Contemporary Morphology 9783110874082, 9783110123494

Citation preview

Contemporary Morphology

Trends in Linguistics Studies and Monographs 49

Editor

Werner Winter

Mouton de Gruyter Berlin · New York

Contemporary Morphology Edited by

Wolfgang U. Dressler, Hans C. Luschützky, Oskar E. Pfeiffer, John R. Rennison

Mouton de Gruyter Berlin · New York

1990

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

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

Library of Congress Cataloging-in-Publication

Data

Contemporary morphology / edited by Wolfgang U. Dressier ... [et al.]. p. cm. — (Trends in linguistics. Studies and monographs ; 49) "Selected papers from the Third International Morphology Meeting held, under the auspices of the International Association of Morphology, in Krems (Austria) from July 4 to July 7, 1988" - Pref. Includes bibliographical references and indexes, ISBN 0-89925-663-5 (acid-free paper) 1. Grammar, Comparative and general — Morphology — Congresses. I. Dressier, Wolfgang U., 1939— , II. International Morphology Meeting (3rd : 1988 : Krems an der Donau, Austria) III. International Association of Morphology. IV. Series. P241.C66 1990 415 —dc20 90-42745 CIP

Deutsche Bibliothek Cataloging-in-Publication

Data

Contemporary morphology / ed. by Wolfgang U. Dressier ... — Berlin ; New York : Mouton de Gruyter, 1990 (Trends in linguistics : Studies and monographs ; 49) ISBN 3-11-012349-5 NE: Dressler, Wolfgang U. [Hrsg.]; Trends in linguistics / Studies and monographs

© Copyright 1990 by Walter de Gruyter & Co., D-1000 Berlin 30. All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Typesetting: Arthur Collignon GmbH, Berlin. — Printing: Gerike GmbH, Berlin. — Binding: Lüderitz & Bauer, Berlin. — Printed in Germany.

Preface

This volume presents selected papers from the Third International Morphology Meeting held, under the auspices of the International Association of Morphology, in Krems (Austria) from July 4 to July 7, 1988. In contrast to the relatively small previous meetings in Veszprem (Hungary), the Krems conference was an open congress which attracted many participants from the immediately preceding Sixth International Phonology Meeting. 1 In effect, the last day of the Phonology conference overlapped with the first day of the Morphology meeting. Accordingly our volume opens with the topic "Interface" (mainly between morphology and phonology) where Spencer's morpholexical approach to morphophonemics contrasts with Shapiro's semiotic approach to isomorphism of rule types. Carstairs' paper on suppletion focuses on phonological triggers, Bayer and Lahiri's on morphosyntactic constraints on Bengali clitics. Within the second topic "Word formation" a wide range of subjects is covered: Corbin presents her own rule-based model of complex words, whereas Mötsch discusses conflicting proposals for word-structure theory (e.g., analogy vs. constraints vs. rules). Principles of headedness in compounds are in Di Sciullo's contribution on argument inheritance and in Vogel's account of shortening of English loans in Italian. Beard argues against morphemes as lexical items and for a strict separation of meaning and form in derivational morphology, Dressier and Kiefer deal with the morphopragmatics of Austrian and Hungarian diminutives, Warren with types of phonologically modified English compounds. Zwanenburg discusses the order of compounding and inflection in French, Scalise argues why Italian adverbs are derivational, not inflectional. Although belonging to the third topic "Inflectional morphology and clitics", Zwicky's presentation of his modular approach to both subtopics includes their relations to other modules as well; Wurzel's approach to inflection arrives at comparable conclusions although couched in a model of markedness. Morin adduces arguments for lexicon-internal 1. The selected papers of that conference will be published under the title Phonologica 1988 by the Cambridge University Press.

VI

Preface

inflection from the history of French, P. K. Andersen against a universal morphological category "passive". Joseph applies Zwicky's classificatory criteria on Modern Greek clitics reinterpreted as affixes, whereas Manoliu-Manea analyses pronominal and pragmatic functions of Rumanian clitics. From the conference topic "Computer morphology" only the paper by Meijs (on his ASCOT model) has been included in this volume; from the topic "The psycholinguistic study of morphology" there are Derwing's experimental evidence for a full-listing hypothesis about the representation of morphology in the mental lexicon and van Marie's reappraisal of analogy with evidence from Dutch derivational morphology. The last topic "Typology and non-Indo-European morphology" is represented by Hagege's substitution of classical morphological types with intralingual polytypical complexity due to phonetic evolution, and S. Anderson's comparison of Sapir's morphological typology with recent theoretical conceptions. In this way our volume offers a representative cross-section of contemporary developments in theoretical morphology. Papers given at workshops are published separately: on Aphasia by J.-L. Nespoulous and P. Villiard,2 on Natural Morphology by J. Mendez Dosuna and C. Pensado Ruiz.3 The discussion papers have already been published in two separate volumes.4 Wolfgang U. Dressier Hans C. Luschützky Oskar Ε. Pfeiffer John R. Rennison

2. Morphology, Phonology and Aphasia (New York: Springer). 3. Naturalists at Krems (Universidad de Salamanca). 4. Wiener linguistische Gazette, supplement 7 Interface, supplement 8 Morphology.

Contents

Topic 1: Interface

1

Josef Bayer and Aditi Lahiri Bengali emphatic clitics in the lexicon-syntax interface

3

Andrew Carstairs Phonologically conditioned suppletion

17

Michael Shapiro On a universal criterion of rule coherence

25

Andrew Spencer The advantages of morpholexical phonology

35

Topic 2: Word formation

41

Danielle Corbin Associativite et stratification dans la representation des mots construits

43

Anna-Maria Di Sciullo Formal relations and argument structure

61

Wolfgang U. Dressier and Ferenc Kiefer Austro-Hungarian morphopragmatics

69

Wolfgang Mötsch Problems of word structure theories

79

Sergio Scalise Constraints on the Italian suffix -mente

87

Irene Vogel English compounds in Italian: the question of the head

99

Beatrice Warren The importance of combining forms

Ill

Wiecher Zwanenburg Compounding and inflection

133

VIII

Contents

Topic 3: Inflectional morphology and clitics

139

Paul Kent Andersen Arguments against the passive as a universal morphological category

141

Robert Beard The empty morpheme entailment

159

Brian D. Joseph The benefits of morphological classification: on some apparently problematic clitics in Modern Greek 171 Maria Manoliu-Manea Case markers and pragmatic strategies: Romanian clitics

183

Yves-Charles Morin Parasitic formation in inflectional morphology

197

Wolfgang Ullrich Wurzel The mechanism of infleciton: lexicon representations, rules, and irregularities 203 Arnold M. Zwicky Inflectional morphology as a (sub)component of grammar

....

217

Topic 4: Computer morphology

237

Willem Meijs Morphology in LDOCE and in the ASCOT system

239

Topic 5: The psycholinguistic study of morphology

247

Bruce L. Derwing Morphology and the mental lexicon: psycholinguistic evidence . . 249 Jaap van Marie Rule-creating creativity: analogy as a synchronic morphological process 267 Topic 6: Typology and non-Indo-European morphologies

275

Stephen R. Anderson Sapir's approach to typology and current issues in morphology . . 277

Contents

IX

Claude Hagege Do the classical morphological types have clear-cut limits? . . . .

297

Index of languages

309

Subject index

311

List of contributors

315

Topic 1: Interface

Bengali emphatic clitics in the lexicon-syntax interface* Josef Bayer and Aditi Lahiri

1. Introduction In this paper, we advocate the view that to provide an adequate account of certain clitic constructions, one must refer to more than one component of the grammatical system. We will argue that the emphatic clitics in Bengali must be licensed by both the morphophonemics of the lexicon and the syntax of logical form, where the latter largely obeys the constraints on overt syntactic movements. After presenting some of the core facts, we will first discuss the lexical account and then turn to the complementary logical-form account. In the last section, we will present our solution to a paradox that appears to arise when both the morphophonemic and semantic aspects of these clitics are considered together.

2. The core facts joj and j\j are the so-called emphatic clitics in Bengali which mean something like 'too' and [ + emphatic] respectively. Since only joj introduces a new lexical meaning, we will mostly use joj for the examples, but /i/ behaves alike in all important respects. At a first glance, /o/ and /i/ seem to adjoin as enclitics to an element of type X° which is then the focus of the clitic, e.g.,

* We wish to thank Probal Dasgupta for his advice and Jogamaya Bayer for discussing the Bengali data presented here. We are also grateful to Wim van der Wurff and an anonymous reviewer for their suggestions. Realizing that proposals independently made in Dasgupta (1984, 1987, in press) turned out to be similar in spirit to ours gave us encouragement to pursue this work.

4

Josef Bayer and Aditi Lahiri

(la)

b)

babul- ο kha- be Babul- too eat- [fut, 3rd pers] 'Also BABUL will eat' babul kha- beο Babul eat- [fut, 3rd pers]- too 'Babul will also EAT'

In (la) we see that /o/ can adjoin to a syntactic word and (lb) shows that it can attach to an inflected verb. The clitic can also attach to an overtly inflected N. (2)

babul chele- keο mereche Babul boy- [obj]- too beaten-has 'Babul has beaten also his SON'

The facts concerning the attachment of the clitic to the verb are more complicated. As we have seen in (1), jo/ can come after the inflected verb, and in most instances, it cannot be added between the stem and the ending. Compare the following pairs of sentences where the second member is ruled out. 1 (3a)

b) (4a)

b)

mar- iο beat- [lpers]- too '(I/we) also BEAT' *mar-o-i mar- chiο beat- [prog]- [lpers]- too '(I/we) am/are also BEATING' *mar-o-ch-i

Now consider instances where the clitic can be added before the inflectional ending is attached. Examples are given in (5b) and (5d).2 (5a)

b) c)

d)

babul cheleke mereche-o Babul boy[obj] beaten- has-too 'Babul has also BEATEN the boy' babul cheleke mere-o-che mer- echiIamο beat- [prt]- [prog]- [link]- [past]- [lpers]- too '(I) have also BEATEN' mer-e-o-ch-i-l-am

Bengali emphatic clitics in the lexicon-syntax

interface

5

Although the clitic can be attached between the stem and the inflection, it cannot be inserted between affixes. The choice is binary — either the clitic comes right after the stem, or it must come after all the affixes are added. Thus the forms in (6) are ungrammatical in Bengali. (6a) b)

*mer-e-ch-i-o-l-am *mer-e-ch-i-l-o-am

It must be noted that there are no instances of categories other than verbs in which the emphatic clitic could be "infixed". For instance, (7) is ungrammatical. joj appears here between a noun stem and the casemarker -ke, which is arguably an inflectional ending. (7)

*babul chele-o-ke mereche 'Babul has beaten also the BOY'

This restriction also holds for compound-like word formations. Bengali has a verbal noun, which is derived by attaching the suffix -a to a Vstem, e.g., por 'read' + a —• pora 'reading'. Similar to German infinitives such as rad+fahrert 'bike riding', the verbal noun can incorporate an N°object into the verb stem involved, e.g., golpo por-a 'story reading'. As (8a) below shows, joj can adjoin to the verbal noun and select its focus inside, but as shown in (8b), it cannot adjoin to the focused Ν incorporated. For these examples, imagine a preceding discourse in which someone states that (s)he liked somebody's reading of stories very much. (8a)

b)

tader [kobita por- a]- ο bhalo laglo their poetry read-ing- too pleased-has '(I) was pleased by also their reading of POETRY' * tader [kobita-o por-a] bhalo laglo

This also holds true for the so-called dvandva constructions such as in (9) below. The contrasting sentences with different clitic placements are given in (10). (9a)

b)

(10a) b) c) d)

bap ma father mother 'parents' uttor dokkhin north south [bap ma]-o *[bap-o ma] [uttor dokkhin]-o *[uttor-o dokkhin]

6

Josef Bayer and Aditi Lahiri

The interpretation of dvandvas such as (10a) suggests that /o/ attaches to the whole dvandva as indicated by the bracketing, not to its second constituent. Thus (10a) can only mean 'the parents too', but not 'father and also mother'. Given these observations there seem to be two obvious questions that come to mind. First, what is the domain to which the clitics can attach? Second, if this domain is not identical to the stem to which all affixes can attach, how do the clitics differ from regular affixes?

3. Emphatic clitics in the lexicon The fact that (3b) and (4b) are bad but not (5b) can be accounted for by examining the minimal phonological unit that /o/ can attach to. Observe that joj can attach to mere but not to mar. We will argue that the host of the clitics must minimally be a phonological word. Under this view, mere constitutes a phonological word while mar is merely a stem. In the following discussion, we will focus on the facts which constitute evidence for differentiating phonological words from stems, indicating also how the clitics themselves are different from regular affixes. 3.1. Bengali has a rule which degeminates syllable-initial geminates. The underlying form of the progressive affix is /cch/, a geminate affricate. After a vowel-final stem the geminate is retained (since it can close the preceding syllable), but is degeminated when preceded by a consonantfinal stem. (11a)

b)

kha- cchi —• eat- [prog, lpers] '(I) am eating' mar- cchi —*•

khac.chi

(closure of preceding syllable)

mar.chi

(C deleted)

The same holds true after the causative affix /a/ is added to the stem; the geminate is retained as after a stem final vowel. (12)

mar- acchi —• ma.rac.chi beat- [caus]- [prog, lpers] '(I) am having (him) beaten'

Bengali emphatic clitics in the lexicon-syntax

interface

7

Resyllabification is, however, blocked outside the domain of a phonological word; after mere, the geminate remains syllable initial and is degeminated. 3 (13)

mar- ecchi —• me.re.chi (not: *me.rec.chi) beat- [past prt]- [prog, lpers]

Implicit in this view is the claim that affixes can be added to stems as well as to larger units like words. The compound-like constructions (cf. 9 — 10) also have affixes added at the end ([bap ma]r 'of parents', [*baper ma]). Moreover, the prosodic unit after the addition of a clitic to a phonological word, is still a word to which an affix can be added. 3.2. Evidence that mere is indeed a phonological word and that there is less cohesion between word + affix than stem + affix comes from reduplicating echo words. In Bengali, an echo word can be formed by reduplicating the entire word except for the initial consonant which is usually replaced by a coronal. The echo word could be interpreted as 'X and so forth' with perhaps a slight pejorative tinge. There is a constraint, however, in what can be reduplicated. All stem + affix constructions can be reduplicated, but no stems alone. (14a)

b)

c)

Nominal forms pa — ta bari — tari chele — tele kobita pora — lobita pora

'leg' 'house' 'boy' 'poetry reading'

Verb forms: mare — tare khae — tae mere — tere

'[3rd pers] beat' '[3rd pers] eat' '[past part] beat'

Inadmissible verb stems: *kha — ta *mar — tar

'eat' 'beat'

The affixation pattern also shows the distinction between the stem and the word. The past-participial form mere can have the affix -che added to it and then become reduplicated; but the -che can also be added after reduplication has taken place. Compare the forms under (15).

8

Josef Bayer and A did

(15a) b) c) d)

Lahiri

mereche tereche '[past part + 3rd pers] beat' [mere tere]che marche tarche '[3rd pers pres] beat' *[mar tar]che

This gives more evidence that mere is a unit which can stand on its own. 3.3. Evidence that the clitics differ from superficially similar affixes can be obtained from phonological rules of deletion and shortening. Derivative vocalic suffixes can trigger vowel shortening in stem vowels and deletion of vowels; clitics, however, do not trigger such processes. The examples in (16) are taken from Dasgupta (1984). (16a) b) c) d) e) 0 g) h) j)

na:k naki ra:g ragi na:k-o ra:g-o no:t noti no:t-i

k) pagol 1) pagli m) pogol-i

'nose' 'nasal' 'anger' 'angry' 'the nose too' 'anger too'

(not: *nak-o) (not: *rag-o)

'(male) dancer' '(female) dancer' '(male) dancer/[+emphatic]' 'idiot' 'mad woman' 'idiot/[ + emphatic]'

To summarize, we have shown that the clitic jo/ and jij are different from derivational suffixes and have as their minimal host category the phonological word. The latter fact, however, still does not account for the fact that sentences such as (7a) are ungrammatical. It seems unreasonable to suggest that chele does not constitute a minimal phonological word. 4 In the next section, we will therefore explore a completely independent line of reasoning.

