement by Equi-NP Deletion or Identity Erasure (Rosenbaum 1967). More recently the infinitival is analyzed as a type of control structure (Radford 1997). In terms of the dependency theory proposed here (cf. [10]) the parsing of the infinitival verb to go must access the matrix subject you in order to assign a subject to it and thereby satisfy the lexically specified syntactic co-occurrence requirements of this verb. The infinitival complement therefore involves a strong dependency on the matrix subject and this dependency is processed within a connected domain of surface elements and associated properties that includes the matrix verb promised, i.e. you promised to go (cf. section 4.2). The finite complement involves no such dependency since the co-occurrence requirements of the embedded verb are satisfied within the finite clause itself and the parser does not need to look outside this clause in order to assign the subject to it. The verb promise also permits a second NP argument, a personal object in Rohdenburg's (1999) terms, as in You promised your mother that you would go to the doctor and You promised your mother to go to the doctor. The parsing of the understood subject in the infinitival complement here still involves accessing the matrix subject and promise, and this (commissive) class of verbs differs from verbs like persuade whose matrix objects are parsed as understood subjects of the infinitival verb, as in You persuaded your mother to go to the doctor. The parsing domain for these subject assignment dependencies to the infinitival verb must therefore access both the matrix verb and the controlling argument, within the smallest possible sequence of connected surface elements that are sufficient for this assignment. The matrix verbs, on the other hand, are dependent on their co-occurring NPs and finite or infinitival complements for selection of the intended co-occurrence frame from among the various options listed in their lexical entries.

196

John A. Hawkins

Since there are more dependency relations for the less explicit infinitival complements than for finite complements there will be a greater preference for domain minimization, according to (9), and tighter adjacency is predicted between infinitival complement and matrix verb. If no other constituent intervenes, the parsing domain for subordinate subject assignment can be minimal, and the infinitival should be highly preferred. If an adverbial phrase or clause intervenes (You promised if necessary that .../to ...) the parsing domain will no longer be minimal and we expect a higher proportion of finite complements, whose parsing involves fewer dependencies and hence less pressure for adjacency. When a personal object intervenes (You promised your mother that .../to ...) parsing domains for subject assignment will again not be minimal, more finite complements are predicted, and there may be an additional functional pressure for finiteness resulting from the fact that the closest argument in the matrix for subject assignment (your mother) is not the understood subject here, whereas for the majority of these surface structures (the persuade type) the closest argument is the understood subject. Rohdenburg (1999: 106) gives the following data (written journalistic American English) that confirm these predictions: (15) a. Complements adjacent to V (promise) Infinitival VP = 94% (490) Finite sentence = 6% (29) b. Complements non-adjacent to V (intervening adverbial phrase or clause) Infinitival VP = 73% (30) Finite sentence = 27% (11) c. Complements non-adjacent to V (intervening personal object of promise) Infinitival VP = 15% (4) Finite sentence = 85% (22) The presence of an intervening adverbial increases the proportion of finite complements from 6% to 27%, while the presence of an intervening and non-controlling NP object increases it even more, to 85%. Consider now the alternation between infinitival to and zero with the matrix verb help in English, as in They helped (us) (to) finish the paper. Infinitival to functions like a complementizer to make the VP inherently subordinate, which we can define as: attach VP to some non-S mother. A main clause VP, by contrast, is immediately attached to an S, as in You (will) finish the paper. When to is omitted (They helped [wä] finish the paper) there is more dependent processing on the verb for subordination

Why are zero-marked phrases close to their heads?

197

assignment, just as there is when the complementizer that is omitted from finite complements (see section 5.1). The VP complement is also shorter without to. Hence tighter adjacency to the matrix verb help is predicted for zero infinitival complements. The following data (journalistic British English) from Rohdenburg (1999: 104) confirm this: (16) a. Infinitival VP complements adjacent to help (± object of help) TO = 45% (822) ZERO = 55% (1001) b. Infinitival VP complements with intervening negation (not) or -ly adverbs or complex objects between help and infinitival VP TO = 87% (176) ZERO =13% (26) Only 26/1027 (2.5%) of the more dependent and shorter zero-marked complements are non-adjacent to V. 5.4. Other alternations There are numerous examples of this sort involving zero-marked and explicitly-marked counterparts in English and Rohdenburg and his collaborators have shown conclusively that the zero forms prefer less complex and smaller environments. It remains to be seen whether all his examples can be subsumed under the theory presented here, in terms of minimal domains for the processing of combinatorial and dependency relations (cf. [7]) and the quantitative adjacency prediction derived from it (cf. [9]). Testing it will require detailed analysis of the syntactic and semantic properties of each construction, and the precise manner of, and domains for, their assignment in parsing. One further example that lends itself well to the analysis proposed here involves the variable use of prepositions as in The police stopped the fans (from) entering the grounds. The preposition from heads a PP complement in the lexical co-occurrence frame for the verb stop. Its absence leaves the gerundial phrase entering the grounds highly ambiguous as to its precise role in the structure of the sentence. By accessing the verb stopped the parser can recognize the gerundial phrase as a (reduced) ^ram-complement of this verb, and assign it this syntactic property and an associated semantics. The explicit from entering the grounds can simply be matched with a ^row-complement option activated by the verb and is less dependent on the verb for property assignments. The verb will be dependent in either case on co-occurring phrases for selection of the intended lexical co-occurrence

198

John A. Hawkins

frame from among the options listed. Once again there are more dependencies between zero and the verb than between ^ram-complements and the verb, and tighter adjacency is predicted for the former. In this case we can predict that the longer the direct object X is {the police stopped X [from] ...), the more explicit ^/rom-complements we will find, with fewer dependencies requiring access to an increasingly distant verb. Rohdenburg (1999: 107) gives the following data (journalistic British English) that support this prediction: Table 3. Gerundial complements associated with the verb stop zero complement of stop

yrom-complement of stop

Personal pronoun object

83% (85)

17% (17)

Other NP objects

72% (159)

28% (61)

Complex NP objects modified by ve/io-relatives

40% (12)

60% (18)

As the direct object size increases from single-word personal pronouns to other NPs to complex NP objects containing wAo-relatives the incidence of zero declines and explicit from increases. 5.5. Zero versus formal marking in grammars: NP-agreement Across languages grammatical rules regularly conventionalize the preferences of performance that we have seen exemplified in English. Zeromarked dependencies may require adjacency to their heads, with nonadjacency being ungrammatical. And grammars may permit only one possibility, zero or explicit marking, in adjacent domains, thereby conventionalizing one of the preferred options of performance. Consider the following universal from Moravcsik (1995: 471) for case copying, an agreement device that is common in a number of languages: (17) If agreement thru case copying applies to NP constituents that are adjacent, it applies to those that are non-adjacent. Compare the following data from two Australian languages, Kalkatungu (a word-marking language, Blake 1987), and Warlpiri (a phrase-marking language, Blake 1987):

Why are zero-marked phrases close to their heads?

(18) a. thuku-yu dog-ERG b. thuku-yu dog-ERG

yaun-tu big-ERG yanyi white-man

yanyi white-man itya-mi bite-FUT

199

itya-mi (Kalkatungu) bite-FUT yaun-tu big-ERG

(19) a. tyarntu wiri-ngki+tyu yarlki-rnu (Warlpiri) dog big-ERG+me bite-PAST b. tyarntu-ngku+tyu yarlku-rnu wiri-ngki dog-ERG+me bite-PAST big-ERG Zero case copying has been conventionalized in Warlpiri when Ν and Adj are adjacent in (19a), but explicit marking must be used with non-adjacency in (19b). Kalkatungu has conventionalized explicit marking in both (18a) and (18b). Hence Kalkatungu and Warlpiri have each fixed one of the two options that we have seen in the relative clause data of English when the adjunct is adjacent to the head noun (cf. [13]), Kalkatungu in favor of explicit marking, Warlpiri in favor of zero. And both languages have fixed the same (explicit marking) option in non-adjacency environments that is almost exclusively preferred in English performance. Grammatical rules have conventionalized the preferences of performance here, eliminating the dispreferred option, while selecting from the options that are highly preferred. 6. Conclusions I conclude, first, that phrases are adjacent or close to their heads in proportion to the number of combinatorial and/or dependency relations whose processing domains can be minimized in an adjacent or proximate order versus a non-adjacent or distant one, and in proportion to the strength of each minimization preference (cf. [7] and [9]). Second, zero-marked phrases involve more dependencies on their heads, explicitly marked phrases more independent processability. Zero forms are also shorter. Hence their tighter adjacency is predicted. The primary descriptive generalization I propose is that zero-marked phrases are adjacent to their heads for the same reason ultimately that semantically dependent and shorter PPs are adjacent to their verbs (cf. [8]). In both cases there are multiple preferences for adjacency, and only very few of the relevant data occur in a non-adjacent order (cf. [8], [lib], [13b], [15b,c], and [16b]). Third, this adjacency prediction holds both for performance and for grammars, and for English as well as other languages. Performance data were summarized in sections 2.1, 3.1, 5.1, 5.2, 5.3 and 5.4, and corre-

200

John A. Hawkins

sponding grammatical data in sections 2.2, 3.2, 5.5, and in sections 4.1 and 4.2. Fourth, this approach leads to a new way of exploring the ultimate causality of all these facts, and this exploration will require interdisciplinary collaboration. One idea that is often appealed to when motivating explicit forms in non-adjacent positions is that such forms provide a clearer link between dependent and head and they therefore avoid ambiguous attachments in parsing. There is some truth in this, but it is not general enough. Explicit forms are systematically preferred in non-adjacent positions even when there is no attachment ambiguity, as in the Danes came to stay who(m) you taught (in which who(m) you taught can only be attached to the subject Danes). Explicit forms also occur productively in adjacent positions, in which attachments are regularly unambiguous (e.g. the Danes who(m) you taught came to stay), cf. (13a). And zero forms are extremely rare in non-adjacent positions that would be unambiguous (Ithe Danes came to stay you taught). Instead the theory proposed here links the degree of adjacency to the number of combinatorial and dependency relations to be processed and to the preference for minimal domains in each case. The explicit forms involve more independent processability. That is why they can occur further from their heads. There are fewer interdependencies that the parser has to access over a larger domain and in the process the greater explicitness may or may not resolve attachment ambiguities. The zero forms involve more dependent property assignments in parsing and these exert a collectively stronger pressure for adjacency or proximity, which is why they occur closer to their heads. Dependency assignments with zero actually result in a larger domain for parsing the relevant properties than if these properties were assigned within a single word or morpheme in a more explicit form. An adjective without agreement marking needs to access a noun for NP construction and attachment, an adjective with explicit agreement can assign these properties within the adjective itself, and so on. Zero does, of course, have a compensating advantage: there is less linguistic form to process phonologically, morphologically, syntactically and semantically. Explicit marking involves fewer property assignments through dependent processing, but there are more linguistic forms to process and assign properties to. Within small surface domains the data of performance and grammars indicate that language users have a choice: between less form processing (zero) but more dependent processing on the one hand, and more form processing (explicit marking) with less dependent processing on the other. The respective loads of form processing versus dependent processing appear to balance each other

Why are zero-marked phrases close to their heads?

201

out under conditions of adjacency, and both options are productive. But in increasingly non-adjacent domains, dependent processing is increasingly dispreferred and the balance shifts towards explicitly marked forms with greater independent processability. One can speculate that the working memory demands of dependent processing across large domains exceed the processing cost of additional form (and meaning) processing through explicit marking. As domain sizes increase in the processing of the relevant relation, property assignments across distant categories would have to be made simultaneously with an expanding number of phonological, morphological, syntactic and semantic processing decisions for all the items in the domain. The efficiency of dependent processing declines accordingly and the independent processability of more explicit and grammatically richer categories becomes preferred. This and many other issues raised in this paper need to be investigated further, and I invite others to join me in doing so. Note 1. The corpus for the Heavy Noun Phrase Shift data in Hawkins (1994) consisted of 200 pages of written English (150 pages fiction, 50 pages non-fiction), see Hawkins (1994 : 129, 183).

References Behaghel, Otto 1932 Deutsche Syntax: Eine geschichtliche Darstellung. Wortstellung. Periodenbau. Heidelberg: Carl Winter. Bhat, D.N. Shankara 1991 Grammatical Relations. London: Routledge. Blake, Barry 1987 Australian Aboriginal Grammar. London: Routledge. Broschart, Jürgen 1997 Why Tongan does it differently: Categorial distinctions in a language without nouns and verbs. Linguistic Typology 1: 123-165. Chomsky, Noam 1981 Lectures on Government and Binding. Dordrecht: Foris Publications. Dowty, David R. 1991 Thematic proto-roles and argument selection. Language 67: 547-619. Dryer, Matthew S. 1992 The Greenbergian word order correlations. Language 68: 81-138. Fillmore, Charles J. 1968 The case for case. In: Emmon Bach and Robert T. Harms (eds.), Universals of Linguistic Theory, 1-88. New York: Holt, Rinehart and Winston.

202

John Λ. Hawkins

Givon, Talmy 1991 Isomorphism in the grammatical code: Cognitive and biological considerations. Studies in Language 15: 85-114. Greenberg, Joseph H. 1966 Some universals of grammar with particular reference to the order of meaningful elements. In: Joseph H. Greenberg (ed.), Universals of Language, 72-113. Cambridge, MA.: MIT Press. Grimshaw, Jane 1990 Argument Structure. Cambridge, MA.: MIT Press. Guy, Gregory and Robert Bayley 1995 Relative pronouns in English. American Speech 70: 148-162. Hawkins, John A. 1978 Deflniteness and Indefiniteness: A Study in Reference and Grammaticality Prediction. London: Croom Helm and Humanities Press, Atlantic Highlands, New Jersey. 1986 A Comparative Typology of English and German: Unifying the Contrasts. London: Routledge. 1990 A parsing theory of word order universals. Linguistic Inquiry 21: 223-261. 1991 On (in)definite articles: Implicatures and (un)grammaticality prediction. Journal of Linguistics 27: 405—442. 1993 Heads, parsing, and word order universals. In: Greville G. Corbett, Norman M. Fraser and Scott McGlashan (eds.), Heads in Grammatical Theory, 2 3 1 265. Cambridge: Cambridge University Press. 1994 A Performance Theory of Order and Constituency. Cambridge: Cambridge University Press. 1995 Argument-predicate structure in grammar and performance: A comparison of English and German. In: Irmengard Rauch and Gerald F. Carr (eds.), New Insights in Germanic Linguistics, Volume 1, 127-144. Berlin: de Gruyter. 1998 Some issues in a performance theory of word order. In: Anna Siewierska (ed.), Constituent Order in the Languages of Europe, 729-781. Berlin: de Gruyter. 1999a The relative order of prepositional phrases in English: Going beyond Manner-Place-Time. Language Variation and Change 11: 231-266. 1999b Processing complexity and filler-gap dependencies across grammars. Language 75: 244-285. 2001 Why are categories adjacent? Journal of Linguistics 37: 1-35. in prep. Efficiency and Complexity in Grammars. Oxford: Oxford University Press. Hays, David G. 1964 Dependency theory: a formalism and some observations. Language 40: 511525. Jackendoff, Ray 1977 X-bar Syntax: A Study of Phrase Structure. Cambridge, MA.: MIT Press. Just, Marcel Adam and Patricia A. Carpenter 1992 A capacity theory of comprehension: Individual differences in working memory. Psychological Review 99: 122-149. Keenan, Edward L. 1979 On surface form and logical form. Studies in the Linguistic Sciences 8.2, Dept of Linguistics, University of Illinois. Reprinted in: Edward L. Keenan (1987) Universal Grammar: 15 Essays, 375—428. London: Routledge.

Why are zero-marked phrases close to their heads?

203

Kuno, Susumo 1973 The Structure of the Japanese Language. Cambridge, MA.: MIT Press. Levelt, Willem J.M. 1989 Speaking. Cambridge, MA.: MIT Press. Lohse, Barbara 2000 Zero versus explicit marking in relative clauses. MS, Dept of Linguistics, University of Southern California. MacDonald, Maryellen C., Neal J. Pearlmutter and Mark S. Seidenberg 1994 The lexical nature of syntactic ambiguity resolution. Psychological Review 101:676-703. Moravcsik, Edith A. 1995 Summing up Suffixaufnahme. In: Frans Plank (ed.), Double Case: Agreement by Suffixaufnahme, 451-484. Oxford: Oxford University Press. Pollard, Carl and Ivan A. Sag 1987 Information-based Syntax and Semantics, Volume 1, Fundamentals. (Center for the Study of Language and Information Lecture Notes 13.) Stanford: Center for the Study of Language and Information. 1994 Head-driven Phrase Structure Grammar. Chicago: University of Chicago Press. Quirk, Randolph 1957 Relative clauses in educated spoken English. English Studies 38: 97-109. Radford, Andrew 1997 Syntactic Theory and the Structure of English. Cambridge: Cambridge University Press. Rohdenburg, Günter 1974 Sekundäre Subjektivierungen im Englischen und Deutschen: Vergleichende Untersuchungen zur Verb- und Adjektivsyntax. Bielefeld: CornelsenVelhagen und Klasing. 1996 Cognitive complexity and grammatical explicitness in English. Cognitive Linguistics!·. 149-182. 1998 Clarifying structural relationships in cases of increased complexity in English. In: Rainer Schulze (ed.), Making Meaningful Choices in English: On Dimensions, Perspectives, Methodology and Evidence, 189-205. Tübingen: Gunter Narr. 1999 Clausal complementation and cognitive complexity in English. In: FritzWilhelm Neumann and Sabine Schülting (eds.), Anglistentag 1998, Erfurt, 101-112. Trier: Wissenschaftlicher Verlag. 2000 The complexity principle as a factor determining grammatical variation and change in English. In: Ingo Plag and Klaus Peter Schneider (eds.), Language Use, Language Acquisition and Language History, 25—44. Trier: Wissenschaftlicher Verlag. Rosenbaum, Peter S. 1967 The Grammar of English Predicate Complement Constructions. Cambridge, MA.: MIT Press. Stallings, Lynne M. 1998 Evaluating heaviness: Relative weight in the spoken production of HeavyNP Shift. Ph.D. dissertation, University of Southern California. Tesniere, Lucien 1959 Elements de Syntaxe Structural. Paris: Klincksieck.

204

John A. Hawkins

Tomlin, Russell S. 1986 Basic Word Order: Functional Principles. London: Croom Helm. Wasow, Thomas 1997 Remarks on grammatical weight. Language Variation and Change 9: 8 1 105.

Cognitive complexity and horror aequi as factors determining the use of interrogative clause linkers in English* Günter Rohdenburg

Abstract This paper surveys the evolution and present behaviour of dependent interrogative clauses in English, focusing on the way these are joined to their governing expressions. The analysis distinguishes between zero links, prepositional and verbal links as shown in (i)-(iii). (i) (ii) (iii)

She was at a loss 0 what to do/?what could be done. She was at a loss about/as to what to do/what could be done. She was at a loss to know what to do/what could be done.

The variants concerned are approached from the perspectives provided by the complexity principle and the horror aequi principle. The complexity principle states that in the case of more or less explicit constructional alternatives as in (i) - (iii), the more explicit option(s) will tend to be preferred in cognitively more complex environments. The principle also predicts that more explicit novel constructions are first established in more complex environments and that more explicit recessive (or obsolescent) constructions survive longer in exactly the same types of context. The horror aequi principle involves the widespread (and presumably universal) tendency to avoid the repetition of identical and adjacent grammatical elements or structures. Accordingly, verbal links containing ίο-infinitives are dispreferred in those cases where the (immediately preceding) governing or superordinate expression is represented by a marked infinitive itself. From an evolutionary perspective, such contexts may constitute functional niches in which obsolescent constructions can survive longer and where incipient constructions become established earlier than elsewhere.

1. A survey of relevant changes 1.1. Major changes in the complementation system Over the last few centuries, the grammar of English complement clauses has experienced a massive restructuring process, resulting in a much closer fit between the relevant forms and functions.1 On the one hand, there are a variety of developments which appear to be motivated primarily in semantic or lexical terms. They include the following:

206

Günter Rohdenburg

-

Certain classes of verbs (e.g. non-implicative ones like avoid or verbs of beginning) have established the gerund as an exclusive or regular sentential complement structure (Fanego 1996a, 1996b; Rudanko 2000; Mair 2002, this volume; Rohdenburg and Schlüter 2000: 462-465). - At the same time, both the /o-infinitive and the (finite) that-clause have drastically contracted their ranges of application (Rohdenburg 1992, 1995a, 1995b, 1996, 1998a, 1999, 2000). On the other hand, English has undergone a series of syntactically motivated regularization processes in its complement system. With numerous verbs, adjectives and nouns selecting infinitival (or finite) complements, the presence of entrenched prepositional objects has given rise to increasingly well established prepositional gerunds, which - in the end - may have ousted the earlier forms of sentential complementation (Rudanko 1998, 2000; Rohdenburg 1995b: 384-385, 1998a).2 This is true, for instance, of the items listed in (1), which - apart from prepositional objects - only select corresponding prepositional gerunds today. (1)

a. verbs: despair (of/about), delight (in), persist (in), thank s.o. (for), blame s.o. (for), praise s.o. (for) b. adjectives: instrumental (in), (un)used (to) c. nouns: (have) the virtue (of), (take) delight (in)

Notice that the prepositions in question generally represent the type characterized by Rauh (1996, 2002) as governed or case prepositions, whose function is simply to mark the complement of a lexical head. Obviously, the changes concerned have extended the complement marking function of prepositions to the area of sentential complementation. Yet though such governed prepositions have become grammaticalized to varying degrees, they still preserve a variety of subtle semantic contrasts. In addition, any existing contrasts between direct objects and prepositional complements may be reflected in the parallel contrast between directly linked gerunds (as in the case of avoid) and prepositional gerunds. This is why the replacement of the /o-infmitive by the prepositional gerund may be regarded as having produced a noticeably more explicit type of complementation structure. A related kind of change, which has been generally overlooked up to now (but see Rohdenburg 1998a, 1999: 108-109, 2002a: 86-94), concerns the way indirect interrogative clauses may be joined to nouns, verbs, adjectives and various phrasal expressions on which they are dependent in

Cognitive complexity and horror aequi

207

terms of selectional restrictions. In Middle English and the earliest stages of Early Modern English, the situation was not much different from that in present-day German. It was possible to join almost any interrogative to a large, varied and presumably (still) productive class of governing expressions, and that without resorting to the use of any prepositions or other linking elements. This earlier state of affairs is illustrated in examples (2)(5). (2)

...; he soberly ask'd our advice 0 what he should do in this intricacy. (Richard Head/Francis Kirkman, The English Rogue, Part 4, 1671)

(3)

.../ these misfortunes greatly stird the resolution of the King of Spaine, so that ignorant in himself 0 what to doe,... (J.S., Clidamas, 1639)

(4)

..., he called his counsel together to heare their advice, 0 what course he should take in this extremitie; ... (J.S., Clidamas, 1639)

(5)

.../ let us now consult 0 what is necessary to be done for our distressed friend; ... (Richard Head/Francis Kirkman, The English Rogue, Part 2, 1668)

In the meantime, English has established more and more governed prepositions in such cases, and with most nouns, adjectives and verbs including advice, ignorant and consult as in (2)-(5), the use of prepositional links has by now become the statistical norm or even obligatory. Tables 1 and 2 chart these developments in the quotations of the Oxford English Dictionary (OED) for a set of semantically related nouns as well as a representative number of adjectives.3/4/5 Table 1. The use of prepositions introducing interrogative clauses dependent on a set of related nouns containing account, description, narration, relation, report, story and tale in the quotations of the OED a) I

0 1 1450-1699 2 1700-1899 3 1900-

15 (8/7) 9

(1/8)

11

III

prepositions

total

9 (9/0)

24 (17/9)

12 (8/4)

21 (9/12)

25 (11/14)

25 (11/14)

% prepositions 37.5% (52.9%/0%) 57.1% (88.9%/33.3%)

100%

The bracketed figures in columns I-IV distinguish between two categories of interrogatives: what versus how + any others.

208

Günter

Rohdenburg

Table 2. The use of prepositions introducing interrogative clauses dependent on selected adjectives in the quotations of the OED between the 16th and 20th centuries a)

ϊ

ΪΪ

ΪΪΪ

IV

0

prepositions

total

% prepositions

1 c l 5 0 9 - 1 6 9 9 (16 types)

60

3

63

4.8%

2 1700-1899 (20 types)

61

16

77

21.1%

3 1 9 0 0 - ( 1 3 types)

16

16

32

50%

a)

The analysis involves the following items: afraid, aware, baffled, careful, careless, cautious/cautionate, divided, dubious (excluding if/whether-c\a.uscs),foggy, ignorant, insensible/unsensible, indifferent, irresolute, reckless, sensible, scrupulous, the wiser, unaware, uncertain (excluding if/whether-clauses), undecided, unresolved/irresolved, unsure, vague, wary.

1.2. Major contextual constraints affecting the use of prepositional links An important factor playing a role in the establishment of prepositional links is provided by the type of interrogative phrase itself. The major contrasts which have been discovered so far may be rendered in terms of the two implicational scales presented in (6a-b). (6) a. what (nom) >>

b. what (nom) >>

r

how ^ > whether > who(m) whose < which > where when etc.

if prepositional phrases involving initial prepositions

what (det)

The claim expressed by (6a-b) is that elements to the left of any other(s) have always shown a stronger tendency to evolve prepositional links than those further to the right. In the case of if and phrase-initial prepositions (e.g. to what extent/by what means) the use of governed prepositions is generally excluded even today. This is why any relevant examples have usually been left out of consideration in the analyses presented in this paper. The bracketed information supplied in Table 1 suggests that what is

Cognitive complexity and horror aequi

209

generally the first interrogative to establish the more explicit prepositional alternatives. This assumption is clearly supported in Table 3, which deals with the adjective ignorant in a collection of historical corpora. Table 3. The use of prepositions introducing interrogative clauses dependent on ignorant in selected historical corpora (EEPF, ECF, EAF, NCF) a)/b) I 0

II prepositions

III total

IV % prepositions

234 (36)

159 (0)

393 (36)

40.5% (0%)

68 (9)

142 (0)

210 (9)

67.6% (0%)

b) nominal use

46 (7)

135 (0)

181 (7)

74.6% (0%)

c) determiner use

22 (2)

7 (0)

29 (2)

24.1%

166 (27)

17 (0)

183 (27)

9.3% (0%)

1 all examples 2 what a) all examples

3 remaining interrogatives a)

b)

The analysis excludes 2 //-clauses and 44 interrogative phrases containing initial prepositions. The figures in brackets give the number of infinitival interrogative clauses.

As is done in (6a-b), Table 3 draws a distinction between two functions of what, nominal uses as in examples (2), (3) and (5) and determiner uses as in (4). This grammatical contrast involves an implicational scale of its own (cf. [6b]): Nominal what has consistently been more strongly attracted to prepositional links than what in its determiner function. Further confirmation of the two basic splits (nominal what vs. all other interrogatives and nominal what vs. determiner what) is provided in Table 4, which deals with the adjective aware in the last two centuries. These findings leave no doubt that the contrast between what (nom) and what (det) has remained relatively stable over something like four centuries. However, what is less clear at present is whether the contrast between what (det) and the remaining interrogatives shown in Table 3 has been preserved equally well. The comparison of sections A and Β in Table 4 and similar analyses of other governing expressions indicate that how (the most frequently occurring item in the set of remaining interrogatives) has at least in some cases caught up with what (det) and even overtaken it. This is why we have expressed the contrast what (nom) vs. what (det) separately in (6b) and refrained from including it within (6a).

210

Günter

Rohdenburg

Table 4. The use of prepositions introducing what- and Aow-clauses dependent on aware in the Nineteenth-Century Fiction Corpus and parts of The Guardian (including The Observer)**

A NCF 1 what a) all examples b) nominal use c) determiner use 2 how Β selected years of The Guardian (including The Observer for 1994-2000) 1 what (g90-g00) a) all examples b) nominal use c) determiner use 2 how (g90-g95) a) all examples b) non-premodifying use c) premodifying use a)

ϊ 0

ΪΪ prepositions (of, about, as to)

38

58 55

96

60.4%

81

67.9%

3

15 106

20% 0%

26 12

ΪΪΪ total

IV % prepositions

106

—

59

704

763

92.6%

42

682

724

94.2%

17

22

39

56.4%

53 3 50

148 47 101

201 50 151

73.6% 94% 66.9%

The analysis excludes fourteen interrogative phrases in the NCF and one example in The Guardian (to what extent) which contain initial prepositions.

Section B2 in Table 4 draws attention to a further contrast, which seems to have become effective only in more recent times, the use of how as a premodifier of adjectives, adverbs and quantifiers and all remaining cases. The distinction is exemplified in (7a-b) and (8a-b). (7)

a. ...I became aware for the first time of how much contact there had been between us: ... (g90) b. On its slopes Tomba is only too well aware 0 how many good slalomers he has to beat. (g92)

(8)

a. ... but the others were very aware of how the banking operates. (g94) b. British coaches are not aware 0 how this workload can be achieved naturally. (g93)

The evidence in Table 4 indicates that it is again the qualifying use of the interrogative (as in [7a-b]) which is less far advanced in the establish-

Cognitive complexity and horror aequi

211

ment of prepositional links. At this stage, Table 4 represents the only analysis displaying this kind of contrast. If it can be supported on a regular basis, the implicational scales in (6a-b) will have to be extended as follows: (6) a.

what(nom)

r

how > whether > if who(m) prepositional phrases whose involving initial which > prepositions where when νetc. j

b. what (nom) > what (det) c. non-premodifying how >

premodifying how

With many nouns any earlier differences between what and (nonpremodifying) how in terms of prepositional vs. zero links have become obliterated by now. A case in point is provided by instructions (pi.) and instruction (sg.) (used far less frequently) in Table 5. Table 5. The use of prepositions introducing interrogative what- and /zow-clauses dependent on instructions) in selected English newspapers (t90-t94, g90-g94) I 0

1 what 2 how

1 (1.9%) 5 (2.5%)

II prepositions a) on 25 (47.2%) 154 (77.8%)

b) about 13 (24.5%) 18 (9.1%)

c) as to 8 (15.1%) 10 (5.1%)

III total d) of/for/in 6 (11.3%) 11 (5.6%)

e) total 52 (98.1%) 193 (97.5%)

53 198

With both what (represented mostly by the nominal use) and how (overwhelmingly represented by the non-premodifying use) the adoption of prepositional links has virtually gone to completion. On closer analysis, it is seen, however, that the contrast between what and how may still be reflected in the choice of prepositions. Items like as to and also about, which in general are more unusual, more informative or more prominent prosodically than most others, appear to be attracted to the what-clause, the more demanding environment in terms of preposition insertion (see Tables 1, 3 and 4). By contrast, on, the preposition most frequently associated inside and outside of w/z-clauses with instructions, in particular, is favoured

212

Günter

Rohdenburg

by less demanding Aow-clauses. In later sections, an attempt will be made to strengthen the hypothesis that more demanding or difficult contexts tend to trigger not just more prepositions but also prepositions like as to which signal interrogative complements more effectively than others. Finally, a large number of analyses dealing mostly with early and late 20th century corpora have shown that - disregarding if and phrases containing initial prepositions - whether has consistently been the last interrogative to become firmly associated with prepositional links. Consider, for instance, the data displayed in Table 6. Table 6. The use of prepositions introducing interrogative clauses dependent on the noun question in the LOB and Brown Corpora I 0 A LOB (1961) 1 what (5 nominal uses and 1 determiner use)

Β Brown (1961) 1 what (5 nominal and 3 determiner uses)

IV % prepositions

100% 5 (plural: 1)

5

100%

8

1

9

11.1%

1

7

8

87.5%

4

4

100%

10 (plural: 1)

16

62.5%

2 how (2)/who (1 )/why (1) 3 whether

III total

-

2 how (\)!who (1)!where (1) /why (2) 3 whether

II prepositions {of, about, as to)

6

In the case of question + whether, one of the last strongholds of zero links, the results in Table 6 suggest that the adoption of prepositional links, which surely corresponds to a syntactic regularization process, has been carried out a bit faster in (written) American English than in (written) British English. At this stage, we can only speculate as to what motivates the interrogative hierarchies set out in (6a-c). So far, three different kinds of explanation have suggested themselves, yet none of them seems to be capable of accounting for the whole range of facts known to us. -

It might be thought that the hierarchies in (6a-c) somehow reflect the contrast between complements (arguments) and adjuncts (non-

Cognitive complexity and horror aequi

-

-

213

arguments) within the interrogative clause. Unfortunately, this assumption fails to account for the hierarchies in (6b-c). Moreover, it cannot explain the fact that who(m), whose (+ N) and which generally used to show a strikingly different behaviour from nominal or determiner uses of what in earlier centuries. Another explanation ultimately derived from Poussa (1988) might attribute the special role played by nominal what to a local avoidance strategy: With nouns as governing expressions, the extra preposition immediately rules out any undesirable association with the non-standard relative clause marker what. While this assumption seems to take care of the contrast between nominal what and all other interrogatives, it does not explain the behaviour of whether or the hierarchy in (6c). What is more, the explanation is incapable of accounting for the fact that (at least) the hierarchies in (6a-b) are not confined to nominal governing expressions but also found with adjectives (cf. Table 3) and verbs (see section 2.2.1.). A third theory might propose to link the (initial) advantage which nominal what has over all other interrogative items (as shown in Tables 1 and 3) to the rise of the nominal (or free/headless) relative what in the second half of the 17th century as described by Kemp (1979). According to this view, the nominal relative (corresponding to that which) provides a model for its interrogative counterpart. While governed prepositions were still infrequent in the first half of the 17th century preceding interrogative w/ia/-clauses, they were hardly ever dropped when marking complements consisting of the much more frequently occurring nominal relative. This theory seems to be the best of a bad lot. It provides an explanation for (6b) and most of (6a). In addition, it would cover all kinds of governing expressions and not just nominal ones. However, the theory still fails to account for (6c), the behaviour of whether, and the obvious parallel between (6b) and (6c). Moreover, the initial behaviour of what (det) displayed in Table 3 and its decreasing strength in relation to how (and possibly other interrogative items) as seen in Table 4 remain unexplained.

Another important factor interacting with the interrogative hierarchies in (6a-c) involves the contrast between finite clauses as in examples (2), (4) and (5) and infinitival ones as in (3). Consider in this respect the bracketed information supplied in Table 3. In this collection of historical corpora the use of prepositional links is as yet confined to finite clauses. The assumption that infinitival clauses were comparatively slow to introduce preposi-

214

Günter

Rohdenburg

tional links is clearly corroborated by the bracketed information supplied in Table 7, which displays the evolution of interrogative clauses dependent on advice over the last four centuries. Table 7. The use of prepositions introducing interrogative clauses dependent on the noun advice in a series of British and American narrative corpora between the 17th and 20th centuries'0 I 0

II prepositions (as to, about, of, on)

III total

IV

%

prepositions

1 17th century publications of the EEPF

40 (24/16)

1 (1/0)

41 (25/16)

2.4% (4%/0%)

2 18th and early 19th century corpora (ECF, NCF1, Ε AFI)

45 (15/30)

5 (5/0)

50 (20/30)

10 (25%/0%)

3 mid and late 19th century corpora (MNC, LNC, NCF2, EAF2)

15 (4/11)

26 (21/5)

41 (25/16)

63.4% (84%/31.3%)

—

24 (6/18)

24 (6/18)

100%

4 fictional component of the (late 20th century) BNC (= wridoml) a)

The figures in brackets refer to finite and infinitival clauses, respectively.

Indeed, as has been shown by many additional and informal soundings, the connection between infinitival interrogative clauses and governed prepositions did not become firmly established until the middle of the 19th century. The intriguing resistance to the use of prepositional links exhibited by infinitival clauses will be explored in more detail in section 2.2.6. 1.3. Interpretative verbal links Another major category of linking devices competing to some extent with governed prepositions selected by certain (classes of) nouns, verbs and adjectives is represented by marked infinitives (or ίο-inflnitives) as in (9) and (10). (9)

They carried out an experiment as to/to see what pressure could be sustained.

(10)

There was a lot of competition/They competed with each other as to/to see who/which could produce the sentence with most prepositions at the end.

Cognitive complexity and horror aequi

215

Typically, we are dealing here with marked infinitives of fairly general (and possibly grammaticalized) verbs which, according to Liidtke (1984), provide an explicit interpretation of a following subordinate clause. In this area, strikingly similar changes to those displayed in Tables 1-7 can be shown to have occurred. Consider, for instance, the evolution of interrogative clauses like those in (10) associated with nouns of competition over the last four centuries (Table 8). Table 8. Dependent interrogative clauses involving who/which/whether {of them) in subject function associated with selected nouns of competition in the quotations of the OED, the EEPF and the Imaginative Prose section of the BNC (= wridoml) a ) I 0 1 EEPF (1518-1699) + OED (1507-1699)

II III prepositions /o-infinitives (as to, over, of) (see etc.)

IV total

V % II + III

32

1

1

34

5.9%

2 OED (1700-1899)

5

5

1

11

54.5%

3 BNC (wridoml) + OED (1900-)

—

8

10

18

100%

The analysis involves the following nouns: competition, contention, contest, disputing, emulation, quarrel, (rat) race, rivalry, strife, striving, struggle.

dispute,

While Early Modern English almost without exception (and like present-day German) joins its interrogative clauses directly to their respective governing expressions, present-day English employs either a governed preposition or some interpretative marked infinitive. In fact, in both (9) and (10) as well as many similar cases, an explicit link of some sort seems to be required in present-day English. We have seen, then, that from a crosslinguistic and historical point of view, which is the perspective adopted in this paper, verbal links like those in (9) and (10) function in similar ways to governed prepositions.

1.4. Exceptional reversals Despite the fact that the range of contexts compatible with directly linked interrogatives has contracted dramatically, there are still dozens of governing expressions left like those in (11) and (12) which may be found to occur with either a straight interrogative clause or an explicitly linked one.

216

Günter

Rohdenburg

(11)

a. The decision (on) whether they should accept the plan was postponed. b. The decision (on) whether to accept the plan was postponed.

(12)

There is no indication (of) what the Government has decided.

No doubt, the vast majority of such variation phenomena are the result of a long term tendency towards more explicit grammatical structures. Even so, developments showing the opposite trend may be observed in one or two exceptional cases. A relevant example is provided by the verb depend as in (13) where a striking increase in its frequency of occurrence has been accompanied - at least in British English - by a corresponding decrease in prepositional links. (13)

It depends (on) what they want.

Table 9 comparing the Nineteenth-Century Fiction corpus with the fictional component of the BNC indicates that the percentage of prepositional links has been halved in little more than a century. Table 9. The use of prepositions introducing interrogative clauses dependent on it (+ intervening elements) + depends in the Nineteenth-Century Fiction Corpus and the Imaginative Prose section of the BNC (= wridoml) a)

1 NCF (37.5m words) 2 BNC (wridoml) (18.9m words) a)

I on(upon)

II 0

III total

IV % on(upon)

20 [1819]

3 [1825]

23

87.9%

22

34

56

39.3%

The figures in angled brackets specify the average years of birth of the authors concerned.

2. Implications of the complexity principle 2.1. General remarks In analyzing constructional variants like those described in sections 1 . 1 1.4. and many other kinds in different areas of English grammar the Paderborn project concentrates on three extra-semantic, extra-stylistic and generally neglected determinants: phonological tendencies, cognitive complexity and horror aequi. This paper explores the last two factors to account for a

Cognitive complexity and

horror aequi

217

number of contextual constraints in the area of competing interrogative clause linkers. The complexity principle, whose implications have been investigated in a series of studies (e.g. Rohdenburg 1995b, 1996, 1998b, 1998c, 1999, 2000, 2002a; Rohdenburg and Schlüter 2000), states that in the case of more or less explicit constructional options the more explicit one(s) will tend to be preferred in cognitively more complex environments. While not necessarily responsible for the actuation of grammatical change, the complexity principle does allow us to make two evolutionary predictions: -

In the case of changes tending towards greater formal explicitness as in examples (2)-(5) and Tables 1-8, the novel and more explicit structures are first established in more complex environments. - With changes tending towards reduced grammatical explicitness as in example (13) and Table 9, the more explicit recessive structures survive longer in contexts involving an increased processing load. As far as differences in processing complexity are concerned, we cannot as yet support them by experiments of our own. We can, however, refer to a vast body of broadly psycholinguistic and typological research centring, in particular, round concepts such as syntactic weight (e.g. Hawkins 1990; Wasow 1997; Wasow and Arnold, this volume), referent accessibility (e.g. Ariel 1990; Gundel et al. 1993) and markedness (e.g. Croft 1990; Givon 1991; Wurzel 1998). The dominant type of change establishing prepositional and verbal links provides a great variety of complexity effects, and only some of the more important ones can be illustrated in the following sections. 2.2. Changes tending towards increased explicitness 2.2.1. Argument complexity We may begin by noting that the expansion of explicit links is syntactically motivated to the extent that verbs involving direct objects have been affected much earlier and much more drastically than corresponding intransitive uses.6 Compare in Table 10 the strikingly different speeds at which transitive uses like those in (14)—(16) and intransitive ones as in (17)—(18) have evolved prepositional or verbal links in the course of the 18th and 19th centuries. η

218

Günter

Rohdenburg

(14) I showed this strange epistle to my mother and consulted her on what I ought to do. (Anne Bronte, Agnes Grey, 1847). (15) I never consult other men 0 how they would have acted. (George Meredith, The Egoist, 1879). (16) Mrs. Margery Mumbleby ... consulting sporting records to see whether foxes were in the habit of doing such things. ... (Robert Smith Surtees, Handley Cross, 1854). (17)

When the time approached for him to leave prison, his mother and father consulted as to what course they should adopt. (Samuel Butler, The Way of All Flesh, 1903).

(18) Next day they consulted together 0 how they were to live. (Charles Reade, Hard Cash, 1863). Table JO. Prepositional, verbal and zero links with wAai-clauses and other interrogative clauses dependent on the verb consult in British and American historical corpora (ECF, EAF, NCF) a) I 0 1 intransitive uses a) all examples

III io-infinitive

IV and see

V total

VI % II-IV

133

29

-

2

164

18.9%

79

29

-

2

110

28.2%

54

-

-

54

0%

11

14

4

30

63.3%

what-clauses

1

10

-

11

90.9%

c) other interrogatives

10

4

19

47.4%

b) what c) other interrogatives 2 transitive uses a) all examples b)

a)

II prepositions

4

1 -

1

The analysis excludes four i/-clauses and 16 interrogative clauses containing initial prepositions.

What is more, the contrast applies consistently to two kinds of context, the more demanding category of wAa/-clauses and the set of remaining examples. With both transitive and intransitive uses of consult, the class of remaining interrogatives displays a very much reduced rate of preposition insertion compared to the category of what-clauses. We seem to be dealing, then, with two cross-cutting (complexity) dimensions. The prepositional variant is still absent where the two less demanding contexts (the intransitive use and the remaining interrogatives) cooccur (in lc). But its

Cognitive complexity and horror aequi

219

percentage is highest where the more complex transitive use coincides with the more demanding w/zaf-clauses (in 2b). By now the intransitive use (which has become comparatively infrequent) has all but caught up with the transitive use, whose development has been virtually completed. Unlike consult, most intransitively used verbs still favour the use of straight interrogatives. Yet with very few exceptions (cf. Rohdenburg 1998a: 240), their transitive counterparts cannot dispense with a prepositional or verbal link either. Table 11, which analyzes the verb check, illustrates this kind of contrast. Comparing rows 1 and 3 we find the expected asymmetry: Transitive uses like that in example (19) involve either verbal or prepositional links while intransitive uses as in (20) still prefer the straight interrogative. (19) He checked the car to seelon whether its cornering performance was adequate. (20) He checked (to see) whether the car was in good condition. Table //.Interrogative clauses associated with the verb check in selected parts of The Times and The Daily Mail

I

II

III

IV

V

0

prepositions

verbal links (to see etc.)

total

% prepositional and verbal links

1

29

30

100%

1 transitive uses involving direct objects (t91, t93, M93) 2 examples involving non-instrumental wiYA-phrases (t9Q-t94, M93-94)

26

21

50

48%

3 remaining intransitive uses (t91, M93)

192

31

224

13.8%

Consider next the use of non-instrumental with-phrases as in example (21). (21) Check with your bank (to seelon) what its charges are. Although functionally close to corresponding direct objects, these prepositional phrases are clearly indicative of a reduced degree of transitivity (in the sense of Hopper and Thompson 1980). As is shown in row 2 of Table 11, this transitional transitivity status is reflected in an intermediate proportion of explicit links.

220

Günter Rohdenburg

2.2.2. The there-clause as an example of a cognitively simple clause type The observations made in the previous section lead us to expect that there may be many other complexity contrasts concerning the structure of the superordinate clause which could play a role in determining the incidence of prepositional or verbal links. As is well known, the distribution of variable that in complement or adverbial clauses is extremely sensitive to the contrast between pronominal or NP subjects in the subordinate clause.8 Furthermore, there-subjects have generally been found to show a similar behaviour to personal pronouns. Other reasons for assuming existential there-clauses to involve a relatively low processing load include the following: -

their high frequency of occurrence the phonetic reducibility of there and the copula the low degree of semantic specificity associated with the copula

It should be instructive, therefore, to compare the existential thereclause, whose logical subject slot would be filled by governing nouns selecting interrogative clauses, with the set of all remaining clauses. To guarantee at least a small portion of straight interrogative clauses, it was decided to restrict the search to whether-clauses. Four of the nouns found suitable for such a comparison in present-day English are illustrated in (22M25). (22) a. There is uncertainty 0 whether a retrospective levy will apply. (t92) b. his days are spent in the endless struggle to pass muster, his nights in the endless uncertainty about whether he has passed it. (t92) (23) a. There was speculation among observers 0 whether the Congress or the Janata Dae would garner the greater share of the Muslim vote. (t91) b. Mr. Marshall's resignation, ..., renewed speculation about whether Mr. Bush will try to replace him with a black. (t91) (24) a. There is scepticism in the market 0 whether the further payments will be triggered. (t91) b. Britons can be forgiven their scepticism about whether this distinction holds. (t94)

Cognitive complexity and horror aequi

221

(25) a. The reason he delayed going off was that there was some confusion 0 whether it was him or Graham Dawe who was being dismissed. (d94) b. Some confusion has existed over whether clubs from the same nation can be drawn against other [sic], (d97) All of these nouns were analyzed in their capacity to select interrogative clauses both inside and outside of there-clauses in the British newspapers available to the Paderborn project. The results, which are presented in Table 12, generally confirm our working hypothesis. Table 12. The use of prepositions introducing interrogative whether-clauses dependent on uncertainty (sg.), speculation (sg.), scepticism (sg.) and confusion (sg.) in selected British newspapers (t90-t00, g90-g00, d91^d00, M93-M00) I 0 A uncertainty (sg.) 1 /Aere-clauses 2 remaining cases Β speculation (sg.) 1 fAere-clauses 2 remaining cases C scepticism (sg.) 1 iAere-clauses 2 remaining cases D confusion (sg.) 1 /ftere-clauses 2 remaining cases

II prepositions

III total

IV % prepositions

34

129

163

79.1%

16

355

371

95.7%

21

174

15

415

195 430

96.5%

15

93

83.9%

3

78 62

65

95.4%

9 1

188 244

197 245

95.4%

89.2%

99.6%

All of the individual analyses reveal more or less robust contrasts in the expected direction: There-clauses are indeed found to involve lower rates of preposition insertion than the set of remaining container clauses. Since the ongoing change establishing prepositional links is approaching completion even with whether-clauses, it is only to be expected that the effect of (clause-structural) complexity contrasts should eventually be minimized (see, in particular, section D). 2.2.3. Noun phrase complexity A number of previous studies in other areas of grammar have shown that object complexity regularly determines the distribution of more or less

222

Günter Rohdenburg

explicit constructional alternatives.9 Since direct objects, however, generally constitute a knock-out context with respect to prepositional or verbal links, we turn to prepositional objects or quasi-prepositional objects instead. This section takes a brief look at the type illustrated in (26). (26) It is up to them/the people concerned (to decide) how they want to go about it. Example (26) features an extraposed subject clause, which practically rules out the addition of ordinary prepositions. As is shown by the bracketed material, the interrogatives may, however, be introduced explicitly by a more or less specific verb of deciding. The use of such verbal links clearly represents an innovation which seems to be favoured by written registers rather than informal spoken ones. In addition, the contrast is seen to be sensitive to the complexity principle. Consider the analysis carried out in Table 13. Table 13. Interrogative clauses associated with (and following) the type it is up to NP/Pro in The Daily Mail and The Sunday Mail for 1993-1995 I 0

II as to

III marked infinitives (to decide and 26 others )

IV total

1 all examples

73

2

111

186

60.8%

2 (personal) pronouns

36

1

11

48

25%

3 remaining NPs

37

1

100

138

73.2%

4 average number of words in 3

1.92

7

3.02 a)

2.66

a)

V % II + III

The figure takes into account the expressions for example/he said7 under the War Crimes Act/in such an event following four of the NPs in question.

The evidence in rows 2 and 3 reveals a striking contrast between pronouns and full noun phrases governed by be up to. With full noun phrases, the proportion of interpretative verbs is over three times as high as with personal pronouns, which may be assumed to generally involve a much lighter processing burden. Furthermore, there is a parallel division within the class of full noun phrases - between shorter or longer ones (row 4)·

Cognitive complexity and horror aequi

223

2.2.4. Number contrasts Proponents of markedness theory generally assume that the morphologically marked plural noun represents a cognitively more complex category as well. It may be hypothesized, therefore, that constructions associated with plural nouns tend to exhibit a greater degree of grammatical explicitness than their singular counterparts. The assumption has been supported by several analyses. In Rohdenburg (2002a: 80-81), for instance, it is shown that with gerundial structures like those in (27) the plural form is definitely more likely to delay the ongoing loss of the preposition in than the singular. (27) He had (great/some/no) difficulty/difficulties (in) finishing his paper. In the case of dependent interrogatives, we would accordingly expect plural governing nouns to represent a more advanced stage in the evolution of prepositional links. Unfortunately, it is not easy to find many suitable governing nouns which meet the following requirements: Singular and plural uses must be functionally equivalent, and plural uses should occur in sufficient numbers. This is why the noun question, for instance, which only rarely functions as a plural noun selecting interrogative complements, had to be discarded as a potential item. As regards its near-synonym inquiry, such a comparison would be equally futile in present-day English, since (outside of //-clauses and interrogative phrases involving initial prepositions) both singular and plural nouns are virtually always followed by prepositional links. However, in 19th century English we regularly find contrasts such as the following: (28) Mrs. Wake's enquiry 0 what had happened was answered by the evidence of her own eyes. (Stanley John Weyman, From the Memoirs of a Minister of France, 1895) (29) He tried to make audible inquiries of what was required of him. (Herbert John Wells, When the Sleeper Wakes, 1899) The analysis summarized in Table 14 clearly supports the initial hypothesis for British English. The American data, though inconclusive, appear to point in the same direction. The comparison of the two standard varieties is revealing, however, since it parallels our findings concerning question + whether in section 1.2. (cf. Table 6): (written) American English

224

Günter

Rohdenburg

has again been shown to be implementing a regularization process more quickly than British English. Table 14. The use of prepositions introducing interrogative clauses dependent on the noun inquiry (enquiry) in 19th century British and American English a)

A

Β

a)

British (NCF1, MNC/B, LNC/B) 1 all examples 2 inquiry (sg.) 3 inquiries (pi.) American ( E A F l . M N C / A , LNC/A) 1 all examples 2 inquiry (sg.) 3 inquiries (pi.)

I 0

II prepositions

III total

IV % prepositions

27 23 4

45 25 20

72 48 24

62.5% 52.1% 83.3%

4 3 1

26 16 10

30 19 11

86.7% 84.2% 90.9%

The analysis excludes any if-, whether- and wAy-clauses as well as any interrogative phrases containing initial prepositions.

Additional evidence confirming for present-day English the correlation between the use of plurals and an increased proportion of prepositional links is provided by the noun indication, which is illustrated in examples (30H32). (30)

There was no indication 0 where financially strained Pyongyang would find the extra money. (t92)

(31) Seles's comeback has been the subject of protracted speculation but there are still no clear indications 0 when she will bring an end to an absence which ... (t94) (32) He had been alerted by the police after his name, a description of his car and indications of where he parked it were found on a list of Eta targets ... (t96) The analysis presented in Table 15 reveals two kinds of complexity effects.10 To begin with, there is the expected asymmetry between singular and plural nouns (cf. rows la and 2a). In addition, the contrast is found in two kinds of environment, existential there-clauses and the set of remaining clauses. With both singular and plural forms of indication, there-clauses are observed to trigger a very much smaller proportion of prepositional links than the set of remaining clauses. Not surprisingly, the percentage of the prepositional variant is lowest where the two easy-to-process elements

Cognitive complexity and horror aequi

225

(singular indication and there-clauses) cooccur (in row lb or example [30]). It is highest and close to 100% where the more complex plural coincides with the more complex remaining sentence structures (in row 2c or example [32]). Table 15. The use of prepositions introducing interrogative clauses dependent on the noun indication in selected British newspapers"* I 0

II prepositions (of, as to, about, on)

III total

92

572

664

86.1%

b) iAere-clauses

46

57

103

55.3%

c) remaining cases

46

515

561

91.8%

9 6

191 34

200

95.5%

40

3

157

160

85% 98.1%

1 indication (sg.) (t92, t98, g94, g98, d94, d98) a) all examples

2 indications (pi.) (t90-t00, g90-g00, d91-d00) a) all examples b) there-clauses c) remaining cases a)

IV

% prepositions

The analysis excludes any ;/-clauses and two interrogative phrases containing initial prepositions (to what extent, on which side).

2.2.5. Discontinuous structures In keeping with the complexity principle, a large number of previous investigations in other areas of English grammar (some of which are discussed in Hawkins, this volume) have demonstrated that structural discontinuities of various kinds tend to be correlated with strikingly higher degrees of (lexico-)grammatical explicitness. It is found, for instance, that the insertion of elements between governing verb (+ object) and complement clause increases the proportion of the variable complementizer that.11 In addition, we have regularly observed that the choice between (more explicit) finite and (less explicit) infinitival complements may be sensitive to the occurrence of non-adjacent or adjacent governing verbs.12 Again, there is a noticeable tendency for the cognitively more complex alternative to favour the more explicit grammatical option. In view of the overwhelming evidence confirming the complexity principle in this area, we expect to find similarly striking and regular correlations between the occurrence of non-adjacent governing expressions as in

226

Günter

Rohdenburg

examples (3), (7a), (18), (23a), (24a), (25b) and the use of prepositions introducing interrogative clauses. Analyses investigating this kind of phenomenon have turned out to be extremely time-consuming, and we have not yet been able to explore the issue to any great extent. The limited investigations carried out so far, however, suggest that the effect of nonadjacent governing expressions on the choice of prepositional links is - at best - very much weaker than might have been assumed. Admittedly, certain analyses like those in Table 16 (reproduced from Rohdenburg 2002a: 88) and Table 17 (section B) reveal more or less robust tendencies in the expected direction. Yet many (most?) others, like that in section A of Table 17, do not show any remarkable effects. Table 16. The effect of structural discontinuity on the use of prepositional and zero links with interrogative clauses dependent on the verb speculate in selected British newspapers (d91-d94, M93-M94) a) I 0

II prepositions

III total

IV % prepositions

1 elements invervening between the verb speculate and the interrogative clause

3

21

24

87.5%

2 adjacent interrogative clauses

55

124

179

69.3%

a)

This analysis excludes any if- and w/zei/ier-clauses.

Table 17. The effect of structural discontinuity on the use of prepositional links with interrogative whether-clauses dependent on doubts in British and American newspapers I 0 A British (d91-d00) 1 all examples 2 elements intervening between doubts and (preposition +) interrogative clause 3 remaining cases Β American (L92-L95, DFP92-DFP95, W90-W92) 1 all examples 2 elements intervening between doubts and (preposition +) interrogative clause 3 remaining cases

II prepositions

III total

IV % prepositions

161

580

741

78.3%

26

105

131

80.2%

135

475

610

77.9%

70

269

339

79.4%

14

30

44

68.2%

56

239

295

81.0%

Cognitive complexity and horror aequi

227

While case studies convincingly demonstrating the expected correlation between structural discontinuities and an increased rate of preposition insertion are still at a premium, virtually all large-scale investigations have shown that the contrast between adjacent and non-adjacent interrogative clauses tends to be reflected in the choice of different prepositions. Consider in this respect the data in Table 18. While far from conclusive, they do suggest that as to may play a special role in signalling more difficult-to-process dependent interrogative clauses. Table 18. The effect of structural discontinuity on the use of prepositional or zero links with interrogative /low-clauses dependent on the noun decision (sg. + pi.) in The Times and The Sunday Times for 1990-2000 a) I 0

II prepositions

III total

a) as to

b) others

c) total

1 elements inserted between decisions) and the Aow-clause

1 (1.5%)

9 (13.8%)

55 (84.6%)

64 (98.5%)

65

2 adjacent interrogative clauses

3 (1.3%)

11 (4.9%)

211 (93.8%)

222 (98.7%)

225

The analysis includes a few combinations of interrogative items involving how in initial position (e.g. how and when).

The more detailed probes undertaken in Rohdenburg (2002a: 91-93), which compares as to with the three prepositions most commonly introducing interrogative clauses, allow us to arrange them in a hierarchy of non-adjacency: (33) as to

>

about

>

J on\ \°fj

(33) embodies the claim that as far as the association of these items with non-adjacent interrogative clauses is concerned as to holds priority over about, which in turn holds priority over on and of. There is no doubt that as to occupies a unique position in the hierarchy: Not only is it the item used most frequently with discontinuous structures, but it also tends to become even more frequent with increasingly large insertions between governing expressions and dependent interrogatives. As is pointed out in Rohdenburg (2002a: 91-92), there are a number of other contextual constraints supporting the implicational scale in (33). They include the following:

228

Günter Rohdenburg

-

Interrogative clauses introduced by as to are much more easily fronted than those introduced by its closest competitor about. - As to is generally far less restrictive in the choice of governing expressions than any other preposition. - Unlike other prepositions, as to is used occasionally to signal extraposed interrogatives associated with copula verbs (e.g. it is up to him [as to] what he makes of it as in Table 13). - In corpus studies of present-day written English, almost half of all instances of the preposition as to have been found to be associated with a w/z-clause. All other prepositions display very much lower ratios with that of the runner-up about ranging around 15%. All of this suggests that (33) should be treated more generally as a scale of explicitness (reflective of a scale of complexity) and that as to constitutes by far the most effective signal of dependent interrogative clauses.

