Markedness in synchrony and diachrony 9783110862010, 9783110117806

Table of contents :
Preface
Contributors
Introduction
Markedness Theory – the First 150 Years
Defining Markedness
Markedness and Linguistic Change
Markedness, Sound Change and Linguistic Reconstruction
Markedness and Naturalness
1. Thoughts about markedness and normalcy/naturalness
2. Markedness and naturalness in phonology; the case of natural phonology
Towards a Theory of Semantic Markedness
Internationalisms: Marked or Unmarked
Markedness, Productivity and Naturalness in L2 Learner Lexis
Markedness and Grammaticalization
On the Assessment of the Markedness Status of the exponents of a Grammatical Category
Markedness and the Category of Case in Polish
Inflectional Class Markedness
On the Distinction Marked/Unmarked and Primary/Secondary in a Linguistic Description
Markedness and the Grammar of INFL and Verb in English
The Marked-Unmarked Distinction in the Grammar of the German Ergative Verb
Transitivity and Markedness: the Antipassive in Accusative Languages
Markedness and Clause Structure
On the Markedness of “Narrative Temporal Clauses”
Author Index
Language Index
Subject Index

Citation preview

Markedness in Synchrony and Diachrony

Trends in Linguistics Studies and Monographs 39

Editor

Werner Winter

Mouton de Gruyter Berlin · New York

Markedness in Synchrony and Diachrony

edited by

Olga Miseska Tomic

Mouton de Gruyter Berlin · New York

1989

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

Library of Congress Cataloging-in-Publication Data Markedness in synchrony and diachrony / edited by Olga Miseska Tomic. p. ca. - (Trends in linguistics. Studies and monographs : 39) Bibliography : p. Includes index. ISBN 0-89925-504-3 1. Markedness (Linguistics) I. Tomic. Olga Miseska. II. Series. P299.M35M37 1988 415—dec!9

Deutsche Bibliothek Cataloging-in-Publication Data Markedness in synchrony and diachrony / ed. by Olga Miseska Tomic. - Berlin ; New York : Mouton de Gruyter, 1989 (Trends in linguistics : Studies and monographs ; 39) ISBN 3-11-011780-0 NE: Tomic, Olga Miseska [Hrsg.]; Trends in linguistics / Studies and monographs Printed on acid free paper. © Copyright 1989 by Walter de Gruyter & Co., Berlin. All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced in any form - by photoprint, microfilm, or any other means - nor transmitted, nor translated into a machine language without written permission from Mouton de Gruyter, a Division of Walter de Gruyter & Co., Berlin. Typesetting and Printing: Tutte-Druckerei, Salzweg-Passau. - Binding: Lüderitz & Bauer, Berlin. - Printed in Germany.

Preface The idea about the publication of this title originated at the 19th Annual Meeting of Societas Linguistica Europaea, held in Ohrid, Yugoslavia from August 31 to September 4,1986. Only a limited number of the contributions, however, are (revised and extended) versions of papers presented at that conference (whose main topic was "Markedness in synchrony and diachrony"); a vast majority of the chapters has been specifically designed and written for the present volume. I have endeavoured to put together a book which is representative of current European work on markedness, both with respect to substance and approach. It is truly European in geographic representation, as well; the authors come from nineteen linguistic centers in the East and West, North.and South of "the Old Continent". Since twenty individuals from twelve countries have been involved, the book has taken somewhat longer to prepare than some of us counted it would; my apologies to the authors who kept their deadlines for those who did not. I wish to express deepest gratitude to Werner Winter for his continuous support. Olga Miseska Tomic

Table of Contents Preface Contributors Introduction Olga Miseska Tomic

V IX 1

Markedness Theory - the First 150 Years Henning Andersen

11

Defining Markedness Jadranka Gvozdanovic

47

Markedness and Linguistic Change Dieter Stein

67

Markedness, Sound Change and Linguistic Reconstruction Thomas V. Gamkrelidze

87

Markedness and Naturalness 1. Thoughts about markedness and normalcy/naturalness Werner Winter 2. Markedness and naturalness in phonology; the case of natural phonology Wolfgang U. Dressier

103

Ill

Towards a Theory of Semantic Markedness Ferenc Kiefer

121

Internationalisms: Marked or Unmarked Vladimir Ivir

139

Markedness, Productivity and Naturalness in L2 Learner Lexis Rüdiger Zimmermann

151

Markedness and Grammaticalization Christian Lehmann

175

On the Assessment of the Markedness Status of the exponents of a Grammatical Category Olga Miseska Tomic 191

VIII

Table of Contents

Markedness and the Category of Case in Polish Roman Laskowski

207

Inflectional Class Markedness Wolfgang Ullrich Wurzel

227

On the Distinction Marked/Unmarked and Primary/Secondary in a Linguistic Description Jarmila Panevovä

249

Markedness and the Grammar of INFL and Verb in English Frits Beukema and Bob Rigter

265

The Marked-Unmarked Distinction in the Grammar of the German Ergative Verb Werner Abraham

291

Transitivity and Markedness: the Antipassive in Accusative Languages Gilbert Lazard

309

Markedness and Clause Structure Martin Harris

333

On the Markedness of "Narrative Temporal Clauses" Elizabeth Couper-Kuhlen

359

Author Index

373

Language Index

377

Subject Index

379

Contributors Professor Werner Abraham Vakgroep Duitse Taalen Letterkunde Fakulteit der Letteren Rijksuniversiteit te Groningen The Netherlands Professor Henning Andersen Department of Linguistics State University of New York Buffalo, NY USA Dr. Frits Beukema & Dr. Bob Rigter Vakgroep Engels Fakulteit der Letteren Rijksuniversiteit te Leiden The Netherlands Dr. Elizabeth Couper-Kuhlen Institut für Anglistik Universität Konstanz Federal Republic of Germany Professor Wolfgang U. Dressier Institut für Sprachwissenschaft Universität Wien Austria Professor Thomas V. Gamkrelidze Institut Vostokovedenija imeni Cereteli Akademija nauk GSSR, Tbilisi USSR Dr. Jadranka Gvozdanovic Slavisch Seminarium Universiteit van Amsterdam The Netherlands Professor Martin Harris Vice-Chancellor University of Essex Wivenhoe Park Colchester C04 3SQ Essex Great Britain Professor Vladimir Ivir Odsjek za anglistiku Filozofski fakultet

X

Contributors

Sveuciliste u Zagrebu Yugoslavia Professor Ferenc Kiefer Nyelvtudomanyi Intezete A Magyar Tudomanyos Akademia, Budapest Hungary Professor Roman Laskowski Institutionen för Slaviska Spräk Göteborgs Universitet Sweden Professor Gilbert Lazard Academie des Inscriptions et Belles-Lettres Institut de France, Paris France Professor Christian Lehmann Fakultät für Linguistik und Literaturwissenschaft Universität Bielefeld Federal Republic of Germany Dr. Jarmila Panevovä Matematickä lingvistika Matematicko-fizikalni fakulta Karlovä universita, Praha Czechoslovakia Professor Dieter Stein Institut für Anglistik und Amerikanistik Justus-Liebig-Universität, Glessen Federal Republic of Germany Professor Olga Miseska Tomic Katedra za engleski jezik i knjizevnost Institut za strane jezike Filozofski fakultet Univerzitet u Novom Sadu Yugoslavia Professor Werner Winter Institut für Sprachwissenschaft Universität Kiel Federal Republic of Germany Professor Wolfgang Ullrich Wurzel Akademie der Wissenschaften der DDR Berlin German Democratic Republic Professor Rüdiger Zimmermann Institut für Englische und Amerikanische Philologie Philipps-Universität, Marburg Federal Republic of Germany

Introduction Olga Miseska Tomic

1. The notion of markedness

"Markedness" has recently received considerable attention in linguistic literature. Attitudes towards it, however, vary from its recognition as one of the most important conceptual achievements of contemporary linguistic thought to its rejection as a vacuous makeshift. The authors of the present volume 1 seem to agree that this divergence of attitudes is due to changes in the understanding of the notion, as it was being carried over from one metatheoretical paradigm to another. Underlying "markedness" is undoubtedly the notion of "opposition". But the "opponents" have never been defined clearly and unambiguously. Initially, Trubetzkoy (1931) distinguished between "mark" and "distinctive property" and thus indicated the possibility of working with two notions of markedness, which could be characterized through the oppositions [ +feature]/[ — feature] and [ + marked]/[ —marked], respectively. His subsequent (Trubetzkoy 1958 [1939]) use of the term "unmarked" as a cover term for both (a) a sign characterized by its form and its oppositive meaning, and (b) a form which shows up in the position of neutralization, where the oppositive sign meaning, and hence the sign value connected with it, is omitted (cf. Gvozdanovic sect. 1), blurred this distinction. In post-Trubetzkoyan literature the oppositions [-f feature]/[—feature] and [+marked]/ [—marked] were conflated and the notion of "markedness" became inseparable from the contradictory relation coded in the terms "marked" and "unmarked". While it was constrained to the domains of phonology and morphology and as long as it was interpretable and interpreted as a distinction between morphonologically simple and morphonologically complex forms, the distinction marked/unmarked was operative. As the notion of markedness was being carried over from one metatheoretical paradigm to another and extended over the entire domain of linguistic analysis, the originally strictly

2

Olga Miseska

Tomic

formal opposition evolved into a set of correlations which, in addition to the formal complexity, involved typological considerations, functional load, frequency of occurrence, learnability and interpretability. In general, it is assumed that the morphonologically simple forms, which are qualified as "unmarked", have heavier functional load and higher frequency of occurrence and are learnt and interpreted relatively easier. There have appeared, however, numerous counterexamples to this assumption, which cast doubt on the justifiability of the markedness hypothesis itself. The contributors to the present volume maintain that the doubts are not due to the markedness relationship per se, but rather to a conflation of oppositions and indiscriminate application in different domains. Andersen suggests that, instead of the all-embracing opposition marked/unmarked, one should adopt Hjelmslev's distinction between the inclusive relationships of participation, where the reference potential of one term includes that of another, and the exclusive (contrary and contradictory) relationships, with which the reference potential of each term excludes that of its opposite. 2 The two types of relations are not mutually exclusive; as a matter of fact, exclusion is merely a special case of participation, in which certain areas of the extensive term are empty so that the exclusive relations can be reduced to inclusive ones, though not vice versa. Gvozdanovic proposes a comparable distinction. Stressing the fact that a definition of markedness which can be assumed to provide a workable basis for linguistic analysis should be based on a asymmetry which manifests itself either syntagmatically or paradigmatically, she distinguishes between syntagmatic and paradigmatic markedness. Syntagmatic markedness is encountered whenever (a) both members of a single correlation can occur within the same context and by doing so one member adds information which is not added by the other or else (b) a set of mutually related correlations can be either present or absent within the same context, and either its presence or its absence adds to the given context the information which in the opposite case is not added. Paradigmatic markedness is encountered when (a) the members of a correlation are mutually asymmetrical so that one of them allows for a further division, whereas the other does not and/or (b) the form of one member of a correlation can be used both with or without its sign value, whereas the form of the other member can be used only with its sign value. The distinctions inclusive/exclusive and syntagmatic/paradigmatic are highly reminiscent of Trubetzkoy's distinction between mark and distinctive feature and may be treated as intraparadigmatic developments, triggered through the use of the term "unmarked" as a cover term for both (a) a sign characterized by its form and its oppositive meaning, and (b) the same form

Introduction

3

which shows up in the position of neutralization, where the oppositive sign meaning, and hence the sign value connected with it, is omitted (cf. Gvozdanovic, sect. 1). The distinction between markedness and grammaticalization, which Lehmann makes, is, in turn, an interparadigmatic distinction, which delimits the scope of the analysed notions.

2. Criteria for assessing the unmarked member of the opposition

Lehmann actually points out that the problems of establishing operational criteria for determining the unmarked member of an opposition are due to the interpretation of grammaticalization relationships as markedness relationships: In a dynamic framework based on language as an activity, markedness should, according to him, be seen as pertaining to the choice of the speaker within a paradigm, whereas grammaticalization perrtains to the choice among different paradigms. Very much in line with this view is Panevova's interpretation of the relationship of markedness as distict from the relationship of representation, holding between the different levels of the Functional Generative Description:3 while the former holds between the members of a single category on the same level, the latter is a relationship between forms and functions of units at different levels. Ivir, in his turn, states that the standard with respect to which markedness is measured varies with the changing perspective and purpose of analysis. In the same vein, Tomic insists that markedness relationships obtain only within a given system, while Beukema-Rigter point out that intra-language markedness relationships should be set off from (inter-language) universal principles. Tomic shows how when systems change, existent markedness relationships are disturbed. Very often these disturbances lead to markedness reversal. Thus changes in grammatical systems may cause changes in markedness relationships. The latter changes should, however, be distinguished from the former. In early Praguian linguistics, when markedness was mainly restricted to phonology, "marking" meant increasing the formal complexity of the linguistic unit. When the distinction "marked/unmarked" was extended over a variety of linguistic domains, the simple correlation unmarkedobstruents, taken together, are more frequent than bdgobstruents, just as in Russian; but in French the ptk-series is marked (the opposition is based on tenseness), whereas in Russian it is unmarked (the opposition is based on voicing). This example suggests an important source of divergences between markedness and frequency. In the historical development of languages, changes may occur through which, for instance, the distinctive properties on which an opposition is based may be reinterpreted. Such "rephonologizations" may entail inverse markedness values, and when this occurs, relative text frequency cannot be expected to correlate with markedness values. Greenberg presents a convincing picture of the interrelations between markedness values and universal diachronic tendencies (63 if.). But it is clear that in some of these, markedness values motivate changes in frequency, and not vice versa. A paradigm case is changes involving stylistic variation. A reinterpretation of the markedness values of stylistic variants can explain changes in their relative frequency. But without such (covert) changes in valuation, the observable changes in frequency are inexplicable. One may conclude that text frequency may be a useful, but not a universally reliable indicator of markedness values. Where it does not conform with markedness values, the independence and the primacy of the values is evident. Where text frequency does correlate directly with markedness values, it seems, there are two possibilities. The markedness values may be language specific, and the relative frequency is the result of a historical development which has adjusted usage to the norms or the system of the language. Alternatively, the markedness values may be universal, as is presumably the case with the basic polarity adjectives (52 f.). In such instances the question of whether markedness values or relative frequency is primary appears to be moot. But here relative frequency is a consequence of a difference in distribution which is the primary manifestation of the different values of the two opposites. It is notable that Lyons, in his exposition of markedness, does not speak of relative text frequency, but of differences in distribution (cf. section 1.2). Differences in distribution are a matter of grammar, which can be correlated with markedness values. But text frequencies reflect a variety of different factors, linguistic as well as extralinguistic. If one defines markedness in terms of text frequency - as Greenberg explicitly wished to do (but in the end shied away from doing), and many linguists have been prone to do since - one merely empties the concept of all meaning other than "relative text frequency", and the term might as well be dispensed with (cf. Lass 1975).

Markedness

theory - the first 150 years

31

4.2 Neutralization and participation Greenberg compares neutralization in phonology with what he calls contextual neutralization in grammar and lexis (58). In environments where a phonological opposition is neutralized, usually, the unmarked term occurs (13 f.). In grammar and lexis, the opposition between two or more categories may be suppressed, and it is then the unmarked member which appears. Greenberg identifies this phenomenon with Hjelmslev's participation, but prefers to call it "contextual neutralization" (28 f.). His only grammatical example of this is the use of the singular form of nouns in phrases with cardinal numerals in Hungarian, Turkish, and other languages (which will be mentioned again in section 4.5), but in the discussion of lexis he gives a variety of other examples (52f.). Greenberg's exposition of phonological neutralization is couched in terms of the simplified, received understanding, which was mentioned in section 3.3 and reflects the pre-1938 conception of segments as the ultimate phonological components. Viewed in these terms, which do not disturb the picture here, the distribution of phonetic segments that results in cases of neutralization is parallel to the reference values of antonymic pairs: we have, say, [b] : [p] where a voicing opposition is distinctive (corresponding to bitch 'female dog' : dog 'male dog', when used contrastively) and phonetic [p] where the opposition is suspended (corresponding to dog 'dog, irrespective of sex'). One can even elaborate the parallelism with further distinctions, in positions of neutralization, between [p] as the outcome of irresolvable neutralization (cf. "generic dog"), [p] representing underlying jbj ('female dog'), and [p] representing /p/ ('male dog') {cf. the examples in section 1.3). It appears that the different reference potential associated with markedness values in grammatical and lexical antonyms is entirely isomorphic with the different "realization potential" of (neutralized) phonological opposites. There is an apparent discrepancy, which needs to be specified and explicated. The grammatical and lexical antonyms are, so to say, born with these differences in reference potential; they form genuine inclusive relations. In phonology, where oppositions are functionally exclusive, the similar distributional configuration depends on the existence of a neutralization rule (which deletes phonological specifications) and an additional rule by which the resulting "archiphonemes" are consistently assigned the phonetic properties of the unmarked term of the opposition. This apparent lack of parallelism disappears, however, the moment "reference potential" and "realization potential" are considered from the

32

Henning

Andersen

encoding point of view. From this point of view, contextual neutralization in grammar or lexis means the omission of semantic specifications for which there is no communicative need and, concomitantly, the obligatory selection of the unmarked term of the relevant grammatical or lexical opposition (e.g., when we talk of a dog having puppies). Similarly, neutralization in phonology means omission of a phonological distinction and the obligatory selection of the phonetic properties associated with the unmarked term. The omission of information is in both cases sanctioned or stipulated by the norms of the language. One may have the impression that in grammar and lexis the omission of semantic specifications is optional, whereas in phonology neutralizations are obligatory. In fact, however, this is not a difference in principle, but one of degree. The two parallel neutralization phenomena can be understood as arising in the mapping between levels A and Β and between levels C and D in (10).

4.3 Allophonic and allomorphic variability Greenberg cites C. F. Hockett for the observation that phonemes which are unmarked for a certain feature tend to show greater allophonic variation than their marked counterparts (21). In morphology, as a rule, there is less morphological irregularity in marked than in unmarked categories (29). Greenberg's examples here involve stem alternations and alternations in affixes. This generalization may only be a tendency; yet, it is clearly related to the established fact that marked (derived) categories are more subject to analogical leveling than unmarked ones (69). Although the similarity between these two, phonological and morphological, phenomena is clear enough, they are also clearly different from the phenomena reviewed in section 4.2. Neither allophonic variation nor allomorphy have to do with the omission of information. On the contrary. The difference between the single, invariant morpheme shape as bearer of a certain meaning and the contextually conditioned alternation of two or more allomorphs is that the latter carry more information. They have their proper grammatical or lexical meaning, but in addition indicate features (be they lexical, grammatical, or phonological) of the environments in which they occur, or, in other words, besides their symbolic content they have specific indexical content. Similarly in phonology. When in Nootka the unglottalized stops have both aspirated and unaspirated allophones (21), the different degrees of aspiration are in effect subsidiary indexical signs which serve to specify the phonological surroundings.

Markedness

theory - the first 150 years

33

Interpreted in these terms, the two phenomena at hand can be restated as the observation that unmarked terms of phonological or morphological oppositions are more compatible with subsidiary, indexical signs than are their marked opposites. We will return to this generalization in section 4.4. The two phenomena considered here concern the mapping relations between levels Β and C, respectively levels C and D in (10).

4.4 Unpaired phonemes and syncretism The number of phonemes in a language with a certain marked feature is always less than or equal to the number with the corresponding unmarked feature but not greater. Greenberg's key example is the universal numerical relation between nasal and non-nasal vowels (21). A morphological and lexical counterpart to this, in Greenberg's view, is the greater susceptibility to syncretism of marked than of unmarked categories, as when masculine and feminine gender is distinguished in the singular, but not in the plural, or when cases that are morphologically distinct in the singular are syncretized in the plural (27). This comparison is not felicitous. The phonological side of it has to do with the combination of distinctive features into simultaneous syntagms (or "bundles") and amounts to an abservation that a simultaneous syntagm of distinctive features with a given marked term will not be expanded with subordinate feature oppositions unless the corresponding syntagm with its unmarked opposite is. This observation goes back (in essence) to Brondal (1943:105 f.), who spoke of it as "a principle of compensation" (cf. Jakobson 1971b: 214, 487 and passim, Andersen 1974, Gvozdanovic 1985: 37f.). It is notably similar to the generalization about subsidiary indexical signs in section 4.3. The morphological and lexical side of the comparison cannot be viewed, as Greenberg did, as a reduction in the number of "semantemes" in marked categories (59). Nouns retain their lexical gender in the plural even when their morphological markings do not reflect gender distinctions; and a noun phrase is in whatever syntactic case it has been assigned regardless of whether its morphological case affix is distinct from other case affixes; and similarly with all the other examples. Syncretism is not a reduction in meaning or content, but a reduction in the number of distinct expressions relative to the number of distinct combinations of grammatical or semantic features. The phonological side of the comparison, then, is a matter of the segment

34

Henning

Andersen

internal phonotactics of distinctive features; it concerns the inventory of readymade distinctive feature syntagms available for phonological representations (level C in (10)). Syncretism, on the other hand, is a peculiarity of the mapping relation between levels Β and C.

4.5 Basic allophone and agreement a potiori When a phoneme has several allophones, it may be possible to distinguish one or more non-basic variants which occur in specific environments, whose features they may share, and a basic variant which is independent of its environments and is unmarked with respect to the features carried by the nonbasic variants (22f., 59). Greenberg sees a parallel to this in the default agreement rules, common in languages with gender distinctions, which apply, for instance, to adjectives modifying conjoined nouns of different gender (59 f.). The phonological side of this comparison clearly has to do with the mapping relation between levels C and D and should be connected with the statement about neutralization in section 4.2. In his discussion of phonological neutralization, Greenberg mentions only the type in which an "archiphoneme" is consistently represented by the unmarked term of the opposition (13 f.). More commonly, however, the phonetic properties of both opposites are distributed in complementary environments. In such cases, apparently, the generalization at the beginning of this section holds, as, for instance, in Russian, where in positions where the voicing opposition is suspended, obstruents are voiced before a following voiced obstruent (regardless of intervening boundaries), but voiceless everywhere else, that is, before voiceless obstruents (regardless of intervening boundaries), at enclitic and word boundaries before an initial vowel or sonorant. and before pause. The morphosyntactic side of the comparison, by contrast, is internal to level B. It concerns the selection of the grammatical category to be expressed in a syntagmatic environment where the grammatical opposition in question is neutralized. This is in principle no different from the neutralization of the category of number in numeral phrases, mentioned in section 4.2. Note that in numeral phrases the number in the noun is redundant. If, for morphological reasons, number cannot be left unspecified in the noun (say, by the use of a bare noun stem or of a special numerative form of the noun), there are three logical possibilities: the noun may be in the unmarked number (paradigmatically motivated neutralization); it may be in the marked number (syntagmatically motivated neutralization); or the different numbers may be complemen-

Markedness

theory - the first 150 years

35

tarily distributed depending on the magnitude of the numeral or on the adjacent constituent of the numeral (as in Russian, which uses the singular with numerals ending in 2, 3, or 4 and the plural with higher numerals; or contrast Danish and English enogtyve natter '21 nights', but tusind og en nat Ί001 nights'). Mutatis mutandis, default agreement rules for gender show a similar diversity. But in all such cases, the grammatical neutralization is operated on the level of semantic representations (level Β in (10)), prior to the selection of the appropriate elements of expression. The apparent parallelism between phonology and morphology here veils a basic difference.

4.6 Zero expression and facultative expression Greenberg defines two morphological phenomena involving markedness, which seem to lack phonological counterparts. The first of these is the high incidence in languages of zero morphemes as expressions for unmarked categories, which involves the mapping between semantic and phonological representations (62), levels Β and C in (10). This is Lyons' first type of "marking (or markedness)" (section 1.1), which was noted by Peskovskij {cf. section 2.1) and was thematicized by Jakobson in "Signe zero" (cf. section 2.2) and exploited in his many papers on morphological patterns of the Slavic languages. By "facultative expression" Greenberg means the omission of semantic specification which is not communicatively necessary, as in the use of author for a writer regardless of sex, or the Korean optional use of the singular form of nouns also where more than one is referred to. In essence, this is only weakly differentiated from his "contextual neutralization", which was discussed in section 4.2.

4.7 Summing up We can now summarize the different phenomena considered by Greenberg in terms of the mapping relationships mentioned in section 4.0. A -*• B. Contextual neutralization (section 4.2) and facultative expression (section 4.6) concern obligatory or optional omissions of semantic specifications. They arise in the mapping of referential representations into semantic representations and can be understood in terms of the inclusive paradigmatic

36

Henning

Andersen

relations described by Hjelmslev (cf. section 2.3). Since the reference potential of the unmarked member of an opposition encompasses that of the marked one, it can be selected whenever there is no need for the specific reference of the marked member. Such omissions may be more or less normative, they may be grammaticalized or reflect genuine pragmatic options, the latter being motivated by the maxim of quantity. If one views them as neutralizations, they can be said to be paradigmatically motivated. B. Agreement a potiori (cf. section 4.5) arises in the concatenation of grammatical categories and can be seen as a syntagmatically motivated kind of neutralization. But when default agreement rules stipulate the selection of the unmarked member of a gender opposition, they are evidently paradigmatically motivated as well: here, too, the undelimited reference potential of the unmarked member is essential. Β -*• C. The assignment of zero expression to unmarked categories (cf. section 4.6) is a simple example of iconicity (cf Jakobson 1965): the quantitative opposition between zero depth (unmarked) and definite depth (marked) (cf. section 2.2) is diagrammed by a similarly quantitative distinction between zero and real expression. But it must be noted that this iconic relation obtains also in the numerous cases where the value relation for semantic reasons cannot be realized as a difference in reference potential, viz. when semantically non-inclusive relations are involved (e.g., married : unmarried (cf. section 2.4). The greater amount of allomorphy in unmarked categories was seen to be a special manifestation of Brondal's principle of compensation: unmarked members of oppositions being more hospitable to the addition of indexical content (cf. section 4.3). Syncretism may be understood in similar, though inverse, terms. While allomorphy provides symbolic signs which carry subsidiary, indexical information about their context, syncretism provides less specific symbolic signs, which for their specific interpretation are dependent on the context (cf. section 4.4). Both allomorphy, slanted in favor of the unmarked categories, and syncretism, biased towards the marked ones, produce skewed mappings between content and expression and thereby diagram markedness values. C. The fact that languages never have more, but often have fewer phonemes with a marked feature than with a corresponding unmarked one was interpreted as a consequence of Brondal's principle of compensation, which acts as a constraint on the expansion of distinctive feature syntagms with subordinate distinctive feature oppositions (cf. section 4.4). C —* D. The same principle constrains the expansion of distinctive feature syntagms; with (subordinate) allophonic differences unmarked phonemes are

Markedness theory - the first 150 years

31

more hospitable to the addition of subsidiary indexical phonic signs than their marked opposites (cf section 4.3). Phonological neutralization was viewed as the normative (but not necessarily involuntary) omission of phonological specifications and the subsequent assignment of phonetic properties to the thus reduced simultaneous syntagms (cf. section 4.2). Greenberg discussed only the cases where the segments in question are consistently assigned the properties of the unmarked member of the phonological opposition. However, as soon as one considers what Trubetzkoy termed "externally conditioned" realizations, the rules that specify these are clearly shown to be of the same kind as the ones that reflect Greenbergs generalization about non-basic and basic variants (cf. section 4.5): they assign marked properties to specified (marked) environments, and unmarked properties elsewhere (22f., 63 f.). This is a convenient place to return to the three types of "marking (or markedness)" described by Lyons 1977. Lyons' "formal marking" (cf. section 1.1) corresponds to Greenberg's "zero expression" (B ). To the extent that they are related to markedness, distinctions between zero and real expressions are an iconic reflection of markedness values. Lyons' "semantic marking" (cf section 1.3) covers the markedness values on the semantic side (level B) of the phenomena summarized above under A -»• B, whereas his "distributional marking" (cf. section 1.2) concerns the differences in extension (level A) engendered by these asymmetrical values. In view of the fact that Lyons' topic is semantics, it is natural that he would limit his exposition to phenomena that in one way or other manifest the value relations at level B. But his failure to introduce hierarchical relations among his three concepts of "marking (or markedness)" is a serious shortcoming. Evidently, markedness values ("semantic marking") are primary. But they could hardly be exhibited if they were not, in conjunction with some semantic oppositions, manifested in differences in distributions. Without doubt, some of these distributional differences are universal. By contrast, differences between zero and real expressions may, but do not necessarily reflect markedness values. Their possible correlation with markedness values, is in every case a language particular matter, which can be ascertained only provided that the markedness values have been established irrespective of their zero or real expressions. It appears from the preceding review of the markedness phenomena discussed by Greenberg that they can all be explicated in terms of one and the same formal principle. All paradigmatic relations in language, both semantic (grammatical and lexical) and phonological distinctions - and, in phonology,

38

Henning

Andersen

both phonemic distinctions and allophonic differences - are established as inclusive oppositions. As a consequence, they all incorporate asymmetrical value relations even though, from a functional (semantic or phonological) point of view, many of these paradigmatic relations are non-inclusive and hence symmetrical. The asymmetry of these value relations is reflected in skewed mappings between referential and semantic representations, between semantic content and phonological expression, and between phonological and phonetic representations. These are stown e. g. in default agreement rules, by preference to the unmarked term in certain syntagmatic environments, in the difference of readiness of marked and unmarked terms to be combined with subordinate distinctions, whether phonemic oppositions or allomorphic/allophonic variation. All these observable phenomena, it must be noted, presuppose, but are not identical to the asymmetrical value relations they reflect. The asymmetry of the "singular" : "plural" opposition may in a given language be reflected, in case syncretism in the plural paradigms or in greater allomorphic variation in the singular one. But it obtains independently of these specific manifestations. Similarly with lexical oppositions. These may be functionally incompatible with the inclusive relation of markedness values (as we saw in section 2.4) and may not necessarily, in any given language, be reflected by any of the types of phenomena reviewed above, not even by what Greenberg called contextual neutralization (section 4.2) or facultative expression (section 4.6). That they do indeed obtain can often be inferred only from more subtle manifestations, such as their preferred sequence in binomials (cf. section 1.3), or through word association tests, as Greenberg suggests (53 f.). Only occasionally do we find such spectacular demonstrations of the pervasiveness of markedness relations as the regular lexical distribution of Spanish and Philippine etyma in Zamboangueno, mentioned in section 2.4. But by whatever means markedness relations are manifested, their existence is logically prior to their manifestation, and they may consequently be presumed to obtain even when their existence has not yet been ascertained.