4. Emphatic clitics in syntax and logical form Let us make the assumption that the clitics under consideration impose quantificational properties on their morphological/syntactic domain, similar to only and even in English. As Rooth (1985) and others before him

Bengali emphatic

clitics in the lexicon-syntax

interface

9

have argued, a phrase narrowly focused by only, even, etc., must be interpreted with respect to a quantificational domain. Such a domain is naturally provided by the verb, although not necessarily by the verb. Let us assume that (17a) below is an S-structure, (17b) is the logical form derived from it, and (17c) is a rough semantic representation which transduces the logical form into a proposition with a universal quantifier having scope over it. (17a) b) c)

we saw only John [only John], [we saw Xj] For all χ [we saw χ —> λ: = John]

Only John like John-o is a quantifier which must be assigned scope over the clause (proposition) at the level of logical form. Scope assignment, however, is constrained in language-specific ways. For instance, even in English cannot appear in an unconstrained fashion, although there is no prima-facie semantic reason which could prevent this. (18a) b)

They have killed [NP even [NP my dog]] *They have killed [NPmy [N' even [N- dog]*\\

As Bayer (1988, 1990) has shown, only, even, etc., and their respective correspondents exhibit different island effects in Dutch, English, and German. (See also Longobardi (in press) for Italian.) Once they are adjoined to an XP which does not correspond to the predicate of a root sentence, this XP must be canonically governed by a verb, or it must connect to a "dynasty" of uniformly oriented governors in the sense of Köster (1986). According to Köster, with the exception of VP, all maximal projections XP of lexical categories are virtual bounding nodes. A bounding domain can, however, be extended when XP is governed by an element which conforms to the basic orientation of government in the language. The direction that counts as basic in a VO-language such as English is — w h i l e it is Β / X as a set of allomorphy statements plus selectional statements. Essentially, we provide for the existence of the A and Β allomorphs (either by direct listing or by generating one allomorph from a basic form by means of a directionally triggered, but otherwise context-free lexical redundancy rule) and specify that the Β allomorph is selected in the X context. Concrete proposals for achieving this without losing phonological generalizations are presented in Spencer (1988a, b) with additional details in Spencer (ms). In a sense the approach can be seen as a radical extension of Kiparsky's (1988) proposals on suppletive allomorphy. The result of this

36

Andrew

Spencer

notational recoding is that phonologically sensitive morphological processes do not have to precede the phonology proper, for in a sense, all cyclic phonology now precedes all morphology. However, there are a number of other advantages to the morpholexical framework, some descriptive, some conceptual. The descriptive gains center on the possibility of stating morphological generalizations about "derived" allomorphs in the lexicon (say, of the form "the present participle stem is always identical to the third-plural stem"). One obvious spin-off is that certain forms of analogical leveling, which are notoriously difficult to handle satisfactorily in a derivational theory, can be seen as straightforward morpholexical rule simplification. Thus, the leveling of the paradigm of Early Latin honos, honorem, ... to Classical honor, honorem, ... is no more than generalization of the honorstem allomorph for the relevant (sub)paradigm, leaving the adjectival honestus 'honest' as an irregular remnant. (Cf. also Spencer 1988 a on Czech conjugation.) The morphophonemics of palatalizations in Slavic illustrate a related advantage. The Slavic languages share a core of distinct palatalizations. The surface inventory of consonant and, crucially, vowel phonemes shows important differences from language to language, even in the case of closely related ones (e.g., Czech and Slovak). Particularly interesting are the interactions between palatalizations and vowel-zero alternations. Some alternating vowels trigger palatalization, while others do not. In some languages the two vowels surface as different, in others only one vowel surfaces. In some cases, this is a front vowel (e.g., Polish, Czech / e/), in others a back vowel (e.g., Serbo-Croatian /a/). Despite the differences, the morphologically motivated palatalizations and vowel-zero alternations are virtually identical across all the languages, applying in roughly the same morphological contexts (modulo leveling) and to the same lexical items (modulo morphological class shifts). The problem for a derivational (SPE-type) account is to provide a set of grammars for Slavic in which the palatalizations result from essentially the same set of rules for all the languages. For if the rules generating the palatalizations are substantially different, we have a curious translingual conspiracy on our hands. Why should Slavic language learners so construct their grammars as to preserve a roughly constant morphophonemic surface pattern? If we assume that the palatalizations and a large part of the vowelzero alternations are morpholexical alternations, then there is no conspiracy. For the acquisition of morphophonemics is the acquisition of allomorphic variants and the extraction of redundancy rules relating

The advantages of morpholexical phonology

37

them. It is thus lexical learning rather than (phonological) rule learning. Variation amongst related languages can therefore be seen as the result of morphological and lexical drift plus the effects of categorial phonological changes, such as restructuring of the phoneme inventory, or underlying distinctive feature set. On this perspective we would expect a basic core to be shared, namely common vocabulary elements and common morphosyntactic and morpholexical classes. Since there are no derivational rules, there is no need to manipulate them to ensure a particular type of output. The conceptual arguments are all ultimately connected with learnability. The first two concern Kiparsky's (1985) strict cyclicity condition and structure preservation. The strict cyclicity condition prevents a cyclic rule, e.g., English velar softening, from applying to a single, underived, morpheme such as king to produce sing. If all cyclic rules are actually morpholexical relations then this is an automatic consequence of universal grammar. For a morpholexical relation is a type of allomorphy statement, not a rewrite rule. Hence, by definition it only applies to items which alternate. Since king fails to alternate, the language learner will not have reason to put king into the velar-softening class. This means that a morpheme such as -cept (as in concept, accept) is also non-alternating (underlyingly it is /sept/). The prefix allomorphy found in accept (vs. assess) and voicing differences between re[z]ignjcon[s]ign and re[s]eive/ con[s\eive must be regarded as facts about allomorph selection, with redundancy rules stating that the morphemes which select the /ak/ prefix allomorph are also the ones which do not have a voiced initial alternant. The trade-off here is between arbitrary, largely non-phonetic, redundancy statements of this sort for small portions of the vocabulary, as against extrinsic rule ordering and opacity.2 In a structure-preserving phonology, lexical contrasts which are not present underlyingly are not created in a derivation. To some extent the explanatory force of this notion hinges on the constraints put on underlying phoneme inventories and absolute neutralization. In any case, it follows automatically from the present proposals since the allomorphic variation is stated at a level which precedes phonology proper. Thus, the only segment types over which it can be defined are phonemes and archiphonemes (cf. Sproat 1985). Much effort has been expended on constraining the degree of "abstractness" a phonology can exhibit. The rationale is that abstract and opaque grammars are harder to learn than concrete and transparent ones. It is worth contrasting the learning of morphophonemics with the learning of syntax here. Current syntactic theory (e.g., Chomsky 1986) includes a

38

Andrew

Spencer

good many abstract elements (e.g., empty categories, restructurings, and so on) but has reduced opacity by eliminating the rule component of grammar in favor of the representational component. The abstract elements are dictated by the form of universal principles and subcomponents of grammar. Their ultimate conceptual justification lies in the "poverty of the stimulus" problem: learners seem to hit on certain classes of grammars, excluding other, apparently equally accessible, classes, even though the type of data that would normally motivate such a decision is either very rarely found or is in principle unavailable (for example, due to the unbounded nature of syntax). Morphophonemics is exactly the opposite of syntax in crucial respects. First, the domain is essentially finite. Therefore in principle learners could simply memorize all alternations. Thus, there can be no genuine "poverty of the stimulus" problem. This deprives morphophonemics of the need for a principles-and-parameters approach. However, an SPE-based approach to morphophonemics introduces opacity (through extrinsic rule ordering), and abstract elements (through absolute neutralizations). This poses a learnability problem which has never been seriously tackled. Consider the concrete case of Slavic. To figure out the palatalizations, the learner has to have worked out the abstract underlying vowel system and the complex pattern of vowel-zero alternations. In some cases, this means correctly analyzing portions of the vocabulary which is of very low type or token frequency. But the vowel-zero alternations themselves cannot be correctly deduced without at least partially figuring out the palatalizations. This situation is completely typical of rich morphophonemic systems. The kind of bootstrapping deduction it requires is typical of the hypothesis-testing mode of scientific inquiry, of the kind which Fodor (1983) associates with central cognitive systems. It is not the kind of operation we associate with an input module. In other words, SPEbased morphophonemics takes a finite domain which should thus pose a trivial learnability problem, and introduces its own, completely artificial "poverty of the stimulus" problems, without providing satisfactory universal principles for their resolution. In the morpholexical approach, these problems do not arise. Most learning is lexical learning, supplemented with very simple universal principles of selection. Phonologically describable relationships are a consequence of the representational aspects of phonological theory and not the result of derivational devices. This means that the only artificial "poverty of the stimulus" problems which can arise concern abstract underlying segments. Since the domain is finite, and since lexical listing

The advantages of morpholexical phonology

39

is therefore always open as a possibility, the extent to which learners project such abstractions is an empirical question, possibly more the concern of the psycholinguist than the phonologist. The final aspect of the learnability problem concerns the nature of the data to which the child is exposed. Morphophonemic systems are typically riddled with exceptionality and complex patterns of subregularity. A learning theory for a derivational morphophonemics would have to show how it is possible for a child to handle such patterns of exceptionality by means of ordered rules. The main question concerns repair of errors: how does the child restructure his grammar when he finds he has made a mistake in the rule formulation, the underlying format, or the ordering of rules, given the immense choice of hypotheses? Lexical learning in morpholexical phonology, as in syntax, encounters no such problem since lexical learning is essentially a question of establishing a taxonomy. If an incorrect taxonomy is projected (or if the child is confused by errors in the input itself), then this can be corrected by simply changing morpholexical class membership (and perhaps redefining the criteria for membership of a class). Thus, the adult grammar can be approached (more or less) asymptotically. It is hard to see how a derivational morphophonemics could be so learned. Certainly, the onus of proof must lie on advocates of derivational approaches.

Notes 1. I am inclined to say that the postcyclic and postlexical components take the form of output constraints governed by universal and language-particular principles such as syllable structure, metrical structure, autosegmental structure, and so on. This would make phonology look rather like syntax on the "principles-and-parameters" approach. 2. The strict-cyclicity condition also prevents a rule A which counterfeeds a rule Β from being fed by Β on a subsequent cycle. In a morpholexical framework, this effect can only be reconstructed using complex diacritics. It seems to me undesirable that a principle of universal grammar should have the effect of maintaining rules in a marked order, so I am happy with this situation.

References Booij, Geert —Rubach, Jerzy 1987 "Postcyclic versus postlexical rules in lexical phonology", Linguistic Inquiry 18: 1 - 4 4 . Chomsky, Noam 1986 Barriers. (Cambridge, MA: MIT Press).

40

Andrew Spencer

Fodor, Jerrold 1983 The modularity of mind. (Cambridge, MA: MIT Press). Kiparsky, Paul 1985 "Some consequences of lexical phonology", Phonology Yearbook 2: 83 — 136. 1988 "Two approaches to suppletive allomorphy" [Paper presented at the 3rd International Morphology Meeting, Krems, Austria]. Lieber, Rochelle 1982 "Allomorphy", Linguistic Analysis 10: 2 7 - 5 2 . Spencer, Andrew J. 1988a "Arguments for morpholexical rules", Journal of Linguistics 24: 1 —30. 1988b "Morpholexical rules and lexical representations", Linguistics 26: 619 — 640. ms "Morpholexical phonology". Sproat, Richard 1985 On deriving the lexicon [Dissertation, MIT, Cambridge, MA].

Topic 2: Word formation

Associativite et stratification dans la representation des mots construits Danielle Corbin

Au risque de paraitre aller a contre-courant de bon nombre de recherches actuelles en grammaire generative, qui dissocient et traitent dans des modules differents la representation de la «syntaxe interne» des mots construits et celle de leur interpretation semantique 1 , je defendrai ici le bien-fonde d'un composant derivationnel associatif ä l'interieur d'un composant lexical autonome et stratifie. « Associatif» signifie que le role assigne au composant derivationnel est d'engendrer conjointement et de fagon associee la structure morphologique et Interpretation semantique des mots construits. «Stratifie» signifie que l'organisation interne du composant lexical et la hierarchie des operations touchant les mots construits qu'il comporte refletent la complexite des associations forme/ sens qui caracterise les mots construits, et qui ne saurait se reduire ä la dichotomie regulier/irregulier. Pour des raisons de place, ce texte ne fait qu'esquisser le cadre general de cette theorie associative, en se concentrant sur la fagon dont celle-ci peut traiter les diverses distorsions forme/sens; il ne presente pas de fagon detaillee les arguments justifiant ses analyses2. Tous les exemples traites sont empruntes au frangais. Apres une mise au point sur la notion meme d'interpretation semantique des mots construits (§ 1), je passerai en revue les reanalyses que le modele presente permet d'effectuer face aux divers types de distorsions apparentes entre la structure morphologique et l'interpretation semantique des mots construits (§ 2). Dans chaque cas, je montrerai que la distorsion n'est qu'apparente, c'est-a-dire qu'il est possible d'apporter des reponses modeliques ä ces distorsions sans remettre en cause le principe de l'associativite, comme cela a souvent ete fait dans la litterature 3 . J'expliquerai pour terminer la stratification correspondante du modele de composant lexical que je propose (§ 3).

44

Danielle

Corbin

1. Mise au point sur la notion ^interpretation semantique (Tun mot construit Je propose d'introduire deux distinctions fondamentales portant sur I n terpretation semantique des mots construits. D'une part entre la notion de «sens atteste» (SA) et celle de «sens predictible» (SP): j'appellerai sens atteste le sens d'un mot construit tel qu'il figure dans un ou plusieurs dictionnaire(s) de langue contemporain(s)4. Au sens atteste directement observable, j'opposerai le sens predictible, c'est-a-dire le sens d'un mot construit tel qu'il est formulable ä partir de l'operation semantique propre ä la regle qui a construit le mot. Comme cette definition le fait apparaitre, le sens predictible est le resultat d'une construction hypothetique du linguiste, en fonction de l'idee qu'il se fait de l'economie generale de la grammaire derivationnelle de la langue, en fonction de sa theorie du lexique. Dans ma theorie, le sens predictible a deux proprietes fondamentales: sa compositionnalite obligatoire par rapport a la structure morphologique du mot construit, et son independance relative par rapport au sens atteste. D'autre part, ä l'interieur du sens predictible, entre le sens predictible construit par la regle (SPcr) et le sens predictible herite de la base (SPhb). Cette distinction vise ä clarifier les divers composants entrant dans l'interpretation semantique predictible d'un mot construit. Le sens predictible construit par la regie est le sens commun a tous les produits d'une meme regle de construction de mots, independamment des types de bases et des procedes morphologiques utilises. Le sens predictible herite de la base est le sens commun ä tous les produits d'une regle de construction de mots construits sur des bases ayant le meme type semantique pertinent; il represente une specification du sens predictible construit par la regie particuliere ä un type semantique de base donne 5 . J'illustrerai cette stratification du sens predictible par un exemple simple, dont je ne developperai pas les justifications 6 : la regle de construction de mots ä laquelle est associe (entre autres procedes) le suffixe -eux en frangais attribue ä ses produits un sens predictible construit par la regle de type relationnel, que l'on peut representer par la paraphrase « en relation avec [nom de base]» (simple transposition en termes semantiques du changement categoriel d'un nom en adjectif)· Quand la regie s'applique ä des bases designant des maladies, le sens predictible herite de la base des adjectifs construits peut prendre la forme « atteint de [nom de base]». L'exemple de l'adjectif coquelucheux illustre ci-dessous ces diverses strates:

Associativite

(1)

SPcr ex. SA SPhb ex. SA

et stratification

45

«en relation avec la coqueluche» une toux coquelucheuse = une toux de coqueluche le bacille coquelucheux — le bacille de la coqueluche « de la coqueluche » « atteint de coqueluche » un enfant coquelucheux — un enfant atteint de coqueluche « atteint de coqueluche »

Cet exemple est simple, car il ne fait guere apparaitre de distorsion entre les sens predictibles et les sens attestes. Comme on va le voir ci-dessous, ce n'est pas toujours le cas.