2.2.6. Finite and infinitival interrogative clauses This takes us back to the contrast between finite and infinitival clauses as in (34a-b), which seems to represent the most pervasive complexity dimension in the area of dependent interrogatives. (34) a. She was at a loss (about/as to) what to do. b. She was at a loss about/as to/to know/to discover/to ascertain what could be done. While markedness theories usually treat infinitives as derived and marked structures, I have suggested elsewhere (Rohdenburg 1999: 108— 109, 2002a: 89-90, 94) that finite interrogative clauses are in general more difficult to process than corresponding infinitival clauses. The contrast between the two types of clauses resembles that between full noun phrases and pronouns. In both cases, the reduction of linguistic structure may be taken to be indicative of a reduced processing load. This is in keeping with most kinds of markedness correlations assumed in the typological literature and also with accessibility theory (Ariel 1988, 1990, 1996), which says that minimal expressions tend to retrieve highly accessible entities. There are a number of additional considerations supporting such a treatment. Among them are the following:

Cognitive complexity and horror aequi

229

-

Comparing the transformational potentials of finite and non-finite complements, Ross (1974: 117) observes that "non-finite clauses are less restrictive than that-clauses, for all languages". - In the same vein, Hawkins (1999) shows that infinitives are much easier to extract out of than finite clauses. - Christofaro (1998) argues that, cross-linguistically, reduced deranked forms (like infinitival interrogatives) apart from being iconic of the semantic integration of two states of affairs are more likely to be used in contexts where the omitted information is easily recoverable. Under these circumstances, we should assume finite interrogative clauses to be further advanced than infinitival clauses in the establishment of more explicit links. We have already seen in section 1.2. that this assumption is indeed borne out by two large-scale historical analyses (cf. Tables 3 and 7): Prepositional links are introduced more easily and extensively with finite clauses than with infinitival ones. Further support in present-day English for the affinity between finite interrogative clauses and more explicit links is provided by (the British phrase) (be) in two minds and (the noun) choice as illustrated in (35) and (36). (35) a. He was in two minds about/as to what should be done. b. He was in two minds (about/as to) whether to go ahead with the plan. (36) a. She was given a choice as to what should be done next. b. She had a choice (of) where to begin. Table 19 dealing with (be) in two minds distinguishes between two classes of interrogatives, whether and all other items. Notice first of all that for both categories the incidence of prepositional links is very much higher in finite clauses than in infinitival ones. Moreover, the percentage of prepositional use is highest (100%) where the two more demanding contexts (the finite clause + other interrogative items) are combined. By contrast, the rate of prepositional insertion is lowest (37%) where the two easy types of context (the infinitival clause + whether) cooccur. In Table 20, which is concerned with choice, the analysis has been confined to wAef/zer-cIauses for the usual reason: The ongoing change regularizing prepositional links has virtually gone to completion outside of whether-clauses.

230

Günter

Rohdenburg

Table 19. The use of prepositions introducing finite and infinitival interrogative clauses dependent on (be) in two minds in selected British Corpora (t90-t00, g 9 0 - g 0 0 , d91-d00, M 9 3 - M 0 0 , BNC) a )

1 finite clauses a) all examples

II prepositions (as to, about, over, on)

III total

IV % prepositions

12

Al

59

79.7%

b) whether

12

30

42

71.4%

c) others

-

17

17

100%

80

70

150

46.7%

72

43

115

37.4%

8

27

35

77.1%

2 infinitival clauses a) all examples b) whether c) others a)

I 0

The analysis excludes one /^clause.

Table 20. Prepositional and zero links with whether-c\auses dependent on the noun choice in selected British newspapers (t90-t94, g90-g94, d91-d94, M 9 3 - M 9 4 ) a ) I 0

1 finite clauses

2 infinitival clauses a)

II prepositions

III total

IV % prepositions

a) of

b) as to

c) others

d) total

5 (7.2%)

11 (15.9%)

28 (40.6%)

25 (36.2%)

64

69

92.8%

28 (20.9%)

59 (44.1%)

8 (6.0%)

47 (35.1%)

106

134

79.1%

Apart from seven instances of the plural form choices, the analysis includes the following combinations involving choice·, right of choice (1), freedom of choice (2) and choice and concern (1).

The data in Table 20 reveal two kinds of contrast between finite and infinitival clauses. First, the proportion of prepositional options is distinctly higher with finite clauses than with infinitival ones. The second contrast concerns the overall distribution of the prepositions selected. Notice that here too the most effective link between various governing expressions and their associated interrogative clauses, namely as to, occurs much more frequently with finite clauses than with infinitival ones. Conversely, what is arguably the least marked preposition selected by choice, namely of, is found to be more heavily attracted to infinitival clauses.

Cognitive complexity and horror aequi

231

As has been shown above (Tables 5 and 18), this kind of prepositional patterning tends to be preserved in those cases where the (basic?) contrast involving prepositional and zero links has been given up or is about to disappear. Compare in this respect the data in Table 21, which provide additional confirmation of the complexity hierarchy outlined in (33). Table 21. Prepositional and zero links with interrogative clauses dependent on instruction(s) in The Times for 1990-1994 and The Guardian for 1990-1994 (including The Observer for 1994) I 0

II prepositions

a)

c)

III total

e)

on

b) about

as to

d) of/for/in

total

1 all examples

8 (2.8%)

194 (69.0%)

39 (13.9%)

21 (7.5%)

19 (6.8%)

273 (97.2%)

281

2 finite clauses

4 (5.6%)

29 (40.3%)

18 (25%)

12 (16.7%)

9 (12.5%)

68 (94.4%)

72

3 infinitival clauses

4 (1.5%)

165 (78.9%)

21 (10.0%)

9 (4.3%)

10 (4.8%)

205 (98.1%)

209

Finally, a few further remarks should be made in regard to the treatment of the contrast between finite and infinitival clauses as a major complexity dimension. Would it not be more appropriate to view the special affinity between finite clauses and the use of prepositional or verbal links simply as an example of markedness reversal? And would it not be much more natural to regard the resistance shown by infinitival clauses to prepositional or verbal links as simply reflecting an iconically motivated tendency for (semantically) closely integrated states of affairs to favour syntactic integration as well? While I cannot see any virtue in postulating a markedness reversal, the (second) iconic argument, which was also mentioned above, will have to be treated more seriously (cf. Christofaro 1998). However, if we decided to attribute the numerous asymmetries between finite and infinitival clauses exclusively to an iconic tendency we would be missing an important generalization. As has been demonstrated throughout this paper, the contrast between finite and infinitival clauses produces the same general kind of (lexico-)grammatical patterning as is induced by the set of established complexity factors (or contrasts between more or less demanding alternatives) listed in (37).

232

Günter Rohdenburg

(37) a. b. c. d. e. f.

transitive vs. intransitive uses (2.2.1.) complex clauses vs. simpler there-clauses (2.2.2.) noun phrase complexity: fxill NPs vs. pronouns (2.2.3.) plural vs. singular nouns (2.2.4.) discontinuous vs. continuous structures (2.2.5.) (nominal) what vs. how + other interrogative items (1.1 -1.2.)

It should be remembered that these dimensions have been shown to be reflected in two different ways: -

in the choice between zero links and more explicit prepositional or verbal links - with prepositional links, in the choice between more or less effective/explicit markers of dependent interrogative clauses (e.g. as vs. on/of) Since virtually all of the relevant distributional patterns happen to be duplicated in the case of finite vs. infinitival clauses, it is only natural to treat this contrast as a parallel complexity dimension. 2.3. Changes tending towards decreased explicitness As has been indicated, the verb depend provides a counter-example to the dominant tendency resulting in the establishment of novel prepositional links. During the 20th century and in certain environments, the omission of an originally obligatory preposition has become a perfectly normal feature in British English (cf. Table 9 in section 1.4.). There is no doubt that this development is spearheaded by informal colloquial styles. A first analysis of the phenomenon is presented in Table 22, which deals with a large corpus of written English. The table distinguishes five subject categories, and it is only the first four which are regularly found without a prepositional link: zero subjects, it, it all and that. It is seen, then, that zero links have become established first with only those kinds of subject which arguably involve the lowest degree of semantic specificity and the highest degree of accessibility/givenness or processing ease (cf., e.g., Ariel 1988, 1990, 1996; Givon 1991: 354 ff., 1992; Gundel 1996; Gundel et al. 1993; Yule 1981). Moreover, it could be argued that the ordering within the group of four, which conforms to Ariel's accessibility scale and to Gundel's givenness hierarchy,

Cognitive complexity and horror aequi

233

represents an inverse scale of processing complexity. For instance, compared with it, the demonstrative pronoun that has been found "to signal prominent or complex foregoing information" (Montgomery 1989: 253), and demonstrative pronouns have also been shown to favour more distant antecedents than personal pronouns (e.g. Ariel 1988: 69-72, 1996: 2224).14 Table 22. The choice of subject determining the omission of prepositional links with depends + (immediately) following interrogative clause in selected British newspapers (g90-g91, t90-t92, d91-d94) a) I on(upon) 1 0 2 it 3 it all 4 that 5 other subject expressions a)

II 0

III total

IV % on(upon)

3

22

189 106

203 63

25 392

12% 48.2%

169

62.7%

37 732

6 5

43 737

86.0% 99.3%

The analysis excludes any /^clauses and any examples of interrogative phrases containing initial prepositions.

In addition, accessibility theory predicts that zero anaphors retrieve the most accessible mental entities and that the strengthened variant of it, namely it all, should be lower on the accessibility scale than it itself (e.g. Ariel 1988, 1996). Notice in this connection that, unlike that, neither it nor it all lend themselves to being highlighted by a variety of focusing devices like those illustrated in (38a-c). 15 (38) a. *It/It all we have to reject, (object-fronting) b. *What we have to reject is it/it all. (pseudo-clefiting) c. *It is it/it all (that) we have to reject, (it-clefting) Comparing the four easy-to-process subject types we find that the postulated differences in processing complexity are matched fairly closely by their rates of prepositional retention. The incidence of prepositional links is lowest with zero subjects and highest with that}6 It and it all occupy an intermediate position with the more complex variant generally involving a higher proportion of prepositional links. The use of straight interrogatives with the verb depend is subject to many other constraints which are naturally accounted for in terms of the complexity principle. First, in keeping with general typological findings,

234

Günter

Rohdenburg

the phenomenon of preposition deletion is very much further advanced in the simple present than in all other verb categories. So far, only structures consisting of a modal auxiliary + infinitive as in (39) have been subjected to a closer analysis (cf. Table 23). (39) It would depend (on) what you had to do. The comparison of Tables 23 and 22 yields the following observations: -

With both it and that, the proportions of retained prepositions are very much higher in modal constructions than in the present tense. 17 - This means that modal constructions preserve the striking contrast between it and that the latter subject expression is still virtually confined to prepositional uses. Table 23. The choice of subject determining the omission of prepositional links with the combination auxiliary + (adjacent) depend (inf.) + (immediately) following interrogative clause in selected British newspapers (t90-t94, g90-g94, d91-d94, M 9 3 M94) a) I on (upon)

II 0

III total

IV % on (upon)

1 it

92

20

112

82.1%

2

52

1

53

98.1%

a)

that

The analysis excludes any (/"-clauses and any examples of interrogative phrases containing initial prepositions.

Second, the use of more explicit variants is generally encouraged by structural discontinuities. With the present tense and auxiliary + infinitive constructions, the tendency applies to - mostly adverbial - insertions like those found in (40a-b) and (41a-b): (40) a. It entirely depends (on) how they go about it. b. It depends, after all, (on) how they go about it. (41) a. It would, however, depend (on) what they had in mind. b. It would depend, however, (on) what they had in mind. Concerning the simple present, the effects of these complications are summarized in Table 24. Comparing rows 3 and 2a, it is seen that the use of such insertions increases the overall proportion of prepositional links by

Cognitive complexity and horror aequi

235

a striking 20%. Intriguingly, the adverb just (in row 2b) reverses the general trend. Here, the percentage of prepositional retention is even lower than in canonical structures generally. I would suggest that the exceptional status ofjust is explained by its strong association with informal colloquial styles. If the style-marker just is excluded from the analysis of inserted adverbials (as is done in row 2c), the proportion of on rises by another 10% to a value approaching even that of canonical (or continuous) /forf-clauses (as shown in Table 22). Table 24. The effect of elements inserted between it + depends + interrogative clause on the choice of prepositional links in selected British newspapers (g90-g91, t90-t92, d91-d94) a) I on(upon)

II 0

III total

IV % on(upon)

248

230

478

51.9%

59 8 51

27 13 14

86 21 65

68.6% 38.1% 78.5%

189

203

392

48.2%

1 all examples (excluding it all) 2 intervening elements (on either side of depends) a) all examples b) it just depends + wA-clause c) remaining examples 3 remaining straightforward cases a)

The analysis excludes any (/^clauses and any examples of interrogative phrases contain ing initial prepositions.

Allowing for very much higher levels of retained prepositions in the case of auxiliary + infinitive, a similarly striking contrast is seen in Table 25. Table 25. The effect of elements inserted between it + auxiliary + depend (inf.) + interrogative clause on the choice of prepositional links in selected British newspapers (t90-t94, g90-g94, d91^194, M93-M94)

1 all examples (excluding it all) 2

intervening elements (on either side of depend or the auxiliary/non-canonical word orderings in questions [3])

3 remaining straightforward cases

I on(upon)

II 0

III total

IV % on(upon)

143

22

165

86.7%

51

2 (just: 1)

53

96.2%

92

20

112

82.1%

236

Günter Rohdenburg

3. Implications of the horror aequi principle 3.1. Introduction Very briefly, the horror aequi principle involves the widespread (and presumably universal) tendency to avoid the use of formally (near-) identical and (near-)adjacent (non-coordinate) grammatical elements or structures (cf., e.g., Brugmann 1909, 1917/1920; Dressier 1976; Stemberger 1981; Plank 1981: 149-153; Menn and MacWhinney 1984; McCawley 1988: 305-308; Plag 1998; Hinderling 1998; Yip 1988, 1998; Tang 2000; Rohdenburg and Schlüter 2000: 461—467; Mondorf, this volume).18 As is well known, there are many and various restrictions on the use of immediately successive -ing forms in present-day English and earlier stages of English (cf., e.g., Ross 1974; Bolinger 1979; Gramley 1980: 166-167; McCawley 1988: 306-308; Westney 1992; Yip 1998: 232-234; Pullum and Zwicky 1998; Rudanko 2000; Rohdenburg and Schlüter 2000: 452-465; Mair, this volume; Vosberg, this volume). In a similar way, English presumably has always shown a more or less pronounced aversion to sequences of (noncoordinated) to-infinitives, in particular to those not separated by any intervening material (cf., e.g., Gramley 1980: 166-167; Lind 1983a, 1983b; Kjellmer 1985; Rohdenburg 1995b: 380-382; Fanego 1996a; Rudanko 2000; Rohdenburg and Schlüter 2000: 462-465; Mair 2002: 125-126, this volume; Vosberg, this volume). No doubt the tendency to establish to-infinitives as verbal links has greatly increased the likelihood of producing an undesirable sequence of marked infinitives. Accordingly, the horror aequi principle would lead us to predict that such to-infinitives should be dispreferred in those cases where the (immediately preceding) governing or superordinate expression was represented by a marked infinitive itself. In principle, we can expect to find two general avoidance strategies: -

The introduction of the (interpretative) to-infinitive could be delayed following a fo-infinitive. Some other verbal or prepositional alternative might be chosen instead.

3.2. Case studies So far, my investigations of the problem have focused on three (intransitively used) verbs: wait, try and check. We have already seen that in the

Cognitive complexity and horror aequi

237

case of intransitive check, the evolution of verbal links has not advanced very far (cf. Table 11 in section 2.2.1.)· With wait and try, however, the quotations from the OED analyzed in Tables 26 and 27 permit us to infer that the change was largely completed in the 17th and early 20th centuries, respectively. Table 26. Interrogative clauses associated with and following the verb wait in the quotations of the OED from 1500 onwards I 0

II /o-infinitives (see, espie, know)

1 to-infinitive

1 (al500-34)

2 remaining uninfected verb forms

3 (1508, 1598, 1671-2)

3 inflected verb forms

6 (al500, 1526, 1596,1645, 1667, 1837)

—

4 (1633, 1647, 1887, 1958) (50%) 5 (1609, 1870, 1940, 1960) (45.5%)

III and see

IV total

4 (1719, 1825, 1875,1964) (80%) 1 (1836)

5

8

(12.5%) 11

Table 27. The decline of interrogative clauses linked directly to the intransitive verb try in the quotations of the OED I 0 1 1600-1799 2 1800-1849

96 16

3 1850-1899

16

4 1900-1949

3 (1911,1921, 1923)

5 1950-1983

II and see/and find out -

—

1 (1881) 1 (1943) -

IV total

V % II + III

4 (= trying) 10

96 20

0% 20%

27

37%

5

9

66.7%

32

32

100%

III to-infinitives (see/find out etc.) -

Examples (42) and (43) illustrate the earlier states of affairs. (42)

..., and all their attendants stood without, waiting 0 when Plivio would dismiss their Lady, ... (John Crowne, Pandion and Amphigenia, 1665)

238

Günter Rohdenburg

(43) He thought it much wiser to try 0 whether the workhouses could not be ameliorated, ... (The Times, 1843) The general predictions derived from the horror aequi principle are indeed confirmed by the available evidence. We do not know for sure whether the marked infinitive to wait managed to preserve the zero link very much longer than other uses of the verb. The data in Table 26 suggest, however, that in this environment the generally used infinitival link was largely replaced at least two centuries ago by the unproblematic alternative shown in example (44). (44) However, we had no Remedy, but to wait and see what the Issue of Things might present; ... (Daniel Defoe, Robinson Crusoe, 1719) At any rate, the assumption that the use of the linker and see is motivated to a large extent by a desire to avoid a succession of two marked infinitives is supported both by 19th/early 20th century and present-day corpora (cf. Tables 28 and 29). Notice first of all that at least in present-day English the linker and see (etc.) tends to be adjacent to the relevant form of wait. Thus examples of type (44) usually contrast with cases like (45). (45)

You will have to wait a bit to see how it works.

While the exclusive use in both Tables 28 and 29 of and see with imperatives may be due to as yet unexplained semantic (or pragmatic) tendencies, the striking contrast in all of these corpora between the behaviour of marked and unmarked infinitives surely must be accounted for in terms of horror aequi. In the earlier and smaller database of Table 28, there is not a single case of two immediately successive marked infinitives. Wherever other provisions to avoid such an identity effect are not made, the linker and see comes to the rescue in a total of 11 instances. By contrast, half of the unmarked infinitives are immediately followed by interpretative to-infinitives like to see. Similar asymmetries can be observed in a presentday newspaper corpus (Table 29). As is shown in Table 29, the linker and see (etc.) has spread by now to all other uninflected forms of wait. Even so, with the peculiar exception of the imperative, the highest proportion of and see (etc.) is still found with to wait itself. Moreover, the distributional pattern today points to a functionally motivated analogical extension: The unmarked infinitive displays a very much smaller percentage of and see, and finite verbs use even fewer of these links.

Cognitive complexity and horror acqu i

239

Table 28. The distribution of interpretative verbs of cognition (and perception) linking interrogative clauses to the verb wait in three historical corpora of British English (MNC/B, LNC/B, ETC/B) a) I 0

II for

III and + verb stem (see: 35, others: 2)

IV toinfinitives (see: 84, others: 60)

V total

VI % adjacent infinitives

2 (3.2%)

3 (4.8%)

21 (E: 3) (33.3%)

63

28.6%

b) to wait

1 (6.3%)

1 (6.3%)

1 (E) (6.3%)

16

0%

c) wait = unmarked infinitive

1 (2.7%)

2 (5.4%)

37 (E: 10) (58.7%) 13 (E: 2) (81.3%) 15 (E: 3) (40.5%)

19 (Ε: 1) (51.4%) 1 (E)

37

48.6%

1 wait a) all examples

d) wait = finite verb e) wait = imperative 2 remaining (inflected) verb forms a)

-

-

-

9 (E: 5) (100%)

-

2 (1.4%)

21 (14.4%)

1

—

123 (E: 41) (84.2%)

9

0%

146

56.2%

The figures following the bracketed label Ε refer to the number of instances involving additional elements intervening between the verb wait (+ linking device) and the (dependent) interrogative clause.

Table 29. Interrogative clauses associated with the uninflected verb form wait in The Guardian for 1990, 1992, 1993 and 1994 (including The Observer for 1994)a)/b) II Ill IV I to-infinitives total % and see and see

a) 1 to wait 2 wait = unmarked infinitive 3 wait = finite verb 4 wait = imperative a)

b)

62 42 7 6

c)

see 20 52

b) others 1 10

total 21 62

22

2

24

-

-

83 104

74.7% 40.4%

31 6

22.6% 100%

The analysis is restricted to interrogative clauses immediately following wait (+ interpretative verb). The analysis excludes one instance of a /o-infinitive involving the passive prepositional verb to be enlightened about.

240

Günter

Rohdenburg

As is well-known, the replacement of the to-infinitive in (46a) by the pattern and + verb stem in (46b) has been frowned upon for quite some time by norm-oriented speakers. (46) a. One has to try to find out what happened. (Biber et al. 1999: 738) b. One has to try and find out what happened. (Biber et al. 1999:738) While the and see type has become well established with wait even in written English, similar examples with try as in (46b) "are relatively common in conversation but generally avoided in formal written registers" (Biber et al. 1999: 739). Biber et al. go on to point out that "Nearly all occurrences of try + and + verb in news and academic prose are used to avoid a sequence of to-clauses ...". Similar though weaker horror aequi effects had already been noted in Lind (1983b). It is against this background that we should view the change leading from directly linked interrogative clauses as in (43) to examples like (46a/b) which contain an interpretative verb of cognition (or perception). Consider first the data in Table 30, which are based on extracts from The Times between 1785 and 1992. Table 30. The distribution of interpretative verbs of cognition (and perception) linking interrogative clauses to the (intransitive) verb try in the Changing Times database (1785-1992) a V b ) III io-infinitive {see/find out/ discover etc.) 28 [1946]

IV total

V % toinfinitive

10 [1860]

II and + find out/ understand clearly 2 [1937]

40

70%

8 7 [1857] 1 [1898]

2 1 [1938] 1 [1936]

9 4 [1970] 5 [1941]

19 12

47.4% 33.3%

7

71.4%

19 [1942]

21

90.5%

I 0

1 all examples 2 try a) all examples b) to try c) 0 try (= infinitive and imperative) 3 trying/tried/tries

a) b)

2 [1792, 1905]

The bracketed figures refer to the (average) years of publication involved. Negated verbs (try not to see how (1) and verbs of describing (2), showing (1) and explaining (4) have been excluded from consideration. The infinitives in column III include the following verbs: see (5), find out (6), discover (6), guess, grasp, read, understand, figure out, imagine, decide, establish, ascertain.

Cognitive complexity and horror aequi

241

The evidence strongly suggests that the reluctance to adopt the type and see in (extremely) formal written English had given the zero link a much longer lease of life after to try than elsewhere. Notice, in particular, that in this environment the to-infinitive is introduced much later than after other uses of uninflected try. Compare these findings with the analysis of a presumably much more informal corpus of novels published mostly in the second half of the 19th century (cf. Table 31). Table 31. The distribution of interpretative verbs of cognition (and perception) linking interrogative clauses to the (intransitive) verb try in the authors of the NineteenthCentury Fiction corpus who were born between 1820 and 1859 a) I 0

II and + verb stem

Ill /o-infinitive

IV total

103 (57.9%)

11 (6.2%)

64 (40%)

178

83 (79.0%)

11 (10.5%)

11 (10.5%)

105

36 (80%)

9 (20%)

c) 0 try (= infinitive and imperative)

47 (78.3%)

2 (3.3%)

trying/tried/tries

20 (27.4%)

1 all examples 2 try a) all examples b) to try

3

a)

—

—

V % I + II 64.0%

89.5% 45 100%

11 (18.3%)

60

53 (72.6%)

73

81.7%

27.4%

Negated verbs and verbs of describing, showing and explaining have been excluded from consideration.

This represents the first generation of authors that showed a sizeable decline of zero links within the category of uninflected forms of try}9 Again, there is a noticeable contrast between marked infinitives and other uses of uninflected try. This time it concerns the distribution of the two competing types of verbal links rather than zero links, which had been retained to almost the same extent by both categories of uninflected try. With to try, the adoption of verbal links is exclusively accounted for by the type and + verb stem. By contrast, other uninflected uses of try clearly favoured the infinitival link over the type and + verb stem. Compare, for instance: (47)

We could not, of course, help peering over the bank to try and see what was going on; ... (George Tomkyns Chesney, The Battle of Dorking, 1871)

242

Günter

Rohdenburg

(48) But since you are awake, I will try to find out where we are. (William Clark Russel, The Wreck of the Grosvenor, 1879) This brings us to a more detailed analysis of (intransitive) check in Table 32. Table 32. Interrogative clauses associated with the intransitive use of the verb check in selected British newspapers I 0 A

Β

The Times & The Sunday Times for 1991 1 to check

40

II prepositions

III and decide

1 1

IV to see etc.

V total

VI % to see etc.

-

41

0%

16

62

25.8%

2 remaining uses

45

The Daily Mail & The Sunday Mail for 1993 1 to check

43

-

-

43

0%

2 remaining uses

64

-

14

78

18.0%

2

231

C The Daily Telegraph & The Sunday Telegraph for 1991-1994 1 to check 2

checking

-

209

19 (8.2%)

73

5 (5.4%)

I (0.4%) —

0.9% 15

93 16.1%

Here, one observation stands out amongst all others. Verbal links containing ίο-infinitives are still practically avoided after to check itself. Moreover, in the two apparent counter-examples in section C, there happens to be some adverbial material between the two marked infinitives. At this stage, any other available links, prepositions or and + verb stem, appear not to be used much more frequently in this environment than elsewhere. 4. Conclusion After surveying the changes undergone by dependent interrogative clauses over the last four centuries, this paper discusses the implications of the complexity principle and the horror aequi principle. Both the synchronic and diachronic predictions derived from the two principles are shown to be confirmed by a great variety of case studies. Concerning the complexity

Cognitive complexity and horror aequi

243

principle, we have seen that more explicit novel constructions are established earlier and faster in more complex environments. On the other hand, more explicit recessive constructions tend to be preserved better and longer in the same types of context. The horror aequi principle is concerned with structures potentially involving undesirable identity effects. Such contexts may constitute functional niches in which obsolescent but contextually well adapted constructions (like those involving zero links) can survive a bit longer and where adaptive novel constructions (like and + verb stem) may become established earlier than elsewhere.

Notes * This study was carried out within the Paderborn research project Determinants of Grammatical Variation in English, which is supported by the German Research Foundation (Grant Ro 2271/1-1). 1. These developments restrict the range of application of Hawkins' (1986) explicitness thesis which says that English grammar may never involve more explicit structures than German. 2. Cf. also Kjellmer (1980), who has shown for present-day English that the survival of the to-infinitive is to a very large extent correlated inversely with the entrenchment value of the prepositional object. 3. Most of the analyses described in this paper have been carried out by means of the search program MicroConcord. The program allows the analyst to restrict the search by using a context word within a window of nine words to the left and nine words to the right of the search word in question. This might have resulted in a very small number of omissions in some analyses. 4. The usual caveats apply concerning the delimitation of dependent interrogative clauses from so-called free (or headless/nominal) relatives (cf., e.g., Huddleston 1971). In addition, any examples displaying a so-called non-restrictive or appositive interpretation as in (i) have been excluded from consideration, (i) "My original question, why he did it at all, has not been answered." (Quirk/Greenbaum/Leech/Svartvik 1972: 736) 5. All of the analyses carried out in this study represent either single interrogative clauses or initial ones in the case of coordinated complements. This restriction is motivated by the fact that non-initial interrogative clauses even today tend not to repeat a governed preposition introducing either a nominal or interrogative complement as in the following example: (i) The question was raised as to what data she prefers and 0 how they should be presented. 6. Intransitive uses are here defined as those not involving any direct objects. 7. Other areas in which argument complexity plays an important role in determining grammatical change and variation in English have been identified in Mondorf (this volume) and Rohdenburg (1995b, 1998a, 2002b). 8. Cf., e.g., Elsness (1984); Rohdenburg (1996: 161-164, 1998c: 197, 1999: 102-103, 2000: 38-39); Rohdenburg and Schlüter (2002: 457-458) and Wasserman (1976). 9. Cf., e.g., Rohdenburg (1995b: 375-376, 1996: 158, 161-162, 1998b: 217-218, 1999: 103-104, 107, 2000: 26-31, 36-37, 2002a: 85-86, 90-91) and Rohdenburg and Schlüter (2000: 452-454).

244

Günter Rohdenburg

10. This is an extended analysis of that shown in Table 10 in Rohdenburg (2002a: 87). 11. Cf., e.g., Wasserman (1976: 147-148), Elsness (1984), Rohdenburg (1995b: 383-384, 1996: 160-161, 1999: 102-103). 12. Cf., e.g., Rohdenburg (1995b: 376-378, 1996: 166-167, 1998c: 192-193, 1999: 104106). 13. Its American equivalent {be) of two minds, which seems to display similar tendencies, is used much less frequently. 14. According to Biber et al. (1999: 239), however, personal pronouns display a larger anaphoric distance than demonstrative pronouns outside of conversational English. 15. This certainly relates to the fact that it represents the pronoun "that is lowest on the scale of accent" (Bolinger 1977: 516). For an analysis of special contexts allowing it to be stressed see, in particular, Seppänen (1997). 16. It should be pointed out, however, that unlike British newspapers, some other corpora (both British and American) mysteriously may show higher prepositional retention rates for zero subjects than for (situational) it. This certainly constitutes an intriguing problem which merits further investigation. 17. It all happens to be unavailable with modal auxiliaries in Table 23. 18. The term itself was presumably coined by Brugmann (1909, 1917/1920). 19. The fact that inflected forms of try, in particular trying, which like inflected forms of wait are incompatible with and + verb stem, are much more progressive in this respect, would seem to be irrelevant to our present concerns and will accordingly be neglected.

Primary Sources BNC BROWN

British National Corpus Brown University Corpus 1991 ICAME Collection of English Language Corpora Changing Times: History in the Making on CD-Rom 1998 News Multimedia d91-d00 The Daily Telegraph & The Sunday Telegraph on CD-Rom 1991-2000 Chadwyck-Healey DFP92-DFP95 The Detroit Free Press on CD-Rom 1992-1995 Knight-Ridder Information Inc. EAF Early American Fiction 2000 Chadwyck-Healey EAF1 1st part of the EAF containing only those authors born in the 18th century (1744-1799) EAF2 2nd part of the EAF containing only those authors bom in the 19th century (1801-1827) ECF Eighteenth-Century Fiction 1996 Chadwyck-Healey EEPF Early English Prose Fiction 1997-2000 Chadwyck-Healey ETCa) Early Twentieth Century Corpus - a selection of British and American writings by authors born between 1870 and 1896 Source: Project Gutenberg ETC/B British writings in the ETC g90-g00 The Guardian (including The Observer 1994-2000) on CD-Rom 19902000 Chadwyck-Healey L92-L95 The Los Angeles Times on CD-Rom 1992-1995 Knight-Ridder Inc. LNCa) Late Nineteenth Century Corpus - a selection of British and American writings (complementary to the EAF and the NCF) by authors born between 1830 and 1869 Source: Project Gutenberg

Cognitive complexity and horror aequi LNC/A LNC/B LOB M93-M00 MNCa)

MNC/A MNC/B NCF NCF1 NCF2 OED t90-t00 W90-W92 wridoml a ·*

245

American writings in the LNC British writings in the LNC Lancaster-Oslo/Bergen Corpus 1991 ICAME Collection of English Language Corpora The Daily Mail & The Mail on Sunday on CD-Rom 1993-2000 ChadwyckHealey Mid-Nineteenth Century Corpus - a selection of British and American writings (complementary to the EAF and the NCF) by authors born between 1803 and 1828. Source: Project Gutenberg American writings in the MNC British writings in the MNC Nineteenth-Century Fiction 1999-2000 Chadwyck-Healey 1st part of the NCF containing only those authors born in the 18th century (1728-1799) 2nd part of the NCF containing only those authors born in the 19th century (1800-1869) The Oxford English Dictionary, 2nd edition (Version 1.13) 1995 Oxford University Press The Times & The Sunday Times on CD-Rom 1990-2000 Chadwyck-Healey The Washington Times (including Insight on the News) on CD-Rom 19901992 Wayzata Technology fictional/narrative component of the BNC The list of texts making up this corpus is available on request from the Paderborn project (email: [email protected]).

References Ariel, Mira 1988 1990 1996

Referring and accessibility. Journal of Linguistics 24: 65-87. Accessing Ν Ρ Antecedents. London: Routledge. Referring expressions and the +/- coreference distinction. In: Thorstein Fretheim and Jeannette K. Gundel (eds.), Reference and Referent Accessibility, 12-35. Amsterdam/Philadelphia: Benjamins. Biber, Douglas, Stig Johansson, Geoffrey Leech, Susan Conrad, Edward Finegan 1999 Longman Grammar of Spoken and Written English. Harlow: Longman. Bolinger, Dwight 1977 Another look at main clause phenomena. Language 53:511-518. 1979 The jingle theory of double -ing. In: David J. Allerton, Edward Carney and David Holdcroft (eds.), Function and Context in Linguistic Analysis: A Festschrift for William Haas, 41-56. Cambridge: Cambridge University Press. Brugmann, Karl 1909 Das Wesen der lautlichen Dissimilationen. Abhandlungen der philologischhistorischen Klasse der königlich-sächsischen Gesellschaft der Wissenschaften 27: 141-178. Leipzig: Teubner. 1917/1920 Gleichklangvermeidung in der lautgesetzlichen Entwicklung und in der Wortbildung. Indogermanische Forschungen 38: 117-128.

246

Günter

Rohdenburg

Christofaro, Sonia 1998

Deranking and balancing in different subordination relations: A typological study. Sprachtypologie und Universalienforschung 51: 3—42.

Croft, William 1990 Typology and Universals. Cambridge: Cambridge University Press. Dressier, Wolfgang U. 1976 Phono-morphological dissimilation. In: Wolfgang U. Dressier and Oskar E. Pfeiffer (eds.), Phonologica, 41-48. Innsbruck: Institut für Sprachwissenschaft. Elsness, Johan 1984 That or zero? A look at the choice of object clause connective in a corpus of American English. English Studies 65: 519-533. Fanego, Teresa 1996a The development of gerunds as objects of subject-control verbs in English (1400-1760). Diachronica 13: 29-62. 1996b On the historical development of English retrospective verbs. Neu-philologische Mitteilungen 97: 71-79. Givon, Talmy 1991 Markedness in grammar: Distributional, communicative and cognitive correlates of syntactic structure. Studies in Language 15: 335-370. 1992 The grammar of referential coherence as mental processing instructions. Linguistics 30: 5-55. Gundel, Jeannette 1996 Relevance theory meets the givenness hierarchy: An account of inferrables. In: Thorstein Fretheim and Jeannette K. Gundel (eds.), Reference and Referent Accessibility, 141-153. Amsterdam/Philadelphia: Benjamins. Gundel, Jeannette K., Nancy Hedberg, Ron Zacharski 1993 Cognitive status and the form of referring expressions in discourse. Language 69: 274-307. Gramley, Stephan 1980 Infinitive and gerund complements with the verbs begin and start. Arbeiten aus Anglistik und Amerikanistik 5: 159-186. Hawkins, John Α. 1986 A Comparative Typology of English and German: Unifying the Contrasts. London/Sydney: Croom Helm. 1990 A parsing theory of word order universals. Linguistic Inquiry 21: 223-261. 1999 Processing complexity and filler-gap dependencies across grammars. Language 75: 244-285. this vol. Why are zero-marked phrases close to their heads? Hinderling, Robert 1998 Huckepackformen in der deutschen Gegenwartssprache. In: Karin Donhauser and Ludwig Μ. Eichinger (eds.), Thema in Variationen: Festschrift für Hans-Werner Eroms zum 60. Geburtstag, 223-231. Heidelberg: Winter. Hopper, Paul J. and Sandra A. Thompson 1980 Transitivity in grammar and discourse. Language 56: 251-299. Huddleston, Rodney D. 1971 The Sentence in Written English. Cambridge: Cambridge University Press.

Cognitive complexity and horror aequi

247

Kemp, William 1979 On that that that became that which which became what. In: Paul R. Clyne, William F. Hanks and Carol L. Hofbauer (eds.), Papers from the Fifteenth Regional Meeting, 185-196. Chicago: Chicago Linguistic Society. Kjellmer, Göran 1980 Accustomed to swim: Accustomed to swimming. In: Jens Allwood and Magnus Ljung (eds.), ALVAR: A Linguistically Varied Assortment of Readings. Studies Presented to Alvar Ellegärd on the Occasion of his 60th Birthday, 75-99. (Stockholm Papers in English Language and Literature 1.) Stockholm: University of Stockholm. 1985 Help to/help 0 revisited. English Studies 66: 156-161. Lind, Age 1983a The variant forms of help to/help 0. English Studies 64: 263-273. 1983b The variant forms of try andltry to. English Studies 64: 550-563. Lüdtke, Jens 1984 Sprache und Interpretation. Semantik und Syntax reflexiver Strukturen im Französischen. Tübingen: Narr. Mair, Christian 2002 Three changing patterns of verb complementation in Late Modern English: A real-time study based on matching text corpora. English Language and Linguistics 6: 105-131. this vol. Gerundial complements after begin and start: Grammatical and sociolinguistic factors, and how they work against each other. McCawley, James D. 1988 The Syntactic Phenomena of English, Volume 1. Chicago/London: The University of Chicago Press. Menn, Lise and Brian MacWhinney 1984 The repeated morph constraint: Toward an explanation. Language 60: 519541. Mondorf, Britta this vol. Support for /nore-support. Montgomery, Michael 1989 Choosing between that and it. In: Ralph Fasold and Deborah Schiffrin (eds.), Language Change and Variation, 241-254. Amsterdam: Benjamins. Plag, Ingo 1998 Morphological haplology in a constraint-based morpho-phonology. In: Wolfgang Kehrein and Richard Wiese (eds.), Phonology and Morphology in the Germanic Languages, 199-215. Tübingen: Niemeyer. Plank, Frans 1981 Morphologische (Ir-)Regularitäten. Aspekte der Wortstrukturtheorie. Tübingen: Narr, Poussa, Patricia 1988 The relative WHAT: Two kinds of evidence. In: Jacek Fisiak (ed.), Historical Dialectology, Regional and Social, 443—474. Berlin: Mouton de Gruyter. Pullum, Geoffrey and Arnold Zwicky 1998 Gerund participles and head-complement inflection conditions. In: Peter Collins and David Lee (eds.), The Clause in English: In Honour of Rodney Huddleston, 251-271. Amsterdam/Philadelphia: Benjamins.

248

Günter Rohdenburg

Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech and Jan Svartvik 1972 A Grammar of Contemporary English. London: Longman. Rauh, Gisa 1996 Zur Struktur von Präpositionalphrasen im Englischen. Zeitschrift für Sprachwissenschaft 15:178-230. 2002 Prepositions, features, and projections. In: Hubert Cuyckens and Günter Radden (eds.), Perspectives on Prepositions, 3-23. Tübingen: Niemeyer. Rohdenburg, Günter 1992 Bemerkungen zu infiniten Konstruktionen im Englischen und Deutschen. In: Christian Mair and Manfred Markus (eds.), New Departures in Contrastive Linguistics/Neue Ansätze in der Kontrastiven Linguistik, Volume 1, 187— 207. Universität Innsbruck: Innsbrucker Beiträge zur Kulturwissenschaft. 1995a Betrachtungen zum Auf- und Abstieg präpositionaler Konstruktionen im Englischen. North-Western Language Evolution 26: 67-124. 1995b On the replacement of finite complement clauses by infinitives in English. English Studies 16: 367-388. 1996 Cognitive complexity and increased grammatical explicitness in English. Cognitive Linguistics 7: 149-182. 1998a Subordinate clauses introduced by interpretative verbs in English and their less explicit counterparts in German. In: Wolfgang Börner and Klaus Vogel (eds.), Kontrast und Äquivalenz. Beiträge zu Sprachvergleich und Übersetzung, 233-249. Tübingen: Narr. 1998b Syntactic complexity and the variable use of to be in 16th to 18th century English. Arbeiten aus Anglistik und Amerikanistik 23: 199-228. 1998c Clarifying structural relationships in cases of increased complexity in English. In: Rainer Schulze (ed.), Making Meaningful Choices in English, 189205. Tübingen: Narr. 1999 Clausal complementation and cognitive complexity in English. In: FritzWilhelm Neumann and Sabine Schülting (eds.), Anglistentag Erfurt 1998, 101-112. Trier: Wissenschaftlicher Verlag. 2000 The complexity principle as a factor determining grammatical variation and change in English. In: Ingo Plag and Klaus P. Schneider (eds.), Language Use, Language Acquisition and Language History: (Mostly) Empirical Studies in Honour of Rüdiger Zimmermann, 25-44. Trier: Wissenschaftlicher Verlag. 2002a Processing complexity and the variable use of prepositions in English. In: Hubert Cuyckens and Günter Radden (eds.), Perspectives on Prepositions, 79-100. Tübingen: Niemeyer. 2002b Aspects of grammatical iconicity in English. In: Wolfgang G. Müller and Olga Fischer (eds.), From Sign to Signing, 265-287. (Iconicity in Language and Literature 3.) Amsterdam/Philadelphia: Benjamins. Rohdenburg, Günter and Julia Schlüter 2000 Determinanten grammatischer Variation im Früh- und Spätneuenglischen. Sprachwissenschaft 25: 443-496. Ross, John Robert 1972 Doubl-ing. Linguistic Inquiry 3: 61-86. 1974 Three batons for cognitive psychology. In: Walter B. Weimer and David S. Palermo (eds.), Cognition and the Symbolic Processes, 63-124. Hillsdale, N. J.: Lawrence Erlbaum.

Cognitive complexity and horror aequi

249

Rudanko, Juhani 1998 7o-infinitive and to -ing complements: A look at some matrix verbs in Late Modern English and later. English Studies 79: 336-348. 2000 Corpora and Complementation: Tracing Sentential Complementation Patterns of Nouns, Adjectives and Verbs over the Last Three Centuries. Lanham: University Press of America. Seppänen, Aimo 1997 Sentence stress and unstressable words. Studia Neophilologica 69: 43-53. Stemberger, Joseph Paul 1981 Morphological haplology. Language 57: 791-817. Tang, Sze-Wing 2000 Identity avoidance and constraint interaction: The case of Cantonese. Linguistics 38: 33-61. Vosberg, Uwe this vol. The role of extractions and horror aequi in the evolution of -ing complements in Modern English. Wasserman, Robert David 1976 Theories of linguistic variation. Unpublished Ph.D. dissertation, Indiana University. Wasow, Thomas 1997 Remarks on grammatical weight. Language Variation and Change 9: 81105. Wasow, Thomas and Jennifer Arnold this vol. Post-verbal constituent ordering in English. Westney, Paul 1992 Revisiting -ing -ing. In: Rosemarie Tracy (ed.), Who Climbs the Grammar Tree?, 493-503. Tübingen: Niemeyer. Wurzel, Wolfgang 1998 On markedness. Theoretical Linguistics 24: 53-71. Yip, Moira 1988 The obligatory contour principle and phonological rules: A loss of identity. Linguistic Inquiry 19: 65-100. 1998 Identity avoidance in phonology and morphology. In: Steve G. Lapointe, Diane K. Brentari and Patrick M. Farrell (eds.), Morphology and Its Relation to Syntax and Phonology, 216-246. Stanford: Center for the Study of Language and Information. Yule, George 1981 New, current and displaced entity reference. Lingua 55: 41-52.

Support for more-support* Britta Mondorf

Abstract Empirical data from computerized corpora buttress the claim that cognitively complex environments favour the analytic comparative (more proud) over its synthetic variant {prouder). The /nore-variant is better suited to environments of increased processing complexity - presumably owing to the greater explicitness produced by the additional lexeme, the more transparent one-to-one relation between form and function and possibly because more introducing a Degree Phrase can serve as a structural signal foreshadowing cognitive complexity. Once the competing and synergetic effects of several potentially interacting determinants have been accounted for, an in-depth treatment of argument complexity reveals that the underlying force pertaining to all determinants that invoke the analytic comparative is to mitigate increased processing demands - a strategy referred to as more-support. A bird's eye view of 21 determinants from all core levels of linguistic analysis illustrates that the different degrees of processing effort mirrored in comparative alternation emanate from structures that are phonologically, morphologically, syntactically, semantically and pragmatically complex.

1. Introduction The choice between the synthetic comparative form {prouder) and its analytic variant (more proud) is subject to forces of an intricately intertwined network of determinants. As the extent of variability in this area appears to have been underestimated in the past, the present paper sets out to provide a quantitative and qualitative in-depth account of the multifarious forces that continually and systematically shape and design this area of morphosyntax. In deciding between the two structural alternatives language users will be shown to aim at a trade-off between concerns of explicitness and economy. Section 1 provides an outline of the research question explored and introduces the structure of the article. The notion of more-support is introduced and defined in section 2. Following that, the methodological prerequisites and the database are outlined. A quantitative and qualitative indepth account of a hitherto neglected syntactic determinant is provided in section 3: the effect of the presence or absence of infinitival complements. In the course of isolating effects of argument complexity, several

252

Britta Mondorf

interacting determinants require weeding out. Some of these have a long tradition of being mentioned in the literature, and the present paper sheds new light on their actual effects. Section 4 is devoted to a weighted comparison of several determinants, which provides information on the relative strengths of the factors involved. It reconsiders earlier treatments of the complexity of to-infinitives, and illustrates the explanatory potential of argument complexity for hitherto unexplained findings reported in the literature. Section 5 presents a bird's eye view of the influence exerted by 21 determinants from all core levels of linguistic analysis and discusses to what extent more-support relates to processing complexity. In addition, it introduces a range of novel and hitherto neglected determinants and offers new and theoretically unaligned explanations for the (presumably universal) forces underlying the systematic variation encountered in comparative alternation. Finally, section 6 puts forward explanations for the link between argument complexity and more-support, and relates the results to ongoing discussions of language processing, cognition and linguistic theorybuilding. 2. more-support Rohdenburg's (1996: 151) complexity principle states that: Complexity principle: In the case of more or less explicit grammatical options, the more explicit one(s) will tend to be favored in cognitively more complex environments. I want to argue that the analytic comparative variant {more) is resorted to, whenever a structure requires more processing capacity, be it for matters of phonology, morphology, lexicon, syntax, semantics or pragmatics. Language users, when faced with the option between the synthetic and analytic variant, prefer the latter in environments that are for some reason more difficult, more complex, less entrenched, less frequent, less accessible or in any way cognitively more complex. The mechanism by which the analytic variant apparently serves to mitigate complexity effects will be subsumed under the notion of more-support. more-support: In cognitively more demanding environments which require an increased processing load, language users tend to make up for the additional effort by resorting to the analytic {more) rather than the synthetic (-er) comparative.

Support for more-support

253

This compensatory strategy can most likely be extended to other kinds of variation phenomena that draw on the synthetic-analytic distinction. As more-support merely represents one instance of a presumably universal tendency, the phenomenon might even more appropriately be defined as periphrasis-support or analytic support. Analytic support: In cognitively more demanding environments which require an increased processing load, language users when faced with the option between a synthetic and analytic variant - tend to compensate for the additional effort by resorting to the analytic form. For instance, the analytic o/-genitive is frequently recruited in the presence of syntactically complex possessors in environments which otherwise favour the synthetic s-genitive (for a discussion see Rosenbach, this volume). Likewise, certain tendencies described for do-support may be attributed to complexity issues (cf. Stein 1985). As the present investigation focuses on comparative alternation, the notion of more-support will be preferred throughout.

3. The effect of infinitival complements As early as 1914, Poutsma pointed out that adjectives taking a complement tend to favour the analytic comparative. "It is chiefly a matter of convenience that periphrastic comparison is preferred of relative adjectives, i.e. such as make complete sense only when followed by some adjunct"1 (Poutsma 1914: 477). Similarly, Jespersen related argument complexity to the use of the analytic variant by stating that "... periphrasis seems commoner than comparison with -er, -est... when full is followed by a prepositional phrase" (Jespersen [1909] 1956: 348). Previous research has pointed towards a marked positive correlation between the complexity of the syntactic environment expressed by prepositional complements and the use of the analytic variant (cf. Mondorf 2002). Although at this stage there is no independent empirical evidence that the analytic variant serves as a signal foreshadowing complex structures, is easier to process or in other ways more suited to complex environments, the results are highly suggestive. If increased complexity or higher processing demands are assumed to cause an incremental use of the analytic variant, an analogous effect should be observable for infinitival complements.

254

Britta Mondorf

The present section sets out to show that increased complexity as effected by the presence of a complement raises an adjective's proclivity towards the analytic comparative. In the case of infinitival complements, the effect of more-support is threefold: 1) The analytic variant has the advantage of unambiguously signalling at the beginning of the Degree Phrase that a comparative follows. Thus phrase structure is easily identifiable. 2) The analytic variant is more explicit and probably easier to parse since it disentangles a complex lexeme consisting of a base plus inflectional suffix by assigning each function a separate form. 3) Simply by choosing the analytic variant as a signal, a language user can alert the addressee to the fact that a cognitively complex Adjective Phrase follows, so that some extra processing capacity can be allotted to that phrase. If syntactic complexity is responsible for the fact that comparatives taking complements favour the analytic variant, constituent structure might play a crucial role. A complement attached to a predicative adjective increases the length of the Degree Phrase considerably and renders its structure more complex. Using a separate lexeme as degree marker rather than an inflectional suffix can serve both as an unambiguous signal indicating increased processing load to the reader and as a less condensed and more explicit way of structuring a complex phrase. Both strategies are suited to facilitate parsing. As regards constituency in a Principles and Parameters framework, both comparative variants are fairly similar. The crucial difference lies in the placement of the adjective and the existence of a filler-gap dependency in the synthetic construction (cf. Hawkins 1999: 246). In the following tree diagram the adjective of the analytic comparative remains in situ. In the synthetic form the adjective is taken to move before the degree marker -er, thereby leaving a trace in the AP. The filler precedes the gap when synthetic comparatives are formed, whereas a filler-gap structure is altogether avoided in the case of analytic comparatives. This avoidance is accomplished by wore-support. If we assume a top-down parser, in terms of linear processing a Degree Phrase introduced by the degree marker more can facilitate recognition of the DegP. After all, it has been shown that "Some orderings reduce the constituent recognition domains for their containing phrases, and it is systematically these orders that are preferred in performance when languages

Support for more-support

255

have a choice ..." (Hawkins 1999: 251-252). The synthetic variant in -er allows phrase structure recognition only after the ADJ and its inflection have been processed. While generally the late identification of the comparative marker is unlikely to impede parsing, complex environments should call for early recognition, in this case by means of /«ore-support.

Analytic Comparative

Synthetic Comparative

DegP

AP

Deg

ADJ

more

DegP

ready

AP

Deg

IP

Δ

to do

ADJ

ready, -er

tj

IP

Δ

to do

A related difference can be described in terms of the preference of head adjacent orderings as laid down in Greenberg (1966). In the analytic comparative the head of the AP {ready) is adjacent to the head of the IP {to). Merely two words, i.e. ready and to have to be read before the immediate constituents of the AP can be recognized. In the synthetic comparative both heads are non-adjacent, separated by the degree marker -er.

(1)

a. more ready to do b. readier to do

OtgV[more ^[ready

jp[to do\\\ DegpUp[ready -er] IP[/o do]\

The following section introduces empirical support for the effect of the presence or absence of infinitival complements on the choice of comparative forms. It will be argued that the presence of an infinitival complement significantly increases an adjective's proclivity towards the analytic variant. In order to be able to attribute predispositions towards the analytic variant to syntactic complexity effects triggered by infinitival complements, the competing or synergetic effects of the following determinants will have to be accounted for: length (section 3.2), final segment (section 3.3), frequency (section 3.4), position (section 3.5) and end-weight (section 3.7).

256

Britta Mondorf

3.1. Methodology Using the concordance software Wordsmith, all analytic comparative forms that directly preceded a to-infinitive were retrieved in a pilot corpus (described below) by using the search string "more * to". The analogous procedure for synthetic comparatives, i.e. "*er to" rendered too many hits for manual editing. Therefore, only those adjectives which occurred at least once in the analytic form were subjected to a case by case retrieval in the synthetic form. Since the British National Corpus (BNC) software Sara does not allow search operations with initial wildcards, the procedure for this corpus had to be modified in such a fashion that each adjective that occurred at least once in both variants in the pilot corpus was searched for individually in the BNC. In order to maximize corpus size, the pilot corpus was then extended so as to encompass several additional years of newspapers. Each of the adjectives that occurred at least once in both comparative forms in the pilot corpus was checked individually in the extended corpus. Unless indicated otherwise, the results documented are based on the extended corpus. 2

Table 1. The corpora Pilot corpus

Mio Extended corpus Words

Mio Words

British National Corpus (BNC)

100

British National Corpus (BNC)

100

Guardian 1990-94 (incl. Observer 1994)

141

Guardian 1990-97 (incl. Observer 1994)

260

Daily Mail and Mail on Sunday 1993-97

106

Daily Mail and Mail on Sunday 1993-94

38

Daily Telegraph and Sunday Telegraph 1991-94

128

Daily Telegraph and Sunday Telegraph 1991-94

128

Times and Sunday Times 1990-94

192

Times and Sunday Times 1990-97

320

Total Number of Words

599

Total Number of Words

914

During the editing process, all concordance results that were not comparative forms of mono- or disyllabic adjectives accompanied by a /o-infinitive were manually eliminated from the tally, as well as the following constructions: Substantival uses of the comparative where the adjective was the head of an NP, correlative constructions and sentence adverbials as e.g. Perhaps the mood is more fragile to start with, ... (Times 1993). This procedure rendered 33 adjectives which occurred with a /o-infinitive in both

Support for more-support

257

the synthetic and the analytic form. Among these, five extremely frequent ones (easy, hard, happy, likely, simple) had to be temporally discarded from further analyses in order to work with a manageable set of data3. The adjectives selected for further analysis are: 12 monosyllables: apt, cool, cruel, fit, free, hard, keen, proud, rare, right, sound, sure, true. 16 disyllables: clever, costly, cosy, crazy, easy, handy, happy, hungry, likely, lucky, pleasant, polite, ready, risky, sexy, silly, simple, subtle, trendy, tricky. 3.2. Isolating length effects The best-known determinant in the discussion of comparative alternation is length measured in terms of the number of syllables. Traditionally, monosyllabic adjectives are assumed to take the synthetic form. Trisyllabic adjectives are taken to be formed analytically with disyllabic adjectives being subject to variation. There are, however, four aspects which attest that the picture is far more complicated than is suggested by this rule of thumb. Firstly, the crucial parameter is not the length of the adjective's positive, which is ubiquitously referred to in the literature, but the prospective length in the synthetic form. Thus, trisyllabic adjectives for which the addition of the comparative suffix -er causes elision of a syllable, e.g. sensible /sensibsi/- sensibler /sensibls/, are acceptable in the inflected form. Secondly, an exception to the general rule of thumb concerning length is presented by adjectives with the initial prefix un- in e.g. unfriendlier (cf. Quirk et al. 1985: 462). Thirdly, depending on the assignment of lexical status, certain compounds (e.g. environment-friendlier) are able to take the synthetic comparative thereby exceeding the usual length requirements (cf. Mondorf 2000). And finally, though being a powerful determinant, the adjective's length is by no means the only or even the most decisive factor operating in comparative variation of mono-, di- and trisyllabic adjectives. Length can occasionally be overridden by more salient factors, such as argument complexity (cf. section 3.6). A comparison of the 28 adjectives singled out for further analysis in the present section does not invite the formulation of general tendencies, even if the number of syllables is kept constant. In Figure 1 the light sections of the columns provide the percentages of the synthetic comparative in -er. The darker sections give the corresponding figures for the analytic variant more. In order to indicate whether the sample is sufficiently large to allow

258

Britta Mondorf

meaningful comparisons, each column is labelled with the total number of synthetic plus analytic forms. Thus, the figure reads as follows: Apter occurs in 8%, more apt in 92% of all cases - and these percentages refer to an overall occurrence of 396 instances of both variants. 396

2271

424

2569

1777

1691 342

1632

125

579

664

1562

100% 90%

£ \

60%

1

50%

g

40%

u°

30%

I

I

80% 70%

20% 10% 0%

οο ο

υυ «

ID

C υ ω

• Synthetic (%)

-Ο J3 3 00 Ο ο. I Analytic (%)

χ) ο

3 ζ/1

υ

S

Figure 1. Synthetic vs. analytic comparatives of monosyllabic adjectives in the extended corpus ( N s y n t h e t i c + A n a l y t i c = 14032)

100%-! 90%80% s N 0® ω > ta c3 CL, 6ο U

1068 892 288 197 65 375 759 782 312 755 1329 709 430 3246 287 1035

m i Mil

II

70%60%50%40%30%20%10%0%

< >u υ

!Λ

ου

S feiff 33

•σ λ

.-ä οCX, —;

u

-g c3 u u.

-ST tn "d^

X 4» ">

S 73 Ή — c o m« j5 3 S -c Μ .fcS +•·

"CL.

• Synthetic (%

I Analytic (%)

Figure 2. Synthetic vs. analytic comparatives of disyllabic adjectives in the extended corpus (Nsynthetic+Analytic = 9644)

Support for more-support

259

Figure 1 demonstrates that for the 12 adjectives investigated the number of syllables is by no means the primary determinant operative in the choice of comparative form. Though monosyllabic adjectives are generally regarded as taking the synthetic comparative, apt and right predominantly use the analytic form. Cruel, proud, sure and true still take the analytic comparative to a considerable extent.4 As Figure 2 illustrates, a similar situation is encountered for the disyllabic adjectives investigated. The amplitude of percentages for the analytic form ranges from 78% for pleasant to 3% for lucky. It becomes apparent that for the 28 adjectives investigated, the number of syllables is not the determining factor for the choice of comparative form.

3.3. Isolating the effect of different final segments The literature is replete with comments on the correlation between final segment and comparative form (cf. Kytö and Romaine 1997: 336 for a diachronic approach). It is well-known that certain suffixes defy the addition of the -er inflection altogether, e.g. -al, -ish. Other suffixes are reported to trigger -er to a considerable extent, e.g. -y, -ly (cf. Leech and Culpeper 1997: 358). However, such quantitative statements are rarely, if ever, accompanied by qualitative analyses and hence devoid of explanatory power. After all, why should adjectives in , such as clever and mature, favour the synthetic variant - as has been claimed by Quirk et al. (1985: 462)? Later sections will reveal that for mature this claim does not stand up to scrutiny. Similarly, the positive correlation between synthetic comparative and disyllables in postulated in Leech and Culpeper (1997) remains unexplained. In order to arrive at an explanation for the diverging percentages among the 28 adjectives, the effect of an adjective's final segment has been quantified. Evidently, even adjectives ending in exhibit a large amount of variation. Figure 2 reveals that the group of disyllabic adjectives in is relatively heterogeneous with respect to comparative form. While ready takes the analytic comparative in the majority of cases (56%), silly (7%) and lucky (3%) strongly disprefer this variant. Thus, even for adjectives which are identical in terms of both length and final segment, widely divergent percentages for the synthetic and analytic comparative forms can be observed. Unless explanations in terms of origin, phonology or - most unlikely grapheme type are intended, we are well-advised to treat the adjective's

260

Britta Mondorf

final segment merely as a formal categorization parameter rather than elevating it to the status of an explanation. After all, apart from well-known knock-outs, final segment is presumably often a corollary of other primary factors, such as requirements of morphological complexity (see section 5.2) and avoidance of identity effects (cf. sections 5.1.1 and 5.1.2).