5. Understanding markedness 5.1 Markedness as a formal principle In the preceding sections I have distinguished between markedness as a formal principle which determines the form or paradigmatic relations in language

Markedness theory - the first 150 years

39

and the numerous - as yet far from fully charted - manifestations this principle has in languages. In this section I will briefly touch on the nature of this principle. In his 1966 monograph, Greenberg discussed some of the asymmetries he described, in terms of implicational relations, noting that whenever one of the observable phenomena can be stated as a universal implication, "it is the unmarked member which is the implied or basic term and the marked which is implying or secondary". This led him to the observation that the distinction between the marked term and the implied, fundamental character of the unmarked term appears very similar to the familiar distinction in Gestalt psychology between figure and ground (60). There is no doubt that this is a fruitful comparison, particularly if the distinction between perception and what seems to be involved in markedness relations can be specified. Note that in visual perception, the ground in fact includes the figure, but figure and ground are experienced as contradictories. In the framing of linguistic oppositive concepts (lexical, grammatical, or phonological), it appears as if there is an initial division of an experiential dimension into a salient, delimited area and a less salient, unbounded one which uncludes it. Only subsequently it comes to an identification of the functional character of their relation, which may be inclusive or exclusive, and if exclusive, contrary, contradictory, converse, etc. With this interpretation we are not very far from Levy-Bruhl's notion of participative relations, particularly as this notion was harnessed by Hjelmslev to account for the paradoxical conjunction in grammatical oppositions of logically diverse relations in meaning with inclusive relations in value (c/. section 2.3). But where Levy-Bruhl spoke of language as bearing the imprint of a pre-logical mentality, it would seem more to the point to recognize in the ubiquitous markedness values the effect of a cognitive strategy which takes precedence, ontogenetically, over the functional (and logical) analysis of the experiential dimensions encoded in language and culture. The inclusive relations which this cognitive strategy imposes on all experiential dimensions are apparently not superseded, in the life of the individual, by the results of later cognitive activity. It is for this reason that the linguist can conclude that all oppositions in language "fall under the law of participation" (Hjelmslev 1935: 102), and the logician who ventures beyond the confines of traditional logic must recognize that "exclusion is merely a special case of participation, in which certain areas (Fr. cases) of the extensive term are empty" (1939: 87). The cognitive interpretation of markedness suggested here forms a good basis for understanding Brondal's principle of compensation as well, for it can

40

Henning Andersen

be expanded into an account of both the conceptual elaboration of individual, homogeneous experiential dimensions and the formation of complexes of heterogeneous concepts. In terms of visual perception, it stands to reason that once a figure-ground distinction has been established, it is easier to perceive secondary differences in saliency in the (primary) ground than within the (primary) figure. Correspondingly, in the conceptual differentiation of any single experiential dimension, the less salient, undelimited area will be a more fertile ground for secondary divisions than the more narrowly delimited one. The difference carries over into the formation of conceptual complexes (semantic or phonological), where the undelimited area of an unmarked term of an opposition will form a better background for secondary, subordinate distinctions than the narrowly delimited one of its marked counterpart. Considering the fundamental importance of the figure-ground distinction for perception, it is not surprising if the most basic paradigmatic relation, the inclusive opposition, is founded on a homologous cognitive operation.

5.2 Definitions Against the background of this understanding, one can evaluate the different attempts that have been made in the history of the markedness concept to capture its essence. To Roth (1815), the unmarked vs. marked (U/M) relation was between a complex (U) and a simple (M) term (dependence and independence vs. dependence; cf. section 2.1), a view which adequately describes the different reference potential of the two terms. Kalepky (1901) emphasized the vagueness and indeterminacy of the unmarked and the saliency of the marked term in words that very neatly circumscribe the ground vs. figure relation. When he defined the difference between two opposites in terms of their privileges of occurrence, assigning one to specific conditions and the other to elsewhere environments, he in effect performed a similar analysis on the total range of relevant environments {cf. section 2.1). Peskovskij (1928) spoke of zero (U) vs. definite (M) meaning and drew the parallel to zero vs. real (segmental) morphemes. As I have tried to show (section 2.2), this quantitative conception of inclusive semantic relations corresponds perfectly to the opposition in semantic depth such relations embody, and it accounts well for their widespread iconic representation in quantitative distinctions between zero and real morphemes.

Markedness

theory - the first 150 years

41

Hjemslev, in his early writings on the subject, spoke of markedness as a relation between a vague term (U) and a precise term (M) (thus in 1933:102 ff.), a conception very similar, in essence, to Kalepky's. His ultimate understanding of participation as a value relation, distinct but inseparable from the semantic (or phonological) relations with which it is conjoined, has proven superior, I think, to all the other views that have been examined here. Both Roth and Kalepky explicitly viewed grammatical oppositions as sui generis contradictory relations. Jakobson initially contrasted the logicians' exclusive relations with those of language, but in the end was unable to free himself from the straitjacket of traditional logic (cf. section 3.1); In attempting to dissolve inclusive oppositions into conjunctions of exclusives he grappled unsuccessfully with the impossible. Hjelmslev saw correctly that exclusive relations can be reduced to inclusive ones, but not vice versa (1939: 87).

5.3 Applications With the understanding of markedness sketched in section 5.0, one can see that the range of phenomena to which the concept is applicable cannot be exhausted with the manifestations surveyed in Greenberg's monograph. If it is true, as Hjelmslev maintained, that all paradigmatic oppositions in language are framed in terms of inclusive relations, then markedness relations obtain in all cases where a language presents its speakers with a choice. 5.3.1 Synchrony Phonemic distinctions have traditionally been recognized as paradigmatic relations. But also allophonic distinctions form paradigmatic sets. And the range of environments in which any phonological sign may occur can be construed as a paradigm (cf. section 4.3). In morphology and lexis, not only do the content categories form paradigmatic relations, but allomorphs form paradigmatic sets, and their selection is dependent on a construal of their ranges of occurrence, such that one or more specified (M) allomorphs are assigned to specified (M) environments, and a single (U) allomorph is assigned to (U) elsewhere environments. Sentence syntax offers paradigmatic choices among grammatical constructions (active vs. passive, nominative vs. ergative) and element orders (direct vs. inverted) and text syntax among pragmatically conditioned focusing, topi-

42

Henning Andersen

calization, and cohesion devices (element order, iteration, substitution, anaphora, ellipsis). Individual speech act types present paradigmatic choices between indirectness and directness. Speaking involves paradigmatic choices among speech genres, and these among different registers and styles. In all such cases one can reasonably presume that asymmetrical value relations obtain, and it makes good sense for the linguist to posit them and, to the greatest possible extent, determine their manifestations - individually as well as in their interaction in complexes governed by the principle of compensation. Among the manifestations of markedness in texts one can recognize juxtapositions of opposite values such as in the irreversible binomials, first described by Malkiel (1967; cf. section 1.3), whose sequence relations (unmarked before marked) are paralleled in phonology by the juxtaposition of opposite phonemic values in diphthongs (cf. Andersen 1972). On the other hand, when in discourse the paradigmatic relations (in grammar) are transposed into syntagmatic relations (in speech); the process is in part governed by a principle by which members of different paradigms which have identical markedness value tend to be distributed in contiguity relations. This principle was first formulated by Jakobson (1960). Some of its manifestations are unwittingly mentioned by Greenberg (cf. sections 4.3,4.5). Its generality and possible cognitive basis is discussed in Andersen 1987.

5.3.2 Diachrony The relevance of markedness relations for linguistic change in semantic and phonological oppositions was succinctly sketched by Greenberg (1966: 69 ff.). In an entirely different dimension, speakers of a language evidently perform choices between what is innovative and what is not, between what is archaic and what is not, and they may have a more or less clear conception of differences in productivity, just as they may have an awareness that some structural features of their language are more, while others are less congenial to its general plan. As it is well known, Sapir's conception of drift rested on the assumption that such long term developments in languages result from generations of speakers' subconscious, but similarly weighted choices between alternative means of expression, among which they favor innovations tending to change the system of the language consistently in the same direction, into greater conformity with a definite general plan or type.

Markedness

theory - the first 150 years

43

This conception, certainly, presumes the existence of a kind of value relations in internalized grammars of speakers. But whether such relations can properly be subsumed under markedness must depend on how Sapir's idea of drift can be made explicit as a theory of long term change in language.

6. Conclusion

In the sketch of the early history of the concept of markedness offered here, I have tried to show both the progressive clarification of the concept (summed up in section 5.1) and some of the unclarities that developed - as a consequence of the terminology established in the Prague school, and in the process of the transfer of the concept from one metatheoretical paradigm to another. It was Togeby who hailed the discovery of inclusive relations as "un evenement comparable ä la decouverte de l'Amerique par Christophe Colomb en 1492 - ou peut-etre plutöt l'oeuf de Christophe Colomb" (1965: 71). Effusive comparisons aside, Hjelmslev's definition of participation - and particularly his insight that the exclusive types of opposition are special cases of (are included by) the inclusive type - these were significant achievements. One can only regret that they did not become widely known at the time. Hjelmslev found the Prague school terminology apt to confuse, and he was right. Trubetzkoy's initial, tentative distinction between "an active mark" (M) and a "passive mark" (U) might have been improved on and might have led to a clarification of the concept involved. But the "marked vs. unmarked" terminology that carried the day in the 1930s at first led to a conceptual conflation of value relations with the distinction between real and zero expressions and in the longer run made it practically impossible to liberate the concept of markedness from the exclusive, contradictory relation coded in the terms "marked" and "unmarked". Some of the most vehement denouncements of markedness have been put forward by scholars who have equated markedness with text frequency and then discover a vicious circle, the "statistical fallacy", as Lass and Anderson have called it (1975: 290). In my critique of Greenberg's monograph (section 4.0) I suggested that his concern with text frequency as a heuristic criterion for markedness reflected an inductivist approach, radically different from the hypothetico-deductive one of the European structuralists. Since markedness

44

Henning Andersen

is a formal principle and not an immediately given, quantifiable observable, one should not expect to be able to define it by boiling data down to statistics. But only the inductivist need fear the "statistical fallacy". If one distinguishes properly between the formal principle, the phenomena it determines, and the epiphenomena these phenomena entail, there is no circularity. On the contrary, the appropriate distinctions will make it possible to undertake serious investigations of how, in individual language states, the diverse skewed mapping relations between referential and phonetic representations may conspire to produce differences in text frequency which correlate with the markedness values of linguistic entities and how, in diachrony, markedness values and the frequencies of semantic and phonological entities can mutually influence each other. A confusion of markedness with text frequency obviously stands in the way of such investigations. The sketch of the first 150 years of markedness theory which I have offered here has deliberately been centered, as much as possible, around the notions, the terminology, and the phenomena that were discussed within that period. In the decades since the publication of Greenberg's monograph, markedness theory has flourished in part on terms independent of the earlier history of the concept. An account of this development, which belongs to the contemporary history of linguistics, will obviously require a very different approach. But perhaps a retrospective like the one presented here can offer a useful perspective also for that future account. References Aksakov, K. 1875 Socinenija filologiceskie [Philological studies], I. Andersen, Henning 1968 "IE *s after i, u, r, k in Baltic and Slavic", Acta Linguistica Hafniensia 11:171 - 1 9 0 . 1972 "Diphthongization", Language 48: 11 - 5 0 . 1974 "Markedness in vowel systems", Proceedings of the Eleventh International Congress of Linguists, edited by L. Heilmann (Bologna: II Mulino): 1136-1141. 1975 "Variance and invariance in phonological typology", Phonologica 1972, edited by Wolfgang U. Dressier & Frantisek V. Mares (München: Fink): 67-78. 1980 "Morphological change: towards a typology", Historical Morphology, edited by Jacek Fisiak (The Hague: Mouton): 1 - 5 0 . 1987 "On the projection of equivalence relations into syntagms", New vistas in grammar: invariance and variation, edited by Stephen Rudy and Linda R. Waugh, (Amsterdam: John Benjamins). Brendal, Viggo 1943 Essais de linguistique generale (Kobenhavn: Munksgaard). Fortunatov, F. 1899 "O russkix zalogax" [On Russian voice], Izvestija Otdelenija russkogo jazyka i slovesnosti Akademii Nauk.

Markedness theory - the first 150 years Fox, James 1971

45

"Semantic parallelism in Rotinese ritual language", Bijdragen tot de Taal, Land- en Volkenkunde, 127: 215-255. 1974 "Our ancestors spoke in pairs: Rotinese views of language, dialect and code", Explorations in the ethnography of speaking, edited by R. Bauman & J. Sherzer (Cambridge: University Press): 65—85. 1975 "On binary categories and primary symbols", Interpretation of symbolism, edited by R. Willis (London: Malaby Press): 99-132. Frake, Charles O. 1971 "Lexical origin and semantic structure in Philippine Creole Spanish", Pidginization and creolization, edited by Dell Hymes (Cambridge: University Press): 223-242. Greenberg, Joseph H. 1966 Language Universals, with special reference to feature hierarchies (= Janua Linguarum, ser. min., 59), (The Hague: Mouton). Gvozdanovic, Jadranka 1985 Language system and its change. On theory and testability (= Trends in Linguistics. Studies and Monographs, 30) (Berlin: Mouton de Gruyter). Hjelmslev, Louis 1928 Principes de grammaire generale (= Det Kgl. Danske Videnskabernes Selskab, Historiskfilologiske Meddelelser, 16,1) (Kobenhavn: Hast & S0n). 1935 La categorie des cas. Etude de grammaire generale (= Acta Jutlandica, 7,1) (Aarhus: Universitetsforlaget). 1939 "Notes sur les oppositions supprimables", Travaux du Cercle linguistique de Prague, 8:51-57. Reprinted in his Essais linguistiques (= Travaux du Cercle linguistique de Copenhague, 12), 2nd edn. (K0benhavn: Nordisk Sprog- og Kulturforlag). 1973 "Structure generale des correlations linguistiques" (1939), Essais linguistiques, II (= Travaux du Cercle linguistique de Copenhague, 14) (Kobenhavn: Nordisk Sprog- og Kulturforlag). Jakobson, Roman 1921 Novejsaja russkaja poezija [Recent Russian poetry] (Praha). 1932 "Zur Struktur des russischen Verbums", Jakobson 1971 b: 3 - 1 5 . 1939a "Signe zero", Jakobson 1971b: 211-219. 1939b "Das Nullzeichen", Jakobson 1971 b: 220-222. 1960 "Linguistics and poetics", Style in language, edited by Thomas A. Sebeok (Cambridge ΜΑ: M.I.T. Press): 350-377. 1971a Selected writings, I. Phonological studies (The Hague: Mouton). 1971b Selected writings, II. Word and language (The Hague: Mouton). Jakobson, Roman & Linda Waugh 1979 The sound shape of language (Bloomington: Indiana University Press). Kalepky, Theodor 1901 "Konträre oder kontradiktorische Gegensätze", Zeitschrift fur romanische Philologie 25: 339-340. Karcevskij, Serge 1927 Systeme du verbe russe (Praha). Lass, Roger & John M. Anderson 1975 Old English phonology (Cambridge: University Press). Lyons, John 1977 Semantics, / - / / (Cambridge: Cambridge University Press). Malkiel, Yakov 1967 "Studies in irreversible binomials", in his Essays on linguistic themes (Oxford: Basil Blackwell): 311-356.

46

Henning Andersen

Nekrasov, N.P. 1865 Ο znaceniiform russkogo glagola [On the meaning of Russian verb forms] (Sankt Peterburg). Peskovskij, Aleksej M. 1956 Russkij sintaksis ν naucnom osvescenii [Russian syntax in scientific light] 7th edn. (Moskva: Ucpedgiz). Roth, G.-M. 1815 Grundriss der allgemeinen Sprachlehre (Frankfurt). Sapir, Edward 1921 Language: an introduction to the study of speech (New York: Harcourt, Brace and World). Saussure, Ferdinand de 1916 Cours de linguistique generale (Paris: Librairie C. Klincksieck). Saxmatov, A. M. 1927 Sintaksis russkogo jazyka [Russian syntax] II. Togeby, Knud 1965 "Theodor Kalepky et les oppositions participatives", Acta Linguistica Hafniensia 9:71-76. Trubetzkoy, Nikolaj Sergeevitsch 1931 "Die phonologischen Systeme", TCLP 4:96-115. 1933 "La phonologie actuelle", Journal de Psychologie 30: 219-246. 1957 Principes de phonologie, trad, par J. Cantineau (Paris: Librairie C. Klincksieck). 1958 Grundzüge der Phonologie (Göttingen: Vandenhoeck & Ruprecht). 1969 Principles of phonology, translated by Christiane A.M. Baltaxe (Berkeley & Los Angeles: University of California Press). 1985 N.S. Trubetzkoy's letters and notes, prepared for publication by Roman Jakobson (Berlin: Mouton de Gruyter). Vostokov, A. 1831 Russkaja grammatika [Russian grammar].

Defining markedness Jadranka Gvozdanovic

1. A brief conceptual history

Structuring is an important aspect of biological functions, language being one of them. It involves analysis in terms of relations among the distinguished units, which can be either unrelated or related, and if related, either symmetrically or asymmetrically related. Symmetry is defined as mutual implication, and asymmetry, as implication in one direction, e. g. Β implies A but not vice versa. If Β is in such a case the common denominator of e. g. Bx and By, whereby (x) and ( y ) are the additional specifications of Β as compared with A, then the asymmetrical implication between Β and A can be viewed as a hierarchical relation by which A is dominating and Β subordinated. In studies of language systems, relations among the units of a system (termed "paradigmatic" relations) and distribution possibilities of these units (termed "syntagmatic" relations) play a central role. A systematic investigation of paradigmatic and syntagmatic relations comes into the picture with early structuralism and with the chronologically first area in which comprehensive generalizations within and across languages have been made: that of phonology. Phonological units are those sound units which are capable of distinguishing meaningful units in a given language. Their distinctive capacity is based on oppositions with the other sound units which could possibly occur in the same environment and by occurring there produce forms which are systematically associated with different meaningful units. Two types of classification of phonological units are possible: one based on their systematic phonetic properties and one based on the relational evaluation of these systematic phonetic properties within the sound system of a given language. These two types have been distinguished by Trubetzkoy (1939/1958 2nd: 67 etc.), who called the first type "factual" and the second type, "logical". In both types, the following classes occur: (a) privative, in which one member of a binary opposition is characterized by

48

Jadranka

Gvozdanovic

the presence of a mark (which is Jakobson's translation of Trubetzkoy's priznak in Russian and Merkmal in German, cf. Jakobson 1971,2nd: 735), whereas the other member is characterized by its absence (e.g. voiced vs. voiceless in consonants, or nasalized vs. unnasalized in vowels), (b) equipollent, in which the members of an opposition are equal (e.g. front vs. back in vowels), and (c) gradual, in which the members of a ternary or a more complex opposition are characterized by various degrees of the same distinguishing characteristic (e.g. high vs. mid vs. low in vowels). In Trubetzkoy's (1939/1958 2nd: 69) view, factually privative oppositions correspond with logically privative ones, factually gradual oppositions with logically gradual ones, and factually equipollent oppositions with logically privative, equipollent or gradual ones. This logical evaluation depends partly on the structure and partly on the functioning of a given phonological system. Privative oppositions are of particular interest to us here, because this is where markedness shows up in phonology: it is only in these oppositions that one member of the opposition is characterized by the presence of a mark, i. e. it is marked, whereas the other member is characterized by the absence of the mark, i.e. it is unmarked. And it is privative evaluation of factually (i.e. phonetically) nonprivative oppositions which is crucial to our understanding of markedness as an aspect of language structuring. When are phonetically nonprivative oppositions evaluated in a phonological system as privative? Throughout the development of Trubetzkoy's ideas of markedness (ending with the references mentioned above), there is no clear answer to this question. The only indication can be deduced from his statement (1939/1958 2nd: 72 ff.) that when an opposition is neutralized and its representative does not depend on the neutralizing context in the sense of either assuming the same phonetic value or a value intermediate between the two neutralized values, but is rather system-internally determined, then there are two possibilities: (a) if the opposition is privative, the unmarked member shows up, and (b) if the opposition is gradual, the "outer" or "extreme" member shows up. The unmarked member is thereby analysed as equalling the archiphoneme (i. e. the common denominator of the two minimally opposed members of the opposition which undergoes neutralization) plus zero, whereas the marked member is analysed as the archiphoneme plus a given mark. Appearance of the unmarked member is phonetically clearly establishable in the case of context-independent neutralization of phonetically privative oppositions. In context-independent neutralization of phonetically gradual

Defining markedness

49

oppositions, however, how can we know that the "outer" or "extreme" member showing up is not system-internally the unmarked one of a privatively evaluated opposition out of a sequence of two or more such oppositions? And if neutralization affects a phonetically equipollent opposition, is this neutralization the only basis for stating that this opposition is system-internally evaluated as privative? Trubetzkoy did not formulate answers to these questions. We can see that at the beginning of the history of markedness in linguistic theory, there was a lack of clarity as to the extent of markedness which reflects the lack of clarity in its definition. Is neutralization the indicator of markedness? And if so, is it the unmarked member which shows up under the conditions described above, or the form equalling that of the unmarked member? In other words, does the form identifiable as that of the unmarked member stand for the unmarked member there, or for the common denominator of the two members the distinction between which is neutralized in the given position? Trubetzkoy's formulation (1939/1958 2nd: 73) by which "that member of the opposition which is admitted in the position of neutralization is from the viewpoint of the given phonological system unmarked, whereas the opposed member of the opposition is marked", can be understood as the first alternative mentioned above. It was not meant as such, however, since the given question arises only against the background of understanding phonological systems as sign systems, as formulated explicitly by Jakobson (1939). Jakobson showed that phonemes are simultaneous syntagms of distinctive features, which are signs with a merely oppositive and negative meaning (in the sense of Saussure's (1916) "value") and a form determined by the relevant phonetic characteristics, capable of distinguishing signs with a positive meaning (/'. e. morphemes etc.). It is against this background that the question of the sign value of the phonological unit showing up in the position of neutralization can be formulated and answered in a clear-cut way. Given the definition of sign as a unit of meaning and form, and given the definition of distinctive feature as merely oppositive in meaning, it follows that whenever an opposition is dismissed, i. e. neutralized, the given distinctive feature is dismissed as a sign. What then remains in the position of neutralization, equals the common denominator of the two minimally opposed phonemes, whereas the distinctive feature specifications distinguishing them in other positions are lacking. This common denominator, called archiphoneme, is a complex sign unit which contains one constitutive sign less than either of the two corresponding minimally opposed phonemes, irrespectively of the fact that its form may

50

Jadranka

Gvozdanovic

equal that of one member of the corresponding minimally opposed pair. For example, in languages with voice neutralization at the end of the phonological word, the voiceless consonants which occur in that position contain all the signs which specify them as consonants of a certain articulation place and manner, but lack the sign of voicelessness, as there is no opposition with voicedness in that position. This means that a voiceless implementation in that position has no sign value, whereas a voiceless implementation in the remaining positions does have a sign value, with its oppositive (and hence also negative) meaning. Trubetzkoy used the term "unmarked" as a cover term for both (a) a sign characterized by its form and its oppositive meaning, and (b) the same form which shows up in the position of neutralization, where the oppositive sign meaning, and hence the sign value connected with it, is omitted. Marked, on the other hand, was used only with the oppositive (and in the sense of excluding the opposite value also negative) meaning, i.e., it was used only when the sign value was present. Trubetzkoy did not answer the question of whether the occurrence of "unmarked" in the position of neutralization is the necessary and sufficient condition for distinguishing "unmarked" from "marked", and for the applicability of these terms in general. This is why he has not given us a clear definition of markedness in his writings. In the further development of phonological theory, highlighting distinctive features as components of phonological signs, criteria for determining markedness and the extent to which it is relevant continued to play a major role. The following criteria were considered: (a) the marked member involves some deviation from the norm, which is absent in the unmarked member, (b) the marked member has a well-defined characteristic or characteristics, whereas the unmarked member may vary considerably, and (c) (as a consequence of its occurrence in neutralization positions,) the unmarked member has a wider distribution than the marked member. The given criteria do not uniquely determine the possible solutions, however, as shown by alternative results reached. Inconsequent treatment of markedness is in fact found throughout the development of the distinctive feature theory, starting from Jakobson's early writings. Jakobson (1949) assumed that each' + ' specification of a distinctive feature (implicitly assumed to equal a well-defined phonetic characteristic) is marked, whereas its opposite, i.e. ' —' specification and predictability of a

Defining markedness

51

feature are unmarked (equalling zero). He applied markedness to the entire distinctive feature system of a language (in the given case, Serbo-Croatian), disregarding problems connected with establishing the phonetic characteristics which he rendered as "marked" (e.g. presence of both vocality and nonvocality in /r/, /l/ and jYj, and absence of this characterization in /m/, /n/ and /n'/, which can be justified neither phonetically nor distributionally). Jakobson's failure to make his criteria explicit can be seen as a serious shortcoming, because there is not any straightforward connection between his usage of markedness and the sound reality assumed to underlie it. This does not hold only for the beginning of his phonological ideas, but also for their later development. In the second phase of the development of his ideas, he came in a comparably implicit way to the insight that some' —' specifications are unavoidable and do not consequently equal zero (cf. the 1962 reprint of the same article, discussed also in Gvozdanovic 1985: 8 f.), without giving an account of the necessity and conditions for their introduction (concerning their questionability, cf. Gvozdanovic 1985: 8ff.). In further phonological development, markedness relations between the members of distinctive feature oppositions received due attention, especially in generative phonology. Markedness was there explicitly viewed as universally specifiable on the basis of phonetic criteria of the types (a) and (b) mentioned above, in combination with the third, distributional criterion as formulated under (c). The "phonetic norm" was thereby understood mainly as "ease of articulation", revealed by distributional frequencies in the languages of the world as well. Syntagmatic and paradigmatic distribution possibilities were viewed as indicative of markedness not only in the sense of considering predictable feature specifications unmarked, but also in the sense of considering distributionally most frequent feature combinations unmarked. In the first step, universal markedness conventions were formulated, and in the second step, language-specific deviations from them were established and treated as adding to the complexity of those lexical entries in which they are encountered (cf. especially Chomsky and Halle 1968: 402 etc.). However, "the markedness conventions reached on the basis of phonetic and distributional criteria appeared not to be uniquely determined" (cf. Kean 1981 vs. Van Lessen Kloeke 1981, discussed also in Gvozdanovic 1985: 72 etc.). At best, they are probabilistic statements about distributions, predicting that "when a marked situation develops it should not be particularly stable" (Herbert 1986: 50), i.e. stating that there is a high probability that marked situations will not occur often, and if they do occur, that they will not be particularly stable. Apart from circularity (as markedness is not established independently of distribution), this approach suffers from the shortcoming