2. Reponses aux distortions apparentes entre l'interpretation semantique et la structure formelle (Tun mot construit On partira de l'idee, generalement admise, que la situation la plus «naturelle» est celle ού la structure morphologique d'un mot construit et son interpretation semantique se correspondent biunivoquement 7 . Mais il n'en est pas toujours ainsi, apparemment du moins. J'envisagerai successivement quatre types de distorsions possibles entre l'interpretation semantique et la structure ou la forme du mot construit qui ont pu ou pourraient servir d'arguments empiriques contre l'associativite, et je tenterai de leur proposer des reponses. Deux principes methodologiques guideront mon expose: (i) un meme phenomene apparent peut recevoir plusieurs interpretations differentes; (ii) il est necessaire, pour atteindre quelque generalite, de depasser le niveau de l'evidence observationnelle du lexique atteste. Je m'en tiendrai ici aux arguments empiriques, ayant developpe ailleurs des arguments theoriques contre les modeles dissociatifs 8 . 2.1. Premier type de distorsion: non-conformite apparente de l'interpretation semantique ä la structure morphologique Le type le plus evident de distorsion apparente est celui ou l'interpretation semantique parait ne pas etre conforme ä la structure morphologique du mot construit. Je prendrai deux exemples, qui illustrent deux fapons de resoudre le probleme.

46

Danielle

Corbin

2.1.1 Four che tie a pour structure morphologique (2a), done pour sens predictible (2b)9, mais pour sens atteste (2c): (2a) b) c)

[[fourche]N(ette)af]N SP « petite fourche » (cf. flechette, hachette, serpette, etc.) SA «1° ustensile de table (d'abord ä deux, puis a trois, quatre dents), dont on se sert pour piquer les aliments »

Le fait que le sens predictible ait ete applique a un domaine particulier de l'experience extra-linguistique ne peut etre ni predit ni explique linguistiquement, et Ton doit considerer ce mot comme superficiellement idiosyncratique, sans que cette idiosyncrasie, posterieure a la construction du mot, puisse constituer un argument contre l'associativite10. 2.1.2 Le sens atteste actuel (3d) de publiciste ne correspond pas ä sa structure morphologique apparente (3a). Le sens predictible construit par la regle correspondant ä cette structure est (3b), et le sens predictible specifie par le processus morphologique (SPsp) quelque chose comme (3c)11: (3a) b) c) d)

[[[[public]A]N(iste)af]A]N Spcr « N en relation avec le public» SPsp « Ν [ +hum] specialiste du public » SA «3° [...] agent de publicite»

Ici, contrairement au cas precedent, le sens atteste est irreductible au sens predictible, meme au prix d'un ajustement superficiel. Je poserai qu'il y a en fait en frangais deux noms (provenant par conversion de deux adjectifs) homonymes qui ont la forme publiciste: l'un a pour structure (3a) et pour sens predictible (3b) et (3c), mais n'est pas atteste, l'autre, qui correspond au sens atteste (3d), a pour structure (4a) et pour sens predictible (4b) et (4c): (4a)

b) c)

[[[[public] A (ite) a f < + T>]N(iste) a f < T + > ] A ]N

SPcr SPsp

« Ν en relation avec la publicite » « Ν [ + hum] specialiste de la publicite»

Dans la structure (3a), le nom °publiciste/2 est converti ä partir de l'adjectif °publiciste,, lui-meme construit sur le nom public, converti ä partir de l'adjectif correspondant. Dans la structure (4a), le nom publiciste2 est converti ä partir de l'adjectif °publiciste2, lui-meme construit

Associativite

et stratification

47

sur le nom publicite, dont le suffixe -ite a subi une troncation devant -isten. Dans cet exemple, c'est l'homonymisation de la forme publiciste qui permet de conserver Γ associativite. 2.2. Deuxieme type de distorsion: ä une seule structure morphologique paraissent correspondre plusieurs interpretations semantiques Le deuxieme type de distorsion apparente est celui ou ä une seule structure morphologique paraissent correspondre plusieurs interpretations semantiques. Si tel est le cas, une theorie associative repond, soit que l'interpretation semantique est unique, malgre les apparences, soit, comme on l'a vu egalement a propos de publiciste, qu'a chaque interpretation semantique correspond une structure morphologique differente. 2.2.1. Du premier cas de figure (interpretation semantique unique sous plusieurs sens apparemment differents), je ne donnerai qu'un exemple. En frangais, les noms suffixes par -oir construits sur des bases verbales (structure (5a)) sont susceptibles de recevoir des sens attestes de type instrumental (5b) et de type locatif (5c). Un certain nombre d'arguments, que la place m'interdit de developper ici14, permettent de montrer que, dans cet exemple, la distinction instrumental/locatif n'est qu'apparente, et qu'en realite, la ou eile apparait irreductible dans le lexique atteste (mouchoir par exemple peut difficilement etre interprete comme un locatif dans l'univers de reference), eile provient de la projection sur le derive de proprietes semantiques et argumentales du verbe de base. Les sens instrumental et locatif peuvent done etre consideres comme les sens predictibles herites de la base des mots construits ä l'aide de ce suffixe (5d), et le sens predictible construit par la regle associe ä la regle de construction de mots en question etre formule par la paraphrase (5e): (5a)

[[X]v(oir)af]N

b)

SA

c)

SA

d)

SPhb

e)

SPcr

mouchoir: «1° petite piece de linge, generalement de forme carree, qui sert ä se moucher, a s'essuyer le visage » dortoir: «1° grande salle commune ού dorment les membres d'une communaute » « Ν qui sert ä [verbe de base]» «lieu ού Ton [verbe de base]» « Ν pour [verbe de base]»

48

Danielle

Corbin

2.2.2. Le deuxieme cas de figure (c'est-a-dire en fait l'homonymie structurelle) peut s'expliquer de diverses fapons: — Dans l'exemple dejä cite de publiciste, l'homonymie de surface etait due ä la troncation d'un affixe. — II peut s'agir aussi de l'application de deux affixes homonymes ä deux bases homonymes de categories differentes: par exemple, la forme plumage a deux sens attestes (6b) et (6d), irreductibles ä un meme sens predictible, ä chacun desquels correspond une structure morphologique differente (respectivement (6a) et (6c)): (6a) b)

[[plume]N(age,)af]N SA «1 0 l'ensemble des plumes recouvrant le corps d'un oiseau [...]»

c)

[[plume] v (age 2 )af]N

d)

SA

«2° [...] action de plumer»

Ou bien encore il peut s'agir de l'application des memes regies de construction de mots dans un ordre different. Par exemple la forme invalidable15 peut recevoir deux interpretations, (7b) et (7d), dont chacune est compositionnelle par rapport ä une structure morphologique differente: ä (7b) (a laquelle repond un sens atteste) correspond (7a); ä (7d) (ä laquelle ne repond pas de sens atteste) correspond (7c): (7a) b) c) d)

[[[(in)af{valide]A]A]v(able)af]A (valide —> invalide —> invalider —•» invalidable) SPcr « Qui peut etre invalide » ( = qu'on peut rendre invalide (l'etat initial du Ν predique est suppose valide)) [(in)af{[[valide]A]v(able)af]A]A (valide —• valider —> validable —> °invalidable) SPcr « N o n validable» ( = qu'on ne peut pas rendre valide (l'etat initial du Ν predique est suppose non valide))

Dans tous ces cas d'homonymie, on fait done bien correspondre un seul sens ä une seule structure morphologique. 2.3. Troisieme type de distorsion: a une seule interpretation semantique paraissent correspondre plusieurs structures morphologiques Lorsque, ä l'inverse de ce qui vient d'etre expose, ä une seule interpretation semantique paraissent correspondre plusieurs structures morphologiques, la reponse d'une theorie associative est, soit que, dans certaines conditions, des structures differentes formellement peuvent etre considerees

Associativiti

et

stratification

49

comme equivalentes semantiquement, soit que l'analyse fondee sur la perception apparente est fausse, c'est-a-dire que sous un seul sens atteste se cachent plusieurs sens predictibles differents. 2.3.1. On peut illustrer le premier cas par la construction des verbes de changement d'etat en frangais: sur des bases adjectivales de sens resultatif, on peut construire des verbes soit ä l'aide de prefixes (les principaux sont α-, έ-, en-), soit ä l'aide de suffixes (-ifi(er), -is(er)), soit par conversion (structures (8a)). Tous ces procedes donnent des verbes au meme sens predictible construit par la regle (8b), souvent attestes avec des definitions comparables (8c): (8a)

b) c)

[(Y)af{X]A]v (ex. appauwir, elargir, enrichir) [[X]A(Y)af]v (ex. humidifier, immobiliser) [[X]A]v (ex. blanchir) SPcr « rendre (plus) [adjectif de base]» SA appauvrir: «1° Rendre pauvre» elargir. «1 ° Rendre plus large » enrichir: « 1 0 Rendre riche » humidifier. « Rendre humide » immobiliser: «2° [...] rendre immobile [...]» blanchir. «1 ° rendre blanc »

Je propose dans ce cas de considerer que les trois structures (8a) sont equivalentes, c'est-a-dire que tous les procedes morphologiques auxquels elles sont associees font partie du meme «paradigme morphologique» associe ä la meme regle de construction de mots 16 , meme si certains different entre eux par des proprietes specifiques (disponibilite, contraintes particulieres sur la base, etc.)17. Les deux implications — importantes — de cette proposition sont d'une part que ce ne sont pas les affixes, dans ma theorie, qui sont fondamentalement responsables du sens des mots construits, mais les regies de construction de mots 18 , d'autre part qu'une telle regie est identifiee avant tout par l'association d'un rapport categoriel et d'une operation semantique, et non prioritairement par une forme affixale. 2.3.2 II y a des cas ού il ne faut pas se fier ä la similitude des sens compositionnels attestes de derives, meme s'ils paraissent correspondre a un meme sens predictible. L'exemple de chenaie et sapiniere illustrera l'apparence d'une meme interpretation semantique correspondant ä des procedes morphologiques differents (les suffixes -aie et -ierj-ieref9, sans que la situation soit celle, exposee au paragraphe precedent, de procedes appartenant au meme paradigme. Les sens attestes de ces mots sont en

50

Danielle

Corbin

effet comparables (9a), et on pourrait penser que les suffixes -aie et -ierj -iere appartiennent tous deux au paradigme morphologique associe ä la regle de construction de mots qui construit, sur des bases nominales designant des vegetaux, des noms dont le sens predictible peut se paraphraser par «lieu plante de [nom de base]». Mais, si ces proprietes paraissent bien etre celles de -aie, elles ne conviennent pas ä -ierj-iere, dont on peut dire tres brievement qu'il construit fondamentalement des adjectifs ä sens relationnel sur des bases nominales 20 . La structure morphologique (9b) et les sens predictibles de sapiniere (9c) ne sont done pas directement deductibles de l'apparence: (9a) b) c)

SA

chenaie « plantation de chenes » sapiniere « bois, foret, plantation de sapins » [[[sapin]N(ier/iere)af]A]N SPcr °sapinierA « en relation avec [nom de base]» sapiniereN « N en relation avec [nom de base]» SPhb « Ν contenant des, producteur de sapins »

Pour repondre a l'argument, souvent avance contre l'associativite, de la «classification croisee des affixes et des sens» 21 , j'ai propose dans ce paragraphe que les procedes morphologiques et Interpretation semantique soient associes de fagon asymetrique: ä un procede morphologique donne ne peut correspondre qu'un sens predictible construit par la regie, mais ä un tel sens peuvent correspondre plusieurs procedes morphologiques appartenant au meme paradigme. 2.4. Quatrieme type de distorsion: les accidents formels Pour en terminer avec la fa^on dont une theorie associative peut repondre aux divers types de distorsions forme/sens, il faut parier rapidement des accidents formels qui, etablissant une distorsion entre la forme predictible du mot construit et sa forme attestee, provoquent indirectement des distorsions entre la forme du mot construit et son sens. Ces accidents formels, allomorphies et troncations, seront analyses comme des modifications superficielles apportees ä une structure morphologique reguliere. Je me borne ici ä donner le principe d'une theorie exposee dans D. Corbin (1987: 283 — 370). Je distinguerai deux cas, en citant a chaque fois un exemple d'allomorphie et un exemple de troncation. 2.4.1. Si la modification de la forme attestee par rapport ä la forme predictible du mot construit est recurrente dans le lexique, je suppose que

Associativite

et stratification

51

cette derniere, reguliere mais bloquee ä la sortie du composant derivationnel, est modifiee par une regie mineure, posterieure aux regies de construction de mots, declenchee par des traits diacritiques dont sont porteurs les constituants du mot construit. Par exemple, le suffixe -ite provoque toujours la posteriorisation vocalique de la syllabe finale -eux de sa base, que cette finale soit affixale (verbe —• verbeux —• verbosite) ou non (genereux —• generosite). Comme le montrent les schemas (10), les formes *verbeusite, *genereusite sont regulierement produites par les regies de construction de mots, bloquees parce que deux constituants successifs de leur structure interne sont marques de traits allomorphiques complementaires, et corrigees en verbosite, g0n0rostä par une regie mineure, posterieure aux regies de construction de mots: (10a) b)

[[[verbe]N(eux)af< + PV>]A(it£)af * verbeusiti; regie mineure —• verbosite) [[genereux]A< +Pv>(ite)af]N (RCM —•» *genereusiti; regie mineure —»· generosite)

Le schema (11) rappelle la troncation du suffixe -ite par le suffixe iste dans l'exemple deja rencontre de publiciste22: la regle de construction de mots engendre *publiciteiste, bloque parce que deux constituants successifs de sa structure interne sont marques de traits de troncation complementaires, et la regie mineure de troncation corrige cette forme en publiciste: (11)

[[[public] A (ite) af< + T>]N(iSte)af]a

(RCM —• *publiciteiste; regie mineure —* publiciste) 2.4.2. Si la modification n'est pas recurrente, le cas est different pour rallomorphie et la troncation. Si la modification est de type allomorphique, je suppose que le mot construit a pour base un item non atteste suppletif du mot atteste correspondant: par exemple, la variation consonantique qui affecte la paire nager/natation est unique, ä ma connaissance (comparer obliger —* obligation, propager —> propagation, etc.). Natation n'est done pas derive de nager, mais de °nater, synonyme de nager23. Si la modification est de type troncatoire, le comportement de l'affixe tronque est exceptionnel par rapport ä son comportement habituel. Je propose alors de marquer idiosyncratiquement < + T> le mot construit, regulier mais non bloque ä la sortie des regies de construction de mots. On peut ainsi comparer verbosite, dans lequel le suffixe -eux a ete

52

Danielle

Corbin

normalement allomorphise devant -ite (cf. §2.4.1), et αηχϊέίέ24, lequel le meme sufflxe a ete tronque dans le meme contexte: (12)

dans

[[[anxie]N(eux)af]A(ite)af]n (RCM —*• *anxieusite; regie mineure d'allomorphie: °anxiosite\ troncation idiosyncratique —• anxiete)

2.5. Conclusion Pour conclure cette partie, je rappellerai les principes essentiels qui y ont ete defendus: — A toute structure morphologique est associee une interpretation semantique compositionnelle et un seul sens predictible construit par la regie, qui ne correspond pas necessairement au sens atteste. — S'il apparait une distorsion entre la structure morphologique et l'interpretation semantique, ou bien la distorsion est reelle (exemples (8), (10) ä (12)), et des dispositifs modeliques (paradigme morphologique, regies mineures d'allomorphie et de troncation) permettent de ne pas remettre en cause l'associativite, ou bien la distorsion n'est qu'apparente (exemples (2) ä (7), (9)), et il faut deplacer l'analyse. — II n'y a pas de bijection entre l'ensemble des affixes et celui des regies de construction de mots. — Ce n'est done pas l'unite de forme affixale qui delimite une operation derivationnelle, mais l'association entre un rapport categoriel, une operation semantique et un Operateur morphologique (affixe, conversion) appartenant ä un paradigme dont le nombre d'unites est variable. — II en decoule que la morphologie associative que je preconise est fondamentalement homonymique, dans la mesure ou le critere d'identification d'un element (affixe ou mot construit) est l'association de proprietes formelles et de proprietes semantiques. Au contraire, les morphologies dissociatives sont polysemiques.