3.4. Isolating frequency effects Whereas the relation between frequency and the synthetic comparative has been occasionally mentioned (cf. Sweet [1891] 1968: 327; Bolinger 1968: 120; Quirk et al. 1985: 463), empirical validation of this claim has hitherto only been provided by Braun (1982: 101). All investigators agree that frequently used adjectives, which are typically also the shorter ones (cf. Zipf 1929), tend to favour the -er variant, while rarer adjectives exhibit a slight preference for the more-variant. There is even an inherent connection between frequency and the -er inflection: "... inflectional morphemes have their origins in full words that develop a high frequency of use. These frequent items are gradually reduced both phonologically and semantically, and are simultaneously gradually fused, again phonologically and semantically, with lexical matter contiguous in the syntactic string" (Bybee 1984: 29). Fast access to morphological structures is affordable for well-entrenched lexemes. Their retrieval from the lexicon, or their on-line construction by means of adding -er, implies less processing effort. Rare lexical items, in turn, are harder to access, which is why the syntactic option with more might be preferred. Moresupport by separating form and function then serves to apportion the otherwise complex expression of adjective and degree marker in one lexeme and has the added advantage of leaving the internal structure of the lexeme unaffected and easily recognizable. As Figure 3 illustrates, the heterogeneous profiles for monosyllabic adjectives observed in section 3.2 can partly be attributed to frequency effects.5 The grey area represents the overall frequency of both the synthetic and analytic comparative forms. It is observable that those adjectives which score lowest on comparative frequency are located at the left end of the continuum, and vice versa. However, the percentages for more proud (34%) and more true (32%) are roughly the same, even though true is more than four times as frequent as proud. Still, there is a slight negative correlation between frequency and the use of the analytic variant. The less entrenched an adjective is in any of the two comparative forms, the more

Support for more-support

261

likely will it require more-support, i.e. an explicit fully-fledged lexeme to form the comparative.

«2 c3 I" ο ϋ χ; ξ

100% 90% so 80% S 70% · |

6000 5000 4000 3000

Iu 2000 §" 1000 -

— i Uh

- 60%

I

- 50% 40% - 30% 20% 10%

§ ^ f |
70% I 60% d. 50% I 40% ο - 30% 20% ^ 10% 0%

&

U VI ecx Ο a Ο "C ·Ρ "in XI 9 Analytic Comparatives %

Figure 4. Analytic comparatives of disyllabic adjectives in according to frequency of the comparative in the extended corpus (Ν^,^+α,,,,^,. = 26426)

262

Britta Mondorf

Thus, frequency is by no means one of the more salient factors. It appears to be easily overridden by other determinants, such as the number of syllables or the complexity of the syntactic environment. However, it is theoretically possible that retrieving a rare adjective, e.g. apt, might in itself be a difficult process requiring more processing effort, so that once retrieval has been accomplished, speakers seek to reduce the processing load by using the easier-to-process more variant. 3.5. Eliminating attributive uses The following quotation by Quirk et al. (1985: 420) suggests that the most obvious factor that requires weeding out in the investigation of argument complexity is position. "Adjectives with complementation ... normally cannot have attributive position but require postposition". For this reason attributives were entirely omitted from the count. 3.6. The effect of to-infinitives Figure 5 attests the effect of the presence of a to-infinitive on the choice of the analytic comparative form. Since comparatives can be either analytic or synthetic, the synthetic percentages are identical to the missing segments of the columns. Thus, for apt which has 95% analytic comparatives {more apt) when not followed by a to-infinitive, the remaining 5% must be synthetic (apter). For 9 out of the 12 monosyllabic adjectives investigated, the use of the more-variant is higher when a to-infinitive immediately follows the adjective. The three exceptions to the general pattern are free, right and true. While no explanation for the deviant behaviours of free and true is offered at this stage (but see section 3.7), the pattern for right appears to be attributable to this adjective's special status in terms of gradability. The adjective right has already received particular attention in the literature. Jespersen ([1909] 1956: 349) claimed that "right if compared at all would take more and most". Likewise, Quirk et al. (1985: 461) point out that right exclusively takes the analytic form. Leech and Culpeper (1997: 356) observe 9 analytic as opposed to 4 synthetic comparatives of right in their data. All authors attribute the predisposition towards periphrasis to semantic factors. With a ratio of 82% when no infinitive follows, the percentage for more is always high, disregarding syntactic complexity. The deviant behaviour of right is attributable to its limited gradability. If right is graded at all, it requires more-support, no matter whether a to-infinitive further complicates matters or not.

Support for more-support

263

100% 205 38 vp

90%

3>

80% —

~

70%

g_ 60% § ^ £

50% 40% 30%

g 20%


ο ο ο

υ

C ο U

I % - to-infinitive

Ό

3

Ο

ω S3

-C

00

•α c

ω I3

Ο οο

I % + to-infinitive

Figure 5. Analytic comparatives of monosyllabic adjectives in non-attributive position +/ίο-infinitive in the extended corpus (NAnalytic= 1821)

Among the 12 disyllabic adjectives ending in , four adjectives decrease their use of more when a ίο-infinitive follows and one is indifferent to the presence of a to-infinitive. For this group the presence of absence of a /o-infinitive does not appear to exert any influence on the choice of the comparative variant. 100% x

90%

o cn 80% ω > 70% αϊ Sj 60% Q. £ 50% Ο Ο υ 40%

455 34

II

% 30% C3

C
2 9 3 ( 7 3 8 % ) 0 5 NCF2 (* 1800-* 1869) (72.8%) 0(0.0«/ 6

III -ing

II to have

^ > 1 8 9 4 )

212(68.4%) 0 ( 0 . 0 ^

VI Total

In American English the evolution and distribution of non-finite complements was almost exactly the same as in British English. At the beginning of our American corpus (EAF1), the gerundial variant had already reached its peak in the evolution of non-finite complements. Now, what factors account for this temporary rise of the -ing form? Table 4. The distribution of non-finite complements and direct objects following the verb attempt in selected American historical corpora attempt

ϊ ίο

Π to have -ed

ϊϊϊ -ing

IV NP

V Total

J

> 4 33 (71.3%).

0(0.0%)

58(2.9%)

520(25.9%)

2,011

1

'108(71·4%)

1(0.1%)

24(1.5%)

419(27.0%)

1,552

0(0-0%)

4(0.4%)

186(19.7%)

943

1 FAF1 (*1744-*1799) 9 FAF9 (*1801-*1827) 3 (*1870-*1896) FTP/A

753

(79

9%

)

The answer is fairly obvious: it is the morpho-syntactic category to attempt, because the establishment of the -ing form is virtually restricted to this matrix verb form (cf. Figure 5 for British English with the dashed line representing the ideal structure to attempt + V-ing and the solid line indicating the double -ing situation; the dotted line refers to all of the remaining uses; cf. also the complementary distribution in the examples given in [9c] and [9g] both by Daniel Defoe). This is a clear manifestation of the horror

318

Uwe Vosberg

aequi principle. In all our corpora the -ing variant never occurred after the verb form attempting. The remaining morpho-syntactic categories are also virtually irrelevant to the establishment of gerundial complements. Despite having only very few (six) examples, Fanego (1997: 65; cf. also Fanego 1996a: 51), quite correctly, even goes so far as to call this a "knockout factor". However, this principle is not (only) a "stylistic constraint" (Fanego 1997: 65). Rohdenburg (1995: 381-382) argues that a sequence of two infinitives causes difficulties for both processing (recovering the understood subject) and for speech and production (serial infinitives may be misinterpreted as various hesitation phenomena). 60 t

25/47

50

Ν 54/133

40

bo Ef

11/39

\

/

.30

20 V16/103 10 0/17 4/2 1 1

1/203

0/36 Ö/35 1518-1700 * 16601699

8/380 0/79 »1700Ί752

9/998 " 0/225 * 1728' 1799

4/909

S

" 0/223 * 1800* 1869

0/28 \

l/13i 0/47 * 1870•1894

— - · - - to attempt (the) V-ing — — · — attempting (the) V-ing - - Ο - - remaining uses (the) V-ing Figure 5. The distribution of ίο-infinitives vs. -ing forms as complements of the verb attempt in various British historical corpora8* a)

ρ < 0.1% *** for each period except 1518-1700 [not significant] and *1870-*1894 [not significant]

American usage shows very similar tendencies, although the authors born in the second half of the 18th century seem to be a bit less sensitive to horror aequi than their British contemporaries (compare Figures 5 and 6).

Extractions and horror aequi in the evolution of -ing complements

319

This is again correlated with the overall entrenchment value as suggested by the remaining uses outside the horror aequi contexts. 60 50 40 54/182 30 0s NO

20

X

21/123 % Ν

10

— - · - • to attempt V-ing — · — attempting V-ing - - Ο - - remaining uses V-ing

4/1029 Ο— 0/280 *1744•1799

3/785 —-Ο-· 0/224 * 1801•1827

3/76 ^ 1/471 0/204 * 1870* 1896

Figure 6. The distribution of fo-infinitives vs. -ing forms as complements of the verb attempt in various American historical corpora

The -ing form did not represent the only means of avoiding the toinfinitive after the matrix to attempt. In British English, the -ing form reached its peak with the authors born between 1700 and 1752, and Table 5 shows a more detailed picture of the situation in this period: Table 5. The distribution of non-finite complements and direct objects following the verb attempt in the Eighteenth Century Fiction corpus 2 a) ECF2 (*1700-*1752)

I to

II

-ing

1 to attempt

22 (23.7%)

22

(23.7%)

2 attempting

79 (74.5%)

0

(0.0%)

3 0 attempt

117 (75.5%)

2

(1.3%)

2

(1.3%)

(21.9%)

A attempted

241 (83.7%)

3

(1.0%)

1

(0.3%)

(14.9%)

14 (77.8%)

0

(0.0%)

0 (0.0%)

(22.2%)

473 (71.7%)

27

(4.1%)

6

5 attempts 6 Total a)

ρ < 0.1% *** with 3-5 being collapsed

3

(3.2%) (25.5%)

(0.9%)

154

(23.3%)

660

320

Uwe Vosberg

Here, the infinitival matrix to attempt triggers an NP object in almost 50% of all cases, which is nearly twice as much as with all other verb forms. The remaining 50% of all complements after the form to attempt are realized by non-finite structures with the -ing form being even as strong as the /o-infinitive. Corpus analyses regarding many other verbs governing sentential complements have shown that one or the other non-finite complement variant attained its peak with the authors born in the first half of the 18th century (represented by the ECF2 corpus) if later - most often owing to semantic reasons - it was again replaced by one of the other two sentential rivals. In other words, the spectrum of variation in non-finite complementation was widest in this period. The verbs concerned include the following: forbear, omit, fail, refuse or prefer. Functionally, (non-sentential) NP objects do not represent a direct alternative to sentential complementation, but very often they can serve as an indirect avoidance strategy in this context as shown in example (9b). Here, the NP the translation is possibly used in order to avoid two successive infinitives {to attempt to translate).13 In Early Modern English times when the -ing form was not yet available as an established grammatical option, NP objects provided practically the only choice after the matrix to attempt, although the /o-infinitive was a traditional non-finite complement of all other morpho-syntactic categories. The findings for the 2nd part of the Nineteenth Century Fiction corpus are even more revealing from a statistical point of view but will not be presented here for the sake of brevity. However, they confirm the figures shown in Table 5 (row 1): Whenever the governing verb occurred in the marked infinitive itself, an infinitival complement was largely avoided. In that case writers particularly preferred either the -ing form or an NP object. In all our historical data, the morpho-syntactic category attempted particularly favours infinitival complements rather than NP objects. Being highly plot advancing in fictional texts, this verb form is the most frequent of all morpho-syntactic categories and at the same time the least susceptible one in terms of horror aequi. This is why this form is not forced to find alternatives other than the well-known infinitive. The morpho-syntactic category 0 attempt can, of course, represent an unmarked infinitive as part of a verb phrase involving auxiliaries (as in / will attempt ...). However, such a matrix construction - although containing an infinitive - does not particularly attract the -ing complement. If the to-infinitive is avoided at all, the alternative is practically always a nonsentential NP object. In the same way, the question of whether a matrix in -ing functions as a progressive tense form, a present participle or a gerund

Extractions and horror aequi in the evolution of -ing complements

321

has no bearing on the choice of verb complements. What counts in terms of horror aequi is merely surface identity. Finally, it should be pointed out that horror aequi can be weakened by a variety of other factors: -

Our corpus analyses have always shown that sentential complements separated from the superordinate clause by any intervening material tend to favour the finite variant (cf. Rohdenburg 1995: 376) or, in the case of non-finite complementation, the old and better-known to-infinitive rather than the newly established -ing form - regardless of whether or not the matrix form is a marked infinitive itself (cf. example [9d]). This phenomenon is most likely due to an increase in cognitive complexity provided by these discontinuous constructions (cf. the complexity principle in Rohdenburg 1995: 368; Rohdenburg 1996: 151; also Rohdenburg, this volume). - Another complexity factor which is able to overrule the horror aequi principle concerns the negation of the subordinate clause (cf. Rohdenburg 1995: 378). Complements involving not-negation are quite rare in the case of the verb attempt but again always trigger the to-infinitive. 3.2. bother The verb bother (cf. OED: s.v. bother, v., sense 3) shows very clearly that the horror aequi principle operates below the threshold of consciousness and does not only apply to a particular verb or a particular period in the history of the English language, because, contrary to the verb attempt, this verb is highly colloquial. (10) a. She contrived to sack him, ..., without bothering to tell him. (The Guardian 1994) b. "..., I suspect he just didn't want to bother reading it." (The Guardian 1994) However, selected British newspapers reveal that (as in the case of attempt in the eighteenth and nineteenth centuries) the distribution of nonfinite complements immediately following the verb bother is mainly determined by this "knockout factor": the -ing form generally being relatively weak except after the verb form to bother, where it is used in order to avoid a sequence of two infinitives.14 The few prepositional -ing complements

322

Uwe Vosberg

coming to light in our corpora also mainly occur after the infinitival matrix to bother. Table 6. The distribution of non-finite complements immediately following the verb bother in selected British newspapers for 1994 (G, T, DM, DT) a ) bother: Br. newspapers

I to

1 to bother

18(33.3%)

2 bothering

139 (99.3%)

II -ing 31(57.4%)

2 (3.7%)

IV with -ing 3 (5.6%)

V Total 54

(0.0%)

0 (0.0%)

1 (0.7%)

140

3 remaining uses 1,028 (82.6%)

204(16.4%)

5 (0.4%)

8 (0.6%)

1,245

4 Total

235 (16.3%)

7 (0.5%)

12 (0.8%)

1,439

a)

1,185 (82.3%)

0

III about -ing

For columns I, II: p topicality > possessive relation. While this hierarchy as such turns out to be unaffected by Standard variety and diachrony, a significant increase of the s-genitive with inanimate possessors can be observed as an ongoing change-in-progress phenomenon, both in British and American English (though more pronounced in the latter). Comparing the results of the Modern English data with a corpus analysis of Late Middle/Early Modern English, finally a diachronic scenario is proposed showing how iconicity and economy interact diachronically, leading to a greater preference of the s-genitive.

1. Introduction The present paper is an empirical study on genitive variation in English, i.e. on the alternation between the s-genitive (the boy's mother) and the ofgenitive {the mother of the boy), and it links to the question of what determines grammatical variation in English in a number of ways.1 The analysis focuses on the factors animacy, topicality and possessive relation. These factors allow for predictions as to why the s-genitive should be preferred in certain contexts rather than in others. In this respect, the present paper will go beyond merely descriptive studies of grammatical variation and offer a possible explanation for why the distributional pattern found is the way it is. The underlying hypothesis is that the choice of the ί-genitive is determined by iconic principles, and the predictions are derived on the grounds of general cognitive, psychological principles and psycholinguistic evidence (for such a cognitive/processing approach to grammatical variation see also e.g. Bock 1982; Hawkins 1994, this volume; Rohdenburg 1996, this volume; Wasow 1997; Arnold et al. 2000; Mondorf, this volume). I

380

Anette

Rosenbach

also consider how the factors animacy, topicality and possessive relation interact with each other when a speaker has to choose between the 5-genitive and the o/-genitive, thereby also evaluating their relative importance. Moreover, in the present study genitive variation is not only looked at from a purely synchronic angle but also includes a diachronic perspective. For this reason, two types of diachronic data are included. The main body of data presented in this paper stems from a Modern English experimental study with British and American subjects. Apart from a synchronic analysis of the experimental data also an apparent-time approach is applied, which helps to track ongoing change-in-progress. Pursuing a more long-term diachronic perspective on genitive variation, the results of the Modern English experimental study are then, in a second step, compared with the results of a Late Middle/Early Modern English corpus analysis covering the period between 1400 and 1630 (cf. Rosenbach and Vezzosi 2000; Rosenbach, Stein and Vezzosi 2000). I will show how the 5-genitive has systematically extended its range of application from Late Middle to present-day English both in terms of compatible contexts and frequency. As a possible explanation for this development, I will finally propose a diachronic scenario of how iconicity and economy may have interacted leading to a greater preference of the ^-genitive.

2. Some structural and methodological preliminaries Before turning to the analysis proper some structural and methodological preliminaries are in order. In the following I will identify those cases in which there is in fact a real choice between the 5-genitive and the of-genitive and in which the choice between these two constructions represents a proper instance of grammatical variation. I will begin by pointing out any relevant morphosyntactic differences, and then move on to the question in which contexts they can be matched in a quantitative analysis. 2.1. Morphosyntactic

differences

The morphosyntactic differences between the 5-genitive and the of-genitive can be summarized as follows (cf. Table 1). First, there are two different relational markers linking the two elements of a possessive construction, which I will call here possessor and possessum: (1) the possessive 's in the 5-genitive (today regarded as a clitic rather than the Old English inflectional case ending), and (2) the preposition of in the o/-genitive. Therefore, the s-

Iconicity and economy in the choice between the two genitives

381

genitive can be regarded as more synthetic or more bound than the ofgenitive. Table 1. Morphosyntactic differences between the s-genitive and the o/-genitive Morphosyntactic differences

s-genitive the man's head

of-genitive the head of the man

Relational marker

POSS 's

preposition of

more synthetic

Grammatical function of determiner: possessor [the/a king] 's daughter modifier (I): a [king's daughter]

Word order

possessor-possessum

more analytic

complement: the daughter of the/a king modifier (I): the daughter of a king modifier (II): a king of honour possessum-possessor

Second, the linear order of possessor and possessum is exactly complementary: while in the s-genitive the possessor precedes the possessum, it follows it in the o/-genitive. And third, for both the s-genitive and the o/-genitive the same surface form can encode various grammatical functions. In the example [the king] 's daughter the possessor functions as a determiner (e.g. Huddleston 1984: 233; Quirk et al. 1985: 326), while in the corresponding o/-genitive (the daughter of the king) it is generally analyzed as a complement (e.g. Huddleston 1984: 262). In both the s-genitive and the o/-genitive the possessor can, however, also serve as a modifier (e.g. a [king's daughter]!the daughter of a king), specifying not the referent of the possessum but rather its denotational class, thereby answering the question what type of daughter rather than whose daughter it is (cf. Huddleston 1984: 258; Quirk et al. 1985: §5.122). There is yet a second type of modification occurring only with the of-genitive and never with the sgenitive, as in a king of honour, where the possessor describes a property of the possessum and where there is no corresponding prenominal nominal modifier (cf. also Huddleston 1984: 262).2 2.2. Delimiting the range of variation 2.2.1. Categorical versus choice contexts Not every s-genitive can be expressed by an o/-genitive, and vice versa. The identification - and subsequent exclusion - of those contexts in which

382

Anette

Rosenbach

there is no choice (= categorical contexts) is therefore a crucial precondition for any quantitative analysis comparing the frequency of the two genitive constructions.3

choice contexts

categorical contexts

categorical contexts

Figure 1. Categorical versus choice contexts

Table 2 gives a brief overview of those contexts which are at least in principle compatible with both genitive constructions (= choice contexts). Any other contexts will be excluded from consideration in this study. Table 2. Choice contexts for genitive variation Genitive function/meaning Noun class Definiteness Grammatical function of possessor

possessive possessor = full lexical NP whole genitive NP = [+definite] no reference tracking devices for whole NP no modification (II)

First, pronominal possessors are usually realized as possessive pronouns and not by an of-genitive (my house vs. *the house of me).4 Therefore, our analysis should focus on full-noun-phrase possessors only. Second, partitive genitives, such as some of my students, where the possessum narrows down the referent of the possessive NP, can only be realized by the ofgenitive and never by the s-genitive. In principle, the analysis should therefore be restricted to possessive relations (in a broad sense; I will come back to this notion in §3.2. below).5 Third, since the possessor in the 5-genitive already occupies the determiner slot rendering the whole possessive construction definite,6 only such possessive constructions may be compared in which the whole possessive construction is definite and where no other

Iconicity and economy in the choice between the two genitives

383

reference tracking devices are needed. Note, finally, that modification of the type as in a king of honour is, as mentioned above, a categorical context for the o/-genitive only; a further restriction as to the grammatical function of the possessor will be made below. 2.2.2. Comparable versus non-comparable contexts within choice contexts So far, it has been shown that a quantitative analysis of genitive variation as a case of grammatical variation should be restricted to choice contexts. In addition, bearing in mind that even within such choice contexts there probably can never be complete synonymy between two alternating constructions (a view expressed in the principle of isomorphism [Haiman 1985], or see also Bolinger [1977: 1]), there appear to be further factors which influence the likelihood of one construction or the other, which has the following implications for our empirical analysis: either we explicitly investigate those contexts as factors, or, otherwise, we need to keep their effects controlled. In the following I will briefly outline the factors which have been controlled for in this study; the factors investigated in this study, i.e. animacy, topicality and possessive relation, will be introduced and operationalized in §3 below. Table 3. Comparable versus non-comparable contexts within choice contexts Factor

Comparable contexts within choice contexts

Grammatical function of possessor

possessor = determiner (in s-genitive) possessor = complement (in o/-genitive)

Phonological

possessor not ending in /s/, /ζ/, /Θ/

Morphological

singular possessor noun

Syntactic

non-complex, non-branching possessor and possessum non-consecutive genitive constructions

Socio-stylistic

data must be either balanced or controlled for style

First, although modification in principle permits both constructions (a driver's seat versus the seat of a driver), there appear to be other potential adnominal variants here, i.e. a compound {the driver seat) or other prepositional constructions (a seat for drivers/a driver), which may obscure the quantitative analysis. In addition, there are genitive constructions, which although they may be subject to similar restrictions as the s-genitive/o/genitive (cf. Biber et al. 1999: §4.6.1 (Μ.6.14) - are not structurally

384

Anette

Rosenbach

equivalent to them, such as double genitives {a friend of John 's) or elliptic genitives {Let's meet at John 's). In this study I will therefore focus solely on determiner s-genitives and complement o/-genitives, as outlined in Table 1 above. Second, possessors ending in /s/, /z/ or /Θ/ are not included in this study as it has been shown that the s-genitive is generally avoided in this phonetic environment, either resulting in the alternative o/-genitive or a zero-genitive {Jones' house), cf. Altenberg (1982: §2.5); Quirk et al. (1985: §5.114). Likewise, plural possessors are not dealt with either. Besides the phonological property of regular plurals ending in {S} there seems to be also evidence pointing to a general avoidance of s-genitives with plural possessors even if not purely phonologically conditioned (i.e. with irregular plurals), cf. Jahr Sorheim (1980: 113). Third, the choice of genitive construction seems to be sensitive to the branching direction and weight of the possessor and the possessum. While the s-genitive has been shown (cf. Altenberg 1982: §3; Jucker 1993: §5) to become increasingly infrequent with right-branching possessor nouns (e.g. the man who lives next door's cat),1 it is generally said to increase with a right-branching possessum noun. As is, however, shown by the frequency scores given in Biber et al. (1999: 304-305), the syntactic complexity of the possessor - apart from the branching direction involved - seems to play a crucial role as well, with the s-genitive being still frequently used with simple premodification. It is only with increasingly complex possessors that the ^-genitive becomes more and more unlikely to occur at all. In the present study, both the possessor and the possessum may therefore be premodified by one word; syntactically more complex constructions are excluded. Moreover, recursive genitive constructions (e.g. Annie's husband's sister's trousers), generally regarded as stylistically odd (cf. Zachrisson 1920; Quirk et al. 1985: §17.118), will not be considered here.8 Fourth, the s-genitive has also been shown to be highly sensitive to socio-stylistic factors, such as text type, formality and standard variety, with the 5-genitive being more frequent in informal text types (cf. Altenberg 1982: §6; Jucker 1993: §7), particularly journalistic language (cf. Jahr Sorheim 1980: §3.5.2) and news (cf. Biber et al. 1999: 302), and it is used more frequently in American than in British English (cf. Jahr Sorheim 1980; Hundt 1998: §3.3). While the present experimental study is neutralized as to the factors of text type and formality by using text passages taken from novels sharing the same formal level in the questionnaire elicitation, I will address the question of which standard variety is the more progressive, British or American English, in the analysis in §4 below.

Iconicity and economy in the choice between the two genitives

385

3. Iconic motivation for the choice of s-genitive: factors animacy, topicality and possessive relation Previous empirical studies (cf. Jahr Sorheim 1980; Altenberg 1982; Jucker 1993; Leech, Francis and Xu 1994; Raab-Fischer 1995; Anschutz 1997; Hundt 1998: §3.3) as well as grammars of English (Quirk et al. 1985: §§5.115-5.118, 17.45; Biber et al. 1999: §4.6.12) have already shown that animacy, topicality and the type of possessive relation play some role in the choice of genitive construction (see also Stefanowitsch, this volume). As is well known, these three factors often go hand in hand: only humans typically possess things and topics are usually animate (cf. Yamamoto 1999: 60-67); the correlation between animacy and topicality is so strong that in Taylor's (1996) cognitive account of the English s-genitive the effect of animacy is in fact subsumed under a superordinate concept of topicality. There is also cross-linguistic evidence for possessor-splits induced by animacy and topicality. That is, if a language has two constructions for expressing possessive relations, it is both the animacy and the topicality of the possessor which determines which construction will be used, cf. Koptjevskaja-Tamm (2001, forthcoming) for presenting evidence for such possessor-splits from some European languages, and Stiebels (2000) for evidence from Guerrero-Nahuatl. In addition to showing that the effects of the individual factors can be separated from each other I shall also try to provide a possible explanation as to why it is the s-genitive that is favoured over the o/-genitive with animate and topical possessors and with more prototypical possessive relations. For this reason, the three factors will be linked to iconic/natural principles, which allow us to predict why we should expect to find more s-genitives in some contexts than in others, and which may, possibly, also account for the diachronic development of the s-genitive, as I will argue in §5 below. In this study I will refer to a more indirect notion of iconicity, proceeding from a resemblance between the relation of linguistic signs and their referents (i.e. a notion of diagrammatic iconicity) rather than a direct resemblance between the two,9 and assuming that it is the way language is conceptualized and processed rather than properties of the world itself that can determine linguistic structure.10 In the latter sense iconicity as used in this study is closely tied to the notion of naturalness. In approaches to naturalness (cf. Dressler et al. 1987) it is generally assumed that language is to a large extent - although not exclusively so - determined by the way it is conceptualized and processed, and what is easier to process for the human brain is usually regarded as more natural. If we apply now such an

386

Anette

Rosenbach

understanding of iconicity/naturalness to grammatical variation, we may, very generally, predict that given two alternative constructions the more iconic/natural should be preferred. In the following I will invoke two iconic/natural principles to predict the distribution of the s-genitive: (1) a word order principle, which will account for the complementary distribution of the possessor and the possessum in the s-genitive and the o/-genitive due to the factors animacy and topicality, and (2) the principle of conceptual distance (cf. Haiman 1985), which, based on the different types of possessive relation holding between the possessor and the possessum, accounts for the different degree of bondedness between the s-genitive (= more bonded) and the o/-genitive (= less bonded). 3.1. Serialization ofpossessor and possessum: animacy and topicality Animacy as used in this study refers to the distinction between a human possessor and a genuine inanimate possessor and is operationalized as illustrated in Table 4 below. Table 4. Animacy of possessor: operational definition [+animate] possessor

[-animate] possessor

[+ human], [-animal], [-collective]

[-human], [-collective]

no proper nouns

no proper nouns

I

no geographical and temporal nouns

personal nouns only e.g. girl, mother, boy, man

concrete nouns only e.g. chair, bed, door

i

1

Among animate noun classes, proper nouns (John) stand out in being the most likely to occur with the ί-genitive (cf. Jucker 1993: 126-127; Hundt 1998: 4 4 ^ 5 ; Biber et al. 1999: 302). Note, that proper nouns are - inherently - highly topical as well, which again shows to what extent animacy and topicality interact. While the s-genitive is said to tend not to occur with inanimate possessors (cf. Biber et al. 1999: 303), it has been reported to be quite common with certain inanimate noun classes, such as geographical nouns (London's weather, the city's shops) or temporal nouns (today's weather), cf. Jahr Sorheim (1980); Jucker (1993: 126-127); Hundt (1998: 44^45). Collective-noun possessors (e.g. the committee's decision) are still

Iconicity and economy in the choice between the two

genitives

387

fairly common with the s-genitive, although this noun class, vacillating between an animate and an inanimate interpretation, resists a clear categorization as to animacy. To avoid any specific lexical effect, I will therefore only regard common nouns, comparing personal (human) nouns with truly inanimate concrete nouns; these two noun classes have been shown to display the least propensity for taking the s-genitive in the animate, respectively the inanimate domain. Note that an exclusion of proper nouns/names (e.g. John, London) is essential for any analysis which tries to investigate the factors animacy and topicality independently from each other. Since such proper nouns/names are inherently topical, it would otherwise not be clear whether in cases such as John's book it is the high topicality of John which induces the s-genitive or the fact that it is a human possessor. Likewise, in London's underground we simply cannot tell whether the s-genitive is chosen because the possessor is a proper (geographical) name or because - or rather despite of the fact that - it is inanimate.11 With topicality I refer to the distinction between referentially given and new possessors.12 A [+topical] possessor was always a second-mention, definite expression (e.g. the girl, his father, the chair)', a [-topical] possessor always a first-mention, indefinite expression (e.g. a girl, one man, some composer), as illustrated in Table 5 below. Table 5. Topicality of possessor: operational definition [+topical] possessor

[-topical] possessor

preferential]

[^referential]

second mention

first mention

definite expression

indefinite expression

Having established an operational definition for the factors animacy and topicality, the question now is how a [+animate] and [+topical] possessor should influence its serial position within a possessive construction, and how this should be regarded as reflecting iconic/natural principles? In psycholinguistic research it has been argued that concepts are processed and then serialized in the order in which they become available to the mind. In particular, animates and topics have been shown to be highly accessible and to occur early in utterances (cf. Bock and Irwin 1980; Bock and Warren 1985; Fenk-Oczlon 1983, 1989; Bock and Levelt 1994: 965966; McDonald, Bock and Kelly 1993). Accordingly, a [+animate] and [+topical] possessor should occur early in a possessive construction, which is only possible in the s-genitive. Therefore, we can predict to find more sgenitives with [+animate] and [+topical] possessors, which can be taken to

388

Anette

Rosenbach

be iconic/natural since in this case linguistic serialization reflects the order of processing, which in turn reflects the order of conceptualization. [+animate] [+ topical] possessor

highly accessible ^

early processing ^

early serialization

4

s-genitive

Note, however, that while there is compelling cross-linguistic evidence for animates to occur early in linear order (cf. Siewierska 1988: 56-60; Ortmann 1988: 75-76; Yamamoto 1999: 52-56), and there seems to be good reason to assume that this word order preference is a cognitive universal (cf. Dahl and Fraurud 1996: 58; and see Haspelmath 1999b: 199 for discussing ANIM > INANIM as a constraint used in Optimality Theory) provided a language allows for such a flexibility - , the preference for putting topics (i.e. given information) first, most probably is not. As is, for example, shown by Hawkins (1994: §4.4), the relative order of given and new information is sensitive to the branching direction of the language involved. While given > new is typical of right-branching languages, such as English, new > given seems to be more frequently found in left-branching languages, such as Japanese. This directly concerns the question to what extent it may be legitimate to stretch the notion of iconicity as to 11 encompass matters of conceptualization which may be language-specific. For the purpose of the present study, however, which investigates a word order choice in English, it will be sufficient to note that given > new is a processing principle operating in the English language. 3.2. The principle of conceptual distance: possessive

relations

The classification of the semantic relations holding between the possessor and the possessum is a notoriously difficult enterprise. The traditional taxonomies of the genitive functions (e.g. Quirk et al. 1985: §5.116; Biber et al. 1999: 303-304) are quite arbitrary and ad hoc and do not entail any predictions as to which functions the ^-genitive should be expected to favour. Such predictions can, however, be derived from the typological work on possession, where a distinction is made between alienable and inalienable possession (cf. e.g. Nichols 1988), which, although not grammaticalized in English, may nonetheless well reflect preferences in the choice between the ^-genitive and the o/-genitive, i.e. may predict distributional frequency patterns. In this study I will regard kin terms and body parts, which are the most uncontroversial members of the inalienable class cross-linguistically

Iconicity and economy in the choice between the two genitives

389

(cf. Seiler 1983: 13; Nichols 1988: 572-573) and legal/permanent ownership, which ranks highest in Taylor's (1989a, b) prototype account and in Heine's (1997: 39—40) possession scale, as the most prototypical possessive relations for animate possessors, while abstract possession or states may be regarded as less prototypical instances of animate possession. Note, that these possessive relations (kin terms, body parts, legal ownership) have also recently been regarded by Koptjevskaja-Tamm (2001, forthcoming) as the prototypical members of possession. For inanimate possessors the situation is slightly complicated by the fact that the prototypical possessor is by definition human (cf. Seiler 1983: 4; Taylor 1989a: 679). Nonetheless I will consider part/whole relations as prototypical instances of inanimate possession, since here the possessum forms an inseparable part of its possessor, while non-part/whole relations will be regarded as [-prototypical]; for an overview see Table 6 below. Table 6. Type of possessive relation: operational definition [+animate] [+human] [+prototypical]

[-animate]

[-prototypical]

Body parts: hand, eyes,... Kin terms: father, brother, ... Permanent/legal ownership: car, house,...

States: exhaustion, pride, joy, ... Abstract possession: future, career, ...

[+prototypical]

[-prototypical]

Part/whole: frame/chair; bonnet/car; door/building

Non-part/whole: contents/bag; condition/car

Based on the principle of conceptual distance as proposed by Haiman (1985) we can now link the type of possessive relation to the choice of genitive construction. This iconic principle states that "the greater the formal distance between X and Y, the greater the conceptual distance between the notions they represent" (Haiman 1985: 106). That is, the s-genitive as the structurally more bonded, tighter construction should be more likely to occur with [+prototypical] possessive relations, representing a close connection between possessor and possessum, than with [-prototypical] relations, as illustrated in Figure 2 below. Possessive Relations Conceptual Distance

I

Linguistic Form

[+prototypical] = close relation

I

NP 's NP s-genitive

Figure 2. Possessive relation: predictions

1 [-prototypical] = loose relation

I

NP o / N P o^genitive

390

Anette Rosenbach

4. Modern English: experimental study To test the predictions for the factors animacy, topicality and possessive relation I conducted an experimental study with 56 British and 48 American native speakers,14 a methodological tool, which has to my knowledge hitherto not been applied to the English genitive variation. Although an experimental study does probably not reflect actual performance in the same way as naturally occurring speech data, it seemed promising for the following reasons. First, given the multitude of factors involved, an experimental design allows the analyst to investigate the effect of the factors animacy, topicality and possessive relation solely, focusing on real choice contexts and controlling to the best of our knowledge all other factors known to bias the choice between the two genitive constructions. Second, given the interaction of the three factors, it allows us to keep their effects apart in the empirical analysis, thereby considering not only their interaction but also enabling us to evaluate their relative importance. In the empirical analysis below, the following questions will be addressed. 1) I shall test whether the predictions derived from iconic/natural principles in §3 above are supported by the relevant evidence. To what extent is the 5-genitive more frequently realized with a [+animate], [+topical] possessor in a [+prototypical] possessive relation? 2) The relative importance of the factors animacy, topicality and possessive relation will be assessed, i.e. the question of which factors turn out to be more important than others. 3) The results for 1) and 2) will be given both a diachronic and a varietal perspective. I shall investigate to what extent any change-in-progress phenomena can be discerned and whether there are any differences between British and American Standard English.

4.1. Procedure Each subject was asked to fill in a questionnaire, which contained small text passages, adapted from novels, providing contexts for adnominal genitive constructions.15 The subjects had to choose as spontaneously as possible between the use of the s-genitive or the o/-genitive in the given contexts. To prevent the subjects from being tempted into giving stereotype responses, the order of presentation of the two genitive constructions was

Iconicity and economy in the choice between the two

genitives

391

randomized so that not always the same genitive construction occurred as the first choice in brackets. An example of what the task looked like is given in (1): (1)

A helicopter waited on the nearby grass like a sleeping insect, its pilot standing outside with Marino. Whit, a perfect specimen of male fitness in a black flight suit, opened [the helicopter's doors/the doors of the helicopter] to help us board, (adapted from: Patricia Cornwell, The Body Farm [1994: 52])

4.2. Conditions and items The three nominal variables tested resulted in eight conditions; in Table 7 below these conditions are specified, illustrated by an example for each condition. There were at least 10 items per condition, altogether 93 items. Table 7. Conditions and items [+animate] [+topical] [+proto]

a)

[-animate] [-topical]

[-proto]

[+proto] [-proto]

the boy's eyes/

the mother's future/

a girl's face/

the eyes of the boy

the future 0fthe mother

a)

°f

[+proto]

a woman's the chair's shadow/ frame/

face a

[+topical]

shadow of a

woman

[-topical]

[-proto]

[+proto]

[-proto]

the bag's contents/

a lorry's wheels/

a car's fumes/

^e frame of the

the the contents of wheels of

the fumes

chair

the bag

of a car

a lorry

proto = prototypical

Testing all eight logically possible combinations of the factors animacy, topicality and possessive relation allows us to keep their effects apart in the empirical analysis, i.e. study these factors in isolation, and, in addition, make predictions as to the relative importance of these three factors. Note, that in Table 7 the conditions are arranged in such a way as to stipulate animacy as the most important factor, followed by topicality and then possessive relation. If this ordering holds true we would expect to find the sgenitive decreasing in its relative frequency - and the o/-genitive increasing - along this scale from left to right.

392

Anette Rosenbach

4.3. Analysis and results In the following I will first present the results for the synchronic state of affairs and then, in a second step, look at possible change-in-progress phenomena, both separately for the British and the American subjects. Finally, the results for the two standard varieties will be compared, both synchronically and diachronically. 4.3.1. Synchronic state of affairs Figure 3 below shows the relative frequency of the s-genitive and the ofgenitive for the British subjects according to the eight conditions (and in the same ordering) as given in Table 7 above.

+a/+t/+p +a/+t/-p +a/-t/+p +a/-t/-p -a/+t/+p -a/+t/-p • s-gpnitive

-a/-t/+p

-a/-t/-p

• of-genitive

Figure 3. British subjects (n = 56) - interaction of factors: relative frequency of the s-genitive versus the o/^genitive according to the eight conditions"^ a) b)

a = animate; t = topical; ρ = prototypical possessive relation token frequencies indicated above each column

Figure 3 shows that the relative frequency of the s-genitive decreases steadily from left to right; the differences between the single conditions are all statistically significant (at least chi-square, ρ < 0.01), except for the difference between the last two conditions.

Iconicity and economy in the choice between the two

genitives

393

As can be seen in Figure 4 below, the picture is exactly the same for the American subjects, except that apart from the difference between the last two conditions, for the American subjects also the difference between the [+a/-t/-p] and [-a/+t/+p], i.e. the condition at the borderline between the [+animate] and the [-animate] conditions turns out to be not statistically significant. 16

+a/+t/+p +a/+t/-p +a/-t/+p

+a/-t/-p

-a/+t/+p

-a/+t/-p

-a/-t/+p

-a/-t/-p

• s-genitive • of-genitive Figure 4. American subjects (n = 48 - interaction of factors: relative frequency of the sgenitive versus the o^genitive according to the eight conditions a ) / b ) a) b)

a = animate; t = topical; ρ = prototypical possessive relation token frequencies indicated above each column

Therefore, for both the British and the American subjects the relative importance of the three factors is as indicated in Table 7 above, i.e. animacy > topicality > possessive relation. Note, that it is not at all surprising to find most s-genitives in the [+a/+t/+p] condition, i.e. where all positive values of the factors are combined, as well as most o/-genitives with the negative clustering of factor values in [-a/-t/-p]. Yet, by taking into account the interaction of factors, we can pinpoint the contexts in which the s-genitive is favoured over the o/-genitive and vice versa. As is apparent from both the British and the American data in Figures 3 and 4, it is, for example, not just any [+animate] possessor that will favour the s-genitive; for a [-topical] possessor in a [-prototypical] possessive relation the of-genitive will become a more highly favoured choice.

394

Anette

Rosenbach

4.3.2. Change-in-progress In order to detect any potential changes-in-progress I adopted an apparenttime approach (cf. Labov 1972), dividing the subjects into an older subject group (age > 40) and a younger one (age < 40), assuming that the language of the older subjects should be indicative of an older language state than the language of the younger ones. Figure 5 illustrates the relative frequency of the s-genitive according to the two age groups for British English (for conditions, see again Table 7).

+a/+t/+p +a/+t/-p +a/-t/+p

+a/-t/-p

• younger subj ects

-a/+t/+p

-a/+t/-p

-a/-t/+p

-a/-t/-p

• ol der subj ects

Figure 5. British subjects - interaction of factors: relative frequency of the s-genitive for the younger and the older subjects a) b)

a = animate; t = topical; ρ = prototypical possessive relation token frequencies indicated above each column

While both age groups apparently use the s-genitive according to the same hierarchy (again, the 5-genitive decreases from left to right), there is a clear difference in the frequency of usage, with the younger subjects using significantly more s-genitives in the four [-animate] conditions (chi-square, ρ < 0.001).17 And again, the American subject groups show very similar results as can be seen in Figure 6 below. Both age groups follow the same hierarchy

Iconicity and economy in the choice between the two genitives

395

(again, from left to right there are fewer s-genitives), and again, the younger subjects use more s-genitives in the four [-animate] conditions, although for the American subjects the difference is not as pronounced as in the British data.18 1 0 0 -ι

+a/+t/+p +a/+t/-p +a/-t/+p +a/-t/-p -a/+t/+p -a/+t/-p • younger subjects

-a/-t/+p

-a/-t/-p

• older subjects

Figure 6. American subjects - interaction of factors: relative frequency of the s-genitive for the younger and the older subjects"^ a) a = animate; t = topical; ρ = prototypical possessive relation b) token frequencies indicated above each column

4.3.3. Comparing British and American English So far, we have only looked at the distribution of the s-genitive, both synchronically and taking a short-term diachronic perspective, according to the factors animacy, topicality and possessive relation within British and within American English. In a next step let us now compare the British and American data. As a first approximation, Figure 7 shows the relative frequency of the s-genitive according to the three factors for all British subjects as opposed to all American subjects. It is, again, only the [-animate] conditions which are conspicuous in that the American subjects use the j-genitive significantly more often with [-animate] possessors (26.5%) than the British subjects (21.9%), (chisquare, ρ < 0.001).

396

Anette Rosenbach

1414 1218

+animate

-animate

+topical

• British subjects

-topical

+prototyp.

-prototyp.

• American subjects

Figure 7. British versus American subjects: relative frequency of the s-genitive according to the factors animacy, topicality and possessive relation a) a)

token frequencies indicated above columns

It is certainly striking that for both the factors change-in-progress and standard variety the [-animate] conditions stand out in the analysis in that in these conditions the s-genitive is used more frequently by the younger subject groups and by American rather than British subjects. 35 30 >> ^ c £w 25 ω 3 3, ω ω er r •έ 20 u • *^ "Ξc I & 15

358 441 232

-

-

133

•g «s 10 5 0

-

'

"" topicality > possessive relation. This proves to be a robust hierarchy which is unaffected by standard variety and change-in-progress. There was also evidence for ongoing change in the use of the s-genitive with [-animate] possessors. Although American English is still leading the trend towards an increased use of the s-genitive with [-animate] possessors, the difference between the two standard varieties appears to be petering out in present-day English. Note, finally, that the increasing use of the s-genitive with inanimate nouns has long been noted as a recent development in 20thcentury English (cf. Zachrisson 1920: 39-49; Jespersen 1949: 327-328; Thomas 1953; Barber 1964: 132-134; Dahl 1971; Jahr Sorheim 1980;

Iconicity and economy in the choice between the two

399

genitives

Raab-Fischer 1995; Denison 1998: 119; Hundt 1998). In all these studies this development is, however, described as being primarily confined to certain conspicuous inanimate noun classes, particularly geographical and temporal nouns. The present study provides additional evidence that this ongoing change towards an increasing use of the s-genitive with [-animate] possessors is not lexically restricted but in fact much more productive, by showing that the s-genitive has also become more frequent in an inanimate noun class hitherto assumed not to be participating in this change, i.e. concrete nouns.

5. The interplay of iconicity and economy in the long-term diachronic development of the English ϊ-genitive So far, choice of genitive construction has been exclusively motivated in terms of iconic/natural principles. In the following, I will argue that both the choice and the historical development of the ί-genitive is not only driven by iconic but also by economical tendencies. To put the results of the present study into a more long-term diachronic perspective I will now refer to the results of a joint project on genitive variation in Late Middle/ Early Modern English (cf. Rosenbach and Vezzosi 2000; Rosenbach, Stein and Vezzosi 2000). In these studies we have shown that the s-genitive, after a steady decline during the Middle English period, increases again in its relative frequency in the period between the 15th and the early 17th century. On the basis of this work, I argue in Rosenbach (2001, 2002) that the s-genitive has extended its range of applications along the following preference structure, which needs to be read as a kind of decision tree and which is simply another way of illustrating the hierarchical order of the factors animacy, topicality and possessive relation. Preference structure for English s-genitive

I

I

[+animate]

[-animate]

I

1

[+topical]

[-topical]

I [+proto] s-genitive

I [-proto]

I [+proto]

I

1

[+topical]

I I

[-topical]

I

I

[-proto] [+proto] [-proto] [+proto]

I [-proto] o/-genitive

The observed diachronic extension of the s-genitive along this preference structure can be briefly summarized as follows. The new productivity of the

400

Anette Rosenbach

s-genitive seems to have started in the early 15th century. If the s-genitive was found at all at that time, it was still confined to the most iconic/natural context on the left of this preference structure, i.e. with a [+animate], [+topical] possessor in a [+prototypical] possessive relation. Yet, even in this context the o/-genitive was the more frequent option at that time. During the 16th century the ^-genitive extends its use along this preference structure, both in compatible contexts and in frequency, and in the early 17th century it became practically everywhere the more frequent option with [+animate] and [+topical] possessors. Note, however, that the ^-genitive keeps being restricted to the animate domain during the Early Modern English period (cf. also Altenberg's 1982 data for 17th-century English); the extension of the s-genitive to inanimate possessors is a Modern English development, a process which, as the present study has shown, is still going on.21 But how does all this relate to the interplay between iconicity and economy? In the following I will proceed from a notion of economy as it is embedded in human cognition, assuming that it is the way language is conceptualized and processed by individual language users that should be as economical as possible, which we may call cognitive economy. When applied to grammatical variation, this predicts that - given two structural alternatives - the option demanding less mental effort will be preferred to that option requiring more mental effort. Note, that such an assumption is also implicit in the approaches to grammatical variation by Bock (1982, and, partly co-authored, subsequent work), Hawkins (1994, this volume) and Rohdenburg (1996, this volume). Given such a definition of economy the interplay of iconicity and economy in the diachronic extension of the sgenitive can now be summarized (cf. Table 8). Synchronically, it is not only iconic but also highly economical for the human processing system to process and serialize concepts in the order in which they become available to the mind instead of storing them in a buffer and wait until they can be released. It is also economical to use the s-genitive as the shorter construction where this can be done, that is, with prototypical possessive relations (cf. also Rohdenburg's 1996 complexity principle). Note, that it is exactly in this synchronically most optimal context that the s-genitive begins to become productive again in the early 15th century. While it is economical to choose the s-genitive in this most optimal context, it is even more economical to have this choice automatized over time.23 As suggested above, the s-genitive was becoming more and more frequent in Early Modern English, eventually ousting the o/-genitive as the preferred option in the [+animate] [+topical] [+prototypical] context, and

Iconicity and economy in the choice between the two

genitives

401

today the s-genitive is almost obligatory in this context at a probability of almost 90% (see §4.3.1. above), unless, of course, other factors call for the o/-genitive. This then constitutes a type of economy which in addition to being perfectly well compatible with iconicity already has a certain amount of diachrony built into it, in the sense that what is synchronically more optimal in performance will, diachronically, become more frequent. 4 Note, that these first two types of economy assign a different role to frequency. In the first case a higher frequency of the s-genitive is motivated by an economical principle (i.e. synchronic user-optimality); in the second case this higher frequency - in turn, as a kind of snowball effect - leads to the increasing automatization of the s-genitive.25 Table 8. Interplay of iconicity and economy in the diachronic extension of the .s-genitive iconicity = economy Synchronically

Ε m

Τ) §

Μ α. χ W

« c

iw ifi. υο

Diachronically

economical (I)

economical (II)

economical (III)

most optimal context for the s-genitive (early 15th century): [+animate] & [+topical] possessor in a [+prototypical] possessive relation

s-genitive becomes more frequent in optimal context (15th -early 17th century)

s-genitive extends to less and less optimal contexts (16th century - present-day English): [-animate] & [-topical] possessor in a [-prototypical] possessive relation

^ synchronic ε optimality

.2 uο •a TO u

iconicity * economy

user-

automatization (automatic processing) of cognitively most optimal and efficient context

still ongoing extension with [-animate] possessors! anal ogical/metaphorical extension: same strategy becomes eventually applied to similar contexts

The extension of the s-genitive to less and less optimal contexts while no longer iconic, is economical in the sense that the same linguistic strategy is actually applied to similar contexts (a process also known as analogical, metaphorical extension).26 Note, finally, that such a scenario of economically-driven language change towards the higher frequency of one genitive variant, i.e. the sgenitive, as such cannot really answer the question of when this change should have started, i.e. in itself it cannot account for why these economical forces should have become effective from the 15th century onwards. It is

402

Anette Rosenbach

only when we connect the diachronic extension of the s-genitive with the structural change of possessive 's from an inflection to a clitic that the timing makes sense. First evidence for a clitic-like behaviour of the s-genitive as in the so-called group genitive, where the -s-suffix attaches to the whole NP, is attested for the late 14th centuiy in Chaucer (e.g. [The grete god of Love]s name), but there seems to be a considerable period of transition until the 17th century, in which the s-genitive shows traces of both a fullyfledged inflection (as in "split constructions", e.g. the fkingjs daughter of Ethiopia) and of the newly evolving clitic (cf. Allen 1997; Rosenbach and Vezzosi 1999). The chronological correlation between the revival of the sgenitive on the one hand and the structural change of possessive's from an inflection to a clitic on the other, is certainly more than striking to be mere chance, and we may speculate that it may have paved the way for the sgenitive to become more frequent.

6. Conclusion and outlook In this paper I have investigated English genitive variation as a case of grammatical variation, focusing on the effect of the factors animacy, topicality and possessive relation. Apart from showing that and how these factors influence the choice of genitive construction I also set out to weigh the effect of these factors, i.e. animacy > topicality > possessive relation. While so far in variation studies factors are usually weighed (if at all) by means of elaborate statistical procedures, such as logistic modelling (see e.g. Leech, Francis and Xu 1994; Arnold et al. 2000) or multivariate analysis (see e.g. Tagliamonte 2000, this volume), this study offers an alternative way of assessing the relative importance of several factors by (i) keeping their effects apart, and (ii) testing all logically possible combinations of factors in a controlled experimental setting. Linking the effects of the factors animacy, topicality and possessive relation to iconic/natural principles, I have, moreover, made a case for a speaker-based approach to grammatical variation, arguing that the choice of genitive construction is determined by the needs of speakers to place easily available information first in linear order and to encode more prototypical, inherent and therefore more predictable relations in the more bounded construction (i.e. the s-genitive). Defining both iconicity and economy in cognitive-psychological terms, I have shown how they interact leading to an increasing use of the s-genitive along the proposed preference structure from Late Middle English onwards. While iconicity and economy are usually regarded as two opposing forces, with

Iconicity and economy in the choice between the two

genitives

403

the iconic mode gradually turning into the economic mode, an assumption underlying, for example, Haiman's (1994: 1634-1635) notion of routinization,28 in the present argumentation they start off as two sides of the same coin; only diachronically and only for the process of analogical/metaphorical extension, what is economical becomes diametrically opposed to what is iconic. It is precisely such an antagonism that ensures that, overall, the English language has not become more economical or more optimal in the use of genitive constructions. While the speaker-approach to grammatical variation advocated in the present paper is certainly an essentially functionalist position towards grammatical variation, it is worthwhile noting that very recently within the formal framework of Optimality Theory (OT) a novel way of analyzing grammatical variation has been introduced, which is close in spirit to the analysis offered here (if, naturally, differing in the technical details of analysis). An evaluation of the relative importance of factors, or rather in OT terminology, a ranking of (violable) constraints, lies at the very heart of OT analysis in general. What is new in this recent Functional Optimality Theoretic (FOT) approach (see particularly Bresnan and Aissen 2002 for a programmatic sketch of this approach) is that it tries to reconcile both formal analysis and functional considerations by assuming OT constraints to have an underlying cognitive-functional motivation. That is, OT constraints as proposed in such functional OT accounts are based on iconic and economic principles, similar to the account offered in this paper (for this, see e.g. Aissen 1999, 2000; Bresnan, Dingare and Manning 2001). In this respect it looks as if formal and functional approaches to grammatical variation, while always at odds with each other in the past, may - at least in principle, and to some extent - be reconcilable in the future. Notes *

I would like to thank the participants of the Paderborn symposium Determinants of grammatical variation in English for stimulating discussions, and the editors of this volume for valuable improvements of an earlier version of this paper. 1. For an in-depth treatment of the English genitive variation according to the factors animacy, topicality and possessive relation (and the data presented in this paper) as well as a more-detailed introduction of the particular type of speaker-based approach to grammatical variation and change as advocated in the present paper, I refer to Rosenbach (2002). 2. Prenominal modification in such cases would usually result in an adjective {an honourable king). 3. See, e.g., also Sankoff and Rousseau (1980: 12) for emphasizing that knock-out factors need to be excluded beforehand.

404

Anette Rosenbach

4. But see Jucker (1993: 131) for giving an example where an alternative o/-phrase can be used (This could be the death of him). As his corpus data show, however, such examples are extremely rare. 5. Also subjective (John's love) and objective genitives (John's murder), in which the possessum is a deverbal noun, are potential choice contexts. The present study focuses, however, on possessive meanings since it is not clear how such nominal valency relations should be treated in accounts of possession as introduced in §3.2 below. 6. This structural explanation is the traditional one (cf. Huddleston 1984: 233; Quirk et al. 1985: 326); for an alternative view which calls for a semantic-pragmatic motivation of article-possessor complementarity see Haspelmath (1999a). 7. Branching possessor constructions are generally called group genitives. Although they are hardly attested in written language, they seem to be more common in dialectal and colloquial language (see e.g. Carstairs 1987: 152; Jespersen [1918] 1960: 296-297). This calls into question any processing-based account for the alleged avoidance of the sgenitive with weighty possessors. Note also, that it is exactly such group genitives, which have, diachronically, emerged in the English language as a new construction from the late 14th century onwards (cf. Janda 1980; Carstairs 1987; Allen 1997); for this see also §5 below. 8. Note, that such recursive genitives may also be avoided due to the horror aequi principle proposed by Rohdenburg (this volume). 9. For a discussion of different types of iconicity see e.g. McMahon (1994: 84-86) and Fischer and Nänny (1999). 10. See also Dotter (1990: §4.4.2) for allowing an extension of "constructional iconicity" as to encompass the way the language user perceives, processes and utilizes concepts as to achieve the most efficient information flow. 11. This study is restricted to [+human] possessors because animals are usually regarded to rank lower on animacy scales (cf. Silverstein 1976: 122; see also the gender scale given in Quirk et al. 1985: §5.104), For evidence that the s-genitive figures lowest with concrete nouns, see e.g. Jucker (1993: 126-127). 12. Note, that non-referentiality of the possessor results in a modifying genitive construction (e.g. a bird's nest, women's clothes), which, as argued in §2.2.2 above, is ruled out in the present study as a non-comparable context. 13. I am grateful to William Croft for drawing my attention to this. 14. The subjects were all monolingual speakers of English (in the sense of having acquired English as their one and only first language), all having an advanced educational background. 15. Note, that the provision of context was particularly important for the factor topicality which is not only defined in terms of definite vs. indefinite expressions, but also anaphorically, in terms of first- and second-mention. 16. The differences between all other conditions are significant with at least chi-square, ρ < 0.001. 17. In the [-a/-t/+p] condition the difference in the relative frequency between the two British age groups is not statistically significant, yielding, however, a strong tendency below the 0.10 level (chi-square, ρ < 0.10). 18. For the American subjects, it is only the difference in the [-a/-t/+p] condition that reaches any statistical significance (chi-square, ρ < 0.01); in the [-a/+t/+p] and the [-a/+t/-p] conditions still a relatively strong tendency can be observed (chi-square, ρ < 0.10).