52

Jadranka

Gvozdanovic

that language-specific deviations from universally predominant patterns and counterpredicted developments (discussed e.g. by Lass 1980) cannot be explained, which may be considered a serious problem for a theory which aims at being a universal theory of language. We must conclude that this problem is rooted in the basis of markedness, namely its definition. The problem is not limited to phonological signs, whose meaning is essentially oppositive. It is found also with lexical and grammatical language signs which are characterized by a positive meaning of their own. The only difference between phonological signs on the one hand, and grammatical and lexical ones on the other, is found in the applicability of the criteria (a) and (b) mentioned above to the form of phonological signs, and to the meaning of grammatical and lexical ones (and the distribution of their forms in accordance with their meanings). The history of markedness at the level of meaning starts with Jakobson's correspondence with Trubetzkoy in the early thirties, reported by Jakobson and Waugh (1979: 90 f.). Trubetzkoy wrote to Jakobson in 1930: "Statistics has nothing to do with it. And the essence lies in the so-to-speak "intrinsic content" of the correlation. Apparently any (or might it not be "any"?) phonological correlation acquires in the linguistic consciousness the form of a contraposition of the presence of a certain mark to its absence (or the maximum of a certain mark to its minimum)." And Jakobson answered: "I am coming increasingly to the conviction that your thought about correlation as a constant mutual connection between a marked and unmarked type is one of your ost remarkable und fruitful ideas. It seems to me that it has significance not only for linguistics but also for ethnology and history of culture, and that such historico-cultural correlations as life ~ death, liberty ~ nonliberty, sin ~ virtue, holidays ~ working days, etc., are always confined to relations α ~ non-a, and that it is important to find out for any epoch, group, nation, etc., what the marked element is." Jakobson thus recognized that markedness at the level of meaning is comparable to markedness at the level of form, and that its evaluation is system-dependent. He subsequently applied the notion of markedness to the grammatical level to language (1932), showing that in categories based on binary correlations ("correlation" being a term more general than "opposition", as the former is applicable to the level of meaning as well), one of the members can be viewed as characterized by the presence of a mark, whereas the opposite member is characterized by its absence. So he wrote that the general meaning of the Russian word oslica 'she-donkey' signifies only a female donkey, whereas the general meaning of its Russian opposite osel 'donkey' does not signify any gender of the animal. If, however, somebody

Defining markedness

53

asks eto oslica! 'is this a she-donkey?' and the answer is net, osel 'no, a donkey', then the usage of osel for the male gender is according to Jakobson a case of narrowed meaning. It is not the general meaning of osel, Jakobson (1932) proceeded to argue, because then the usage of osel for both genders would have to be a case of broadened meaning, with the same effects as brought about by figurative or metaphorical usage, which are not attested in the usage of osel for both genders. Unfortunately, Jakobson apparently did not distinguish between meaning and interpretation at this point. Otherwise, he would presumably also have discussed the logically alternative solution to the one proposed, namely that osel used for both male and female is a case of narrowed meaning, whereas its usage for male only is general. In the narrowed meaning, gender is missing as a sign, thus allowing interpretations referring to both genders. Whenever gender is present as a sign, interpretational possibilities are restricted by it. In the article which starts with the definition of markedness presented above, Jakobson applied markedness to the Russian verb system. And comments similar to those presented above apply to markedness in the verb system as well. The term "general meaning" is there misleading, too, and we can better proceed to discuss "meaning" as distinguished from "interpretation". Take, for example, the tense correlation between the preterite and the present in Russian. Jakobson defined the preterite as signifying that the action belongs to the past, whereas the present does not signify any tense; the preterite is marked and the present unmarked. Given this absence of the mark of tense, the present can be used as historical, gnomic etc., whereas the preterite has relatively restricted possibilities of usage. Jakobson's nondistinguishing between "mark", characteristic of the marked member, and "sign" was a source of confusion here as much as it was in phonology. In order to avoid this terminological vicious circle, I shall proceed by discussing "signs". In the sense of signs, it is correct to state that the present is used when the sign is missing, but it is incorrect to state that the sign of tense is always missing in the present. A test for this can be set up by constructing sentences containing a past temporal adverb or another verbal form in the preterite in a coordinated or subordinated clause, and adding to it a verbal form in the main clause in the preterite on the one hand, and the present on the other. An investigation of the meaning of the present in such sentences, as opposed to the preterite, shows that it is incorrect to say that the present does not signify any tense. On the contrary, it does signify the present, at least partly coinciding with the speech moment or another temporal orientation point indicated as relevant (whereas the preterite necessarily precedes the speech moment or the orientation point), and usage of the present

54

Jadranka

Gvozdanovic

in such sentences has the effect of transposing the speech moment or the relevant orientation point onto the past. We can conclude that there are two clearly distinguished usages of the present: one in which the sign of tense is absent (as in the gnomic present, with its timeless validity), and one in which the sign of tense (that of nonpast) is present. The historical present belongs to the second group. Similar comments apply to Jakobson's discussion of verbal aspect in Russian. Aspect is a binary correlation, its members being the perfective and the imperfective aspect. Jakobson defined the perfective aspect as signifying the absolute boundary of the action, whereas the imperfective aspect does not signify it. The perfective aspect is consequently marked and the imperfective aspect is unmarked. Later investigations (cf. Barentsen 1985) have shown that this 'absolute boundary' semantically equals a terminus and the corresponding situation change, such that there is a situation with the action and a situation without it. In addition, there is a sequential connection of this situation change onto an orientation point: a Russian does not use (1) if the window is closed at the speech moment or another orientation point, but rather (2), with the English translation of the Russian aspectual difference in the given usage as Ί had opened the window'. (1)

Jä otkryl

okno

1

window

en

°P perfpret Ί opened the window.'

(2)

·

Jä otkryväl

okno

1

window

en

°P imperfpret Ί was opening the window.' / Ί had opened the window.'

In other words, if the sequential connection of the situation change is lacking, the perfective aspect cannot be used. The meaning of the imperfective aspect is absence of a situation change and/or of the sequential connection. There are two variants: either the terminus connected with the potential situation change is not reached, or it is reached repeatedly. In Jakobson's terminology, either no boundary is reached, or only a relative boundary or relative boundaries are reached. Can this be viewed as the absence of a sign? The problem with Jakobson's analysis is here again that it is incomplete. It does not reveal the fact that in the imperfective pair of the perfective aspect, the terminus is potentially present, and explicitly denied in the final result of the action expressed by means of the given verb form. This general meaning of

Defining markedness

55

the imperfective aspect is a denial of the meaning of the perfective aspect, not the absence of the given linguistic sign. Only in those imperfective verbs which have no perfective counterpart, the given linguistic sign is missing as a part of the meaning, but then markedness is not at issue as there is no correlation. The imperfective aspect can be used in two ways: either as a denial of the perfective aspect (e.g. (3)) or when the aspect is not at issue (e.g. (4)). (3)

Jä dolgo perepisyval

etu

I

this article

no

long c o p y i m p e r f ne

pret

uspel

but not manage p e r f

pret

stat'ju

perepisät'

ee

copy

it

Ί was copying this article for a long time, but did not manage to copy it all out.' (4)

Etot skol'nik

ploxoj:

ot

this pupil

bad:

from others c o p y i m p e r f

drugix

perepisyvaet pres

'This pupil si a bad one: he copies from the others.'

The examples given above show that in aspect correlations in Russian, the imperfective verb form can be used either (a) to signalize that the terminus and/or the sequential connection is not reached (as in (3)), or (b) that this sign is not at issue (as in one of the interpretive variants of (4)), but (b) can occur only as a contextually conditioned variant of (a). The imperfective verb form, on the other hand, always signalizes and the terminus and the sequential connection are always reached. In view of these contextually determined variants of the imperfective aspect, and their absence with the perfective aspect, it seems appropriate indeed to call the former "unmarked" and the latter "marked". This differs, however, from Jakobson's way of reasoning. Jakobson (1932: 8) also analysed the distinction between the modal vs. indicative finite verb forms in Russian as a case of "marked" vs. "unmarked", as the modal forms denote that there is no real connection between the action and the acting person (the action might be unmotivated or arbitrarily ascribed to the acting person or forced onto him), whereas the indicative finite verb forms lack such denotation. In view of the analysis of the aspect correlation presented above, we can now ask ourselves whether this is a comparable case. The Russian modal examples (according to Jakobson, all containing singular imperative verb forms, as in (5) and (6)) can be analysed as meaning that the action does not take place at the time axis at which the speech moment (or the relevant temporal orientation point) is situated, and/or that it is not controlled by the agent (not necessarily restricted to "the acting person"), as in (5) and (6) below.

56 (5)

Jadranka

Gvozdanovic

Pridi

on, vse by

ulädüos'

comehe all would get in order 6 imper 'If he would come, all would get in order.' (6)

Vse govorjät,

a

my

molci

all talkand we shut upimperff pres Mmper 'All are talking, and we must shut up.'

The indicative forms of the same verbs, on the other hand, can be analysed as placing the action at the time axis at which the speech moment (or the relevant temporal orientation point) is situated, and viewing the action as controlled by the acting person. All of Jakobson's examples of the modals are in the imperative, without distinction of person in the ending, which accounts for the absence of control by the (intended) agent, and without distinction of tense, which accounts for the absence of placement at the time axis at which the speech moment (or the relevant temporal orientation point) is situated. By virtue of this fact, the modal vs. indicative distinction is not a case of correlation based on a single distinction of meaning. In the examples discussed by Jakobson, it is rather a complex distinction between modal vs. indicative syntactic constructions, characterized by a different relation between the meaning of the verb form and the meanings of the remaining constituents within the same syntactic unit: in indicative syntactic constructions, the signs of person (in the nonpast) and of tense are present in the verb ending, and there is coreferentiality with the person of the agent, whereas in modal syntactic constructions (in the given examples, statements with the verbal predicate in the imperative), the signs of person and tense are lacking in the verb ending, and there is nonidentity of the time axis and/or of control as compared with the rest of the given syntactic unit. Modal verb forms thus point to a different dimension than the one expressed by means of the remaining constituents within the same syntactic unit, and in this sense add to the information expressed within the same syntactic unit in the way in which indicative verb forms do not do. It is in this sense that statements and questions containing imperative verb forms may be termed "marked", whereas they may be termed "unmarked" if containing indicative verb forms. As to the verb forms themselves, the (synthetic) imperative has two variants: one containing the sign of the second person, and one lacking a sign of person (as in the examples mentioned above), whereas the indicative always contains a sign of person. If containing the sign of the second person, the imperative still fundamentally differs from the indicative in that the synthetic imperative does not express distinction of person, whereas the indicative does express it. It is in this sense that the verb forms themselves do not express a single

Defining markedness

57

correlation of meaning which would fit into the definition of markedness, but only the given syntactic constructions in which they can be used to fit into such a definition. Jakobson's presentation of the data was here oversimplified and his usage of markedness confusing. Comparable comments apply to calling the passive in Russian "marked", and the reflexive "unmarked". In Jakobson's own formulation, the passive means that the action is not brought by the subject, but rather exerted onto him from the outside. The reflexive means mere intransitivity in all cases, but can in some contexts be interpreted as passive. By stating this, Jakobson in fact stated that markedness is in this case applicable to syntactic constructions, not to isolated passive vs. reflexive verb forms. If we investigate the subject matter along these lines further, we can see that a more adequate formulation of the reflexive meaning would be that one of the actants combinable with the lexical meaning of the verb is left unspecified. Either the agent or the goal can be lacking in this way (or they may be mutually undistinguished), and if the agent is lacking, then the action is not interpreted as actually occurring at the time axis at which the speech moment or the relevant temporal orientation point is situated, but rather as a potential one. As for the passive, on the other hand, we can see that the referential agent is either left unspecified or expressed by means of the instrumental case, characterizing the referential agent as a linguistically marginal actant (concerning "marginality" as the meaning of the instrumental case in Russian, cf. Jakobson 1936). We can consequently say that the agent is linguistically lacking in passive syntactic constructions, whereas the goal of the action is expressed as the central actant, in the nominative case and fulfilling the syntactic role of subject. An additional semantic characterization of the passive is connected with its syntactic usage, by which passive verb forms are used either attributively or as a part of the nominal predicate. Corresponding with this syntactic usage, the passive denotes the result of an action exerted onto a goal, and the goal expressed as the syntactic subject of a passive sentence is presented as bearing the characteristic which results from the action exerted onto him. This differs fundamentally from syntactic usage possibilities of reflexive verb forms, which function as verbal predicates, with the corresponding semantics. Again, the relation between the passive and the reflexive as mentioned by Jakobson is not a simple correlation in which one of the members is characterized by the presence of a sign, whereas the opposed member is characterized by its absence. We must conclude that the first application of markedness to grammatical data was based on an incomplete analysis of the investigated data, and on an insufficiently operationalized definition of markedness as concentrating on a

58

Jadranka

Gvozdanovic

greater complexity of the marked member, and the usage possibility of the unmarked member to be interpreted so as to denote the appropriate referent(s) of the marked member as well. Owing to its incomplete analysis of the data, Jakobson's work did not shed enough light on the semantics of the unmarked member, as I hope to have shown above. I paid so much attention to these beginnings of the history of markedness because they contributed to the full applicability of markedness to a given level of language to be seen as a fundamental principle, such that the remaining ones could be derived from it. This can be illustrated by Van Schooneveld's (1978:11 f.) formulation of what he considers to be Jakobson's most important contributions to the theory of semantic structure. As Van Schooneveld put it, "it is possible to discover, among the morphemes of a given language (in casu Russian) recurring relations of a semantic nature; since these relations are based upon the presence or absence of semantic "distinctive features", occurring in different combinations within the units involved, these units are opposed to each other as "marked" vs. "unmarked". A marked unit carries a specific piece of information as given by a particular distinctive feature, whereas no such information is given in the unmarked unit. The unmarked unit contains no positive statement with respect to the specific item of information involved, i. e., it remains uncommitted... The alternatives "presence of specific information" vs. "absence of specific information" being only two, we are necessarily dealing with binary oppositions." Van Schooneveld's reasoning that the recurring semantic relations to be discovered among the morphemes of a given language are based on markedness, and that these relations are consequently by necessity binary, manipulates semantic theory into a vicious circle, which can be solved only if the binary nature of the assumed correlations is investigated separately from the investigation of markedness, and both are based on a thorough investigation of the meaning of each unit. Due to lack of clarity at the beginnings of the history of markedness, the terms "marked" and "unmarked" have been used as nothing more than descriptive labels for language phenomena which require further clarification.

Defining markedness

59

2. Defining markedness A definition of markedness which can be assumed to provide a workable basis for linguistic analysis, and account for various sorts of language data which cannot be accounted for otherwise, should be based on an asymmetry which manifests itself either syntagmatically or paradigmatically. Syntagmatic markedness is encountered whenever both members of a single correlation can occur within the same context which is definable in terms of form (e. g. sentence intonation and congruence phenomena) and of meaning, and by doing so one of them adds to it the information which is not added by the other member (which also does not add to it any incomparable sort of other information). This can be illustrated by the distinction between the present vs. the preterite verb forms (with the corresponding meaning correlation of past vs. nonpast) used in sentences with a past temporal adverbial, as mentioned in the preceding chapter. Whereas the preterite does not add in such cases any new information on the placement of the given action at the time axis, the present tense does add the information on a temporal transposition of the speech moment (or the relevant temporal orientation point) at which the action takes place onto the past. Syntagmatically seen, the present tense is in such sentences marked, whereas the preterite is there unmarked. Syntagmatic markedness is encountered also whenever a set of mutually related correlations (e.g. past vs. nonpast, and in the nonpast, correlations of person) can be either present or absent within the same context, and either its presence or its absence adds to the given context the information which is not added in the opposite case (which also does not add to it any incomparable sort of other information). This can be illustrated by the distinction between the imperative vs. the indicative verb forms used in statements (and possibly also questions) as discussed in the preceding chapter. The imperative, in which the correlations of tense and person are absent, adds to the information on tense and person in the given context by pointing to the absence of coreferentiality, and thus the absence of placement at the same time axis and/or of control by the intended agent. The indicative, on the other hand, in which the correlations of tense and person (in the nonpast) are present, denotes placement at the same time axis and coreferentiality with the agent. By doing so, it does not add to the information on the dimensions present in the given context. This is why the imperative is syntagmatically marked in statements (as in (6) mentioned above), whereas the indicative is there unmarked (as in (7) below).

60

(7)

Jadranka

Gvozdanovic

Vse govorjät, a my molcim all talk, and we keep „ , F silent· c imperf pres 3. pi imperfc pres 1. pi 'All are talking, and we keep silent.'

Paradigmatic markedness is encountered whenever the members of a correlation are mutually asymmetrical so that one of them allows for a further division, whereas the other member does not allow for it. This allowance for a further division can be either one of usage possibilities or one of patterning with another correlation. The aforementioned further division in the sense of usage possibilities of one of the members which are not found with the other member will be discussed first. I shall henceforth call it "paradigmatic markedness within a correlation". Paradigmatic markedness within a correlation is encountered whenever the form of one of the correlation members can be used both with its sign value and in the absence of the given sign value, whereas the form of the other member of the correlation can be used only with its sign value. The former is then paradigmatically unmarked and the latter is paradigmatically marked. This definition, as an alternative to Jakobson's approach, can be acceptable only if it can be shown that the presence of a given oppositive sign value in the unmarked member is not added by the context and thus present only at the level of interpretation, but rather an actual feature of meaning contributed by the given sign. In order to test this, let us go back to Jakobson's examples of the type osel 'donkey' vs. oslica 'she-donkey'. As we know, Jakobson wrote that osel always lacks the mark of gender and can denote either a male or a female, whereas oslica always contains the mark of female gender. By using this formulation, he did not distinguish between 'mark' and 'sign', as I have pointed out in the preceding chapter. So we can proceed by investigating the presence vs. the absence of the sign of gender in such examples. In order to investigate this, I shall now discuss comparable pairs in another Slavonic language, Serbo-Croatian, which is my mother tongue, so that I am best aware of the meaning possibilities there. Consider: (8)

*To nije magarac, nego je muzjak this not-is donkey, but is male 'This is not a donkey, but a male.'

(9)

To nije magarac, nego je zenka this not-is donkey but is female 'This is not a donkey, but a female.'

(10)

To nije magarac, nego je magarica this not-is donkey but is she-donkey 'This is not a donkey, but a she-donkey.'

Defining markedness

(11)

61

To nije magarac, nego je konj this not-is donkey but is horse 'This is not a donkey, but a horse.'

If magarac, i. e. the Serbo-Croatian pendant of osel, were to lack the feature of gender, then the negative sentence presented under (8) should be as equally possible as the negative sentence presented under (9). This, however, is not the case: unacceptability of (8) as opposed to acceptability of (9) points to the presence of the gender feature for male in (8), (9), and (10), and its positional absence, i.e. neutralization, in (11). On the basis of the fact that the form of one of the members (i. e. "male") of the given correlation (i.e. "male" vs. "female") is systematically used in cases of neutralization of the given correlation, i.e. whenever the given sign value (of either "male" or "female") is absent, next to its being used whenever the sign value of the given member (i.e. "male") is present, it is possible to classify the given member as a negation of the opposite member (i.e. "male" as a negation of "female"), with the two variants of the negation described above (i.e. "male" as the opposite of "female", and as absence of "female", without opposition in the given context). If we now call the former member of the correlation (i.e. "male", classified as [— female]) "unmarked", and the latter member of the correlation (i. e. "female", classified as [ + female]) "marked", and if we restrict the applicability of markedness to correlations (as Jakobson did in his writings mentioned above); we can see that "unmarked" cannot be equalized with the absence of a sign, but rather with the two variants described above. In other words, if markedness is applicable to a correlation, it is a mere descriptive label for the observed asymmetry. If no asymmetry can be observed between the members of a correlation, or if there is no correlation (as in most cases among the lexical items of a language), then markedness is not applicable. Examples of the type magarac 'donkey' vs. magarica 'she-donkey' are not isolated in the Serbo-Croatian system. As a result of cultural determination, 'fox' is represented by the female animal, lisica, which is unmarked in the same way as magarac is, whereas its counterpart, lisac, is marked as the male fox. The test presented above applies to these examples as well: (12)

*To nije lisica, nego je zenka this not-is fox but is female 'This is not a (female) fox, but a female.'

(13)

To nije lisica, nego je muzjak this not-is fox but is male 'This is not a (female) fox, but a male.'

62

(14)

Jadranka

Gvozdanovic

To nije

lisica, nego je lisac

this not-is fox but is male fox 'This is not a (female) fox, but a male one.' (15)

To nije

lisica, nego je macka

this not-is fox but is cat 'This is not a fox, but a cat.'

These examples show the same type of asymmetry as observed above. Both (8) and (12) are unacceptable (in the sense of being self-contradictory) in the original meaning of their constitutive parts. The sentences would be acceptable only if interpreted as containing a metaphorical usage of magarac and lisica on the one hand, and muzjak and zenka on the other, e. g. as referring to a living being's acting in the given ways. In these examples, lisica contains the sign of gender which is opposite to that of lisac in (12), (13), and (14), whereas this sign of gender is positionally absent, i. e. neutralized, in (15). Reversal of the asymmetry pattern in gender in examples such as lisica '(female) fox' vs. lisac 'male fox', or macka '(female) cat' vs. macak 'male cat', as compared with magarac 'donkey' vs. magarica 'she-donkey', shows that the classification of the gender correlation as [ + female] vs. [—female] in Serbo-Croatian is in fact arbitrary as far as lexical items are concerned, as either gender is a negation of the opposite gender, and there are lexically determined cases of either of them being used when the sign of gender is missing. The only argument in favour of this classification can be found in the general regularity in the grammar of this language with which the male form is used for designating sets consisting either of male entities or of entities of different gender, i. e. if gender is not at issue, as in the following example. (17)

Stari

su lisica

i

lisac

masc pi old are female fox and male fox prosli

kroz

sumu

masc pi passed through wood 'The old female and male fox have passed through a / the wood.'

By this lexically independent rule of the grammar of Serbo-Croatian, the male form is used when the sign value of gender is missing, thus rendering "male" as unmarked, whereas "female" is marked. However, certain lexical contexts can condition so-called "markedness reversal", by which "female" is unmarked and "male" is marked, as in the case of fox or cat mentioned above. Generally speaking, "markedness reversal" can be conditioned either by the paradigmatic context, in the sense of cooccurring sign values characteristic

Defining markedness

63

of the same unit, or by the syntagmatic context, in the sense of sign values characteristic of cooccurring units. "Markedness reversal" is in fact "markedness" characteristic of a given context, in which the asymmetry found in the remaining contexts is reversed, as in the case of "male" vs. "female" discussed above. In addition to asymmetry within a binary correlation, asymmetry between binary correlations is connected with markedness in a way which is illustrated by the following correlations of number (discussed in the various relevant aspects in Gvozdanovic 1985: 117 etc.). (18)

mone

singular dual plural where: m = marked - u = unmarked1

In a language with a grammatical dual vs. plural correlation, e.g. in the Slavonic language Slovenian, the plural is unmarked as compared with the dual, because whenever choice is not at issue and the given semantic sign is thus neutralized (e.g. with "eyes", "ears", "arms", "hands", "legs" and "feet", usually in connection with one person), the plural is used. This is an example of markedness within a binary correlation. Markedness between two binary correlations is found if we compare this binary correlation with the correlation between "one" and "not one". There is asymmetry between these two correlations in the sense that whereas "one" does not allow for further distinctions, "not one" does allow for them, further specified as either "not one, two", or "not one, not two". On the basis of this asymmetry, "not one" is unmarked, whereas "one" is marked. This is an example of "paradigmatic markedness between correlations". An asymmetry of the type presented under (18) constitutes a hierarchy. At the lowest level of hierarchy, markedness is determined in the way outlined in the preceding part of this paper. At each higher level of a given hierarchy, the member which allows for more distinctions at the next-lower level is considered unmarked. Both types of markedness, i.e. markedness within a binary correlation and markedness between binary correlations (as illustrated by (18)), are partly universal and partly language-specific, depending on conditions on rules of paradigmatic and syntagmatic combination possibilities. In the absence of asymmetry, no markedness can be established.

64

Jadranka

Gvozdanovic

Finally, one could ask the question of relevance of the definition of markedness proposed here with respect to competence, i. e. language ability of the native speakers of various languages. It is language change which gives us crucial insights into competence, and in the case of asymmetries it shows that they can be either further extended or abandoned through a process which, at least in its initial stage, follows the direction of asymmetry if there is one present already (numerous examples of this can be found in Gvozdanovic 1985). Thus, if we take the example of the singular vs. dual vs. plural distinction mentioned above and follow its development in various Slavionic languages, we can see that the dual is abandoned exclusively by merging in meaning and in some cases also in form with the plural. This shows that the proposed analysis of the singular vs. dual vs. plural distinction in terms of two hierarchically structured correlations can be viewed as the correct one indeed. If no asymmetry has been established, then the original sign value still forms the basis for further development. An example of this can be found in the development of Dutch names of professions such as chirurg 'surgeon' which used to have no gender distinction at the beginning, corresponding with the absence of choice in the real world, this profession being a typically male one. We can assume that from the beginning on, chirurg could be used to denote either a male person of the given profession or simply a person of the given profession, depending on the context. With the allowance of women into this and comparable professions, it became necessary to express the gender difference as well, and the female form chirurge was introduced. Nowadays, a markedness relation in the sense described above holds between them, such that chirurge can be used only with reference to a female surgeon, whereas chirurg can be used either for a male surgeon or when gender is not at issue. Defining markedness on the basis of asymmetry provides a well-defined linguistic analytical tool which appears, moreover, to be relevant to the competence of native speakers. Asymmetry can occur in correlations, based on a common denominator. It is connected with the possibility of choice between the presence and absence of a given sign value associated with the form of one of the members of a correlation, and absence of such a choice associated with the form of the other member. Asymmetry is an aspect of economy characteristic of biological structuring of information, and markedness which is defined on the basis of asymmetry has the value of a descriptive label covering this aspect of structuring.

Defining markedness

65

Note 1. In contrast to Gvozdanovic 1985, I have chosen to formulate the meaning correlations differently from the resulting units; in Gvozdanovic 1985, "singular" and "dual" were used both at the level of the constituting meaning correlations and at the level of the resulting units of grammatical number.

References Barentsen, Adriaan A. 1985 'Tijd', 'Aspect' en de conjunctie poka ['Time', 'Aspect and the conjunction pokci~\ (Amsterdam). Chomsky, Noam and Morris Halle 1968 The Sound Pattern of English (New York: Harper & Row). Gvodzdanovic, Jadranka 1954 Language System and Its Change: On Theory and Testability (Berlin: Mouton de Gruyter) Herbert, Robert K. 1986 Language Universals, Markedness Theory, and Natural Phonetic Processes (Berlin: Mouton de Gruyter). Jakobson, Roman 1932 "Zur Struktur des russischen Verbums", Charisteria Gvilelmo Mathesio qvinqvagenario a discipulis et Circuit Lingvistici Pragensis sodalibus oblata (Pragae): 74-84; reprinted in Jakobson (1971): 3-15. 1936 "Beitrag zur allgemeinen Kasuslehre: Gesamtbedeutungen der russischen Kasus", Travauxdu Circle Linguistique de Prague 6,240-288; reprinted in Jakobson (1971): 23-71. 1939 "Zur Struktur des Phonems", Vorträge an der Universität von Kopenhagen (Kopenhagen); reprinted in Jakobson (1971, 2nd): 280-310. 1949 "On the Identification of Phonemic Entities", Travaux du Cercle Linguistique de Prague 5, 205-213; reprinted in Jakobson (1962) and (1971, 2nd): 418-425. 1962 Selected Writings I, Phonological Studies (Berlin: Mouton de Gruyter). 1971 Selected Writings II, Word and Language (Berlin: Mouton de Gruyter). 1971 2nd Selected Writings I, Phonological Studies (Berlin: Mouton de Gruyter). Jakobson, Roman and Linda Waugh 1979 The Sound Shape of Language (Brighton: The Harvester Press). Kean, Mary-Louise 1980 The Theory of Markedness in Generative Grammar (Bloomington: The Indiana University Linguistics Club). 1981 "On a Theory of Markedness: Some General Considerations and a Case in Point", Theory of Markedness in Generative Grammar, edited by Belletti Α., L. Brandi and L. Rizzi (Pisa: Scuola Normale Superiore): 559-604. Lass, Roger 1980 On Explaining Language Change (Cambridge: Cambridge University Press). Lessen, Kloeke, Willem U.S. van 1981 "How Strident is Raspberry? Likely, Unlikely, and Impossible Feature Configurations in Phonology", Theory of Markedness in Generative Phonology, edited by A. Belletti, L. Brandi and L. Rizzi (Pisa: Scuola Normale Superiore): 363-406.