3. La stratification du composant lexical Je me bornerai ici ä citer l'ensemble des proprietes qui font du modele de composant lexical que je propose, et dont une representation figure ci-apres25, un modele stratifie conformement aux principes de l'associativite enumeres ci-dessus.

Associativite

et

stratification

REGLES D'INSERTION LEXICALE Figure 1. Organisation du composant lexical

53

54

Danielle

Corbin

L'hypothese fondamentale est que les operations s'appliquent de fagon hierarchisee en fonction de leur caractere plus ou moins predictible et plus ou moins regulier, les moins regulieres etant toujours subordonnees aux plus regulieres. Du point de vue derivationnel, la stratification du composant lexical s'illustre de trois fagons. 3.1. La stratification des composants La stratification la plus visible est le fait que le composant lexical est divise en trois sous-composants: — Le composant de base, dans les «entrees lexicales de base», liste les morphemes (entrees majeures non construites et affixes) et toutes leurs proprietes, ainsi que les regularites semblables ä Celles des mots construits qu'ils sont susceptibles de presenter («regies de base»). — Le composant derivationnel engendre les mots construits et toutes leurs proprietes regulieres. II definit ce qui est possible linguistiquement. — Le composant conventionnel rend compte des sous-regularites et irregularites touchant les mots construits. II definit, a partir du possible, ce qui est atteste. Ainsi, le caractere bien forme d'un mot n'est pas indexe sur l'attestation de celui-ci. 3.2. La stratification des operations Une deuxieme manifestation de la stratification du modele est la hierarchie des types d'operations qu'il comporte. — Les regies de base (niveau 2) sont congues comme des regies de redondance a valeur evaluative, ne s'appliquant qu'aux entrees lexicales de base. — Les regies de construction des mots (niveau 3), sont des regies generales s'appliquant, si les conditions de leur application sont reunies, et chaque fois qu'elles le sont, aux entrees lexicales de base et a leurs propres produits (recursivite). On peut les concevoir comme un ensemble d'operations modulaires associees, dont je ne developperai pas ici le detail26. Elles construisent tout ä la fois la structure morphologique des mots construits, leur interpretation semantique (sens predictible construit par la regle et herite de la base27), une partie de leurs proprietes syntaxiques (celles qui sont predictibles a partir de leur caractere construit), en respectant d'une part les contraintes qui leur sont imposees, d'autre part

Associativite

et stratification

55

Celles, plus speciflques, qui sont imposees a l'application de chacun des procedes appartenant ä leur paradigme morphologique. — Les regies mineures, representees aux niveaux 5, 7 et 8 du composant conventionnel, ont un domaine d'application defini ä l'avance, et rendent compte des sous-regularites que presentent les mots construits et dont ne sont pas responsables les regies de construction de mots. — Les operations de l'applicateur d'idiosyncrasies (niveau 6) sont des operations ad hoc touchant les mots construits. C'est ä ce niveau, par exemple, que se decide la troncation de °anxiosite en anxiete (cf. (12)). — Les operations du selectionneur (niveau 9), sont des operations pour la plupart ad hoc, touchant ä la fois les entrees lexicales de base et les mots construits. Elles ont pour täche de selectionner les formes et les proprietes attestees (lexique de fait) parmi les formes et les proprietes possibles (lexique de droit). Ainsi, seul nager, et non °nater, et, pour fourchette, seul le sens idiosyncratique « ustensile de table », et non le sens « petite fourche », seront selectionnes pour faire partie du lexique atteste. 3.3. La stratification des sorties Une troisieme illustration de la stratification du modele est la possibilite donnee aux regies d'insertion lexicale (ou ä leur equivalent) d'aller chercher des mots a trois niveaux de sortie differents: — Au niveau des entrees lexicales de base: ä part les affixes, que leur categorisation specifique empeche d'etre inseres sous des nceuds syntaxiques28, toutes les entrees, attestees ou non (nager et °nater), ont ainsi la possibilite linguistique d'etre inserees dans des phrases. — Au niveau des mots construits possibles, attestes ou non. Rappelons que l'une des restrictions qui jouent ä ce niveau est le blocage de mots comme *verbeusite (cf. (10)) ou *publicitiiste (cf. (11)), dont les constituants internes sont marques pour subir des regies mineures. — Au niveau du lexique conventionnel, pour ce qui concerne le lexique atteste. Ne sont done pas confondues dans ce modele une propriete de la langue, qui est la possibilite linguistique qu'ont les mots d'apparaitre dans des phrases, et une propriete des locuteurs, qui est celle d'exploiter, de fa^on variable selon les individus, les potentialites de la langue.

56

Danielle Corbin

4. Pour conclure Les principes gouvernant le composant lexical stratifie dont j'ai esquisse les grandes lignes sont ceux d'une morphologie derivationnelle associative. Le modele propose reste generatif, dans la mesure ού la description a un caractere explicite, et ou l'objectif qu'elle se fixe est de rendre compte de la competence lexicale des locuteurs. Quelles que puissent etre ses imperfections actuelles, son originalite consiste a deplacer les ordres de priorite habituels, en affmant la typologie des associations forme/sens qui caracterisent les mots construits. C'est pourquoi il ne pourra veritablement etre confronte aux propositions dissociatives faites dans un cadre generatif que pour autant que celles-ci accepteront de se preoccuper de la representation de Interpretation semantique des mots construits, ce qu'elles ne font pas encore, a ma connaissance 29 .

Notes 1. 2. 3. 4. 5.

6. 7. 8. 9.

10. 11.

Voir par exemple Selkirk (1982), Toman (1987). Pour un expose detaille, voir Corbin (1987). Voir par exemple Jackendoff (1975), Lieber (1980). Ici, le dictionnaire de reference sera le PR86, sauf indication contraire. Dans les cas de concurrence entre plusieurs procedes morphologiques associes ä la meme regle de construction de mots (voir infra §2.3.1.), il peut s'ajouter a ces deux strates fondamentales du sens predictible un troisieme niveau, celui du sens predictible specifie par le procede morphologique, qui represente le sens attribue par un procede morphologique (affixe, conversion, composition) ä tous les mots qu'il sert a construire, sous la forme d'une specification du sens predictible construit par la regie. Une illustration en sera donnee ci-dessous avec les exemples (3) et (4). Voir, sur ce point, Puchulu (1987, 1988) et Corbin —Corbin, ä paraitre. Cf. Dressier (1987) par exemple. Voir Corbin (ä paraitre). Si Ton admet que le sens predictible construit par cette regie est diminutif, le sens (2b) represente ce sens; si Ton pense que le sens est hyponymique («Nb d'une certaine sorte»), (2b) represente le sens predictible specifie par le processus morphologique. Je ne discuterai pas ici cette question. Je suis d'accord sur ce point avec Booij (1979). On admettra ici (cf. Corbin 1988) que le suffixe -iste sert a construire fondamentalement des adjectifs de relation sur des bases nominales, et que les noms en -iste sont des conversions des adjectifs correspondants. Contrairement ä d'autres procedes morphologiques associes ä la meme regie, le suffixe -iste, applique ä des bases designant des personnes ou des doctrines, a pour particularite de favoriser la specification du sens predictible construit par la regie sous les formes « specialiste de [nom de base]» (dans un contexte nominal [ + hum] et « partisan de [nom de base]» (dans un contexte nominal [±hum]). Comparer par exemple un parti anarchiste ( = «partisan de l'anarchisme (ou de l'anarchie)») et un parti anarchique ( = «qui manifeste de l'anarchie»).

Associativite et stratification

57

12. Le signe 0 precede des formes ou des sens grammaticaux quoique non attestes, par opposition au signe * qui precede des formes ou des sens agrammaticaux. 13. Cette troncation est representee dans le schema (4) par les traits < + T>, signifiant que -ite est predispose ä subir une troncation dans un contexte approprie, et , signifiant que -iste est predispose a provoquer la troncation d'un suffixe predispose a la subir. Voir sur ce point Corbin (1987: 341 —370). 14. Certains sont developpes dans Corbin (1987: 247 — 248). 15. Encore attestee dans le LXXe et son successeur le GLE avec le sens «Qui peut ou doit etre invalide», mais plus dans le GDEL, ni aucun dictionnaire frangais strictement contemporain. Validable est atteste dans le GRLF. 16. Ma proposition differe sur plusieurs points de ce que Zwanenburg (1984) appelle «types de derivation »: — Peuvent etre inclus dans le meme paradigme morphologique des prefixes, des suffixes, la conversion, alors que Zwanenburg n'inclut dans ses «types » que des suffixes. — Zwanenburg propose qu'ä l'interieur d'un type donne, les suffixes soient exclusifs Fun de l'autre, c'est-ä-dire que son dispositif lui sert ä expliquer certains blocages, alors que je ne pense pas que ce principe de blocage soit adequat: si les procedes appartenant ä un meme paradigme obeissent aux memes contraintes, ils peuvent s'appliquer ä une meme base remplissant ces contraintes (en temoignent par exemple, dans le lexique atteste: abonnir / bonifier; clarifier / eclaircir, etc., et dans le lexique possible °laidifier face ä enlaidir, °sauvagiser face ä ensauvager, etc.). — Zwanenburg raisonne dans un cadre dissociatif, moi dans un cadre associatif. 17. Certaines de ces contraintes sont explicitees dans Corbin (1987: 450). 18. Ce qui s'oppose ä l'hypothese «One affix, one rule» formulee notamment par Aronoff (1976). 19. Le genre feminin de sapiniere n'est pas plus predictible que celui de saliere compare a poivrier. 20. Par exemple, un argument contre l'assimilation du suffixe -aie et du suffixe -ier/iere est que, contrairement ä ce qui se passe pour -aie, on trouve des noms de lieu en -ierj-iere sur des noms de base ne designant pas des vegetaux {sablonniere («lieu d'ou Ton extrait le sable»), heronniere («endroit amenage pour l'elevage des herons»), etc.). On trouvera une argumentation plus detaillee concernant le suffixe -ierj-iere dans Corbin — Corbin (ä paraitre). 21. Voir par exemple Jackendoff (1975), Zwanenburg (1982, 1984, 1987: 64). 22. Cette troncation est recurrente, comme le montre l'exemple finalite —> finaliste,, au sens « Qui croit ä Taction des causes finales et, en general, ä la finalite comme explication de l'univers » (sens predictible specific: « Partisan de la finalite »). 23. Ce verbe, non atteste aujourd'hui, etait atteste au XIVC siecle (Greimas, 1968). 24. Si ce nom est derive d'anxieux. Je simplifie ici le traitement de ce mot. Pour des details, voir Corbin (1987: 505-508). 25. Extrait de Corbin (1987: 417). 26. Voir sur ce point Corbin (1987: 476-504). 27. Et, si necessaire, sens predictible specifie. 28. Je m'oppose sur ce point ä des propositions recentes comme celle de Roeper (1988) qui propose d'inserer des affixes, qui selon lui sont des «tetes», sous des nceuds syntagmatiques. 29. A l'exception, remarquable, de Botha (1988).

58

Danielle Corbin

References

bibliographiques

Aronoff, Mark 1976 Word formation in generative grammar (Linguistic Inquiry Monograph 1). (Cambridge, Mass.: MIT Press). Booij, Geert 1979 "Semantic regularities in word formation", Linguistics 17: 985—1001. Botha, Rudolf P. 1988 Form and meaning in word formation. A study of Afrikaans reduplication. (Cambridge: Cambridge University Press). Corbin, Danielle 1987 Morphologie derivationnelle et structuration du lexique, 1—2. (Tübingen: Niemeyer). 1988 «Une hypothese a propos des suffixes -isme, -ique, -iste du frangais: la troncation reciproque », in: R. Landheer (ed.) Aspects de linguistique franQaise. Melanges Q. I. M. Mok (Amsterdam: Rodopi), 63 — 75. a paraitre «Contre une transposition de la theorie X a la morphologie derivationnelle», Acta Linguistica Hungarica 37. Corbin, Danielle —Corbin, Pierre ä paraitre «Pour un traitement unifie du suffixe -ier/-iere», Lexique 10. Dressier, Wolfgang U. 1987 "Word formation (WF) as part of natural morphology", in: W. U. Dressier et al., Leitmotifs in natural morphology (Amsterdam: Benjamins), 99 — 126.

1982—1985 Grand dictionnaire encyclopedique Larousse. (Paris: Larousse), 10 vol.

1960 — 1964 Grand Larousse encyclopedique en dix volumes. (Paris: Larousse), Supplements 1968, 1975. Greimas, Algirdas Julien 1968 Dictionnaire de l'Ancien Frangais. (Paris: Larousse).

1985 Le Grand Robert de la langue Frangaise. Dictionnaire alphabetique et analogique de la langue frangaise de P. Robert, 2e ed. entierement revue et enrichie par A. Rey. (Paris: Dictionnaires Le Robert), 9 vol. Jackendoff, Ray 1975/1977 « Regularites morphologiques et semantiques dans le lexique », in: M. Ronat (ed.) Langue. Theorie generative etendue (Paris: Hermann), 65 — 108.

1928 — 1933 Larousse du XX" siede en six volumes publie sous la direction de P. Auge (Paris: Larousse). Lieber, Rochelle 1980/1981 On the organization of the lexicon [Dissertation, MIT] (Bloomington: Indiana University Linguistics Club).

1986 Le Petit Robert. Dictionnaire alphabetique et analogique de la langue frangaise. (Paris: Le Robert) [1(1967), 2(1977)]. Puchulu, Agnes 1987 Etude des suffixes -aire dans le cadre de la morphologie derivationnelle [Memoire de maitrise , Universite de Lille III]. 1988 Propositions pour une interpretation semantique des adjectifs denominaux [memoire de DEA , Universite de Lille III].