Iconicity and economy in the choice between the two genitives

405

19. The difference in the relative frequency of the s-genitive vs. the o/-genitive between the four subject groups are all statistically significant with at least chi-square, ρ < 0.05. 20. Again, it is the [-a/-t/+p] condition which does not fit in the otherwise neat picture; for the older subject groups the difference between the two standard varieties is not significant here. 21. Note, that the preference structure itself, i.e. the relative importance of the three factors, remains stable over time. What changes and varies is to what degree this preference structure is explored. This corresponds to the findings reported by Tagliamonte (this volume), who shows in a multifactorial analysis, weighing several factors, that the ranking of the tested factors remains the same, unaffected by variation between groups and over time, with only the locus on the scale varying for the different subject groups. 22. Note, that this is not an altogether uncontroversial assumption. Some scholars see no reason at all to believe that biological systems in general and language in particular should be subject to considerations of economy (see e.g. Johnson and Lappin 1997: 328-329). Cognitive economy is, however, an assumption implicit in many orientations of linguistics and cognitive science, as e.g. in Relevance Theory (Sperber and Wilson 1986), Fodor's (1983) modularity hypothesis and Chomsky's (1995) Minimalist Program to name just a few and illustrate how widespread it is. 23. The distinction between automatic and controlled processing (cf. Schneider and Shiffrin 1977) is an assumption implicit in production frameworks such as Bock (1982) and Levelt (1989). While controlled processes are intentional, strategic devices, involving speakers' consciousness and therefore drawing on working memory capacity, automatic processes are employed unintentionally, run very fast, do not need much capacity and are therefore an extremely economical processing mode. 24. Note, that this is an assumption also underlying Hawkins' (1994) Performance Grammar as well as the mechanism of diachronic adaptation recently proposed by Haspelmath (1999b). For a critical view on Haspelmath (1999b) see also the peer commentaries in Zeitschrift für Sprachwissenschaft 18 (2). 25. On the role of frequency see also Krug (1997, this volume). 26. See e.g. Ronneberger-Sibold (1980) for regarding analogy as an economical principle. 27. For a more detailed account of the change of possessive 's towards a clitic (as well as a determiner) and how this links to the observed diachronic extension of the .s-genitive, see Rosenbach (2002: §6.5). On the question of how this development is to be interpreted in terms of grammaticalization theory, see Rosenbach (2002: §7.6.3). 28. But see also Fischer (1999) for arguing that "we are still always at the crossroads of both possibilities" (348), i.e. the iconic and the symbolic (i.e. economic) mode, and for showing that in the development of infinitival to in English indeed a kind of Ikonisierung as described by Plank (1979) has taken place (cf. also Fischer 2000). Fischer (this volume) further explores this issue, elaborating on the relation between grammaticalization and iconicity.

References Aissen, Judith 1999 Markedness and subject choice in optimality theory. Natural Language and Linguistic Theory 17: 673-711. 2000 Differential object marking: iconicity vs. economy. Draft on-line, University of California, Santa Cruz (http://ling.ucsc.edu/~aissen).

406

Anette Rosenbach

Allen, Cynthia 1997 The origins of the 'group genitive' in English. Transactions of the Philological Society 95: 111-131. Altenberg, Bengt 1982 The Genitive v. the Of-Construction. A Study of Syntactic Variation in 17th Century English. Malmö: CWK Gleerup. Anschutz, Arlea 1997 How to choose a possessive noun phrase construction in four easy steps. Studies in Language 21 (1): 1-35. Arnold, Jennifer E., Thomas Wasow, Anthony Losongco and Ryan Ginstrom 2000 Heaviness vs. newness: the effects of structural complexity and discourse status on constituent ordering. Language 76 (1): 28-55. Barber, Charles 1964 Linguistic Change in Present-day English. Edinburgh: Oliver and Boyd. Biber, Douglas, Stig Johansson, Geoffrey Leech, Susan Conrad and Edward Finegan 1999 Longman Grammar of Spoken and Written English. London: Longman. Bock, J. Kathryn 1982 Toward a cognitive psychology of syntax: information processing contributions to sentence formulation. Psychological Review 89: 1-47. Bock, J. Kathiyn and D. E. Irwin 1980 Syntactic effects of information availability in sentence production. Journal of Verbal Learning and Verbal Behavior 19: 467-484. Bock, J. Kathryn and Willem Levelt 1994 Language production: grammatical encoding. In: Morton Ann Gernsbacher (ed.), Handbook of Psycholinguistics, 945-984. San Diego: Academic Press. Bock, J. Kathryn and Richard K. Warren 1985 Conceptual accessibility and syntactic structure in sentence formulation. Cognition 21: 47-67. Bolinger, Dwight 1977 Meaning and Form. London: Longman. [Third impression 1983]. Bresnan, Joan and Judith Aissen 2002 Optimality and functionality: objections and refutations. Natural Language and Linguistic Theory 20 (1): 81-95. Bresnan, Joan, Shipra Dingare and Christopher Manning 2001 Soft constraints mirror hard constraints: voice and person in English and Lummi. In: Miriam Butt and Tracy H. King (eds.), Proceedings of the Lexical Functional Grammar 01 Conference, University of Hong Kong, on-line proceedings. Stanford: Center for the Study of Language and Information Publications (http://csli-publications.stanford.edu/). Carstairs, Andrew 1987 Diachronic evidence and the affix-clitic distinction. In: Anna G. Ramat, Onofrio Carruby and Giuliano Bernini (eds.), Papers from the 7th International Conference on Historical Linguistics, 151-162. Amsterdam: Benjamins. Chomsky, Noam 1995 The Minimalist Program. Cambridge, MA.: MIT Press. Dahl, Liisa 1971 The s-genitive with non-personal nouns in modern English journalistic style. Neuphilologische Mitteilungen 72: 140-172.

Iconicity and economy in the choice between the two genitives

407

Dahl, Osten and Kari Fraurud 1996 Animacy in grammar and discourse. In: Thorstein Fretheim and Jeanette Κ. Gundel (eds.), Reference and Referent Accessibility, 47-64. Amsterdam: Benjamins. Denison, David 1998 Syntax. In: Suzanne Romaine (ed.), The Cambridge History of the English Language, Volume 4 (1776-1997), 92-329. Cambridge: Cambridge University Press. Dotter, Franz 1990 Nichtarbitrarität und Ikonizität in der Syntax. Hamburg: Helmut Buske. Dressier, Wolfgang U., Willi Mayerthaler, Otto Panagl and Wolfgang U. Wurzel 1987 Introduction. In: Wolfgang Dressier, Willi Mayerthaler, Otto Panagl and Wolfgang U. Wurzel (eds.), Leitmotifs in Natural Morphology, 3-22. Amsterdam: Benjamins. Fenk-Oczlon, Gertraud 1983 Ist die SVO-Wortfolge die 'natürlichste'? Papiere zur Linguistik 29: 23-32. 1989 Word frequency and word order in freezes. Linguistics 27: 517-556. Fischer, Olga 1999 On the role played by iconicity in grammaticalisation. In: Olga Fischer and Max Nänny (eds.), Form Miming Meaning. Iconicity in Language and Literature, 345-374. Amsterdam: Benjamins. 2000 Grammaticalisation: unidirectional, non reversable? The case of to before the infinitive in English. In: Olga Fischer, Anette Rosenbach and Dieter Stein (eds.), Pathways of Change. Grammaticalization in English, 149-169. Amsterdam: Benjamins. this vol. Grammaticalization and iconicity. Fischer, Olga and Max Nänny 1999 Introduction. In: Max Nänny and Olga Fischer (eds.), Form Miming Meaning. Iconicity in Language and Literature, xv-xxxvi. Amsterdam: Benjamins. Fodor, Jerry A. 1983 The Modularity of Mind. Cambridge, MA.: MIT. Press. Haiman, John 1985 Natural Syntax. Iconicity and Erosion. Cambridge: Cambridge University Press. 1994 Iconicity. In: Ronald E. Asher and J.M.Y. Simpson (eds.), The Encyclopedia of Language and Linguistics, 1629-1637. Oxford: Pergamon. Haspelmath, Martin 1999a Explaining article-possessor complementarity: economic motivation in noun phrase syntax. Language 75 (2): 227-243. 1999b Optimality and diachronic adaptation. Zeitschrift für Sprachwissenschaft 18 (2): 180-205. Hawkins, John A. 1994 A Performance Theory of Order and Constituency. Cambridge: Cambridge University Press. this vol. Why are zero-marked phrases close to their heads? Heine, Bernd 1997 Possession. Cognitive Sources, Forces and Grammaticalization. Cambridge: Cambridge University Press.

408

Anette

Rosenbach

Huddieston, Rodney 1984 Introduction to the Grammar of English. Cambridge: Cambridge University Press. Hundt, Marianne 1998 New Zealand English Grammar. Fact or Fiction. A Corpus-Based Study in Morphosyntactic Variation. Amsterdam: Benjamins. Jahr Sorheim, Mette-Catherine 1980 The s-genitive in present-day English. Department of English, University of Oslo. Janda, Richard D. 1980 On the decline of declensional systems: the overall loss of OE nominal case inflections and the ME reanalysis of -es as his". In: Elizabeth C. Traugott, Rebecca Labrum and Susan Shepherd (eds.), Papers from the 4th International Conference on Historical Linguistics, 243-252. Amsterdam: Benjamins. Jespersen, Otto 1949 A Modern English Grammar on Historical Principles. Part VII, Syntax. London: Allen and Unwin. [repr. 1961] 1960 Chapters on English. Selected Writings of Otto Jespersen, 153-345. London: Allen and Unwin. [1918] Johnson, David and Shalom Lappin 1997 A critique of the minimalist program. Linguistics and Philosophy 20: 2 7 3 333. Jucker, Andreas 1993 The genitive versus the o/-construction in newspaper language. In: Andreas Jucker (ed.), The Noun Phrase in English. Its Structure and Variability, 121136. Heidelberg: Carl Winter. Koptjevskaja-Tamm, Maria 2001 Adnominal possession. In: Martin Haspelmath, Ekkehard König, Wulf Oesterreicher and Wolfgang Raible (eds.), Language Typology and Language Universals, Volume 2, 960-970. (Handbooks of Linguistics and Communication Science 20. 1,2). Berlin/New York: Mouton de Gruyter. forthc. Adnominal possesion in the European languages. In: Thomas Stolz (ed.), Inalienable Possession in the European Languages. (Sprachtypologie und Universalienforschung, special issue.). Krug, Manfred 1998 String frequency. A cognitive motivating factor in coalescence, language processing, and linguistic change. Journal of English Linguistics 26 (4): 286-320. this vol. Frequency as a determinant in grammatical variation and change. Labov, William 1972 Sociolinguistic Patterns. Philadelphia: University of Philadelphia Press. Leech, Geoffrey, Brian Francis and Xunfeng Xu 1994 The use of computer corpora in the textual demonstrability of gradience in linguistic categories. In: Catherine Fuchs and Bernard Victorri (eds.), Continuity in Linguistic Semantics, 57-76. Amsterdam: Benjamins. Levelt, Willem J.M. 1989 Speaking. From Intention to Articulation. Cambridge, MA.: MIT Press.

Iconicity and economy in the choice between the two genitives

409

McDonald, Janet L., J. Kathryn Bock and Michael Kelly 1993 Word and world order: semantic, phonological, and metrical determinants of serial position. Cognitive Psychology 25: 188-230. McMahon, April M.S. 1994 Understanding Language Change. Cambridge: Cambridge University Press. Mondorf, Britta this vol. Support for more-support. Nichols, Johanna 1988 On alienable and inalienable possession. In: William Shipley (ed.), In Honor of Mary Haas: From the Haas Festival Conference on Native American Linguistics, 557-609. Berlin: Mouton de Gruyter. Ortmann, Albert 1998 The role of [± animate] in inflection. In: Ray Fabry, Albert Ortmann and Teresa Parodi (eds.), Models of Inflection, 60-84. Tübingen: Niemeyer. Plank, Frans 1979 Ikonisierung und De-Ikonisierung als Prinzipien des Sprachwandels. Sprachwissenschaft 4: 121-158. Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech and Jan Svartvik 1985 A Comprehensive Grammar of the English Language. London: Longman. Raab-Fischer, Roswitha 1995 Löst der Genitiv die o^Phrase ab? Eine korpusgestützte Studie zum Sprachwandel im heutigen Englisch. Zeitschrift für Anglistik und Amerikanistik 43 (2): 123-132. Rohdenburg, Günter 1996 Cognitive complexity and increased grammatical explicitness in English. Cognitive Linguistics 7: 149-182. this vol. Cognitive complexity and horror aequi as factors determining the use of interrogative clause linkers in English. Ronneberger-Sibold, Elke 1980 Sprachverwendung - Sprachsystem. Ökonomie und Wandel. Tübingen: Niemeyer. Rosenbach, Anette 2002 Genitive Variation in English. Conceptual Factors in Synchronic and Diachronie Studies. Berlin/New York: Mouton de Gruyter. 2001 The English j-genitive: animaey, topicality and possessive relationship in a long-term diachronic perspective. In: Laurel Brinton (ed.), Historical Linguistics 1999: Selected Papers from the 14th International Conference on Historical Linguistics, 277-292. Amsterdam: Benjamins. Rosenbach, Anette, Dieter Stein and Letizia Vezzosi 2000 On the history of the ί-genitive. In: Ricardo Bermudez-Otero, David Denison, Richard M. Hogg and Chris B. McCully (eds.), Generative Theory and Corpus Study: A Dialogue from 10 ICEHL, 183-210. Berlin: Walter de Gruyter. Rosenbach, Anette and Letizia Vezzosi 1999 Was the ί-genitive a traveller through England? In: Alexander Bergs, Monika Schmid and Dieter Stein (eds.), LANA-Diisseldorf Working Papers on Linguistics 1: 35-55. (http://ang3-ll.phil-fak.uni.duesseldorf.de/ ~ang3/LAN A/LAN A. html)

410

Anette Rosenbach

2000

Genitive constructions in Early Modern English: new evidence from a corpus analysis. In: Rosanna Sornicola, Erich Poppe and Ariel ShishaHalevy (eds.), Stability, Variation and Change in Word-Order Patterns over Time, 285-307. Amsterdam: Benjamins. Sankoff, David and Pascale Rousseau 1980 Categorical Contexts and Variable Rules. In: Sven Jacobson (ed.), Papers from the Scandinavian Symposium on Syntactic Variation, Stockholm, May 18-19, 1979, 7-22. Stockholm: Almqvist and Wiksell. Schneider, Walter and Richard M. Shiffrin 1977 Controlled and automatic human information processing: I. Detection, search and attention. Psychological Review 84: 1-66. Seiler, Hans Jakob 1983 Possession as an Operational Dimension of Language. Tübingen: Gunter Narr Verlag. Siewierska, Anna 1988 Word Order Rules. London: Croom Helm. Silverstein, Michael 1976 Hierarchy of features and ergativity. In: Robert M.W. Dixon (ed.), Grammatical Categories in Australian Languages, 112-171. Canberra: Australian Institute of Aboriginal Studies. Sperber, Deirdre and Dan Wilson 1986 Relevance. Oxford: Blackwell. Stefano witsch, Anatol this vol. Constructional semantics as a limit to grammatical variation: the two genitives in English. Stiebels, Barbara 2000 Typologie des Argumentlinkings: Ökonomie und Expressivität. Habilitation thesis, Department of General Linguistics, University of Düsseldorf. Tagliamonte, Sali Α. 2000 The grammaticalization of the present perfect in English: tracks of change and continuity in a linguistic enclave. In: Olga Fischer, Anette Rosenbach and Dieter Stein (eds.), Pathways of Change. Grammaticalization in English, 329-354. Amsterdam: Benjamins. this vol. "Every place has a different toll": determinants of grammatical variation in cross-variety perspective. Taylor, John 1989a Possessive genitives in English. Linguistics 27: 663-686. 1989b Linguistic Categorization: Prototypes in Linguistic Theory. Oxford: Clarendon Press. 1996 Possessives in English. Oxford: Clarendon Press. Thomas, Russel 1953 Notes on the inflected genitive in modern American prose. College English 14:236-239. Wasow, Thomas 1997 Remarks on grammatical weight. Language Variation and Change 9: 81105. Yamamoto, Mutsumi 1999 Animacy and Reference. Amsterdam: Benjamins.

Iconicity and economy in the choice between the two genitives Zachrisson, R.E. 1920 Grammatical changes in present-day English. Studier i Modern vetsenskap (Stockholm Studies in Modern Philology.) 7: 19-61.

411

Spraek-

Constructional semantics as a limit to grammatical alternation: The two genitives of English Anatol Stefanowitsch

Abstract It has often been claimed that the distribution of the s-genitive and the o/-genitive is determined by considerations of information structure, more specifically by linear precedence preferences related to animacy, givenness, or syntactic weight. This paper shows that such claims are untenable on empirical as well as theoretical grounds. First, corpus analyses simply do not bear out the predictions made by these claims. Second, such claims assume that the two genitives are semantically equivalent. I show that this assumption is wrong and offer a systematic account of the s-genitive and the o^genitive as distinct semantic-role constructions, arguing that the former encodes a possessor-possessee relation and the latter a part-whole relation unless the head noun itself inherently specifies a different relation. Only in the case of such an inherently specified relation does the possibility arise that information structure may play a role. I then show that in such cases animacy and (to a lesser degree) length have an influence, but that givenness has an optional stylistic influence at best.

1. Introduction Wherever a language has two constructions with the same apparent syntactic function, the question arises as to what conditions their distribution. Unless they are in free variation, there seem to be two main possibilities: either the two constructions differ in their discourse-functional properties (i.e. they encode alternative ways of structuring the information flow), or they differ in their semantics (i.e. they either have different constraints on the lexical items they occur with, or they differ in their semantic import). English has two nominal modification constructions that are traditionally referred to as genitives: one where the modifier is morphologically marked with the possessive clitic -'s and precedes the head noun, and one where the modifier is syntactically marked by the preposition of and follows the head noun. The two constructions, traditionally referred to as s-genitive and o/-genitive, are shown in (la) and (lb) respectively:

414

Anatol

(1)

a.

Stefanowitsch

[NPmod'-S Nhead]

e.g. the university's budget b. [DET Nhead ofNP mod ] e.g. the budget of the university It is received wisdom that the distribution of these two constructions is governed by information structure (in a broad sense of the term). It has repeatedly been claimed that they are primarily discourse-pragmatic alternatives, conditioned by linear precedence preferences related to animacy (e.g., Jespersen 1949; R. Hawkins 1981; cf. also Quirk et al. 1985: §17.39; Jucker 1993), topicality (e.g., Standwell 1982; Osselton 1988; cf. also Altenberg 1980; Quirk et al. 1985: §17.45), or a combination of the two (Deane 1987, 1992). I will refer to this hypothesis as the discoursefunctional hypothesis. In this paper, I will take issue with these claims. I will show that there are two problems with an analysis of the two genitives that accounts for their distribution primarily in terms of information structure. First, any analysis of a pair of constructions in terms of information structure has to assume that the two constructions are semantically more or less equivalent; this assumption is not warranted in the case of the two genitives. Second, text counts simply do not bear out the predictions concerning linear precedence made by such an analysis. I will argue instead that the two genitives are semantically distinct constructions, whose primary function is the assigning of semantic roles to their head and modifier slots. This claim, which I will call the semantic hypothesis, may seem difficult to substantiate in light of the fact that the meanings of the two constructions are notoriously hard to pin down. It has been claimed that they encode "a grab-bag of relations" (Givon 1993: 264) and even that "any attempt to sum up 'the meaning' of the genitive is doomed" (Strang 1962: 93). I will show that while the two constructions indeed appear to encode a "grab-bag" of relations, their semantics can be accounted for in a principled way in the framework of Construction Grammar (cf. Goldberg 1995), although such an analysis requires a relatively abstract approach to semantic roles. I will propose an analysis which accounts for the fact that there is a vast variety of relations encodeable by both constructions as well as for the fact that for each construction there is a core set of semantic relations that cannot be encoded by the other. I will not dismiss information structure entirely as a factor in the distribution of the two genitives. On the contrary, I will show that once the two constructions are properly characterized semantically, it is possible to

Constructional

semantics as a limit to grammatical

alternation

415

delimit a subset of genitives whose distribution is to some degree governed by factors related to information flow (namely animacy and length). A word on abstractness seems in order, since it has become a concept non grata to many linguists. Even within the field of cognitive linguistics, which I consider Construction Grammar to be a part of, there is considerable disagreement as to how abstract an analysis may be. At one extreme are discourse-centered approaches like Hopper's Emergent Grammar that eschew any kind of abstraction (e.g., Hopper 1998), at the other extreme is Langacker's Cognitive Grammar, which proposes highly schematic analyses of many linguistic phenomena (e.g., Langacker 1990). Both approaches have their merits. The first essentially regards language as a repository of remembered pieces of discourse and views any regularities as partial and emergent generalizations over this repository. The second approach focuses on these generalizations, partial though they may be, and attempts to uncover the systematicity inherent in them (and to relate this systematicity to general cognitive principles). In this paper, I focus on the systematic, general aspects of the two genitives rather than on clusters of concrete manifestations. This is not to deny that there are such clusters, i.e. that for both constructions there are recurring, highly entrenched instances and low-level generalizations. However, this paper aims to uncover the general properties of the two constructions and thus to delineate the semantic space within which such clusters may occur. The paper is structured as follows: Section two will outline the basic tenets of Construction Grammar, focusing on those aspects relevant to the issue at hand. Section three will discuss the discourse-functional and the semantic hypothesis in some more detail and outline the a priori advantages and problems of each. Section four will show that an analysis of the two constructions in terms of information structure is empirically inadequate, and section five will propose a semantic analysis that overcomes the apparent problems outlined in section three. Finally, section six will return to the issue of information structure and draw some general conclusions about the relation between semantics and discourse-pragmatic factors. 2. Some basic tenets of Construction Grammar Construction Grammar is a non-derivational, non-modular theory of the grammatical knowledge of speakers. Unlike most current linguistic theories (both generative and discourse-analytical ones), it views the construction as

416

A natol Stefanowitsch

the fundamental unit of grammatical organization, where construction is defined as follows: "C is a construction iffdef C is a form-meaning pair such that some aspect of F, or some aspect of S, is not strictly predictable from C's component parts or from other previously established constructions." (Goldberg 1995: 4) Meaning must be understood here in a broad sense, as encompassing semantic properties in the traditional sense of the term as well as frame-semantic encyclopedic knowledge, pragmatics, and information structure (Goldberg 1996: 69). In other words, a construction is any formal element that is directly associated with a particular meaning, pragmatic function, or discourse context. Such formal elements may be single morphemes (like give), multimorphemic words, like care-giver, or fully or partially filled idioms (like Give me five!, or [SUBJ be given to Nactjvity], as in Sam is not given to selfanalysis). Crucially, they may also be abstract syntactic patterns. Two abstract construction types that are important to the following discussion are argument-structure constructions, i.e. formal patterns whose elements are directly linked to particular configurations of semantic roles, and information-structure constructions, i.e. formal patterns whose elements are linked to particular ways of packaging information structure. As an example of an argument-structure construction, take the English double-object construction [SUBJ V OBJ OBJ]. The double-object (or ditransitive) construction assigns the semantic roles of agent, recipient, and theme to the subject, first object, and second object respectively, irrespective of the particular verbs which occur in this construction. This means that the construction itself imparts the meaning 'transfer' even with verbs that do not specify this notion as part of their lexical semantics. This is shown by the use of hit in Pat hit Chris the ball. Hit is a two-participant verb whose meaning can be roughly glossed as '(some part of) X comes into forceful contact with (some part of) Y \ Clearly, nothing in its meaning points to a transfer of Y to some third participant. However, a sentence like Pat hit Chris the ball will consistently receive the interpretation 'Pat transferred the ball to Chris by coming into forceful contact with it' (cf. Goldberg 1995: 34-35). In other words, constructions may add properties that are unspecified or underspecified in more specific constructions or lexical items. For example, the verb hit only specifies an Agent (a Hitter) and a Patient (a Hittee). These are compatible with two of the roles specified by the double object construction. Since hit does not specify a third role, this can be added by the double object construction itself. As a well-known example of an information-structure construction, take the two verb-particle constructions [SUBJ V OBJ PRT], as in Diane pushed

Constructional semantics as a limit to grammatical

alternation

417

Billy over, and [SUBJ V PRT OBJ], as in Diane pushed over Billy. These constructions differ in terms of the activation state (cf. Chafe 1994; Lambrecht 1994) of their constituents. The first construction assigns the activation state active to the object and is thus only compatible with discourse contexts where the object refers to given information, while the second construction assigns the activation state inactive to the object and is thus used in situations where the object refers to new information (cf. e.g., Chen 1986 or Gries, this volume, for a more detailed analysis of these constructions). Any actual utterance larger than a word is a simultaneous manifestation of several constructions. For example, the sentence Pat hit Chris the ball instantiates the subject-predicate construction (i.e. [SUBJ PRED]), the double-object construction (i.e. [SUBJ V OBJ OBJ]/'X transfers Y to Z'), the past tense construction (i.e. \W-ed\rpast'), the noun-phrase construction, and the lexical constructions corresponding to the individual words (cf. Goldberg 1996: 68). A construction that is a (full or partial) manifestation of a more general construction is said to inherit that more general construction. For example, the double-object construction inherits the subject-predicate construction: it is a more specific construction which inherits the form [SUBJ PRED] and adds its own specifications, namely the exact type of verb-phrase instantiating the predicate (i.e. [Vp V NP NP]), as well as a particular configuration of semantic roles (i.e.

Act

Syn

of

i

Nhead

i

NPmod

(b) of-genitive Figure 2. The two genitives as prototype semantic-role constructions

Second, a prototype analysis (with or without metaphorical mappings) cannot account for the fact that the set of semantic relations encodeable by the two genitives is almost open-ended. For example, given the right context, Kate 's shoes could refer to the shoes she owns, the shoes she wears, the shoes she dreams of wearing, the shoes she likes, the shoes she regrets not having bought when they were on sale, etc. Not all of these examples can plausibly be related to Taylor's prototype, and of course it is logically impossible to posit an open-ended set of metaphors to account for them. To avoid problems like this, Langacker (1992, 1993, 1995) takes the second type of approach mentioned above, i.e. a schematic approach, to the two genitives. He analyzes both constructions as manifestations of a general cognitive ability which he calls the reference-point function. He characterizes this function as follows: "one entity ... is invoked as a reference point for purposes of establishing mental contact with another" (Langacker

424

Anatol

Stefanowitsch

1993: 8). The first entity, the reference point, corresponds to the modifier in both genitive constructions. The second entity, which Langacker calls target, corresponds to the head. Langacker assumes that the s-genitive evokes the reference-point relation directly (Langacker 1993: 11) and thus has no additional semantic content. The o/-genitive, on the other hand, evokes the reference-point relation by virtue of the fact that it encodes an intrinsic relationship between two entities (cf. Langacker 1995: 69). This account can be represented in the Construction Grammar framework as shown in Figure 3.

(a) i-genitive

Sem

INTRINSIC RELATION

reference point > I

{

Act

Syn

< target Ι

Jf

} Nhead

NPmod

(b) o/-genitive Figure 3. The two genitives as abstract semantic-role constructions

This account also runs into several problems. First, the difference in the semantic value of the two genitives is extremely tenuous. Langacker himself comments that the s-genitive is "quite analogous to the o/-construction" (Langacker 1995: 69), and that, for example, in Kennedy's assassination vs. the assassination of Kennedy, the "only difference is that Kennedy's profiles the reference-point relationship per se, whereas of Kennedy profiles the relationship of intrinsicness (which has a reference-point relationship as a consequence)" (Langacker 1995: 69). It is unclear what it is about these two examples that justifies this claim, but more importantly, this difference cannot account for the different sets of semantic roles encoded by the two constructions. Following Langacker's account, the s-genitive should be

Constructional semantics as a limit to grammatical

alternation

425

able to encode any reference-point relationship, and it is thus hard to see why it cannot encode THING-CONSTITUENT (cf. *the silk's dress), SUBCATEGORY-CATEGORY (cf. *the wood's dark kind), SUBPART-WHOLE (cf. *the population's 50%), or DEPICTION-DEPICTED (cf. *the accident's footage). The o/-genitive, on the other hand, should only encode a subset of those semantic relations encodeable by the s-genitive, namely those that are intrinsic. However, note that the o/-genitive does not encode a subset of the relations encoded by the s-genitive. Furthermore, it is unclear what is meant by intrinsic. Langacker does not define the term, and it is hard to grasp intuitively what is more intrinsic about, for example, DEPICTION-DEPICTED than POSSESSEE-POSSESSOR or INTERPERSONAL RELATION (both of which allow the of-genitive only under very specific circumstances to be discussed below). A second problem is that this analysis is overly general: it does not delimit in any way the set of semantic relations that should be encodeable by the two constructions. Where the prototype account (especially in the version with metaphorical mappings) is unable to account for the fact that this set of relations is almost open-ended, the schematic analysis is unable to account for the fact that it is almost, but not entirely, open-ended. I will argue that it is possible to combine the advantages of the prototype account and the schematic account in a way that avoids their problems, and that this can be accomplished naturally in the framework of Construction Grammar. First, however, I will return to the discourse-functional hypothesis. 4. The discourse-functional hypothesis As suggested above, the discourse-functional hypothesis makes one wrong prediction with respect to the two genitives, namely that they encode the same set of relations. Still, it seems worthwhile to investigate it a little more closely rather than dismissing it immediately, if only because it has been proposed again and again in the literature, often seemingly independently. In order to test the hypothesis that the two genitives are informationstructure constructions, I selected three well-established correlates of information structure for investigation: first, animacy of the referents of head and modifier (on the scale human > other animate > inanimate > abstract); second, givenness2; and third, length (measured in number of syllables). I extracted fifty examples of each construction from a corpus of spoken American English (the CSPAE, Barlow 1998). I only chose exam-

426

Anatol

Stefanowitsch

pies that theoretically allow an alternation, i.e. that occur in the other construction at least under some circumstances.3 Finally, I coded them for all three features mentioned. The predictions of the discourse-functional hypothesis are straightforward: following the principle that (in English) active referents typically precede inactive ones, the s-genitive should be used in the majority of cases where the modifier NP is respectively higher on the animacy hierarchy, more given, or shorter than the head noun, and the of-genitive should be used where the opposite is the case. If it is true that it is the primary function of the two genitives to package information structure, then we should not only expect the majority of cases to follow these predictions, but this majority should be near-categorical. Consider Table 2, which shows the results of the corpus analysis for each of the three parameters. Clearly, the predictions are not borne out. Beginning with animacy, it is true that for the majority of s-genitives the modifier has a higher animacy value than the head. However, the same holds true for the o/-genitive. The only significant difference between the two constructions is that in the case of the of-genitive there is a higher proportion of cases where there is no difference between head and modifier in terms of animacy. While this is an interesting finding, it is not the result predicted by the discourse-functional hypothesis. Table 2. Results of the corpus analysis Table 2.1.

Animacy (χ 2 = 19.16 (2), ρ < 0.001) Nhead higher

NPm0d higher

N o difference

Total

s-genitive

0%

(0)

90%

(45)

10%

(5)

100%

(50)

of-genitive

2%

(1)

50%

(25)

48%

(24)

100%

(50)

Table 2.2.

Givenness (χ 2 = 16.93 (2), ρ < 0.001) Nhead higher

NPm0d higher

No difference

Total

5-genitive

8%

(4)

88%

(44)

4%

(2)

100%

(50)

o/-genitive

30%

(15)

50%

(25)

20%

(10)

100%

(50)

Table 2.3.

Length ( χ 2 = 1.00 (2), ρ > 0.05, n.s.) N head shorter

NP mod shorter

N o difference

Total

s-genitive

34%

(17)

56%

(28)

10%

(5)

100%

(50)

of- genitive

42%

(21)

46%

(23)

12%

(6)

100%

(50)

Constructional semantics as a limit to grammatical alternation

427

The findings for givenness are similar. Again the modifier is more given than the head for both constructions. The significant difference between the two constructions is that in the case of the s-genitive there is a very clear majority of such cases, while in the case of the o/-genitive the differences between the three possibilities are much smaller (cf. Altenberg 1980 for similar results). Again, this is an interesting finding, but again, it is not the one predicted by the discourse-functional hypothesis. Finally, with respect to length there is a trend toward shorter modifiers for both constructions, and the difference between the two constructions is not significant at all. The discourse-functional hypothesis is clearly disconfirmed by these results: the two constructions are not alternative ways of packaging information flow. However, the results do show that the s-genitive strongly prefers highly active modifiers. This preference has been observed before (cf. Brown 1983 and Taylor 1994a), and it makes sense given Langacker's analysis of the s-genitive as a reference-point construction: if a referent A is to function as a means of accessing another referent B, it is to be expected that A should be highly active. The fact that the of-genitive shows the same preference also makes sense under the reference-point analysis: for the ofgenitive, it is again the modifier that functions as a reference point. However, the fact that the preference is much weaker also points to the fact that the o/-genitive is more than just a reference-point construction.

5. A semantic account of the two genitives of English Despite the apparent difficulties, a semantic account of the two genitive constructions is possible if we accept the idea of inheritance with overrides introduced in section two. Beginning with the s-genitive, note that there are two fundamentally different kinds of examples in Table 1 above. On the one hand, there are those cases where the specific nominals occurring in the construction do not themselves evoke a particular semantic relation: Kate's shoes is understood to mean 'the shoes belonging to Kate', but this relation is not inherently specified by the word Kate or the word shoes. In other words, we can conceptualize Kate without necessarily conceptualizing any of the things she owns, and we can conceptualize a pair of shoes without conceptualizing their owner. On the other hand, there are those cases where one of the nominals does evoke a particular relation: the University's budget is interpreted to mean 'the budget belonging to the university', but here the word

428

Anatol Stefanowitsch

budget itself actually inherently specifies a relation of ownership. In other words, we cannot conceptualize a budget without conceptualizing its owner (i.e. the person or group of persons in charge of spending it). Shoes may or may not belong to someone, but a budget that does not belong to anyone is not a budget but simply an amount of money. Slightly extending traditional terminology, I will refer to words like budget as relational nouns and words like shoes as non-relational nouns. Strictly speaking, it is only in the case of non-relational nouns that the sgenitive can be claimed to encode a relation of ownership. In the case of relational nouns, there is no way of telling whether it is the j-genitive that encodes a relation of ownership or whether it is, for example, the word budget that evokes this relation. Now, note that there are only two relations that can actually be encoded by the s-genitive if the head noun is non-relational: ownership and kinship. If the head noun is inanimate and the modifier animate, the default interpretation is one of ownership, as in Kate's shoes above.4 Where the head noun and the modifier are both animate, the default interpretation is one of kinship: take the examples your Emily or Diane's girl, which will be interpreted to mean 'your daughter Emily' and 'Diane's daughter' respectively.5 Of course, these relations can also be encoded if the head noun is relational, as in the university's budget or Billy's wife. For all other relations, however, the head noun must specify the relation that holds between the referents of the two nominals. A COMPONENT-WHOLE relation can only be encoded if the head noun's semantics includes the fact that it is a component of something: the baby's eyes means 'the eyes that are a part of the baby', and the earth's surface means 'the surface that is a part of the earth', but the words eyes and surface already include the COMPONENT-WHOLE relationship as part of their meaning. If a head noun does not specify this relationship, the s-genitive is odd even if a COMPONENT-WHOLE relation is known to exist between the referents of the head and the modifier. It is difficult to come up with an example, because most words that refer to a component of something do specify such a relationship. But take an example like ??the steel's iron, which is odd because the word iron does not have to be conceptualized as a component of something. In the same context, however, the steel's main component would be fine, because here the head noun does evoke the relevant relationship. The same argument applies to ATTRIBUTE-HOLDER. This relation can only be encoded by the s-genitive if it is inherently specified by the head noun. Kate's coldness means 'the coldness that is an attribute of Kate' but the word coldness (and all other words referring to attributes) inherently

Constructional semantics as a limit to grammatical

alternation

429

specify the ATTRIBUTE-HOLDER relation, i.e. it is not possible to think of an attribute without the entity of which it is an attribute. Finally, the same argument can be made for the PARTICIPANT-EVENT relation: event nominals obviously specify their participants as part of their meaning, and thus expressions like the fire department's investigation will be interpreted as 'the investigation in which the fire department plays a role'.6 To summarize the argument so far, the s-genitive encodes possession (i.e. ownership/kinship) when neither of the nominals evokes a particular semantic relation, and it encodes other relations only if those relations are inherently specified by the semantic class of the head noun or by the individual lexical item functioning as the head noun. In addition, as shown in Table 1 above, there are a number of relations that cannot be encoded by the 5-genitive, namely SUBCATEGORY-CATEGORY, SUBPART-WHOLE, and THING-CONSTITUENT. I will return to this issue at a later point, but first, I will suggest how the facts discussed so far can be naturally accommodated by a Construction Grammar approach. Essentially, what needs to be accounted for is the fact that the i-genitive is interpreted as encoding possession by default, i.e. unless the head noun specifies a different semantic relation. Consider Figure 4, which shows what such an account might look like. Sem REFERENCE-POINT RELATION

< possessee

possessor > ι




t Nhead

Nptod

Figure 4. The s-genitive

The analysis follows Langacker (1993, 1995) in assuming that the sgenitive encodes a reference-point relation between the modifier and the head, but it differs from this analysis in that, in addition, it assigns the role POSSESSEE to the head noun and the role POSSESSOR to the modifier (the modifier simply encodes the most natural reference point for the head, which in the case of a possessee is a possessor). This construction accounts for the interpretation of s-genitives with non-relational nominals, such as Kate's shoes and Diane's girl, which receive their semantic roles from the construction itself.

430

Anatol Stefanowitsch

The fact that an s-genitive with a relational head noun is not interpreted as encoding the POSSESSOR-POSSESSEE relation is due to the principle of inheritance with overrides. A relational noun, like any lexical item, is a construction in its own right, and it evokes a semantic frame in which it occupies a particular semantic role, but which in addition includes one or more other participants. Since lexical items (or classes of items) are more specific than unfilled grammatical constructions, the semantic roles specified by a relational noun will override the roles specified by the construction. For example, a noun encoding an attribute, such as coldness, assigns the role attribute to itself and evokes a frame that includes at least one other participant, namely the holder of the attribute. The reference-point relation encoded by the s-genitive picks out this second participant as the most natural reference point, and assigns its semantic role to the modifier. This state of affairs is shown in Figure 5. Sem REFERENCE-POINT < RELATION

{ Syn

I

possessee

possessor >

I attribute

holder

Nhead

NPmod

}

Figure 5. The s-genitive of attribute

The o/-genitive can be accounted for in a parallel fashion. This requires a well-defined notion of intrinsic relation. I will use this term in a very limited way: an intrinsic relation is the relation between an entity and the smaller entities which it consists of or the larger entity which it is a part of. For example, the concept wall has an intrinsic relation to the bricks out of which it is built as well as to the building of which it is a part. In other words, an intrinsic relation is the relation between any two contiguous entities in a chain of part-whole relations. Given this definition, the o/-genitive can be characterized as a construction encoding a reference-point relation between one entity and another entity which is intrinsically related to the first entity. This analysis can be represented as shown in Figure 6 (intrinsic entity here stands for the second entity which is intrinsically related to the first). This construction accounts for those uses of the o/-genitive where the nominale themselves do not specify a particular semantic relation: SUBPART-WHOLE and THING-CONSTITUENT, which are simply two different

Constructional semantics as a limit to grammatical

alternation

431

manifestations of intrinsic relation. Beginning with the latter, consider the example a dress of silk. Neither dress nor silk evoke a THING-CONSTITUENT relation, i.e. we can conceptualize a dress without paying any attention to the material of which it consists, and we can conceptualize silk without conceptualizing an entity consisting of silk. It is thus the o/-genitive itself which provides the meaning THING-CONSTITUENT. Of course, head nouns that do specify this relation as part of their semantics can also occur in the construction, as in a constellation of stars or an array offlowers. Sem

REFERENCE-POINT RELATION

Syn

Of

t


entity

I

NPmod

Figure 6. The o/-genitive as a semantic role construction

Turning to the SUBPART-WHOLE relation, note that it often occurs with head nouns that seem to specify this relation: a big chunk of the company means 'the big chunk that is a (sub)part of the company', and the head noun chunk already evokes a relation to some larger entity. Similarly, fifty percent of the population means 'the fifty percent that are a subpart of the population', and again, fifty percent inherently specifies the SUBPARTWHOLE relation as part of its lexical meaning. However, where the head noun does not specify this relation, the construction provides it. Consider examples like a glass of water and a bowl of oranges. Glass and bowl do not inherently specify that they are (sub)parts of a larger entity, but when they occur in the o/-genitive, they are interpreted as such: they can be paraphrased as 'the subpart of all water which is contained in the glass' and 'the subpart of all oranges which is contained in the bowl'. 7 1 will return below to seemingly relational quantity nouns like chunk and percent and show that even for these, the interpretation as subparts is in part due to the construction. First, note that relations other than SUBPART-WHOLE or THINGCONSTITUENT can be accounted for in the same way as they were for the sgenitive. They can only be encoded if the head noun specifies a relation, which then overrides the construction's semantics. This is shown in Figure 7 for the ATTRIBUTE-HOLDER relation, accounting for examples like the beauty of the desert.

432

Anatol

Stefanowitsch

Sem REFERENCE-POINT < RELATION

entity

< attribute Syn

t of

f

Nhead

intrinsic > entity I

holder >

I

NPmod

Figure 7. The o/-genitive of attribute

This analysis also explains why the o/-genitive can encode POSSESSEEPOSSESSOR and INTERPERSONAL RELATION only in some instances, but not in others: the budget of the university is possible, since, as noted earlier, budget already evokes a relation of ownership and can thus override the semantics of the o/-genitive. In contrast, *the shoes of Kate is not possible, because shoes does not evoke an ownership relation. Therefore, the ofgenitive assigns the semantic relation SUBPART-WHOLE or THINGCONSTITUENT to the expression, neither of which is readily interpretable (and neither of which, of course, expresses the intended relationship of possession). Two issues remain to be addressed. First, there are some restrictions on the set of semantic relations encoded by the s-genitive that I have referred to above, but not yet accounted for. Second, section three pointed out a problem for the prototype analysis that has not been resolved: on the one hand, for both constructions there are semantic relations that cannot be encoded; on the other hand, both constructions can encode a seemingly open-ended set of relations given the right context. The s-genitive cannot encode SUBPART-WHOLE, SUBCATEGORYCATEGORY, and THING-CONSTITUENT. For the first two relations, this may seem problematic, since for both of them there are nouns that seemingly specify the respective relation as part of their meaning. For example, percent or chunk seem to evoke a SUBPART-WHOLE relation, and thus they should be able to encode this relation when they occur in the s-genitive. They should be able to override the construction's semantics just like nouns evoking ATTRIBUTE-HOLDER or PARTICIPANT-EVENT do. However, if these nouns occur in the s-genitive, they do not in fact override its semantics, but actually encode a POSSESSOR-POSSESSEE relation. Expressions like the population's fifty percent and the company's big chunk are not actually generally unacceptable, they are only unacceptable under a SUBPARTWHOLE interpretation. They are acceptable if the intended interpretation is 'the fifty percent that belong to the population' and 'the big chunk that

Constructional semantics as a limit to grammatical

alternation

433

belongs to the company'. Similarly, nouns like kind or brand seem to evoke a SUBCATEGORY-CATEGORY relation and should also be able to override the semantics of the ί-genitive to encode this relation. However, to the extent that they can occur in the s-genitive, they encode interpersonal relations, as in examples like his kind, his type, and his brand (of people), meaning 'the kind of people that are like him'. The obvious solution to this problem would be to claim that nouns like percent, chunk, kind, and brand are not inherently relational. This would still allow them to be used as terms of quantity in the o/-genitive, where the construction adds the SUBPART-WHOLE relation, while it would also account for the fact that they receive a reified interpretation as a possessee in the 5-genitive. However, this claim is somewhat counter-intuitive. Instead, I would make the weaker claim that such nouns do not uniquely evoke the SUBPART-WHOLE or SUBCATEGORY-CATEGORY relation in the way that a noun like beauty uniquely evokes the ATTRIBUTE-HOLDER relation. Take quantity nouns like percent and chunk. The difference between the referents of these nouns and those of nouns which actually refer to a component of a larger entity is that the latter are clearly differentiated from the larger entity. The leg of a table, for example, has a certain internal structure that makes it different from other parts of the table. Nouns of quantity, in contrast, do not refer to such a component, but to an arbitrary portion of some larger mass that is not objectively different in any way from other portions of that larger mass. This portion only becomes a subpart by virtue of a conceptualizer who imposes a division onto the undifferentiated mass. This means that nouns like percent or chunk evoke not just a larger entity, but also a person who imposes a division (physically or conceptually) or for whose benefit such a division is imposed. The o/-genitive naturally picks out the relation to the larger mass, since this is the one that is compatible with the meaning of the construction. The s-genitive, on the other hand, picks out the relation to the person for whose benefit the division is made. If someone owns fifty percent of something, then those fifty percent become a subpart precisely because they are owned by that someone, i.e. their relation to their owner is the only thing that makes them different from the rest of the entity or mass. In the case of terms that denote a (sub)category, like kind or sort, the relation to a larger category that they evoke is also not unique: they evoke, in addition, a relation to their individual members. In other words, a category can be defined by reference to a larger category that includes it, or by reference to the individual entities that it itself includes. The o/-genitive naturally picks out the relation to a larger category, since this corresponds

434

Anatol

Stefanowitsch

to the construction's meaning, while the s-genitive picks out the relation to the individual member (in those cases where this relation is an interpersonal one as in his kind). The last relation to be accounted for is DEPICTION-DEPICTED. This relation can always be encoded by the o/-genitive, as in a picture of Lisa, a picture of the table, the footage of the riots. In contrast, it can be encoded by the s-genitive only under certain circumstances, namely when the DEPICTED is a human being and when the picture either serves as a means of identification {They took our pictures) or has special significance to someone close to the DEPICTED {She kept his picture close to her heart). Where this is not the case, the s-genitive is unacceptable {??the table's picture, ??the riots 'footage). Assuming that nouns referring to depictions are inherently relational, i.e. that they necessarily evoke the thing they depict, we would expect them to be able to occur in both genitives, overriding their respective semantics. In the case of the of-genitive, this is indeed the case. The question is why the s-genitive does not allow this. I have no final answer, but it seems to me that the fact that the pattern [NP's Ndepiction] has restrictions that are not predictable from the s-genitive or from nouns of depiction is evidence that this pattern is a construction in its own right. This construction, being more specific than the s-genitive, overrides the semantic roles provided by the latter as well as adding semantic content of its own. While the account developed here cannot explain the behavior of nouns of depiction in the s-genitive, it can naturally accommodate it: systematic exceptions to general constructions in the form of more specific constructions are expected.8 Finally, let us turn to the issue of the potential open-endedness of the set of relations encoded by the two genitives. This can be accounted for by looking more closely at the difference between relational and non-relational nouns. This distinction is not a binary one, as has been implicitly assumed in the above discussion. Instead, nouns evoke relations to other participants to varying degrees. Some nouns evoke such relations so strongly that they cannot be conceptualized without their relation to other participants, for example (i) nouns referring to attributes, component parts, and events, (ii) kinship terms and many other words from the domain of interpersonal relations, such as colleague, opponent, etc., as well as (iii) individual lexical items from many semantic domains, e.g., container, assets, constellation, and many others. In contrast, nouns like shoe or glass do not evoke a relationship to another entity so strongly that they cannot be conceptualized without it, but they weakly evoke a number of such relationships by virtue of our world knowledge about them. It is a salient aspect of our knowledge

Constructional

semantics as a limit to grammatical

alternation

435

about shoes that people wear them, and increasingly less salient aspects are that some people have strong feelings about them and that they can be bought on sale, etc. A salient part of our knowledge about trains is that people ride on them to get somewhere, and less salient aspects are that someone services them regularly, that they are driven by someone, etc. Given the right context, any of these relations can be evoked as strongly as if such nouns were true relational nouns, and the relations thus evoked, like Shoe-Wearer, Shoe-Admirer, Train-Passenger, or Train-Engineer, can override the semantics of the construction.

6. Information structure revisited Some of the authors who have analyzed the two genitives as informationstructure constructions implicitly or explicitly recognize that there are semantic relations that can only be encoded by one of the two constructions (e.g., Altenberg 1980; Rosenbach, this volume). These authors often discard those relations from consideration, focusing on (a subset of) the relations for which there is a choice between the two constructions. This is a valid strategy, since of course the question remains as to what determines the choice in such cases. However, this strategy essentially recognizes the fact that the two genitives are primarily semantic-role constructions without providing an explicit account of their semantics. The preceding section has provided such an explicit analysis and explains why some semantic relations can be encoded by both constructions and some cannot. However, it does not account for the choice between the two constructions in those cases where a relation can be encoded by both. In order to address this issue, I chose a relation that always and unambiguously overrides the semantics of the two genitives, ATTRIBUTE-HOLDER. I extracted 50 examples of each construction encoding this relation from the same corpus used in section three. The same criteria for selection were used. I then coded the examples for the same three parameters as before. The results are shown in Table 3. Table 3. Information structure for the ATTRIBUTE-HOLDER relation Table 3.1.

Givenness (χ 2 = 2.35 (2), ρ > 0.05, n.s.) Nhead

s-genitive o/-genitive

higher

NPmod

higher

N o difference

Total

12%

(6)

64%

(32)

24%

(12)

100%

(50)

4%

(2)

66%

(33)

30%

(15)

100%

(50)

436

Anatol

Table 3.2.

Stefanowitsch

Length (χ 2 = 23.12 (2), ρ < O.Ol,***) N head shorter

NP mod shorter

N o difference

Total

s-genitive

10%

(5)

82%

(41)

8%

(4)

100%

(50)

of- genitive

50%

(25)

36%

(18)

14%

(7)

100%

(50)

Table 3.3.

Animacy ( χ 2 = 84.72 (2), ρ < 0.01,***) NP m o d animate

NP mod inanim

NP mod abstract

Total

s-genitive

96%

(48)

2%

(1)

2%

(1)

100%

(50)

o/-genitive

2%

(1)

26%

(13)

72%

(36)

100%

(50)

The results clearly show that givenness is in general not a deciding factor even in choice contexts. The modifier is more given in the majority of instances for both constructions, and the difference between them is not significant. However, the other two parameters now show the distribution predicted by the discourse-functional hypothesis: in the majority of the cases of the sgenitive the modifier is shorter than the head, and vice versa for the ofgenitive (although this majority is much smaller in the case of the latter). More tellingly, the majority of s-genitives have animate modifiers while the majority of o/-genitives have inanimate or abstract modifiers. In fact, the difference between the two constructions with respect to this parameter is near-categorical. A closer look reveals that this difference is in fact also responsible for the distribution of long vs. short modifiers: the average length of all animate nominals in the sample is 2.9 syllables, the average length of all inanimate and abstract nominals is 4.5 syllables. The difference in length thus merely reflects the difference in animacy. The two constructions thus seem to reflect the general preference of English (and other topic-initial languages) for animate referents to precede inanimate referents where possible (Rosenbach's experimental results confirm this)9, moreover this preference is near absolute. In concluding this section, let me briefly return to the issue of givenness, though. Note that there are three exceptions to the general pattern of animate modifiers occurring with the s-genitive and inanimate modifiers occurring in the o/-genitive: there are two examples of s-genitives with inanimate/abstract modifiers, and there is one example of the o/-genitive with a human modifier. Let us look at these examples in detail, beginning with the s-genitive. The relevant examples are shown in (2):

Constructional

(2)

semantics as a limit to grammatical

alternation

437

a. The National PTA believes there are important roles for the federal government to play, as well as state and local education agencies, in ensuring assessment integrity. NAEP's value is in providing national and state trend data... b. There are two general issues there ... [3 clauses omitted] So the issue there is ... [2 clauses omitted] And the other concern is ... [1 clause omitted] I think of those two concerns ... [9 clauses omitted] The next issue ... [6 clauses omitted] As we move through the report, you can see there is a section on pages 3 and 4. That is the issues' relevance specifically to the grade 8 mathematics test

Both examples would be encoded by the o/-genitive if they followed the general animacy preferences: the value of the NAEP (i.e. the 'National Assessment of Educational Progress') and the relevance of the issues. Note that without an accompanying context, these actually sound more natural. However, in both examples, the referent of the modifier is highly active: in (2a) the topic of the discussion is educational assessment, and the NAEP has been mentioned countless times in the preceding discourse; in (2b), the speaker has been talking about the referent of the modifier for twenty-one clauses at the time that the s-genitive occurs. It seems plausible, then, to claim that the violation of the animacy preference in these examples is due to the extremely high activation of the modifier's referents. However, note that there is no general tendency to violate animacy preference with highly active modifiers: there are six cases in the sample where the head is more highly activated than the modifier but that are encoded by the ^-genitive anyway, subordinating the given-first principle to the animacy preference of the j-genitive. Turning to the o/-genitive, a similar argument can be made. Consider the relevant example in (3): (3)

And actually, the composition of the panel... There are people who have been involved in the NSF initiative curricula writing. There are people who have been involved in writing basal text, algebra. There are people who have worked in systemic initiatives both at state or urban systemic initiatives. That there are people representing the mathematics community. If you look at the backgrounds of the people...

438

Anatol

Stefanowitsch

Here, the more natural choice in terms of animacy would be the people's backgrounds or their backgrounds. However, the referents of backgrounds are highly active, since the speaker talks about these backgrounds in the preceding four sentences. Again, it is plausible that the high activation of the head influences the choice of construction here, but again, this is an individual example, not a general principle (note that Standwell 1982 and Osselton 1988 base their claims on the discussion of precisely this type of isolated example). 6. Conclusion The two genitives of English are clearly not information-structure constructions, i.e. their elements are not inherently associated with particular activation states. Instead, they are semantic-role constructions: the s-genitive assigns the roles POSSESSEE and POSSESSOR to its head and modifier respectively, and the of-genitive assigns roles that I have called, for want of a better term, ENTITY and INTRINSIC ENTITY. The fact that these two constructions have so often been analyzed as information-structure constructions is due to two facts about their semantics: first, as for other unfilled constructions, the semantic roles they assign can be overridden by those associated with particular semantic classes of words or with individual lexical items occurring in them. Since the two genitives are used to express nothing more than a relation between two participants and since a vast number of nouns evoke specific relations, the meanings of these two constructions are overridden much of the time. This fosters the impression that they do not actually have any meaning at all, which encourages an explanation in terms of information structure. However, the meaning of the two constructions becomes apparent when nonrelational nouns occur in them. Second, the fact that the semantics of the two constructions can be overridden means that they can in many cases encode the same semantic relation. Such cases again encourage an explanation in terms of information structure. A closer look at one such case has shown that animacy is a strong factor in determining the choice between the two constructions where they encode the same semantic relation: the constructions follow the general preference of English to put animate nouns before inanimate ones. The fact that this preference is virtually absolute, at least with the ATTRIBUTE-HOLDER relation, indicates that it may not be an on-line processing phenomenon, but may actually be grammaticized to a large degree.

Constructional semantics as a limit to grammatical

alternation

439

Givenness was not found to be a decisive factor in this study. This is not to say that it cannot be a factor in individual instances, as the discussion of examples (2) and (3) suggests: it seems that with highly active modifiers speakers sometimes optionally choose an s-genitive where an o/-genitive is expected or, with highly active heads, an o/-genitive where an ί-genitive is expected. The fact that such choices are optional shows that, at least in the case of the two genitives, the influence of on-line information packaging is heavily limited by semantic factors. It seems that information structure can influence linguistic structure in two different ways. On the one hand, it can manifest itself in the form of information-structure constructions like the English verb-particle constructions. In such cases, its influence is substantial (though perhaps not exclusive, cf. again Gries, this volume), and language structure will reflect the activation state of referents very directly. On the other hand, information structure can manifest itself as an optional preference in the choice between constructions whose primary difference is to be found elsewhere. In such cases, its influence is extremely tenuous, making it look like a stylistic principle rather than a fundamental cognitive mechanism.

Notes 1. Not all versions of Construction Grammar assume this type of inheritance. However, it follows naturally from the assumption that constructions are abstracted over specific instances in language learning. If a number of structurally similar expressions share a particular semantic property, then their formal and semantic similarities will be abstracted as a more schematic representation and henceforth categorize the more specific instances (e.g., Langacker 1987: 66-71). However, any specific instances which do not fit this more general schema will of course retain their conflicting properties. 2. In order to calculate givenness, a hierarchical combination of methods was used: first, a Givön-style text count of (i) number of clauses since the last mention and (ii) number of mentions in the subsequent ten clauses; second, a careful interpretative assessment of which of the two nominals in a given example referred to what the current stretch of discourse was about (if there was a difference between the two nominals in this respect). If the two nominals differed with respect to the first text-count criterion, they were coded according to this criterion. If there was no difference, they were coded according to the second criterion. If there was again no difference, they were coded according to the interpretative assessment. 3. This was determined by checking for each example from the CSPAE whether the 250million-word North-American News Corpus contained at least one example of the opposite construction with the same head noun and a modifier of exactly the same semantic type. Examples with possessive pronouns were ignored, as were fixed expressions (like Chairman of the Board), since for both types of case there is no possibility of alternation. Also ignored were immediate verbatim repetitions and tokens in the immediate

440

4. 5.

6.

7.

8.

9.

Anatol Stefanowitsch

vicinity of performance errors, self-corrections etc., since the increased cognitive load associated with these discourse phenomena could presumably interfere with information flow. I will return to the fact that, as noted earlier, an expression like Kate's shoes can actually refer to any number of semantic relations. The examples actually do not necessarily encode a parent-child relationship, they could also encode a relationship between spouses or lovers. However, they cannot encode any kind of INTERPERSONAL RELATIONSHIP, say, one between colleagues or acquaintances. I will not discuss the difference between subjective and objective genitives here. Typically, both for s- and o/-genitives the participant encoded by the modifier could be the Agent or the Patient given the right context. This does not mean that both constructions encode both relations equally frequently, but this does not bear on the issues discussed here (cf. Taylor 1994b for discussion). At first glance, this analysis may seem unnecessarily complicated: glass and bowl may be argued to evoke a CONTAINER-CONTAINED relationship, and thus these examples may simply be analyzed as examples of this relation. However, if this were the case, then we should be able to refer to a bowl containing a single orange as *a bowl of an orange. The fact that this is not possible shows that the word bowl does not evoke a CONTAINERCONTAINED relation, but instead is assigned the role Subpart by the construction. In a bowl of oranges, oranges refers to an undifferentiated whole from which a bowl can pick out a subpart. In *a bowl of an orange, the single orange is not conceptualized as an undifferentiated whole, and hence a bowl cannot pick out a subpart. Another example of such a more specific construction is the Time-Event relation, which can only be encoded by the s-genitive, as in last year's Olympics, or last night's arrest. There are two reasons for positing [NPtjme's N] as a construction in its own right: first, in all other cases it is the head that evokes the relation that overrides the s-genitive's semantics, but here it seems to be the modifier. Second, the construction has heavy restrictions on the kinds of temporal nouns that can occur in it: note the unacceptability of *midnight's arrest, *June 25th's explosion. An example of a more specific construction related to the o/-genitive is the one instantiated by expressions like an angel of a woman, where the modifier NP is semantically the head. Rosenbach also finds an influence of topicality (i.e. givenness). This discrepancy between her results and mine may be due to the different text types under investigation. Her study is based on a questionnaire giving a minimal context and requiring subjects to choose between the two constructions. This is an off-line task based on written language, and I would assume that the subjects tap into a different kind of knowledge than the speakers in the CSPAE, who are producing spontaneous discourse. A possible explanation is that the subjects in Rosenbach's experiment are influenced by stylistic considerations that reflect knowledge of writing conventions rather than reflecting linguistic processing.