66

Jadranka Gvozdanovic

Saussure, Ferdinand de 1916 Cours de Linguistique Generale (Lausanne/Paris: Klincksieck). Schooneveld, Cornells H. van 1978 Semantic Transmutations, Prolegomena to a Calculus of Meaning (Bloomington: Physsardt). Trubetzkoy, Nikolay S. 1939/1958 Grundzüge der Phonologie, Travaux du Cercle Linguistique de Prague 7; reprinted 1958 (Göttingen: Vandenhoeck und Ruprecht).

Markedness and linguistic change Dieter Stein

0. Introduction

The notion of markedness has recently experienced such a widening of scope that the need is being increasingly felt both to define a core common to all variant notions and to define and tighten up individual domains of the field of application. We shall address one aspect of the second task, the application of the notion of markedness to language change, especially syntactic change. At the bottom of the markedness notion applied to language change is the idea that there are preferred and dispreferred states of the grammar of a language, an idea that is a long conceptual distance away from the original structural notion. Markedness in language change implies treating (at least historical) linguistics as a prognostic science: a statement that the state of grammar or language at a given time is marked implies that there is a diachronic directionality towards unmarking. It also raises the question how, if a marked state is a kind of pathological state of a language to be in, these dispreferred states of the grammar arise in the first place. Often, there is a formal basis for classifying a given state of a language as marked. Just as in the original conception of markedness the point of departure was a formal statement as to distribution, syntagmatic and paradigmatic complexity (Eckman et.al., 1986: 3), so the markedness conceptions in diachronic syntax were based on statements of quantitative preferences such as statistical universale, as in the area of word order universale. Underlying the conjunction of markedness and diachrony is the dialectic between synchrony and diachrony in linguistic analysis. In most cases and to the extent that individual structures are concerned, interpreting a markedness evaluation diachronically involves the methodological tour de force of turning a synchronic analysis into a diachronic prediction. This is problematic to the extent that the synchronic analysis depends on a synchronically oriented theoretical tool kit. The ideal case would be a synchronic analysis that has diachrony built into it. This would then be the prototypical markedness

68

Dieter

Stein

theory. As indicated above, such a synchronic theory with diachronic vectors built into it exists only for phonology (Bailey 1973) and morphology (Dressier 1970, Mayerthaler 1981, Wurzel 1984) in an elaborated version. These approaches from outside syntax stand at the positive end of a scale of affinity towards a diachronic interpretation of a synchronic analysis. The other end is typified by diachronic interpretations of generative transformational analyses such as Lightfoot (1979) and Kiparsky (1968). The problems arising from their diachronic interpretation have been amply discussed (e. g. Romaine 1981) and have indeed served to catalyse the awareness of the dialectic between synchronic analysis and language development in the postgenerative resurgence of interest in diachronic analysis. There is therefore no need to recapitulate this discussion in detail in this place, although some points will be taken up in section 1. But, with most of the approaches to be discussed below, it must be clear that, if an approach is interpreted diachronically, this does not mean that it was originally meant to have diachronic implications or a directional diachronic implication. The purpose of this paper is to show how markedness and diachrony could be brought together to the enlightenment of each other; probing the ground to what extent markedness can be seen as an ordering principle in diachronic research, what markedness can look like in diachronic research, whether and in which ways preference in language can define the direction of linguistic change, and indicate whether there can be preferred or dispreferred modes of linguistic change, and what are the prerequisites for markedness to act as a force in linguistic change. In section 1 we shall discuss markedness conceptions based on generative grammar, section 2 will treat word order universale, section 3 will look at certain aspects of drift, section 4 will deal with evidence from syntactic variation and change, section 5 will be concerned with the origins of marked structures, while section 6 will take up questions of the heuristic and epistemological status of markedness in diachrony. Our aim being to explore the perspectives of the field, no attempt can be made to do full justice in presentation or references to the individual approaches.

Markedness and linguistic change

69

1. Generative Grammar The classical case of a synchronic theory interpreted diachronically is generative transformational grammar. Like word order universals, it is a diachronic interpretation grafted onto a synchronic analysis. As far as the status of the diachronic argumentation with this approach is concerned, two aspects stand out. The first one lies in the fact that only those types of language change can be registered which are captured by the specific nature of the synchronic descriptive apparatus. Basically, every language theory defines a theory of language change. No other theory, however, has revealed this connection so strikingly. Occasionally, the impression was created that there are models of language description with and ones without theory, where the latter is a kind of standard case and the former a case that has to be specially justified. Generative theory highlighted the fact that every language description is dependent on theory. The first generative approach to diachrony captures change from one language state to the next, without claiming diachronic directionality in a synchronic structure, as adding, reorganization and simplification of rules (King 1971). The second approach (Kiparsky 1968 and Lightfoot 1979) claims that certain constellations in the formal structure of grammar - as formulated by the theory - are not wanted, and a diachronic directionality is given by the fact that the grammar will organize itself better and differently so that this unwanted constellation will be abolished. The extension into the diachronic dimension must be regarded as follows: a primary synchronically oriented theory includes secondarily additional types of data with the aim of obtaining criteria for the validation of the synchronic theory from these secondary types of data. In this respect, language acquisition and language change have the same ancillary status. According to the famous window quotation of Kiparsky (1968), processes of language change - always within the theory - can be used as criteria for the psychological plausibility of synchronic constructs for example for the use of braces or the concept of the "natural class". Lightfoot, in his turn, wants to adduce diachronic data - again as formulated by the theory - to test the limits of the formal structures of grammars as to the amount of strain they will bear until there is remedial change. A synchronic formal structure of grammar, which is not wanted and is therefore marked, contains the diachronic prognosis of restructuring. The aspect of markedness he is concerned with is how much opacity the language can bear. There had been earlier attempts to describe language change as simplification. Language change was localized in first language acquisition: Children

70

Dieter Stein

learn the easiest grammar as the most highly valued one and thus restructure it. Additional principles that have been developed later, above all by Kiparsky (1968), are first of all the "feeding" and "bleeding" relationships of individual rules to each other. "Feeding order", for example, takes care that a rule creates structures that serve as input for another rule. Generative transformational grammar makes use of the term "markedness" as a preference concept and calls a "feeding order" optimal and unmarked. Unfortunately, Kiparsky deals with phonological processes only. An additional principle of opacity and transparency is seen by Lightfoot as a criterion for the markedness of a grammar state. The so-called transparency principle states that a synchronic grammar state is subject to reorganization if there is too much opacity and derivational complexity, or when the surface and deep structures have become too divergent. A well-known example is the disappearance of the impersonal verbs in English, like in (1)

The kyng dremed a merveillous

dreme.

With regard to this sentence type an interpretation as SVO as well as OVS is possible in the 15th/16th century: there is thus an ambiguity in comprehension. The historical development until this point is well-known: due to syncretism, many of these impersonal verbs are no longer used, while some have developed the so-called "dummy" subject it. Above all, the SVO word order is increasingly grammaticalized. To Lightfoot this means that the above sentence type is opaque, it can be analyzed with kyng as object or with kyng as subject. The grammar is opaque in this respect, not transparent and two derivations can be related to one surface structure. The marked state of grammar predicts directionality towards the abolishment of such structures to create transparency. In English this prediction has come true: a structure of this nature can only be interpreted as SVO. This argumentation has been criticized on a number of grounds (cf. Romaine 1981), both concerning the analysis of this particular case and on general grounds. In the present context, only a few points can be mentioned. The restructuring of abstract grammar is conceived as a sudden event. The relationship to the graduality of syntactic change processes that is factually observed is unclear. It is uncertain when this point is assumed, as it is not recognizable on the surface. An additional problem is that no parallel process is observed under similar circumstances in other languages. If syntax and syntactic language change were as autonomous as Lightfoot assumes this should have been expected. Other theories do not have this problem: there, syntax ist not autonomous, but pragmatic and sociocultural factors, for

Markedness and linguistic change

71

example, are present as explanatory dimensions. One could also point out that opacity is only a problem for relatively decontextualized written language. Of particular importance for our argumentation is the nature of opacity. As a criterion for the psychological plausibility of this approach, the diachronic vectors must be present in the linguistic competence of the individual and as a factor in the realization of the individual and utterance. At the point of the decision about the shape of the individual utterance the speaker must consider whether a particular choice will be one which eventually makes for a more transparent grammar. Being a process grammar, an important point is to what extent the form of the postulated process is isomorphic to what actually happens in production. If a certain configuration of the processes is claimed to be marked or unwanted, then an ontological problem emerges: unwanted for whom or what? If these processes and categories do not have a psychological reality, the question is how and where the locus for this unwantedness should be located. In other words: if markedness or dispreferredness is supposed to be sensitive to something that does not have psychological reality, but is a construct of descriptive synchronic linguistics, then how can it be an operative force in linguistic change? This points to the central difficulty of diachronic interpretations of originally synchronic concepts: the more purely descriptively synchronically orientated they are the less explanatory in terms of actual language behavior and the greater the problems in their diachronic interpretation. It seems, then, that a central and general problem of diachronic directionalities is the extent to which system-teleologies can be translated to the level of the decision by the individual speaker in the individual utterance. To the extent that they can be made plausible at this level they may have predictive value.

2. Word order universals

The type of directionality in the synchronic language which is most widely discussed at present concerns the so-called word order universals, which are above all linked with the name of Greenberg. These universals have the logical form of synchronic implicational findings that concern the position of verb, subject and object or complement as well as of other elements in the sentence. There are a number of languages, for instance, which have the verb in

72

Dieter Stein

sentence-initial position. Languages of this type have prepositions to mark the syntactic relationship between the nominal phrases. The reverse conclusion is not true: English does have prepositions, but is not VSO. On the one hand, languages with final verb position typically have postpositions, i. e. function words which follow their NP. VSO languages have auxiliaries preceding the main verb while with SOV-languages the auxiliary follows the main verb. Admittedly, the relative position of verb and object as well as the position of prepositions is certainly a main criterion, but there are also implicational statements about other sentence elements, such as the relationship of the genitive-modifier with its head on the one hand and pre- or pospositions of function words on the other. Where the genitive-modifier follows its head it is highly probable that the language has prepositions like in French: (2)

la plume de ma tante

'the pen of my aunt' Vice versa, languages with the genitive-modifier before the head have generally postpositions, like in Japanese: (3)

Taro no hon

'Taro his book', where no is a postpositive functional word and Taro no the genitive-modifier. These word order implications are one of several possible parameters of the typological classification of languages. In terms of word order, languages are divided into the basic types of SOV, VSO and SVO with implicational consequences with regard to the order of other elements of the sentence. The SVO-type, which includes the English language, is the laxest one as far as the implications for the position of other elements in the sentence are concerned. It is true that English has prepositions but there are also genitive-modifiers before the head in this language. There are constructions like (4)

the chairman of the workshop

with postmodifying position, but also constructions like: (5)

John's book

where the genitive s is postpositive. These word order universals are first of all descriptive statements on cooccurrence of linguistic forms which are too frequent to be produced by accident. It is very significant for our argumentation that we are dealing with statistical and not with absolute universals. Word order universals are relative and statistical, open to exceptions and inconsistencies. It is presumed that natural or unmarked word order patterns exist and that languages without the effect and interference of sociocultural,

Markedness and linguistic change

73

external factors like language contact or standardization would develop diachronically in their direction. Apart form the binary dimension "possible" or "not possible" a concept of "markedness" must exist which generally designates the occurring preference structures of co-occurrence that clearly exist in every language (Vennemann 1983). In general terms, this version of a markedness concept implies that a word order structure that is not impossible but is dispreferred has a tendency to become regular and common. The synchronic approach of Greenberg has been refined by Hawkins (1980) who says that all languages put the modified and the modifying element basically in the same order: (6)

AN GN where: A = adjective G = genitive Ν — nominal head

Hawkins' refinement of Greenberg's universale roughly means that a language with the verb at the end of the sentence puts the modifying element before the modified one: (7) SOV (AN GN) SOV -+ (postposition) where: means "includes" A language with this type of word order and modification relationships is fully "harmonious". As a SOV-language it has postpositions. In showing the same modification structure in different syntactic structures (subject, object, genitive) it is "cross-categorically" harmonious. This principle of "crosscategory-harmony" means in principle the same as Vennemann's principle of "natural serialization" which refers to the linear order of modified and modifying elements. Vennemann calls these elements "operand" and "operator". If the same order of operand and operator exists in all parts of the sentence one has natural serialization. From this principle one can derive the prediction that all languages which do not show harmonious serialization are therefore marked and diachronically tend towards serialization. For Vennemann all languages with synchronic disharmonious structures have the diachronic tendency towards a harmonious structure: (8)

SOV where: AN -*• GN.

A disharmonious structure is a marked structure, a structure which is not preferred by languages. If we look at word order universale from the perspective of the psychological reality sketched at the end of section 1, the tendency to abolish marked

74

Dieter

Stein

structures seems a plausible agent. If can be argued that the tendency towards a unitary strategy of comprehension (i.e. towards expecting the operand always before or always after the operator) is indeed psychologically plausible. A harmonious language would basically require a smaller effort of comprehension and would be less complex than languages where, for example, the modified element appears in some cases before the modifying element and in some cases behind it. It should be noted that markedness based on word order universals is a factor for linguistic change only as long as there are marked, non-harmonious, inconsistent languages.

3. Drift

The previous discussion of word order development was based on an assumed tendency of languages to be "consistent". The same psychologically interpretable tendency may be assumed to be at work in the phenomenon of "drift", an area which is not normally discussed under the heading of markedness and diachrony. It could be argued that drift may be interpreted as the manifestation of a tendency of a language to be consistent and by that token achieve an unmarked state. There is in fact a close connection between the concepts of drift and word order universals, as pointed out by Comrie (1981: 204if.). Apart from that, drift is similar to diachronic markedness conceptions in illustrating the central problem for diachronic markedness conceptions that was addressed in sections 1 and 2, namely how the individual speakers "set themselves a target" (Comrie, 1981: 206) and follow it in the individual speech act. To Yakov Malkiel we owe an extensive definition of "drift", as well as the terms "slant" and "slope", which we will follow in the present discussion. For Sapir, The drift of a language is constituted by the unconscious selection on the part of its speakers of those individual variations that are cumulative in some special direction (Sapir 1921: 22). Language moves down time in a current of its own making (Sapir 1921: 25). The basic features of "drift" according to Sapir are the following: (a) language change is internally controlled, more precisely: not controlled by

Markedness and linguistic change

75

sociocultural factors such as linguistic contact, prestige or standardization, but language-internally: "a current of its own making"; (b) the processes are slow, gradual, over several hundred years, of unconscious character and limited to one single directionality. (a) (b) (c) (d)

Sapir gives the following examples for "drift": whom becomes who deletion of the formal difference between subject and object substitution of case inflections by word order increasing consolidation of word order.

Subsequently, the term "drift" experienced an expansion and came to mean parallel development in genetically related languages, which Malkiel suggests to call "slant". Among other things, Malkiel gives the example of the derivational affix in -ese (Bostonese, journalese), which is as productive in contemporary English as -eis, or -ois in Old French (cf. lourdois, mocois). In Chaucer we have goliardeys, "a professional jester" ("He was a janglere and a goliardeys", Prologue A 560), deriving from goliardeys and from gula. An additional possible candidate for slant is the variable deletion of the inflectional dental suffixes (t/d) in different Germanic languages - English, Swedish, Dutch, which can be quoted as a real panchronic rule. Finally, phenomena like the innovative occurrence of the same linguistic form in different historical periods discussed by Diller (1985) can also be called a slant of the English language in its historical continuity, if this term is to be extended to different periods of the same language. The further history of the term is well known. Robin Lakoff extended the concept to a development from synthetic to analytical and focused on a number of sub-phenomena like definite and indefinite articles, the development of periphrastic forms instead of inflections, the use of prepositions instead of case endings and the development of anaphoric pronouns in subject function. The obvious question is how the correlation of phenomena with parallel directionality is controlled. Lakoff calls this central steering force a "metacondition" but does not explicate it. There are some aspects of drift which seem relevant to the points raised in sections 1 and 2. First of all, "drift" - in its original as well as in its later, extended conception - obviously deals with the directionality "of language", with the developmental vectors given "in language". If "drift" is a gradually oriented long-term change, every synchronic speaker must - in Chomsky's terms - somehow have a certain "knowledge" of the evolutionary direction. All speakers have to chose the same varieties, otherwise no common collective direction can result (after all we have to visualize language use as realization

76

Dieter

Stein

with considerable variation in performance). The average of these performance fluctuations only changes when the majority of realizations shows a tendency towards the same direction. In order to be able to explain the phenomenon of language change one has to explain the general tendency (of drift) in the individual utterance. A change of "language" as "language" abstraction can only result from the fact that the individual speakers have to chose the same variants. This is how the explanation of the phenomenon of drift highlights a central problem of diachronic interpretation of the markedness conceptions: Some kind of "pressure" must be unconsciously felt by the speaker to realize an utterance in a certain way. In principle, if this "pressure" or the forces behind the markedness status of a structure (its functional basis) are present in every individual, the change is polygenetic. Once a group of speakers has completed a minimal innovatory step, one has reached the center of gravity for social diffusion, the "slope" in Sapir's terminology, the final point of which is a change of langue. But exactly what kind of functional basis is involved in the case of drift phenomena, e.g. whether it can indeed be reduced to unitary perceptional strategies, as in the case of word order universals, is much less clear (Itkonen 1982: 100).

4. Syntactic variation

So far diachronic prediction pertained to the developmental direction of the system. As such it was a case of diachronic interpretation of markedness implying a long-term directionality. In what follows I will discuss a case involving directionality not as the developmental tendency of the structure of a syntactic subsystem as a whole, but of a variational pattern. The developmental stretch does not represent the future patterning of the whole subsystem but the way a subcategorization pattern of variable realizations evolves. It is by now a truism that syntactic change is gradual and variationally finely structured in its transition from one state of langue to the next. I will argue that it is possible to extend the term markedness in such a way as to refer to a dispreferred variational structure with inherent diachronic vectors. The process at issue is the variable realization of the verb phrase in Early Modern English in finite form and with do, go you vs. do you go in questions (and other

Markedness

and linguistic change

77

Wh-questior>s 100·/.

507.-

10·/.-

"Ί dst

1 weak

^V— past

1

Γ

strong

J

st

you

V.

ν

J

present

Figure 1: Periphrasis frequency in w/i-questions in Early Modern English texts.

78

Dieter

Stein

100%—

50%

10%—

strong verbs

weak verbs

dst / didst

you • d/did

—ν— past

d/did

st/dost J

\

you + 0/do

0 / do

ν

s/doth J

present

Figure 2: Periphrasis frequency for questions in Shakespeare.

syntactic contexts, which are not at issue here). Fig. 1 represents one aspect of empirical data presented in Stein (1985). For a number of Early Modern English corpora (roughly 1550-1700, including Shakespeare and other roughly contemporaneous corpora) it gives the relative frequencies of whquestions with do in relation to their realizations in finite forms. It should be read for example like this: for Dekker (corpus nr. 13) periphrasis frequency is

Markedness and linguistic change

79

75 % for the person marked preterite with weak verbs (thou received'st vs. thou did'st receive, represented in Fig. 1 by dst), 55 % for the not person marked preterite with weak verbs, 15% with strong verbs, 55% with the person marked present (thou dost sing, represented by st), 30 % for the present with you, and 10% for 3rd person singular (represented by s). There are many aspects of this graph which could be discussed (and which are discussed elsewhere), but the present discussion will focus on two: the effect of phonotactics on periphrasis frequency and the subcategorizational differentiation. The effect of phonotactics on periphrasis frequency can be seen by comparing the periphrasis frequencies of the individual tense/person categories; the size of the consonantal cluster that would arise with the finite form being an important factor determining periphrasis frequency. At this point it must suffice to point out that there are numerous ways of demonstrating the dramatic effect of the phonotactic factor (e.g. quality and size of cluster arising through suffixation with st, the ending cooccurrent with thou). The effect can also be demonstrated by showing periphrasis frequencies of the individual tense/person categories without syntactic subdifferentiation as in Fig. 2. Do is used here as a syntactic strategy to avoid word-final consonantal clusters which can be considered phonetically marked structures while the processes that simplify or abolish them can be considered strategies to reduce phonological markedness, such as those arising through the reduction of the vocalic part of inflectional endings in English. The use of do, or one of the functions of do, was to prevent phonologically marked structures. If the prevention of phonologically marked structures is treated as trigger for theriseof a finely subcategorized realization of questions, we get markedness at another linguistic level. If one linguistic category - w/i-questions - is realised differently in different tense/person categories we have a marked situation at the level of form/function associations. Again, there is a motivation to level out the differences, "economy" in form/function association, which is indeed what happened. Though "economy" has experienced a certain inflation, there can be no doubt that it is a major force in linguistic change (RonnebergerSibold 1980). Looking at the process as a whole, we notice two diachronic directionalities, both made up by marked structures. The strategy employed to get rid of the diachronically earlier one on the phonetic level creates a situation that is marked in the perspective of economy of form/function associations. The whole process is then a case of markedness reversal with diachronic directionality in the area of variability. The directionality is not long-term in the sense of remedying a pathological state of the system, rather it

80

Dieter Stein

is short-term. In addition, there is no formal basis of markedness as in the cases discussed in section 1 and section 2, but appeal is made to general behavioral principles which are larger than the linguistic ones and which can therefore be assumed to be present in every speaker and thus present as a factor in the realization of every utterance. This is why these principles may be termed "natural" factors.

5. Origin of marked structures

If diachronic development goes from marked to unmarked then the question arises how marked structures originate in the first place. Basically there are two ways: language internally and language externally. If we wish to maintain a heuristic distinction between biological, psychological and more strictly language internal forces - whatever their nature may be - on the one hand, and forces that may be termed societal in the largest sense (such as prestige, acts of identity) on the other, then we may call the first group internal and the second group external, corresponding to nature and nurture. The classical theory of the rise of marked structures in languages is the analytic - synthetic cycle in its various forms (e.g. Hodge 1970), which is too well known to require detailed recapitulation here. Lüdtke (1980) may be regarded as the hitherto best elaborated version of this type of theory which sees language development as a continuous process where the abolishment of marked structures on one level (e.g. "difficult" phonetic structures and too much articulatory effort are marked) leads to marked structures on another level: it is dysfunctional for languages to lose, by phonetic attrition, too much sign material. From the point of view of articulatory effort it is seen as marked because it is uneconomical. Let us remember that when we discussed marked structures at the syntactic level the variational case of section 4 was interpreted as caused by phonetic (phonotactic) markedness, triggering a relief strategy to avoid it. Even within "individual compartments" it is to be assumed that unmarking on one level may cause marked structures on another (cf. Bailey 1973). In terms of its mechanics and functions language is "built" in such a way that at any historical moment there are of necessity bound to be marked structures at some level. The cost for abolishing marked structures on one level is the creation of marked structures on another level. Cyclicity in word order developments is another aspect of the built-in tendency to create

Markedness

and linguistic change

81

marked structures. What is here at work are the language internal resources of marked structures and unmarking tendencies. The second major source of marked structures is of course borrowing on all linguistic levels, which makes language "inconsistent". I will briefly discuss a case from English historical syntax, where it can be argued that borrowing introduces into the language an element of markedness, with concomitant directionality. The discussion is based on Dekeyser (1984), who in turn refers to Keenan and Comrie's (1977) Accessibility Hierarchy, which predicts that subjects, direct objects, indirect objects, obliques, genitives and objects of comparison - in that order - are less accessible to relativization, i. e. from left to right are increasingly marked with respect to relativization strategies. This is a typical descriptive, statistical universal as a formal basis of markedness. Among the predictions associated with it are those concerning frequency in discourse in general and frequency of types in the directional rear end of the list; it is concluded that in relatively more complex discourse relativized obliques or indirect objects are more frequent than in less complex discourse (Keenan 1975). These synchronic findings could be used to derive a diachronic dimension in the following way: if a language acquires a new relativization strategy we would expect it to enter at the least marked point and then work its way up towards more marked ones. This path of innovation could be called an unmarked type of linguistic change. In the case at hand, in Early Modern English (between 1500 and 1600) several available sets of data (cf. Dekeyser 1984, section 5) show that the wh series of relative pronouns, which is available from the 12th century onwards, starts at the marked end of the accessibility hierarchy {who is slowest and enters the language in the most complex styles (Romaine 1980). By contrast, there is evidence that the 0-relativization strategy proceeds in English from the "proper", unmarked end. The 0-strategy therefore does not require comment. But the fact that the w/j-strategy reverses the proper order may be seen as symptomatic for borrowing, in this case from Latin and French. The indigenous 0-strategy starts at the unmarked end, but borrowing from a nonGermanic language leads to a marked local state of markedness, illustrative of extrinsic sources of markedness. A well-known example from morphology of a marked structure that must make reference to societal factors in the explanation of its existence, i. e. in the explanation of its evolution, is the third singular marker in what is now Standard English. From a universality point of view this is a highly marked feature (e.g. Markey 1987) since the other two persons have no person markers. Its synchronically highly marked status defines both directionality of change predictions and an a posteriori explanandum for diachronic analysis.

82

Dieter

Stein

6. Marked and unmarked linguistic change In the discussion of the relativizer example, a path of innovation that starts from the least marked end of the accessibility hierarchy was itself termed unmarked. This constitutes a further development in the application of markedness in diachrony, and a further extension or metaphorization of the term. Up to this point the application of markedness to diachrony meant interpreting synchronic preference structures as diachronically directional. At this point in the discussion of the relativizer example the metaphor is taken yet one step further: the domain of markedness is the mechanism of change or the diffusion through linguistic context itself. In other words, extending markedness in diachrony from a preferred or dispreferred state of grammar as the teleological, synchronic endpoint of a change process to diffusional modes involves claiming preferred or dispreferred modes of diffusion. The example from the spread of do may be an example of an unmarked path of evolution of a new state of grammar. If we stick to the distinction between external and internal forces in linguistic change, the unmarked type of linguistic change would be characterized as being driven by internal forces. Linguistic evolution would be marked to the extent that it is influenced by external forces. If we express this definition in naturalness terms the marked (externally steered) linguistic change would be unnatural. The spread of the wh series of relative pronouns would be a case of marked (unnatural) linguistic change, the spread of do in questions would be an unmarked (natural) one. Logically, the abolition of markedness as a system-teleological concept is separate from the (marked or unmarked) way in which the innovation arises, but it is inconceivable that, given synchronically marked results of a change (such as 3rd person s in English), an external influence should not have been at work (i.e. should not have at least a certain share in marked or unnatural development). Also, many change processes may well be societally triggered, but their internal diffusional directionality may be internally motivated. Possible candidates here are vowel chain shifts such as the English one. The application of markedness as developed here defines a heuristic and methodological priority for diachronic research: external factors (marked change) are called upon as explanation only at that point at which the assumption of internally determined change leads no further: it is assumed that externally determined (i.e. marked) change is the dispreferred mode of explanation, i. e. the default mode of explanation, which is carried through as long as there is no reason to consider an externally motivated one. It is very important at this point not to confuse the methodological or heuristic level

Markedness and linguistic change

83

with the level of facts. Establishing a heuristic precedence for the search for internal factors does not mean that these factors are in any sense more important or more frequently operating in linguistic change but that, given the presumed, more erratic and unpredictable character of societal forces, it seems at the moment preferable to assign internally determined change an unmarked status. The study of these internal factors as language-inherent redundancies is of no less interest than the study of societal factors, especially if the former can be correlated with results of studies of such inherent redundancies of domains other than diachrony (such as first language acquisition, cf. Wode 1984). The study of unmarked linguistic change can provide important access to the nature of the internal forces if there are only purely internal change processes and if the effects of internal and external factors can be segregated in cases in which the two interact. The segregation of the factors constitutes a major problem if one takes into consideration that normative influences, standardization and written language add to the range of societal factors. The hope to eliminate those additional societal factors is certainly one aspect that motivates much work that is being done on material prior to the appearance of these factors. As it will have become obvious from the previous discussion, the problems in applying the markedness concept to diachrony are many-faceted. They range from the general question of transferability of originally synchronic concepts to diachrony to the difficulty of seperating internal from external factors. In between are the "normal" difficulties of segregating the interaction of the effects of the individual internal factors. This is connected with the problem that the synchronic markedness concepts are defined locally and the "limits of tolerance" (Moravcsik 1978) for a synchronically marked state must be felt as pressure for remedy by the individual speaker in the individual act of shaping his utterance. In Lüdtke's global concept of linguistic change, it links with determining how much sign material must be eroded for the nesessity for replacement to be felt to prevent dysfunctionality. But the most important criterion for the success of the markedness approach to diachrony, its explanatory and predictive power and its insights through change into language, is whether or not the factors and categories postulated can claim to have psychological reality, or whether the markedness statements can be described on a psychologically interpretable functional basis. Finally, a point common to all applications of markedness to diachrony is its negative definition: if directionality can indeed be derived, it abolishes structures, rather than strives for "beautiful" ones. This is why diachrony based on markedness establishes not teleologies, but finalities. The evolutionary aim is not to achieve a pre-defined ideal structure at which point linguistic

84

Dieter Stein

evolution would stop, but to abolish disfunctive or (for whatever reason) dispreferred structures. Markedness defines not what it wants but what it does not want. By providing relief on one level it creates problems at another - which keeps languages changing and linguists busy.