Associative

et stratification

59

Roeper, Thomas 1988 «Arguments implicites et la relation Tete-Complement», Lexique 7: 121-141. Selkirk, Elizabeth 1982 The syntax of words (Linguistic Inquiry Monograph 7). (Cambridge, Mass.: M I T Press). Toman, Jindrich 1987 Wortsyntax. Eine Diskussion ausgewählter Probleme deutscher Wortbildung. (Tübingen: Niemeyer). Zwanenburg, Wiecher 1982 « Types de derivation comme universaux », Recherches de linguistique franςaise d'Utrecht (Utrecht: Institut d'Etudes frangaises), 57 — 66. 1984 "Word formation and meaning", Quaderni di Semantica 5: 130 — 142. 1987 «Structure et forme des adjectifs denominaux», Recherches linguistiques d'Utrecht·. 5 5 - 8 0 .

Formal relations and argument structure* Anna-Maria Di Sciullo

0. Introduction This paper is a refinement of the argument structure calculi of Di Sciullo —Williams (1987), where the argument structure of a morphological object 1 is calculated in terms of the arguments of its parts. 2 We will identify relations between the elements of argument structures, preserving the idea that there is a basic difference between the external argument (the italicized variable in an argument structure) and the internal arguments. We will first discuss the notion of head, then consider different ways in which a head may be related to a non-head, and propose wellformedness conditions for argument structures.

1. The head relation We will assume that the external argument is the head of the argument structure. But how can we define the head with respect to the argument structure in morphological objects which contain more than one argument structure? A possible definition is found in (1), if we define feature F as being the argument structure: (1)

Head F (read: head with respect to the feature F) The head F of a word is the rightmost element of the word marked for feature F. (Di Sciullo-Williams 1987: 26)

In English, (1) holds for a large class of morphological objects. Therefore, in (2), for instance, the suffix -ed has no argument feature, and thus, * This work was conducted as part of the Argument Structure Project at the Linguistics Department of the Universite du Quebec ä Montreal. Support for the project was provided in part by the SSHRCC grant no. 4 1 0 - 8 6 - 0 7 6 0 .

62

Anna-Maria

Di Sciullo

even if it is the rightmost element of the word, it is not the head with respect to that feature. Of the two constituents with the relevant features, it is the rightmost which is the head with respect to the argument structure. Thus, the external argument of formalize is the external argument of the verbal suffix -ize and not the external argument of the adjective formal? (2)

formal\A ω

ize]v ed\w (x, y) (*, y)

There are languages where (1) is empirically false. Lieber (1988), for instance, reports cases of Vietnamese and Breton compounds, as well as affixed forms in Tagalog, where the head, with respect to the categorial features, is on the left. Moreover, (1) may fail within the same language. Consider the fact that, in Italian, the head with respect to the category, and for our purpose the head with respect to the argument structure, is on the right in affixed forms and generally on the left in compounds. (3a) b)

dimostrazione/ mangiabiL·/ giocare 'demonstration' 'eatable' 'to play' «ave-tragetto/ blu-notte/ «ma-forte 'ferry-boat' 'night-blue' 'safe'

We propose a revision of (1) in terms of types of words (henceforth word T ) as in (4), where word T includes suffixed forms and compounds. (4)

The head F of a word T is the X-most element of a word marked for feature F. Parameter: X-most: initial/final position.

It has been suggested in the literature, in Lieber (1988) for instance, that the head of a word could be defined only in terms of inherent features, and not in terms of positions. The definition in (5) expresses this view: (5)

The head of a word is the element whose features F project to the word.

However, (5), as opposed to (4), does not account for the cases where two elements of a word are marked for a given feature F, as in (2) for instance. It is not clear how a proposal which does not define the head in terms of positions can account for these cases. Furthermore, (4) accounts for the language variation noted above. This is done by fixing the variable X to initial or final position, according to the word T .

Formal relations and argument

structure

63

Basically, the definition in (4) develops the idea of relativized head, and for our purposes, it defines the head with respect to the argument structure. This notion of head is specific to the morphology. There can be no relativization of the head within the syntax, because there is only one potential head in a phrase.

2. Control relation In Di Sciullo — Williams (1987), a head with respect to argument structure can be related to a non-head by controlling an argument of the nonhead, as in (6). We will assume here, as in Levin — Rapaport (1987), that control applies to the predicate argument structure variables. Thus in (6a), the head controls the external predicate-argument structure variable, and in (6c), the head controls the internal predicate-argument structure variable. Similar cases are found in French, as in (7), as well as in other languages. (6a)

employ er ( * i , Y )

b) c) d) (7)

(r, y)

fa)

the employer of John employ ee (r, x) (*. yd fa) the employee of John agresseur / agressi, 'aggressor' 'aggressed person' employeur / employe 'employer' 'employee' hacheur / hachis 'chopper' 'chop'

Let us define the control relation as stated in (8): (8)

In a morphological object, A controls Β iff, A is an argument of the headcontroi A controls an argument Β of the non-head control A and Β are co-indexed.

The control relation is determined by the lexical property of a listeme, which is the headcontroi. Control in morphological objects is distinct from

64

Anna-Maria

Di Sciullo

syntactic control, since in the latter case it is only the external argument that is controlled. The -e (-ee) nominals show that the internal argument can also be controlled, which is not the case in syntax.

3. Binding relation There is another type of relation which holds between the elements of argument structures, which is not discussed in Di Sciullo — Williams (1987), and which involves two predicate-argument structure variables of the same type, 4 such as the [Χ Χ] χ compounds in (9). (9a) b) c

)

d)

telephone book (r), producer-distributor / plumber-contractor (r i) (rO Italian English (y), deaf-mute j bitter-sweet (yd

(yd

up Oj, w i) blow (xj, yj)

on (y, w), onto / into (Vj, Wi) dry (x, y), stir-fry j slam-dunk (xj, yO

In (9a) and (9b), two external predicate argument structure variables are bound. The examples in (9c) and (9d) include the binding of two internal variables. These variables must be bound within the word. If it were otherwise, their derived predicate argument structures would be illformed, as in (10). Furthermore, according to the projection principle (Chomsky 1986), lexical structures must be represented categorially at every syntactic level. Thus, all the arguments of the predicate argument structures in (10) must be projected in the syntax. However, the syntactic realization of these structures is excluded since there is only one external argument position for every argument taking lexical head. (10a) b) c) d)

telephone-book: *(r, r) Italian-English: *(y, y) into: *(y, w, y, w) blow-dry: *(x, y, x, y)

We will define the binding relation which applies in morphological objects as in (11):

Formal relations and argument

(11)

structure

65

In a morphological object, A binds Β iff, A and Β are predicate-argument structure variables of the same type, A and Β are co-indexed.

A possible explanation for the existence of (11) is that it results from the following well-formedness condition of argument structures: (12)

Distinctivity All the variables of a predicate-argument structure are distinct.

(12) states that there is no word whose predicate-argument structure includes more than one instance of the same type of variable. This is the case for external arguments, since predicators may have only one external argument. One consequence of (12) is that the predicate-argument structure of verbs does not include two direct internal arguments. This is the case if we assume binary branching structures for the syntactic realization of arguments. Furthermore, the predicate argument structure of verbs does not include two indirect internal arguments associated with the same preposition, which is the case for verbs such as walk and fly, for instance. Distinctivity forces the binding of two instances of the same type of predicate-argument structure variable. When two variables are bound, they constitute a unique variable. Thus, the facts in (9) are a consequence of the interaction of (11) and (12). Morphological binding, as defined in (11), is formally distinct from syntactic binding. The syntactic binding of lexical anaphors, for instance, typically involves arguments that are not of the same type: an external argument and a direct internal argument.

4. Function-composition relation In suffixed forms, the head can functionally combine with the non-head, as in (13). As defined in Di Sciullo —Williams (1987), when function composition occurs, the argument of the head becomes the external argument of the whole, and the argument of the non-head becomes the internal argument of the whole. (13a)

lazi

ness

y)

construct

ion (r, x, y)

(*, y)

W

the construction of the city by the enemy

66

Anna-Maria

Di

Sciullo

In Di Sciullo —Williams (1987), function composition is dependent on specific suffixes, which we have called "functors". For instance, -ness, -ude, -ion, and -ment are functors, but not suffixes such as -ee or -er, which do not combine with the non-head, but control an argument of the non-head. However, note that function composition does not hold for examples such as those given in (14), since the argument structures in (15), which are obtained by function composition, are not proper argument structures for (14). (14a)

univers al (r)

b)

(15a) b)

(y), rudimentary, courageous, alpine

(y)

box er (χ, y), monopoliser, coudoyer (r) (*, y) 'to monopolize' 'to elbow' 'to box' * 0 , r) *(x, y, r)

We want to explain this fact and reduce the reference to specific functors in the system, by proposing that function composition is the general case for the combination of arguments in suffixed forms, and that its non-occurrence is attributed to (16): (16)

Specificity The semantic nature of the external predicate-argument structure variable of a word is specific to its category.

The semantic nature of an external predicate-argument structure variable is fixed at the lexical-conceptual structure level. For a nominal such as construction, for example, the r predicate argument structure variable is projected from a semantic concept which is "the process r" or "the result r"; for a derived adjective such as fearful, the y variable is projected from a semantic concept which is "the affected thing y", and so on. 5 According to (16), there are semantic types of arguments which cannot be the external arguments for certain categories. This is the case for r, which cannot be the external argument for words other than nouns. Moreover, r cannot be the internal argument for any other category, as it is specific to the external argument of nouns. Thus, specificity blocks function composition in (14), and prevents r from being the internal predicate argument structure variable of adjectives and verbs.

Formal relations and argument

structure

67

5. Summary In this paper, we developed the theory of morphology proposed in Di Sciullo —Williams (1987), which calculates the categorial and argument structure properties of derived words from the properties of their parts. Formal relations hold between the sub-parts of derived words; refinements to these relations were suggested, in particular with respect to the head and binding relations. For all of these morphological relations between arguments, we indicated ways in which they differed from equivalent syntactic relations. Moreover, the well-formedness conditions of argument structures that we have proposed affect function composition and binding, as well as account for different cases of affixation and compounding which the former argument-structure calculi had left unexplained.

Notes 1. In Di Sciullo —Williams's (1987) theory, a morphological object is a word whose properties are derived from the morphological laws. Regular suffixed forms and compounds are morphological objects. They are not part of the lexicon which includes listemes and syntactic atoms. 2. In Di Sciullo —Williams (1987), an argument structure was defined as a set of theta roles (agent, theme, etc.). We will assume here that a predicate argument structure is a set of variables (r, ..., x, y, z) selected by a predicator, that is an argument taking a lexical head, as in Levin — Rappaport (1987). The variables in the predicate-argument structure serve as placeholders for arguments, which must be saturated at some level of grammatical representation. They are distinguished by the way in which the NPs corresponding to the arguments they stand for are realized in D-structure positions. The predicate-argument structure contains variables which stand for the external argument, the direct internal argument and the indirect internal argument of a predicator. These variables are projected from concepts which can be defined in terms of deeper semantic predicates at the level of lexical conceptual structure. 3. The following interpretations can be used for the predicate-argument structure variables: " r " is realized at D-structure when a nominal head meets the specifer (as in Higginbotham 1985); "x" is agent; "y" is the theme; and " w ' " is the location. I will assume here that theta roles are not primitives in the grammar. 4. We will assume that there are only three types of such variables: external, direct internal, and indirect internal. When two variables are of the same type, they are both either external, direct internal, or indirect internal. Furthermore, two indirect internal variables are of the same type if they are associated to the same preposition. 5. It would be possible to suppose that "event" would be the external argument for verbal predicators. This would make the external argument of nouns and verbs very similar, to the extent that they would not be realized as NPs, but via functional categories: Determiner and Inflection respectively. We will not investigate this hypothesis here, given space limitations.

68

Anna-Maria Di Sciullo

References Chomsky, Noam 1981 Lectures on government and binding. (Dordrecht: Foris). 1986 Knowledge of language. (New York: Praeger). Di Sciullo, Anna-Maria 1988 "Argument satisfaction, lexical and syntactic". [Paper presented at the GLOW 1988 workshop on the interaction of the lexicon with other modules of the grammar, Budapest], Di Sciullo, Anna-Maria—Williams, Edwin 1987 On the definition of word (Linguistic Inquiry Monograph no. 14). (Cambridge, Mass.: MIT Press). Grimshaw, Jane 1987 "Psych verbs and the structure of argument structure". [Ms. Brandeis University], 1988 "Adjuncts and argument structure", Lexicon Project Working Papers 21. (Cambridge, Mass.: MIT Occasional Paper no. 36). Hale, Ken —Keyser, Samuel Jay 1987 "Explaining and constraining the English Middle". [Ms. MIT], Higginbotham, James 1985 "On semantics", Linguistic Inquiry 16: 547 — 593. Jackendoff, Ray 1983 Semantics and cognition. (Cambridge, Mass.: MIT Press). Levin, Beth — Rappaport, Malka 1987 "Non-agentive -er nominals: probe into argument structure". [Ms. MIT], Lieber, Rochelle 1980 On the organization of the lexicon. [Ph. D. dissertation, MIT]. 1988 "Phrasal compounds in English and the morphology-syntax interface", Papers from the Parasession on agreement in grammatical theory, Chicago Linguistic Society, PCLS 24-11. Rappaport, Malka —Levin, Beth 1986 "What to do with theta roles", Lexicon Project Working Papers 11, MIT. Williams, Edwin 1981 "Argument structure and morphology", The Linguistic Review 1: 81 — 114. Zubizarreta, Maria-Louisa 1987 Levels of representation in the lexicon and in the syntax. (Dordrecht: Foris).

Austro-Hungarian morphopragmatics Wolfgang U. Dressier and Ferenc Kiefer

When studying meaning in morphology, we claim that the following four areas must be distinguished: 1. Morphosemantics = semantic meaning of morphological rules, i.e., regular semantic change between inputs and outputs of a morphological rule; here we can differentiate word formation rules and inflectional rules 2. Lexical semantics of morphology = semantics of morphologically complex words, especially of the semantic idiosyncrasies of each output of a morphological rule 3. Lexical pragmatics of morphology = idiosyncratic pragmatic meanings/effects of individual, morphologically complex words 4. Morphopragmatics = general pragmatic meanings/effects of morphological rules, i.e., regular pragmatic change between input and output of such rules. From a semiotic point of view, morphopragmatics (cf. Dressier — Merlini-Barbaresi 1987) comprises (a) the universal pragmatic foundations of morphology (e.g., for the semantic relations among personal pronouns), (b) the relations between morphological rules and their interpreters as well as the interpretant of a potential (or actual) output of such a rule (particularly a word-formation rule). In this dualistic investigation of morphopragmatic phenomena of Viennese German and the Hungarian of Budapest we will deal with the second aspect (b). Hungarian EXCESSIVE - German allerIn Hungarian, the excessive is formed by means of the prefix legesleg(the superlative has the prefix leg-) which is added to the comparative: e.g., nagy 'big' — nagy + obb 'bigger' — leg + nagy + obb 'biggest' — legesleg +nagy + obb 'the very biggest', and csiinya 'ugly' — csünyä + bb 'uglier' — leg + csünyä + bb 'ugliest' — legesleg +csünyä + bb 'the very

70

Wolfgang U. Dressler and Ferenc Kiefer

ugliest'. In contrast to the derived forms of nagy 'big', the derived forms of csünya 'ugly' are not used relatively, the things compared are presupposed to be ugly. What is the difference between the excessive and the superlative? Let us have a look at the pairs legeslegnagyobb — legnagyobb and legeslegcsünyäbb — legcsünyäbb. The excessive legeslegnagyobb can only be used of things which are big, i.e., it introduces a presupposition in contrast to the corresponding superlative which does not. On the other hand, the difference between legcsünyäbb and legeslegcsünyäbb cannot be purely presuppositional since the superlative already introduces the relevant presupposition. One important function of the excessive is, then, to introduce a semantic presupposition in cases where it is absent in the superlative. In order to find out what the pragmatic difference, if any, is between the superlative and the excessive, let us then concentrate on the pair legeslegcsünyäbb — legcsünyäbb. Compare the following sentences (1)

Εζ α legcsünyäbb rajz, amit valaha lättam. 'This is the ugliest drawing I have ever seen'.