References Altenberg, Bengt 1980 Binominal NPs in a thematic perspective: Genitive vs. o/-constructions in 17th century English. In: Sven Jacobson (ed.), Papers from the Scandinavian Symposium on Syntactic Variation, 149-172. (Stockholm Studies in English 52.) Stockholm: Almqvist and Wiksell.

Constructional semantics as a limit to grammatical

alternation

441

Barlow, Michael 1998 Corpus of Spoken, Professional American English. Houston, TX.: Athelstan. Brown, Cheryl 1983 Topic continuity in written English narrative. In: Talmy Givön (ed.), Topic Continuity in Discourse. A Quantitative Cross-Language Study, 315-341. (Typological Studies in Language 3.) Amsterdam/Philadelphia: Benjamins. Chafe, Wallace 1994 Discourse, Consciousness, and Time. The Flow and Displacement of Conscious Experience in Speaking and Writing. Chicago/London: The University of Chicago Press. Chen, Ping 1986 Discourse and particle movement in English. Studies in Language 10: 7 9 95. Chomsky, Noam 1970 Remarks on nominalization. In: Roderick Jacobs and Peter Rosenbaum (eds.), Readings in English Transformational Grammar, 184-221. Waltham, MA.: Ginn. 1986 Knowledge of Language. Its Nature, Origin, and Use. New York: Praeger. Deane, Paul 1987 English possessives, topicality and the Silverstein hierarchy. Proceedings of the Annual Meeting of the Berkeley Linguistics Society 13: 65-76. 1992 Grammar in Mind and Brain. Explorations in Cognitive Syntax. Berlin/ New York: Mouton de Gruyter. Givön, Talmy 1993 English Grammar. A Function-Based Introduction. Amsterdam/ Philadelphia: Benjamins. Goldberg, Adele 1995 Constructions. A Construction Grammar Approach to Argument Structure. Chicago/London: The University of Chicago Press. 1996 Construction grammar. In: Keith Brown and Jim Miller (eds.), Concise Encyclopedia of Syntactic Theories, 68-71. Oxford: Pergamon Press. Gries, Stefan this vol. Grammatical variation in English: A question of'structure vs. function'? Hawkins, Roger 1981 Towards an account of the possessive constructions: NP's Ν and the Ν of NP. Journal of Linguistics 17: 247-269. Hopper, Paul 1998 Emergent grammar. In: Michael Tomasello (ed.), The New Psychology of Language. Cognitive and Functional Approaches to Language Structure, 155-175. Mawah, N.J.: Erlbaum. Hudson, Richard A. 1984 Word Grammar. Oxford/Cambridge, MA.: Blackwell. Jespersen, Otto 1949 A Modern English Grammar on Historical Principles, Volume 7, Syntax. Copenhagen: Munksgaard. Jucker, Andreas H. 1993 The genitive versus the o/-construction in British newspapers. In: Andreas H. Jucker (ed.), The Noun Phrase in English: Its Structure and Variability, 121-136. (Anglistik und Englischunterricht 49.) Heidelberg: Winter.

442

Anatol Stefanowitsch

Kempson, Ruth Μ. 1977 Semantic Theory. Cambridge: Cambridge University Press. Lakoff, George and Mark Johnson 1980 Metaphors We Live by. Chicago: The University of Chicago Press. Lambrecht, Knud 1994 Information Structure and Sentence Form. A Theory of Topic, Focus, and the Mental Representation of Discourse Referents. (Cambridge Studies in Linguistics.) Cambridge: Cambridge University Press. Langacker, Ronald W. 1990 Concept, Image, and Symbol. The Cognitive Basis of Grammar. Berlin/New York: Mouton. 1992 The symbolic nature of cognitive grammar: The meaning of of and ofperiphrasis. In: Martin Pütz (ed.), Thirty Years of Linguistic Evolution. Papers in Honour of Rene Dirven, 483-502. Amsterdam/Philadelphia: Benjamins. 1993 Reference-point constructions. Cognitive Linguistics 4: 1-38. 1995 Possession and possessive constructions. In: John R. Taylor and Robert E. MacLaury (eds.), Language and the Cognitive Construal of the World, 5179. Berlin/New York: Mouton de Gruyter. Levin, Beth 1993 English Verb Classes and Alternations. A Preliminary Investigation. Chicago/London: The University of Chicago Press. Nikiforidou, Kiki 1991 The meanings of the genitive: A case study in semantic structure and semantic change. Cognitive Linguistics 2: 149-205. Osselton, Noel E. 1988 Thematic genitives. In: Graham Nixon and John Honey (eds.), An Historic Tongue: Studies in English Linguistics in Memory of Barbara Strang, 138144. London: Routledge. Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech and Jan Svartvik 1985 A Comprehensive Grammar of the English Language. Harlow: Longman. Rosenbach, Anette this vol. Aspects of iconicity and economy in the choice between the j-genitive and the o/-genitive in English. Siewierska, Anna 1988 Word Order Rules. London: Croom Helm. Standwell, Graham J.B 1982 Genitive constructions and functional sentence perspective. International Review of Applied Linguistics 20 (4): 257-261. Stefanowitsch, Anatol 1997 On the genitive constructions in English. A cognitive linguistic approach. MA thesis, Seminar ftir Englische Sprache und Kultur, Universität Hamburg. 1998 Possession and Partition: The Two Genitives of English. (Cognitive Linguistics: Explorations, Applications, Research 23.) Seminar fur Englische Sprache und Kultur, Universität Hamburg. Strang, Barbara M.H. 1962 Modern English Structure. London: Edward Arnold.

Constructional semantics as a limit to grammatical

alternation

443

Taylor, John R. 1989a Possessive genitives in English. Linguistics 27: 663-686. 1989b Linguistic Categorization. Prototypes in Linguistic Theory. Oxford: Clarendon. 1994a Possessives and topicality. Functions of Language 1: 67-94. 1994b 'Subjective' and 'objective' readings of possessor nominals. Cognitive Linguistics 5 (3): 201-242. 1996 Possessives in English. An Exploration in Cognitive Grammar. Oxford: Clarendon Press.

Principles of grammaticalization and linguistic reality* Olga C.M. Fischer

Abstract This paper considers a number of deterministic conceptions that occupy a central position in current thinking about the process of grammaticalization, both in formal and functional theories of grammar. After a general discussion of the way the phenomenon of grammaticalization is dealt with from the point of view of grammar change and language change, and the explanatory value of these two rather different approaches, the paper turns to an examination of determinants considered to play a role in grammaticalization, i.e. the principle of unidirectionality, the idea of conceptual chains (grammaticalization as a semantically driven process), of grammaticalization as a mechanism or cause in itself, and the so-called parameters of grammaticalization. These assumptions will be critically examined with the help of two case studies, i.e. the grammaticalization of the infinitival marker to and of semimodal have to in the history of English. In addition, other factors will be looked at of an essentially synchronic nature, which may interact in this diachronic process, such as iconic factors and the synchronic state of the grammar/language. Both of these play an important part in the way grammaticalization proceeds. The paper concludes that certain tendencies can indeed be discerned in grammaticalization, but that the process is first and foremost steered by the shape of the synchronic language system. The conclusion also offers some thoughts on how the synchronic factors that steer grammaticalization may yet set off a longterm development through the implicational properties of the structure that is grammaticalizing.

1. Introduction The organizers of the symposium on Determinants of Grammatical Variation in English stressed in their introduction that "rule-based deterministic conceptions of grammar do not adequately reflect linguistic reality" and also that the "[m]ore promising models are those that assume a wide variety of more or less closely linked interacting factors determining grammatical variation". In my contribution to this debate, I will consider a process in which each developmental stage presents a choice of variants, i.e. the phenomenon of grammaticalization. In grammaticalization, layering (that is, the synchronic presence of diachronic variants expressing the same meaning or linguistic function) plays an important role. My present aim is to find out what determines at each stage the choice between variants, and thus what determines the next stage of the process. In doing this I will take issue

446

Olga C. Μ. Fischer

with a number of deterministic conceptions that play a role in the way the process of grammaticalization has been conceived of in the literature, both in formal and functional theories of grammar. Especially I would like to critically consider the ideas of unidirectionality, of conceptual chains (grammaticalization as a semantically driven process), of grammatical ization as a mechanism or cause in itself or as an epiphenomenon, and the socalled parameters of grammaticalization. In addition, I will indicate what other factors may interact in this diachronic process which are of an essentially synchronic nature, such as iconic factors and the state of the synchronic grammar. Both of these play an important part in the way grammaticalization proceeds. What follows will be based on two case studies of grammaticalization in English that I have investigated in earlier work, i.e. the grammaticalization of infinitival to (cf. Fischer 1997, 2000) and of the semi-modal have to (cf. Fischer 1994b). In my discussion I will lean heavily on the data gathered for those two studies. My intention here is to concentrate on their results and to discuss the consequences of these for the determinative factors or principles under discussion here. 2. Determinants in grammaticalization Most functionally oriented linguists stress the fact that grammaticalization processes are unidirectional (for some this is even a principle of grammaticalization, cf. especially Haspelmath 1999), and that they are mechanistic and semantically driven, while more formally inclined (including generative) linguists tend to see grammaticalization as an epiphenomenon: i.e. grammaticalization itself is not a mechanism, nor a cause for change, rather it is the accidental result of a number of common changes such as reanalysis, analogy, phonetic reduction, etc., which cluster together and thus result in a process that could be called grammaticalization (cf. especially Newmeyer 1998: chapter 5, but also less formally oriented linguists such as Harris and Campbell 1995: 20, Campbell 2001). In some recent generative publications (notably Roberts and Roussou 1999, and see also Roberts 1993), grammaticalization seems to have become equivalent to one mechanism in the process namely re-analysis. Whereas Abraham (1993: 13) - who also belongs to the generative school is still more tentative and describes re-analysis as a "subcomponent" of the process of grammaticalization, i.e. that part which "can be handled exclusively by means of clear categorial distinctions and a restructuring of the constituents",1 for Roberts and Roussou it seems to have become the main mechanism. They assume that "grammaticalization involves some sort of Λ

Principles of grammaticalization

and linguistic reality

447

categorial reanalysis of lexical material as functional material" (1999: 1014). Instead of seeing such a re-analysis as the result of a number of different factors or determinants, including as Abraham (1993: 7-8) notes "die Zwischenstufen" ('intermediate steps', involving semantic, iconic and pragmatic factors), Roberts and Roussou do not consider these to be part of the grammaticalization process itself, co-steering the re-analysis. (Or, to put it in generative terms, these factors are not considered to be triggers of change - for a definition of triggers, see note 6). Rather, they explain the re-analysis in terms of "structural simplification", which is a "natural mechanism of change" (p. 1014). It seems to me that this pushes the explanation one step back because the question inevitably arises: where does the necessity for structural simplification come from? For structural simplification to fulfil this natural role, it has to be always present in any impending situation of change, even before the re-analysis. What they need to explain, in other words, is what stops the principle from applying earlier. In their explanation of grammaticalization, they therefore argue from the other direction, as it were. The quest is not so much for triggers that set the grammaticalization into motion, but for factors that prevent the need for structural simplification to apply earlier (cf. p. 1023). In their description of four cases of grammaticalization, there is usually one crucial factor that prevents an earlier re-analysis, and this factor is always syntactic (and unitary). For instance, in the case of the English modals, the crucial causal factor is the loss of the infinitival -en ending (p. 1024), or, to put it differently, the loss of -en caused opaqueness which moved structural simplification into action. In the development of subject agreement markers out of pronouns in North West Romance, it is the loss of a local movement rule due to the paradigm reduction of subject clitics. Movement rules are seen in this framework as "a marked option" (p. 1021) because they involve adjunctions (p. 1020), ergo the loss of such a rule is always a structural simplification. Most of the cases discussed by Roberts (1993) and Roberts and Roussou (1999) indeed involve the loss of a movement rule, as Haspelmath (1999: 1053) notes, "... their central proposal is that grammaticalization changes can generally be understood as reanalyses involving a structural simplification, especially as involving fewer movement operations." It is a neat way of accounting for the process of grammaticalization, because we have to deal with only two syntactic factors, the re-analysis itself and the syntactic factor that prevented it, but can it really also be said to provide "an enlightening account of grammaticalization" (p. 1022)? One wonders whether the strictly formal methodology is an aid to understanding the complex process of grammaticalization or an obstruction.3

448

Olga C. Μ. Fischer

Structural (generative) linguists reject the diachronic process-aspect of grammaticalization. They argue that when we look at this phenomenon from the point of view of speakers or learners who can only base their grammar on or deduct it from the synchronic output they hear and who cannot be aware of historical processes taking place in a language, the only possible conclusion is that grammaticalization cannot exist as a principle or mechanism of change in the form proposed by functional linguists. Where functional linguists working within grammaticalization theory see the process as a "chain", which is "the result of conceptual manipulation" (cf. Heine et al. 1991: 171, 174 and passim), generative linguists are only interested in the links that make up the composite chain, and they see these links as independent of one another, i.e. for them there isn't really a chain. The reason for this is, first of all, that the notion of conceptual manipulation is difficult to incorporate into a formal theory of grammar in which the semantic module is an interpretative one (with pragmatics being more or less ignored), and in which the syntactic module is seen as central.4 Secondly, generative theory accepts a rather simple and strict model for language acquisition, and a highly abstract (simple and elegant) model of grammar; in this model there is no space for variation and diachronic developments, and hence no way in which a (diachronic) chain of changes could be incorporated. Because of these last two factors, there is, first of all, little room for changes taking place in the grammar of the speaker after the so-called critical period (if change is possible after this period, it is of an additive rather than a radical nature), and secondly, because of the distance between the concrete performance data and the abstract grammar, it is difficult if not impossible to account for small changes in the output (consequently, they are often ignored).5 These small changes may not have an immediate effect on the abstract rules of grammar (the question then of course is: how do they arise, how does the grammar generate the changed constructions?), but when these changes on the output level increase, and when variation increases, this may eventually result in a change in the grammar, but only at the end of the line, so to speak. In other words, what generative linguists look at is the purely grammatical or formal result of the various processes that have been playing around, while functional linguists are more interested in these processes themselves, and in the variation, which may eventually lead to a more radical (or deep) formal change in the grammar. These differences can be captured in another way. We can study language change from the point of view of the speaker. We then concentrate on his competence and on the grammar that he deducts from the surround-

Principles of grammaticalization

and linguistic reality

449

ing (synchronic) output. Let us call this the study of grammar (or competence) change. In order to describe grammar change we must take account of the shape of the theory of grammar and how this interacts with the output presented to the learner. Changes on the performance level may then act as triggers for grammar change. The explanation for any change lies in the interaction of these two entities: the speaker's grammar (which includes the principles of UG) and the triggering experience. In other words, the variation that causes language change is not important in its own right unless it provides a trigger,; consequently, according to this view, variation may be ignored if it does not lead to deeper (i.e. grammar) change. The theory of grammar thus plays a most crucial role here because it is the theory that decides which triggers are of interest and which are not.6 Since the emphasis in generative theory is on syntax, it is almost inevitable that the changes that constitute triggers are also themselves syntactic (cf. the work of Roberts and Roussou discussed above). This then leads to a neglect of other factors (semantic, pragmatic, iconic) that functionally oriented linguists see as the main triggers or determinants in grammaticalization. The question next is, what decides whether a change constitutes a grammar change or not? The answer surely is that this depends on the form of the theory of grammar that we have hypothesized. Since there are still many questions to be answered here (for instance, should we opt for fewer rules, constraints and principles, and thus conceive of grammar as highly abstract, or should we allow more semi-automatic processing and a more prominent role to lexical idiosyncracies7), it is not easy to determine what linguistic change constitutes grammar change and what does not. Generative linguists, for this reason, look at cases where a number of surface changes seem to cluster together, which may point to a deeper grammar change (cf. van der Wurff 2000, and the work of Lightfoot referred to in note 5). This is a good working strategy, and such cases, if they can be found, are a clear indication that a grammatical rule at a deeper level exists. Examples of such clusterings have been proposed in the literature, notably in connection with the English modals. It is clear, however, that there is no consensus about the radical nature of the modals change (cf. note 5). It looks, in fact, as if there is an inverse correlation between depth of data and radicality. The more detailed in data a case study is, the less likely it seems to be that the phenomenon in question is presented as a radical change; compare for instance the differences between the results of Warner's and Lightfoot's investigations into the modals (see note 5). Another problem here is the nature of the radical change. If the change does not produce further and simultaneous surface evidence (further changes in

450

Olga C. Μ. Fischer

the output), then it is more difficult to uphold that such a change took place. Turning to the modals again, the greatest problem with Lightfoot's proposal was that the simultaneous surface changes that were presented as evidence of the radical category change, were themselves changes that had been going on for some time. They all concerned the gradual obsolescence of constructions, i.e. they did not constitute new constructions. In order to show that re-analysis has taken place, we need to show that new constructions arise that cannot be generated by the same grammar that produced the old ones. To give an example. Clear evidence that a re-analysis has occurred in the so-called for NP to V construction (as in, It is badfor you to smoke\ cf. e.g. Fischer et al. 2000: 214-220), are cases where for + NP can no longer function as the old benefactive dative phrase, because the new construction disobeys the constraints of the benefactive dative. For example, when there are two ybr-phrases in the clause in the new situation, this constitutes a violation of the so-called θ-criterion (as in Chomsky's [1981: 239] famous example, It is pleasant for the rich for poor immigrants to do the hard work), or when for is followed by expletive it (as in It is essential for it to rain soon), it violates the constraint that the NP must be animate, etc. Such direct evidence is not available for the radical shift proposed by Lightfoot, at least not for the modals as a class in any simultaneous fashion. We can also concentrate on language in use and study changes in the output from the point of view of the output itself. This would be the study of language change, and this is the approach taken by traditional and many functional historical linguists. By looking at how language changes (and here we have the advantage that the data on which we base our description are observable data - unlike the [indirect] grammar data), we may come to learn eventually more about the theory of grammar. In other words, we approach the question from the other direction. Following this approach, we are inclined to study innovations and changes from their very beginning. All noticeable changes are taken into account, not just the triggers that may lead to grammar change. The objection often levelled against this approach is that it reifies language: language does not change, rather, speakers change their language.8 In other words we should concentrate on what speakers do and not on what language does. This may be essentially correct (but see the complications noted in note 8), but at the same time it is worthwhile to point out that at this moment in time (having not yet enough insight into the workings of the brain) we can only know what speakers do by looking at language, how it behaves and how it changes. Both the grammar- and the language-change approaches are indirect as it were; it is

Principles of grammaticalization

and linguistic reality

451

by combining them that we should be able to get a little more insight into our linguistic competence. We have seen how our notion of what constitutes change, how the differences between grammar change and language change, influence our notion of what determines change. Determinants are likely to be smaller in number when change is seen as grammar change, while variation itself will be ignored if it is not considered to be a trigger in the context of the theory of grammar. Furthermore, from this point of view grammaticalization itself cannot be a determinant in change. For those linguists, however, who concentrate on the diachronic process of grammaticalization (i.e. who concentrate on language change), the determinants which drive or steer the process are more local and of a semantic-pragmatic nature, and the process itself is also often seen as a determinant because of its unitary and unidirectional nature. We will now have a look at the two case studies announced in the introduction, to see what factors determine the changes that have taken place there.

3. The grammaticalization of infinitival to It is generally acknowledged in the literature that the allative preposition to (or its equivalent in other Germanic languages) developed into an infinitival marker when it became combined with an infinitive. Jespersen (1927: 10-11) describes it as a process where to became weakened in meaning due to its fixed position before the infinitive, and where the to-infinitive itself encroached on the terrain of the bare infinitive, replacing it in many of its functions so that to became a mere marker of the infinitival form without any meaning of its own. He adds that this process can be seen in English but also in other Germanic languages, such as German, Dutch and Scandinavian. Haspelmath (1989) essentially repeats this story, illustrating the way grammaticalization works with data from (mainly) German. He argues that the to (or zu) development should be seen as "a universal path of grammaticization". In both accounts, the grammaticalization itself is seen as a determinant, it is seen as an inexorable process, that, once started, could not be stopped, and Haspelmath shows that it neatly follows all the parameters involved in grammaticalization (with one exception where the condition is vacuous, see [le] below) as distinguished by Lehmann (1985). Thus, he notes for German (but all this can be silently extended to English in his view):

452

Olga C. Μ. Fischer

(1)

a) an increase in paradigmaticity in that the number of prepositions that can be used with the infinitive decreases to just one, i.e. zu b) an increase in bondedness: i.e. zu and the infinitive start behaving as a unity c) loss of paradigmatic variability ·, the choice of complements after a particular verb becomes reduced to just one d) loss of integrity: both semantically and phonetically zu becomes reduced. The phonetic evidence is not so clear for German zu, but does apply to Early Middle English to (there is spelling evidence for this in the form of te and t' variants) and Dutch te. Semantic reduction is clear, among other things, from the fact that another preposition expressing 'goal' is found to reinforce the earlier zu. e) reduction in syntagmatic variability: not applicable because the zw-infinitive starts out as a PP that already had a fixed order of constituents f) reduction in scope: the scope of zu becomes confined to its immediate constituent, i.e. the infinitive9

In Fischer (1997a), I discuss these parameters one by one in relation to the development of to in English. It is interesting to observe that at first English to more or less follows the above parameters, although for English, not only (e) is vacuous, but also (a), and in a way (b) and (f) as well, (a) is vacuous because the only preposition that occurred with an infinitive from the very beginning was to (I am ignoring the Scandinavian loans cet and till, since they are later and dialectal, cf. Visser 1963-73: §897). As far as (b) is concerned, there was cohesion between to and the infinitive from the very beginning, as there was between any preposition and the NP that it governed in Old English, i.e. no other elements could occur between to and the infinitive (cf. Los 2000: 252-253). Because of this cohesion, it is not surprising that the scope of to was also from the beginning restricted to the following infinitive (cf. [f]). When we look at Old English to, we note that to is normally repeated before a second conjunct. In Fischer (1996), I discussed all the cases of coordinated infinitives in the Old English part of the Helsinki corpus,10 and additional examples found in Callaway (1913: 150-51, 173-74), Visser (1963-73: §967) and Mitchell (1985: §§929, 935, 956), and came to the conclusion that the repetition of to before the second conjunct is the rule (it was repeated in 96.8 per cent of all cases attested in the Helsinki corpus). When to is not repeated, the second infinitive must be seen either as part of the first, i.e. the two infinitives form

Principles of grammaticalization

and linguistic reality

453

a group (examples found here are generally a translation of a single infinitive in Latin), or the second infinitive has a different status. The latter doesn't express an activity different from the first infinitive, or subsequent to it, rather, it expresses the content of the first activity, or the way through which the first activity may be achieved. Thus, this case involves not two coordinate infinitives but a hierarchically ordered set with the second dependent on the first. The following examples show the differences. (2) is the normal case, where the two infinitives, both preceded by to, are coordinated and express two separate activities. Of the instances with a second bare infinitive, (3) is an example of the grottp-infinitive, and (4) shows the second infinitive as dependent on the first: (2)

Wid eagena sare, haran lifer gesoden ys god on wine against of-eyes sore, of-hare liver boiled is good in wine to drittcenne, & mid pam brope da eagan to bepianne to drink and with the broth the eyes to bathe 'Against eye-sore, a boiled hare's liver is good to drink in wine, and to bathe the eyes with the broth (Helsinki Corpus, Quadrupedibus, de Vriend 1972: 27)

(3)

ic cwom for don to delanne vel sceadenne I came therefore to part or separate monnu wid feeder his & dohter wid moder hire man from father his and daughter from mother her Ί came therefore to part or separate a man from his father and a daughter from her mother' (Helsinki Corpus, Rushworth Gospels Skeat 1871-87: 89) (Cf. Latin ueni enim separare hominem aduersus patrem suum et filiam aduersus matrem suam)

(4)

ic eom gearo to gecyrenne to munuclicere drohtnunge, I am ready to turn to monastic way-of-life, and woruldlice deawas ealle forlcetan and worldly practices all leave-off Ί am ready to turn to a monastic way of life and forgo (by forgoing) all worldly practices' (Callaway 1913: 150-51, Mf Horn. I 534)

This leaves us only with parameters (c) and (d). (c) is a somewhat difficult case. It looks indeed as if there has been a reduction in the choice of complements after each particular verb. Verbs that in Old English could take both that-clauses and fo-infinitives (such as the equivalents of 'command',

454

Olga C. Μ. Fischer

'intend', 'plan' etc.) now only use the to-infinitival complement. On the other hand, verbs that could take both a bare and a to-infinitive in Old English, usually can still take two different complements in present-day English, except that the bare infinitive has been replaced by the -ing form (cf. Fischer 1997a: 268-269). It is not the case, in other words that the toinfinitive ousted the bare infinitive as suggested by Jespersen (cf. above). Parameter (d), however, is very clearly present in the development from Old to Middle English. There is evidence for the phonetic reduction of to to te in the spellings used in the Middle English part of the Helsinki corpus. There is also clear evidence that to must have become weakened in meaning: after the Old English period, we frequently see to accompanied by an additional preposition, for, showing that its original meaning of 'purpose' had to be reinforced by a new 'purpose'-marker. The use of a second preposition indicates at the same time that to itself was no longer a preposition. It could of course be argued that for + to, formed a new, combined preposition, but there are further Early Middle English examples which show to in combination with other prepositions, such as with and of indicating that to had become a mere infinitival marker (for examples, see Fischer 2000: 157). That the purpose-sense of to had weakened semantically is also clear from occasional examples such as, (5)

And in my barm ther lith to wepe / Thi child and and in my bosom there lies to weep, your child and (Macauley 1900-01, Gower Conf.Am. Ill 302)

myn mine

To wepe in (5), which expresses an activity simultaneous with that of the matrix verb, clearly does not express purpose or a subsequent activity. Thus, as far as the development in the Middle English period is concerned, we have a clear case of the grammaticalization of to to an infinitival marker. Some of the grammaticalization of to may already have taken place at an earlier, pre-Old English stage, witness the vacuity of four of the six parameters. On the other hand, it is also possible that to and the infinitive were pretty much a fixture from the very beginning. To sum up, the following changes have been shown to mark the development: (6)

the grammaticalization of to in its early stages: a) strengthening of to by for b) phonetic reduction of to c) occurrence of to-infinitives after prepositions other than to d) loss of semantic integrity (loss of directional sense)

Principles of grammaticalization

and linguistic reality

455

With the changes in (6), we have a situation that is typical in any grammaticalization process and that is the layering of the old and the new forms. Next to infinitival marker to, there still exists the preposition to, which has undergone none of the above changes. If we accept the principle of unidirectionality as valid, we must also accept that with the onward process of grammaticalization, this layering will develop into further divergence or the replacement of the original preposition by another form. This has indeed happened in Dutch where the infinitival marker te has no connection anymore with the particle toe,11 from which it first developed, while the preposition itself has changed into tot. In German, the form for infinitival marker and preposition is still the same, but their uses have further diverged, as Haspelmath (1989) has shown. In English, however, the process has not continued in the same direction as German and Dutch. We see a number of new developments in Late Middle English and Early Modern English that all go against the grammaticalization parameters described in (1) above. (7)

degrammaticalization developments involving to: a) appearance of split infinitives b) increase in scope c) strengthening of semantic integrity d) loss of all double prepositions, including/or

The use of split infinitives (7a) in present-day English, such as to not go, to silently abhor, is a well-known fact. They first appear in the Late Middle English period; they are not found in Old English (cf. Visser 1963-73: §§977-982). Scope increase (7b) is clear from the fact that in present-day English there is no need to repeat to before a second infinitive, whereas this was not possible in Old English (if to was not repeated, it involved a meaning difference, see the discussion above), nor is it allowed in either Modern German or Dutch, cf. (8), (8)

He went into the church to pray and (to) light a candle for his sick wife 'Er ging in die Kirche um zu beten und eine Kerze anzustecken für seine kranke Frau' 'Hij ging de kerk in om te bidden en een kaars aan te steken voor zijn zieke vrouw'

Thus, in Modern English to can have scope both over the first and the second infinitive, while in Dutch and German te/zu must be repeated. Con-

456

Olga C. Μ. Fischer

cerning (7c), constructions such as the Middle English one given in (5) (ther lith to wepe) are no longer possible in present-day English. This in contrast to Dutch, where the te-infinitive is common after verbs like to lie when no purpose is intended, as in Hij lag te slapen, 'He lay to sleep [i.e. sleeping]'. As to (7d), the for to infinitive has disappeared from standard English. Similarly, it became impossible to use the /o-infinitive after other prepositions roughly after the Middle English period (cf. Visser 1963-73: 19 §976), whereas this is still possible in Dutch, where the ie-infmitive can appear with such prepositions as zonder 'without', met 'with', door 'through', na 'after' etc., and to a lesser extent also in German. What does all this tell us about determinants or principles of grammaticalization? It shows that the process need not necessarily be unidirectional. The grammaticalization process may proceed regularly up to a certain point (while it does not have to go through all grammaticalization parameters as we have seen), and may then turn backwards reversing (some of) the grammaticalization parameters. Thus, (7a) shows the reversal of parameter (lb); (7b) reverses (If), while (7c) and (d) reverse parameter (Id). It also shows us that the parameters of grammaticalization need not decide the process, they are merely indicators that a process may be under way. In the case of English some of the parameters may indeed have been in place from the very first appearance of the to-infinitive. This case further indicates that there must have been other forces at work that disturbed the grammaticalization process. What is the status of these disturbing factors? Might they not with equal reason be called determinants? I believe that there were two important factors which steered the development of the /o-infinitive in English. The first and most important is the form and shape of the grammar at the time the grammaticalization reversed. Another, auxiliary factor may have been iconic pressure, which is a latent, universal feature in the grammar.13 There is no room to go fully into the developments that infinitival complements underwent in Middle English, so a brief discussion of the various changes will have to suffice. It is important to note, however, that the list of changes discussed all involve changes which are typical of English and not shared by Dutch and German. Concerning the form and shape of the contemporary grammar, there is, first of all, the important question of the category status of the infinitive. Los (2000: 233-283) shows that the traditional view, namely that the Old English infinitives were NPs and that these became more verbal in Middle English, is far too simple, and from a purely Old English point of view, largely incorrect. Their nominal status in Old English had been deduced

Principles of grammaticalization

and linguistic reality

457

from the fact that originally they derived from nouns (as the dative ending in -enne indeed suggests), but Los makes very clear that as early as Old English the /o-infinitive had verbal properties and was more or less equivalent to a subjunctive that-clause. Although the development of verbal properties must have started before the Old English period, there is still some evidence for its origin as a PP, and the absence of a clausal negator, a perfective infinitive and a lexical subject shows that the /o-infinitive did not have the full trappings of a clause in Old English yet. It can also be shown that the /o-infinitive increased some of its nominal properties in Early Middle English, in that it came to be used in subject position in that period for the first time. Soon however, this new use was given up in favour of the gerund, which was on the rise in Middle English, which made it possible for the to-infinitive to develop further in a verbal direction. It is interesting to observe that the Dutch infinitive has retained more of this nominal status to the present day. Unlike in English, where there are severe temporal restrictions on the use of non-extraposed subjects associated with the concept of factivity (cf. Bolinger 1968), the te-infinitive can still be freely used as a subject, and the bare infinitive in Dutch can even be preceded by a determiner and/or adjective (cf. Los 2000: 246-249). No doubt the absence of the verbal gerund in Dutch was conducive to this situation. Thus, what is clear is that the infinitives in the two languages developed differently: in English they remained more verbal (even acquiring further clausal features, see below), while in Dutch they gained additional nominal properties. Furthermore, new structures appear in Middle English involving the toinfinitive. We see the rise of split infinitives, already mentioned above, of independently negated infinitives (as in They warned her not to get involved), perfective infinitives, passive infinitives after verbs other than modals (as in Let them be thrown into the bay), and so-called ECM (Exceptional Casemarking) (or a.c.i.) constructions (as in I believe him to be innocent). It is clear that these new features all involve a fuller clausal range for the infinitive (i.e. the infinitive can now have its own tense features, its own negator, and its own lexical subject). The rise of perfective and negated infinitives is also attested in Modern German and Dutch and may not be truly new features but accidental gaps in Old English (cf. Los 2000: 26), but the other features are all special to English. In Fischer (1991, 1992, 1994a) I have attributed the rise of the passive infinitive and the ECM constructions to the change in word order in English, which became a strict SVO language in the Middle English period; Los (2000: 285-344) too relates these changes to the loss of OV orders and of V2. Since Dutch and German retained OV and V2, this may be the main reason why the three

458

Olga C. Μ. Fischer

languages diverged in the further developments of the fo/te/zw-infinitives. Finally, it must be stated that the spread of the fo-infinitive at the cost of the subjunctive that-clause, whose replacement was very rapid in Middle English (cf. Manabe 1979), must have been partly due to the early loss of the subjunctive inflection in English, again showing a difference with Dutch and German. The latter retained the subjunctive much longer (German still does) and both also make use of ί/ζαί-clauses, where English can often only use a ίο-infinitive. The conclusion that can be drawn from these changes is that the infinitival marker to, instead of remaining an appendage to the infinitive (which was the result of the early stage of grammaticalization), began to be interpreted as an independent element, with a meaning of its own. Los (2000: 352) describes the change in formal terms as follows: "To has the same features to check as the subjunctive, and like the subjunctive it checks them in T. In Old English, it is a clitic on the infinitive, and checks its features covertly; in Middle English, it starts to move to Τ overtly and is no longer a clitic but a free word." To indeed functions as a tense-modality marker, it indicates a Τ domain different from the domain of the matrix verb. This can be seen most clearly in the new ECM constructions that begin to appear after verbs of perception in the Middle English period. Perception verb complements with a bare infinitive had been common in Old English. In these constructions, matrix verb and complement share the same tense domain. In the new Middle English constructions, however (which, incidentally, do not occur in German and Dutch), the use of to indicates a shift in tense domain. In the examples of (9), the activities conveyed by the matrix verb se 'to see' and the infinitival verbs to be forlore and to forgon respectively, are not simultaneous, to points to a future event: (9)

a. it thoghte hem gret pitel To se so worthi on as sehe, it thought them great pity to see so worthy one as she With such a child as ther was bore,I with such a child as there was born So sodeinly to be forlore so suddenly to be totally-lost 'it seemed to them a great pity to see so worthy a woman as she was to be destroyed together with the child that was born to her' (Macauley 1900-1901, Gower, Conf.Am. II, 1239-42) b. 'for certeynly, this wot I wel', he seyde,! for certainly, this know I well, he said

Principles of grammaticalization

and linguistic reality

459

'That farsight of divine purveyaunce/ that foresight of divine providence Hath seyn alwey me to forgon Criseyde,' Has seen always me to forgo Criseyde 'for certainly, this I know well, he said, that the foresight of divine providence has always seen that I would lose Criseyde' (Benson 1988, Chaucer T&C IV,960-62)14 Looking at the change from a semantic point of view, it could be said that this new use of to is close again to the meaning of the preposition to from which the infinitival marker derived and with which it was presumably still 'layered'.15 Layering, as opposed to divergence, must mean in terms of language learning that for the speaker the two items are still associated, that they belong to the same prototype. Presumably the preposition to, being more meaningful, is also more prototypical. It could be said therefore that the degrammaticalization is supported iconically in that the infinitival marker to moves closer again to prototypical to re-acquiring a sense of direction, of goal. Metaphorically, place is often used to indicate time, and here, similarly, goal or direction is used to indicate future (tense) or possibility (modality). The type of iconicity involved here is isomorphism which, in this case, also involves persistence (cf. note 16). The development of English to can be seen as follows: (10) stages of grammaticalization of to (a)

α χ

(b)

α xy

(c)

α

β

χ

y

(α = the signans to; β = the reduced signans of to\ χ = signatum 'goal'; y = signatum 'infinitival marker')

At stage (10a) there is isomorphism because to has only one signatum, i.e. the allative meaning of to. At stage (10b) to has two different signata, functioning as a preposition as well as an infinitival marker; i.e. the isomorphism has been disturbed. In a typical grammaticalization process the new stage would be as in (10c) restoring isomorphism (this represents the Dutch development). In English, however, there was a partial reversal to stage (10a) in order to restore isomorphism. Isomorphism is the most simple type

460

Olga C. Μ. Fischer

of diagrammatic or relational iconicity, i.e. the notion that one form should correspond to one meaning (cf. Haiman 1980).16 4. The grammaticalization of have to The role played by iconicity in the degrammaticalization of the /o-infinitive is not a crucial one, it must be seen rather as an aggravating factor that cosupports the other synchronic syntactic factors that were found to be influential in the process. Iconicity comes more clearly to the fore in the grammaticalization of have and the to-infinitive. In line with similar developments involving a possessive verb like have, where have in combination with an infinitive grammaticalized from a full verb into an auxiliary, it has usually been taken for granted that English have to represents a regular case of grammaticalization. Thus, van der Gaaf (1931), Visser (1969: §1396-1410), and Brinton (1991) all more or less accept the following three developmental stages for the construction, I have my work to doll have to do my work. At the earliest stage the construction has the following features: have at first is used as a full verb, meaning 'to possess', the NP work functions as the direct object of have, the to-infinitive is not obligatory, the infinitive functions as an adjunct dependent on the NP, and word order is not relevant, it does not influence meaning. In a subsequent stage of the development, the meaning of have slowly generalizes and acquires obligative colouring in combination with the toinfinitive, the fo-infinitive becomes obligatory, the infinitive no longer functions as an adjunct to the NP but as an object complement of the matrix verb have, and the original object of have (work) becomes an argument of the infinitive. In the final stage we see the appearance of inanimate subjects (possessive have + infinitive always had animate subjects), and the appearance of intransitive infinitives, i.e. the original object can now be dropped altogether. Re-analysis or rebracketing from (1 la) to (1 lb) now follows, (11) a. / [have [my work to do\~\ b. I [[have to do\ my work] resulting in a fixed have + fo-infinitive + NP word order. Note the sudden shift in word order between (11a) and (b), which is difficult to account for satisfactorily within this account.

Principles of grammaticalization

and linguistic reality

461

It is quite clear in this sketch of the putative development of have to that the grammaticalization proceeds along a path of semantic change - bleaching of possession first, the development of obligative colouring later - and that the syntactic changes - the word order change and the rebracketing are subordinate to it, following hard on the heels of the semantic change. Because the development is seen as gradual, the various stages are extremely difficult to disentangle. This is noticeable also from the fact that van der Gaaf (1931), Visser (1969) and Brinton (1991) do not agree as to when the different stages occur. The change, therefore, is seen as a typical chain, driven semantically. In my own investigation of this case (Fischer 1994b), I considered all the instances in which have is followed by a ίο-infinitive in the Helsinki corpus (which covers the Old, Middle and Early Modern periods). Looking at a total of 643 examples, I came to the conclusion that there is no evidence for a gradual semantic change in have from 'possess' via a more general meaning to an obligative sense as envisaged by the studies reported on above. The generalized meaning of have already existed in the earliest recorded (Old English) period, (12) And her beoö swype genihtsume weolocas... Hit hafad eac And here are very abundant whelks ... It has also pis land sealtseapas, and hit hafap hat wceter this land salt-springs, and it has hot water 'And there are (or: 'it has') plenty of whelks ... the country also has (or: 'there are also') salt springs and hot water' (Bede, Miller 189091:26,11.9-12) and a modal obligative colouring of have was possible in Old English too, but only in constructions where have and the infinitive thematically shared an object, i.e. in constructions like: (13) hcefst du ceceras to erigenne have you acres to plow 'do you have acres you could/should plow?' ( ^ l f . Gram. 135.2) where ceceras is syntactically and thematically the object of hcefst and can also be interpreted as the thematic object of erigenne. Obligative colouring did not occur in the following construction types:

462

Olga C. Μ. Fischer

(14) pcet he stowe hcefde in deem streame that he room had in the stream (Bede, Miller 1890-91: 436,11. 7-8)

to standenne to stand

(15) Ic hcebbe anweald mine sawle to alcetanne I have power my soul to leave Ί have power to lay down my life' (WSGospels, Skeat 1871-1887: 10.18) where the infinitive has no object of its own (14), or where both have and the infinitive have their own objects (15). It is important to note, furthermore, that the modal colouring in construction types like (13) was not necessarily one of obligation. In (16) (of the same type as [13] containing a shared object), obligative meaning is not possible in a context where Christ refers to spiritual food, (16) Ic hcebbe mete to I have food to 'There is food I may {JEHom 5 225, Pope

etene pone pe ge nyton eat that that you not-know eat that you know nothing of 1967: 298)

All the more firm syntactic evidence for the change (the appearance of inanimate subjects, absence of an object of have, double use of have, etc.) is very late, occurring only from the Early Modern period onwards. In fact, it can be shown that these syntactic changes follow upon a (general) word order change. The basic SOV word order of Old English, which persisted quite long in infinitival constructions in Middle English, ensured that the order of the three basic elements involved in type (13) constructions, was in normal circumstances (so when no movement rules were involved) almost always have + NP + fo-infinitive. The medially positioned NP could function equally well as an object of the main verb have (due to the V2 rule in main clauses, which would move have to a position before the NP object), and as an object of the infinitive. (17) below shows the Old English situation (I have used a simplified tree structure). When the word order in Late Middle, Early Modern English became generalized to SVO everywhere, the structurally ambiguous surface position of the object NP was no longer available: that is, the object NP had to shift to a postverbal position with respect to the verb which gave it its semantic role. Since have usually had a generalized meaning in this construction, the usual position for the shared object became the one after the infinitive, with which it had a stronger thematic bond. This change is shown in 18.

Principles ofgrammaticalization and linguistic reality

463

(17) surface order in Old English (an SOV language) in both main and subordinate clauses of type (13)

I

acres to plow •

have

I

1

(heavy NP shift/VP raising in subordinate clauses) (V2 rule in main clauses)

surface order > I have acres to plow

(18)

Old English (SOV basic word order) NPS have NPo/i [0; to infinitive] (main and subclauses) NPS N P o / j have [0, to infinitive] (subclauses)

1 Late Middle English (SVO basic word order)

NPS have NPo/i [0; to infinitive] I have acres to plow

NPS have [NP0 to infinitive] ^ NPS have [to infinitive NP 0 ] I have to plow acres

So it was the SOV > SVO word order change that fixed the order of the have to construction in type (13) to have + to-infinitive + NP, and which ultimately led to the re-analysis described in (11). Or, to put it differently, the word order change caused the adjacency of have and the to-infinitive, which in turn led to a semantic change, in which have and the to-infinitive

464

Olga C. Μ. Fischer

were considered one semantic unit. The other syntactic changes involving have to (intransitive infinitives, inanimate subjects, the use of'double' have to have, the development of have to into an epistemic modal via an earlier deontic stage) all date from after the word order change. What may be learned from this case? What are the determinants in the change? We see, a) no gradual generalization in meaning, no intertwining with gradual syntactic adaptations; the semantic bleaching of have has been around for at least six hundred years b) a syntactic change causing the adjacency of have and the to-infinitive. This change is not linked to an earlier semantic development, but the result of a general word order change c) syntactic adjacency > new structural unit (rebracketing) > new semantic unit d) a semantic change in have to to a modal auxiliary as a result of the new adjacency and of metonymic forces (conversational implicatures), caused by the inherent 'goal' meaning of to (for which see section 2) and the occasional modal colouring that could be present all along in these have constructions depending on the other lexical items in the clause e) analogy? The frequency of other V-V constructions analyzed as Auxiliary-V may have aided the restructuring Further support for this scenario may be found in the fact that in the closely related languages Dutch and German the auxiliarization of the cognates of have to did not take place, at least not to the same extent.17 The reason for this difference may be the fact that German and Dutch did not undergo the SOV > SVO change, which was the cause of the ensuing adjacency. The reinterpretation of the new structural unit as a semantic unit is an iconic phenomenon. It is the reverse of Givon's (1985: 202) proximity principle, derived from what he calls an "iconic meta-principle": "The closer together two concepts are semantically or functionally, the more likely they are to be put adjacent to each other lexically, morpho-tactically or syntactically". One would expect the proximity principle also to be valid the other way around, i.e. the moment two elements are placed together syntactically or formally, it is likely that they will begin to function together semantically or functionally. The determinants in this case therefore are the contemporary shape of the grammar, and iconic pressure. Semantic bleaching is not a determinant,

Principles of grammaticalization and linguistic reality

465

even though it is a prerequisite. There is no conceptual chain, since the change starts off syntactically. Re-analysis is central, but it takes place at the beginning rather than the end of the process. Haspelmath's (1999) idea that the maxim of extravagance is the motor behind grammaticalization, is unlikely here because the adjacency of have and the to-infinitive is clearly an accidental result, and not an innovation by a speaker who wishes "to attract attention" by using "unusually explicit formulations" (Haspelmath 1999: 1043).

5. Concluding remarks I have tried to show by means of these two case studies that it is very difficult to find (a) common denominators) in grammaticalization. The unidirectional development from more concrete to more abstract, from more lexical to more grammatical, from open (noun, verb) to more closed categories (preposition, pronoun etc.) is indeed a strong tendency, but cannot be a principle or determinant unless we try to explain away all counterexamples. It also cannot be denied that pragmatic inferencing and semantic change play an important part, that semantic bleaching may be involved, but again they need not play a primary role; that is, not in the sense of a chain resulting from conceptual manipulation whereby the syntactic developments necessarily follow semantic developments. It was shown in the case of have to that semantic bleaching (i.e. the development towards a more abstract, grammatical meaning) was a prerequisite for grammaticalization but not a result of it - on the contrary, it could be said that the quasimodal have to is an enrichment of the earlier bleached have - , whereas in the early stages of the to-infinitive bleaching must be seen as part of the process itself because it is simultaneous with the other factors involved in the process. The same story holds for Lehmann's parameters or properties of grammaticalization. They are very useful diagnostics for grammaticalization but again it is not necessary that they are all present in the process. As we saw in the early grammaticalization of infinitival to, some of the parameters must have been vacuous from the beginning. Structural and/or categorial re-analysis seems to be a firm component of grammaticalization but it is not necessarily situated near the end of the process (as the analysis of the cases discussed in Roberts and Roussou [1999] suggests), as again the have to case has made clear. Here it was the word order change and the ensuing adjacency of have and the to-infinitive which led to the structural re-

466

Olga C. Μ. Fischer

analysis. All the other aspects of this grammaticalization case, marking have to as an auxiliary rather than a verb (involving categorial re-analysis), occur after the structural re-analysis. It is also important to note that the weight of re-analysis as a component in the process differs according to the theoretical framework used. In generative theory re-analysis is seen as the most important if not the only mechanism in grammaticalization because only structural factors are considered; in other words, it is the result of a number of surface changes (phonetic and semantic reduction, semantic/pragmatic changes) which themselves are not interpreted as changes in the grammar. Most functionally oriented linguists likewise believe that reanalysis is an important mechanism but in their view there is not one, deep structural re-analysis but a chain of smaller ones that may spread lexically by analogical extension. They would see pragmatic inferencing as a type of re-analysis, too, on the semantic level, which produces small-scale structural re-analyses in its wake, so to speak.18 The point then is, do we see these smaller semantic and structural changes as involving (local) grammar changes or do we only see grammar change occurring after the analogical extension of the re-analyses has been played out in full? Linguists who see grammaticalization as an epiphenomenon, as an accidental conglomerate of various mechanisms of change, cannot really explain why the process is so common and to a certain extent predictable (in the sense that it can be predicted of ongoing changes that they are likely to follow a certain path). Such a view of grammaticalization, however, does help to explain why not every process once started runs its full course, and why processes involving cognate elements do not always run the same course in related languages as the cases of infinitival te/zu/to and have to have shown.19 It takes account of the fact that language acquisition is discontinuous, and, most importantly, it takes account of the circumstances of the present speakers' grammar. We have seen that the synchronic contours of the grammar are crucial in the process of both to and have to. However, this way of looking at grammaticalization does not explain why we so often have a grammaticalization chain. The formalist solution could also lead to extreme reductivism, i.e. one could easily argue that each linguistic change is in fact a collection of changes. A sound change, for instance, looks neat and tidy after the event, but it is also a change that is not necessarily purely phonetic: it may diffuse lexically, it may start and not succeed (historical evidence for this would be hard to find because not recorded in the spelling, but there is enough evidence of this from sound change-in-progress investigations), it may even be left in the balance for a long time when the old and the new form have both overt and covert pres-

Principles of grammaticalization

and linguistic reality

467

tige in the speech community. Here, too, other mechanisms may play a role, external ones such as prestige, internal ones such as iconic factors (e.g. phonaesthemes and the avoidance of homonymy, cf. Samuels 1972: 45-48 and 67-75 respectively). How can we decide, in other words, whether any particular sound change is a unitary phenomenon, or a conglomerate of smaller changes? Sound change may look more simple both structurally and semantically than grammaticalization, but is it? Of course the moment that meaning becomes involved (which is less frequently the case in sound change) the situation becomes more complicated, idiosyncratic developments are more likely to creep in, but is that a reason to doubt the possible unitary nature of grammaticalization? There are two areas that need to be investigated in order to get a more complete view of the nature of grammaticalization, i.e. whether it should be seen as an epiphenomenon or as something more solid. First of all, as is emphasized and discussed by Janda (2001), we need to pay attention to the sociolinguistic background and to sociolinguistic mechanisms: "the persistent trend of grammaticalization across generations can perhaps best be accounted for by invoking a model which focuses precisely on the relations between generations in a speech community." He refers to work, among others, by Ohala (1989, 1993) on the nature of sound change, noting that only the beginning of a sound change is phonetically motivated and that hypercorrection plays a far more important role than has usually been assumed. Janda writes, It could thus be said that sound-change tends to be regular, not due to persistent influence from some kind of articulatory or auditory/acoustic phonetic naturalness, but instead because exaggerations and misconceptions of phonetic tendencies tend to involve stepwise generalizations based on the natural classes of phonology .... And the reasons for these (over) generalizations can be sought in the social-groupmarking function so often brought to light in sociolinguistic research. (Janda 2001: 305)

He therefore proposes a new language acquisition model which is much more complex than the well-known Klima-Andersen generative model; a model which incorporates synchronic language variation as well as grammatical innovations within an individual. In this model, the social function of age-group marking plays an important part, i.e. it helps to explain the socially motivated extension of what begins as a simple phonetic change. Janda (2001: 304) writes that the progress of a sound change to other phonological environments may "sometimes obey implicational principles ... but [that they] are often independent." It is quite possible, however, that

468

Olga C. Μ. Fischer

implicational principles are more important in grammaticalization (because semantic factors are much stronger here), and that that may explain why the process so often proceeds in the same direction: the adoption or loss of one property implying the next one as it were. Plank (1995) studies the progress of both grammaticalization and degrammaticalization, and notes that the latter differs from the former not only in its lesser frequency and in direction but also in "die Art des schrittweisen Ablaufs" ('the way in which it devolves stepwise', Plank 1995: 200). He shows schematically how a grammatical structure y may possess an χ number of properties, how these properties are linked internally, and how they are linked externally to difΛΛ

ferent formal categories. He illustrates how this works in practice for the properties that are important in the marking of adnominal relations, showing how the properties are related to each other thematically, and how they are linked and in what order to the forms expressing this relation. That the properties are linked is shown by the neat, stepwise decline (or increase depending which form one takes as basic) of properties between forms in fully synthetic languages and forms in fully analytic languages, with Latin (which uses a genitive case inflexion exclusively) and English (which uses mainly prepositional phrases) serving as endpoints. Grammaticalization then involves, as Plank shows, the gradual orderly loss of the thematically linked properties along this cline. He next studies the degrammaticalization of the genitive in English from a full case ending to a clitic and notes that the same properties are involved but with a different chronology: a stepwise reversal does not take place. In the case of the English genitive two particular properties in the middle of the scheme collapsed through other (external) circumstances, and this led to a disturbance of the original implicational order of properties. What occurs next is a re-establishment of order (which Plank calls "resocialization") as follows: it is not the properties themselves that get re-ordered (they cannot be because they are implicational and two links have been lost), but it is the formal expression that gets re-ordered (changed) in such a way that it fits the left-over, disturbed properties. In the case of the English genitive, the form changes from an affix into a clitic because this was the best way to salvage the properties that it still possessed. Two interesting facts emerge from this study. First of all, it shows that degrammaticalization cannot be a mirror image of grammaticalization because it is caused by the disturbance of properties and not by the orderly loss of one property linked to another.21 Secondly, it shows the importance of implicational properties in the case of adnominal marking. It suggests that the progress of a grammaticalization change from a PP to an inflex-

Principles of grammaticalization

and linguistic reality

469

ional affix follows a specific implicational order. It suggests further that this order ultimately depends on the type of change involved in any grammaticalization. Quite naturally, when an adnominal relation grammaticalizes, the process will be different (because a nominal expression has different implicational properties) compared to when a verb grammaticalizes into an auxiliary or a subject pronoun into an inflexional verbal ending.22 In other words, Plank's study shows the orderly progress in any case of grammaticalization as well as differences between types. We will need to find out, therefore, by means of further investigations to what extent the continuity or chain-like quality found in grammaticalization cases, is a result of sociolinguistic factors (Janda's idea) or structural, implicational ones (or indeed both), and secondly, to what extent the different properties of structures and their formal expressions lead to different types of grammaticalization paths and the use of different types of mechanisms and parameters. If implicational properties are found to play an important role in grammaticalization processes, it would be difficult to uphold that grammaticalization is a mere epiphenomenon. I think it can be concluded that the determinants or principles of grammaticalization are by no means firm or fixed but depend very much on the circumstances of the language or its grammar under investigation, and on general iconic principles that interact with these circumstances. In the real linguistic world many rules are no more than tendencies. Notes *

I would like to thank Adrienne Bruyn, Anette Rosenbach and the editors, Britta Mondorf and Günter Rohdenburg, for their careful reading of this paper and for the very helpful suggestions they have each made to improve it. 1. The full quotation is as follows: "Reanalyse ließe sich leicht als jener Teil dessen beschreiben, was innerhalb des Grammatikalisierungsprozesses ausschließlich mit Mitteln eindeutiger kategorialer Unterscheidungen und mit Restrukturierung von Konstituenz zu erfassen ist. Reanalyse wäre innerhalb eines solchen begrifflichen Rahmens einfach eine Subkomponente der Grammatikalisierung insofern, als die unter Grammatikalisierung klassifizierten Typen des Wandels erst bzw. nur syntaktisch, d.h. durch Restrukturierung oder durch Kategorien-/Wortartwechsel sichtbar werden müssen" ('Re-analysis could easily be described as that part of what within the process of grammaticalization can be grasped exclusively by means of clear categorial differentiation and restructuring of constituency. Within such a terminological frame, re-analysis would simply be a subcomponent of grammaticalization in so far as the types of change classified under grammaticalization must become visible first, or in other words, only syntactically, that is, by means of restructuring or a shift of category/part of speech. (Abraham 1993: 1314).

470

Olga C. Μ. Fischer

2. It is interesting to observe in this respect that Haspelmath (1998) has suggested that reanalysis is not part of grammaticalization, indeed that "[g]rammaticalization and reanalysis are disjoint classes of phenomena" (p. 315). This idea, however, can only be upheld if one defines re-analysis in a way not recognized by other scholars. Campbell (2001: 145-149) discusses the criteria or diagnostics which Haspelmath uses to distinguish the two phenomena in great detail. He shows that they are rather idiosyncratic. They, in fact, seem to be adapted to Haspelmath's 'new' definition of re-analysis (Geurts [2000a and b], in his commentaries on Haspelmath [1999], also notes his tendency to present "an unduly narrow conception" of familiar terms). For instance, when one takes 'ambiguity' as a diagnostic for re-analysis but not for grammaticalization, then one must class the change in the for NP to V construction (see below) as re-analysis, whereas it is quite clear that it involves grammaticalization too, in that a benefactive preposition becomes a (more general) complementizer. Similarly, when a diagnostic for grammaticalization is its unidirectionality (re-analysis being classified as 'bidirectional'), then the loss of the impersonal in English could be seen as grammaticalization as suggested by Haspelmath (1998: 338-340), even though the case doesn't seem to show any of Lehmann's (1985) parameters (no phonetic or semantic reduction, no bonding, no increase in paradigmaticity, etc.). In fact, I would even question its unidirectionality, because in Middle English some original nominative experiencer subjects became dative (cf. Denison 1993: 71-72), and with some impersonals, the dative experiencer remained, while the thematic role of source/cause became subject. 3. There may be another problem related to this, as noted by Geurts (2000a). Grammaticalization is typically a phenomenon of language (as a communicative system), and not of the individual speaker. Geurts writes (p. 784) that there is no continuum between content and function words in the individual speaker (this would agree with the approach taken by Roberts and Roussou), but also that grammaticalization (on the language level) is not a result of an individual's conscious decision (p. 786). In other words, it is a little odd to see a re-analysis that takes place in an individual's competence as the equivalent of grammaticalization. 4. This can clearly be seen in the treatment of grammaticalization by Roberts and Roussou (1999) mentioned above. 5. Generative linguists are mainly interested in how linguistic features and constructions hang together and how this hanging together is shown up by change. Thus their interest is in radical changes rather than superficial ones. How much evidence has been gathered so far for the existence of radical changes? The category change in the English modals, which Lightfoot (1979: 81-120; 1992: 142-54; 1999: 180-85) considers to be of a radical nature in the grammar of 16th-century English (it was the paradigm case in Lightfoot 1979), was quite gradual on the output level both before and after the putative radical change. Warner (1993) indeed shows that the modal verbs were exceptional within the category of verbs already in Old English, and that they still are in present-day English, but to a higher degree, and that the progress of exceptional behaviour was by incremental lexical changes within each modal verb. Reviewers and critics of Lightfoot (1979) have also voiced severe doubts about the radical category change involving the Old English quantifiers (see Bennett 1979; Fischer and van der Leek 1981: 311-317), the ίο-infinitives (Warner 1983: 200-202; Fischer 1996; Fischer et al. 2000: chapter 7, and see below, section 2) and the introduction of a rule of NP Preposing (Fischer and van der Leek 1981: 325-339; Warner 1983: 202-206). In all these cases, the data proved stubborn and the various new constructions were not as neatly simultaneous and related as they should have been to prove the existence of one deeper change.