References Bailey, Charles-James 1973 Variation and linguistic theory (Arlington: Center of Applied Linguistics). Comrie, Bernard 1981 Language universals and linguistic typology, syntax and morphology (Oxford: Basil Blackwell). Dekeyser, Xavier 1984 "Relativizers in Early Modern English: a dynamic quantitative study", Historical Syntax, edited by J. Fisiak (Berlin: Mouton/de Gruyter): 61-87. Diller, Hans-Jürgen 1985 "Archaismus und Innovation in der englischen Literatursprache", Anglistentag 1984 Passau, edited by Μ. Pfister (Glessen: Hoffmann), 116-128. Dressier, Wolfgang 1970 "Towards a semantic deep structure of discourse grammar", CLS VI: 202-209. Eckmann, Fred R. - Edith Moravcsik - Jessica R. Wirth (eds.) 1986 Markedness (New York: Plenum). Greenberg, Joseph 1963 "Some universals of grammar with particular reference to the order of meaningful elements", Universals of language, edited by J.J. Greenberg (Cambridge): 73-113. Hawkins, John 1980 "On implicational and distributional universals of word order", Journal of Linguistics 16:193-235. Hodge, C. 1970 "The linguistic cycle", Language Sciences 13: 1 - 7 . Itkonen, Esa 1984 "Short-term and long-term teleology in linguistic change", Papers from the 3rd International Conference on Historical Linguistics, edited by J. Peter Mäher et al. (Amsterdam: John Benjamins B.V.). Keenan, Edward L. 1975 "Variation in universal grammar", Analyzing variation in language, edited by R. W. Fasold and R.W. Shuy (Washington: Georgetown University): 136-148. Keenan, Edward L. - Bernard Comrie 1977 "Noun phrase accessibility and universal grammar", Linguistic Inquiry 8: 63-99. King, Robert 1971 Historische Linguistik und generative Grammatik (Frankfurt: Athenäum). Kiparsky, Paul 1968 "Linguistic universals and linguistic change", Universals in linguistic theory, edited by E. Bach and R.T. Harms (London: Holt, Rinehart and Winston), 171-204. LakofF, Robin 1972 "Another look of drift", Linguistic change and generative theory: Essays from the UCLA Conference on Historical Linguistics in the Perspective of Transformational

Markedness and linguistic change

85

Theory, February 1969, edited by R.P. Stockwell and R.K.S. Macaulay (Bloomington): 172-198. Lightfoot, David 1979 Principles of diachronic syntax (Cambridge: University Press). Lüdtke, Helmut 1974 "Sprachwandel als universales Phänomen", Kommunikationstheoretische Grundlagen des Sprachwandels, edited by Η. Lüdtke (Berlin: Mouton de Gruyter): 182-252. Malkiel, Yakov 1981 "Drift, slope and slant", Language 57: 535-570. Markey, Thomas L. 1987 "English -s vs. -th in the third person singular: historical contrasts and cross language argumentation", Paper presented at the XXIIIrd International Conference on Contrastive Linguistics Poznan, Poland, Mai 5th-7th 1987. Mayerthaler, Willy 1981 Morphologische Natürlichkeit (Wiesbaden: Athenaion). Moravcsik, Edith 1978 "On the limits of the subject-object ambiguity tolerance", Papers in Linguistics 11, 1-2: 255-259. Romaine, Suzanne 1980 "The relative clause marker in Scots English: Diffusion, complexity, and styles as dimensions of syntactic change", Language in Society 9: 221-247. Romaine, Suzanne 1981 "The transparency principle and why it doesn't work", Lingua 55: 277-300. Ronneberger-Sibold, Elke 1980 Sprachverwendung - Sprachsystem. Ökonomie und Wandel (Tübingen: Niemeyer). Sapir, Edward 1921 Language: An Introduction to the Study of Speech (New York: Harcourt, Brace and World). Stein, Dieter 1985 Natürlicher syntaktischer Sprachwandel (München: Tuduv). Enlarged English Version to appear in the series Trends in Linguistics. Vennemann, Theo 1975 "An exploration of drift", Word Order and Word Order Change, edited by Ch. N. Li (Austin: University of Texas Press). Vennemann, Theo 1983 "Causality in language change. Theories of linguistic preferences as a basis for linguistic explanations", Folia Linguistica Historica VI/1: 5-26. Winter, Werner "Thoughts about markedness and Normalcy/Naturalness", this volume, pp. 103-109. Wode, Henning 1984 "Überlegungen zur Markiertheit aus psycholinguistischer Sicht", Anglistentag 1983. Vorträge, edited by J. Schläger (Glessen: Hoflfmann). Wurzel, Wolfgang U. 1984 "Flexionsmorphologie und Natürlichkeit", Studio Grammatica XXI (Berlin Akademieverlag).

Markedness, sound change and linguistic reconstruction Thomas V. Gamkrelidze

0. Introduction

The undisputed dominance of problems of synchronic linguistics in the first half of the 20th century was followed in its second half by a growing interest in diachronic linguistics - in problems of linguistic change and transformation in time. Thus there seems to be a return - on a new methodological basis - to the treatment of problems that arose in the classical, comparative historical IndoEuropean linguistics of the past century. This enhanced concern for problems of linguistic change and diachronic linguistics stems from the general development of linguistic thinking in recent decades: Overcoming the Saussurean antinomy between synchronic and diachronic linguistics, it seeks to build a linguistic theory that would have greater explanatory power in comparison with the purely synchronic theories of descriptive taxonomic grammar, built strictly on the basis of empirical language data. One of the basic assumptions of language reconstruction, and of comparative-genetic linguistics in general, is the thesis that language change should not be conceived as a movement of language from a simple to a more complex or perfect form, but rather as a diachronic variability of language, as its ability for diachronic transformations at all levels of language structure. On the phonemic level, such diachronic transformations are manifested in the change of certain phonemes into others, essentially involving split or merger of phonemes or phonemic series that had been characteristic of an earlier language stage. Phonemic transformations occur under conditions of redundancy of a language system. Redundancy as an "incomplete system" constitutes the main structural factor that makes possible sound variability in a language. Thanks to it, language is not a fossilized structure resulting from sound changes and diachronic movements of phonemes, but a dynamic system with

88

Thomas V. Gamkrelidze

diachronically varying structures. However, the character of the diachronic movements of phonemes, i. e. the character of the sound changes in a system does not depend on the degree of redundancy of the system but on the hierarchical relationship of "markedness" or "dominance" between the linguistic units in general and the phonological units in particular.

1. Stratification of phonological values 1.1 The scale of markedness There exist universal patterns of combinability (compatibility) of distinctive features in a simultaneous (vertical) sequence or bundles representing definite phonemes. Some features combine with each other on the axis of simultaneity in preference to others, as manifested in the high systemic and textual frequency of the phonemes represented by these features; other features combine in a single bundle in a more limited manner, as shown by a lower relative frequency of the phonemes having the given distinctive features. To the latter group belong the empty slots, the gaps in the paradigmatic system, which can be considered as instances of maximally difficult combinations of features. It is the variable capacity of distinctive features to relate with one another into simultaneous combinations or "vertical sequences" that creates bundles of features differing in character and possessing a varying degree of "markedness", i.e., combinations of features characterized by commonness, naturalness, high degree of occurrence in the system ("unmarked") and less common, less natural combinations of features manifesting a lower degree of occurrence ("marked") (cf. Gamkrelidze 1975). Since the capacity of distinctive features to combine with one another in a simultaneous bundle varies, it is feasible to set up a scale of "markedness" of simultaneous (vertical) combinations of features. The opposite extreme values on such a "scale of markedness" involve: (a) obligatory combination of the distinctive features on the axis of simultaneity, i.e., maximally „unmarked" combinations (e. g., combinations of features like [ + syllabic, — nonsyllabic], [ — syllabic, + nonsyllabic] or [ +discontinuous, + dental] which are represented in any phonemic system, being a constituent part of the phonemes that make up the minimal phonemic inventory of language, and (b) noncombinability i.e. mutual incompatibility of features potentially forming maximally

Markedness, sound change and linguistic reconstruction

89

"marked" combinations (e. g., the features of [+glottalized] and [+voice] or the features [+nasal] and [+fricative], which are incapable of combining into simultaneous bundles. Between such extreme values of "markedness" we have all kinds of possible combinations of distinctive features with varying degrees of „markedness" with a greater or lesser approximation on the extreme values, reflecting the varying capacity of the distinctive features to combine with one another in simultaneous bundles. Such a "scale of markedness" of combinations of distinctive features must, in principle, be characterized by a fairly high degree of universality, for it reflects the capacity of definite phonetic and acousticarticulatory properties to combine more or less freely and form simultaneous articulatory complexes. Because of their acoustic-articulatory peculiarities, the definite phonetic features, combine preferably on the axis of simultaneity. The "marked" bundles of features - in contrast to the "unmarked" ones reflect the limited capacity of given phonetic features to take part in simultaneous combinations, i.e. their lesser tendency to combine with one another. Hence such bundles represent less usual or less natural combinations of features and are placed closer to the maximal value of the "scale of markedness". It is but natural to expect that such bundles (and the phonemes represented by them) will be characterized by a lesser degree of actualization than will the features which, due to their acoustic and articulatory properties, combine easily with each other and thus represent natural or usual combinations of features. The former bundles of features (and, correspondingly, the phonemes represented by them) constitute functionally weak units in the system, with low degree of occurrence (frequency) and with distributional limitations, or are entirely absent and form gaps in the paradigmatic systems of given languages. The latter bundles are more common and natural and, in this sense, "unmarked"; they consist of functionally stronger units of the system, which are characterized by greater distributional freedom and higher degree of occurrence (frequency) - some of them, the maximally "unmarked" combinations of features, have a probability of occurrence equal to one. Thus, some distinctive features combine readily with one another, while others form more complex bundles in terms of artuculation and perception. Being less optional, the latter bundles of features are of limited occurrence in the system - they form less natural phonemic units, which are characterised by lower frequency of occurrence - in certain systems equalling zero and thus yielding phonemic gaps in paradigmatic patterns.

90

Thomas V. Gamkrelidze

1.2 The relationship of paradigmatic dominance The phonemic units representing stable and "unmarked" bundles of features may be characterized as "dominant" as opposed to the less common and less natural (i.e., "marked") units of the system that may be referred to as "recessive". Thus, any two phonemic units which are opposed to each other in the paradigmatic system by the hierarchical relationship of "markedness" may be characterized as "dominant" vs. "recessive", while the relationship of "markedness" itself, implying a dependence between these units, may be referred to as the relationship of "paradigmatic dominance". 1 Such a change of terms and the substitution of "dominant vs. recessive" for "unmarked vs. marked" seems to solve the problem of ambiguity of the notion of "markedness" and its still widespread use in the original sense of "merkmalhaltig/merkmallos" (as differing from that of "common, natural" vi·, "less common, less natural"). It is precisely the establishment of universal patterns of compatibility of distinctive features into simultaneous bundles or into "vertical sequences", with determination of their opposite function of "dominance" in the paradigmatic system that appears to be one of the basic tasks of present-day typological phonology. This will help establish universally relevant hierarchial dependence between the correlative units of a phonological system and to identify the core of phonemic oppositions - a kind of deep phonological structure, that constitutes the basis of the phonemic inventory of human language, invariant in relation to particular phonemic systems in synchrony and the possible phonemic transformations in diachrony. The presence of hierarchical dependences between individual phonological units or between bundles of distinctive features, reflected in the relationships of dominance, testifies to the strict stratification of phonological values in the language system. It is in conformity with such universally valid correlations that diachronic phonemic transformations occur in a language. A number of diachronic phonemic changes in the system which may appear disparate and unrelated - may be conceived of as interdependent, mutually conditioned transformations regulated by the above mentioned hierarchical dependances of phonological values. In particular, the observable dominance of the front series of voiced stops and fricatives as opposed to the back series and, conversely, the dominance of the back series of voiceless stops versus the front series on else the overall dominance of stops with respect to corresponding fricatives, etc. permit to determine the sequence of phonemic changes, in particular language systems, and to establish universally relevant models of diachronic phonemic transformations.

Markedness, sound change and linguistic reconstruction

91

A universally valid hierarchy of phonological values presupposes - as noted above - the presence of phonemes with low relative frequency, reaching zero (empty slots in the phonemic system). These paradigmatic regularities of the system should be constantly borne in mind both in a synchronic description of languages and in language diachrony, particularly while reconstructing earlier language stages. The presence of an empty slot in the paradigmatic system is not an anomaly from the viewpoint of the theory of dominance, and hence it does not imply of necessity whenever comparative data are lacking. 2

2. Reconstruction 2.1 On the interpretation of the correspondence between the sounds of different languages Another basic assumption of comparative historical linguistics is the thesis of the arbitrariness of the linguistic sign. Although the arbitrariness thesis ought to be dealt with in a somewhat different manner from that presented by de Saussure and his followers 3 and ought to be argued in the light of the "principle of complementarity", the relations between the "significant" and the "signifie" may be considered arbitrary in the Saussurean sense (cf. Gamkrelidze 1974). When in two or more languages one encounters similarities in the forms and meanings of the signs, i. e. similarities at the planes of substance and content of the signs, the question naturally arises about the causes of these similarities. If one accepts the thesis about the limited arbitrariness of the linguistic sign, the formal closeness between the signs of different languages (i. e. the phonetic similarity of two or more signs with close or identical meanings) may be interpreted as chance coincidence of two or more signs of different languages. The probability of the hypothesis of chance coincidence for accounting for this kind of similarity will decrease with the growth of the number of languages exhibiting such similarities, the decrease being greater with the growth of the number of signs in these languages which manifest such sound similarities or coincidences. The attempt to account for this kind of formal similarity by historical contacts between languages and by borrowing of words from one language into another or else by loans into two languages from a third source leads to better explanation of the formal coincidences of the respective signs. However, not all types of formal-semantic similarities in

92

Thomas V. Gamkrelidze

the features of two or more languages can be ascribed to loans. There is a type of formal-semantic similarity between the signs of different languages which is manifested in the existence of regular sound correspondences between similar signs of these languages; this type of formal-semantic similarity cannot, generally speaking, be accounted for through borrowing or loans. The sole plausible explanation of a similarity of this type is the assumption of common origin of the language systems under consideration, i.e. the assumption that these systems were derived from a certain common parent language, which in the course of time transformed in various directions and thus resulted in different but historically related languages. An interpretation of the sound correspondence between different languages through the assumption of common origin from a certain parent language speaks in favour of a reconstruction which accounts for the origin and derivation of historically attested related language systems. A comparison of languages that attempts to establish regularities of sound correspondences between these languages must logically lead to the reconstruction of the language model whose transformations in different directions yielded the historically attested related language systems. A comparison of related languages that does not aim at the reconstruction of the parent language system cannot be considered to be the final stage of research into the history of the languages under consideration. A genetic comparison of languages without reconstruction is, as Saussure put it, sterile by its nature.

2.2 Diachronic transformations The unrecorded history of related languages can be reconstructed only if the entire diversity of historically attested language structures proves to be reducible to common initial models. The reconstruction of the origin and development of the systems under consideration should begin with the initial state and should go back to the historically attested language states. This approach to the genetic comparison of languages naturally raises the question of the possibility of the reconstruction of the initial, so called "parent language systems" and of the linguistic methods used for it. A reconstruction is actually achieved through juxtaposition of historically attested related languages and through retrospective movement from a given language state to earlier ones. Retrospection should continue until language state is reached out of which all the historically attested related language systems can be derived through a definite set of consecutive structural transformations, which lead from the initial language system to later language

Markedness, sound change and linguistic reconstruction

93

states, resulting from its structural differentiation. These transformations may be called "diachronic". The explanatory power of the diachronic transformations, which derive the historically attested forms of a language from definite theoretical constructs, considered to be chronologically earlier stages of these forms - their "archetypes", may be compared to the transformations of TG grammar, which generate the observable structures of a language from theoretically postulated basic constructions, constituting its deep structure. A description of the diachronic changes of a language through transformational rules essentially ammounts to a consecutive listing of discrete steps, each correlating with a synchronic stage of the development of a language. The smaller the chronological distance between such steps the more rigorous and adequate the description of the development of the language which reflects its consecutive transition from state to state, starting with the initial one.

3. Typological verification of reconstructed models

When a reconstruction is carried out by definite methods of linguistic analysis one is faced with the methodological question of its plausibility and reality and the extent of structural correspondence to a real language that had existed in space and time and has been assumed to be an initial common linguistic system for the given group of related dialects. This leads to setting up a whole set of methodological principles of comparative-genetic linguistics, which are closely linked with the principles of linguistic typology and language universals. In this sense, genetic linguistics, i.e. the linguistics that establishes genetic relations among groups of languages and carries out reconstructions of their initial models, constitutes essentially an integral discipline with structural-typological linguistics, which explores language types and language universals in the light of the "theory of markedness or dominance". The reconstructed linguistic models of the initial language system - if they claim to reflect a language that had really existed in space and time - must correspond to the typologically determined universal regularities of language in general which are established on the basis of comparison of a set of various language structures. Naturally, a linguistic reconstruction running conunter to language universals cannot claim to reflect a language system that had actually existed. However, the correspondence of the reconstructed models to

94

Thomas V. Gamkrelidze

synchronic language universale cannot serve as a sufficient criterion for assessing the reality and naturalness of these reconstructions. Another necessary condition is the correspondence of the reconstructed models to the diachonic typological data, to the common patterns of change of definite linguistic structures established in studying particular facts of the recorded history of individual languages. Typological (synchronic and diachronic) verification of the reconstructed linguistic models proves to be one of the basic prerequisites in positing initial language structures. It is indispensable for validating the probability of such structures and their conformity with general linguistic reality. A linguistic reconstruction must be based on comparative evidence and at the same time take into account the synchronic and diachronic typological plausibility of a linguistic system arrived at by means of comparative and internal reconstruction. To put it differently, comparative reconstruction must go hand-in-hand with typology and language universals; otherwise, one would get a system which is linguistically implausible and nonverifiable. Consequently, "typological reconstruction" is not distinct from "comparative reconstruction" (as implied by Dunkel 1981). Diachronic linguistics actually deals with comparative language reconstruction; in some cases it reconstructs the Proto-linguistic patterns internally, typology and language universals serving only as "verification criteria". Typologically verified linguistic models arrived at by comparative and internal reconstruction must be given prevalence over typologically rare and implausible patterns that may also be posited theoretically on the basis of language comparison. Among diverse theoretical patterns of linguistic reconstruction arrived at with the aid of genetic comparison of related dialects, typological criteria must give preference to those which are linguistically most plausible and realistic, and which explain a host of historical facts that remain unaccountable from the viewpoint of alternative reconstructed models.

Markedness, sound change and linguistic reconstruction

95

4. The Indo-European consonant system 4.1 The limitations of the comparative method in classical Indo-European linguistics Considerations of synchronic typology in linguistic reconstructions calls for a substantial revision of the traditional schemes of the classical Indo-European comparative historical grammar and for a new interpretation of the established structural correlations on the basis of the principle of typological verification of the system and its diachronic derivability. The one-sidedness and limitations of the historical comparative method in classical IndoEuropean linguistics lies in the fact that the reconstructed models of the common Indo-European language were the result of external comparison of individual cognate systems, supplemented in some cases with internal reconstructions based on an analysis of definite types of relations within a single linguistic system. Furthermore, the linguistic plausibility of the obtained models was not explicitly taken into account. This led to postulating an initial language system which - running counter to the synchronic typological evidence as it does - cannot be considered structurally plausible and historically real. Consequently, the structural patterns postulated for the common (Proto-)Indo-European language are hardly probably from a typological point of view. The subsystem of Proto-Indo-European stops, for example, was traditionally reconstructed through the following three phonemic series: I voiced, I (b) d g

II voiced aspirates, II bh dh gh

and III voiceless: III Ρ t k

This essentially Sanskrit-based reconstruction of the PIE consonant system which may be characterized as "the Neogrammarian Paradigm" survived and flourished in Indo-European comparative linguistics until recently; the presently valid system differs from the Brugmannian one only in the elimination of the fourth series of voiceless aspirates, which have been qualified as having combinatory origin.

96

Thomas V. Gamkrelidze

4.2 The classical Indo-European consonant system challenged The classical system of Indo-European obstruents, however, did not stay unchallenged in the history of IE comparative studies. As noted years ago by Pedersen (1951) the functional weakness or even complete absence of the labial phoneme *b in the reconstructed series of voiced stops does not conform to the universally valid typology of stops, in which the voiced labial phoneme b invariably emerges as a functionally strong, "unmarked" member of the voiced series of stops, characterized by high frequency of occurrence and freedom of distribution. The absence of voiceless aspirated stops in the presence of corresponding voiced aspirates ((1), series II) constitutes another typological incongruity of the traditionally reconstructed system of Indo-European stops (Jakobson 1958). If such a system of Proto-Indo-European with specific and highly exceptional characteristics was a "historically attested" language, we would be called upon to account for its exceptional structural features and set up pre-stages to justify its peculiar, highly "marked" character and typologically exceptional traits. This would be a methodologically acceptable procedure which would account for the typological peculiarity of a historically attested language that served as a Protosystem to a group of related dialects. Some scholars did precisely this in order to justify the peculiar structural characteristions of traditionally reconstructed consonantism and thus "rescue" the time-honoured picture of Proto-Indo-European, as if it were not a theoretically posited linguistic construct, and hence a hypothesis, but a historically attested and recorded linguistic system, whose highly "marked" character and typologically exceptional features should be justified and accounted for. We, however, maintain that the Proto-Indo-European stop series do not pattern as they have been traditionally assumed to do; the traditional patterning is a mere historical chance due to the influence of the once prestigious Old Indian system and to the absence of strict reconstruction methodology. What is actually the unit underlying the IE dental stop series of phonemes; a voiced *d, a voiceless */ or a third unit, different from the historically attested ones? Logically all three possibilities may be envisaged, since any of these entities is a priori not ruled out. The decision in such cases must entirely rest with typological considerations; one would have to opt for the choice which, on the whole, would be linguistically more probable and plausible, and would constitute no exception to general typological evidence. That is why in the given case preference must be given to an entity which is phonemically unvoiced and characterized by an additional distinctive feature of glottalization.

Markedness, sound change and linguistic reconstruction

97

4.3 A reinterpretation of the PIE consonant system To bring the traditionally reconstructed system of PIE stops into correspondence with the data of linguistic typology - both synchronic and diachronic - it is necessary to set up a system with the following three phonemic series: I glottalized (ejectives) ~ II voiced (aspirates) ~ III voiceless (aspirates), with voiced and voiceless stops occurring positionally in the form of aspirated and corresponding nonaspirated allophones (Gamkrelidze Ivanov 1984: 5 f.): (2)

I CP') t' k'

II bh/b dh/d gh/g

III ph/p th/t kh/k

The reinterpreted system of the PIE consonantism offers a natural, typologically verifiable phonetic explanation for a number of features of PIE which could not be satisfactorily accounted for in traditional theory. For example, it accounts for the weak representation or even total absence of the maximally "marked" bilabial ejective stop, the absence in PIE of roots with two heteroorganic ejectives, the absence of ejectives in inflexional affixes (Hopper 1977:44), etc.. These features would be highly uncharacteristic of a system with plain voiced stops, but are very natural in a system with ejectives or glottalized phonemes. To the same direction points a typologically plausible interpretation of the frequency distribution of the three phonemic series in the PIE system. The statistics compiled by Jucquois (1966) show the following ratios for the phonemic series in question: (3)

Series I - 6 , 2 % Series II - 8,9% Series III - 17,7%

It is typologically evident that a series of voiced stops in a phonemic system, being less "marked" than the corresponding voiced aspirates, cannot be characterized by a lower frequency of occurrence. On reinterpretation of Series I as a series of ejectives, the frequency characteristics of these three series in the PIE system are made to fit well with the typologically established ratios of glottalized consonants which are juxtaposed to other phonemic series

98

Thomas V. Gamkrelidze

in a linguistic system.4 Taking into account both synchronic and diachronic typology, the proposed comparative reconstruction of the Proto-IndoEuropean stops differs considerably from the system of the Proto-IndoEuropean consonantism as reconstructed in classical comparative linguistics. In the new interpretation, the Proto-Indo-European system of stops proves to be closer to the systems traditionally defined as those with Lautverschiebung (Germanic, Armenian...), whereas systems which in respect to consonantism were thought to be close to the Common Indo-European system (Old Indian, Greek, Italic, etc.) appear to be a result of complex phonemic transformations of the original language system. In the latter group of languages, the original glottalized phonemes ((3), series I) became voiced (a phonetic process that has a parallel in a number of languages with glottalized consonants). A series of voiced stops thus appears, which is necessarily supplemented by a labial member that was regularly missing (or was weakly represented) in the original glottalized series. With this new interpretation of the Proto-Indo-European phonological system, the traditionally established trajectories of the transformations of the Proto-Indo-European stops into the phonemic units of the individual Indo-European languages acquire a reverse direction.5 The basic phonetic laws of Classical comparative linguistics, such as Grimm's Law, Grassman's Law, etc., are also conceptualized anew and acquire a different meaning. The new methods of comparative reconstruction, supplemented by the findings of modern linguistic typology - both synchronic and diachronic - in effect asks for a revision of the traditional schemes of Classical IE comparative linguistics; they urge the introduction of new comparative historical methods of reconstruction and the establishing of comparative historical IndoEuropean grammar as a distinct discipline.6

Notes 1. The terms are borrowed from molecular biology, but are extensively used in linguistics (cf. Jacob 1977). 2. Filling in by internal reconstruction is frequently the practice in current diachronic linguistics. {cf. the filling in of the so-called cases vides referred to by Martinet 1958). 3. The vertical relations of the "signifiant" and "signifie" are in a sense motivated at the level of horizontal relations. 4. As mentioned, there have been attempts to justify and rescue the traditional Proto-IndoEuropean consonantism, as if it were a historically attested system rather than a hypothetical construct. Thus, the plain voiced stops with highly marked labial *b and very common and unmarked velar *g, have been treated as transformations of a previous pre-Indo-European system with voiced implosives.