(2)

Ez α legeslegcsünyäbb rajz, amit valaha lättam. 'This is the very ugliest drawing I have ever seen'.

(1) is a factual statement about the quality of the drawing, (2), on the other hand, is an emotionally colored statement, it is more naturally interpreted as an exclamation. (1) can easily be denied, it is not the last word to be said about the matter. On the other hand, (3) sounds odd (possible only as a denial) (3)

?Ez nem α legeslegcsünyäbb rajz, amit valaha lättam. 'This is not the very ugliest drawing I have ever seen'.

Apparently, the speaker attaches special importance to his qualification; he does mean it to be the last word on the matter. No further discussion is expected. The meaning of the excessive can thus be characterized in the following manner: (4)

The speaker believes that the thing considered possesses the absolutely highest possible degree of the property in question (and he wants the listener to share his belief).

(4) also explains the presence of the semantic presupposition. The highest possible degree of a property must, in fact, be qualifiable by

Austro-Hungarian

morphopragmatics

71

means of the property in question in an absolute sense. That is, the very biggest house cannot refer to a small house since the very biggest expresses the biggest possible. In other words, whenever the semantic presupposition is not present in the case of the superlative, it has to be introduced via the excessive. (4) formulates the discourse function of the excessive which is in principle independent of semantic presuppositions. One may thus claim that the excessive introduces a discourse function which clearly belongs to morphopragmatics, it is neither lexically conditioned nor a property of morphosemantics. The German correspondence of the Hungarian agglutinating prefix legesleg- is aller- (originally a genitive plural of all- 'all') preposed to a superlative, e.g., in Wissen Sie schon das Aller #neu + este? 'Do you know the most recent news of all?' (neu 'new' superlative neu + est-). Although less grammaticalized than its Hungarian equivalent (as is typical for compounding, vs. derivation, cf. Lehmann 1982), it seems to share the same pragmatic and semantic effects: it expresses the absolutely highest possible degree of a property and it is only compared with items which do have this property to a high degree; it is used for emphasis and for impressing the hearer. Therefore it is often found at the end of a text chunk, as in the "Auszählreim" (nursery rhyme): (5)

Du, bist schön und duj bist schön und duk die aller schönste. 'YoUi are beautiful and youj are beautiful and youk the most beautiful of all'.

If aller- is prefixed recursively (e.g., die alleraller allerschön + sie), then the respective form must be used at the end of the respective text chunk. An exception is — for diagrammatic reasons — aller φ erst 'first and foremost' which usually introduces the enumeration of things to be done first, cf. the Ancient Greek equivalent pän#prötos in Homer (Ilias VII 324, IX 93, XVII 568, Odyssey IV 577, IV 780). Similarly the excessive can be used as a corrective device in discourse: (6a) b)

Das ist sehr schlimm. 'That is very bad', Aber das aller φ schlimmste ist, dass X ... 'But the very worst of all is that X ...'.

Emphasis and interactive orientation of expressing the absolutely highest possible degree also explain the use of the German excessive as equivalent to Neo-Latin (and Hungarian!) superlatives in the following feudal terms:

72 (7a) b) c)

Wolfgang U. Dressler and Ferenc Kiefer

Der aller # durchlauchtig + ste Herr = Lat. serenissimus 'His (Ever) Most Serene Dignity' aller # untertänig + st 'most devote of all' Der aller # christlich + sie König = rex christianissimus 'His Most Christian Majesty' = le roi tres chretien (de France)

In the text worlds where these and similar expressions were used, they underlined the absolutely highest degree of the property present. Cf. allerhöchst 'the very highest' qualifying the absolute monarch or God. Typical for Viennese (but not only Viennese) irony are pragmatic reversals emphasizing the highest degree as in lexicalized Der Allerwerteste 'the dearest of all = buttocks' or in expressions such as Das wär' ja doch das allerschönste 'This would be the most beautiful ( = indecent, impertinent, scandalous) of all'. Thus the morphopragmatic properties of the Hungarian and German excessives are all-important; the introduction of a semantic presupposition is secondary. Diminutives Let us now turn to the Hungarian "diminutive" suffix -i. Nouns are formed from nouns by means of the suffix -i in two different ways: (a) -i is added to the truncated form of the initial noun, cf. csok-i (from csokoläde 'chocolate'), dir-i (from direktor 'director'), ov-i (from ovoda 'kindergarten'), fagy-i (from fagylalt 'icecream'), zong-i (from zongora 'piano'), szak-i (from szaktärs 'fellow-worker'), nyug-i (from nyugalom 'quiet'). (b) the suffix -i is added to the stem of the noun: läb-i (from läb 'foot'), comb-i (from comb 'thigh'), has-i (from has 'belly'), hus-i (from hüs 'meat'), ägy-i (from ägy 'bed'). As can be gathered from the examples cited, the class of nouns in (a) is quite heterogeneous. In contrast, the nouns (b) seem to belong to the semantic classes of body parts, baby food, and objects which are important in the small child's life. The suffix -i has no semantic meaning (it does not denote smallness): both (a) and (b) are nouns with the same denotations as the corresponding simplex nouns. If csoki and csokoläde 'chocolate' and läbi and läb 'foot' are semantically identical, is there any pragmatic difference between the corresponding forms? The forms in (a) presuppose a certain degree of intimacy and have often a jocular touch. They are used in school slang

Austro-Hungarian

morphopragmatics

73

as well as in adult communication. The forms in (b), on the other hand, belong to nursery talk, they are typically used in a discourse in which at least one small child participates (even if only as passive hearer), and they can also occur when addressing animals. It can also be transferred to the purely adult world in certain specific speech situations: lovers in the language of love, in attempts to please. Love and attempts to please are also elements of nursery talk. One may thus generalize the cases of transference by saying that the forms of nursery talk can be transferred to the adult world just in case one of the ingredients of nursery talk becomes the defining property of the speech situation. A detailed investigation of this problem would go beyond the scope of the present paper, however. To summarize, nouns in -i have a clear pragmatic function: they determine the social setting of the speech situation (in the case of (a)), or they determine the relationship between speaker and hearer (in the case of (b)). The suffix -/' has no other contribution to the meaning of the noun. That is, the suffix -i is a morpheme with exclusively pragmatic meaning. Notice that Hungarian -i (e.g., hus-i) does not denote smallness in talking to animals either. Similarly, the equivalent Austrian suffix -i can be extended to speech addressed to animals, e.g.: (9)

Hund-i/ Katz-i, da ist das Fleisch-i, dein Papp-i, 'dogg-ie/ catt-ie, here is the meat-ie your papp-ie, dein Tschapp-i your food-ie'

But only Katz-i may be used in the language of love, otherwise Fleisch + erl, Papp + erl must be substituted. (Tschapp+erl has a different meaning (and etymology) 'awkward person', but Tschapp + erl#salat 'the first leaves of lettuce' is semantically related). The primary locus of -i suffixation is child language and baby talk, cf. the nursery rhyme (10)

Kommt ein Maus-i aus dem Haus-i 'Comes a mous-ie out of its hous-ie'

Both Maus-i and Haus-i can be used when addressing pets, but only Maus-i would seem appropriate for language of love (cf. also dialect names for animals in Weber 1960: 120, 130, 136, 144, 146). In a children's book (Nöstlinger 1987) which captures Viennese language use quite well, diminutives in -i are only used by a four-year old child, by his granny when speaking to him or about him, and — rarely

74

Wolfgang U. Dressler and Ferenc Kiefer

— by the rest of the family when sarcastically referring to them (i.e., there is a pragmatic sanction against the transfer of Austrian diminutives in -i into a purely adult text world). This pragmatic sanction is even stronger for the (nominal) diminutive suffix -erl when attached to auxiliaries, principally hat 'has' and is(t) 'is' — this is only possible when addressing small children or pets in a jocular way, e.g., (11)

Ja, was is-erl denn? 'Oh, what is it?' (Nöstlinger 1987: 30)

All this contrasts strikingly with -erl affixation to nouns which is extremely productive in Vienna. Of course, we must exclude here lexicalized forms such as Mäd-erl (= Mädchen) 'small girl' vs. Mäd-l 'girl' which is typical for /-diminutives being more lexicalized than er/-diminutives. Lexicalized e/7-diminutives are, e.g., Buss-erl 'kiss', Flank-erl 'flake', Nock-erl 'dumpling', Pock-erl 'turkey, fir-cone'. There is no *Buss or *Pock (Buss-i occurs in child language and language of love), Flanke and Nocke have very different meanings (only Nock-i occurs in child language). Cf. the qualification by Willi Schlamm (1938/1987: 150): "Die Sprache der Wiener neigte zum Diminutivum" ('The speech of the Viennese tended towards the diminutive'), which is still valid today. Cf. for Southern German in general Brandstetter (1964), Schirmunski (1958). The er/-diminutives either denote smallness, i.e., a small size or a relatively small degree of the prototypical properties ascribed to the respective nouns, they may connote endearment, intimacy, irony, sarcasm, etc. (cf. Dressier — Merlini-Barbaresi 1987: 9ff.). Due to both meanings they are used more frequently in discourses in which at least one small child participates (even if only as a passive hearer or in his/her absence as a person referred to) and therefore er/-diminutives can be also used metaphorically for evoking childhood or connotations associated with it. The precise connotative effect of an er/-diminutive is text-pragmatically determined. Let us just mention a few generalizations. Whereas in socially convergent verbal interactions er/-diminutives tend to receive positive connotations, they are liable to receive negative ones in socially divergent interactions. When the referent of a diminutive clashes with the denotational meaning of the suffix, there is an effect of strong understatement, e.g., mein Wag-erl 'my car' (Wag-en 'car') when referring to a Rolls Royce.

Austro-Hungarian

morphopragmatics

75

This clash occurs only if the referent has a more than normal size, whereas a normal-sized referent may be easily expressed with a diminutive. This also explains the difficulty of accepting potential diminutives such as Ries-erl 'giant-ie'. This example illustrates also that the presence of the possessive 'my' in the microcontext conveys the connotation of endearment. However, the macrocontext may add negative connotations as in the oral narrative of a Jew who was forced to leave Vienna in 1938: (12)

So hab ich mein Koffer-l gepackt und bin weg. 'Thus I packed my suitcas-ie and went off.'

{Koffer-l is rather a haplology for *Koffer-erl than an instance of an /-diminutive.) Consider next the diminutive suffixes -kaj-ke and -cskaj-cske in Hungarian. The former occurs mainly after noun stems ending in I, r, n, m, ny, s or ό, ö, and i. That is, -kaj-ke also occurs after the suffix i: läbi-ka 'small foot', combi-ka 'small thigh', husi-ka 'little meat', etc. Its primary meaning is diminutive: level-ke 'small letter', asztal-ka 'small table', leäny-ka 'little girl', tänyer-ka 'small plate', kanal-ka 'small spoon', väros-ka 'small town', szekreny-ke 'small wardrobe' (some of these are lexicalized forms). The same with -cskaj-cske, e.g., felhö-cske 'small cloud', bokr-ocska 'small bush', vödr-öcske 'small bucket', läb-acska 'little small foot', vaj-acska 'little butter', terd-ecske 'small/little knee'. The diminutive meaning is exclusive except for the following semantic classes: body parts, food, and familiar objects (this latter class is yet to be defined more precisely). In the case of the above-mentioned semantic classes the diminutive may acquire a purely pragmatic function when it is transferred to the adult world (in the language of love, or if used jokingly in certain speech situations). In such cases denotative diminutive meaning disappears. That is, szäjacska 'small mouth', for example, will no longer mean small mouth but, say, a mouth which I love. Or, läbika may even be a big foot, in the language of love it means something like a foot to be caressed. In nursery talk, it may still keep the diminutive meaning while at the same time expressing affection. That is, depending on the speech situation, the diminutive suffixes may acquire either a purely pragmatic meaning or, in addition to the denotative meaning, certain pragmatic overtones. To sum up, it would seem that both Austrian and Hungarian diminutives prove essentially the same thing, i.e., that diminutives may have purely pragmatic functions. All morphological rules which contain a pragmatic variable in the description of their meaning are morphopragmatically relevant.

76

Wolfgang U. Dressler and Ferenc

Kiefer

On separation of form and meaning If our assumption is correct that morphopragmatics must be separated both from morphosemantics and from lexical pragmatics, then one can find a new argument against the strict separation of morphosemantic derivation rules and morphotactic affixation rules (as postulated, e.g., by R. Beard (this volume) and A. Pounder (1988)). According to Beard's (e.g., 1986, 1987) model there are rather few, general, and possibly universal, semantic derivation rules, but many diverse affixation rules. It would follow from these assumptions for our diminutives that we should try to posit one morphosemantic derivation rule for all diminutives of a language and then several morphotactic affixation rules (e.g., for Hungarian -i, -cska, -ka or Viennese -i, -I, -erl, -lein). But this implies that the meaning effects of these word formation rules should be synonymous. However we have found various (and regular) differences (a) in denotative semantics (e.g., Hung. -/ not denoting smallness), (b) in connotations, (c) in their morphopragmatic potential. Thus, e.g., for Hungarian diminutives, we would have to postulate at least three different rules: two for -i and a third one for -cskaj-ka. In this way an apparent advantage of Beard's model, the simplicity and generality of morphosemantic derivation rules, would vanish. And this situation is not unique for diminutives, because most "synonymous" word-formation rules are not really synonymous (pace Pounder 1987), i.e., their putative synonymy resides in their classificatory labels such as diminutives, agent nouns, action nouns (cf. Benveniste's (1948) classical study). As far as morphopragmatics is concerned, it always depends on the particular combination of form and meaning as embodied in the various usual formats of morphological rules.

References Beard, Robert 1986

On the separation of derivation from morphology: toward a lexeme I morpheme-based morphology. (Bloomington: Indiana University Linguistics Club). 1987 "Morpheme order in a lexeme/morpheme-based morphology", Lingua 72: 1-44. Beneveniste, Emile 1948 Noms d'agent et noms d'action en indo-europeen. (Paris: AdrienMaisonneuve). Brandstctter, Alois 1964 "Semantische Studien zum Diminutiv im Mittelbairischen", Zeitschrift für Mundartforschung 30: 335 — 351.

Austro-Hungarian

morphopragmatics

77

Dressier, Wolfgang U. — Lavinia Merlini-Barbaresi 1987 Elements of morphopragmatics (Linguistic Agency, University of Duisburg, Series A, Paper 194). [also to appear in the 1987 International Pragmatics Conference proceedings] Lehmann, Christian 1982 Thoughts on grammaticalization. (Köln: Arbeiten des Kölner UniversalienProjekts 48). Nöstlinger, Christine 1987 Wetti und Babs. (Wien: Jugend und Volk). Pounder, Amanda 1987 Systemangemessenheit in der Wortbildung am Beispiel desubstantivischer Adjektivableitung im Deutschen. [Phil. Dissertation, Universität Wien.] 1988 "On the status of conversion as a word-formation process", in: W. Dressier et al. (eds.); Discussion papers, Third International Morphology Meeting, Krems 1988 (Wiener Ling. Gazette, Suppl. 8), 1 9 - 2 1 . Schirmunski, Viktor 1950 "Verstärkte Wortformen in den deutschen Mundarten", Zeitschrift fur Mundartforschung 26: 225 — 238. Schlamm, Willi 1938 [1987] "Das war Wien", Das Neue Tagebuch 6,18 (30. 04. 1938): 4 2 2 - 4 2 5 [Österreichs Fall, herausgeg. v. U. Weinzierl (Wien: Jugend und Volk), 147 — 157)]. Weber, Otto 1960 Haustierbezeichnungen in den steirischen Mundarten. [Phil. Dissertation, Universität Graz.]