Principles of grammaticalization and linguistic reality

471

6. It is important to realize that the status of these triggers is different from the status of what Milroy (1992: 169-172) calls innovations or what Harris and Campbell (1995: 54 and passim) call exploratory expressions. The latter two are examples of variations that may lead to linguistic change, but need not. I.e. they do not (yet) constitute change. Triggers in generative theory, on the other hand, are full-blooded changes because they lead to grammar change. However, not all changes (in the sense of Milroy) need be triggers. Thus, we have a scale of changes running from least to most influential: innovations (changes at the individual level) > changes (changes on the language level that are midway in the S-curve) > triggers (changes that lead to further changes elsewhere). The first two may be determined by empirical observation of language data, the last one depends on the theoretical framework one works in. 7. A fairly abstract theory of grammar will work with clear categories and restrictive rules and constraints (restrictive both in number and form) as to how the various categories are ordered syntactically. Rules will be as general as possible and will try to capture the majority of the generated constructions. A less abstract, more surface-related grammar will have rules that apply on the whole less generally; more of the grammar will be part of the lexicon, whose rules are restricted to lexemes or small groups of lexemes and are thus more idiosyncratic. In the latter type of grammar there are more rules to learn, in other words, but the rules themselves are easier because less abstract, and constructions may be processed semi-automatically rather than generated by the grammar each time. The categories too may be less clear-cut, may be fuzzy even, and slight shifts are more easily understandable (and more easily describable in terms of the grammar). These shifts are often driven by context and by the way we perceive the external world (pragmatic inferencing, for instance, will play an important role in the shifts). The most extreme proponent of the latter view is Paul Hopper with his Emergent Grammar, where strict rules and categories have all but disappeared, and grammar is for ever on the move, so to speak, constantly in the making. As always, there is something to be said for both points of view: the model of grammar (if we ever manage to describe it) will no doubt have elements of both. 8. This issue is discussed in great detail in Keller (1994) and Yngve (1996: esp. chapter 3). They both discuss the problems that exist with respect to the scientific approach to language. Language is not a living organism that can be described and analyzed in terms of natural laws, speakers are the real organisms involved. So we should study human brains in their environment in order to find out more about language. In this respect, not only the concentration on parole would be wrong - which is indeed the generative viewpoint - but also the concentration on langue, because in Saussure's opinion langue dealt with the object of language "independent of the individual" (Saussure [1916] 1986: 37; 1983: 19), "[t]he activity of the speaker must be studied in a variety of disciplines, which are of concern to linguistics only through their connexions with linguistic structure." In other words the speaker is not central to Saussure's investigations into language either (see also Yngve 1996: 31). Yngve calls the problem domain confusion and shows that it has deeply affected generative theories. He writes, "In retrospect, and with a sensitivity to domain confusions, it can now be seen that the various criteria and tests [in Chomskian linguistics] that were introduced were all operating in the logical domain. They effectively defined a priori certain characteristics to be attributed to the structure of language. Thus they were disguised assumptions creating and introducing the objects of study in detail. Each criterion characterized an a priori view of some characteristic of language or the objects of language, and since there was no reality, no object with existence prior to tests, they did not describe any existing object and there was no guarantee

472

Olga C. Μ Fischer

that they would even be consistent with one another. With no physical reality, no scientific tests are possible" (Yngve 1996: 45-46). At the same time, there is the paradoxical situation that language change is not directed by the will of man. Keller shows how linguistic changes (in the Milroyian sense, see note 6) may occur whose result is as it were the opposite of what the innovations brought about by speakers set out to do (it is the "non-intended consequence" of an accumulation of individual, intended actions, Keller 1994: 64). Keller refers in this respect to the invisible hand - phenomenon. He shows that, although innovations begin (sub)consciously and are in a sense weakly teleological or final, the linguistic change that may arise out of a number of individual innovations, must itself be seen as a causal phenomenon, i.e. caused by natural laws. Keller writes (1994: 80): "The results of final, or as I prefer to say, intentional actions accumulate under certain conditions and bring about structures which do not lie within the sphere of final individual actions. The accumulation is a causal problem. Thus, both the 'finalists' and the 'causalists' have a share of the truth. Their error lies in the exclusivity of their claims, as both fail to notice the interaction of final and causal processes." This view of Keller's seems to reduce the importance of the individual speaker in change (as opposed to innovation). It seems highly likely that for explanations of long-term processes of change (such as grammaticalization) we must look at sociolinguistic processes next to grammar change. This is in fact the stance taken by Richard Janda in a recent article in Language Sciences: granted that "grammaticalization phenomena ... represent an epiphenomenon which can therefore have no global properties and no long-term 'path(way)s' of its own, ... we are then obliged to explain why grammaticalization as a conjunction of separate processes does tend to proceed in the same direction over time" (2001: 304). His suggestion is that it is "sociolinguistically motivated generalization by successive generations that allows grammaticalization phenomena to show apparent graduality and a predominant directionality despite discontinuous transmission over time" (p. 307). For more discussion of this, see also section 5. 9. William Croft brought to my attention in the discussion of this paper that this parameter has been called into question by Tabor and Traugott (1997). Tabor and Traugott have suggested a new way of looking at grammaticalization, which ignores the notion of unidirectionality for the time being, and instead focuses on a new hypothesis namely that processes that involve the hallmarks of grammaticalization (these hallmarks being: evidence of [1] morphosyntactic change, [2] pragmatic/semantic change, [3] gradualness) involve scope /«crease rather than decrease. I am not yet convinced that this is a good direction to take because it obscures the real differences between grammaticalization and degrammaticalization. The case of to discussed here clearly shows scope decrease in both German and Dutch, where the original preposition has fully grammaticalized, while English to shows scope increase at the moment that it i/egrammaticalizes. The grammatical changes that Tabor and Traugott discuss, all showing scope increase, are not the most typical grammaticalization (in the original sense of the term, i.e. including unidirectionality) cases. One involves the English genitive, which has been seen as a prime example of degrammaticalization, cf. Janda (1980) and Plank (1995); one involves the verbal noun becoming a gerund, which I would see as a case of reanalysis (brought about by, among other things, phonetic confusion between the original verbal participle ending in -ende/inde and the nominal ending -ung > -yng) and not grammaticalization since it shows no sign of semantic or phonetic reduction or any of the other parameters distinguished by Lehmann; while cases which involve subjectification (such as the modals, and the development of sentence adverbials) - which indeed

Principles of grammaticalization and linguistic reality

473

generally show scope increase - seem to me to be basically different from regular grammaticalization instances (i.e. no real loss of integrity, no increase in paradigmaticity, increase in bondedness?). In other words, Lehmann's parameters are still useful diagnostics for the process of grammaticalization, as far as I can see. 10. For more information on this corpus, see Kytö (1991). 11. Except in lexicalizations or fixed expressions such as ten oosten van 'to the east o f , te lijfgaan 'to body go', i.e. 'attack', te Amsterdam 'at Amsterdam'. 12. With the exception of the new prepositional phrase instead of, which takes a /o-infinitive (next to a gerund) even up to the present day. All other prepositions disallow a /o-infmitive from about 1500 onwards, with a few later examples, mainly from Spencer's The Fairy Queen, where he used archaic language on purpose. 13. Cf. Plank (1979: 131): "Der frühkindliche Spracherwerb zeichnet sich durch eine ausgeprägte Präferenz zu ikonischer Zeichenbildung aus, die als 'natürliches' Substrat jeder Zeichenbildung zumindest latent wirksam bleiben dürfte, wenn auch nach Maßgabe von Symbolisierungsnotwendigkeiten" ('language acquisition in young children is characterized by a very strong preference for the formation of iconic signs, which, as a natural substratum of all sign formation, probably remains at least latently active even though tempered by the need to symbolize'), and also Givön (1995: 61): "One must consider the pervasive iconicity of human language merely the latest manifestation of a pervasive preference for isomorphic coding in bio-organisms". 14. For more details on how this rather difficult example (of which there are three in Chaucer) should be interpreted, see Fischer (1995: 10-11). 15. I am using the term layering here in a more narrow way than in Hopper (1991) and Hopper and Traugott (1993: 224), in that the layering is restricted to forms within one cline (i.e. the layering concerns one and the same formal element), and not to layers within a functional domain (including renewal). A layered item becomes divergent when the speaker no longer sees the formal connection between the items in the cline, i.e. in lexical terms the polysemous items have become homonyms. 16. For more information on the role played by iconicity in grammaticalization processes, see Plank (1979), Fischer (1999). Fischer (1999) argues that metaphorical shift, renewal, phonetic reduction and persistence, all processes important in grammaticalization, are iconically based. 17. There is one small group of expressions in Dutch, where hebben 'have' seems to come very close to a modal auxiliary expressing pure obligation; these expressions are commonly accompanied by the word maar 'but/only', as in Je hebt het maar te doen 'you have it only to do'. It seems to me that these expressions are a subtype of the regular use of hebben + te infinitive in Dutch. They always mean 'there is nothing else for you to do/that you can do but ...'. Since the expression always refers to the only thing that one can still do, in practice it comes to mean that one must do it. So far there is no evidence that a pure modal is developing in Dutch from this subtype. The same seems to be true for German, haben zu + infinitive can only be used when there is such a restriction, i.e. when the thing that has to be done is the only possibility. So one can say, Du hast das Buch zu lesen, sonst .... 'you have to read this book, otherwise ....', but one would not normally say Ich habe mit dem Zug zu fahren, weil ich kein Auto habe Ί have to travel by train because I don't have a car'. 18. Cf. Croft (2000: 160), who describes it thus, "Pragmatic inference ... is a type of metanalysis: a contextual ('pragmatic') property of the meaning is reanalyzed as an inherent ('semantic') property of the meaning, and a related inherent property is reanalyzed as a contextual one." Like structural re-analysis, the re-analysis on the seman-

474

19.

20.

21.

22.

Olga C. Μ. Fischer tic/pragmatic level takes place syntagmatically. In that sense it is different from (semantic) metaphor and (structural) analogy, which take place on the paradigmatic axis. I do not agree, therefore, with Haspelmath (1989) who refers to "universal paths of grammaticalization" cutting across related languages; and I would use with great caution "the application of ... cross-linguistic generalizations about grammaticization [as] a standard technique to guide an investigation of grammaticization in a particular language" (Hopper 1991: 20) because this may lead to a preconceived analysis of the facts, as happened for instance in Brinton's (1991) analyis of the grammaticalization of have to (cf. Fischer 1994b). His proposal is similar to the way sounds are analyzed on the phonetic/phonological level. Each phonetic sound (or form) consists of a number of distinctive features (or properties), and these sounds (allophones) are linked to an abstract structure, i.e. the phoneme. In sound change, too, the links between the features play an important role in the way allophones alter, and in the way allophones are linked to phonemes. There are also of course more direct and formal reasons why degrammaticalization cannot be a mirror image of grammaticalization. One can to some extent predict how a full form would reduce phonetically, but one cannot predict the reverse, the choices would be almost unlimited (cf. Geurts 2000a). It may explain, for instance why the scope-parameter works differently in different cases, as noted by Tabor and Traugott (1997). See also note 9.

References Abraham, Werner 1993 Grammatikalisierung und Reanalyse: Einander ausschließende oder ergänzende Begriffe? Folia Linguistica Historica 13: 7-26. Bennett, Paul A. 1979 Observations on the transparency principle. Linguistics 17: 843-861. Benson, Larry D. (gen.ed.) 1988 The Riverside Chaucer. Oxford: Oxford University Press. Bolinger, Dwight 1968 Entailment and the meaning of structures. Glossa 2: 119-127. Brinton, Laurel 1991 The origin and development of quasimodal have to in English. Unpublished manuscript, University of British Columbia. Callaway, Morgan 1913 The Infinitive in Anglo-Saxon. Washington: Carnegie Institution. Campbell, Lyle 2001 What is wrong with grammaticalization? Language Sciences 23: 113-161. Chomsky, Noam 1981 Lectures on Government and Binding. Dordrecht: Foris. Croft, William 2000 Explaining Language Change. An Evolutionary Approach. London: Longman. Denison, David 1993 English Historical Syntax. London: Longman

Principles of grammaticalization and linguistic reality

475

Fischer, OlgaC.M. 1991 The rise of the passive infinitive in English. In: Dieter Kastovsky (ed.), Historical English Syntax, 141-188. Berlin: Mouton de Gruyter. 1992 Syntactic change and borrowing: The case of the accusative and infinitive construction in English. In: Marinel Gerritsen and Dieter Stein (eds.), Internal and External Factors in Syntactic Change, 17-88. Berlin: Mouton de Gruyter. 1994a The fortunes of the Latin-type accusative and infinitive construction in Dutch and English compared. In: Toril Swan, Endre Merck, Olaf Jansen Westvik (eds.), Language Change and Language Structure. Older Germanic Languages in a Comparative Perspective, 91-133. Berlin: Mouton de Gruyter. 1994b The development of quasi-auxiliaries in English and changes in word order. Neophilologus 78: 137-164. 1995 The distinction between to and bare infinitival complements in late Middle English. Diachronica 12: 1-30. 1996 The status of to in Old English to-infinitives: A reply to Kageyama. Lingua 99: 107-133. 1997a The grammaticalisation of infinitival to in English compared with German and Dutch. In: Raymond Hickey and Stanislaw Puppel (eds.), Language History and Linguistic Modelling. A Festschrift for Jacek Fisiak on his 60th Birthday, 265-280. Berlin: Mouton de Gruyter. 1997b On the status of grammaticalisation and the diachronic dimension in explanation. Transactions of the Philological Society 95: 149-187. 1997c Verbal complementation in Early Middle English: How do the infinitives fit in?. In: Derek Britton (ed.), English Historical Linguistics 1994, 247-270. Amsterdam: Benjamins. 1999 On the role played by iconicity in grammaticalisation processes. In: Max Nänny and Olga Fischer (eds.), Form Miming Meaning. Iconicity in Language and Literature, 345-373. Amsterdam: Benjamins. 2000 Grammaticalisation: Unidirectional, non-reversable? The case of to before the infinitive in English. In: Olga Fischer, Anette Rosenbach and Dieter Stein (eds.), Pathways of Change. Grammaticalization in English, 149-169. Amsterdam: Benjamins. Fischer, Olga C.M. and Frederike C. van der Leek 1981 Optional vs radical re-analysis: mechanisms of syntactic change. Lingua 55: 301-350. Fischer, Olga C.M., Ans van Kemenade, Willem Koopman and Wim van der Wurff 2000 The Syntax of Early English. Cambridge: Cambridge University Press. Gaaf, Willem van der 1931 Beon and habban connected with an inflected infinitive. English Studies 13: 176-188. Geurts, Bart 2000a Explaining grammaticalization (the standard way). Linguistics 38: 781-788. 2000b Function or fashion? Reply to Martin Haspelmath. Linguistics 38: 1176-1177. Givön, Talmy 1985 Iconicity, isomorphism, and non-arbitrary coding in syntax. In: John Haiman (ed.), Iconicity in Syntax, 187-220. Amsterdam: Benjamins. Isomorphism in the grammatical code: Cognitive and biological considera1995 tions. In: Raffaele Simone (ed.), Iconicity in Language, 47-76. Amsterdam: Benjamins.

476

Olga C. Μ. Fischer

Haiman, John 1980 The iconicity of grammar: Isomorphism and motivation. Language 56: 515— 540. Harris, Alice and Lyle Campbell 1995 Historical Syntax in Cross-Linguistic Perspective. Cambridge: Cambridge University Press. Haspelmath, Martin 1989 From purposive to infinitive - a universal path of grammaticization. Folia Linguistica Historica 10: 287-310. 1998 Does grammaticalization need reanalysis? Studies in Language 22: 315-351. 1999 Why is grammaticalization irreversible? Linguistics 37: 1043-1068. Heine, Bernd, Ulrike Claudi and Friederike Hünnemeyer 1991 From cognition to grammar: Evidence from African languages. In: Elizabeth C. Traugott and Bernd Heine (eds.), Approaches to Grammaticalization, Volume 1, 149-188. Amsterdam: Benjamins. Hopper, Paul 1991 On some principles of grammaticization. In: Elizabeth C. Traugott and Bernd Heine (eds.), Approaches to Grammaticalization, Volume 1, 17-35. Amsterdam: Benjamins. Janda, Richard D. 1980 On the decline of declensional systems: The overall loss of OE nominal case inflections and the ME reanalysis of -es as his. In: Elizabeth C. Traugott, Rebecca Labrum and Susan Shepherd (eds.), Papers from the 4th International Conference on Historical Linguistics, 243-253. Amsterdam: Benjamins. 2001 Beyond "pathways" and "unidirectionality": On the discontinuity of language transmission and the counterability of grammaticalization. Language Sciences 23: 265-340. Jespersen, Otto 1927 A Modern English Grammar on Historical Principles. Part III. Copenhagen: Munksgaard. Keller, Rudi 1994 On Language Change. The Invisible Hand in Language. London: Routledge. Kytö, Merja 1991 Manual to the Diachronic Part of 'The Helsinki Corpus of English Texts': Coding Conventions and Lists of Source Texts. University of Helsinki, Helsinki. Lehmann, Christian 1985 Grammaticalization: synchronic variation and diachronic change. Lingua e Stile 20: 303-318. Lightfoot, David W. 1979 Principles of Diachronic Syntax. Cambridge: Cambridge University Press. 1991 How to Set Parameters: Arguments from Language Change. Cambridge, M.A.: MIT Press. 1999 The Development of Language. Acquisition, Change, and Evolution. Oxford: Blackwell. Los, Bettelou 2000 Infinitival Complementation in Old and Middle English. The Hague: Holland Academic Graphics.

Principles of grammaticalization

and linguistic reality

All

Macauley, G. C. 1900-1901 The English Works of John Gower, Volumes 1 and 2, 81-82. (Early English Text Society e.s.) London: Oxford University Press. Manabe, Kazumi 1989 The Syntactic and Stylistic Development of the Infinitive in Middle English. Fukuoka: Kyushu University Press. Miller, Thomas 1890-1891 The Old English Version of 'Bede's Ecclesiastical History of the English People', Parts 1,1; 11,2, 95-96. London: Early English Text Society, o.s. Milroy, James 1992 Linguistic Variation and Change. (Language in Society 19.) Oxford: Blackwell. Mitchell, Bruce 1985 Old English Syntax, Volumes 1-2. Oxford: Clarendon Press. Newmeyer, Frederick J. 1998 Language Form and Language Function. Cambridge, M.A.: MIT Press. Ohala, John 1989 Sound change is drawn from a pool of synchronic variation. In: Leiv Egil Breivik and Ernst Häkon Jahr (eds.), Linguistic Change. Contributions to the Study of Its Causes, 173-198. Berlin: Mouton de Gruyter. 1993 The phonetics of sound change. In: Charles Jones (ed.), Historical Linguistics. Problems and Perspectives, 237-278. London: Longman. Plank, Frans 1979 Ikonisierung und De-Ikonisierung als Prinzipien des Sprachwandels. Sprachwissenschaft 4: 121-158. 1995 Entgrammatisierung - Spiegelbild der Grammatisierung?. In: Norbert Boretzky, Wolfgang Dressler, J. OreSnik, K. Teräan and Wolfgang U. Wurzel (eds.), Natürlichkeitstheorie und Sprachwandel, 199-219. (Essener Beiträge zur Sprachwandelforschung.) Bochum: Brochmeyer. Pope, John C. 1967 Homilies of /Elfric. A Supplementary Collection. (Early English Text Society o.s. 259). Oxford: Oxford University Press. Roberts, Ian 1993 A formal account of grammaticalisation in the history of Romance fiitures. Folia Linguistica Historica 13: 219-258. Roberts, Ian and Anna Roussou 1999 A formal approach to "grammaticalization". Linguistics 37: 1011-1041. Samuels, Michael 1972 Linguistic Evolution. Cambridge: Cambridge University Press. Saussure, Ferdinand de 1983 Cours de linguistique generale. Paris: Payot [Translated and annotated by Roy Harris, Course in General Linguistics. London: Duckworth] [1916] Tabor, Whitney and Elizabeth C. Traugott 1997 Structural scope expansion and grammaticalization. In: Anna Giacalone Ramat and Paul J. Hopper (eds.), The Limits of Grammaticalization, 229-272. Amsterdam: Benjamins. Visser, Fredericus Theodorus 1963-73 An Historical Syntax of the English Language, Volumes l-3b. Leiden: E.J. Brill.

478

Olga C. Μ. Fischer

Warner, Anthony 1983

Review article of D.W. Lighfoot, Principles of Diachronic Syntax. Journal of Linguistics 19: 187-209 1993 English Auxiliaries. Structure and History. Cambridge: Cambridge University Press Wurff, Wim van der 2000 Variation and change: Text types and the modelling of syntactic change. In: Ricardo Bermudez-Otero, David Denison, Richard M. Hogg and C. B. McCully (eds.), Generative Theory and Corpus Study. A Dialogue from 10 ICHEL, 261-282. (Topics in English Linguistics 31.) Berlin: Mouton de Gruyter. Yngve, Victor H. 1996 From Grammar to Science. New Foundations for General Linguistics. Amsterdam: Benjamins.

Varieties of English from a cross-linguistic perspective: Intensifies and reflexives* Peter Siemund

Abstract Modern linguistics and the study of varieties are still largely disconnected fields. In discussing regional and historical varieties of English against the background of crosslinguistic typological work, the following article is an attempt to bridge this gap. The empirical domain chosen to demonstrate that the two fields can stimulate each other are selfforms both in reflexive and intensive use (John saw himself vs. the queen herself). It will be shown that the diachronic data are to a large extent consonant with well established paths of grammaticalization and that regional variation mirrors cross-linguistically observable patterns and limits of variation. Moreover, the study of varieties can help to solve extensively discussed theoretical problems such as locally free se^forms of English.

1. Introduction An attempt is made here to demonstrate that grammatical variation found in English and its varieties is to a large degree consonant with rules and regularities that have emerged from cross-linguistic typological work. The empirical domain chosen for the present study is the one of reflexivity and intensification, whereby intensification is here to be understood in a very narrow sense involving expressions like German selbst, Engl, x-self, etc. Central to the subsequent discussion will be the observation that some languages (e.g. German) draw a formal distinction between intensifiers and reflexives (selbst vs. sich), whereas other languages, notably English, do not differentiate between these expressions (cf. [1]). (1)

a. Der Dekan selbst trifft die Entscheidung. / Johan kritisiert sich nicht gern. b. The Dean himself takes the decisions. / John is opposed to criticizing himself.

As is well known, linguistic typology is a methodological approach to the analysis of language that involves the comparison of many languages for practical reasons usually between 60 and 200 - whose central aim is to

480

Peter Siemund

reveal what is common to all languages, i.e. to identify so-called language universals, and along which parameters languages can differ. Once linguistic generalizations, or universals for that matter, have been established on the basis of a representative sample of the world's languages, there are good reasons to believe that they also carry over to other languages and different varieties thereof, even more so if robust functional explanations can be given for them. Being able to show which of the grammatical phenomena found in English varieties are common, widespread and hence more expectable than others is an important asset of the typological approach advocated here because it enables us to distinguish idiosyncratic, language-specific facts from pervasive regularities and allows us to focus our analysis on the important issues. As a matter of fact, many interesting questions can not be raised, let alone be answered, without going beyond the descriptive material made available by one language alone. In contradistinction to traditional and established approaches to the study of varieties (dialectology, socio-linguistics), which have mostly been concerned with phonological variation, we will here exclusively concentrate on morpho-syntactic features. In addition we will restrict the types of varieties to be investigated and only consider the regional as well as the historical dimension. In view of the traditional focus on phonological variation, it is noteworthy that we are currently witnessing a major upsurge in the importance assigned to dialect syntax with methodological inspiration being taken both from current syntactic theorizing (cf. Beninca 1989; Abraham and Bayer 1993; Black and Motapanyane 1996) as well as language typology (cf. Kortmann 1997, 2001). Enormous efforts are currently being made to systematically enlarge the empirical basis available for dialectal data and to make it available in a machine-readable form (cf. Kortmann 2001). As a result of this new interest in morpho-syntactic variation the recent past has seen the emergence of several in-depth studies dedicated to dialectal variation, for instance the study by Anderwald (2001) on negation and multiple occurrence of negative elements, by Klemola (1996) on tense and aspect, specifically do as simple tense carrier and on pronominal gender (Siemund, forthcoming). To summarize the major aims and claims, what I would like to achieve in this article is first of all to provide a thorough account of regional and historical variation in the domain of intensifiers and reflexives. Including data from several different historical stages as well as regions of the English speaking world it will be shown that grammatical variation in this domain is highly systematic and not at all idiosyncratic. Finally, by embed-

Varieties of English from a cross-linguistic perspective

481

ding the data from English varieties in cross-linguistic typological work we will be able to show that the grammatical phenomena found are frequently in harmony with cross-linguistic generalizations, in some respects even to a greater extent than the standard variety. 2. Intensifiers and reflexives: some basic facts In this section some background information is provided concerning the empirical domain studied as well as the theoretical tenets that are assumed. As far as reflexives, or more specifically reflexive anaphors, are concerned, it is sufficient to point out that these are expressions that in their most prototypical use indicate that the object of a transitive predicate is bound by the subject, as shown in (2).1 (2)

John hurt himself.

Apart from this central, argumental use of reflexives, the relevant expressions can also occur in various derived constructions (cf. [3]), in which the main function of the reflexive is to intransitivize the predicate (derived intransitivity). The subsequent discussion will show that there is a connection between these derived uses of reflexives and identity of reflexives and intensifiers. (3)

a. Paulfürchtet sich vor Mäusen. 'Paul is afraid of mice.' b. Dieses Buch liest sich leicht. 'This book reads easily.'

An extensional definition of intensifiers across different languages is given in (4). Since a thorough description of meaning and use of intensifiers is a relatively complex issue, only a few salient properties will be discussed here (cf. Siemund 2000 for the details). (4)

Engl, x-self, own, Germ, selbst, Mand. ziji, Indon. (sen)diri, Russ. sam, Arab, nafs, Bengali nije, Dutch zelf Fin. itse, Lith. pats, Malayal. tanne, etc.

Intensifiers always interact with an NP, which is made clear by overt agreement in some languages (English). We can distinguish between

482

Peter Siemund

adnominal (juxtaposed, part of NP) and adverbial (non-juxtaposed, part of VP) uses. Most languages appear to draw this distinction (cf. [5]). In their adverbial use intensifies can often be paraphrased by 'alone' or 'without help'. (5)

a. The General himself has calledfor peace talks. b. He changed the nappy himself. (« alone)

It is important not to confuse such se//1 intensifiers with scalar focus particles like Germ, sogar or Engl, even, although these may have the same form (cf. [6]). It can be shown that intensifiers do not induce the scalar nor the quantifying effects indicative of these expressions. (6)

Sogar / selbst Riesen haben einmal klein angefangen. 'Even giants started from small beginnings'

Intensifiers always evoke alternatives to the denotation of the NP they are construed with so that they can plausibly be analyzed as focus particles. In contrast to focus particles like also, even and only the scope of adnominal intensifiers is restricted to the head NP. Adnominal intensifiers characterize the value of their focus as the contextual point of reference (centre) and relate the set of alternatives (periphery) to this reference point (König 1991; Baker 1995; Siemund 2000), thus ordering them as peripheral values around the central value. This general contribution of structuring a contextually given set of values into a central one and associated peripheral values is instantiated by essentially three different contexts (cf. [7a-c]). An entity may serve as the contextual point of reference (i) because it takes up an extreme point on a hierarchy of the real world, (ii) because it serves as the point through which other entities are identified and (iii) because the relevant entity is the centre of perspective.2 (7)

a. The director himself will talk to us. (centre because of high rank) b. Mary's husband looks after the children. Mary herself works in a hospital, (centre for the identification of others) c. He was not particularly tall, a little taller than Jemima herself perhaps, but his shoulders in the tweed suit were broad, giving an air of authority, and he himself, if not exactly heavy, was cer tainly a substantial man. [Antonia Fräser, A Splash of Red, p. 88] (centre of perspective)

Varieties of English from a cross-linguistic perspective

483

Adverbial intensifiers as in (8) have a lot in common with manner adverbs in terms of distribution and partly also in meaning. However, there are also important differences, such as low positional mobility, agreement with the subject and the evoking of alternatives to the subject. Moreover, adverbial intensifiers cannot be conjoined with other manner adverbs. Again, a precise characterization of the syntax and semantics of adverbial intensifiers is beyond the scope of this article and also not necessary.3 Suffice it to say that they characterize the referent of the subject as the point of reference relative to the situation described by the sentence. Notice that in the negative version of (8) the man referred to by John is characterized as being interested in or benefiting from the situation described by the sentence. (8)

John has cleaned the bathroom himself. / John has not cleaned the bathroom himself.

3. Lack of reflexive anaphors These brief remarks should suffice as background information and in the following we will turn to a discussion of some interesting pieces of variation in this grammatical domain, starting with the encoding of reflexive relations. Moving through the historical and regional dimensions, we can observe considerable differences in the expression of reflexivity. The basic empirical observation is that different varieties of English lack special reflexive anaphors completely or in various structural positions (non-direct or indirect object positions, adjunct positions, first and second person, certain types of predicates) - mainly substituting them and thus expressing reflexive relations by simple personal pronouns - and that this kind of variation is highly systematic as well as compatible with well known typological hierarchies and paths of grammaticalization. In the following we will discuss these cases one by one.

3.1. Complete lack of reflexives Like in Modern Frisian today, there were no separate reflexive pronouns in Old English and personal pronouns were used to express co-reference (binding) in a local domain. In those cases in which the intended interpretation could not be inferred from the context, this double duty of personal

484

Peter Siemund

pronouns led to an ambiguity between coreferential and non-coreferential or disjoint reading of the personal pronouns (cf. [9]).4 (9)

Old English hine he beweraö mid wcepnum ('se defendit armis') [7E1G 96.11] 'he defended himself with weapons'

To enforce local binding it was possible to add the intensifier seolflselflsylf thus co-indexing the personal pronoun in object position with the subject NP. Case marking makes clear that it was the adnominal intensifier that was added (cf. [10]).5 (10) Old English Hannibal... hine selfne mid atre acwealde [Or 4 11.110.2] 'Hannibal killed himself with poison.' According to the semantic contribution of adnominal intensifiers outlined in the previous section, we can assume that in examples like (10) the intensifier co-indexes the referentially unspecified object pronoun with the contextual point of reference, which, everything else being equal, will be the subject.6 As a result of coalescence and univerbation (pronoun + intensifier) complex reflexive anaphors (and intensifiers) develop.7 The original situation found in Old English as well as intermediate steps in the development are still visible in (conservative) regional varieties (cf. [11] taken from Wright 1898-1905). (11)

Yorkshire a. He has cut him [himself]. b. He went to bathe him [himself].

English based Pidgins and Creoles frequently go back to the default state and start a new cycle in the development of reflexives. In Sranan we essentially find the situation of Old English with reflexive relations being expressed by personal pronouns (cf. [12]) and the intensifier being optionally available to indicate binding in a local domain unambiguously (cf. [13]).8 (12)

Sranan a. a kil hem 'he has killed himself b. a bron hem 'he has burnt himself

Varieties of English from a cross-linguistic perspective

(13)

485

Sranan a za kiele hem zlifi 'he'll kill himself

The Saramaccan example in (14), by contrast, already exemplifies distributional complementarity of personal pronoun and complex self-iorm, as we are familiar with from Standard English. For this variety it makes sense to claim the existence of a separate reflexive anaphor. (14)

Saramaccan a. Jarii si en*^. 'John saw him.' b. Jarii si en-sinkiiy*y 'John saw himself.'

3.2. Lack of reflexives in non-direct object positions The lack or existence of special reflexive anaphors across varieties of English is made even more interesting by the fact that one can usually observe differences with respect to the distinction between direct and non-direct objects.9 A stable generalization across regional as well as historical varieties is that if there is a separate reflexive in non-direct object position, then there will be also one used in direct object position. In terms of the historical development of English in this domain this means that special reflexives develop first in direct object positions and only later in non-direct object positions, if they develop at all. (15) illustrates this with some examples from Early Modern English.10 (15) a. Not so common as commendable it is, to see young gentlemen choose them such friends with whom ... [E 105] b. I made me a large tent. [RC 45] Some regional varieties, specifically those spoken in the southwest of the United States (cf. Christian 1991), have preserved this intermediate step (cf. [16]). (16) Southern American English a. He got him [himself] some candy. b. He wanted some straw to build him [himself] a house out of. c. I traded it, sold it for twenty-five dollars and bought me [myself] a pony.

486

Peter Siemund

d. He organized him [himself] a band of traders. e. We'd head out up in them trees and roll that stuff up and make us [ourselves] cigarettes, you know, and smoke that. To explain this asymmetry in the encoding of reflexive relations, it would seem plausible to refer to the semantic roles that are encoded by direct and non-direct (indirect) objects in the prototypical case. On the assumption that subjects typically bear the role of an agent whereas the referents of direct objects are typically patients, it is rather obvious that these two roles are difficult to combine in one and the same referent and that special expressions or marking is necessary to encode such a relation. By comparison, the semantic roles that typically go together with nondirect objects are those of recipient or beneficiary and in this case no incompatibility with the agent role arises. Hence, the expression of agentbeneficiary coincidence requires no special marking. This explanation is roughly compatible with Farmer and Harnish's (1987) Disjoint Reference Presumption, which says that the arguments of a transitive predicate are maximally different.

3.3. Lack of reflexives in adjunct positions The observation of asymmetries in the encoding of reflexive relations can be extended further to adjunct positions, where we find personal pronouns used in the prototypical case even in Standard Modern English. The examples in (17) illustrate the much discussed swa^e-sentences, which have become notoriously famous for not requiring a self-form for the expression of coreference between the subject and the oblique NP, although the use of the self-form is possible under certain conditions.11 In a language like German, by comparison, we find the reflexive pronoun sich obligatorily used 19 in these positions. (17) a. He noticed a snake near him. b. She grabbed behind her. c. John has a lot of passion in him. 3.4. Lack of reflexives in first and second person Reflexive marking in English and German manifests an interesting and cross-linguistically well known parameter of variation. In English, as in

Varieties of English from a cross-linguistic perspective

487

Russian, Yiddish, Mandarin and many other languages, complementarity between reflexive anaphors and personal pronouns exists in the first, second and third person, i.e. for all persons a special form is used to indicate coreference in a local domain.13 By contrast, German (as well as Swedish, Spanish, etc.) only requires a special reflexive in the third person, but uses simple personal pronouns for the expression of reflexive relations in first and second person. The distribution of reflexives across these languages is fully compatible with a well known linguistic universal according to which first and second person reflexives require the existence of third person reflexives (cf. [18]).14 (18) 3 > 2 > 1 (English, Huichol, German) When we now turn to reflexivity and intensification in English varieties we find, first of all, that the development of complex reflexives in Old English starts in the third person and only then does it spread to second and first person. These developments are fully compatible with the universal in (18). 5 Moreover, there is a clear functional motivation because the only place where an ambiguity can arise is the third person. After all, there is only one speaker and one addressee in any speech situation.16 Regional varieties also comply with this universal in that there is no such variety with special reflexive anaphors for the first and second person, but not for the third person. As shown in examples (12) and (13), personal pronouns are used to express reflexive relations in the English based Creole Sranan. However, whenever the intensifier is added to indicate reflexivity unambiguously, we only find third person pronouns involved (cf. [19]). (19)

Sranan a. Mi cotte mi. Ί cut myself.' b. Wassijoe! 'Wash yourself!' c. Mi wassi mi. Ί wash myself.'

3.5. Lack of reflexives with non-other-directed predicates A final parameter determining the presence or absence of reflexive anaphors to be considered here concerns the meaning of the predicate

488

Peter

Siemund

involved. A semantic parameter that can be shown to be important for the choice of a particular strategy of reflexivization in many languages is whether the activity denoted by the matrix verb is in the prototypical case directed away from the subject (so-called typically other-directed verbs) or directed towards the subject (so-called typically non-other-directed verbs). According to this classification all verbs of bodily harm (hurt, kill, etc.) belong to the group of typically other-directed verbs whereas verbs of grooming {wash, comb, etc.) are typically non-other-directed. The relevance of this parameter can be seen in Old English, where, as Keenan (2001), following Farr (1905), notes, there are ten verbs with which a reflexively used object pronoun is nearly always followed by the intensifier. Among these ten verbs we exclusively find verbs of bodily harm, inter alia acwellan 'kill', ahon 'hang', fordon 'destroy', forseon 'scorn, renounce', (ge)haelan 'cure, castrate', ofslean 'slay', (ge)swencan 'afflict, oppress', and yreag(ga)n 'threaten, torture'. And even in Modern English typically non-other-directed verbs like verbs of grooming normally do not require the use of a reflexive (cf. [20]). (20) John shaved / washed / combed / dressed. Since the relevance of this semantic distinction has only relatively recently caught the attention of researchers working in the field, there are unfortunately no systematic data available from regional varieties of English, nor are there systematic classifications of verbs based on this distinc17 tion. Essential empirical work is needed here and in order to establish further analogies we have to content ourselves with giving an example from a related Germanic language. As the Swedish example in (21) shows, the intensifier själv has to be added obligatorily to the reflexive anaphor sig with a verb like 'attack', but not with a verb like 'defend'. (21)

Swedish a. Han angrep sig själv/*sig. 'He attacked himself.' b. Han forsvarade sig (själv). 'He defended himself.'

The general tendency, corroborated by language after language, is that complex strategies of reflexivization occur with typically other-directed predicates whereas typically non-other-directed predicates select simplex strategies.

Varieties of English from a cross-linguistic perspective

489

3.6. Summary To sum up the preceding discussion, in the history of English we can identify the following general paths of grammaticalization, which can also be interpreted as synchronic tendencies constraining variation in the domain of reflexive marking. No regional nor any historical variety of English has come to my attention that would not be in harmony with these general tendencies concerning reflexive-marking. to

-

m

argument position of predicates —*· adjunct positions direct objects —> non-direct (indirect) objects 3rd person —> 1st, 2nd person typically other-directed predicates —*• typically non-other-directed predicates

An important result of the historical processes in English is that reflexives and intensifiers are coded identically. Why this kind of identity developed is one of the many unsolved mysteries in the history of English. It is certainly not a necessary consequence tied to the rise of complex reflexives (cf. Afrikaans). Nevertheless, identity of this kind may be rare in Europe, but is nothing unusual from a cross-linguistic perspective.

4. Identity of intensifier and reflexive According to research done by König and Siemund for the World Atlas of Language Structures (Diyer, Haspelmath, Gil and Comrie 2003), intensifiers and reflexive anaphors are completely or partially identical in form in about 55 percent of all languages and are only differentiated distributionally in these cases.19 An example of intensifier-reflexive identity from Mandarin Chinese is given in (22) and (23). Figure 1 visualizes the distribution of languages according to this parameter. There are clear areal clusterings. Major areas showing identity are the Middle East, Asia and Papua New Guinea. English, like Celtic and Finnish, is an exception in Europe. The languages

490

Peter Siemund

of Europe and the Americas mostly differentiate between intensifies and reflexives. Africa shows no clear areal patterning. (22)

Mandarin Zhangsan kanjian-le ziji Zhangsan see-ASP REFL 'Zhangsan saw himself.'

(23)

Mandarin Zhangsan Zhangsan 'Zhangsan

ziji INT himself

kanjian-le see-AS? saw

ta 3SG him/her.'

Apart from the areal clusterings, which would warrant a separate study, there are other interesting facts connected with the identity or formal differentiation of intensifiers and reflexives, one of which will be briefly discussed here. It is well known that three grammatical domains are interrelated, two out of which, namely intensification and reflexivity, we have already shown to be intimately connected. The third domain is what might be called derived intransitivity, by which we will understand the non-referential use of reflexive pronouns with the function of intransitivizing otherwise transitive predicates. Representative examples of this use of German sich were given in (3) above/" With respect to these three domains, the following typological generalization (implicational universal) can be stated: if a language codes reflexives and intensifiers identically, the relevant expression will not be used as a marker of derived intransitivity. In the graphic representation given in Figure 2 this means that only adjacent domains can be encoded by the same expression. This prediction is borne out by data from Modern Standard English, where reflexive pronouns are, apart from a few exceptions (cf. Geniusiene 1987), not used as markers of derived intransitivity (cf. [24]). Λ 1

(24) a. He sat down. / He stood next to me. b. The door opens c. This shirt washes easily. / This book reads easily.

Varieties of English from a cross-linguistic

perspective

491

Interestingly enough, we also find the extension of reflexively used personal pronouns as markers of derived intransitivity (non-referential uses) in Old English. In the older literature (Penning 1875; Farr 1905) these uses of pronouns are referred to as pleonastic pronouns and are treated as an extremely puzzling phenomenon. Some typical examples of such pleonastic uses are shown in (25), but it is currently not clear to which contexts these non-referential uses of pronouns had spread in earlier stages of English. Extremely common are contexts of bodily motion as in (25a). As of today, I have not seen them in anticausatives nor have I seen them in facilitatives (middles), but future examination of data may reveal that they had extended their territory even into these domains. (25)

Old English a. andstodhim underpam treowe [Gen 18.8] 'he stood under the tree' b. he him ondrcet his deapes symble [ ^ I S , Ash Wed, 86] 'He will ever dread his death.'

By contrast, what we do know with certainty is that the use of personal pronouns as markers of derived intransitivity (pleonastic pronouns) suddenly dies out when complex intensifiers and reflexives become grammaticalized (around 1500). Apparently, the complementarity established between the newly developed reflexive anaphors and personal pronouns prevents the latter from being used with an antecedent in a local domain. Hence they had to disappear. As has already been shown for some of the other phenomena surveyed so far, the original situation is preserved in regional varieties. In the following example from Scots personal pronouns are used in typical bodily motion contexts (cf. [26]). (26)

Scots He bouns him to the house, And sit him down upo' the bink.

And again, Creoles replicate these developments (cf. [27]). (27)

Sranan No holle joe na stole. 'Don't lean on your chair!'

492

Peter Siemund

5. Extra-thematic pronouns (benefactive datives) A related but by no means the same phenomenon is the occurrence of dative pronouns in Old English which are not directly licensed by the argument structure of a verb (cf. [28]). (28) Old English wineleas wonscelig mon genimeö him wulfas to geferan [Max I 146] 'The friendless unfortunate man takes wolves as companions.' Such extra-thematic or non-subcategorized pronouns are well known from languages like Modern German, where these uses of pronouns are often referred to as benefactive datives, but can also occur with other functions.22 The peculiar property of Old English again is that simple personal pronouns are used in the third person. The use of the third person pronoun in itself is not surprising, since Old English lacked special reflexives, but it is nevertheless interesting to see that personal pronouns readily assume non-prototypical reflexive functions. This is what benefactive datives have in common with personal pronouns used as markers of derived intransitivity, as discussed in the previous section. The use of personal pronouns as benefactive datives is preserved in some regional varieties of English, specifically Southern American English (cf. Christian 1991). In examples like (29) we would invariably expect a reflexive pronoun to occur in Standard English, which can normally be paraphrased by a^br-construction. (29) Southern American English a. Lotta time, I take me a pound or two of butter and cut me off a chunk of butter. b. We got us some logs, ... put us four big poles around the side of it, and got us logs put over top. Nevertheless, it would be premature to argue that extra-thematic datives in Southern American English are merely a residual feature of some earlier stage of English because we also find these pronouns with predicates with which the addition of benefactive datives is less familiar, specifically verbs denoting psychological states - mainly desires - such as want, love and like (cf. [30]). Apparently, what happens in these cases is that the subject is not characterized as the beneficiary of the situation expressed by the matrix verb (this is simply not possible with psychological states), but as the pos-

Varieties of English from a cross-linguistic perspective

493

sessor-cum-beneficiary of the object desired.23 This explanation, I think, could also account for intuitions of 'heightened subject involvement' which are frequently mentioned in the context of these examples. (30) Southern American English a. We want us a black German police dog cause I had one once. b. He done had him a way figured out to get out. c. I love me some baked beans. d. She wanted her some liver pudding. e. Fran likes her a day off every now and again. Hence, the use of extra-thematic pronouns in contexts like (30) may represent a new development, but this argumentation must remain tenuous as long as the historical sources have not been checked for this phenomenon.24 Another piece of evidence indicating that Southern American English has extended the scope of the original construction is that extrathematic pronouns can also occur in double object constructions (cf. [31]). There are rigid restrictions, though. As the contrast between (31) and (32) makes clear, benefactive datives can only be added if the recipient expressed in the double object construction is encoded as a prepositional object. The addition of extra-thematic pronouns to a double object construction containing two noun phrases as objects leads to ungrammaticality. (31) Southern American English a. I'm gonna write me a letter to my cousin. b. Cindy sent her a letter to Sue yesterday. (32)

*Cindy sent her Sue a letter yesterday.

6. Expressions for body parts Let us now take up a topic again that we have already investigated, although from a different perspective. The historical development of reflexives and intensifiers in English has been shown to be highly systematic, even from a cross-linguistic perspective. Similarly systematic observations can also be made with respect to the lexical sources of reflexives and intensifiers, for which consistently expressions for body parts can be identified, at least in those languages as well as varieties thereof in which the original lexical meaning of these expressions is still transparent or for which we have documents of reasonable time depth. Of special importance

494

Peter Siemund

in this respect are Pidgins and Creoles where the development of reflexives can be monitored in real time, as it were. What one can observe in Creole after Creole is that reflexive anaphors are innovated and renovated via expressions for body parts. For example, the reflexive pronoun of Tok Pisin yet is derived from Engl, head, Saramaccan sinkii comes from Engl, skin and in Nigerian Pidgin the word bodi (< body) is used to indicate conference in a local domain (cf. [33] through [35]). In addition there is frequently a form derived from Engl, self which can be used interchangeably with yet, sinkii, bodi, etc., or be added to these expressions. This suggests a close semantic relationship between intensifiers and reflexives on the one hand and body part nouns on the other. Notice also that the body part nouns in [33] - [35] are appended to personal pronouns, much like in the development from Old to Middle English. (33) Tok Pisin my yet (lSG head) (34)

Saramaccan a. en sinkii (3 skin) b. en seei sinkii (3 self skin)

(35) Nigerian Pidgin Di man bit im bodi. 'The man beat himself.' This general tendency to derive reflexive anaphors from expressions for body parts can be observed in many other Creoles, not just those that are based on English. Thus, the reflexive of Papiamentu is kurpa (< Span. cuerpo), the one of Kabuverdiano is kabesa (< Port, cabega) and Haitian tet evidently is related to Fr. tete. From a cross-linguistic perspective these pervasive tendencies can hardly come as a surprise because expressions for body parts are a widespread source of reflexives and intensifiers. Schladt (2000) investigates the lexical sources of reflexives in 148 languages and found body part nouns to be the clear majority pattern. In my own survey of approximately seventy languages there are 61 languages whose reflexive pronoun is clearly related to a body part expression. Table 1 illustrates a number of source lexemes across various languages. It is noteworthy that only relatively central or important body parts serve as sources of reflexives. Most widespread in terms of distribution are 'body' and 'head'.

Varieties of English from a cross-linguistic perspective

495

Table 1. Lexical sources of reflexives body part

language(s)

body

Burmese, Korean, Swahili

head

Basque, Hausa, Tachelhit

heart

Dongolese Nubian

soul

Arabic, Tetelcingo Nahuatl

bone

Hebrew

skin

Ngiti

The development of reflexive anaphors from body part nouns, however, is only one part of the story because such body part nouns also represent an important and widespread source of intensifiers. Moreover, for many languages showing identity of intensifier and reflexive, and for which the lexical source of the relevant expression is known to be a body part noun, there are good reasons to believe that intensifiers are an intermediate step in the development of reflexives, i.e. that body part nouns first develop into intensifiers and only in a second step into reflexive anaphors. A well known example of a language whose reflexive anaphor is derived from an intensifier is again English (him + self 26 => himself), but there are also languages, notably Basque, of which we know that the reflexive marker is derived directly from the body part noun without the intermediate step via the intensifier. In other words, the second step in Figure 3 given below is optional (but nevertheless widespread).

body part noun

intensifier

reflexive

Figure 3. Development of reflexives

Renovation and reinforcement of reflexives through intensifiers is also widespread in Romance languages (cf. Kemmer 1993: 166ff; Turley 1997:25-26). For instance, in Brazilian Portuguese the reflexive clitic se dropped out of the system and has been replaced by a tonic pronoun in combination with the intensifier mesmo (cf. [36]). Two such cycles of renovation are attested for Rhaeto-Romance (cf. [37]).

496

Peter Siemund

(36) Brazilean Portuguese ele se perguntou / ele perguntou-se —* ele perguntou-se a si mesmo —*• ele perguntou a si mesmo —• ele perguntou α ele mesmo 'he was asking himself (37)

Rhaeto-Romance a. se + ipsu(m) —> sez (renovation through intensifier); se develops into a verbal prefix; b. sei + sez —»· sesez (second cycle of renovation; sei is the disjunctive pronoun) c. el selava 'he washes' (light reflexive marker) d. Jed vesel memez el spieghel. (heavy reflexive marker) Ί see myself in the mirror'

7. Free self-forms: Headless intensifiers or untriggered reflexives? What I have tried to show in the preceding sections is that the phenomena observable in different varieties of English frequently find a natural explanation when we consider them in the light of what we know from other languages and in the context of established cross-linguistic generalizations. In this section I will try to demonstrate that data from English varieties can also have interesting repercussions for current theoretical controversies in the domain of syntax in that they can help to decide between different competing analyses. The problem is the following. Self-forms in Standard English can occur in positions where they are not predicted according to current syntactic theories. Both Chomsky's (1981) Binding Conditions as well as Reinhart and Reuland's (1993) Reflexivity Framework, which represent the two most widely discussed syntactic models, assume that reflexively used selfforms in English (anaphors) require the existence of a syntactic antecedent in a local domain. Moreover, reflexive anaphors are required to be in complementary distribution with pronouns (pronominals). When we now consider the examples in (38) and (39), which are all taken from authentic sources, we find that the ie//~-forms contained therein are not in a local domain with their antecedent, not governed or bound by their antecedent and are not in complementary distribution with pronouns. In (38) the antecedent of the self-form is in a different clause and in the examples shown in (39) there is no antecedent given in the verbal context at all. Therefore, these occurrences of .se/^-forms in English are often referred to as free self-forms or untriggered reflexives.

Varieties of English from a cross-linguistic perspective

497

(38) a. He [Zapp] sat down at the desk and opened the drawers. In the top right-hand one was an envelope addressed to himself. [David Lodge, Changing Places, p. 62] b. At the same time he had the clearest image in his mind of the three of them: Fred, Daisy, and himself, and it was a spectacle of nothing but pleasure. [BNC] (39) a. Of course most of us, including myself will accept the democratic decision. [BNC] b. You may be the one person to bring about improvements which will benefit many others as well as yourself. [BNC] The fascinating piece of evidence that varieties of English contribute to this problem is that they even demonstrate the occurrence of se/^forms in subject positions (cf. [40] - [41]). Of the regional varieties of English Irish and Scottish English are well known to exhibit this phenomenon (cf. Filppula 1999). (40) Irish English a. I'm afraid himself [the master of the house] will be very angry when he hears about the accident to the mare. b. How is herself? 'How is your wife?' / Is herself in? 'Is the lady of the house in?' (41)

Scots The clan must stick together and go out upon the old road when Himself required.

To complicate the problem even further as well as to preclude hasty analyses in terms of language contact with Celtic languages, where a related phenomenon is indeed attested, it should be observed that in earlier stages of English se^forms were equally possible in subject positions (cf. (42)). (42) a. Himself drank water of the wel, As did the Knight Sire Percivel [CT 94] b. For it engenders choler, planteth anger, And better 'twere that both of us did fast Since of ourselves ourselves are choleric Than feed it with such overroasted flesh [Shakespeare, The Taming of the Shrew 4.1].

498

Peter Siemund

The question is how to analyze these free se/^forms. Essentially three different proposals for how to deal with these forms can be found in the literature. According to Zribi-Hertz (1989), who gives an account within Binding Theory, these se/^-forms do not require a syntactic antecedent, but are bound by what she calls the minimal subject of consciousness, i.e. the NP whose referent represents the centre of perspective in the local discourse domain. Although this proposal makes correct predictions for many examples, it clearly fails to cover all cases of free self-forms. In restricting their Reflexivity Framework to argumental occurrences of se^-forms in predicates, Reinhart and Reuland (1993) essentially exclude free self-forms from the discussion of reflexivity, but this is tantamount to ignoring them. Moreover, it can be shown that free self-form?, can even occur in argument positions of predicates (cf. König and Siemund 2000c) although the argumental use of free se/^forms remains a relatively rare phenomenon. The approach to free se/f-forms advocated for here goes back to a proposal by Baker (1995), who in fact picked up ideas by Ross (1970), the central claim being that English free self-forms, should not be analyzed as reflexives, but as intensifiers in a parallel fashion to Germ, selbst or Engl. x-self as in the director himself To substantiate an analysis of free selfforms as intensifiers, specifically in their adnominal use, we have to show two things, namely that (i) free se^-forms and adnominal intensifiers are comparable in terms of distribution and (ii) we have to explain why these adnominal intensifiers can occur without a nominal head. As for the first point, it is easy to show that free se/f-forms predominantly occur in contrastive contexts, i.e. in contexts in which alternatives to the referent picked out by the self-form are provided (Siemund 2000: 69ff., König and Siemund 2000c). As discussed in section 2, this is one of the salient properties of adnominal intensifiers, so that there is a clear parallel. Many such examples can be found in corpora of English (cf. [43]). (43)

a. Her beauty had been something which hadfilled even herself with wonder; ... [BNC] b. The only English people there besides myself were a couple called Keith and Doreen. [LLC] c. Joyce hadn 't expected Barry to follow her, for she knew he was as obstinate as herself [BROWN]

Moreover, it can be shown that free se/f-forms normally pick out the contextual point of reference, the prototypical contexts being coextensive to

Varieties of English from a cross-linguistic perspective

499

the ones identified for adnominal intensifiers in (7) above, i.e. they either refer to persons of high rank or those used as a basis for the identification of others. The third important usage of adnominal intensifiers mentioned in (7) is to mark the centre of perspective in a discourse domain (the so-called logophoric use). There are many minimal pairs (extensively discussed in Kuno 1987: 118ff.) showing that the acceptability of free se^-forms decreases markedly if they do not refer to the centre of perspective (cf. [44], [45]). (44) a. According to John, the article was written by Mary and himself. b. *Speaking of John, the article was written by Mary and himself (45) a. As for myself I won't be invited. b. ? ?As for yourself, you won't be invited. c. *As for himself, he won't be invited. Finally, it is certainly no coincidence that in regional varieties of English free ie/f-forms in subject position must refer to somebody important (cf. [40] and [41] above). The second point that we have to show to make our argument complete is why adnominal intensifiers occur without a proper nominal head, are headless intensifiers as it were. The most important piece of information to motivate this point is that adnominal intensifiers cannot be added to object pronouns. Consider (46) in this context, where a full NP has to be used to allow intensification. (46) a. ???I want to talk to him himself. b. I want to talk to the man himself Native speakers of English normally willingly acknowledge this restriction, but we can easily draw additional support from text corpora. A search through the BNC (100 m. words) has yielded just one such example (cf. (47)). Moreover, it is just as rare for me, you, us, them to occur in object positions followed by the adnominal intensifier (cf. Siemund 2000: 56ff.). (47) And he besought his mother that she would love her even as she loved him himself and that she would do good to her and show her great honour, for which he should ever serve her with the better goodwill. [BNC] By comparison, intensifiers in their adverbial use can readily occur adjacent to object pronouns (cf. [48]).

500

Peter Siemund

(48) She wanted somehow to have her mother for herself, but only so that she could reject her herself (adverbial) [BNC] We conclude that intensifiers in the function of what is commonly known as free self-forms occur without a pronominal head because sequences of object pronoun and intensifier are disallowed by the grammar of English. Of course, this does not explain why they are disallowed. In Siemund (2000: 69ff.) it is argued that such sequences or combinations are ruled out because English intensifiers contain a pronominal head in their morphological structure. Finally, we may turn to other languages to draw support for our line of argumentation from parallel phenomena. And clearly, there are many other languages that allow intensifiers (inter alia) in subject positions as long as they refer to the contextual point of reference (Russ. sam, Lat. ipse, Turk. kendi, Mand. ziji, Lith. pats). (49) a. Russian: sam skazal 'the master/the husband said' b. Latin: ipse/ipsissimus dixit 'the master said' We can conclude that an analysis of free self-forms as (headless) adnominal intensifiers evades many of the problems and complicated assumptions that are necessary if one tries to subsume these occurrences of self-forms under the notion of reflexive anaphor. Nevertheless, an issue that clearly requires further attention, specifically from a cross-linguistic perspective, is the omission of the nominal head. The conditions relevant for Standard English appear to be quite different from those relevant for similar phenomena in, say Russian or Latin, where the possibility to leave pronominal arguments implicit seems to play an important role. 8. Alternative expressions for intensifiers This effectively concludes our survey of intensifiers and reflexives across different varieties of English, but before summing up the preceding discussion let us briefly consider some phenomena which are interesting because they represent something like curiosities in this domain. They are all concerned with alternative expressions for intensifiers. A first interesting observation comes from Tok Pisin where we find the form tasol used as the adnominal and maybe even adverbial intensifier. The surprising fact is that the lexical source of tasol, namely 'that's all', so far is something unique

Varieties of English from a cross-linguistic perspective

501

that has not been attested by other languages. No more than merely mentioning this phenomenon can be done here. (50) Tok Pisin Mi tasol mi mekim. (tasol < that's all) Ί my self did it.' The other phenomenon that will be raised briefly for discussion here is exemplified in (51) and (52), the remarkable property being that combinations of personal pronoun followed by the adverb lane (< alone) are used in the function of adnominal (51) and adverbial (52) intensifiers (data are taken from Wright 1898-1905). (51) Scots a. Speyk a word or twa to God him lane ... b. Yet, quait, aside the fire himlane, Was harmless as the soukin' wean ... (52)

Scots a. She danc 'd her lane, cry'd, Praise be blessed, ... b. Lady Amelia dos not walk her lone as yett.

The use of alone in an intensifying function in itself certainly is not what makes these examples so remarkable: It is well known that expressions like alone warrant an analysis as focus particles and make a semantic contribution similar to x-self in an intensifying function.30 For many contexts alone is the closest paraphrase to se^-intensifiers, particularly in their adverbial use. Some languages lost their se/^intensifiers during the course of their history and have filled the resulting gap with an expression corresponding to Engl, alone, cf. the examples from Yiddish in (53).31 (53) Yiddish a. Di photo hob ikh aleyn gemakht. b. Der direktor aleyn vet undz ufitemen. To be sure, what makes (51) and (52) so remarkable and quite exceptional in the context of other varieties of English is the occurrence of alone together with the pronominal form. This kind of evidence may retain important clues for our understanding of why English developed complex intensifiers in the first place.