Markedness, sound change and linguistic reconstruction

99

New suggestions and alternative theories for the Proto-Indo-European consonantism have become very popular this last decade (cf. Gamkrelidze and Ivanov 1972; Hopper 1973). In connection with the pre-Indo-European voiced implosives proposed by Haider (1985) instead of the glottalized stops, we would like to point out that Greenberg (1970) showed that the series of voiced implosives is characterised by the same hierarchical relationship of markedness or dominance as the "plain" voiced stops: "unmarked (dominant) labial" vs. "(recessive) velar" or else vs. a gap. This is in contradiction with the traditional Proto-Indo-European "evidence" of plain voiced stops with highly labial *b and unmarked velar *g. The pre-IE voiced implosives could not have yielded what is traditionally known as the series of "plain voiced stops". It seems to be untenable to try and account for this fact by assuming a change of the postulated pre-IE implosive *'b into PIE *m (Haider 1985:12), along with respective changes of *'d and *'g to PIE *d and *g, which leave in the new series of PIE plain voiced stops a gap at the bilabial point. The latter point, it should be remembered, is just as favoured a point of articulation in the series of voiced stops as it is in the series of voiced implosives. (Considering the highly dubious PIE root *bel- 'force' as an instance of voiced *b will in no way save the situation.) Apart from this, positing voiced implosives, even for the pre-IE stage, leaves unexplained the root-constraint, which rules out the cooccurence of two voiced stops - one of the most conspicous typological inconsistencies of the classical PIE system. Haider (1985) would like to see unsurmountable difficulties with the reduplication structure of the type of Gr di-dö-mi, ΟΙ da-dä-mi. But even if we consider this type of verbal reduplication to be of PIE origin, there is no difficulty in positing a PIE structure t' V-t'oH-, with two homorganic glottalized stops in a sequence, since the typological constraint concerns the tendency of non-cooccurrence of two heterorganic glottalized stops, while two homorganic glottalized consonants may combine freely in a root or a word-form. The attempts to account for the typological inconsistencies of the traditional system by setting up for Proto-Indo-European a pre-stage with different sorts of phonemes are as old as the attempts to introduce such changes at the pre-Indo-European stage and leave intact the traditional system of Proto-Indo-European {cf. Pedersen: 1951). Such internal reconstructions of different typologically consistent pre-Indo-European stages leave unexplained the transition from presumably stable configurations to the highly unstable system known traditionally as "Proto-Indo-European", which is assumed to have subsequently transformed once again into the typologically stable systems of the historical Indo-European dialects (cf. Cowgill 1984 [1985]). The same considerations refute Szemerenyi's recent objections to the new system of PIE consonantism. Szemerenyi actually tries to reject the thesis of the absence of the voiced labial *b from PIE, referring to forms with b in medial position: Lat lubraicus, libo; Goth diups. He admits that "initially b is rare, perhaps not to be acknowledged at all; but internally it is vigorously represented" (cf. Szemerenyi 1985: 12). But his "vigorous representation" of internal *b is restricted mainly to Western ("ancient European") dialects, thus casting doubts on its PIE character. Furthermore, dealing with root-restrictions in PIE Szemerenyi tries to account for the "voiced aspirate - voiceless" and "voiceless - voiced aspirate" restrictions from traditional point of view, through assimilation (which, by the way, is our explanation, too), leaving unaccounted the "voiced - voiced" restriction, which has no satisfactory traditional explanation. Reading Szemerenyi, one cannot get rid of the impression that one of the earliest proponents of the "new outlook" on PIE has abandoned his previous approach to the theoretical problems of reconstruction and is rejecting the attempts of others to follow that approach. 5. The reinterpreted Indo-European system of stops appears to be typologically close to respective systems of the languages in adjacent areas - South Caucasian and Semitic ones. This suggests that one should study the interrelationships of the Indo-European, South Caucasian and Semitic languages within a common cultural area. 6. The contours of this grammar can actually be discerned in the Indo-European Glotalic Theory (Bomhard 1975) - a fundamentally different approach to the reconstruction of PIE, which

100

Thomas V. Gamkrelidze

along with the Palatalgesetz and the Laryngeal Theory is said to constitute a paradigm in IndoEuropean comparative linguistics (cf. Baldi 1981; Palome 1982), one which represents a final stage in the process of divergence of the Indo-European comparative studies from the Old Indian Pattern as a model for Proto-Indo-European (cf. Mayrhofer 1983,1986). As Lehmann (1983) put it: "Major contributions of the past five decades have modified extensively the view on Proto-Indo-European phonology presented in the standard handbooks by Brugmann, Hirt and Meillet. The contributions result on the one hand from a different approach to the parent language, on the other from two far-reaching theories. The Laryngeal Theory and The Glottalic Theory... What had seemed one of the most solid achievements of the 19th century linguistics is now modified in every section". How different this attitude is from the one governing linguistic thought at the beginning of our century, when Meillet (1903: Introduction) summing up his view of the situation in comparative linguistics, stated: "En un sens au moins, il semble qu'on soit parvenu a un terme impossible a depasser"!

References Baldi, Philip 1981 [Review of festschrift Szemerenyi], General Linguistics 21, 1: 47-62. Bomhard, Allan R. 1975 "An Outline of the historical phonology of Indo-European", Orbis Vol.26, No. 2: 354-390. Cowgill, Warren 1984(1985) [Review of memorial volume Kerns], Kratylos 29: 6. Dunkel, George 1981 "Typology versus reconstruction", in Bono Hamini Donum Essays in historical linguistics in memory of Alexander Kerns, edited by Y. L. Arbeitman and A. R. Bomhard (Amsterdam: Benjamins), Vol. V, Part II: 559-569. Gamkrelidze, Thomas V. 1974 "The problem of l'arbitraire du signe", Language Vol. 50, No. 1: 102-110. 1975 "On the correlation of stops and fricatives in a phonological system", Lingua Vol. 35: 231-261. Gamkrelidze Thomas V. & Vjaceslav V. Ivanov 1972 "Lingvisticeskaja tipologija i rekonstrukcija sistemy indoevropejskih smycnyh" [Linguistic typology and the reconstruction of the system of Indo-European obstruents], in Konferencija po sravnitel'no-istoriceskoj grammatike indo-evropejskih jazykov. Gamkrelidze. Thomas V., Vjaceslav V. Ivanov 1984 Indoevropejskij jazyk i indoevropejcy [Indo-European and the Indo-Europeans] (Tbilisi: Izdatel'stvo Tbilisskogo univerziteta), Vols. I and II. Greenberg, Joseph 1970 "Some generalizations concerning glottalic stops, especially implosives", I J AL 36:123-145. Haider, Hubert 1985 "The fallacy of typology. Remarks on the PIE stop-system", Lingua 65:1-27. Hopper, Paul 1977 "The typology of Proto-Indo-European segmental inventory", The Journal of Indo-European Studies 5:41-53. Jacob, F. 1977 "The linguistic model in biology", in Roman Jakobson. Echoes of his Scholarship, edited by D. Armstrong and C.H. van Schooneveld (Lisse: Peter de Ridder), 185-192.

Markedness, sound change and linguistic reconstruction

101

Jakobson, Roman 1958 "Typological studies and their contribution to historical comparative linguistics", Reports for the VIHth International Congress of Linguists (Oslo), 17-25. Jucquois, G. 1966 "La structure des racines en indo-europeen envisagee d'un point de vue statistique", in Linguistic Research in Belgium (Wetteren). Lehmann, Winfred P. 1983 Directions for Historical Linguistics (Austin: University of Texas Press). Martinet, Andre 1958 "Function, Structure and Sound Change", Word 8: 1 -32. Mayrhofer, Manfred 1983 "Sanskrit und die Sprachen Alteuropas. Zwei Jahrhunderte des Widerspiels von Entdeckungen und Irrtümen", NAWG 5: 121-154. Mayrhofer, Manfred 1986 Indogermanische Grammatik, Bd. 1, 2. Halbband: Lautlehre (Segmentale Phonologie des Indogermanischen (Heidelberg: Winter) Meillet, Antoine 1908 Les dialectes indo-europeens (Paris). Pedersen, Holger 1951 "Die gemeinindoeuropäischen und die vorindoeuropäischen Verschlusslaute", Det Lgl. Danske Vidernskabernes Selskab. Hist.-Filol. Meddelelser, XXXII/5 (Kobenhavn) Polome, E. 1982 "Preface", in The Indo-Europeans in the Fourth and Third Millennia edited by E. Polome (Ann Arbor: Karoma). Szemerenyi, Oswald 1985 "Recent developments in Indo-European linguistics", in Transactions of the Philological Society 1-71.

Markedness and naturalness

1. Thoughts about markedness and normalcy /naturalness Werner Winter In recent years, the notion of markedness has received a good deal of attention. It has been noted repeatedly that in the course of the past decades there have been considerable changes in what "markedness" was supposed to mean. In the early days of the Prague school, markedness was a formal property in the narrowest possible sense: a marked form was taken to be characterized by the presence of a feature absent in the corresponding unmarked form - in other words, marking was the same thing as increased complexity of a form. Thus, a plain voiceless consonant would lack the marking that a voiced consonant had by virtue of the addition of voice; an aspirate would be marked over against a nonaspirate; a form containing a future suffix would be marked when viewed against a present form lacking this suffix. The simple correlation of complexity and markedness was destroyed when an extension of the notion of markedness to a greater variety of domains of human language was introduced. In Jakobson's view, linguistic categories which formed related parts of a pattern could be classified as marked or unmarked with respect to each other, and this could be done without any regard for the formal properties of the items compared and with results that were thought to retain validity provided the scope of the comparison was kept constant. Thus, a singular form would always be held to be unmarked when compared with a plural form, no matter whether the two forms were of identical complexity, whether the plural was more complex, or even whether the singular was formally more complex than the plural. A preterite form would always be marked against a present tense, a perfective against an imperfective, a first or second person against a third. At this point then, the notion of markedness had ceased to be strictly formal one and had become a functional notion: it could, in some way, be said that category A was simpler functionally than category B, and there seemed to be reason to believe that the earlier (Trubetzkoyan) notion of markedness could

104

Werner Winter

be considered to be so similar to the new (Jakobsonian) notion of lesser or greater simplicity that no harm would be done if the earlier terminology was retained. Some empirical findings seemed to support the assumption made. It could be noted that there was a certain correlation to be found between frequency of occurrence of marked and unmarked forms in the old and the new sense: highly complex sounds or morphological constructs tended to occur with lesser frequency than the less complex, "unmarked" ones; likewise, functionally simple, or in the new sense "unmarked", forms turned out to be more common than functionally more complex ones. Moreover, reduction processes seemed to lead toward unmarked forms in both contexts: neutralization of phonemic contrasts would generally let the unmarked (in a Trubetzkoyan sense) member of a pair survive, neutralization of a functional contrast would lead to the elimination of the functionally marked member of the set. Further progress seemed possible when some concepts of information theory were introduced: it seemed that there was merit in saying that a marked form carried a higher informational load than its unmarked partner. Somehow, however, the introduction of notions of function and informational yield carried with it seeds of destruction for central aspects of what once had been a relatively straightforward markedness theory. It was known that the informational yield could be said to increase if the signal chosen was not the expected one - if it was not what one would normally encounter. Thus, "unmarked" and "normal" could be taken to be almost, if not altogether, interchangeable terms. And indeed, proposals were made to replace the notion of markedness and unmarkedness with one of degrees of normalcy; more recently, this notion in its turn has given way, at least in the diction of a number of linguists, to one of degrees of naturalness. Of the last two alternatives, the first may appear to be the more appealing one, as normalcy, if defined in terms of frequeny of occurence, can be said to be an empirically founded and testable notion, whereas naturalness, for all its apparent cogency, still has to be localized in the realm of intuition - not a bad place in the incipient stages of a new approach, yet part of a domain definitely in need of objectification. Whether one wants to prefer normalcy or naturalness, in both cases one does not get very far by asking the question: "Is the use of form A more normal (or more natural) than that of form B?", although this appears to be precisely what is implied in Jakobson's approach. (Note that the question: "Is form A more normal/natural than form B?" cannot even be asked with any hope for an appropriate answer.) It rather seems necessary to modify the question introduced to read as follows: "Is the use of form A more

Markedness and naturalness

105

normal/natural than that of form Β under a set of conditions C?". The conditions may be of very different nature. They may be textual - it is of greater significance to switch, in a narration, from preterite forms to a historical present than to continue with a preterite. They may be sociolinguistic - in present-day German universities, a student will address a person whom he takes to be a fellow student with the allegedly familiar du, even if he does not know him personally at all, while a person of the same age group whom he does not identify as a student, will usually be addressed by Sie; du thus is "unmarked/normal/natural in one social context, "marked"/nonnormal/nonnatural in the other. Historical conditions will affect the status of a member of a set of forms: in the days when the Quakers set out to fight what they considered a lack of simple manners in the society around them by refusing to employ the illogical, unnatural you in addressing a single individual, they selected the seemingly appropriate, plain thee as what must have had the status of the "unmarked" member of the pair thee : you; when you lost all characteristics of an honorific and just became plain second person, the humble thee all of a sudden came to signal something else, namely that the speaker considered the person addressed a member of the small community of plain people (this pattern is what has developed among Quakers in the United States who will use you freely with all outsiders). Viewed in this way, "markedness", or degree of normalcy, or degree of naturalness, is not the property of a form, but of a form when used - even more: when used under ultimately specifiable conditions. One may thus say that "markedness" as understood here is a context-sensitive feature, whereas in the early Praguian versions of the theory markedness was there or not there, regardless of how a form was used - in other words, markedness of the early kind was context-free. What seems to have happened in the transition from Trubetzkoyan markedness to Jakobsonian "markedness" is that the assumption of nonsensitivity to context was retained as presumably valid even though essential properties of markedness and "markedness" differed greatly. One might stop here and say that due to historical developments the nature of a notion seems to have undergone a profound change, so that when using the term markedness one should take care to make clear one's point of reference. One could go further and say that as a result of deepening insights a notion had been modified and improved, so that there would be no point in going back to Trubetzkoy's and Jakobson's early views at all. Or finally, one could find merit in both early and later versions of the theory, but doubt the value of the suggested continuity and argue in favor of keeping the notions of markedness and of "markedness"/degrees of normalcy/degrees of naturalness strictly apart.

106

Werner Winter

It should be obvious that I am opposed to an exclusive concentration of our attention on formal complexity, and to a concomitant system-bound disregard for the use of forms. Likewise, it should be clear that I do not favour a lighthearted identification of a particular usage as normal/natural; only to the extent that claims in this respect can be supported by empirical findings would I want to see these categories retained at all. This means that for me a twofold approach has the highest degree of attractiveness: I would want to be able to classify manifestations of language with respect to their overt complexity, and I would be interested in precise statements about the relative normalcy/naturalness of a particular use of linguistic forms. The first interest could be called an interest in paradigms, the second, one in syntagms. Both interests are fully legitimate for a linguistic and neither should be made exclusive at the expense of the other. This means, among other things, that I will certainly want to know whether, and to what extent, there is a correlation between degree of formal complexity and degree of normalcy/ naturalness of use of items from a language. The few examples alluded to above will suffice to show that this cannot be a question of a very simple correlation: if that is the case. I cannot dispense with an independent study of both aspects: the early Jakobsonian solution of investigating only function cannot be applauded, nor can the easy assumption of a continuity between Praguian „markedness of form" and recent "degree of normalcy/naturalness" be adopted. As I do not want markedness as a formal property excluded from consideration, I have no choice but to take a stand about as follows: A linguist has a legitimate interest not only in the description of the structure of an item of a language, but also in the comparison of the structure of this item with that of other items. In pursuing this interest, he will discover that some items can be fruitfully compared and that frequently it can be said that one item has a higher degree of complexity than another one. This higher degree of complexity can be usefully referred to by the term markedness. Once the term has been adopted in this particular sense, it should be blocked for all other uses in the interest of a lucid scientific discussion. Linguistic items will be identifiable as marked or unmarked only in the context of a comparison; pairs of forms are needed to talk about markedness (but "pair" may very well be a member of the set "more than one form" - there is no intrinsic need for a binary approach). A linguist should not only be interested in the study of forms and their structure, he should also pay attention to their use. Here the question of normalcy or naturalness of a particular use of a particular form under particular conditions may be raised profitably. But in my view, there is no need for believing that "unmarked" in the sense of "simpler in form" should

Markedness and naturalness

107

be coextensive with "more normal/more natural". These is too much that argues against such an assumption. For instance, Russian ljudi 'people' is certainly marked as compared with Russian / / W people'; still, the plural and not the singular form is the commonly used and therefore normal/natural term. In an Indo-European language, as elsewhere, b is certainly the marked counterpart of unmarked ρ; still, in the development of Celtic, ρ is lost. If, however, "less normal/natural" may refer to phenomena quite different from those covered by "marked" in its original sense, it seems appropriate not to use the same term in both cases; it seems unwise to speak of "marked" when "simpler in form" is addressed. I would suggest that we had better speak of "marked" and "unmmarked" members of a set of forms when referring to their properties of form and of "normal" vs. "nonnormal" when discussing the use of forms. Markedness would then be measurable in terms of complexity, normalcy in terms of frequency of occurrence in a given context. It is immediately obvious that it is infinitely easier to observe and prove markedness than to observe and prove normalcy. In very many cases we will hardly be able to transcend the state of intuitive judgement that seems to be characteristic of an assessment as "natural". Still, if we are at all interested in the functioning of language and not just in some bloodless abstract grammar, determining degrees of normalcy remains of paramount importance. As is usual with all empirical research, progress will be slow, and the final goal will very possibly never be reached. But in pursuing the goal of determining what is normal usage, we will notice linguistic variation, both synchronic (on a social scale for instance) and diachronic, much more clearly than if we are not forced to weigh our observations and can be content with fitting them into some mere description. Concern with the study of usage and the determination of normalcy should not prevent us from continuing the investigation of markedness. A few brief comments will suffice here: Jakobson's discussion of "marked" and "unmarked" forms in the Russian verbal paradigm disregarded the formal side - it was only concerned with function, and thereby became a precursor to the normalcy/naturalness approach, with the level of intuitive assessment not exceededed. The disregard for form led to results that were undesirable, for all their apparent elegance. Jakobson's classification of imperfective forms as "unmarked" and of perfective ones as "marked", with "markedness" as a property attached permanently to one member of this pair, failed to offer a basis for an explanation of the fact that there were pairs of forms in which the perfective was more complex (that is, formally marked), as in napisaf vs. pisaf 'write', and others where an increase in complexity occurred in the imperfective, as in daf vs.

108

Werner Winter

davaf 'give'. For Jakobson, at least at the time when this analysis was proposed, the evidence of form did not seem to count; it could not be allowed to count, for had he acknowledged it, he would have had to give up his beautifully simple appraisal of the use-determined classification of the constituents of a verbal paradigm. But in giving up the evidence of form, he deprived himself of the chance to state that in Russian - as in other IndoEuropean languages - there were verbs which, because of their basic meaning, had their focus in the durative present and others which had their simplest form and their center of use in the nondurative parts of the verb system. The gain in simplicity, paid for with a striking disregard for the evidence of form, was offset by a loss of insight into the internal dynamics of a typical Indo-European verb paradigm with its meaning-conditioned distribution of simple basic forms and its use of shifters in the creation of a completed, balanced paradigm. Parallel cases are easily found. Armenian has a special singular suffix that occurs in nouns denoting paired body parts: akn 'eye', unkn 'ear\ jern 'hand', otn 'foot', etc. In terms of ProtoArmenian morphology, the accusative plural ots 'feet' (which replaced an older dual) is less complex than the singular otn 'foot': ots consists of plain stem plus ending, otn of extended stem plus ending (which of course was lost as a result of final-syllable reduction). It one were to accept Jakobson's dictum that dual forms are always "marked" in the juxta-position with singular ones, morphological markedness ( = complexity) and functional "markedness" would be at odds. If one, however, separates the two approaches, one can state that in this class of Armenian nouns, the dual forms were less complex, hence unmarked, if compared with their singular counterparts. If one then considers the possibility that the unmarked member of the pair might also be the more basic one (a most tempting conclusion as morphological derivative processes normally are more readily additive than subtractive in the Indo-European languages), the interpretation readily comes to mind that indeed in the domain of paired body parts a reference to the pair, and not to its individual member, may be basic, and that therefore Armenian otn, etc., should be identified not as primary singular forms, but as derived singulative ones. Taking again the evidence of form to be more essential than that derived from a universalist appeal to function, means that the seemingly irreducible conflict between markedness in a formal and "markedness" in a functional sense can easily be eliminated, and again valuable insights into the dynamics of a language system are gained. A closely similar case can be adducted from Russian again. If one were to accept Jakobson's view that plural is always more "marked" than singular, a

Markedness and naturalness

109

pair like dom : doma 'house : houses'would pose no problem: the plural form would be both morphologically more complex, hence marked in the original sense, and functionally nonbasic in the Jakobsonian view. However, if one compares Russian bojarin 'nobleman' with its plural bojare, one finds that the latter form contains the stem bojar- plus an ending, while the singular forms bojarin, bojarina, bojarinu etc., show a suffix preceding the ending. Again a recourse to the primacy of the evidence of form yields a desirable result: bojarin can be taken to be a singulative form, which could be described as meaning 'member of the nobility', and the term for the group again is more basic than that for the individual. As in the case of the Armenian nouns, respect for the evidence of form helps reveal a pattern which, to be sure, is much less common in Indo-European languages than is its counterpart with basic singular forms. But the pattern is there nevertheless, and in terms of a functional approach to "markedness" or normalcy/naturalness, it makes very good sense. When criteria of form ceased to be all-important, the notion of "markedness" became vague to a considerable extent, with a claim to a very high generality, but also unverifiable in too many details to make one feel comfortable with it. The Jakobsonian notion of a universally valid, permanent surplus of "markedness" of one category over another, intellectually exciting as this notion may have seemed, could not possibly be retained. Instead, a return to the formbound notion of markedness seemed called for, combined with a search for degrees of normalcy to be determined on the basis of actual usage. Once the two had been investigated independently of one another, the results could be viewed together and possible correlations could be studied. It seems that we stand at the beginning of a very long road, but of one that promises to lead toward highly interesting findings.

Markedness and naturalness

2. Markedness and naturalness in phonology; the case of natural phonology Wolfgang U. Dressier Phonological markedness has been understood in many diffeerent ways in different theories {cf Klein 1982; Brakel 1983; Shapiro 1983; Herbert 1985; papers in Eckmann et al. 1986). In Natural Phonology (henceforth NatPh) 1 as founded by Stampe (1969) and further developed in the Americas and in Europe {cf. Dressier 1984), markedness is equated with phonological unnaturalness. Similar to Natural Morphology we must differentiate naturalness/markedness on the levels of phonological universals of typology, and of language specific systems. On the level of universals marked options are the opposite of universal preferences {cf. Mayerthaler 1982) and of the application of universal natural processes {cf Stampe 1973; Hurch 1985). On the typological level markedness refers to dispreferred vs. preferred phonological options of language types {cf. e.g. Donegan Stampe 1983; Dressier 1984:4.2). Little work has been done on language specific phonological system adequacy within NatPh {cf Dressier 1985: 348 ff.). Emphasis has always been laid on establishing what is phonologically natural of truly phonological in a given language. Two of the most common misinterpretations of naturalness are, first, the implicit equation of natural with intuitively plausible, based on impressionistic intuitions of linguists, second, the (often explicit) equation with crosslinguistic high frequency. This latter misinterpretation reverses the logical order of inquiry: Parameters of universal naturalness/markedness first must be deduced from the extralinguistic bases of phonology {cf. Dressier 1984: 2.3.4), and only afterwards can they be subjected to inductive falsification tests in crosslinguistic investigations of as many types (or domains) of evidence as possible. "Natural" has been used in more precise ways by other phonologists, e.g. in Schane's (1973) concept of "natural rules". However as one critic of NatPh, Bauer, (1982: 21) explicitlly stated, "Stampe's theory of naturalness is far superior to Schane's". 2

112

Wolfgang U. Dressler

What does language specific phonological naturalness mean within NatPh ? Within any language, both its phonemes and their phonetic realizations are the outputs of phonological processes which are subsets of universal natural processes inhibited in language acquisition, to which I shall refer as phonological rules (henceforth PRs). I distinguish prelexical and postlexical PRs. Prelexical PRs merge conceivable sounds into the phoneme inventory of a given language (e.g., in many languages, all obstruents are merged into underlying unvoiced obstruents) and determine its phonotactics. Postlexical PRs derive phonetic outputs from phonemic inputs. However, PRs may apply both in prelexical and postlexical functions. E. g., in many languages, the same PR of nasal place assimilation excludes */np/ from its phonemic input in favour of /mp/ (prelexical function) and derives [rjk] from underlying /nk/ (postlexical function). In contradistinction to Generative Phonology (e.g. Anderson 1981; Gussmann 1984), NatPh acknowledges and investigates the extralinguistic bases of its principles (cf. Dressier 1984,1985), and these bases lie in phonetic, neurological, psychological, and social phenomena, but also in general properties of semiosis. For example, physiological limits of perceptual space and of articulation restrict the number of possible sounds, of distinctive features and of their combinations. And among the articulatory options available there are best points, optimum positions which have the preceptual effect required while entailing minimum physiological energy, i. e. "minimum displacements of lips and tongue from their natural positions" (Lindblom 1972: 79) so that they achieve maximal perceptual contrast with least effort. Thus the extralinguistic bases of phonology underdetermine phonological structure, but they limit the choice of phonological processes open to languages and favour/disfavour alternative processes in a given constellation of factors. Therefore there are fewer absolute external constraints than relative ones, which makes many conceivable phonological structures and processes highly marked. This is the extralinguistic basis of phonological markedness. NatPh and phonetics share a functional perspective (cf. Lindblom 1983). In order to avoid proliferating functional ad hoc assumptions it seems useful to subsume functional considerations into a semiotic metatheory (cf. Dressier 1984, 1985). Clearly all phonetic acts involved in production and perception are parts of semiosis, because acts of semiosis (or sign chains in the Jakobsonian sense) mediate between "substance and form". In this perspective the pernicious dichotomy between sound substance as a domain of phonetics and sound form as a domain of phonology can be eliminated and phonological markedness/naturalness can be founded on phonetic operations.