Problems of word structure theories Wolfgang Mötsch

1. In recent theoretical approaches to word formation a particular level of word syntax has often been adopted. There are two major arguments in favor of this assumption: (1) Particular rules defining well-formed word structures seem to offer a proper basis for semantic interpretation; and (2) a representation of word structure seems to be necessary to demonstrate the operation of some general grammatical principles like thematic-structure satisfaction and feature percolation within word structures. However, if we look seriously at the theories proposed so far, some fundamental doubts will arise as to whether these arguments really hold. In this paper, I want to point out some crucial problems concerning this approach. I am going to argue in favor of a revised version of redundancy rules. 2. In my argumentation, I am going to touch three implications involved in the assumption of a separate level of word-syntactic structure: (i) There are rules determining well-formed syntactic words. These rules are either based on separate principles or turn out to be determined by general principles of phrase structure. (ii) Representations on the level of word structure are the basis to which processes of semantic interpretation apply. (iii) Word-structure representations allow for the application of general grammatical principles. 3. As to the first implication mentioned above, that is, the type of rules, adopted to account for properties of compounds and derived words, the most general device are rewriting rules of the type

X°

Y° x°

80

Wolfgang

Mötsch

There are more refined versions of this rule scheme, but they may be neglected for the purpose of my argumentation. Rewriting rules of this form have some general properties in common with phrase-structure rules of the X type. They are, however, restricted in some fundamental respects: (i) Only a limited subset of lexical categories may enter into these rules. (ii) Only zero-bar categories may appear in word structures, that is, no phrasal structures are allowed within word structures. (iii) Word-structure rules generate binary structures. (iv) One of the constituents of a word structure is identical with the dominating category. In our version of the rule scheme it is the right-hand constituent, which may be defined as the head of the construction. All predictions of this rule scheme are empirically more or less problematic. There are many counterexamples to each of them. This, however, is not necessarily a sufficient argument against the theory, because it may turn out to be an interesting generalization which represents the unmarked properties of word structure. This presupposes that deviating cases might be interpreted as marked exceptions. Unfortunately, a proper theory of markedness which excludes purely intuitive judgements on markedness is not available. Despite the lack of a theory of markedness, I think, there are some problematic predictions which — as far as I can see — may not be resolved by markedness assumptions. One general problem of wordstructure theories is the fact that some of the presumably marked cases cannot be described other than in an ad hoc way. Difficulties arise with: (i) The prominent role of affixation among other kinds of devices involved in morphological processes. (ii) Synthetic compounds and other types of words containing phrases, cf. German Arbeit-nehm-er, Saure-Gurken-Zeit English car driv-er, peace lov-ing, generative-grammar approach. (iii) The restriction to binary branching. Word-structure rules of the above-mentioned type generate hierarchical structures with lexical categories as final elements and as dominating nodes. It is assumed that the lexical insertion rule applies to these structures. This, in turn, presupposes that affixes — like roots — are to be analyzed as lexical entries. Affixes, of course, are considered to be

Problems

of word structure

theories

81

lexical entries requiring a sister constituent which belongs to a certain class of words. The assumption that affixes have a separate status as lexical entries, to make it clear, follows from the theory. In my view, the treatment of affixes as separate dictionary entries has two serious disadvantages: (i) It rules out morphological processes like conversion, discontinuous affixes (cf. German Gt-schrei-€), modification of the base string, reduplication, and subtractive techniques. (ii) It blurs the difference between two types of restrictions which affixes impose on their base words. There are restrictions which ultimately depend on the semantics of a derivational process and which are represented in terms of lexical categories and subcategories of the base word, and there are restrictions on purely morphological and perhaps phonological grounds. In German, to give some examples, the distribution of -heit und -keit depends on properties of the latter kind. The derivational process (entpref last N ) v en inf excludes complex words. Cf. entflecken but *entfettflecken. Furthermore, derivational affixes in German may be divided into two types. One of them attaches the affix directly to the stem, the other one requires an extended form of the base. Cf. frau-lich, weib-isch, Bieg-ung frauen-haft, weiber-mäßig, biegt.-fähig The distribution of class I and class II affixes in English seems to belong to this type of restrictions too. If this is correct, many of the so-called bracketing paradoxes may be simply paradoxical descriptions of the facts, because a description of these morphological regularities does not necessarily presuppose hierarchical structures. All attempts to extend the lexical entry approach to morphological processes other than affixation involves exceptions to general properties of lexical entries. We may, for example, admit phonologically empty affixes in order to account for conversion. Attempts to represent the regular nature of processes like modification of the base string by affixes may only lead to purely technical solutions. In other words, it seems to be impossible to reduce all types of morphological processes involved in derivational morphology to separate affixes, even if we adopt a very abstract concept of affix. A second exception to the standard conditions on lexical entries arises with prefixes. Prefixes are morphemes which typically lack an indication

82

Wolfgang Mötsch

of lexical category. This is simply a consequence of word structure properties. In word structures, only that constituent must be lexically categorized which bears the properties of the entire newly coined word. The category membership of the non-head constituent only plays a role in the formulation of conditions on the base to which a certain affix is attachable. It is redundant, as far as I can see, in the representation of the derived word. There are, apparently, no rules referring to the lexical category or subcategory of the non-head constituent of a complex word. One strong argument in favour of the lexical-entry approach to affixes seems to be the possibility to apply the concept "head" to word structures. We have to take into account, however, that the notion "head of a word" may not be defined without exceptions in many languages. In German and English, for example, there are prefixes which should be analyzed as heads. This, of course, would be an exception to the general definition of the head as the right-most constituent of a word. In German ent-laus-en, ent-wanz-en (which correspond to English debug) and be-stiick-en, be-last-en are examples of prefixes indicating the categorial status of the entire derived word. Thus, a dilemma remains in any case. Prefixes without category marking are an exception to the general property of lexical entries to be a member of a certain class of words. Prefixes with category marking are an exception to the assumed principle of the head position. One consequence of the assumption that prefixes must not be marked for lexical categories is the rejection of rules involving lexical categories in order to derive prefixed words. Lieber's (1980) modification of the word structure approach avoids this problem with prefixes. It also avoids a further obscurity of rules containing lexical categories. These rules predict subclassifications like [A - Ν AffixA ] parallel to subcategorization in phrasal structures, cf. [VP - V NP]J However, whereas verbs requiring NPs as direct objects are an important syntactic generalization, no class of "adjective suffixes requiring nouns as co-constituents" is relevant in word structures. Restrictions concerning the base word in derivational processes are a purely individual property of affixes, not of classes of affixes. A crucial problem for all sorts of word-structure theories evolves from synthetic compounds which may be observed in many languages. In

Problems

of word structure

theories

83

English and German, there are many highly productive types of synthetic compounds, that is, processes which attach an affix to a phrasal base. However, this phrasal base is restricted to a concatenation of words, that is, no normal phrase is allowed. As far as I can see, no word-structure approach has a sufficient solution to the problem of how restricted phrases may enter into word-structure trees. Even if we agree with an analysis like A

/ gle

it will remain a secret how phrasal substructures of this restricted sort may enter into word trees on the basis of solutions that are other than ad hoc. 4. Let me now turn to questions of semantic interpretation. The first observation concerning word structures is that they do not offer any direct basis to semantically necessary distinctions between subordinated, coordinated, and exocentric compounds. The interpretation depends on information available from the lexical entries which are part of a compound, in particular on information about the thematic structure of relational elements. If a compound lacks a relational element, a relation will have to be added by semantic processes. I think it is quite clear that the concatenation of lexical categories in words neither depends on genuine word-syntactic principles nor does it offer interesting restrictions to semantic interpretation. Syntactic information involved in rules of the above-mentioned type appears to be only a reflection of underlying semantic processes. 5. It should furthermore be realized that the concept "head of a word" is of little help if we want to try to extend devices of feature percolation to the description of argument-inheritance phenomena. In these cases the

84

Wolfgang

Mötsch

verb, that is, the non-head constituent of a nominalization, transmits its thematic structure to the whole word. In any case, argument-structure inheritance in nominalizations is only one case of the reconstruction of theta-grids involved in derivational processes.

6. To conclude, some of the implications of word-structure theories seem to be in a serious conflict with facts we want to be covered by a theory of word formation. Some of the attempts to account for these facts force us to accept deviations from standard assumptions. Facts like purely morphological devices and restrictions cannot be sufficiently accounted for in this theoretical framework. A level of word-syntactic structure seems neither to be justified by processes of semantic interpretation nor by the description of thematic structure reconstruction involved in derivational processes. One major disadvantage of the word-structure approach is the blurring of the relatively independent purely morphological processes in word formation on the one hand and the semantically-based processes on the other. In my view, redundancy rules offer a descriptive device which avoids most, perhaps all, of the problems mentioned. Redundancy rules of the type sketched by Jackendoff (1975) following Chomsky's (1970) proposal enable us to make explicit all aspects of relatedness between base words and derived words, as well as between pairs of words and compounds. Redundancy rules may be divided into two parts as Jackendoff has already observed. One part accounts for morphological conditions on the concatenation of stems and affixes. The other one is organized by properties of the semantic, and in part of the thematic, structure involved in word-formation processes. Redundancy rules describe types of word formation as global processes of sign creation. In derivational processes, there is no need to designate a head, because the category marking of the derived word is a property of the entire word. Each rule application results in a representation of the lexical-entry properties of its products. There is no need for special rules generating hierarchies. The existence of more complex words is simply a consequence of the fact that some affixes (and compounds in general) allow complex bases. Synthetic compounds may be analyzed as a combination of compounding and derivation, quite in the spirit of the traditional analysis, that is, there is no syntax involved. The concatenation can be described by special types of redundancy rules.

Problems

of word structure

theories

85

Of course, a theoretical approach has to offer restrictions to the very powerful device sketched by Chomsky and Jackendoff. In general, these restrictions must account for principles underlying the morphological and semantic aspects of word-formation processes. Only if this has been done can the two approaches be compared more thoroughly. What we also want to be added to a theory of word formation is an evaluation of different realizations of possible structural types or processes. A scale of markedness seems appropriate to do that. The proposals by Dressler et al. (1987) concerning markedness involved in derivational processes is an illustration of what I have in mind. There is still one further argument in favor of redundancy rules which I want to point out. Rules of this sort enable us to account for lexical relatedness of all kinds in lexicalized complex words. The prediction of new words and the application of redundancy rules to coin new words may be considered to be a special function of these rules which presupposes some particular properties of redundancy rules, cf. Mötsch (1988). Thus, we may in principle use the same theory to account for highly productive processes and for restricted relations between lexicalized words which, in turn, may be models for analogical word formation.

References Chomsky, N o a m A. 1970 "Remarks on nominalization", in: R. A. Jacobs — P. S. Rosenbaum (eds.) Readings in English transformational grammar (Waltham, MA: Ginn), 184 —

221. Dressler, Wolfgang U.— Willi Mayerthaler —Oswald Panagl — Wolfgang U. Wurzel 1987 Leitmotifs in natural morphology. (Amsterdam: Benjamins). Jackendoff, Ray 1975 "Morphological and semantic regularities in the lexicon", Language 51: 639-671. Lieber, Rochelle 1980 On the organisation of the lexicon (Cambridge, M A ) [MIT dissertation], Mötsch, Wolfgang 1988 "On inactivity, productivity, and analogy in derivational processes", Linguistische Studien 179: 1—30.

Constraints on the Italian suffix -mente* Sergio Scalise

It has been maintained that the suffix -ly in English, being fully productive, could be considered "inflectional". 1 In this paper I intend to demonstrate that the Italian suffix -mente, which may be viewed as equivalent to English -ly, although very productive, is not "fully" productive. On the contrary, it is subject to a number of subtle semantic and morphological constraints which cannot be interpreted as "inflectional". It is therefore claimed that -mente (and consequently -ly) are derivational affixes. Constraints on simple adjectives and complex adjectives (compounded and derived) will be examined in that order.

1. Negative Constraints 1.1 Simple adjectives -mente cannot be attached to possessive, demonstrative, indefinite, or numeral adjectives: (la) b) c) d)

mio quest ο qualche due

—• *miamente —*• *questamente —• * qualchemente —• *duemente

'my' 'this' 'some' 'two'

As far as qualifying adjectives are concerned, in general, the most significant constraints are semantic. That is, -mente cannot be attached to adjectives denoting physical properties: (2)

brutto bello calvo

—• *bruttamente —• *bellamente —> *calvamente

'ugly' 'goodlooking' 'bald'

* This paper is based on Scalise et al. (in press).

88

Sergio

Scalise

Nor can the suffix ever be attached to adjectives denoting color: (3)

giallo viola

—> *giallamente —> *violamente

'yellow' 'purple'

When the base adjective has more than one meaning — a "literal" one and a more abstract "metaphoric" one — m e n t e always selects the latter (cf. also 1.3.3.1): (4)

arido 'arid'

1: 'free from water' 2: 'lacking feeling'

—> *-mente —» aridamente

Finally, there appears to be a constraint relating to the "spatio-temporal" meaning of the base. If an adjective has two readings, one referring to time and the other to space, -mente selects the temporal meaning: (5)

lungo 'long' 1: 2:

(rel. to space) (rel. to time)

—* *-mente —> lungamente

1.2 Compounded adjectives -mente cannot be attached to compounded adjectives even if it can be attached to the second element of the compound. This is true both in the case of "strict" compounds (6a) and in the case of "loose" compounds (6b): (6a)

b)

[dolce] [amaro] + *mente but cf. amaramente dolcemente [storico] [critico] + *mente but cf. criticamente storicamente

'sweet and sour' 'bitterly' 'sweetly' 'historical-critical' 'critically' 'historically'

This appears to be a "morphological" constraint. However, another explanation may be offered: in a model of ordered morphology such as "lexical morphology", composition follows derivation and therefore the derivation of compounds is excluded a priori. 1.3 Complex adjectives Three subclasses of complex adjectives will be considered here separately: (a) superlative/comparative adjectives, (b) adjectives modified by an evaluative suffix, and (c) adjectives deriving from a lexical category X by adding an adjective-forming affix.

Constraints on the Italian suffix -mente

89

1.3.1 Comparative/superlative adjectives Only the "synthetic" comparatives from this class will be tested since analytic comparatives cannot receive a suffix (except inflection for gender and number). The suffix -mente can be attached to a superlative but not to a comparative: (7)

ottimamente *migliormente pessimamente *peggiormente

ottimo migliore pessimo peggiore

'best' 'better' 'very bad' 'worse'

As regards the superlative form of the adjective, on the other hand, the adverb formation is fully productive: (8)

leggero leggerissimo tranquillo tranquillissimo

leggermente leggerissimamente tranquillamente tranquillissimamente

'light' 'very light' 'quiet' 'very quiet'

This pattern is very regular and can be summarized as follows: if A + mente, then A + issima +mente will always be possible whereas if *A + mente, then it will hold true that *A + issima + mente. 1.3.2 Adjectives with evaluative suffixes -mente cannot be attached to adjectives modified by an evaluative suffix: (9)

grande grandicello leggero leggerino

—> grandemente —> *grandicellamente —> leggermente —> *leggerinamente

'big' 'quite big' 'light' 'rather light'

This case as well may be explained in terms of order. In Scalise (1988), an attempt was made to demonstrate that the evaluative suffixes occupy a particular position in the system, that is, they apply "after" the normal word-formation rules. The output "adjective + evaluative suffix" therefore is not available for the rules adding -mente. 1.3.3 "Derived" adjectives What follows is a list of the suffixes forming Italian adjectives and the results of -mente attachment:

90

Sergio

Scalise

(10) *

Yj*

1. 2. 3. 4. 5.