502

Peter Siemund

9. Summary and conclusion In discussing varieties of English from a cross-linguistic perspective I have mainly tried to show that the facts reported from different varieties in the domain of reflexivity and intensification represent no idiosyncrasies or arbitrary deviations from the norm set by Standard English, but are frequently in harmony with cross-linguistics findings, generalizations and widely attested paths of development, and find their natural position in the great underlying groundplan. In addition, I hope to have convinced the reader that the restriction to the study of phonological variation imposed upon the field by traditional dialectology and socio-linguistics is too rigid and not necessary. The often quoted argument that there simply is not enough interesting morpho-syntactic variation around, or that there is a severe problem of authentic data, proves to be ill-founded once we move beyond the study of separate varieties in isolation and start investigating what different varieties have in common and in which areas they differ. The methodological toolkit made available by language typology is an important aid in this quest for the patterns and limits of variation. Notes *

1. 2.

3. 4. 5.

6.

7.

I would like to thank the audiences at the universities of Konstanz and Hamburg as well as the participants of the symposium Determinants of Grammatical Variation (Paderborn, June 2000) for helpfiil comments and suggestions. Alternatively we may say that subject and object are co-indexed or coreferential. Notice that adnominal intensifiers are difficult to contextualize when they are combined with foci the values of which are not the contextual point of reference: IThe porter himself will talk to us. / IMary's husband himself looks after the children. Mary works in a hospital. The interested reader is referred to Siemund (2000). In view of the well known historical (biblical) circumstances no ambiguity can possibly arise in the following example: Judas aheng hine. 'Judas hanged himself.' Case marking is quickly lost in the further development of English and it cannot be excluded that the adverbial use of intensifiers also contributed to the rise of complex 5e//-forms (cf. König and Siemund 2000b). In OE documents there are also combinations of object pronoun and intensifier which are not coreferential with the subject, but in these cases the person picked out by the pronoun can be shown to be the contextual point of reference. One noteworthy result of these processes is an incongruity in the paradigm of complex self-forms with object forms of the pronoun being used in the third person and possessive forms in first and second person. Non-standard varieties typically have possessive forms also in the third person: Standard: myself, yourself, himself ourselves, yourselves, therrwe/ves. Varieties: myself, yourself hisself ourselves, yourselves, theirse/ves

Varieties of English from a cross-linguistic perspective

503

8. The data are taken from Muysken and Smith (1995). 9. I avoid using the term 'indirect object' because for some of the pronouns found in the subsequent examples it is not entirely clear that they belong to the argument structure of the respective verbs. 10. It is noteworthy in this context that Old High German had a special reflexive only in the accusative case. In the dative personal pronouns were used to express reflexive relations. 11. The use of a self-form is not possible in (17c) for reasons that cannot be discussed here. Susan Carrol (p.c.) points out that the use of complex se//-forms in adjunct position is the unmarked case in Canadian English: John put a blanket under himself 12. In German the reflexive is required as long as its antecedent is in the same minimal clause. 13. Note that English reflexives agree with their antecedent in person, number and gender. 14. The ordering of first and second person relative to one another is not without problems. Huichol, to the best of my knowledge, is the only language for which the existence of second and third person reflexives to the neglect of first person reflexives has been claimed, but even this is not undisputed. A hierarchy like 3 > 1, 2 is probably more adequate. 15. See Plank and Schellinger (2000) on the relationship between synchronic and diachronic universals. 16. A structural explanation of these developments can be found in van Gelderen (2000). Interesting observations concerning reflexive uses of the personal pronouns of the first and second person {I criticize me) are found in Haiman (1995). 17. For a first attempt one may consult König and Siemund (2000a). 18. The first two items in this list represent parts of the well known grammatical relations hierarchy (cf. Croft 1991). 19. This study is based on a sample of almost 200 languages which can claim representativeness. 20. It is well known (cf. Haspelmath 1990) that reflexive pronouns may develop into such markers of derived intransitivity (cf. German, Spanish, Russian, etc.), a frequently attested path of grammaticalization being the following: reflexive anaphor —• bodily motion —> anticausatives —• facilitatives —• passives The following examples from German illustrate these developments: Er sieht sich. 'He sees himself —> Die Tür öffnet sich. 'The door opens.' —» Das Hemd wäscht sich gut. 'The shirt washes easily.' 21. In addition, the relevant expression will not undergo cliticization nor morphologicization. 22. See Wegener (1985) for an extensive discussion of extra-thematic pronouns (datives) in German. 23. Notice the existence of this phenomenon in French: Je me mange une pizza, 'literally: I eat me a pizza.' 24. As far as I know this has not been done so far. 25. Less widespread strategies are expressions meaning 'return' and 'reflection'. 26. There is a proposal by Grimm (1967) according to which the etymology of Engl, self, Germ, selbst, etc. can be reconstructed as reflexive plus body part noun. 27. The reflexive pronoun of Basque is buru (< 'head') whereas the intensifier is ber(etymology unknown). 28. There are important differences between these theories which are impossible to discuss here.

504

Peter Siemund

29. Andrew Pawley (p.c.) informs me that the reflexive of Tok Pisin (yet) may also be used as an intensifier. 30. Useful in this context is a distinction between core intensifiers (x-self) and non-core intensifiers (alone). 31. Semantically there seems to be a connection between 'alone' and Tok Pisin 'that's all'.

Primary Sources |7E1G] [JEIS]

[BNC] [BROWN] [CT] [E]

[Gen]

[LLC] [Max] [Or] [RC]

Aelfrics Grammatik und Glossar, ed. by Julius Zupitza. Berlin: Weidmannsche Verlagsbuchhandlung. 1966. Aelfric's Lives of Saints, ed. by Walter William Skeat. London: Eearly English Text Society 76. 1881. The British National Corpus The Brown Corpus The Canterbury Tales by Geoffrey Chaucer, ed. by Norman F. Blake. London: Edward Arnold. 1980. Euphues - The Anatomy of Wit by John Lyly (1578). In: The Complete Works of John Lyly, ed. by Richard Warwick Bond. Oxford: Clarendon Press. 1902, 177-375. The Old English Version of the Heptateuch. Aelfric's Treatise on the Old and New Testament and his Preface to Genesis, ed. by Samuel J. Crawford. London: Early English Text Society 160. 1922. The Longman-Lancaster Corpus Maxims I. In: The Exeter Book, ed. by Elliot Van Kirk Dobbie and George Philip Krapp. New York: ASPR 3. 1936. King Alfred's Orosius, ed. by Henry Sweet. London: Early English Text Society 79. 1883. Robinson Crusoe, by Daniel Defoe (1719). London: Dent (Everyman's Library). 1972.

References Abraham, Werner and Joseph Bayer (eds.) 1993 Dialektsyntax. Opladen: Westdeutscher Verlag. Anderwald, Lieselotte 2001 Negation in Non-Standard British English. London: Routledge. Baker, Carl L. 1995 Contrast, discourse prominence, and intensification, with special reference to locally free reflexives in British English. Language 71: 63-101. Benincä, Paola (ed.) 1989 Dialect Variation and the Theory of Grammar. Dordrecht: Foris. Black, James R. and Virginia Motapanyane 1996 Microparametric Syntax and Dialect Variation. Amsterdam: Benjamins. Chomsky, Noam 1981 Lectures on Government and Binding. Dordrecht: Foris.

Varieties of English from a cross-linguistic perspective

505

Christian, Donna 1991 The personal dative in Appalachian speech. In: Peter Trudgill and Jack Chambers (eds.), Dialects of English: Studies in Grammatical Variation, 11-19. Longman: London. Dryer, Matthew, Martin Haspelmath, David Gil and Bernard Comrie (eds.) 2003 World Atlas of Language Structures. Oxford: Oxford University Press. Farmer, Ann K. and Robert M. Hamish 1987 Communicative reference with pronouns. In: Marcella Bertuccelli-Papi and Jef Verschueren (eds.), The Pragmatic Perspective, 547-565. Amsterdam: Benjamins. Farr, James Marion 1905 Intensives and Reflexives in Anglo-Saxon and Early Middle-English. Ph.D. dissertation. Baltimore: Fürst. Filppula, Markku 1999 The Grammar of Irish English. London: Routledge. van Gelderen, Elly 2000 A History of Reflexive Pronouns. Amsterdam: Benjamins. Geniuäiene, Emma 1987 The Typology of Reflexives. Berlin: Mouton. Grimm, Jacob 1967 Deutsche Grammatik. Hildesheim: Georg Olms Verlagsbuchhandlung [reprint], Haiman, John 1995 Grammatical signs of the divided self. In: Werner Abraham, Talmy Givön and Sandra A. Thompson (eds.), Discourse Grammar and Typology, 21334. Amsterdam: Benjamins. Haspelmath, Martin 1990 The grammaticization of passive morphology. Studies in Language 14: 2572. Keenan, Edward L. 2001 Explaining the creation of reflexive pronouns in English, University of California at Los Angeles, ms. Kemmer, Suzanne 1993 The Middle Voice. Amsterdam: Benjamins. Klemola, Juhani 1996 Non-standard periphrastic do: A study in variation and change. Unpublished Ph.D. dissertation, University of Essex. Kortmann, Bernd 1997 Typology and dialectology. In: Bernard Caron (ed.), Proceedings of the 16th International Congress of Linguists, Paris 1997 (CD-ROM). Amsterdam: Elsevier Science. Kortmann, Bernd 2001 New prospects for the study of English dialect syntax: Impetus from syntactic theory and language typology. In: Sjef Barbiers (ed.), Syntactic Microvariation. Amsterdam: SAND. König, Ekkehard 1991 The Meaning of Focus Particles. A Comparative Perspective. London: Routledge.

506

Peter Siemund

König, Ekkehard and Peter Siemund 2000a Intensifies and reflexives: a typological perspective. In: Zygmunt Frajzyngier and Traci S. Curl (eds.), Reflexives: Forms and Functions, 4 1 74. Amsterdam: Benjamins. 2000b The development of complex reflexives and intensifiers in English, Diachronica XVII: 39-84. 2000c Locally free se//"-forms, logophoricity and intensification in English. English Language and Linguistics 4 (2): 183-204. Kuno, Susumo 1987 Functional Syntax: Anaphora, Discourse and Empathy. Chicago: Chicago University Press. Muysken, Pieter and Norval Smith 1994 Reflexives. In: Jacques Arends, Pieter Muysken and Norval Smith (eds.), Pidgins and Creoles, 271-288. Amsterdam: Benjamins. Penning, Gerhard E. 1875 A History of the Reflective Pronouns in the English Language. Inaugural Dissertation, University of Leipzig. Bremen: Heinrich Frese. Plank, Frans and Wolfgang Schellinger 2000 Dual laws in (no) time. Methodology in Linguistic Typology. Special Issue of Sprachtypologie und Universalienforschung 53 (1): 46-52. Reinhart, Tanya and Eric Reuland 1993 Reflexivity. Linguistic Inquiry 12: 657-720. Ross, John R. 1970 On Declarative Sentences. In: Roderick A. Jacobs and Peter S. Rosenbaum (eds.), Readings in English Transformational Grammar, 222-72. Waltham, MA.: Ginn, and Co. Schladt, Mathias 2000 The typology and grammaticalization of reflexives. In: Zygmunt Frajzyngier and Traci S. Curl (eds.), Reflexives: Forms and Functions, 103-124. Amsterdam: Benjamins. Siemund, Peter 2000 Intensifiers in English and German: A Comparison. London: Routledge. forthc. Pronominal Gender in English: A Study of English Regional Varieties from a Cross-linguistic Perspective. London: Routledge. Turley, Jeffrey S. 1997 The renovation of Romance reflexives. Romance Philology LI: 15-34. Wegener, Heide 1985 Der Dativ im heutigen Deutsch. Tübingen: Narr. Wright, Joseph 1898-1905 The English Dialect Dictionary. London: Harry Frowde. Zribi-Hertz, Anne 1989 Anaphor binding and narrative point of view. English reflexive pronouns in sentence and discourse. Language 65: 695-727.

Non-standard English and typological principles: The case of negation Lieselotte Anderwald

Abstract In this paper, I will try to answer the question of why older traditional dialect forms are being supplanted by non-standard forms in general use rather than the relevant Standard English forms, even though Standard English is available too. The inspiration for this investigation is provided by the recently established research paradigm combining functional typology and the analysis of dialect grammar (in the sense of Kortmann 1999, 2002 and Siemund, forthcoming). Concentrating on the domain of negation, this paper is based on a first in-depth study in this new paradigm. The spoken sections of the British National Corpus (BNC) have served as the main data base, as they - for the first time - make possible quantitative comparisons across Great Britain. In particular, some non-standard features in morphological paradigms of everyday spoken British English are investigated that are so frequent and widespread that they have been called general sociolinguistic markers: the use of don't in the third person singular (he don't, she don'?), the use of ain 't for all persons for have and be, and the non-standard use of was and were (i.e. we was, you was, they was·, I were, he were, she were). These features of general non-standard spoken English are linked to general typological principles which suggest a cognitive explanation for the pervasiveness of these phenomena.

1. Introduction Functional (Greenbergian) typology and dialect grammar have not been linked so far, although a combination of these two fields promises very interesting results (cf. Kortmann 1999; Anderwald and Kortmann 2002). Applying the methods and insights from functional typology not just to unrelated languages, or to a group of related languages, but to several dialects of a single language should reveal the breadth of variation that is possible within a language as well as suggest functional explanations for dialectal variation that have not been considered before. It is generally accepted that dialects are as complete and logical as languages - indeed it is very difficult to find purely linguistic criteria for what constitutes the cut-off point between different dialects (of the same language) or different languages. This differentiation is very often dictated by non-linguistic considerations (Chambers and Trudgill 1998: 4 enumerate

508

Lieselotte

Anderwald

political, geographical, historical, sociological and cultural criteria) - so much so that in comparative linguistic work, a difference is usually not even made, and authors speak of languages or dialects interchangeably. As is also well known, it is mostly a result of historical coincidence which dialect becomes the basis for a standard language (although during the process of standardization recurrent patterns can be encountered, cf. Stein 1998). Despite these difficulties, it is uncontroversial that dialects are naturally evolved varieties of language, and general findings made on the basis of cross-linguistic comparisons, claimed to hold for all human languages, must therefore also hold for dialects. On the other hand, the investigation of several dialects of one language is intuitively different from the investigation of several unrelated or related languages, and dialectologists in particular have tried to capture this difference. Dialects of one language can perhaps be expected to be more similar to each other in several respects than different languages. Thus, we would expect the same basic word order, mutual comprehensibility, heteronomy with respect to the same standard language etc. (cf. the discussion in Chambers and Trudgill 1998: 3-5). Nevertheless, recent studies in dialect grammar have shown that non-standard dialects can differ in interesting and non-trivial ways from each other or from their standard roof, which calls for more comparative studies on more grammatical phenomena. This paper uses some of the explanatory tools provided by functional typology to analyze certain aspects of negation. 2. Negation in functional typology Negation has been called a pragmatic universalevery language needs to be able to express negation to be fully functional. This even holds for artificial languages - the negative operator is one of the basic logical functions and cannot be expressed by means of other, more basic or atomic operators. Strategies for negation are present in every language - and, thus, in every dialect - and should therefore be easily comparable. Surprisingly enough, even functional typology has so far largely neglected the area of negation. The first cross-linguistic comparative studies did not appear until the 1980s: first overviews were provided by Dahl (1979) and Payne (1985). In a similar vein, Dryer (1988) relates negative strategies to basic word order, but all three studies seem to have evolved independently of each other. A notable collection of papers of more in-depth studies on a range of languages appeared in 1994 (Kahrel and van den Berg [eds.]). Finally, the

Non-standard English and typological principles

509

most detailed analysis is Bernini and Ramat's Negative Sentences in the Languages of Europe from 1996 (the Italian version dates from 1993). Bernini and Ramat in particular have developed detailed and explicit criteria for the comparison of European languages, of the kind "if a language is N1 [i.e. allows only one negative item per clause, LA], then it must also have postverbal negation" (Bernini and Ramat 1996: 187), and they compare their languages along several parameters: whether the same morph is used for negative answers and sentence negation, whether sentence negation and prohibitions (i.e. negative imperatives) use the same morph etc. (1996: 210-219). When we try to apply these criteria to English dialects, however, it becomes apparent that these varieties are simply not varied enough on the very basic criteria that Bernini and Ramat have developed so successfully for the European languages. For example, the basic word order is the same, the basic negative strategy - negation by addition of a particle - is the same in all dialects of English, including the standard, etc. Paradoxically, therefore, the results from typological studies of negation are both too specific and too general to find much application in this investigation. Instead, we shall have to fall back on very general typological criteria, in particular in the field of markedness, which will then lead to interesting results. The following paragraphs investigate negation patterns, mainly morphological paradigms, in non-standard British English especially in comparison with the standard that is well-described and, incidentally, the usual basis for typological comparisons. Differences between the systems are well known; features like ain't for example are very wide-spread (not only in Great Britain), and are also highly stigmatized; other features, for example the use of he don 't/she don't or past tense was or were where they do not belong, are not mentioned as often, but are just as frequent. The emphasis of this investigation therefore lies less in the discovery of new phenomena than in the collection and quantification of well-known features of the negative system. These features are compared on a regional scale that has not been possible before. They are then analyzed against the background of functional typology, which can provide cognitive motivations for the very general underlying trends that can be discovered. 3. Markedness patterns Negation is generally treated as marked in the typological literature, when compared to affirmation (cf. Greenberg 1966: 50; Croft 1990: 93) and there

510

Lieselotte A nderwald

are a number of general criteria that support this analysis. The general criteria for markedness are set out in Greenberg's Language Universals (1966), and they are summarized and discussed in Croft (1990), Typology and Universals, where chapter four is devoted to "Markedness in typology". It will be uncontroversial to call negation a feature of morphosyntax; for this reason I will not consider the criteria for phonological markedness patterns in any detail here. The criteria set out by Greenberg and repeated (and grouped) in Croft are the following: Table 1. Markedness criteria Name

Explanation

SI

ZERO VALUE

the unmarked value is typically realized by zero

S2

SYNCRETIZATION

S5

IRREGULARITY

S8

FREQUENCY

S3

FACULTATIVE USE

S4

CONTEXTUAL NEUTRALIZATION

S6

DEFECTIVATION

the unmarked value has greater range of grammatical behaviour

S7

DOMINANCE

number/gender only

the unmarked value typically has more distinct forms in the paradigm the unmarked value typically has more allomorphs or is more irregular the unmarked value typically is more frequent in text counts the unmarked value can refer to both in certain contexts only the unmarked value appears in certain contexts

The last four criteria will be excluded from the following discussion for a variety of reasons: Croft argues that the two criteria applying to neutralization (S3 and S4) do not belong to markedness criteria in general, as neutralization seems to have a different theoretical status (Croft 1990: 89-91); criterion S6 (defectivation) even for Greenberg "can be considered a form of syncretism" (1966: 29), and can thus be subsumed under S2, and the last criterion S7 is a specific one that applies to the categories of number and gender only and therefore does not play a role in this investigation. In the following discussion, I shall therefore consider the first four criteria of this table (SI, S2, S5 and S8) only. Let us examine now how they apply to Standard English.

Non-standard English and typological principles

511

4. Negation in Standard English Negation in Standard English is effected in the following way: The negative particle not is added to a positive clause, if this clause contains an auxiliary or a form of be. Otherwise, a form of the so called dummy do or dosupport is added as well. This procedure is summarized in a flow chart in Figure 1.

Figure 1.

Negation in Standard English

This leads to the following asymmetries between affirmative and negative clauses displayed in examples (1) to (3). (1)

She works hard.

She does not work hard.

(2)

I have finished.

I have not finished.

(3)

The children are playing.

The children are not playing.

Already visually it is clear that the negative clauses are longer, Standard English thus clearly conforms to markedness criterion SI: the unmarked value (a positive clause) is not marked explicitly by an affirmation marker, but by zero, whereas the negative clause is clearly marked by the addition of (at least) the negative morpheme, in many cases by the addition of an auxiliary do as well.3

512

Lieselotte

Anderwald

Another criterion is also clearly fulfilled for Standard English, the frequency criterion S8. Negative clauses are much rarer than their positive counterparts; although text counts vary considerably, figures from my investigation suggest a ratio of between 1:7 and 1:10 for contemporary spoken English (one negative clause for ten positive ones). The remaining two criteria however are not fulfilled: positive and negative paradigms of Standard English verbs have an equal number of distinctions, and the positive paradigms are by no means more or less irregular than the negative ones, as Figure 2 shows.

Figure 2. Symmetrical Standard English paradigms

Strictly speaking, an equal number of distinctions does not constitute a counterexample to a markedness pattern; the exact definition is that the unmarked value has "at least as many distinct forms in the same paradigm" (Croft 1990: 79) as the marked one, and this is clearly the case for Standard English. A counterexample undermining the proposed markedness relation would only be constituted by a reversal of the expected order (in this case, if one found more distinctions in the negative paradigm than in the positive one). On the other hand, a simply equal number of distinctions offers no positive support for assuming a markedness relation between positive and negative contexts. Although we can say that for Standard English, negation is clearly marked (namely by the two criteria S1 and S8), it does not fulfil all criteria for markedness, but is simply neutral with respect to criteria S2 and S5. 5. Negation in non-standard English Let us now turn to negation in non-standard English. More precisely, the following discussion will focus on present-day colloquial British English. The data comes from the British National Corpus (BNC), particularly from its section of spontaneous speech, where selected respondents taped their everyday conversations for a number of days.4 At roughly five million

Non-standard English and typological principles

513

words this represents a sizeable corpus that is unparalleled by any other database of unmonitored everyday speech. More importantly for our present purposes, most speakers are assigned an accent label that gives some information about their regional background so that regional comparisons with older data become possible. This investigation will concentrate on the primary verbs (for obvious reasons: only they - and the auxiliary verbs can take negation, and they are frequent enough to allow meaningful quantitative comparisons to be carried out across regional variants). Because the verb paradigms investigated here are particularly frequent words, automatic searches had to be restricted in a principled way, and as a result only combinations of personal pronouns or there with the verb forms were searched. The BNC subsample was investigated per dialect area, and where possible the results were compared with the older data from the nation-wide Survey of English Dialects (SED), collected in the 1950s and 60s, in order to detect indications of diachronic change.

5.1. Don't The use of positive do in the third person and also of negative don't has been reported for a fairly long time. Current examples are provided in (4) and (5), their non-standard status is also demonstrated by the simultaneous presence of multiple negation5: (4)

He don't have no dinner. (KCT 4276)

(5)

Don't she know nothing? (KC2 3159)

It is obvious that this use of don 7 is restricted to the auxiliary do, as the main verb is always negated with c/o-support. By contrast, positive thirdperson do used to be relatively restricted regionally in the rural dialects, basically to East Anglia and a large area of the south-west of England. For East Anglia, third-person do (instead of does) is part of a wider dialect phenomenon as all present-tense verbs lack third-person -s. In 1974, Trudgill still recorded almost categorical lack of third-person -s in Norwich for the lower working class (Trudgill 1974). Negative don't on the other hand is much more wide-spread, even in the traditional dialects, and links the two areas East Anglia and the south-west where we find positive third-person do, as is shown in Map 1, combined from Maps 34, 35 and 37 in Orton et al. (1978).

514

Map 1.

Lieselotte

Anderwald

Third person singular do and don 7, SED

3rd person do + don Ί

If we compare this older data with results obtained from the spoken sections of the BNC, we can see clearly that in spontaneous spoken language

Non-standard English and typological principles

515

today, positive do has all but disappeared: The whole BNC spoken section records only 32 occurrences of positive do in the third person singular, most of them in East Anglia and the south-west, against a total of almost 3,000 instances of negative don't, even though positive verbs are several times more frequent than negative ones; negative don't on the other hand seems to be on the increase. The detailed figures in Appendix 1 show that third-person don't today is present in practically every dialect area throughout Great Britain, and with an average of 28% is a highly frequent phenomenon indeed. There are no significant regional differences in the relative frequency with which third-person singular don't is used. We can observe, however, that there is a difference in how widespread this feature is: in some dialect areas a wide range of environments of subjects and sentence types allows third-person singular don't, whereas in others thirdperson don't occurs only in a very restricted subset of environments: the rare interrogatives have the smallest range of subject expressions involving third person singular don't. Conversely, the most frequent syntactic environment of declaratives has a much wider range of subject pronouns with third person singular don't. The regional distribution of these differences is displayed in Map 2. Although some details of this distribution, in particular the white areas, are due to the lack of reliable data, what becomes apparent is a north-south divide for this phenomenon that still corresponds roughly to the older SED map for third-person singular don't. The south and the Midlands area seem to use don't extensively, in all environments with all subjects. The north of England, Scotland and Ireland use this feature in a much more restricted way, as detailed above, which is not surprising if we assume that it may have been spreading from the south over the last few decades. In view of the fact that the BNC speakers are roughly representative of all social and regional groups of Great Britain, this feature displays a remarkably high average. It is equally noteworthy that third-person don't today is found almost everywhere. Particularly remarkable is the fact that this spread has only occurred for the negative form; positive he do/she do, which even at the times of the SED had a much narrower regional distribution, has virtually died out today. In addition, other, especially main verbs also show no weakening of third-singular -s. For this reason one cannot say that the motivation for this change is straightforward simplification or a general weakening of the rules of Subject-Verb Concord ; after all, the person/number distinction do-does has not been abolished systematically. It has been weakened, but this weakening is restricted to negative environments. Before we look for explanations for this phenomenon, we shall

516

Lieselotte

Anderwald

investigate two other non-standard negative paradigms where we encounter similar patterns. Map 2.

Regional distribution of don't, BNC

Non-standard English and typological principles

517

5.2. ain't Ain 7, probably the best-known shibboleth of non-standard English, is used world-wide as the negative form of both be and (auxiliary) have. Its history is still debated, as it can be plausibly derived from contractions of am not, are not, is not, have not and has not, i.e. from all forms where it is used today.7 Ain't today is highly stigmatized, but nevertheless widely used. Overall, ain't presents a picture similar to third-person don 7, also when we look at regional distributions. Again, data from the SED will be presented first. In those maps where forms of negated have are displayed, no forms of ain't occur.8 The following discussion will therefore be restricted to ain't used for the paradigm of be, as in examples (6) to (9)9. These examples show that ain't can be used both for copula be, and for the auxiliary. (6)

It ain 7 a bad night. (KB7 9365)

(7)

She ain't really worried. (KBE 2257)

(8)

It int [sic] gonna take much sorting. (KBD 7503)

(9)

Everybody 's so poorly, ain 7 they? (KCP 6509)

Map 3 gives details of ain't and in 7 used for forms of present tense be in the older data from the SED. Again we see a regional restriction; the heartland of ain't and in't seems to be the south-east, but also extends into the east Midlands. Comparing the data from the SED with present-day material, data from the BNC again shows that ain't today is spread through the whole of Great Britain10; with an average of around 10% however it is not nearly as frequent as third person don't, which is used in over one quarter of all cases. This may indeed be an indication of the stigmatized status of ain 7. What becomes apparent from the regional comparison is a striking difference in frequency, as Map 4 shows: there is one group of dialect areas - the majority - (marked by a darker grey) where the frequency of ain 7 is persistently high. It is at least in double figures, in fact clustering around 15%. This group extends throughout England in a homogeneous area, stretching from the central north over the east and central Midlands to East Anglia and to the lower south-west, the traditional heartland of ain't. For the second group of dialect areas, marked by a lighter grey, ain 7 only occurs with a frequency of around 5%, significantly less frequently than for the first group.

518

Map 3.

Lieselotte A ndenvald

A in't and in't, SED

Non-standard English and typological principles

Map 4.

A in't and in't for be, BNC

519

520

Lieselotte

Anderwald

This second group is represented in particular by the mid-West (Wales, west Midlands, Merseyside and the upper south-west) as well as Scotland and Ireland, traditionally completely ain't-free dialects, which show the lowest percentages.11 Again, we see a remarkable regional spread of a nonstandard form which levels the person/number-distinctions, in this case of present-tense be (and, of course, also of have), but again this levelling only takes place under negation. Even more so than for third-person do, where we do after all find a minute amount of person levelling in the positive paradigm, for present-tense be and have the positive paradigms are remarkably intact and show no sign at all of losing the distinction third person singular-non-third person singular. The pervasive regional spread of ain't then must be due to other reasons and cannot be described as straightforward simplification of the third-singular-non-third singular distinction.

5.3. was/were Generalization of was and were (as well as wasn 7 and weren't, which are more interesting for our purposes) to contexts where they are rejected in Standard English is a slightly more complicated feature than don't and ain't, because the past tense paradigm of be is the only paradigm in the past that has two morphological forms instead of just one (I was, he was but we were, you were, they were - and the same for the negative forms of course). In other words, there is a lot of room for variation, and indeed every possible combination is attested in non-standard spoken English: I were, he were, as well as we was, you was, they was, and the same for negative 10 forms. Some examples are provided in (10) to (13) . (10)

She were up sick all night. (KB8 6727)

(11) It was in the paper, weren't it? (KRO 2550) (12) All I know is that we was first. (KDA 7419) (13) So you were lucky to be in today really wasn'tyou? (KBC 5584) Unfortunately, this phenomenon is not dealt with in map form in the published SED material. The published maps are exclusively confined to positive contexts and neglect any negative equivalents. This is a pity since a

Non-standard

English and typological principles

521

comparison of the two strategies is the locus of interest from the point of view of this investigation. The most striking feature of irregular forms of past tense be today is that singular and plural pronouns display extremely divergent frequencies, as is evident from the analysis of the BNC (see Appendix 3). Were used with singular pronouns (/ were, she were) is used on average in only about 7% of all cases, whereas its plural counterpart (we was, they was) is almost twice as frequent (12% on average). Its negative equivalent we wasn't or you wasn't, however, is extremely rare, contrary to what one might perhaps expect: the generalization of negative wasn't only amounts to an average of 5% or a total of 30 occurrences in 595 - in a 5 million word corpus. Finally, I weren't, he weren't is extremely frequent, with an average rate of occurrence of almost 30%. This extremely high rate might be indicative of its sociolinguistic status, which does not seem to be particularly stigmatized. (This is in contrast to the highly marked ain't, but perhaps in parallel with third-person singular don't, as noted above.)13 Interesting patterns emerge when we look at combinations of these strategies. It is striking that out of the four logically possible patterns, only three are attested: some dialects generalize was to all persons, some (very few) generalize were, thus leading to straightforward simplification of the paradigms, but the majority seem to employ a mixed strategy: these dialects seem to prefer generalized was in positive contexts, but have generalized weren't in negative ones (for further details of this distribution see Anderwald 2001). For each dialect area, non-standard forms are much more frequent in negative contexts than in positive ones (in terms of relative frequency, not in absolute figures). The overall figures illustrate this: taken together, wasand were-generalization in positive clauses average at almost 8%. Generalization of wasn't and weren't in negative clauses on the other hand average at 24%, a highly significant difference. Generalization in negative clauses is therefore roughly three times as likely as in positive clauses, although negative clauses are much rarer - in the BNC data they account for only one in seven clauses, as the total figures in Appendix 3 show. It is striking that this simplification strategy is highly sensitive to the distinction of positive and negative contexts. Although the person/number distinction is not as robust for was/were in positive contexts as for the present tense paradigms of be or do above, negative contexts exert a remarkable influence on the frequency of generalized forms. This pervasive trend can be cognitively motivated, as the following section shows.

522

Lieselotte

Anderwald

6. Summary and discussion For all three phenomena, we can see that an older, regionally restricted pattern has spread to the rest of the country. Admittedly, there are still some regional differences in particular with respect to the use that is made of these three strategies: typically, in the traditional heartland many more combinations are possible, while in the newer areas only a more restricted use is permitted. Even so, there must be a reason why these three strategies are so successful that - even in the age of mass media and high mobility they are actually on the increase rather than dying out. Let us consider now how we can interpret these three phenomena in the light of the markedness patterns presented in section 3 above. There is no question that the two criteria that are met by Standard English are also fulfilled by non-standard English: positive clauses are unmarked (they have zero value), whereas negative clauses are marked by the addition of the negator. Although there are reliable indications that negation is much more frequent in spoken language than in written texts14, it is nevertheless still the case that in non-standard English, positive clauses are much more frequent than negative ones: the past tense paradigm of be suggests a ratio of 1:7. Non-standard English, just like Standard English, thus clearly conforms to markedness criteria SI and S8. Let us now look at the two remaining criteria, syncretization and irregularity. Recapitulating Greenberg (1966) and Croft (1990), we can state that the marked member, in our case the negative paradigm, typically has fewer allomorphs or distinct forms than the unmarked member (as stated by criterion S2). A trend towards this strong asymmetry in morphological paradigms between positive and negative clauses is exactly what we find in the non-standard system. In each of our three cases, there is a very strong tendency in nonstandard English to fuse or merge inflectional categories in negative contexts that are still kept distinct in positive ones. In this way, the only remaining person/number distinction in the positive paradigm do/does is fused or syncretized in the single form don 7; the contrast between was and were is neutralized in the case of weren7 in the majority of dialect areas, as is the distinction of am, is and are in the case of ain7, as Figure 3 shows. Although not discussed in detail here, the same holds true for positive has and have vs. negative ain 7, as I have shown elsewhere (Anderwald 2002: 116-150). This is why the motivation behind these phenomena cannot be a general trend towards simplification, since only the negative paradigms are simplified. The distinction singular-plural or third person-non-third

Non-standard English and typological principles

523

person for example is still available in the system, and it is still fully intact in positive contexts - perhaps with the exception of past tense be where, as we have seen, generalization in positive contexts still amounts to almost 8%, a not negligible amount. On the other hand, we have also noted that past tense be is already highly irregular in that it is the only past tense paradigm that possesses more than one form - a situation otherwise only found in the present tense paradigms, which might explain the pressure towards change in this particular positive paradigm, too.

Figure 3. Asymmetrical non-standard English paradigms

At the same time, the simplification of the negative paradigms makes them more regular than their positive counterparts, and in this way the nonstandard paradigms also conform to criterion S5 (according to which the marked member tends to be more regular, the unmarked member tends to preserve more irregularities). This is particularly visible in the case of waswere, where all other verbs only have one form for the past tense, and the irregularity of having two forms is only preserved to a degree in positive contexts (even though generalization is much more frequent in negative than in positive contexts). In this way we can draw a final diagrammatic comparison of Standard and non-standard English with respect to these markedness criteria: Table 2. Fulfilled markedness criteria Name SI S2 S5 S8

Standard English

Nonstandard English

ZERO VALUE

V

SYNCRETIZATION

-

IRREGULARITY

-

FREQUENCY

V

V V V V

What has become apparent from the preceding discussion is that there is a very strong trend for non-standard English to develop an asymmetrical system which conforms to all four markedness criteria, rather than just two as in Standard English. Several explanations have been proposed for this

524

Lieselotte Anderwald

phenomenon. Miestamo (2000) gives an ontological reason for asymmetrical paradigms as they are found cross-linguistically between affirmative and negative contexts: "In the asymmetric paradigms there is a 'vertical' analogy (or iconicity): the ontology of non-fact is less differentiated than the ontology of fact, and linguistic structure reflects this distinction" (Miestamo 2000: 78). These asymmetries can thus be iconically motivated. In addition, this asymmetry increases the number of differences between negative clauses and positive clauses in non-standard systems, where the standard often makes no distinction. In other words, we could describe the function of this asymmetry as making negation "more marked" in nonstandard systems. In this sense, increased morphological marking corresponds to the semantic status of negation as the marked member of the opposition affirmative-negative. Following Mayerthaler (1988), who explicitly makes the distinction between morphological featuredness and semantic/functional markedness, we can say that, again, this asymmetry is iconically motivated. Mayerthaler calls this constructional iconicity·, he claims that semantically more marked categories "are more often coded with features than are base categories. ... The intuitive basis for this is that what is semantically 'more', should also be constructionally 'more'" (Mayerthaler 1988: 18, cf. also Wurzel 1984: 203). Ultimately, increasing the difference between positive and negative clauses hugely increases the cognitive salience of negation. In addition to all social and psychological reasons that have undoubtedly furthered the spread of these non-standard forms they can therefore also be said to be functionally well-motivated, which might explain the pervasiveness of don't, ain't and weren't.

Non-standard English and typological principles

525

Appendix 1. Data from the BNC: Third person singular don't BNC-code XMS XEA XSS XMC XSL XNO XNC XLO XMI XWA XSU XME XNE XLC XHC XSD XMW XMD XIR XHM Total

Dialect area South Midlands East Anglia Central south-west England Central Midlands Lower south-west England Northern England Central northern England London Midlands Wales Upper south-west England North-east Midlands North-east England Lancashire Home Counties Scotland North-west Midlands Merseyside Ireland Humberside

don't!Total 33/51 113/183 123/236 76/162 34/79 8/22 53/151 184/543 22/70 45/180 7/32 24/111 18/96 25/163 43/294 5/49 27/272 5/56 2/35 0/210

% 64.7 61.7 52.1 46.9 43.0 36.4 35.1 33.9 31.4 25.0 21.9 21.6 18.8 15.3 14.6 10.2 9.9 8.9 5.7 0

847/2995

028.3

Appendix 2, Data from the BNC: A in't for be BNC-code XNC XMS XMC XMI XLC XNO XEA XSL XME XSS XLO XHC XMW XMD XWA XIR XNE XSU XSD XHM Total

Dialect area Central northern England South Midlands Central Midlands Midlands Lancashire Northern England East Anglia Lower south-west England North-east Midlands Central south-west England London Home Counties North-west Midlands Merseyside Wales Ireland North-east England Upper south-west England Scotland Humberside

ain V/Total 134/767 44/253 141/840 46/285 99/656 22/186 75/715 39/409 58/616 86/953 169/2447 100/1564 79/1250 12/202 31/756 6/200 14/468 3/150 4/306 0/172 1162/13195

% 17.5 17.4 16.8 16.1 15.1 11.8 10.5 9.5 9.4 9.0 6.9 6.4 6.3 5.9 4.1 3.0 3.0 2.0 1.3 0 0 8.8

526

Lieselotte Anderwald

Appendix 3. Data from the BNC: Non-standard was/were BNC-code XEA XHC XHM XIR XLC XLO XMC XMD XME XMI XMS XMW XNC XNE XNO XSD XSL XSS XSU XWA Total a)

Dialect areaa) East Anglia Home Counties Humberside Ireland Lancashire London Central Midlands Merseyside North-east Midlands Midlands South Midlands North-west Midlands Central northern England North-east England Northern England Scotland Lower south-west England Central south-west England Upper south-west England Wales

%

way/Total 88/214 35/522 2/76 7/97 21/276 88/874 37/358 2/46 21/230 6/77 7/134 36/421 32/337 19/158 13/75 35/231 8/106 68/282 13/55 64/354

41.1 6.7 2.6 7.2 7.6 10.1 10.3 4.3 9.1 7.8 5.2 8.6 9.5 12.0 17.3 15.2 7.5 24.1 23.6 18.1

602/4,923

0 12.2

were/Total 6/851 27/1,759 0/230 1/411 59/817 451/2,964 81/1,042 0/252 31/765 8/278 59/385 26/1,703 206/888 137/886 28/231 5/904 1/390 17/1,147 3/265 19/1,152 1165/17,320

% 0.7 1.5 0.0 0.2 7.2 15.2 7.8 0.0 4.1 2.9 15.3 1.5 23.2 15.5 12.1 0.6 0.3 1.5 1.1 1.6 0 6.7

Dialect areas in alphabetical order

Appendix 4. Data from the BNC: Non-standard wasn 't/weren 't BNC-code XEA XHC XHM XIR XLC XLO XMC XMD XME XMI XMS XMW XNC XNE XNO XSD XSL XSS XSU XWA Total

Dialect area East Anglia Home Counties Humberside Ireland Lancashire London Central Midlands Merseyside North-east Midlands Midlands South Midlands North-west Midlands Central northern England North-east England Northern England Scotland Lower south-west England Central south-west England Upper south-west England Wales

wasn 7/Total 0/16 1/64 0/8 2/6 3/41 6/114 2/49 0/8 3/27 0/6 2/23 6/47 0/50 0/14 0/10 1/12 0/18 1/39 0/4 3/39 30/595

% 0 1.6 0 33.3 7.3 5.3 4.1 0 11.1 0 8.7 12.8 0 0 0 8.3 0 2.6 0 7.7 0 5.0

weren 't/Total 61/114 39/231 0/30 1/49 63/214 133/409 36/136 4/24 24/101 26/49 26/63 13/226 59/134 19/79 9/41 5/77 33/67 110/211 5/46 37/165 703/2,466

% 53.5 16.9 0 2.0 29.4 32.5 26.5 16.7 23.8 53.1 41.3 5.8 44.0 24.1 22.0 6.5 49.3 52.1 10.9 22.4 0 28.5

Non-standard English and typological principles

527

Notes 1. Cf. Ramat (1987: 42), Bernini and Ramat (1996: 1). 2. Cf. Quirk et al. (1985: 775-778). The only exception to this figure is the use of full verb be and, marginally, have which both do not take do-support. There are also some other marginal cases (exceptions to the exceptions) that I shall not discuss further in this context. 3. This is not as obvious as it may sound; there are indeed languages (for example, Welsh) that possess a marker of affirmation (cf. Bernini and Ramat 1996: 110). I will not be concerned with the difference between example (1), which is asymmetrical between affirmative and negative clause through the addition of a form of do, and examples (2) and (3), which are not (in the sense of Miestamo 2000). Cf. Miestamo (2000) for an interpretation of this kind of asymmetry. 4. For more details cf. Aston and Bumard (1998). 5. Examples from the BNC; text codes in brackets. 6. As argued for example by Hudson (1999). 7. In African American Vernacular English, ain't is also regularly used for negative forms of do. As there is no derivation of ain't from do not/does not, this seems to be the result of an extension by analogy from the all-purpose use for be and have. This use is largely restricted to this dialect. 8. There are however a number of forms without /v/ or /z/, which still retain the initial /h/, e.g. in the SED transcription HAN'T HAAN'T HEN'T HAIN'T HANNO (cf. Orton et al. 1978, maps 42—44). I regard these as examples of secondary contracted forms and therefore as pre-forms of ain't/in't. For this reason they will not be considered further in this discussion. 9. Examples from the BNC. 10. Cf. Appendix 2. 11. Ain't is of course not a part of the traditional dialects of Scotland or Ireland, and even in the BNC the handful of instances may be due to freak occurrences, although there are some indications that ain't is currently spreading at least to Glasgow. 12. BNC. 13. This impression is also supported by Wolfram and Schilling-Estes, who similarly note a lack of stigmatization for non-standard forms of past tense BE, in particular for weren Ί as opposed to wasn't, in their data from Ocracoke, North Carolina (see Wolfram and Schilling-Estes 1996: 140-143). 14. See Biber et al. (1999: 159) or Tottie (1991: 17) for word counts.

References Anderwald, Lieselotte 2001 Was/were-variation in nonstandard British English today. English WorldWide 22: 1-21. 2002 Negation in non-standard British English: Gaps, Regularizations and Asymmetries. (Studies in Germanic Linguistics.) London/New York: Routledge.

528

Lieselotte

Anderwald

Anderwald, Lieselotte and Bernd Kortmann 2002 Typology and dialectology: a programmatic sketch. In: Jan Berns and Jaap van Marie (eds.), Present Day Dialectology: Problems and Findings, Volume 1, 159-171. (Trends in Linguistics 137.) Berlin/New York: Mouton de Gruyter. Aston, Guy and Lou Burnard 1998 The BNC Handbook: Exploring the British National Corpus with SARA.. Edinburgh: Edinburgh University Press. Bernini, Giuliano and Paolo Ramat 1996 Negative Sentences in the Languages of Europe: A Typological Approach. (Empirical Approaches to Language Typology 16.) Berlin/New York: Mouton de Gruyter. Biber, Douglas, Stig Johansson, Geoffrey Leech, Susan Conrad and Edward Finegan 1999 Longman Grammar of Spoken and Written English. Harlow: Longman. Chambers, Jack K. and Peter Trudgill 1998 Dialectology. Second edition. Cambridge: Cambridge University Press. Croft, William 1990 Typology and Universals. Cambridge: Cambridge University Press. Dahl, Osten 1979 Typology of sentence negation. Linguistics 17: 79-106. Dryer, Matthew S. 1988 Universals of negative position. In: Michael Hammond, Edith A. Moravcsik and Jessica R. Wirth (eds.), Studies in Syntactic Typology, 93-124. Amsterdam/Philadelphia: Benjamins. Greenberg, Joseph 1966 Language Universals. The Hague/Paris: Mouton. Hudson, Richard 1999 Subject-verb agreement in English. English Language and Linguistics 3: 173-207. Kahrel, Peter and Ren6 van den Berg (eds.) 1994 Typological Studies in Negation. Amsterdam/Philadelphia: Benjamins. Kortmann, Bernd 1999 Typology and dialectology. In: Bernard Caron (ed.), Proceedings of the 16th International Congress of Linguists, Paris 1997. CD-ROM. Amsterdam: Elsevier Science. 2002 New prospects for the study of English dialect syntax: Impetus from syntactic theory and language typology. In: Sjef Barbiers (ed.), Syntactic Microvariation, 185-213. Amsterdam: SAND. Mayerthaler, Willi 1988 Morphological Naturalness. Ann Arbor: Karoma. Miestamo, Matti 2000 Towards a typology of standard negation. Nordic Journal of Linguistics 23: 65-88. Orton, Harold, Steward Sanderson and John Widdowson 1978 The Linguistic Atlas of England. London: Croom Helm. Payne, John 1985 Negation. In: Timothy Shopen (ed.), Language Typology and Syntactic Description: Clause Structure, 197-242. Cambridge: Cambridge University Press.

Non-standard English and typological principles

529

Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech and Jan Svartvik 1985 A Comprehensive Grammar of the English Language. Harlow: Longman. Ramat, Paolo 1987 Linguistic Typology. Berlin: Mouton de Gruyter. Siemund, Peter forthc. Pronominal gender in English: a study of English regional varieties from a cross-linguistic perspective. Habilitationsschrift, Freie Universität Berlin. Stein, Dieter 1998 Syntax and varieties. In: Jenny Cheshire and Dieter Stein (eds.), Taming the Vernacular: From Dialect to Written Standard Language, 35-50. London: Longman. Tottie, Gunnel 1991 Negation in English Speech and Writing. San Diego: Academic Press. Trudgill, Peter 1974 The Social Differentiation of English in Norwich. Cambridge: Cambridge University Press. Wolfram, Walt and Natalie Schilling-Estes 1996 Dialect change and maintenance in a post-insular island community. In: Edgar W. Schneider (ed.), Focus on the USA, 103-148. Amsterdam/ Philadelphia: Benjamins. Wurzel, Wolfgang U. 1984 Flexionsmorphologie und Natürlichkeit: Ein Beitrag zur morphologischen Theoriebildung. Berlin: Akademie-Verlag.

'Every place has a different toll': Determinants of grammatical variation in cross-variety perspective* Sali A. Tagliamonte

Abstract Use of have, have got or got for stative, possessive meaning is said to distinguish varieties of English. However, this variation may also be viewed as grammatical layering. Using the comparative method and quantitative methodology this paper provides a comparative analysis of the internal and external constraints operating on this variability in three British communities. The communities differ in their favoured variant, however have is consistently favoured with abstract objects and generic readings. These are the very effects that have been attested in the history of the English language and thus can be interpreted as persistence. However, variable inter-variety distributions across generations as well as cross-dialectal differences suggest that the change is not progressing at the same rate in all circumstances. Another observation is that when the development of a form involves specialization, then a consequence may be that grammatical constraints on its distribution will strengthen as the change progresses.

1. Introduction Use of have, have got or got for stative, possessive meaning is a feature typically thought to distinguish varieties of English, particularly British vs. American dialects (Quirk et al. 1985: 131-132; Biber et al 1999: 159-163; 466). However, in Britain alone there is variability for this function, as illustrated in (1): (1)

a. He's got bad-breath; he has smelly feet. (YRK/i/521,14)1

This is one of the areas of English grammar that has been in the process of tremendous grammatical change, or drift, over the past four centuries. Yet as the examples in (2) attest, contemporary varieties have robust variation of forms, indicating grammatical layering. This presents an excellent opportunity to examine language variation and change which is ongoing at the present moment.

532

Sali A.

Tagliamonte

(2)

a. He 's got it but he hasn't got the will to force himself. (WHL/c/506,46) b. She's got such a coarse tongue in her. (BCK/s/498,0) c. We got eight wooden swords. (YRK/#/276,69) d. I have my moments where I have my doubts. ( WHL/e/204,11) e. I've arthritis of the spine. (WHL/e/868,52) f. What a bonny face she has. (BCK/Z/1249,10) g. And they haven't - don't have any roots to cling to. (YRK/h/535,11)

The construction with have is the oldest of the variants. Attestations of its use range from the late 10th century, in (3 a), right through to the 1960s, as in (3g) (all from Visser 1963-73: 1475) (3)

a. Nu we sind hlcene ncebbe we nan ping to etanne buton Manna. OElfric c. 970-1000, Num. 11,9) 'Now we are lean, haven't we no thing to eat except Manna.' b. Bot he haf wit to steir his stede. (c. 1375 Barbour, Bruce VI, 334) 'Unless he has wit to stear his steed.' c. A pouere ne3ebour... pat hath nat scarseliche a penny, (c. 13581400 Three ME Sermons (ed. Gisdale) 78, 910) Ά poor neighbour that has not scarcely a penny to spend.' d. Thei han no thing to ete. (c. 1380 Wycliff, Mt. 15, 32) e. What has pou in pi howse at sell? (c. 1425 Metric. Paraphr. Old Testament III (ed. Ohlander) 12183) f. He that had lyttle to spende, hath not much to lose. (1580 Lyly, Euphues and His England (Arb.) 243) g. I hae a penny to spend. (1788, Burns, Naebody 12) h. "If we can get horses", sighed he, "I have two to sell. " (1847 Thackeray, Vanity F. (Everyn.) 314) i. If riches have wings to fly away form their owner, they have wings to fly also to escape danger. (Oxford English Dictionary 'have' 3) j. Have you anything to hide? (1964 R. de Rouen, The Heretic (Heinemann) 72)

Around about the sixteenth century got was added which formed the construction have got and this started to be used for the same function (Crowell 1959: 280; Jespersen 1961: 47; Visser 1963-73: 2202).2 Use of have got can be found in Britain from at least the early 1600s onwards, as in (4)% (Visser 1963-73: 2203-2204).

Every place has a different toll

(4)

533

a. Hodge, what a delicious shop you have got! (1600 Dekker, Shoemaker 's Holiday [ed. Wheeler, in Six Plays by Contempo. of Shakesp] IV, ii, 51) b. Fie, th 'art a churle, ye haue got a humour there does not become a man. (1607 Shakespeare, Tomon I, ii, 26) c. I have got such a way of evasion upon the road, that I don't care for speaking truth to any man (1707 Farquhar, The Beam' Stratagem [Mermaid] III, ii, p. 393) d. He has got a tail (1827, Disraeli, Vivian Grey 2429) e. I've got a little of my own back. (1912, G.B. Shaw, Pygmalion IV)

The origins of the have got construction are controversial. Its earlier meaning, 'to have acquired', can be traced back to early Middle English (Visser 1963-73: 2202). In the sixteenth century, however, it came to mean simply 'possess' (Jespersen 1961: 47). Thus, a sentence such as The girl has got brown eyes means that she simply has them rather than that she has acquired them by some means. From the 18th century, Dr. Johnson's Dictionary 1755 prescribes that "He has got a good estate does not mean he has acquired it, but that he simply possesses it." (Visser 1963-73: 2202) According to Jespersen (1961: 47-48) use of have got arose because have was frequently being used as an auxiliary and people felt a need for another word to express or emphasize the idea of 'possess'. 4 Another hypothesis is that have got developed as a resolution of a conflict between stress and syntax (Crowell 1959). At least as early as the middle of the fifteenth century vowels in unstressed syllables were shortened or reduced in English (Wyld 1936: cited in Crowell 1959: 280). Then, in the sixteenth century the unstressed vowel schwa often dropped entirely; and the contracted forms in English developed.5 Thus, main verb have was often reduced, as in (5). (5)

a. b. c. d.

I've a very loud voice by the way. (BCK/e/2,9) He's a heart of gold. (BCK/p/451,24) I've no idea. (WHL/i/16,47) We've a very good relationship. (YRK/a/275,69)

According to Crowell (1959: 283) this reduction of have to a single sound [v] or [z], paved the way for the insertion of got as speakers responded to a need to mark the subject/verb relationship more overtly. The last variant to appear on the scene is the construction with got alone, as in (6) (Quirk et al. 1985: 132; Visser 1963-73: 2206).

534

Sali A. Tagliamonte

(6)

a. She's upstairs, she got her telly on. (WHL/a/319,0) b. They 're far far better than the stuff they got. (BCK/d/329,11) c. They got dodgy scenery. (YRK/®/3 50,40)

This variant is thought to be typical of "very informal English" (Quirk et al. 1985: 132), but is also said to distinguish British from American dialects. In the latter it is reported to be "particularly frequent" (Jespersen 1961: 53) and "widely used" from the mid-19th century (Visser 1963-73: 2205). It is often quoted for American literary works, as in (7). (7)

a. I got a right to know what she said. (1913, Booth Tarkingston, The Flirt [Lond, Hodder and St.] 36) b. Got your press card with you, just in case? (1940 Ayn Rand, The Fountainhead [Signet Bks] 379) c. We got some mail (1948, Irwin Shaw, The Young Lions [Signet Bks.] 450) d. I got this raking to do. (1964, H. Kemelman, Friday the Rabbi Slept Late [Penguin] 27) (all from Visser 1963-73: 2206)

Yet many commentators do not mention this variant, particularly for British English. Visser (1963-73: 2206) suggests that it is extremely difficult to collect as "many English speakers, who frequently employ the idiom in unconstrained and unpremeditated diction, change it into has got when putting pen to paper". By the late 1800s the variants of have got had attracted a great deal of social stigma. It was condemned as "vulgar", "not very beautiful" (cited in Rice 1932: 291) and got was even considered a "bogie word", a source of fear, perplexity, or harassment (Rice 1932: 292). This overt stigmatization is observable in the following quotes (cited in Rice 1932): Many persons, who consider themselves well educated, often commit... an error by using the participle got in connection with the verb to have; as, "I have got the book in my library"; "Who has got a pencil?" "I have got one." This form of speech should never be used to express possession, as the verb to have conveys that idea, in these and similar phrases. (Gwynne 1855) The most common misuse of this word is to express simple possession. It is said of a man that he has got this, that, or the other thing, or that he has not got it; what is meant being simply that he has it, or has it not - the use of the word got being not only wrong, but if right, superfluous. (White [1870] 1927) Get means to secure; got should not therefore be used unless the intended meaning is secured, nor have got unless the intended meaning is has secured. Wrong: Have you got a knife with you: Right: Have you a knife with you? (Wooley 1907)

Every place has a different toll

535

In 1915, Faulty Diction as Corrected by the Funk and Wagnalls New Standard Dictionary (cited in Rice 1932: 284) reports got as: "Properly having the sense of acquired, procured, and the like, but improperly used to express mere possession. Not"The hound has got long ears," because he had done nothing to get them; he has them." In fact, early 20th century investigations into the mis-use of English grammar record that the fourteenth most frequent error in English is the use of have got (cited in Rice 1932: 292). Given this backdrop, precisely how, and why, the have got variant was selected as the favoured variant (Rice 1932: 288) is not well known. Despite the condemnation of the use of have got by early 19th century writers as we just saw, British writers had been using the construction for at least 100 years or more. For example, Shakespeare, Lewis Carroll, Dickens, and Malory all used the construction and so had grammarians like Lindley Murray ([1795] 1968) and William Cobbett ([1818] 1983). Nor is it entirely clear when and how the bare got variant entered into this envelope of variation. Kroch's (1989) research has revealed that grammatical factors constraining this variation are consistent over time, corroborating the constant rate effect. However, little further attention has been devoted to a consideration of the grammatical determinants of this linguistic change, nor whether the historically attested effects for written corpora can be confirmed on contemporary dialect data. In this paper my aim is first to provide an accountable examination of all the variants of stative possessive meaning - have, have got and got. Further, I aim to compare the distribution and grammatical conditioning of these forms across a number of British dialects where this variability is current. My goal is to explore the possibility that this type of comparative analysis may facilitate tracking how the grammatical development from have to have got may have taken place and, if possible, how it is proceeding within the communities where it is still underway.

2. Data and methods The data on which this research is based were extracted from three corpora. Each represents a variety of contemporary (northern) British English in three different geographic locales. Each corpus was collected using standard sociolinguistic methodology (e.g. Labov 1970; 1971). They consist of dozens of hours of tape-recorded conversations, which include discussions about local traditions, narratives of personal experience, group interactions

536

Sali A.

Tagliamonte

and local gossip. All of these materials are highly informal and as far as possible represent the typical discourse found in each community. In addition, the corpora are stratified by age and sex. Stratification by age permits comparison of the pattern of variation in apparent time, both within and across each community, thus providing a view of possible linguistic change. Comparison of male and female speakers permits a rudimentary assessment of the social evaluation of the forms, a factor which has figured prominently in this variation at least over the last several hundred years. I now turn to a brief description of the communities. -

Buckie, Scotland

Buckie (BCK) is a small town on the far north shore of Scotland. While most of the population was traditionally involved in the now diminishing fishing trade, the community has not suffered from depopulation or unemployment because there are plenty of jobs on the oil rigs. For these and other reasons the community has had a long history of cultural cohesiveness (see Smith 2001; Smith and Tagliamonte 1998). -

Wheatley Hill

Wheatley Hill (WHL) is a village in County Durham in Northeast England. It was once an insular mining community. However, in the last few decades many of the mines have closed down, forcing the inhabitants to find work in nearby urban centres (Tagliamonte 2000-2001). A counter effect has come from a steady influx of urban populations into outlying villages. Both these socio-economic trends have led to a dilution of the traditional community networks as well as increasing extra-community connections. -

York

Finally, I will consider the city of York (YRK). This community has not undergone the same degree of demographic upheaval and dialect contact found in other British cities and so it still retains a somewhat conservative and unique character (for discussion see Tagliamonte 1998). Each context of stative possessive have, have got, or got alone in these materials was extracted from the data files, leading to a grand total of 1380 instances. Each context was then coded for a series of factors implicated in the trajectory of development of the forms. These factors were first analyzed individually in a distributional analysis and then simultaneously by means of Goldvarb 2.0 (Rand and Sankoff 1990), a logistic regression software package.