Markedness and naturalness

113

Any functional explanation has to account for functional conflicts. NatPh has highlighted two conflicts well-known to many linguistic and phonetic schools for a long time: (a) the conflict between phonological-phonetic and morphological tendencies, whose diachronic version has become prominent under the Neogrammarian labels "Lautgesetz und Analogie" {cf. Dressler 1985, Wurzel 1982), and (b) the antagonism between tendencies towards "ease of articulation" vs. "optimal perception". These concepts have been shunned as too vague by many investigators. Lindner's (1975) model of context-free and context-sensitive articulations gestures has represented an important step forward by operationalizing concepts of articulation-based markedness/naturalness. Clearly "ease of articulation" may mean two things: (a) the universal (cross-linguistic) likelihood of a universal phonological process locally applied irrespective of its contextual effects, e. g. universal hierarchies of consonant palatalization, vowel centralization/deletion etc., or (b) some more "global" effects in terms of prosodical structure and consonant/vowel clusters, e. g. avoidance or elimination of consonant clusters resulting from unstressed vowel deletion. On the basis of this antagonism NatPh has elaborated on a classification of opposed universal phonological process types with opposed hierarchies {cf. Donegan and Stampe 1979; Dressier 1984,1985). Phonological processes are either of a dissimilatory nature (foregrounding processes in Dressier 1984, 1985), e.g. diphthongization, insertion, lengthening, strengthening process types, or of an assimilatory nature (backgrounding processes), e. g. shortening, weakening, centralization, deletion, fusion, assimilation process types {cf. Kiparsky 1986: 2.2). The Phonological processes exhibit their communicative function through their proper functions: pronounceability and perceptibility, i.e. the foregrounding processes primarily serve perceptibility, while the backgrounding ones serve, pronounceability by changing marked into unmarked values.3 Most naturalist analyses seem to be rather concrete, and often {cf Hooper 1976) the difference in concreteness is seen as the major distinction, in contrast to Generative Phonology. But in NatPh concreteness can be deduced from the sign character of processes and rules: the input of a process, e. g. a phoneme, is the signatum of the process, its output, e.g. an allophone, its signans. The semiotic principle of sign efficiency renders distinctive and well perceivable signantia preferable (relatively unmarked). This property of efficient signantia of being distinctive and perceptible makes intermediate false steps in phonological derivations costly. Gussmann's (1984) analysis of Modern Irish lenition

114

(1)

Wolfgang U. Dressler mäthair

[ma:r']

'mother'

an mhäthair

'the mother'

[wa:r']

with the rules m - » b , b->v, ν w involves two intermediate false steps; for a concrete alternative see below. Therefore the relative concreteness of analyses in terms of NatPh represents neither an ad hoc assumption (although more abstract analyses always have to carry the burden of proof) nor a research strategy designed to restrict the number of possible competing analyses of the same data (as in T. Vennemann's and Hooper Bybee's, 1976, Natural Generative Phonology), but follow from a higher-order principle. One of the most common misunderstandings of NatPh by its detractors lies in lacking, or defective, distinction between PRs and MPRs (morphonological rules) or even AMRs (allomorphic rules), cf. countercritiques in Dressier (1984: 3.9.5,1985). Many phonologists think that PRs should have a phonetic basis, but MPRs should not. However a phonetic basis is just a necessary property of PRs, but not a sufficient criterion for distinguishing them from MPRs or even AMRs. Note that even lexical suppletion between the English indefinite articles a and an has a phonetic basis. Among the many criteria for distinguishing PRs, MPRs, and AMRs (Dressier 1985), the following ones must be stressed: a) If phonology has the function of making language pronounceable and perceptible, PRs should aid pronunciation or perception via backgrounding and foregrounding respectively. Thus of all optional PRs only those phonostylistic PRs are truly phonological that follow this predicition. b) The obligatory PRs must represent a constraint on production and perception, whereas MPRs or AMRs need not. c) Thus PRs must be general, i. e. exceptionless, MPRs and AMRs need not be, and usually are not. d) A PR should apply productively to loanwords, neologisms, abbreviations etc. e) The phonetic distance between input and output of a PR must be small, but it is usually much bigger with an AMR; MPRs are usually in between. f) From (b) it follows that PRs must be phonologically conditioned, because the constraint is phonological. For MPRs the morphological conditions must always be present, while phonological conditions may be. Distribution is therefore a more important criterion than alternation for PRs, the reverse for MPRs. In this respect NaTPh resembles structural phonemics more closely than Generative Phonology {cf. Zwicky 1975:155f). A strict consideration of these criteria has allowed Dressier (1984,1985) to refute criticisms levelled against NatPh by Anderson (1981), Drachman

Markedness and naturalness

115

(1981), Hellberg (1980), and Lass (1980, 1981). In this paper I want to rebut the criticisms by Gussmann (1983, 1984) and Bauer (1982). First let us apply the above mentioned criteria to Icelandic palatalization, which has been identified by Wurzel (1980, 1981) as an MPR, but by his critic Gussmann (1984) as a phonological rule. Icelandic palatalization 4 of /k,g/ [k',g'] always applied before /j/ as long as /j/ was not obsorbed by the preceding consonant, and it still applies before /i/, e. g. in bök 'a book', definite bokin 'the book', ekki 'not', but not always before /e/. This phonetically based phonological hierarchy and the criterion of small structural change/phonetic distance between input and output exclude the assumption of an AMR; the existence of exceptions points to an MPR rather than to a PR. Exceptions are recent borrowings such as okei 'O.K.', geim 'game', the acronym ΚΕΑ, the affectionate term of address gey from grey 'poor thing' (more in Ärnasson 1978: 186f., Oresnik 1977: 138f)· Thus Icld. palatalization before /e/ is not fully productive in loanword integration either, nor is it productive in language internal neologisms or expressive language. Therefore palatalization before /e/ cannot be a PR, but may safely be identified as an MPR. Moreover, palatalization applies before the primary diphthong [ai], but not before secondary diphthongs [ai] derived from other sources. Thus this subpart of palatalization is not general and the contextual distinction between a primary and a secondary diphthong is not phonetically based. These criteria again exclude the status of a PR. More precisely, Icld. palatalization has remained a PR before [i], but changed to a MPR in other contexts (cf. below). While the answers to Gussmann's (1984) other ciriticisms follow from the framework described in Dressier (1984, 1985), there remain two specific problems regarding the history of Icelandic which seem to undermine the assumption of an older PR of palatalization: a) Gussmann (1984:160) claims "that at no stage was there a neat natural PR of the type proposed by Wurzel (1981: 426) ... since at the time when /j/ palatalised, the vowels did not, whereas when the vowels did, β ] presumably no longer appeared phonetically after velars" (i.e. it had already been absorbed by the palatalized velars). However NatPh claims in fact that usually a PR starts to apply in the most favorable phonetic environment, and for palatalization a following glide /j/ is more favorable than a following front vowel. Thus the first part of Gussmann's claim in fact strengthens Wurzel's analysis. The second assumption of Gussmann's first claim is rather striking (as Anatoly Liberman, Madison, underlined in private communication): 14th century spellings kje, gje {cf. Einarsson 1941: 47) just show the terminus ante quem for palatalization before /e/ (which presumably occurred later than

116

Wolfgang U. Dressler

before /i/ due to the intrinsic palatalization hierarchy /j > i > e/), and usually such spellings should indicate (at least incipient) phonemization of palatalized velars (cf. Penzl 1957). Then there were never spellings kji, gji. On the other hand spelling evidence for absorption of /j/ is hard to come by. Therefore Gussmann's first claim does not hold. b) Gussmann's (1984:160 f) second claim seems to be better founded: "the pi. participle suffix -end- [of -endur, sg. -andi] never caused palatalisation ... This one suffix in WurzePs model turns the cross-lexical and otherwise unexceptional generalisation ... into an MPR". I.e. the failure of palatalization before the umlauting form of this one suffix would exclude Icelandic palatalization from ever having been a PR. However this does not follow. The only conceivable consequence could be that palatalization before /e/ can never have been a PR, i.e. the present situation (PR before /i/, but MPR before /e/) would have been true from the very beginning. If we replace Gussmann's static view with a dynamic approach, we see that PRs can have lexical exceptions in two situations (cf. Dressier 1985: 67, 84 if): i) when a PR of casual speech diffuses lexically or by phonetic analogy (cf. Kiparsky 1986: 3.1) (i.e. we may assume that the PR of Icelandic palatalization spread from the position before /j/ to the position before /i/ and then before /e/ but never reached the umlauted plural form of the participle.) ii) When a phonemic PR is on the verge of becoming an MPR (see below). This seems to have actually happened; otherwise the spellings kje, gje would not have been introduced (see above). Without going into further details I hope to have shown that by now the distinction between PRs (the proper domain of NatPh) and MPRs is usually rather simple, although there are a few difficult instances where a phonemic PR is on the verge of becoming an MPR (see above). One such difficult instance is Russian final obstruent devoicing (cf. Dressier 1985: 93). Phonemic PRs change diachronically into MPRs and even AMRs, the socalled morphologization of PRs (cf Dressier 1985, chapter 5; Wurzel 1981; Kiparsky 1986). Gussmann (1983) has misunderstood the profound nature of change in morphologization when he anachronistically analyzes Modern Irish lenition via the single lenition rule: (2)

[+obstruent]

-*

[+continuant],

preceded and followed by 5 other abstract rules. A concrete analysis would also need 6 rules, but very concrete ones, which I express in phonemes rather than in features, for expository reasons, but also because I agree with J. Kurylowicz that PRs are better described in terms of features, and MPRs

Markedness and naturalness

117

and AMRs in terms of phonemes (for similar Breton lenition cf. Dressier 1985: 59f, 62, 66, 92, 94): (3)

(a) (b) (c) (d)

(e)

b,m ->· w e.g. mac 'son' - mo mhac [wa:k] 'my son' b',m' -> ν'; ρ f; p' -»• f e.g. bean 'woman' - def. an bhean [v'] f , f -> 0 e.g. fear 'man' - Gen.def. an fhir [ir'] t, t', s, § -> h e.g. teach 'house' - se theach [h] '6 houses' y y.

e.g. crann 'tree' - faoi chrann 'under a tree'

α palatal] ~_ papapapa (clicks tongue) (.5) prefer ahah /prif3:rans/ nee (W1:163ff.)

2.2.2.3.4 Sometimes, though seldom, the students use over-specific forms (hyponyms): (33)

lebensfähig ... and who's only ~ able to live (.3) or can only survive (B9: llOff.) mit Rotlicht ... with - a red signal (W11:117 ff.)

164

Rüdiger

Zimmermann

2.2.2.3.5 Search in the wrong cognitive frame may also lead the learner completely astray. (34)

ins Abseits - geäußerte Gedanken ... Abseits da fallt mir sofort - der Kinofilm ein mit Götz George - Abseits - Fußball denk ich auch dran (B 22 b: 223 ff.) (thoughts spoken aside ... Abseits I immediately think of the movie with G.G. - offside - football I think of too)

2.2.2.3.6 At the nonstrategic level, a very frequent phenomenon has to be mentioned: repetition of the difficult L1 item, often over and over again, as if this might somehow conjure up the L2 form. (35)

Kutscherhaus ... or -*• Kutscher λ Kutscher ~ Kutscher - Kutscher s - (groans) keine Ahnung \ Kutscher /" - fallt mir überhaupt nix zu ein \ - Kutscher (B 3:112ff.)

2.2.2.3.7 Finally, from a sequential point of view, "discontinuous" strategies are unproductive and can be seen as nonnatural in an interactive sense: jumping from one line of search to a completely different one would not be helpful for a listener, as it would postpone the choice of the lexical item: (36)

Kutscherhaus ~ Kutscherhaus (incomprehensible) heißt Kutscher cart ~ horse cart (.7) Chauffeur ~ aber is ja eigentlich kein Kutscher /hat ja mit/ Benzinkutschen (.4) Kutscherhaus (.4) da kann ich ja vielleicht schon mal house aufführen ... hm Kutscher (.12) man kann ja schon mal Chauffeur hinschreiben ... ich weiß /eben/ das Wort für Kutsche nich horse cart horse cart Lenker einen Wagen lenken ... also es geht mir immer noch sowas mit — Pferd und so im Kopf rum ... ich kann mich auch an keine Stelle in der Literatur erinnern wo so'n Kutscher vorgekommen wäre weder in Spanisch noch sonst was -- nein mir fällt's nich ein (B 14 a: 68 ff.) (... means Kutscher cart ... but isn't really a Kutscher /has with/ Benzinkutschen ... perhaps I can write house for a start ... one can just write Chauffeur ... I just don't know the word for Kutsche ... to drive a car(t) ... well there is always something with ~ horse or the like going around in my head ... I also can't remember any passage in literature where something like a coach-man might have appeared neither in Spanish nor anything else — no it doesn't occur to me)

2.2.3 The middle level: natural, but not so productive? 2.2.3.1 Formal decomposition and relexification is a phenomenon which again links pidginization and L2 acquisition and use. It is frequently attested in learner products, where the relexified form presupposes parsing, and planning, where decomposition is often quite piecemeal. 14 There can hardly be any doubt that this is a natural IL strategy. Even so, its degree of strategic

Markedness,

productivity

and naturalness

in L2 learner lexis

165

productivity will again largely depend on the closeness of the LI and L2 as well as on the quality of the decomposition Examine: (37)

Endzeitstück -»• (.7) Endzeit (.6) Zeit ->· - time /" - endtimeplay —• (Β 3:1030 ff.) Kutscherhaus (.3) ts (writing) - also Haus y wäre house τ1 ~ und Kutscher ~ hm Kutscher heißt coach / (.6) nn Kutscher vielleicht — coach driver ... also a coach driver's — (writing:) coach driver's house (B 12a: 88ff.) (... well Haus would be house and Kutscher hm Kutscher means coach a Kutscher perhaps coach driver ... well a coach driver's house)15

2.2.3.2 Onomatopoetic forms play almost no part in our data. They would be natural owing to their iconicity (though in language-specific form), but they are certainly not more than a very loose way of referring, which is therefore avoided by advanced speakers. (38)

blabla (Β 11 a: 394, 'small talk') dingdong (B 21 a: 605, 'small talk'; as 'confessed' in interview)

2.2.3.3 The use of "primary counterparts" (cf. Arabski 1979: 34) can probably not be characterized in terms of naturalness, but it is mentioned as a very frequent strategy of middle productivity, even between languages from similar socio-cultural spheres. Very often at least part of the communicative intention is preserved in the L2: (39)

post ('mail'), money ('cash'), piece ('play'), all repeatedly Endzeitstück ... produced - a pi: a play (B4: 291)

2.2.3.4 On the metalevel of knowledge about languages and translating (often of folkloric nature) the tendency to look for primary counterparts is grounded in the learner's assumption that there exist lexical equations between languages and that you either know them, or there is nothing you can do about it. It is an open question to what extent this is a natural fact or an artifact of language teaching methods. (40)

verhinderten Geldräubern \ (.3) Räubern S robber \ Geldräuber gibt's net \ (W4: 565 ff.) (doesn't exist)

2.2.3.5 Finally, there are less systematic aspects of lexical search which I call "search in the episodic memory" {cf. Tulving 1984). Whatever their different functions in search sequences, I tend to think that - although not untypical or rare - they are rather unproductive, since an individual's experience (a) will usually not be shared by potential listeners or readers, and (b) will not hold for visual memory.

166

Rüdiger

(41 a)

b)

Zimmermann

ausgewählt ... also da fällt mir — (clears throat) direkt — was - ein ... und zwar - Selektion an der Rampe ~ (writes) and selected ~ (Β 18 b: 170 ff.) (... well I just have to think of ... namely - selection at the ramp ...) Abseits ... mit diesem asides s* - wie - ähm - bei Shakespeare so was gemacht wird - auf e Bühne (B21 b: 220 ff.) ( . . . with this /word/ asides - like - this is done in Shakespeare - on the stage) Stilleben also Stilleben stell ich mir jetzt erstmal bildlich vor was das is halt η - Tisch mit Obst oder sowas drauf (B22b: 168 f.) (still life well still life I imagine visually what that is just a — table with fruit or something like that on it)

2.3 Discourse aspects of (lexical) planning in an L 2 There are many aspects of planning in an L 2 which need a lot of study under the general topic of this volume, such as its overall structure, the role of pauses, the interplay of knowledge sources in solving a problem, etc. 16 For reasons of space I will focus on verbal protocols as discourse. As has been shown elsewhere the verbalizable aspects of conscious interior planning can be characterized as the L2 user's "inner dialogue" (cf. Zimmermann-Schneider 1987b). In what follows I will concentrate on those formal and functional features which appear to be candidates for an analysis of inner dialogues in terms of markedness and perhaps of naturalness. 17

2.3.1 Fluency As a point of departure I take it for granted that fluent discourse is easier to understand and reduces the listener's (or reader's) cognitive effort because a fluent story is easier to recall; fluency is, therefore, more natural and unmarked. I tentatively transfer this interpretation of fluency to the fluency of verbalized interior planning in thinking aloud protocols. Apart from the (re)reading of L1 and L2 texts there are as a rule two major kinds of fluent passages: (a) structuring moves (usually in L1) and (b) fluent L2 phrases - "verbal routines" or "islands of reliability", around which the L2 unit is planned. 18 Examples of (a) and (b) are given in (42 a) and (42 b), respectively: (42 a)

jetz setz ich das einfach zusammen (Β 7: 350) (now I'll just put that together) übersetzen wa einfach (Β 6:103)

Markedness, productivity

b)

and naturalness in L2 learner lexis

167

(we'll simply translate that) so jetz laß uns nochma kucken (Β 6: 795) (good let's have another look now) under the influence of alcohol (B 5: 261) his place looks rather abandoned (B 9: 202) this reminds me of that (W1:191)

Lack of fluency must then be considered less natural and more marked; pauses, hesitation phenomena, repetitions, false starts, blends, etc. are signs of planning difficulties. The highest degree of hesitation seems to occur before, during, and after attempts to find proper L2 lexical items. Compare (43 a) to (43 b): (43 a) b)

(.3) a gang of twelve mostly young people with a common (.3) tendency for great -- sums of cash (W 2: 70ff., first attempt) durchsieben ? hm / (incomprehensible) hm (.25) hm (groans) (.20) (clears throat) (.20) first (.5) they were occupied (groans) (.10) to shoot (.35) durchlöchern \ (.5) to shoot holes (.10) ok (.5) wie wär's denn mit shooting holes ^ (.3) (W 2: 260 f.)

2.3.2 Language of planning In our examples L1 is certainly the preferred unmarked language of planning. I doubt, however, whether it would turn out to be typical (and natural?) in more natural kinds of bilingualism and bilingual tasks. 19 In planning passages in nonbilingual situations in which the listeners or the readers can be expected to be L 2 monolinguals, code-switching appears to be most marked. While in diglossic communities code-switching is not unnatural, in situations in which the aim is a monolingual L 2 text it can be characterized as marked, if not fairly unnatural. In the clear cases of code-switching in utterance- or sentence-size passages, some evaluative English words used as terms in the process of language teaching (e. g. awkward, familiar, clumsy) seem to appear almost exclusively in otherwise German utterances. There are also clear cases of more profound code-switching, such as the following ones: (44 a) b) c) d)

irgend so'n special term (W 9: 790) (some kind of special term) well let's see äh ja nochmal (W 5: 225) (... äh once again) die suffern ja nich (W1: 432) (they don't suffer after all) ne role kann man nich vollfillen (B8: 812) (you cannot fulfil a role)

168

Rüdiger

Zimmermann

2.3.3 Sequences of speech acts Typical sequences of speech acts such as (Self-)Request for information Answer, Proposal - Acceptance or Rejection are standard features of interior planning. As opposed to standard discourse, where acceptance is unmarked, in inner dialogues rejection or hesitation seems to be unmarked whereas acceptance is rare and marked. Note examples (45a-d): (45 a) b) c) d)

was heißt denn noch ~ ähm Eisenbahnwagen -> ~ hm (.6) carriage -> carriage glaub ich auch \ (W10: 330ff.) (what on earth is ~ Eisenbahnwagen ... carriage I think so too) die drei Polizisten ... the ~ three cups z1 hm cops policemen \ ja gut (W 4: 338fF.) (the three policemen ... yes good) auf den Boden - on the floor s nee \ on the bottom /" ~ nee \ (W9: 521 f.) Stilleben ... quiet life ->• nee das geht nich \ (B7: 515f.) (... nope that's no good)

2.3.4 Word order The language of thinking aloud protocols over longer passages is even less grammatical (in a traditional sense) than normal spoken discourse. In particular, it contains many nonsentential utterances and seemingly isolated words of an associative character. Nevertheless, there is one feature which stands out against normal sentence structure, namely the frequency of right dislocation (afterthought). The principle of end-weight seems to be much more prominent here, thereby making right dislocation considerably less marked than it appears to be in normal spoken discourse. Consider (46): (46 a) b) c)

is das'n - Kutscher s (B13a: 70) (whats that, coachman?) s'is ja hübsch Ladung (W1: 845) (it's nice, charge) gibt's das denn environmentalists (32:251) (does environmentalists?)

that

exist,

2.3.5 Use of function words Several function words, particularly but, or and and as well as their German equivalents appear in their unmarked use as logical connectors. But at the same time they function as indicators of competing plans (of lexical choice) and as a kind of boundary markers between levels of planning. Thus in: (47)

Musik — mus music — aber sich eingliedern (.3) involve (Β 22 b: 294)

Markedness, productivity

and naturalness in L2 learner lexis

169

Aber marks the transition from the translation of Musik to search for eingliedern. In: (48)

Abseits - beyond - aber hier isses ja 'η Substantiv (Β 19b: 195) (... but then it is a noun here)

however, aber is a boundary marker separating lexical search proper from metalinguistic considerations.

2.3.6 Implicit, ambiguous and unambiguous dialogical aspects 2.3.6.1 Since as yet I know very little about the dialogical character of planning beyond what our data suggests (cf. Zimmermann-Schneider 1987 b, esp. note 4), I can qualify them only by setting them against passages which are not overtly dialogical.19 Actually, our recordings abound in planning passages of an implicitly dialogical nature such as: (50 a) b) c)

inszenierte s öh - directed \ suizidgefährdet s hm - suicidal glaub ich \ (suicidal I think) able to live s nee \ (no)

2.3.6.2 There are, however, planning passages (in general and in lexical search in particular) whose dialogical character is more obvious. We have recorded clear partner-oriented or partner-inclusive forms and structures, such as utterances with wir 'we' or let's, directives with or without du 'you sg.', tags (ne? 'right?') and self-addresses with ones' names. Least marked among these are the ambiguous forms. The most frequent unmarked structures of dialogical nature are, as a matter of fact, those which can be interpreted both as statements and proposals, such as: (51 a)

schreiben jnehmen j ... wir mal Χ (let's write/take ... X)

b)

(da!also) schreiben wir mal Χ (we just write ...)

2.3.6.3 The moves with wir/uns 'we/us' can also be characterised as less marked than those with forms of du 'you sg.' and tags (mostly ne? 'right?'), such as: (52)

... effect - nehm wer des ne (Β 5: 453) (... effect, let's take that, right?)

170

Rüdiger

Zimmermann

Here again a distinction can be made between implicit reference as in (53): (53)

so mach ma'n Punkt (Β 5: 90) (ok, make a full stop here)

and explicit reference as in du 'you sg.' or a name, as in (54): (54 a) b)

superintendent ? — wie würdst de / = du/ das jetz umschreiben ( . . . ~ and how would you paraphrase that?) (W11: 695 f.) ... on Italian's (stops) on - (loud:) Anette (sound of crossing out) Italy's ... ( W l l : 4 2 8 )

\

Most marked are the dialogical sequences containing more than one of the marked forms from this short list, such as: (55 a)

b)

oh ja ästhetischen sind athetic ~ wir wird denn athetic geschreiben Anette - athetic - so jetzt hast du's Stilleben - oh good gracious (.3) Stilleben ? - äh das sind doch Landbilder s -- die also Bilder - die irgend - die ahm — ja ahm - nee wieso kommste auf Landleben s - das is - du ähm ~ Stilleben das sind irgendwie - das sind erstmal pictures ne ? — so (B 17b: 185ff.) (oh yes ästhetischen are athetic ~ how is athetic spelt Anette ... good now you've got it Stilleben ... uh they are country-pictures ~ which well pictures which some - which uh - yes uh ~ no what makes you think of rural life - that's - you uhm ~ Stilleben they are somehow - they are first of all pictures, right ~ good)

3. Conclusion

This first attempt to deal with typical features of one kind of L 2 production in terms of naturalness, markedness and related phenomena has in many respects turned out to be an illustration of problems, rather than a solution. Nevertheless, one conclusion can be drawn: learners must be made aware of and encouraged to prefer those IL-typical strategies which potentially result in unmarked and productive approximations in L2 terms and correspond to natural phenomena. Such strategies will be transferable to other languages. As a corollary to this learners also need to become thoroughly familiar with IL-typical strategies resulting in (less natural) L 2-productive approximations which will normally not be transferable. Transcribing conventions - < 1 second ? = question intonation — = 2 seconds -> = nonfinal intonation

Markedness, productivity and naturalness in L2 learner lexis

171

(.χ) = χ seconds \ = final (statement) intonation underlined syllable = strong stress /.../ = not clearly audible, reconstructed Notes 1. All examples and quotations are from the ALES corpus. 2. Note that such introspective learner data are a better basis for determining the IL rules actually applied than a 'surface' characterization as e.g. in Odlin 1986:161. 3. Note that I have made sure that the learner forms cited are not slips of the tongue or the pen; all of them appear more than once with the same learner. 4. It is obvious from this approach that I share the view of those who see markedness as a scalar phenomenon, (cf. Odlin 1986: 164). 5. This can be formulated more explicitly following Marchand's (1974: 242if.) categories of 'semantic dependence' and 'range'. 6. Sources are given only for larger and individual examples. 7. I have tried to elaborate the notion of productive lexical strategy further in Zimmermann (forthcoming). 8. My short list is based on what seems to be recurrent in recent research, e. g. in Ringbom 1983; Paribakht 1985 and our own data. 9. For this condensation process consider also passage (1) above (containing to bullet, to hole, gun holes and bullet holes). 10. The preference for avoidance over form-orientation may of course not hold for closely related pairs of languages where putative false friends are in reality L 2 equivalents. 11. Of course, this tendency to start from some kind of base form may be a language teaching device. 12. For a more detailed analysis of content- and form-orientation in lexis cf. Zimmermann 1987 a. 13. For (lexical) transfer between cognate languages cf. e.g. Kellermann 1977; Ringbom 1983. 14. Cf e.g. Dechert-Raupach 1985: 264; not to be confused with the developmental notion of decomposition (cf. Wode 1984: 356). 15. I am aware that the L1 form is not fully relexified here, but is rather rendered in what is usually called "analytical translation". 16. For a first treatment of several phenomena in translation see Krings 1986. 17. The phenomena are arranged from more comprehensive textual features to more isolating formal ones. 18. Cf. e.g. Dechert-Raupach (1985, esp. 244-263), who show that such islands can become firmer as the task proceeds; see also Keseling (in press). 19. In L2 planning there are some typical chunklike L2 phrases such as let's see, let's say, whatever X means, let's try that, sounds good/bad to me, etc.

172

Rüdiger Zimmermann

References Adams, Valerie 1973 An introduction to Modern English word-formation (London: Longman). Arabski, Janusz 1979 Errors as indications of the development of interlanguage (Katowice: Uniwersytet Slijski). Bauer, Laurie 1983 English word-formation (Cambridge: University Press). Blum, Shoshana -Eddie Levenston 1978 "Universals of lexical simplification", Language Learning 28:399-415; also in Faerch/Kasper 1983:119-138. Braunwald, Susan 1978 "Context, word-formation and meaning: towards a communicational analysis of lexical acquisition", in Lock 1978:485-527. Chao, Wynn 1986 "Indefinite NPs and the interpretation of discourse based null elements", in Eckman et al. 1986: 65-84. Dechert, Hans - Manfred Raupach 1985 "Hypothesen zur Zweitsprachenproduktion", Lernersprache. Thesen zum Erwerb einer Fremdsprache, ed. by Ralf Eppeneder (Munich: Goethe-Institut), 219-288. Eckman, Fred, Edith Moravcsik and Jessica Wirth (eds.) 1986 Markedness (New York: Plenum Press). Faerch, Claus - Gabriele Kasper (eds.) 1983 Strategies in interlanguage communication (London: Longman). 1987 Introspection in second language research (Clevedon: Multilingual Matters). Haastrup, Kirsten - Robert Phillipson 1983 "Achievement strategies in learner/native speaker interaction", in Strategies in second language research edited by Faerch-Kasper 1983:147-200. Kellerman, Eric 1977 "Towards a characterization of the strategy of transfer in second language learning", Interlanguage Studies Bulletin 2: 58-145. 1979 "The problem with difficulty", Interlanguage Studies Bulletin 4, 27-48. Keseling, Gisbert forthcoming "Pausen und Routinen beim Verfassen von Sachtexten", to appear in Multilingua. Krings, Hans 1986 Was in den Köpfen von Übersetzern vorgeht. Eine empirische Untersuchung zur Struktur des Ubersetzungsprozesses an fortgeschrittenen Französischlernern (Tübingen: Narr). Lapointe, Steven 1986 "Markedness, the organization of linguistic information in speech production, and language acquistion", in markedness, edited by Eckmann et al. 1986: 219-239. Lock, Andrew (ed.) 1978 Action, gesture and symbol. The emergence of language (London: Academic Press). Marchand, Hans 1969 The categories and types of present-day English word-formation (Munich: Beck). 1974 "A set of criteria for the establishing of derivational morphemes", Studies in syntax and word-formation, ed. by Dieter Kastovsky (Munich: Fink), 242-252. Mayerthaler, Willi 1981 Morphologische Natürlichkeit (Wiesbaden: Athenaion) 1986 "Markedness", lecture presented at the 19th Annual Meeting of the Societas Linguisticae Europaeae, Ohrid, Yugoslavia.

Markedness, productivity and naturalness in L2 learner lexis

173

Moerk, Ernst 1977 Pragmatic and semantic aspects of early language development (Baltimore: University Park Press). Mühlhäusler, Peter 1983 "The development of word-formation in Tok Pisin", Folia Linguistica 17:463-487. Nokony, Alicia 1978 "Word and gesture usage by an Indian child", in Lock 1978: 291-307. Odlin, Terence 1986 "Markedness and the zero-derived denominal verb in English: synchronic, diachronic, and acquisiton correlates", in Eckman et al. 1986:155-168. Paribakht, Tahereh 1985 "Strategic competence and language proficiency", Applied Linguistics 6:132-146. Ringbom, Hakan 1983 "Borrowing and lexical transfer", Applied Linguistics 4: 207-212. Rutherford, William 1982 "Markedness in second language acquisition", unpublished paper. Tulving, Endel 1984 "Precis of 'Elements of episodic memory'", The Behavioral and Brain Sciences 7: 223-268. White, Lydia 1986 "Markedness and parameter setting: some implications for a theory of adult second language acquistion", in Markedness, edited by Eckman et al. 1986: 309-327. Wode, Henning 1984 "Psycholinguistische Grundlagen sprachlicher Universalien: Möglichkeiten eines empirischen Paradigmas", Folia Linguistica 18: 345-377. in press "Einige Grundzüge des natürlichen L2-Erwerbs des Wortschatzes", 12. Arbeitstagung der Fremdsprachendidaktiker, ed. by Günther Nold (Tübingen: Narr). Zimmermann, Rüdiger 1986 "Classification and distribution of lexical errors in the written work of German learners of English", Papers and Studies in Contrastive Linguistics 21, 31-41. 1987a "Form-oriented and content-oriented lexical errors in L 2 learners", International Review of Applied Linguistics 25: 55-67. 1987b "Lexical search in an L2 as impeded communication", Folia Linguistica 21:211-224. in press "A partial model of lexical search in L 1 - L2 translation", Interlingual Processes, ed. by Hans Dechert (Tübingen: Narr). forthcoming "Productive and unproductive lexical strategies in L 2 discourse", Paper, International Society of Applied Psycholinguistics, Second International Congress 1987, Kassel/Germany. Zimmermann, Rüdiger-Klaus Schneider 1987 a "The collective learner tested: retrospective evidence for a model of lexical search", in Faerch-Kasper 1987:177-196. 1987b "Dialogical aspects of individual lexical search", Multilingua 6:113-130.

Markedness and grammaticalization Christian Lehmann

0. Introduction

In discussions of grammaticalization, the terminology of markedness has made repeated appearances. A sign affected by grammaticalization has been said to become less marked or to undergo demarking. The marking terminology has not however, always been used with strict definitional implications. We should therefore not wonder why terms that have been on the linguistic market for a long time are used to express the relatively novel ideas that are behind grammaticalization. However, we do not need just another fashionable rephrasing of concepts for which there are precise terms available. If this were all that there is to the relation between grammaticalization and markedness, we might as well dismiss this issue without more ado. On the other hand, given the grammar-theoretical environment to which both of these notions belong, one would in fact expect relations between them. Both grammaticalization and markedness relate to subsystems of linguistic signs, both in the grammar and in the lexicon. Their locus is in inflectional morphology; they are intimately connected with the notion of the paradigm. Given this shared background, it might seem worthwhile to investigate into the nature of their relationship.

1. Markedness

The notion of markedness presupposes the notion of "opposition". If two linguistic elements A and Β occur in identical contexts, but contrasting in function or meaning, they are in opposition. Such elements may be composed

176

Christian Lehmann

of smaller constitutive elements. Given this, "markedness" may be defined as follows: D1.

Let the set of elements constituting A include the set of elements constituting Β plus an additional element e. Then the opposition between A and Β is a "privative" or "markedness "opposition". A is "marked" as against B, e is its mark.

As 1) is meant to illustrate, the English colt contains all the semantic features of horse plus a feature [non-adult]; therefore colt is marked as against horse. (la) b)

horse: [equine] colt: [equine] [non-adult]

While the above is a possible definition of markedness, it is not a methodological criterion, at least not a primary one. Obviously, given the opposition between horse and colt with the differentiating feature [ + / — adult], there is no a priori way of deciding whether colt possesses an additional feature [nonadult] or rather whether horse possesses an additional feature [adult]. The chief operational criterion in establishing a marking relationship between two elements is an implicational one ( c f . Zwicky 1978): D2.

Let there be a binary feature [ + /— f] and two corresponding categories of linguistic elements, defined by [af] and [— af]. Then [af] constitutes a "mark" as against [ — a f ] if for any subcategory [ßg] of [af], there is a corresponding subcategory [ßg] of [— af], but not necessarily vice versa. Correspondingly, an element A belonging to category [af] is "marked" as against an element Β belonging to category [ — af]·

Speaking loosely, the unmarked term exhibits at least as much diversification in terms of subcategories as the marked one. Take (1) again. The feature [f] in question is [ + /— adult]. Within the category [ + adult], there is a subcategorization according to [ + /— female] (instantiating [g] of the definition). In the case of horse, this yields stallion and mare. Within the category [— adult], there is no such corresponding subcategorization, i.e. there are no words for 'female colt' and 'male colt'. Consequently, [— adult] constitutes a mark as against [ + adult]. This is shown in (2) (2 a) b) c)

stallion: mare: colt:

[equine] [equine] [equine] [non-adult]

[male] [female]

In the given context we are dealing with language signs, so we need not detain ourselves with phonological markedness. In morphology, the component elements are morphemes, the features are morphological categories and

Markedness and grammaticalization

177

subcategories. On the morphological level, there is a tendency towards "iconicity" {cf. Haiman 1980) or, as the older term used to be, towards "isomorphism" (cf. Lehmann 1974) between the content and the expression sides. That is, if there is an extra semantic or functional unit in a given structure, it tends to have an extra expression unit associated with it, and vice versa. This is not the case in the pair horse - colt, which belongs to the lexicon. It is, however, the case in pairs such as horse - horse-s, book - book-let etc. Therefore D1 and D2 make reference to "constitutive elements" or "features", respectively, without differentiating between the content and expression sides. Again, to the degree that such iconicity fails, the definition does not provide an operational criterion of markedness. In connection with our present objectives, it will be good to be aware that the above definition is essentially an inversion of what has come to be known as Brondal's principle of "compensation". The principle was named thus by the Danish linguist V. Brendal (1940:107) and may be formulated as in D3: D3

Let there be a linguistic category with A as the marked and Β as the unmarked subcategory. Then Β will exhibit as much or more differentiation in terms of another linguistic category as A.

An example is provided by the tense and mood categories in Latin. Of the primary tenses, the future is most marked. Now the present, as an unmarked tense, is differentiated according to at least two moods, namely indicative and subjunctive. The future, however, has only an indicative one. Another way of putting it would be: the mood paradigm is smaller within the future than within the present tense; or, more generally, the paradigm of a second category is smaller in the marked than in the unmarked subcategory of a first category. We shall see below how this principle serves to put the relationship between markedness and grammaticalization straight.

2. Grammaticalization Traditionally, grammaticalization has been conceived as a diachronic process transforming lexical into grammatical signs. Given that any diachronic change is mirrored in a preceding synchronic variation, this notion is easily widened to comprise both synchronic and diachronic variation of a certain sort.

178

Christian Lehmann

The notion of grammaticalization thus presupposes the notion of "variation", as well as those of basic and derived variant. These are not easily defined, although the idea is clear enough: A linguistic element A is a variant of an element Β if A and Β are in some functional sense identical, but differ in aspects of their expression or their content or both. In this sense, Engl, α is a synchronic variant of an, and the is a diachronic variant of that. On the diachronic axis, the basic and the derived variant are easily distinguished as the earlier and later one, respectively. On the synchronic axis, we will take that variant to be basic which presupposes the other one. 1 Grammaticalization can now be defined as follows: D4.

Let A and Β be language signs. A is grammaticalized to Β if Β is a synchronically or diachronically derived variant of A and more integrated into the grammatical system than A.

Two clear examples may be found in (3) and (4): (3)

truwe-liko > truly true like 'in a true form'

The English adverbial suffix -ly evolved from the Germanic noun *llka 'body, shape' in the ablative, *tiko 'in a form', so that tru-ly derives from *truwe-liko 'in a true form'. (4)

*nasjan

dedum > nasi-

save (inf.) did we 'we did save'

dedum

save (pres.) did we 'we saved'

The English past tense suffix -d derives from *-de, a form of Germanic *dön 'do'. In Gothic, we find forms such as nasi-d.ed.um 'we saved', apparently derived from the combination of the infinitive nasjan 'save' with dedum 'we did'. Just as in the case of markedness, the nominal definition of grammaticalization offered in D 4 may appear to raise problems since we do not know how to verify whether A is more integrated into the grammatical system than B. Therefore this definition should be associated with a set of operational criteria which afford just that. In this sense, grammaticalization is a cover term for a number of constitutive processes that tend to go hand in hand. As the examples show, three processes may be recognized, each of which may be split up into a pair according to its paradigmatic and syntagmatic aspects: (a)

The sign itself - both its expression and its content - shrinks while its syntagmatic scope, i. e., the grammatical level and the construction on which it operates, is reduced.

Markedness and grammaticalization

(b)

(c)

179

*liko forfeits both concrete semantic features and part of its expression on its way to -ly. Also, while the noun could take a complex prenominal modifier, the suffix can only be preceded by a single adjective. Similarly, the Germanic verb form *de meant 'did'. On its way towards -d, it lost all its nongrammatical features and half of its expression. The sign is integrated into a grammatical paradigm and into its syntagmatic context. While lika was not part of any paradigm, the suffix -ly is integrated into the paradigm of adjectival suffixes. It is an affix of the adjective, while Ilka was an independent word. Again, while the form *3e may have been part of an open semantic field of semi-aspectual verbs, -d is tightly integrated into the small paradigm of verbal endings. And while the verb form is an independent verb, the past tense marker is an affix. The sign and its paradigm become increasingly obligatory, and its position in the syntagm becomes fixed. While in Germanic, one was free to combine an adjective with any noun that made sense, in English one has to append the adverbial suffix according to rules of syntax. Also Germanic allowed both for prenominal and postnominal position of adjectives; but in the adverb, the adjective has to precede the suffix. While *de and similar verbs could be chosen according to semantic demands, the choice of the past tense suffix and its whole paradigm is subject to rules of grammar. The same goes for the position of these elements, which was comparatively variable for the Germanic verb form, but fixed for the past tense affix.

From these three pairs of processes we may abstract three pairs of criteria each having a paradigmatic and a syntagmatic member - , by which we may compare two signs which are variants of each other. Our operational definition may then run as follows: D 5.

Let A and Β be two language signs, A being a derived variant of B. Then A is more grammaticalized than Β if it differs from Β by the following set of criteria: i. A displays less phonological and semantic integrity and less syntagmatic scope than B. ii. A displays greater paradigmaticity and syntagmatic bondedness than B. iii. A displays less paradigmatic and syntagmatic variability than B. These criteria correlate at least to the degree of yielding compatible results.

As in the case of markedness, the criteria rely on a certain amount of iconicity in grammar. None of them makes exclusive reference to either the expression or the content side of the language sign; instead, they refer to the sign as a whole.

180

Christian Lehmann

3. Markedness and grammaticalization 3.1 Static and dynamic concepts Before we can go into the content of the two concepts, a methodological problem must be eliminated. The notion of markedness has evolved in the structuralist context and thus reflects an essentially static view of language. The notion of grammaticalization has evolved in the traditions of historical linguistics and evolutive typology and is thus grounded in an essentially dynamic view of language. Markedness designates a certain state of affairs, while grammaticalization designates a process. The two concepts might therefore appear to be incommensurable from the start. Here I will take the following view: If a conception of language is static, it is inadequate and has to be substituted by a dynamic one. However, the concepts that it incorporates may nevertheless be valuable and then have to be reinterpreted on the basis of the dynamic conception. Consequently, concepts such as opposition and markedness have to be seen as constraints that the speaker imposes on his choice of elements. He may contrast two elements by confining them to the same class of contexts, and he may oppose a marked term to an unmarked one by adding a mark to it.

3.2 Demarking 3.2.1 Desemanticization and demarking Our definition of markedness makes reference to an additional component in the marked term, which may be a semantic feature. The opposition of an unmarked to a marked term might therefore be conceived as the subtraction of a semantic feature. Our definition of grammaticalization involves, among other things, the semantic integrity of the language sign. A grammaticalized sign contains fewer semantic features than the sign it derives from (criterion ι in D5). So here we seem to have a close parallel between markedness and grammaticalization. Consider the opposition between the and that as an example (cf. Traugott 1982: 250) This might be analyzed as a markedness opposition, based on the common feature of definiteness. That would be marked by containing an additional feature of non-proximal deixis, which is mirrored by its somewhat fuller expression.

Markedness and grammaticalization

181

Now it so happens that the derived from that by grammaticalization in Middle English, as shown in (5). Clearly, some phonological substance was lost - precisely that semantic feature which makes that a non-proximal demonstrative, - so that only the definiteness remained. (5)

that [definite] [deictic] > the [definite]

On the basis of such examples, it seems understandable that grammaticalization has been described as demarking. However, once we widen our perspective a bit, we start to see the problems behind the conception. First of all, the application of the notion of markedness to a pair of elements A and Β presupposes that they be in a binary opposition. This is not true for the and that. The forms the paradigm of the articles together with a, while that forms the paradigm of the demonstratives together with this and yonder. The two paradigms are in opposition. The markedness opposition of two selected elements of them is a derivative, contingent fact. Take again the example of Gothic nasidedum. There is supposed to have been a stage where this was in synchronic variation with something like *nasjan dedum 'we did save'. Here, I think, it is clearer that while the grammaticalization of *nasjan dedum to nasidedum did in fact involve desemanticization, *nasjan dedum and nasidedum do not form a markedness opposition. It is hard to speak of demarking in any terminologically precise sense if there is no unmarked counterpart to the term being demarked. In fact, if demarking were understood to mean 'loss of a mark, i. e. transformation of a marked term into its unmarked counterpart', and if grammaticalization involved demarking in this sense, it would consist in the reduction of oppositions to their unmarked members. This is not at all what grammaticalization does; quite on the contrary, it creates new paradigms, as we shall see in section 3.3.

3.2.2 Selection restrictions and demarking There is a distinct, but related sense in which one might speak of demarking as a factor in grammaticalization. The obligatoriness of a term increases with grammaticalization. This is intimately tied up with the loosening of selection restrictions which the grammaticalized item imposes on its context. One example of this would be the extension of object agreement on the verb, which moves down the animacy hierarchy: first the verb agrees only with human, definite objects, then the object may be either definite or human, then it suffices for it to be an animal, then it need only be an individual, and in the end

182

Christian Lehmann

the verb agrees with every object. This process can currently be observed in Amharic and, to a certain degree, in Spanish. A similar example is provided by the extension of the preposition a for the expression of the object relation in Spanish. This takes exactly the same course, as is illustrated in (6). (6 a)

b)

c)

Conozco *(a) esse hombre. know I to this man Ί know this man.' Conozco (?a) ese bicho. know I to this creature Ί know this creature.' Conozco (*a) ese tarea. know I to this task. Ί know this task.'

Here, grammaticalization may be conceived as demarking if the selection restrictions inherent in the grammaticalized item are subsumed under its semantic features and conceived of as marks. However, it still does not correspond to D 2, which relies on the implicational relationship between the marked and the unmarked term. Applying this to the example at hand, we would rather say that the use of an originally dative preposit.i.^ with direct objects on a low position of the animacy hierarchy implies its use with direct object higher up on the hierarchy (cf. Bossong 1985). This is correct as far as it goes. On this basis, however, the use of a would be least marked in (6 a) and most marked in (6 c), so that the movement from (a) to (c) would be marking instead of demarking, if anything. I therefore conclude that this talk of demarking in grammaticalization is also not useful.

3.2.3 Paradigmaticization and demarking Finally, consider the conceptual interaction of the compensation principle and the parameter of paradigmaticization. As we saw above, the compensation principle says that the paradigmatic differentiation of an unmarked subcategory is greater than or at least equal to the differentiation of the marked subcategory. Paradigmaticization, on the other hand, means that a more grammaticalized paradigm is more tightly integrated than a less grammaticalized one. Integration of a paradigm implies, among other things, a relatively small number of members. For instance, the English demonstratives outnumber the articles; the paradigm of the French object agreement

Markedness and grammaticalization

183

clitics is larger than that of the more grammaticalized subject agreement suffixes. Cf. also section 3.4 on the consolidation of paradigms. Now suppose the relation of a less grammaticalized to a more grammaticalized item were, at the same time, one of a marked to an unmarked item. Then, according to paradigmaticization, the paradigm of the latter item would have to be smaller than the one of the former, but, according to the compensation principle, it would be larger than the paradigm of the former. Thus, paradigmaticization and compensation would contradict each other. Since both principles are empirically well founded as long as we deal with clear, prototypical cases of grammaticalization and markedness, respectively, we may conclude preliminarily that the interpretation of a grammaticalizational relationship as a markedness relationship introduces incoherence into the overall theoretical framework.

3.3 Markedness reversal When a grammatical category is renewed, the successor necessarily takes some of the field formerly occupied only by the more traditional category. Take the Latin personal and demonstrative pronouns of (7) as an example. (7 a) b)

is 'he' hie

(Dl)

'this one' iste

( D 2)

'that one' ille

(D3)

'that one' In Classical Latin, is belongs into a paradigm with the other personal pronouns ego Τ and tu 'you', while ille is the deictically least marked member of the demonstrative set. The relation of is to ille is thus rather similar to the relation between English the and that. However, in contradistinction to the English personal pronoun, is could not be used deictically. This function was regularly fulfilled by ille {cf. Pinkster 1986). In colloquial Latin, the whole pronominal system was restructured. For one thing, the demonstrative hie was lost, probably in connection with the utter irregularity of its paradigm. For another, ille gradually took on all the functions of is, becoming the normal 3rd person pronoun. Given that, among other things, grammaticalization involves increasing obligatoriness of the item in question, it follows that it also normally involves an increase in frequency. While the original 3rd person pronoun is was ousted, it became

184

Christian Lehmann

oldfashioned, high-styled and obsolete. Its use instead of ille increasingly acquired a striking quality. Consequently, is might now be called marked as against ille. Thus, the relationship between the two terms seems to be exactly reversed. Consequently, some authors speak of "markedness shift" or "markedness reversal" in this connection (e.g. Dik 1978: l l l f ; cf. also Givon 1979: 75f). A parallel example from German is offered by the genitive and its successor, the von-phrase, as in (8). (8 a) b)

der Chef meines Mannes the boss my (Gen.) husband (Gen.) 'My husband's boss' der Chef von meinem Mann the boss of my (Gen.) husband 'the boss of my husband'

Originally, only the genitive was admissible in nominal attributes. Later, the preposition von 'from' was grammaticalized and acquired the function of E. of, being used in nominal attribution instead of the genitive. At first, von was more expressive in attribution than the mere genitive. Nowadays, the genitive is becoming increasingly old-fashioned, and we might again speak of markedness reversal. The idea of markedness reversal as a switch occurring inside a paradigm and, thus, as a leap in linguistic history is in itself not very attractive. Moreover, we easily recognize here the loose usage of marking terminology. It is true that markedness has been associated with frequency, and justifiably so. However, mere frequency in itself is never a sufficient criterion for a certain structural analysis. There is an obvious difference between the pair horse colt, whose second term is less frequent because it contains an additional semantic feature and therefore, marked in the strict terminological sense, and the pair ille - is in Vulgar Latin, whose second term is less frequent because it is obsolete, but which otherwise does not conform to any of the markedness criteria. I therefore conclude that the term "markedness reversal" is not well applied in grammaticalization. While it is true that the frequency relationship between the traditional and the innovated term is reversed during the historical change in question, their paradigmatic relation is not directly affected by this in the sense that the traditional term acquired a mark formerly possessed by the innovated term.

Markedness

and grammaticalization

185

3.4 The consolidation of paradigms A set of oppositions involving the same elements in the same contexts constitutes a paradigm. Consider again Proto-Gothic *nasjan dedum > Gothic nasi-dedum (e). At the time when only the former variant existed, this was a periphrastic locution which may have been in loose association with other verbal periphrases. At the Gothic stage, nasidedum is integrated into the inflectional paradigm and forms a markedness opposition with nasjam 'we save'. The same goes for the Germanic articles. The numeral 'one', the source of the indefinite article, and the demonstrative 'that', the source of the definite article, were just two words that could appear in the same position, but did not form a paradigm. In the course of their grammaticalization, first the demonstrative, then the numeral started to form a privative opposition with the absence of determination. Cf., for the definite article, (9) (from Ramat 1980:1200, and for the indefinite article, (10), all from the Old Saxon Heliand (ca. 830). (9 a)

b)

(10 b)

b)

Thar ina thiu modar fand (Hel.818) there him that mother found 'There his mother found him.' fand thar barn gesund (Hel.) found there son healthy '(he) found there the son healthy.' Ina antleddun thanen drohtines engilos endi him led (pi.) thence God's angels and is dohter twa an enan berg uppen his daughters two to a (Acc.) mountainup 'Him and his two daughters led thence God's angels up to a mountain' Thuo sia thar an griete galgon rihtun (Hel., Aus der Kreuzigung) then they there on sand gallows set (pi.) 'Then they set up gallows there on the sand.'

At this stage, grammaticalization has created two markedness oppositions. Together they form a paradigm of three terms, namely 'definite determiner vs. zero vs. indefinite determiner'. Grammaticalization continues, and the result is an equipollent opposition between the definite and the indefinite article. This paradigm becomes increasingly obligatory and can only be left out under well-defined conditions. As another example, we may briefly consider the category of "aspect" in some languages. In several Romance languages, periphrastic aspects have been formed, as illustrated by Spanish in (11).

186

Christian Lehmann

(IIa)

b)

c)

d)

voy a cantar go (1st sg.) to sing Ί will sing' estoy cantando be (1st sg.) singing Ί am singing' he cant ado have (1st sg.) sung Ί have sung' acabo de cantar finish (1st sg.) to sing Ί have just finished singing'

This system is not yet very tightly integrated, and the oppositions do not seem to be binary. We will come in a moment to the possibility of assuming a privative opposition between each of the forms in (11) and the simple present tense canto Ί sing'. In Russian, there is a paradigm of two aspects, the perfective and the imperfective, as illustrated in (12). (12a)

b)

Cto ze delal Bel'tov ν prodolzenie what then (interj.) did (imperf.) Beltov in continuation etix desjati let? these (Gen.) ten years 'What then did Beltov in the course of these ten years? Cto on sdelal? Nicevo, ili pocti nicevo? what he did (perf.) nothing or almost nothing 'What did he achieve? Nothing or almost nothing.'

By all the criteria introduced in section 2, this system is more grammaticalized than the Spanish one. Here there is a clear privative opposition with, in this case at least,2 "perfective" as the marked term. The point here is that grammaticalization leads to the creation and integration of paradigms. In its course, the chances for two formerly unrelated terms to enter into a markedness opposition are furthered. With grammaticalization going on, more terms are moved into direct oppositions to each other, and former markedness oppositions may give way to equipollent oppositions.

Markedness and grammaticalization

187

3.5 The stratification of paradigms When a semantic field of lexical items is grammaticalized to a loosely integrated paradigm of grammatical words, such as prepositions, auxiliaries, determiners and the like, there may exist, at the same time, an older, more grammaticalized paradigm representing a semantically related category. This is the case of the Spanish aspectuals, which hit upon the synthetic tense system that has long been in existence. Another example are the passive auxiliaries in Italian. Traditionally, the passive was formed with essere 'to be', as in (13). This verbal form is in a paradigm with the other verbal voices, namely the active dice 'says' and the reflexive si dice 'is said'. (13)

Questo e detto senza later implicazioni. this is said (partic.) without further implications 'This is said without further implications.'

More recently, another way of forming the passive has spread in the language, which involves andare and venire as auxiliaries, as illustrated in (14). (14a)

b)

Questo va detto senza later implicazioni. this goes said (partic.) without further implications Ί say this without further implications.' Questo viene detto senza later implicazioni. this comes said (partic.) without further implications 'They say this without further implications.'

This periphrasis exploits the deictic potential of these verbs: andare designates a movement originating at the speakers site, venire designates a movement directed towards the speaker. Thus, the two analytic forms in 14 a, b come to mean 'it is said by me' vs. 'it is said by somebody else' respectively. They form a binary opposition, whose exact nature is yet to be analyzed. It may be equipollent, or it may be privative, with andare as the marked term. The situation is thus similar to the Spanish aspectuals: By the grammaticalization of the verbs andare and venire, a paradigm based on a binary opposition is formed. As a whole it contrasts with an already existing category, the passive with essere, which is slightly more grammaticalized. The effect is the same in both cases: we get a stratification of paradigms representing functionally similar grammatical categories. As a last example, we may again consider the English articles and demonstratives. These are two paradigms of forms which are functionally similar, occur in the same position and therefore contrast with each other.

188

Christian Lehmann

Although here it is not the case that the demonstratives were grammaticalized chronologically after the articles, it is still true that they are on a less advance stage of grammaticalization. In all of the examples, the problem is the same. Given the presupposition that the paradigms are semantically similar and contrast with each other, they contract in a paradigmatic relationship with each other. As a consequence, a term belonging to one paradigm enters into a mediate opposition to a term belonging to the other paradigm. Now this mediate opposition may be reinterpreted as a direct one, so that not only the paradigms as a whole, but also the single forms may appear to contract opposition. That is, not only the group of demonstratives and that of the articles but also the demonstrative that and the article the are in opposition. Not only the standard Italian voices are in opposition with the deictic passives, but also the auxiliary venire with the auxiliary essere. Furthermore, since the paradigms exhibit different stages of grammaticalization, it follows that the forms of the less grammaticalized paradigm are richer in semantic features than those of the more grammaticalized paradigm. This means that the direct cross-paradigm oppositions may be interpreted as markedness oppositions. I suggest that this is why some people conceive of grammaticalization relationships as markedness relationships. A t the same time, it is clear how we may gain precision here. There is a difference between the choice that a speaker makes among the different forms within a paradigm, on the one hand, and the choice among different paradigms, on the other. Obviously, the latter is a choice on a higher level, it is presupposed by the former choice. The freedom that the speaker enjoys in his choice is greater on the higher level that on the lower level. For instance, he has some freedom to choose, according to his semantic exigencies, either a demonstrative or an article. However, once he has chosen a category, the selection of the correct subcategory is less free and more subject to rules of grammar, this again differing depending on the degree of grammaticalization of the paradigm in question. If we make this distinction between the intraparadigmatic and the interparadigmatic levels, we may speak of different degrees of grammaticalization on the interparadigmatic level and restrict the use of markedness terminology to the intraparadigmatic level. The riddles posed by the combination of grammaticalization with markedness then dissolve. In particular, the contradiction emerging from the combination of Br0ndal's compensation principle with paradigmaticization disappears: the compensation principle is not applicable to paradigms of different degrees of grammaticalization.

Markedness and grammaticalization

189

4. Conclusion

Consideration of the facts of grammaticalization strongly argues for a restricted use of markedness terminology. In particular, nothing is gained if markedness is based on text frequency, since this is influenced by factors not directly related to meaning. Also, markedness should not be applied to any two terms one of which gives more information than the other. Instead, in a dynamic framework based on language as an activity, markedness should be seen as pertaining to the choice of the speaker within a paradigm, whereas grammaticalization pertains to the choice among different paradigms.

Notes 1. Within a theory of grammar, the notions of basic and derived variant may be defined with respect to the functioning of the grammar, in particular to a derivational rule type in the grammar. However, in our context, such a definition would beg the question since we are not concerned with theory of grammar, but with theory of language. Any theory of grammar presupposes a theory of language. Thus, any rule type in a grammar must reflect some language operation. 2. Other verbs have the imperfective formally marked. The details of the analysis of the Russian aspects are of no concern in the present context.

References Bossong, Georg 1985 Empirische Universalienforschung·, Differentielle Objektmarkierung in den neuiranischen Sprachen ( = AL 14) (Tübingen: Günter Narr). Brendal, Viggo 1940 "Compensation et variation, deux principes de linguistique generale", Scientia: 101-116. Reprinted in Viggo Br0ndal, Essais de linguistique generale, publie avec une bibliographie des oeuvres de l'auteur (Kopenhagen: Munksgaard): 105-116. Dik, Simon C. 1979 Functional Grammar ( = North-Holland Linguistic Series 37) (Amsterdam etc.: North-Holland). Givon, Talmy 1979 On Understanding Grammar (Perspectives in Neurolinguistics and Psycholinguistics) (New York etc.: Academic Press). Haiman, John 1980 "The iconicity of grammar; isomorphism and motivation", Language 56: 515-540. Lehman, Christian 1974 "Isomorphismus im sprachlichen Zeichen", in Linguistic Workshop II, Arbeiten des Kölner Universalienprojekts 1973/74 (Struktura 8), edited by Η. Seiler (München: Fink), 98-123.

190

Christian Lehmann

Pinkster, Harm 1986 The Pragmatic Motivation for the Use of Subject Pronouns in Latin: The Case of Petronius (Amsterdam: University of Amsterdam, Department of Latin). Ramat, Paolo 1980 Introduzione alia linguistica germanica ( = Linguistica Generale e Storcia 16) (Bologna: Patron). Traugott, Elizabeth Closs 1982 "From propositional to textual and expressisive meanings; Some semanticprgamatic aspects of grammaticalization", in Directions for Historical Linguistics II, edited by W.P. Lehmann and Y. Malkiel (Amsterdam: Benjamins): 245-272. Unbegaun, B.O. 1969 Russische Grammatik (Göttingen: Vandenhoeck and Ruprecht). Zwicky, Arnold M. 1978 "On markedness in morphology", Die Sprache 24:130-143.

On the assessment of the markedness status of the exponents of a grammatical category Olga Miseska Tomic

1. The scope of markedness

In classical Praguian linguistics "marked" meant morphonologically complex. Trubetzkoy's initial (1931) distinction between "mark" and "distinctive property", which in his Grundzüge (1958 (1939)) transpires through the use of the term "unmarked" both in reference to a member of the opposition "marked/unmarked" and to a form which shows up in neutralization, was subsequently blurred. The notion of "markedness" became unseparable from the relation coded in the terms "marked" and "unmarked" and came to be treated strictly formally - as an abstraction over the binary opposition marked/unmarked, made by a simple count of features. When the feature analysis was introduced into syntax and semantics and the notion of markedness extended over the entire domain of linguistic analysis, the simple correlation marked