Every place has a different toll

537

3. Distributional analysis of selected determinants The next section details the distributional analysis of the overall variation between have, have got and got and then considers this variability according to a number of selected determinants extrapolated from the literature. 3.1. General survey Table 1 shows the overall distribution of have/got in each of the communities. Table 1. Overall distribution of variants by community have Community Wheatley Hill York Buckie

have got

got

Total

%

Ν

%

Ν

%

Ν

18 40 47

64 202 245

75 57 52

262 291 269

7 4 1

24 18 5

350 511 519

All three varieties make use of all the forms under investigation - have, have got and got; however, their proportions vary considerably. Got - the form typically labelled "American" is hardly ever used. Still, it is present, albeit infrequently. This use of got in these communities suggests that it may not be an entirely American phenomenon. Instead, it could have been present in British dialects by the turn of the 20th century, perhaps earlier. If so, however, it was clearly "embryonic" (Trudgill et al. 2000) in nature and has remained so. As far as have got is concerned, Table 1 makes clear that it is the most frequent form in every variety, but much more so in Wheatley Hill (75%), the community which has the most well-developed links with the supralocal (urban) community. However, take note of the cross-variety distribution of the oldest form, main verb possessive have, as in (8). (8)

a. b. c. d.

I have a bee in my bonnet about it. (YRK/1/301,29) He has two boys. (BCK/g/895,9) It has a lovely castle up on a hill. (YRK/b/269,26) In the wee bitty garden that we have at the back. (BCK/H/171,39)

This construction is most robust in Buckie (47%), the only variety which is Scots. Is this a reflection of dialect differences, essential differ-

538

Sali A.

Tagliamonte

ences in the underlying grammatical structure of the varieties, grammatical change, or some other mechanism? A dialect divide is not entirely explanatory because the form is nearly as frequent in York (40%). Even in Wheatley Hill, it is fairly frequent overall (18%). The differences amongst varieties are more quantitative than qualitative. Perhaps the most distinguishing aspect of the Buckie data is the fact that here the use of have and have got is nearly equal (47% vs. 52%). In such a system, what can explain the use of one form rather than the other? In fact, the differences in overall rate and type of variants across varieties that can be observed in Table 1 can be indicative of any number of explanations. One way to determine which is most explanatory is to consider evidence from the hierarchy of constraints on the internal factors that give rise to the different variants (see Poplack and Tagliamonte 1991: 124). Therefore, I now test the key linguistic measures of have/got variation that have been observed in the literature (e.g. Kroch 1989; Jespersen 1961; Quirk et al. 1985; Visser 1963-73) and further, I subject these constraints to comparative cross-variety scrutiny.

3.2. Contraction One of the most revealing tests of assessing the status of forms is to consider the effects of contraction. An examination of the forms of have according to whether they have been contracted or not, as in Figure 1, reveals that there is a marked contrast between the full versus contracted variants. The contracted forms are very infrequent, except in Buckie. In York and Wheatley Hill they hardly ever occur. Thus, the forms in (9) sharply differentiate the varieties. In Buckie they are typical; in York and Wheatley Hill they are rare. Moreover, in the latter two varieties they are virtually restricted to the oldest generation. (9)

a. b. c. d. e. f.

You've a miserable face. (YRK/>/83,18) You've nothing now have you, apart from a pub? (YRK/g/323,11) Amy's thirty in May and she's two little boys. (YRK/a/494,39) I've a wee roomy up the stairs. (BCK/e/120,56) She's a lovely house. (BCK/e/281,5) He's nae married to L, but he's two bairns. (BCK/z/233,51)

These results are consistent with the traditional view that contracted have is a feature of Scots dialects; however, the fact that both York and

Every place has a different toll

539

Wheatley Hill have this feature and that the forms are virtually confined to the oldest speakers suggest further that: 1) Buckie preserves an earlier stage in the development of this area of English grammar, and 2) that the forms are gradually disappearing as more have got is being used. Further, these findings provide evidence to support the hypothesis that one of the factors which led to the development of have got was the contraction of have. 60-1

5040% 3020100Buckie

Wheatley Hill • Full have/has

York

• Contracted W s

Figure 1. Distribution of full and contracted have/has by community.

3.3. Negation An important way to observe changes in grammatical status comes from structure changing operations such as negation. Of the available alternatives for negating have and have got consider the examples in (10): (10a) is considered British and formal, (10b) British and informal, and (10c) is thought to be North American (Quirk et al. 1985: 131-132; Visser 1963-73: 2205; Biber et al. 1999: 159-162).6 Of course (lOd) is highly stigmatized. However, this variant is considered American and is often η cited for literary works, as in (11) (Visser 1963-73: 2205). (10) a. b. c. d.

I haven't a headache. I haven't got a headache. I don't have a headache. I ain V got a headache.

540

Sali A. Tagliamonte

(11) a. / hain't got no money. (1884, M. Twain, Huckleberry Finn [Tauchn] I, 44) b. They ain't got no more clothes on them than a plucked duck! (1948 Irwin Shaw, The Young Lions [Signet Bks] 453) Figure 2 displays the proportion represented by each type of negation out of the total number of negated forms in each data set. Haven't and haven't got are robust variants across the board. Note, however, that negation is more frequent with have, as in (12), than the have got construction, as in (13). This contrasts with Jespersen's (1961: 49) observation that have got is more frequent with negatives. Furthermore note that the hierarchy is parallel across the varieties. This shows that although the varieties may be distinguished according to the proportion of have contraction, they are parallel with respect to the patterning of negation. (12) a. As much as I try. I haven't any strength to do it. (YRK/>/318,63 ) b. I've no idea. (WHL/i/16,47) c. You have na [not] a wrinkle. (BCK/9/1386,48) (13) a. I have na got a trade. I'm just a fisherman. (BCK/e/123,52) b. He hasn't got the will to force himself. (WHL/c/506,46) c. Because you haven't got your mates there with you. (YRK/*/l 19,4) 70 η 60 -

50 40 % 30 20 -

10 -

0-

Buckie

York

• haven 't/have no El haven't got/haven't got no

Wheatley Hill

• don't have/don't got

Figure 2. Distribution of different types of negation by community.

Every place has a different toll

541

Finally, note that the construction with do-support - another form regarded as American - is also present in small proportions in York and Wheatley Hill, as in (14): (14) a. We dort't have an automatic, we have a twin-tub. (YRK/e/206,59) b. She's now saying that he doesn't have the bairn [child], (WHL/7520,69) c. I don't really have that many friends at work. (WHL/@/331,43) The presence - albeit rare - of the construction with do-support in these British dialects demonstrates that have may have already begun to lose its auxiliary-like syntax by the turn of the 20th century in Britain. Once again this is supported by the external conditioning, where even the oldest speakers make use of the form. However, the number of instances of contracted have represent very small proportions of the York and Wheatley Hill corpora. Moreover, the number of negative constructions is very small, representing less 10% of each of the data sets. The explanation for the robust variability between have/have got and got must lie elsewhere. In the remainder of this paper I test the applicability of several internal grammatical constraints, each of which have been extrapolated from the literature which discusses the development of the have got construction. 3.4. Abstract/concrete One constraint that is often reported and perhaps the only one to have been systematically tested (Kroch 1989), involves the inherent nature of the object. Jespersen (1961: 47) observed that have got probably entered the language for use with objects denoting physical concrete things, then gradually spread to those denoting non-concrete, abstract objects. This is not surprising since got is historically, and in British dialects to this day, the past participle of get, whose original meaning, as mentioned earlier, is Ho acquire'. This is in accordance with the principle of persistence in grammatical change, which says that traces of the original lexical meanings of forms may be reflected in constraints on their grammatical distribution (Hopper 1991). Thus, we may still be able to observe this tendency in the distribution of have and have got such that have got may exhibit a higher tendency toward collocation with concrete objects and have with abstract objects.

542

Sali A.

Tagliamonte

Indeed, Kroch (1989: 207-209) reports consistent operation of such a constraint on the competition between have and have got in written corpora from 1750 to 1935. Figure 3 tests for this conditioning effect in the data as well as its application across varieties using the comparative method. Operation of this constraint is visible in two communities, York and Wheatley Hill. When the object is abstract have occurs more often than when the object is concrete. Thus, contexts such as in (15) are more likely to be lexicalized with have, than those in (16). Abstract: (15) a. They have the Baptist beliefs. (BCK/H/380,0) b. I've a feeling that they don't know. (YRK/™/526,86) c. I have my doubts whether I should be doing it. (WHL/e/249,71) Concrete: (16) a. I've got a bump [on my head]. (WHL/g/440,80) b. You've got a great big dan pole. (BCK/v/229,31) c. Me sister's got two and my mate's got one [kitten]. (WHL/s/550,50) 60-i

5040% 3020-

100Buckie

Yoric • Abstract noun

Wheatley Hll

• Concrete Noun

Figure 3. Distribution of have by abstract vs. concrete NP by community.

However, put in cross-variety perspective, an interesting patterning can be observed across varieties. First, in Buckie, there is barely any effect. Then, in York abstract nouns have much more of a propensity to be

'Every place has a different toll'

543

encoded with have. Finally, in Wheatley Hill the same tendency is even more pronounced. The dialects differ in terms of how strong this effect is. In Buckie where there is least have got (52%), there is no effect. In York, which has a little more have got (57%) there is a slight effect. In Wheatley Hill, which has the most have got (75%), the effect is visible, and this is where it is statistically significant, as we shall see below. This appears to correspond with the varying rates of have got overall. It is only where the proportion of use of have got is greater (and have lesser) (see Table 1) that the constraint operates with any degree of strength. 3.5. Generic/non-generic Consider another conditioning factor that has been observed in the literature. Jespersen (1961: 51) also noted that in certain contexts have tends to appear for general as opposed to specific reference. In coding this feature we treated all generic readings in one category and all non-generic readings in another. Once again I apply the comparative method and put the data to the test. 60-, 5040% 3020-

100Budde • Generic

Yak INivgeneric

WteatleyHll

Figure 4. Distribution of have by generic vs. non-generic reading by community.

In each variety, generic readings have more of a propensity to occur with have than non-generic readings. This means that contexts such as in (17) are more likely to be lexicalized as have, than those in (18).

544

Sali A. Tagliamonte

Generic: (17) a. A lot people have lots of fixed ideas. (YRK/2/263,65) b. Parents have a lot more money. (YRK/1/451,86) c. You still have your problems even when you 're not married. (WHL/n/153,19) Non-generic: (18) a. She's got really blonde hair. (BCK/p/423,45) b. I've still got all my lego. (YRK/N/35,28) c. Sheila's uncle 's got the small holding. (WHL/t/617,63) Again, however, the different patterning that can be observed from this cross-variety comparative perspective suggests that this effect may be developmental. The same increasing propensity for differentiation between have and have got can be observed in direct proportion to increasing use of have got. The variety which uses the most have got shows the greatest effect: Wheatley Hill. 3.6. NP vi. pronoun While coding this material it also became apparent that the effect of generic/non-generic could also be linked to the distinction between full NPs and pronouns. As observable in (17) and (18) above, generic stative possessive meaning tends to collocate with full NP subjects while non-generic stative possessive meaning tends to collocate with singular personal pronoun subjects. This is confirmed in Figure 5 which shows the distribution of have by type of subject by community. Full NPs, as in (19), are more likely to occur with have. Full NP: (19) a. Each module has either an exam or some essays. (YRK/Z/457,23) b. Every woman has a little secret. (WHL/a/524,84) c. Our Nicholas has a little plastic pinny. (WHL/Z/464,23) d. Maurice has a coat of arms. (YRK/b/241,49) Once again the pattern across varieties is similar in that the use of have and have got in Buckie appears undifferentiated, while in York and Wheatley Hill, the constraints appear to strengthen incrementally.

'Every place has a different toll'

545

60 •

50 40 % 30 • 20 • 10 •

0· Buckie

York D

full NP

WheatleyHill

• pronoun

Figure 5. Distribution of have by type of subject by community.

In sum, the predictors of have/got variability extrapolated from the literature - syntactic and semantic - play a part in constraining have/got variation in these data. However, although their effect is relatively consistent across all communities, their magnitude varies. This calls for an analysis which can determine whether or not these factors are statistically significant and which can assess their relative strength when they are all considered simultaneously. Statistical methods can estimate the true effects of these factors as well as to remove artifacts of poor data distribution, correlated factors, or statistical fluctuation (Sankoff and Labov 1979; Sankoff and Rousseau 1980). Moreover, as Labov (1982: 76) suggests, such an analysis can enable the analyst to tap into the "ordered set of shifts" in the frequency of forms involved in the progression of linguistic change. Extrapolating from this, it may be the case that the incremental shifts in frequencies of have in the three communities can be taken to represent the possible trajectory of change which is manifest differentially due to varying external conditions operating in each locality. If so, then environmental constraints on this variation and the constraint ranking of those constraints can provide critical diagnostics for tapping in to how this change is developing. Similarities and differences in the significance, strength, and ordering of constraints provide a microscopic view of the grammar from which we can infer the structure (and possible interaction) of different grammars through multidimensional space.

546

Sali A. Tagliamonte

4. Multivariate analysis of have/got variation in cross-variety perspective The next section details the results of a multivariate analysis in which the internal factors (type of subject polarity, type of object and type of reference) are considered simultaneously with the external factors (age and sex).8 Table 2 displays the results of a variable rule analysis of the factors contributing to the probability that stative possessive meaning will be marked by have in Buckie, York and Wheatley Hill. A number of different lines of evidence can be inferred from the table (for further discussion see Tagliamonte 2001). First, which linguistic factors are statistically significant and which are not? This is calculated by the multiple regression procedure incorporated in the variable rule program. Higher numbers can be interpreted as favouring have whereas lower ones disfavour it. The higher the figure, the greater the contribution of that factor to have, as determined by the regression procedure. The factors for which the numbers are enclosed in square brackets were considered in the analysis, but found not to be significant. Second, what is the relative contribution of each linguistic factor selected? In other words which linguistic factor is the most significant and which is the least? This is assessed by the range of the factor weights for each effect. Perhaps most important is the order (from more to less) of categories within each of the linguistic factors, i.e. the constraint hierarchy. Finally, it is critical to determine whether this order reflects the direction predicted by one or the other of the hypotheses being tested. In essence, the analysis provides a model of the grammar of the variation between have and have got in each community. Overall, there is remarkable parallelism in terms of the constraint hierarchies, or direction of effect for each of the internal factors. Looking across the rows for each factor considered in the analysis, the ranking of more to less is the same in each variety. However, consistent with the distributional analysis they are very much differentiated in terms of precisely how the constraints manifest - how strong they are and whether they are statistically significant. In Buckie where have and have got are near equal in frequency, none of the internal constraints are statistically significant (the factor weights hover near or around .50). In contrast, in York and Wheatley Hill where have and have got compete vigorously with each other for the same function, some, or all of the internal constraints approach statistical significance. Moreover, it is Wheatley Hill, rather than York that

Every place has a different toll

547

exhibits a stronger partitioning of these constraints and where their effects approach or exceed the strength of external factors. The question is why? Table 2. Three independent variable rule analyses of the contribution of internal and external factors to the probability of possessive have in three dialects in Britaina)

Corrected mean: Total Ν Factors considered: Age 65+ 36-64 35). I suggest, therefore, that the shifting patterns of variability across these communities reflect the progression of change from have to have got in British English. Where once have and have got competed for stative possessive meaning, have got is clearly winning out. Notice however that in Wheatley Hill the effect of generation is much attenuated. There is a relatively even distribution of have across generations. If the rates of have for the youngest speakers in each community are compared, however, note that there is little difference in the frequency of have across the communities. This means that although have got may well be gradually taking over to mark stative possessive meaning, it appears that have decreases to a certain threshold and then levels out. Why would this be so? 100 -

90 \ 80 70 60 % 50 40 30 20 10

0 Buckie

York • 65+

Ξ 36-64

Wheatley Hill S>35

Figure 6. Distribution of possessive have by generation across communities

One of the principles of grammaticalization outlined by Hopper (1991) is specialization, i.e. a gradual narrowing of choices to the point where one form takes over as obligatory. Here, instead, we observe that the older

Every place has a different toll

549

variant have is holding its own. I have already pointed out the main internal constraints governing possessive have in these dialects and suggested that genericness and the type of NP is implicated. Therefore, I now examine the varieties again, this time showing the distribution of have according to these intersecting characteristics of the subject NP. In Buckie the frequencies of have across these varying subject types hover near 45%. Thus, neither genericness of the subject nor the type of NP plays a role in the choice between have and have got. In York and Wheatley Hill however, there is a visible tendency towards the use of have in certain types of NPs. In York, full NPs that are generic have a far greater propensity for have than any other context (-80%). In Wheatley Hill the favouring effect is found for all generic subjects whether full NPs or pronouns, suggesting a generalizing trend. In both varieties these tendencies are statistically significant (see Table 2).

Buckie • NP Generic

York CI NP non-generic

H Pro Generic

Wheatley Hill • Pro non-generic

Figure 7. Distribution of have by type of subject across communities

Thus, I also suggest that have got is not entirely ousting have. Instead have is gradually becoming specialized for a much more highly circumscribed function within the general domain of Stative possessive meaning. That meaning is first, and foremost, genericness. This can be observed in York, as encoded by NPs, but also generic subjects more generally, as can be observed in Wheatley Hill, perhaps as a later stage in the process. Indeed, only in Wheatley Hill does the strength of the type of reference

550

Sali A.

Tagliamonte

constraint (range = 37) outweigh the effect of age (range = 34). Thus, it appears that as have specializes in function the internal factors begin to outweigh the strength of the external factors like age and sex (see Table 2). This has the effect of making sentences such as in (20) all nondescript in these dialects despite the fact that possessive have may be somewhat oldfashioned in stative possessive contexts more generally. (20) a. Every place has a different toll... (BCK/Z/43,36) b. Most people have dreams. (WHL/k/1423,67) c. Each sort of region has a different sort of accent. (BCK/m/163,0) 5. Conclusion What explanation can be offered for these results? Although the varieties differ in their selection of individual variants, on each measure, the consistency of the direction of these effects confirms that the variability between have and the periphrastic construction with got is basically the same across the three communities. These are the very effects that have been attested in connection with the variation and change from have to have got in the history of the English language and thus can be interpreted as persistence. Their consistent patterning at the intra and inter-community level provides some corroboration of Kroch's (1989) constant rate hypothesis which holds that grammatical constraints will hold constant over time despite the fact that a form or construction may be taking over from another. Moreover, these results show that the same trends are strongly implicated in contemporary dialect data. However, instead of ongoing specialization of function in which one form would be expected to take over the entire functional domain, here we have observed co-existence with concomitant divergence. This mutation in functional distribution, however, is not intuitively obvious, though it becomes visible in the nature of the operation of the constraints and their relative strength across these three communities. As the two forms coexist, they are developing their own individual collocation patterns (Hopper and Traugott 1993: 120). More broadly, selection of a periphrastic construction over and above an earlier synthetic form is consistent with the findings of Mair (1997) and Mair and Hundt (1995) who have consistently demonstrated that English is tending toward greater use of periphrasis. The patterns we have observed here fit in with these pervasive trends in the history of English.

Every place has a different toll

551

Similarly, the cross-variety patterning observed here with have/got is remarkably consistent with findings for grammatical change in the future temporal reference system through the grammatical ization of going to (Poplack and Tagliamonte 2000). That analysis demonstrated that a number of enclave communities in rural Canada tended toward the periphrastic verb form going to. Although all the communities exhibited the same basic trends, the constraints had varying degrees of strength and patterning consistent with the general trajectory of linguistic change in that area of grammar. These changes could be correlated with the relative degree of separation from mainstream varieties and hence participation in ongoing linguistic change. The same tendencies are visible here. The grammatical change from have to have got to got is mirrored in the ordered set of shifts visible in Buckie, York and Wheatley Hill. The varieties which have evolved in peripheral as well as culturally cohesive regions - e.g. Buckie - preserve obsolescent forms and patternings from earlier stages in the history of the change. Those which have evolved in situations with great exposure to mainstream developments, like York, exhibit more extensive layering as well as developments which can be interpreted as more fully advanced along the path of change. Those locales which have been subject to external economic pressure and social reorganization, like Wheatley Hill exhibit further degrees of change both with respect to layering as well as in the operation of evolving grammatical determinants of change. I have interpreted this as the result of linguistic change moving at a much faster rate. Notwithstanding the links between ecological circumstances and internal patterns, it appears that when the development of a form involves specialization, then a consequence will be that grammatical constraints on its distribution will strengthen as the change progresses. In conclusion, the diverse dialects of British English bring a tremendous resource into the controversial arena of tracking the grammatical mechanisms of linguistic change. Consideration of the distributions and patterning of linguistic features across dialects uncovers a wealth of evidence to show that language drift proceeds in the same way, but not at the same rate in all circumstances. Moreover, the cross-variety perspective has given us a critical bird's eye view of how this particular change is progressing, at least in the British English dialects studied here. Finally, I hope to have demonstrated the promising new research potential that exists for tracking grammatical change in synchronic dialect data, particularly when viewed from a cross-variety perspective.

552

Sali A. Tagliamonte

Notes *

1.

2.

3. 4. 5.

6. 7. 8.

With thanks to the Social Science and Economic Research Council of the United Kingdom (ESRC) for projects R000221842 and R000238287 which made this research possible. Codes in parenthesis refer to the data set, York = Y R K , Buckie = BCK, Wheatley Hill = WHL, the individual speaker codes from each corpus and the co-ordinates of the particular datum in the transcripts. Crowell cites the Oxford English Dictionary (OED) under get citing passages from Shakespeare. Jespersen (1961: 48) gives quotations from Marlowe and Dekker. Bartlett (1949: 281) reports that have got is used by authors "from Shakespeare through Dr. Johnson to Augustine Birrell" as well as many other novelists. According to Rice (1932: 283) even earlier 1516. Bartlett's Dictionary of Americanisms (cited in Crowell 1959: 280). Indeed, "the loss of an unstressed syllable from the middle of a polysyllabic word appears to be one of the most conspicuous trends of colloquial English in the sixteenth and seventeenth centuries" (Crowell 1959: 281). Another possibility for negation is a negative determiner or pronoun, as in, e.g. I have no headache. There were only 3 instance of this in these data, all from young males in York. Separate runs (not shown) with internal factors and external factors treated separately did not affect the constraint ranking.

References Biber, Douglas, Stig Johansson, Geoffrey Leech, Susan Conrad and Edward Finegan 1999 Longman Grammar of Spoken and Written English. Harlow: Longman. Cobbett, William 1983 A Grammar of the English Language, in a Series of Letters. Amsterdam: Rodopi. [1818] Crowell, Thomas L. 1959 'Have got', a pattern preserver. American Speech 34: 280-186. Gwynne, Parry 1855 A Word to the Wise, or: Hints on the Current Improprieties of Expression in Writing and Speaking. London: Griffith and Farran. Hopper, Paul J. 1991 On some principles of grammaticization. In: Elizabeth Closs Traugott and Bernd Heine (eds.), Approaches to Grammaticalization, Volume 1, Focus on Theoretical and Methodological Issues, 17-35. Amsterdam/ Philadelphia: John Benjamins. Hopper, Paul J. and Elizabeth Closs Traugott 1993 Grammaticalization. Cambridge: Cambridge University Press. Jespersen, Otto 1961 A Modern English Grammar on Historical Principles, Part II, Syntax. London: George Allen and Unwin Ltd.

Every place has a different toll

553

Kroch, Anthony S. 1989 Reflexes of grammar in patterns of language change. Language Variation and Change 1: 199-244. Labov, William 1970 The study of language in its social context. Studium Generale 23: 30-87. 1971 Some principles of linguistic methodology. Language in Society 1: 97-120. 1982 Building on empirical foundations. In: Winfred P. Lehmann and Yakov Malkiel (eds.), Building on Empirical Foundations, 79-92. Amsterdam/ Philadelphia: John Benjamins. Mair, Christian 1997 The spread of the going-to-future in written English: A corpus-based investigation into language change in progress. In: Raymond Hickey and Stanislaw Puppel (eds.), Language History and Linguistic Modelling, 15371543. Berlin: Mouton de Gruyter. Mair, Christian and Marianne Hundt 1995 Why is the progressive becoming more frequent in English? Zeitschrift für Anglistik und Amerikanistik 2: 112-122. Murray, Lindley 1968 English Grammar. Menston, England: Scolar Press. [1795] Poplack, Shana and Sali A. Tagliamonte 1991 There's no tense like the present: Verbal -s inflection in Early Black English. In: Guy Bailey, Natalie Maynor and Patricia Cukor-Avila (eds.), The Emergence of Black English: Texts and Commentary, 275-324. Amsterdam/ Philadelphia: Benjamins. 2000 The grammaticization of going to in (African American) English. Language Variation and Change 11:315-342. Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech and Jan Svartvik 1985 A Comprehensive Grammar of the English Language. New York: Longman. Rand, David and David Sankoff 1990 GoldVarb: A variable rule application for the Macintosh. Montreal, Canada: Centre de recherches mathematiques, Universitd de Montröal. Version 2. Rice, Wallace 1932 Get and got. American Speech 7: 280-296. Sankoff, David and William Labov 1979 On the uses of variable rules. Language in Society 8: 189-222. Sankoff, David and Pascale Rousseau 1979 Categorical contexts and variable rules. In: Sven Jacobson (ed.), Papers from the Scandinavian Symposium on Syntactic Variation, Stockholm, May 18-19, 1979, 7-22. Stockholm: Almqvist and Wiksell. Smith, Jennifer 2001 Synchrony and diachrony in the evolution of English: Evidence from Scotland. Ph.D. dissertation, University of York. Smith, Jennifer and Sali A. Tagliamonte 1998 'We were all thegither ... I think we was all thegither'·. Was regularization in Buckie English. World Englishes 17: 105-126. Tagliamonte, Sali A. 1998 Was/were variation across the generations: View from the city of York. Language Variation and Change 10: 153-191.

554

Sali A.

2000-2001

Tagliamonte

Vernacular Roots: A datebase of British dialects. Arts and Humanities Research Board of the United Kingdom (AHRB) Research Grant. Reference 041R00383. 2002 Comparative sociolinguistics. In: Jack Chambers, Peter Trudgill and Natalie Schilling-Estes (eds.), The Handbook of Language Variation and Change, 729-763. Maiden/Oxford: Blackwell. Trudgill, Peter, Elizabeth Gordon, Gillian Lewis and Margaret Maclagan 2000 Determinism in new-dialect formation and the genesis of New Zealand English. Journal of Linguistics 36: 299-318. Visser, Fredericus T. 1963-73 An Historical Syntax of the English Language. Leiden: E.J. Brill. White, Richard Grant 1927 Words and Their Uses, Past and Present: A Study of the English Language. Boston: Houghton. [1870] Wooley, Edwin C. 1907 Handbook of Composition. Boston: D. C. Heath. Wyld, Henry Cecil 1936 A History of Modern Colloquial English. Oxford: Basil Blackwell.

Author index Abraham, Werner, 446, 447, 469, 480 Agren, John, 40 Aissen, Judith, 403 Aitchison, Jean, 13-15, 46, 94, 96 Allen, Cynthia, 402, 404 Allen, George D., 76, 77, 88, 109 Allerton, David J., 323 Altenberg, Bengt, 384, 385, 400, 414, 427, 435 Anderwald, Lieselotte, 4, 480, 507, 521, 522 Anschutz, Arlea, 385 Anttila, Arto, 106, 107 Anttila, Raimo, 373 Archangeli, Diana, 104, 281 Ariel, Mira, 217, 228, 232, 233 Arnold, Jennifer E., 2, 129, 148, 151, 159, 217, 271, 295, 298, 379, 402 Aston, Guy, 371, 527 Bailey, Guy, 94, 96 Baker, Carl L„ 482, 498 Baldinger, Kurt, 373 Barber, Charles, 19, 32, 80, 398 Barber, Nicholas, 80 Barlow, Michael, 425 Bauer, Laurie, 89 Bayer, Joseph, 480 Bayley, Robert, 194 Beedham, Christopher, 373 Behaghel, Otto, 120, 128, 131, 132, 139, 184,292 Bell, Alan, 76, 77 Benincä, Paola, 480 Bennett, Paul Α., 470 Benson, Larry D., 459 Berg, Thomas, 71, 72, 74-78, 80, 102, 103, 105, 106, 109, 111,299 Bernini, Giuliano, 509, 527 Bertinetto, Pier Marco, 88 Bever, Thomas G., 94, 96 Bhat, D.N. Shankara, 188 Biber, Douglas, 29, 32, 240, 244, 335, 383-386,388, 527, 531,539 Black, James R., 480 Blake, Barry, 198 Blumstein, Sheila E., 77

Bock, J. Kathryn, 129, 158, 286, 299, 379, 387, 400, 405 Boersma, Paul, 105, 107 Bohem-Jemigan, Heather, 137 Boland, Julie E„ 137 Bolinger, Dwight L., 89, 94, 97, 100, 110, 236, 244, 260, 281, 285, 287, 294, 295, 299, 306, 348, 361, 383, 457 Bolkestein, A. Machtelt, 166, 167 Borkin, Ann, 353, 365, 366 Boy land, Joyce Tang, 35 Braun, Albert, 260, 277, 286, 289 Bresnan, Joan, 403 Brinton, Laurel, 330, 460, 461, 474 Broschart, Jürgen, 188 Brown, Cheryl, 427 Brugmann, Karl, 236, 244, 299 Bumard, Lou, 371, 527 Burnley, David, 80 Bush, Nathan, 33-39, 46, 54, 60 Busse, Ulrich, 109 Bybee, Joan, 7, 9, 10, 13-16, 18, 23, 29, 32-35, 39-42, 44, 47, 48, 54-56, 58-60, 76-78, 82, 260, 291 Callaway, Morgan, 452, 453 Campbell, Lyle, 10, 20, 59, 446, 470, 471 Carpenter, Patricia Α., 179 Carstairs, Andrew, 404 Cassidy, Frederic G., 9, 110 Chafe, Wallace, 128, 417 Chambers, Jack K., 10, 507, 508 Chen, Ping, 417 Cho, Young-mee Yu, 106, 107 Chomsky, Noam Α., 69-72, 83, 93, 101, 102, 104, 108-110, 121, 125, 148, 149, 181, 193, 405, 419, 450, 496 Christian, Donna, 485, 492 Christofaro, Sonia, 229, 231 Clark, Herbert H., 147 Closs Traugott, Elizabeth, 17, 28, 48, 472-474, 550 Cobbett, William, 535 Comrie, Bernard, 489 Conrad, Susan, 335 Cooper, William E., 71

556

Author index

Corbett, Greville, 10 Couper-Kuhlen, Elizabeth, 71, 77, 88, 299 Crain, Stephen, 146 Croft, William, 15, 21, 59, 217, 404, 472, 473, 503, 509, 510, 512, 522 Crowell, Thomas L., 532, 533, 552 Culpeper, Jonathan, 259, 262, 277, 286, 299 Curme, George Oliver, 294 Cuyckens, Hubert, 372 Dahl, Liisa, 398 Dahl, Östen, 388, 508 Dauses, August, 368, 369, 371, 373 Davenport, Mike, 281 Davison, Alice, 372, 373 Deane, Paul, 414 Declerck, Renaat, 361, 372 Dell, Gary S., 74, 102, 103, 147, 152 Denison, David, 399, 470 Dik, Simon C., 348 Dingare, Shipra, 403 Dixon, Robert M.W., 1, 333, 334 Dobson, Eric J., 80, 109 Dotter, Franz, 404 Dowty, David R., 183 Dressier, Wolfgang U., 236, 385 Dryer, Matthew S., 180, 489, 508 Duffley, Patrick J., 323, 334, 361 Ellegärd, Alvar, 31, 32, 60 Ellis, Rod, 58 Elsness, Johan, 243, 244 Fanego, Teresa, 206, 236, 306-308, 315, 318, 324, 342 Farmer, Ann K., 486 Farr, James Marion, 488, 491 Fenk-Oczlon, Gertraud, 387 Ferreira, Victor S., 147 Fidelholtz, James, 14 Fijn van Draat, Pieter, 89, 110 Fillmore, Charles J., 186, 187, 293, 347, 349 Filppula, Markku, 497 Fischer, Olga C.M., 4, 306, 330, 342, 372, 404, 405, 446, 450, 452, 454, 457, 461,470, 473,474 Fodor, Janet D., 146 Fodor, Jerry Α., 405 Fowler, Carol, 14

Francis, Brian, 385, 402 Francis, Gill, 371 Francis, W. Nelson, 110 Franz, Wilhelm, 89, 110 Fräser, Bruce, 162 Fraurud, Kari, 388 Frazier, Lyn, 145, 146 Fries, Peter H., 299 Gaaf, Willem van der, (see van der Gaaf, Willem) Gabelentz, Georg von der, 18 GeniuSiene, Emma, 490 Geurts, Bart, 470, 474 Gil, David, 489 Gimson, Alfred Charles, 83, 109 Ginstrom, Ryan, 151 Givön, Talmy, 1, 17, 22, 157, 184, 217, 232, 285, 288, 291, 293, 370, 414, 439, 464, 473 Goldberg, Adele, 133, 173, 414, 416, 417 Gramley, Stephan, 236 Greenbaum, Sidney, 243 Greenberg, Joseph H., 179, 255, 507, 509, 510, 522 Gregory, Michelle, 43, 44, 55, 60 Gries, Stefan Th., 2, 160, 172, 417, 439 Grimm, Jacob, 8, 503 Grimshaw, Jane, 183 Gronemeyer, Claire, 333 Gundel, Jeannette K., 128, 129, 217, 232 Gussenhoven, Carlos, 58 Guy, Gregory R„ 106, 107, 194 Gwynne, Parry, 534 Haiman, John, 1, 285, 383, 386, 389, 403, 460, 503 Hall, Joan Houston, 110 Halle, Morris, 71, 109 Halliday, Michael A. K„ 348, 364 Harnish, Robert M., 486 Harris, Alice, 446, 471 Haspelmath, Martin, 13, 20, 43, 44, 47, 58, 105, 388, 404, 405, 446, 447, 451, 455, 465, 470, 474, 489, 503 Hawkins, John Α., 2, 19, 120, 121, 132, 145, 148, 151, 157-159, 166-171, 176-179, 181-193, 201, 217, 225, 229, 243, 254, 255, 294, 295, 299, 307, 323, 343, 379, 388, 400, 405

Author Index Hawkins, Roger, 414 Hawkins, Sarah, 77 Hayes, Bruce, 71 Hays, David G., 188 Heine, Bernd, 389, 448 Herbst, Thomas, 265, 299 Herschensohn, Julia, 58 Hickey, Raymond, 281 Hinderling, Robert, 236 Hofland, Knut, 10 Holmqvist, Erik, 19 Hooper, Joan (see Bybee, Joan) Hopper, Paul J., 7, 17, 18, 28, 29, 48, 60, 219, 415, 471, 473, 474, 541, 548, 550 Housum, Jonathan, 14 Huddleston, Rodney D., 243, 265, 266, 323,381,404 Hudson, Richard Α., 108, 419, 527 Huelva-Unternbäumen, Enrique, 373 Hundt, Marianne, 337, 384-386, 397, 399, 550 Hunston, Susan, 371 Ineichen, Gustav, 373 Ingram, David, 77 Irwin, David E., 129, 387 Itö, Junko, 105 Jackendoff, Ray, 178, 181, 182 Jacobs, Haike, 58 Jahr Sorheim, Mette-Catherine, 384-386, 398 Janda, Richard D., 404, 467, 469, 472 Jespersen, Otto, 14, 88, 253, 262, 278280, 286, 290, 293, 299, 330, 333, 360, 398, 404, 414, 419, 451, 454, 532-534, 538, 540, 541, 543, 552 Johansson, Stig, 10 Johnson, David, 405 Johnson, Mark, 422 Jones, Daniel, 27, 280 Jordan, Richard, 80 Jucker, Andreas H., 384-386, 404, 414 Jurafsky, Daniel, 14, 37, 43, 44, 46, 55, 60 Just, Marcel Adam, 179 Kager, Ren6, 71, 88, 105, 110, 111 Kahrel, Peter, 508 Kaisse, Ellen M., 71 Kaltenböck, Gunther, 267

557

Kay, Paul, 347, 349 Keenan, Edward L„ 183, 187, 193, 488 Keller, Rudi, 471, 472 Kelly, Michael H., 299, 387 Kemmer, Suzanne, 495 Kemp, William, 213 Kempson, Ruth M., 419 Kjellmer, Göran, 236, 243 Klavans, Judith, 130 Klemola, Juhani, 480 Koester, John, 72, 73 König, Ekkehard, 265, 283, 293, 482, 489, 498, 502, 503 Koptjevskaja-Tamm, Maria, 385, 389 Kortmann, Bernd, 480, 507 Kroch, Anthony S., 535, 538, 541, 542, 550 Krug, Manfred, 1, 10, 13, 23, 26, 28, 29, 32, 43, 47, 48, 54, 60, 405 KuCera, Henry, 110 Kuno, Susumo, 188, 365, 499 Kurylowicz, Jerzy, 277 Kytö, Merja, 19, 259, 473 Labov, William, 13,14, 18, 27, 394, 535, 545 Ladefoged, Peter, 59, 109, 110 Lakoff, George, 422 Lamb, Sidney Μ., 72, 74, 75, 102, 109 Lambrecht, Knud, 417 Langacker, Ronald W„ 1, 15, 347, 349, 363-367, 369, 372, 415, 423-425, 427, 429, 439 Langendoen, D. Terence, 94, 96, 99 Lappin, Shalom, 405 Lass, Roger, 110, 373 Leech, Geoffrey, 243, 259, 262, 277, 286, 299, 385, 402 Legenhausen, Lienhard, 265 Lehmann, Christian, 27, 451, 465, 470, 472, 473 Lengert, Joachim, 373 Levelt, Willem J.M., 189, 387, 405 Levin, Beth, 149, 417 Liberman, Mark, 71 Lightfoot, David W„ 449, 450, 470 Lind, Age, 236, 240 Lindquist, Hans, 266, 267, 286, 299 Lohse, Barbara, 132, 194 Los, Bettelou, 452, 456-458

558

Author index

Losongco, Anthony, 151 Lüdtke, Jens, 215 Lutz, Angelika, 80, 83, 84 Macauley, G. C„ 454, 458 MacDonald, Maryellen C., 133, 146, 148, 183 MacKay, Donald G., 72, 74-77, 102, 109, 299 MacWhinney, Brian, 236 Maddieson, Ian, 20, 59 Mair, Christian, 3, 206, 236, 273, 307, 337, 366, 550 Manabe, Kazumi, 458 Mann, William C., 366 Manning, Christopher, 58, 403 Markus, Manfred, 33, 89, 291, 299 Martin, James G., 76, 77, 88 Matthiessen, Christian Μ. I. M., 366 Mayerthaler, Willi, 524 McCalla, Kim I., 281, 282 McCarthy, John J., 72, 104 McCarthy, Michael, 46 McCawley, James D., 236 McClelland, James L., 74, 102, 103 McDonald, Janet L., 387 McMahon, April M.S., 404 Menn, Lise, 77, 236 Mester, R. Armin, 105 Miestamo, Matti, 524, 527 Miller, Thomas, 461, 462 Milroy, Jame, 471, 472 Mindt, Dieter, 335 Minkova, Donka, 80, 110 Mitchell, Bruce, 452 Mondorf, Britta, 2, 236, 243, 253, 257, 284, 298, 299,315, 379 Montgomery, Michael, 233 Moravcsik, Edith Α., 198 Motapanyane, Virginia, 480 Murray, Lindley, 535 Muysken, Pieter, 503 Nagy, Naomi, 107 Nakamura, Fujio, 60 Nänny, Max, 404 Nespor, Marina, 88, 110 Nevalainen, Terttu, 19 Newmeyer, Frederick J., 446 Nichols, Johanna, 388, 389 Nikiforidou, Kiki, 421-^123

Nogl, Dirk, 3, 349, 364-366, 369, 373 Ntlbling, Damaris, 8 Nunberg, Geoffrey, 131 O'Seaghdha, Padraig G., 74 Ogura, Mieko, 18-20, 60 Ohala, John, 467 Ortmann, Albert, 388 Orton, Harold, 513, 527 Osselton, Noel E., 414, 438 Paccia-Cooper, Jeanne, 71 Palmer, Frank R., 361 Paul, Hermann, 8, 9 Payne, John, 508 Penning, Gerhard E., 491 Peters, Pam, 272, 273 Pharies, David, 373 Pierrehumbert, Janet, 14, 44, 58 Pike, Kenneth Lee, 88 Plag, Ingo, 105, 110, 236, 299 Plank, Frans, 236, 405, 468, 469, 472, 473, 503 Pollard, Carl, 178, 182, 183, 193 Pope, John C., 462 Poplack, Shana, 538, 551 Postal, Paul M., 307 Poussa, Patricia, 213 Poutsma, Hendrik, 253, 298 Previte, Joseph, 72, 73 Prince, Alan S., 71,72, 104 Prince, Ellen F., 128, 365 Pullum, Geoffrey K., 70, 100, 236, 265, 266, 343 Quirk, Randolph, 10, 28, 60, 89, 120, 193, 194, 243, 257, 259, 260, 262, 265-267, 286, 299, 329, 333, 334, 381, 384, 385, 388, 404, 414, 527, 531, 533, 534, 538, 539 Raab-Fischer, Roswitha, 385, 399 Radford, Andrew, 195 Raffelsiefen, Renate, 105, 110 Ramat, Paolo, 509, 527 Rand, David, 536 Rauh, Gisa, 206 Reich, Peter Α., 74, 102, 103 Reinhart, Tanya, 496, 498 Renouf, Antoinette, 11-13, 58 Reppen, Randi, 335 Rericha, Vaclav, 333, 334, 340 Reuland, Eric, 496, 498

Author Index Reynolds, Bill, 107 Rice, Wallace, 534, 535, 552 Ringler, Richard N., 9 Risselada, Rodie, 166, 167 Roberts, Ian, 446,447, 449,465,470 Rohdenburg, Günter, 2, 80, 89,109, 111, 175, 176, 184, 186, 190-192, 195198, 206, 217, 219, 223, 226-228, 236, 243, 244, 252, 278, 281, 294, 295, 299, 315, 318, 321, 323, 338, 343, 370, 373, 379, 400, 404 Rohr, Anny, 266, 275, 280, 282, 286, 287, 292-294, 299 Romaine, Suzanne, 259 Ronneberger-Sibold, Elke, 405 Rosenbach, Anette, 3, 253, 288, 380, 399, 402, 403, 405, 435, 436, 440 Rosenbaum, Peter S., 195 Ross, John R„ 229, 236, 293, 498 Rousseau, Pascale, 403, 545 Roussou, Anna, 446, 447,449, 465, 470 Rudanko, Juhani, 206, 236, 306, 323, 324 Rumelhart, David E., 74, 102, 103 Saffran, Jenny R., 34 Sag, Ivan Α., 148, 178, 182, 183, 193 Samuels, Michael, 467 Sankoff, David, 403, 536, 545 Saussure, Ferdinand de, 471 Schade, Ulrich, 72, 74, 102, 109 Schafroth, Elmar, 373 Scheibman, Joanne, 10, 33, 35, 54, 55. 59 Schellinger, Wolfgang, 503 Schendl, Herbert, 109 Schibsbye, Knud, 286 Schilling-Estes, Natalie, 527 Schladt, Mathias, 494 Schlüter, Julia, 2, 80, 89, 109, 111, 206, 217, 236, 243,299,315, 343 Schmid, Hans-Jörg, 334, 335 Schneider, Walter, 405 Schuchardt, Hugo, 13, 14 Schuhmacher, W. Wilfried, 373 Schütze, Carson T„ 58, 150, 172 Schwartz, Myrna F., 307 Seiler, Hans Jakob, 389 Selkirk, Elizabeth O., 71, 88, 110, 111 Seppänen, Aimo, 244

559

Shiffrin, Richard M., 405 Siegelbaum, Steven Α., 72, 73 Siemund, Peter, 4, 480-482, 489, 498500, 502, 503, 507 Siewierska, Anna, 130, 156, 157, 159, 388,419 Silva, Georgette, 265 Silverstein, Michael, 404 Smith, Jennifer, 536 Smith, Norval, 503 Smolensky, Paul, 72, 104, 106 Snedeker, Jesse, 147 Sperber, Deirdre, 405 Stallings, LynneM., 134, 148, 177 Standwell, Graham J.B., 414, 418, 438 Staun, Jorgen, 281 Stefanowitsch, Anatol, 3, 385, 422 Stein, Dieter, 19, 253, 380, 399, 508 Stemberger, Joseph Paul, 236 Steriade, Donca, 80 Stiebeis, Barbara, 385 Stockwell, Robert P., 80 Strang, Barbara M.H., 414, 419 Svartvik, Jan, 243 Sweet, Henry, 260, 277, 279, 282, 283, 288 Szmrecsanyi, Benedikt, 58 Tabor, Whitney, 472, 474 Tagliamonte, Sali Α., 4, 402, 405, 536, 538, 546, 551 Tanenhaus, Michael K., 146 Tang, Sze-Wing, 236 Taylor, John R., 385, 389, 4 2 1 ^ 2 3 , 427, 440 Tesni£re, Lucien, 188 Thomas, Rüssel, 398 Thompson, Sandra Α., 29, 32, 219, 265, 366 Tieken-Boon van Ostade, I., 60 Tomlin, Russell S., 130, 182 Tottie, Gunnel, 527 Trask, Robert L., 371 Trudgill, Peter, 10, 507, 508, 513, 537 Trueswell, John C., 134, 148 Turley, Jeffrey S„ 495 Vachek, Josef, 80 Vallduvi, Enric, 130 van den Berg, Reni, 508 van der Gaaf, Willem, 460, 461

560

Author index

van der Leek, Frederike C., 470 van der Wurff, Wim, 449 van Ek, Jan Α., 360 van Gelderen, Elly, 503 Vennemann, Theo, 14, 77, 78, 87, 281 Verschueren, Jef, 368, 373 Verspoor, Marjolijn H., 347, 349, 355— 358, 360-364, 369, 371,372 Vezzosi, Letizia, 380, 399, 402 Visser, Fredericus T., 306, 316, 330, 331, 361, 452, 455, 456, 460, 461, 532-534, 538, 539 Traugott, Elizabeth C., 17, 28, 48, 472474, 550 Vogel, Irene, 88, 89, 110 Vosberg, Uwe, 2, 3, 236, 299, 308, 309, 312, 324 Wang, William S-Y., 19, 20, 60 Warner, Anthony R., 307, 449, 470 Warren, Paul, 147 Warren, Richard K., 387 Wasow, Thomas, 2, 120, 121, 131, 132, 134, 147, 148, 151, 177, 181, 183, 217, 271,295, 298, 379 Wasserman, Robert David, 243, 244 Wegener, Heide, 503 Wells, John C„ 27, 58

Westney, Paul, 236 White, Richard Grant, 534 Wierzbicka, Anna, 1, 173, 347-353, 355, 357, 363, 364, 369, 371, 372 Williams, Robert S., 151 Wilson, Dan, 405 Windsor-Lewis, John, 27 Wolfram, Walt, 527 Woods, Anthony, 61 Wooley, Edwin C., 534 Wright, Joseph, 484, 501 Wurff, Wim van der, (see van der Wurff, Wim) Wurzel, Wolfgang U., 217, 524 Wyld, Heniy Cecil, 533 Xu, Xunfeng, 385, 402 Yamamoto, Mutsumi, 385, 388 Yip, Moira, 236, 299 Yngve, Victor H., 471,472 Yule, George, 232 Zachrisson, R. E„ 384, 398 Zipf, George K., 11-13, 16, 18, 23, 28, 44, 45, 58, 59, 260 Zonneveld, Wim, 111 Zribi-Hertz, Anne, 498 Zwicky, Arnold M., 14, 70, 100, 236, 343

Subject index a.c.i. (see raising) accessibility, 129, 130, 147, 217, 228, 232, 233, 252, 387, 388 adjacency, theory of, 2, 175-186 adjacency hypothesis, 176, 184 age, 394, 397, 467, 536, 547-550 ambiguity ambiguity avoidance, 134, 139, 144, 145, 147, 271 syntactic ambiguity, 120, 133, 134— 136, 145-147 global ambiguity, 136 local ambiguity, 137-140 analogy, 7-10, 21, 58, 405, 446, 464, 474, 524 animacy, 156, 165, 334, 379, 380, 385402, 413—415, 418, 425, 426, 428, 436-438, 450, 460, 462, 464 argument complexity, 217, 243, 251-253, 257, 262, 264, 269, 271-273, 294, 298, 299 benefactive dative, 450, 492, 493 British vs. American English, 84—88, 214, 218, 224, 226, 310-318, 336-342, 379, 384, 390-398, 534, 539 (rhythmic), 91, 93, 95, 96, 99, 100, 103, 275, 276, 279, 280

complementation, 2 0 5 , 2 0 6 , 2 6 2 , 3 0 5 - 3 0 7 , 309, 316, 3 2 0 - 3 2 5 , 329, 332, 334, 336-338, 349,350, 371,373 complementation system, 2 0 5 - 2 0 8

sentential complementation, 206, 305, 320, 373

verb complementation, 305, 332, 334, 336, 349

complementation of adjectives, 262 gerundial complementation, 306, 322, 324, 325, 329, 3 3 6 - 3 3 8

infinitival complementation, 306, 316, 324, 329, 3 3 6 - 3 3 8 , 3 4 9

non-finite

complementation,

307,

309-314, 317-322, 329

complexity lexical, 284 morphological, 260, 283 phonological, 84, 282 complexity principle, 2, 3, 175, 205, 216, 217, 222, 225, 233, 242, 252, 294, 297, 299, 321, 370, 373, 400

semantic, 289, 295, 297 syntactic, 2 , 119, 1 2 1 - 1 2 3 , 1 2 5 - 1 2 8 , 155, 156, 161, 1 6 3 - 1 6 7 , 169, 170, 172, 176, 2 0 5 , 2 1 6 , 2 1 7 , 2 1 8 , 2 2 0 , 222, 224, 228, 2 3 1 - 2 3 3 , 2 5 1 - 2 5 5 , 262, 264, 266-275, 284, 286, 2 9 4 297, 305, 306, 3 2 1 , 3 2 2 , 3 8 4

consonant clusters, 69, 78, 81, 87, 105, colonial lag, 310 comparative, synthetic vs. analytic, 2, 251-299 complement, adjectival, 253-255, 265-274, 286, 291,294, 295 infinitival vs. gerundial, 3, 305-324, 330-343 in genitive constructions, 381-384 of begin/start, 3, 329-343 ordering of, 182, 183 prepositional, 206, 253, 265-268, 270, 273, 294, 299 sentential, 3, 175, 189-192, 195-197, 220-225, 229, 321, 349-352, 453

280-283, 256

constituent recognition domain, 1 7 8 - 1 8 0 , 182,187,254

constraints, 1, 2 , 4 , 7, 3 1 - 3 3 , 7 1 , 7 2 , 9 6 , 9 7 , 102, 1 0 4 - 1 0 8 , 119, 1 2 0 , 129, 146, 148, 1 4 9 , 2 0 8 , 2 1 7 , 2 2 7 , 281, 297, 298, 306, 311, 322, 403, 413, 449, 450, 531, 538, 5 4 4 - 5 4 7 , 550, 551 ANIM > INANIM, 3 8 8 • V V , 105 • C C , 105

145, 233, 333, 541,

•CLASH, 1 0 5 , 1 0 7 , 2 9 8 DERIVED OBJECT CONSTRAINT, 3 0 7 DOUBLE -ING CONSTRAINT, 3 1 5 , 3 2 2 •LAPSE, 105

562

Subject Index OCP, 275 SAME VOICE, 17 STAY, 2 9 7

constructional semantics, 4, 413 contractions, 1, 22, 27, 39, 49-51, 89, 181, 517, 533, 538-541 degrammaticalization, 306, 455, 459, 460, 468 derived intransitivity, 481, 490-492 dialects, 150, 507, 508 English dialects, 21, 83, 149, 309, 452, 507-509, 513-522, 531-538, 541-543, 547-551 dialect grammar, 480, 507, 508 dialectology, 4, 480, 502 discontinuous structures, 30, 162, 225227, 232, 234, 285, 321,466 distance principle, 1, 2, 184, 285, 291, 297, 464 double object construction, 133, 137, 139141, 149, 151,416,418, 493 early immediate constituents, 157-159, 167,169-172, 179-181, 185 economy, 7, 8, 11, 19, 78, 251, 297, 298, 379, 380, 399-403 entrenchment, 15, 18, 20, 28, 29, 31-33, 41, 44, 47, 48, 58, 59, 145, 161, 206, 243, 252, 260, 277, 285, 295, 310, 319, 336,415 extractions, 3, 305-315, 323 extraposition, 194, 266, 273, 274 filler-gap dependency, 188, 254, 307 frequency, 1, 2, 7-57, 95, 133, 134, 216, 220, 260-262, 277, 282, 295, 306, 352, 384, 388, 401, 464, 510, 512, 523,545 string frequency, 2, 7, 8, 23-55, 59, 60 transitional probability, 2, 7, 8, 33-57 givenness, 3, 155, 157, 163, 164, 166, 167, 169, 171, 232, 288, 364, 365, 370, 373, 379, 380, 383, 385-388, 390396, 398—404, 413, 414, 425^127, 435, 436, 439, 440

grammaticalization of auxiliaries, 26-29, 38, 349, 460, 461,551 of infinitival to, 306, 451, 452, 454, 455, 458—460 of intensifiers and reflexives, 491, 548 of prepositions, 206 of the indefinite article a, 12 of the nucleus-satellite relation, 366 grammaticalization theory, 4, 47, 405, 448 hiatus, 16, 44, 58, 77, 78, 81 hierarchy of non-adjacency, 227, 228, 231 horror aequi, 2, 3, 205, 216, 236, 238, 240, 242, 243, 278-281, 298, 305, 306, 315, 318-323, 324, 325, 338, 343, 404 iconicity, 1^1, 21, 148, 229, 231, 266, 267, 285, 291-293, 379, 380, 385-390, 398-403, 445-447, 449, 456, 459, 460, 464, 467, 469, 473, 524 constructional iconicity, 404, 524 ideal syllable structure, 71, 72, 76-78, 82, 84, 87, 89, 93, 101, 103, 105, 107, 111

information status (see information structure) information structure, 128-131, 287, 364, 365, 413-418, 425, 426, 435, 438, 439 intensifiers, 4, 131, 479-484, 487-504 isomorphism, 383, 459 language change, 1, 2, 7, 10, 22, 71, 107, 217, 335, 342, 4 4 5 ^ 6 9 and frequency, 8, 10-22, 27, 31, 47, 55, 56, 545 in the complementation system, 206, 215-217, 221, 229, 237, 240, 242, 306, 329, 336-339 morphological, 89, 515, 523 morphosyntactic, 379, 380, 390, 394402, 531-539, 541,547-551 law of growing elements (Gesetz der wachsenden Glieder), 120, 139, 292 layering, 18, 445, 455, 459, 473, 531, 551 lexical bias, 119, 132-134, 142-144, 148

Subject index markedness, 19, 59, 176, 217, 223, 228, 373, 509-512, 522, 524 markedness reversal, 231 more-support, 3, 251-255, 260-264, 267270, 273-277, 280-297 negation, 4, 7, 29-32, 49, 94-100, 103, 108, 197, 321, 323, 369, 457, 480, 507-524, 539-541, 547 (neural) network, 71, 72, 74-77, 102, 103, 106-108, 297 non-standard, 1, 4, 213, 507-509, 512-524 number contrasts, 223-225 obligatory contour principle (see constraints) opaque collocations, 131, 181, 183 optimality theory, 1, 2, 58, 72, 104-108, 281,388, 403 particle movement (see particle placement) particle placement, 121, 123, 125-128, 132, 155, 158, 160-172, 416, 418, 439 persistence, 4, 286, 459, 531, 550 principle of, 541 phonotactic, 19, 76, 78, 82, 83, 101, 103, 105 pleonastic pronouns, 491 principle of rhythmic alternation, 71, 72, 76, 77, 88, 89, 93, 94, 98-101, 103, 105, 107, 108, 275, 279, 287 prosody, 76, 84, 86, 88, 90, 93, 97-99, 103, 105, 145-147,211,267, 288 proximity, principle of (see distance principle) prototype, 389, 420-423, 425, 432, 459 quantity principle, 288, 293 raising, 3, 59, 261, 307, 364, 457, 463 re-analyis, 446, 447, 450, 460, 463, 465, 466, 469, 470, 472, 473 reflexives, 4, 479-496, 498, 500 retrospective verbs, 306 semantic connectedness, 2, 130-132, 148, 295

563

sex, 19, 536, 546, 547, 550 sociolinguistic, 329, 335, 337, 342, 343, 384, 467, 469, 472, 480, 502, 507, 521, 535 sociolinguistics, 4, 480, 502 specialization, 548, 550, 551 stress clash, 69, 88, 91-100, 107, 275, 276, 279, 296 string frequency (see frequency) ίΑαί-clause, 191, 192, 206, 229, 235, 347, 350, 351, 356, 359, 361, 363-373, 453, 457, 458 topicality (see givenness) tough movement, 266, 273, 274 transitional probability (see frequency) typological, 4, 19, 25, 217, 228, 233, 388, 479-481, 483, 490, 507, 509 typology, 180, 479, 480, 502, 507, 508, 509, 510 unidirectionality, principle of, 4, 445, 446, 455, 456, 465, 470, 472 variables cliticization of HAVE and BE, 16-18, 23 cliticization of NOT, 29-33 dative alternation, 119, 121, 123-126, 128, 133, 134, 137, 139, 140, 151, 172 gerundial vs. infinitival complements, 306, 316, 322, 324, 325, 329, 336338, 349 heavy noun phrase shift, 119, 121123, 125-128, 131, 134, 136, 151, 156, 159, 177, 181,463 indefinite article (a vs. an), 83-87 infinitival vs. prepositional clause linkers, 214, 215 infinitival vs. verbal clause linkers, 236, 237, 239 infinitival vs. verbal, zero and prepositional links, 242 negation, 507-524 non-finite complements, 310-314, 317-319, 322 participle alternation (lighted vs. lit), 89-93

564

Subject Index particle placement in verb-particle constructions, 121, 123, 125-128, 132, 155, 158, 160-172, 416, 418, 439 prepositional phrase ordering, 177178 prepositional vs. verbal and zero clause linkers, 218, 219, 239-241 presence or absence of prepositional clause linkers, 207, 214, 216, 221, 224-235 -s genitive vs. o/genitive, 3, 253, 288, 379-402, 405, 413, 418, 420-422, 424—440 stress clash avoidance in Det + not + ADJ + Ν constructions, 94-99 synthetic vs. analytic comparative, 251-299

ίΑαί-clause vs. infinitival clause after believe-type verbs, 366, 367 reflexives and intensifiers, 479-504 verbal ending (-« vs. -0), 78, 79 verbal ending (-s vs. -th) 19, 20, 22 visibility hypothesis, 334 weight, 121, 123, 125, 128-130, 132, 143, 148, 159, 177, 180-182, 184, 191, 217, 266, 384,413 prosodic weight, 84, 86, 267 end-weight, 120, 122, 149, 267, 268, 270-273, 275, 294 zero-marking, 2, 9, 10, 17, 22, 47, 175, 176, 185-200, 205, 212, 218, 226, 230-233, 238, 241, 243, 510, 511, 522, 523