Y/*

6. -aneo

* * *

*

Yj* *

-accio -acco -ace -aceo -ale

7. -ano 8. -ante

YJ*

9. -ardo 10. -are

Yj*

11. -ario

*

Yj*

12. 13. 14. 15.

Yj*

16. -ato

* *

Yj* * * *

Yj*

-asco -astro -ate -atico

17. -bile 18. 19. 20. 21.

*

22. * 23. * 24. Y/* 25.

-eccio -ello -eno -ente -eo -eo -escente -esco

*

26. -ese 27. -etto Y/* 28. -evole *

Y/* 29. -(i)ano *

30. 31. * 32. * 33. Y/* 34. *

-iccio -icello -icino -ico 1 -ico 2

golosaccio polacco seguace cartaceo postale naturale cutaneo temporaneo isolano ignorante brillante savoiardo immobiliare militare funerario arbitrario comasco biancastro arpinate asmatico dogmatico stellato dettagliato utilizzabile amabile mangereccio miserello madrileno adiacente precedente europeo argenteo fosforescente cardinalesco principesco bolognese furbetto pieghevole amichevole transiberiano cristiano torbidiccio grandicello grandicino libico desertico barbarico

— • — •

— > — •

— v — • — •

— V — •

—• — •

— • — •

— • — • — > — •

— • — •

— • — • — • — • — • — • — • — •

— •

—>

— • — » —>·

— • —».

—> —>

—>·

—»· — •

*golosacciamente *polaccamente *seguacemente *cartaceamente *postalmente naturalmente *cutaneamente temporaneamen te *isolanamente *ignorantemente brillantemente *savoiardamente *immobiliarmente militarmente *funerariamen te arbitrariamente *comascamente *biancastramente *arpinatemente *asmaticamente dogmaticamente *stellatamente dettagliatamente * u tilizzabilmen te amabilmente *mangerecciamen te * miserellamente *madrilenamente *adiacentemente precedentemente *europeamente *argenteamente *fosforescen temen te * cardinalescamente principescamen te *bolognesemente *furbettamente *pieghevolmente amichevolmente * transiberianamen te cristianamente *torbidicciamente *grandicellamente *grandicinamen te *libicamente *deserticamente barbaricamente

'greedy' 'Polish' 'follower' 'papery' 'postal' 'natural' 'cutaneous' 'temporary' 'insular' 'ignorant' 'brilliant' 'Savoyard' 'immovable' 'military' 'funerary' 'arbitrary' 'from Como' 'dirty white' 'from Arpino' 'asthmatic' 'dogmatic' 'starry' 'detailed' 'utilizable' 'lovable' 'eatable' 'quite poor' 'Madrilenian' 'adjacent' 'previous' 'European' 'silvery' 'phosphorescent' 'cardinal' 'princelike' 'Bolognese' 'cunning' 'pliable' 'friendly' 'trans-Siberian' 'Christian' 'rather turbid' 'quite big' 'rather big' 'Libyan' 'desert-' 'barbarian'

Constraints on the Italian suffix -mente Y

35. -ido 36. -iero Y/* 37. -ifico *

*

Y/* Y/* * *

* * * *

Y YJ* *

Y/* * * * * * *

*

Y/* *

* γ

1*

* * *

*

38. -igno 39. Hie 40. -He 41. 42. 43. 44. 45. 46. 47. 48.

-ineo -ingo -ino 1 -ino 2 -ino 3 -(i)ota -issimo -istico

49. -ita 50. -ivo 51. 52. 53. 54. 55. 56. 57. 58.

-lento -occio -occo -ognolo -olino -olo -one -oso

59. -otto 1 60. -otto 2 61. -(t)orio 62. 63. 64. 65.

-trice -uccio -uto -uzzo

lucido alberghiero sudorifico pacifico asprigno versatile duttile vescovile signorile sanguineo ramingo salino tunisino beiI ino cipriota grande violinistico idealistico vietnamita purgativo impulsivo sonnolento belloccio bustocco amarognolo marzolino romagnolo furbone erboso coraggioso chioggiotto stupidotto venatorio obbligatorio ricama trice caruccio barbuto avaruzzo

— • — • — • — * — > •

— •

— > — •

— • — • — • — > — •

— •

— • — • — • — » — • — • — > — • — • — •

— >

—v — » — >

— • — • — > — >

— • — • — • —>.

lucidamente *alberghieramente * sudor ificamen te pacificamente *asprignamente *versatilmente duttilmente *vescovilmente signorilmente *sanguineamente *ramingamente *salinamente *tunisinamente *bellinamente *cipriotamente grandissimamen te * violinisticamente idealisticamente *vietnamitamente *purgativamente impulsivamente *sonnolentamente *bellocciamente *bustoccamente * amarognolamen te *marzolinamente * romagnolamen te *furbonamente *erbosamente coraggiosamen te * ch ioggio t tame η te * stupidottamente * venator iamente obbligatoriamente * ricamatricemen te * carucciamente *barbutamente * avaruzzamente

91

'polished' 'hotel-' 'sudorific' 'pacific' 'rather sour' 'versatile' 'ductile' 'episcopal' 'gentlemanlike' 'bloody' 'wandering' 'salty' 'Tunisian' 'quite nice' 'Cyprian' 'very big' 'rel. to violin' 'idealistic' 'Vietnamese' 'purgative' 'impulsive' 'sleepy' 'rather nice' 'from Busto' 'bitterish' 'of March' 'from Romagna' 'cunning' 'grassy' 'bold' 'from Chioggia' 'quite stupid' 'venatorial' 'obligatory' 'embroideress' 'pretty' 'bearded' 'rather stingy'

These data are very clear: 45 out of 65 suffixes never allow the affixation of -mente·, 20 suffixes sometimes allow and sometimes do not allow the affixation and only two "always" allow the affixation (including the superelative suffix -issimo, for which cf. 1.3.1). The quantitative data alone would suffice to show that the claims regarding the "full" productivity of the suffix under analysis prove to be empirically unfounded. -mente "avoids" entire semantic sets: in fact it is not added to adjectives

92

Sergio

Scalise

denoting provenance (2,7,9,14,20,22,26,33,44,49,53,56,59) or to adjectives formed by evaluative suffixes (1,13,19,27,30,31,32,38,45,52,54,55,57,60, 63,65), or to material adjectives (4,23,43). The remaining suffixes include 18,24,36,39,41,51,62, and 64. Of the suffixes belonging to this group, -eccio, -escente, -ineo, -ingo, -lento, and -trice are marginal "adjectival" suffixes in that -ineo is a variant of -eo, -escente is a variant of -ente, -lento and -eccio are no longer productive and only occasionally does -trice form adjectives. As far as the two suffixes left, -uto and -iero, are concerned, see the reasons given in the next section. 1.3.3.1 Noun + adjectival suffix + -mente Among the suffixes listed in (10), there are some such as -oso, -ale, -ario, and -ico to which -mente may or may not be attached. In all these cases the affixation of -mente seems to be possible when the noun of the base is [ +abstract] but is not possible when the noun of the base is [—abstract]. The fact that a feature such as [ + abstract] can be relevant for a derivational process is not surprising: for example, the Italian suffix -aio can be added to [ — abstract] nouns but not to [ +abstract] nouns: (11)

libro pazienza

—> libraio —> *pazienzaio

'book — book seller' 'patience — *patience seller'

What is surprising, though, is the fact that the suffix seems to be sensitive to non-"adjacent" information. This would obviously be a violation of the adjacency condition because -mente would be able to "see" information contained in a non-adjacent cycle2: ]N [ — abstr]

+ Suf]A

+ mente]

However, if the noun of the base is [—abstract] and the derived adjective has two meanings, then the affixation of -mente is possible (but only with the "metaphoric" meaning): (12a) b)

vorticoso 'whirling' teatrale

'theatrical'

1 2 1 2

'full of whirls' 'very quick' 'related to the theater' 'dramatic, exaggerated'

—> * +mente —* + mente —• * -[-mente —> + mente

As can be seen, this constraint is the same as the one applying to simple adjectives (cf. (4) in 1.1).

Constraints

on the Italian suffix -mente

93

1.3.3.2 Verb 4- adjectival suffix + mente The suffixes -bile, -ante, -ente, -evole, -ile, and -ivo are added to verbs. After such suffixes, the derivation in -mente may or may not occur: (13)

asfissiante galleggiante prudent e contenente lodevole sdrucciolevole impulsive purificativo

asfissian temen te *galleggian temen te prudentemente * con tenen temen te lodevolmente *sdrucciolevolmente impulsiv amente *purificativamente

'asphyxiating' 'floating' 'prudent' 'containing' 'laudable' 'slippery' 'impulsive' 'purifying'

All of these cases seem to follow the same strategy: if the derived adjective is very transparent with respect to the verb, then affixation is blocked. If, on the other hand, the adjective has undergone a process of semantic drift, then the affixation is possible.

2. Positive constraints The suffix in point is also subject to positive constraints. One of the constraints affecting the various suffixal categories has to do with the notion of "time", -mente may or may not be added to a certain set of suffixes. If these suffixes form adjectives denoting "time", then -mente may be added freely: (14)

acciden t+ale 4- men te episod + ica + mente quotid + iana + mente regol + are + mente mens + ile + mente

'accidental' 'episodically' 'daily' 'regular' 'monthly'

This constraint must be linked to the one seen above for simple adjectives (cf. 1.1).

3. -mente and prefixation Prefixation may have effects on the attachment of -mente. Consider the following possibilities

94

Sergio

(15a) b) c) d)

if

Scalise

[\X\ +mente] *[[X] + mente] [\X\ +mente] *[[X] + mente]

then then but but

[pre + [X]] + men te * [pre + [X]] + men te * [pre + [X]] + men te [pre + [X]] + men te

My data provide evidence for all of the four cases in the following order of frequency: (b), (a), (c), (d) (only for the prefix in-). Examples of (15a) and (15b) are the following: (16a)

b)

critico acritico estetico antiestetico diluviano antidiluviano navigabile circumnavigabile

criticamente acriticamente esteticamente an tiesteticamen te *diluvianamente * antidilu vianamen te *navigabilmente * circumna vigabilmente

'critical' 'not critical' 'esthetic' 'antiesthetic' 'diluvian' 'antidiluvian' 'navigable' 'circumnavigable'

As regards group (d), the following examples will suffice: (17)

consolabile inconsolabile qualificabile inqualificabile

* consolabilmente inconsolabilmente * qualificabilmente inqualificabilmente

'consolable' 'inconsolable' 'qualifiable' 'not qualifiable'

As can be seen, the suffixation by -mente of an adjective ending in -bile is made possible by the presence of the negative prefix in-. The fact that a prefix can favor the attachment of a suffix is a counterexample to the "atom condition" (cf. Williams 1981), but this possibility has already been observed in connection with the suffix -mento and the prefix a- in Italian (cf. Scalise - Zannier 1982-1983: 195). The situation in (c) is more complex as a greater number of prefixes are involved. Let us consider, however, what happens when the base adjective is semantically complex, that is when it has both a literal and a metaphoric meaning as in the following examples: (18)

spasmodico 'spasmodic' 1 2 rigido 'rigid' 1 2

'affected by spasm' 'frenetic' 'hard' 'severe'

The consequences of prefixation for the attachment of -mente can be seen below:

Constraints on the Italian suffix -mente

(19)

rigido semirigido spasmodico antispasmodico

—> —> —• —>

rigidamente *semirigidamente spasmodicamente *antispasmodicamente

95

'rigid' 'semi-rigid' 'spasmodic' 'antispasmodic'

In general, a prefix selects one (and only one) meaning, actually the more "literal" one. 3 Also in the examples just given the prefix chooses the more literal meaning (antispasmodico 'antispasmodic', for example, refers to products to prevent spasms, not to something to avoid rushing, semirigido 'semirigid' does not mean 'severe but not too much') but doing so it automatically disfavors the attachment of -mente which, as we have seen, prefers metaphoric meanings. In connection with these facts (as well with others as observed in Scalise 1988), my claim is that prefixation can favor or disfavor only the attachment of derivational material and not of inflectional material.

4. Conclusions The suffix -mente seems to draw morphological distinctions between simple, compounded, and derived adjectives. It distinguishes between indicative and qualifying adjectives in the first case; it never applies in the second case, and it triggers a number of subtle distinctions in the third case. Among the derived suffixes, -mente clearly distinguishes between superlative (to which it always attaches), evaluative (to which it never attaches), and derived suffixes proper. In the latter case, -mente makes global semantic choices: it does not attach to adjectives denoting "provenance" or "material". As far as the other classes of adjectives are concerned, -mente is added only if the base is an abstract noun. Furthermore, when the base is an abstract noun and the derived adjective has two meanings, a literal and a metaphoric one, -mente selects the metaphoric meaning. Finally, this suffix is never added to adjectival bases deriving from verbs where the derivation V —> A is still clearly identifiable. Thus a present or past participle does not, as a rule, constitute the base for an adverbial derivation unless it underwent some sort of semantic drift. The positive constraints which have been singled out, on the other hand, are generally related to the notion of "time", including frequency.

96

Sergio

Scalise

As for the relation between -mente and prefixation, in some cases (especially with the prefix in- and the suffix -abile) prefixation favors suffixation which is typical of the relationship between prefixation and derivational suffixation, thus confirming the initial hypothesis that -mente behaves more as a derivational suffix than as an inflectional one.

5. Post-scriptum A reader of the present paper (whom I want to thank for his observations) suggested to me that the restrictions on the attachment of -mente can be explained very easily if one assumes that "A + mente" is possible when the morphological sequence corresponds to 'in an A way' (e.g., semplicemente 'simply' = 'in a simple way'). The reader maintains, furthermore, that in a language with a case "ablativus modi", the constraints on the adjectives are the same and therefore there is no point in arguing in favor of the derivational status of -mente. Even though such a proposal has some advantages, it cannot be accepted here for the following reasons. In the first place, the theoretical framework adopted here, being strictly "lexicalist", rules out the possibility of "matching", so to speak, words with phrases. In the second place, the offered explanation needs, in my opinion, to incorporate some diachronic information which is not within the reach of a synchronic model such as the one adopted here.4 In the third place, not every problem would be explained anyway. In general, if the offered explanation were correct, we would expect both the following implicational patterns: (20a) b)

if if

'in an A way' then A+mente *'in an A way' then *A + mente

This pattern holds true for many cases, but not for all. In fact we also find the following possibilities: (21a) b)

'in an A way' but *'in an A way' but

*A + mente A + mente

Examples of (21a) are the following: (22)

in in in in

modo modo modo modo

rivelatore calzante invidiabile distorto

*rivelatormente *calzantemente *invidiabilmente * dist or lament e

Constraints on the Italian suffix -mente

97

Examples of (21b) are the following: (23)

*in *in *in *in

modo modo modo modo

bimestrale estetico caldo grafico

bimestralmente esteticamente caldamente graficamente

The hypothesis in (20) is thus falsified in both directions. (20a) is falsified when the adjective is deverbal (cf. the examples in (22), and (20b)) is falsified in various cases: (i) when the adverb express "periodicity" (cf. (24a)), (ii) when the resulting adverb modifies adjectives (cf. (24b)), (iii) when the adverb can only be used in metalinguistic contexts (cf. (24c)): (24a)

*in modo settimanale

settimanalmente 'weekly'

b)

* in modo alto

altamente (prevedibile) 'highly (foreseeable)'

c)

* in modo batter iologico batteriologicamente (parlando) 'bacteriologically (speaking)'

I will not try to account for these data here. Clearly some aspects of the problem are open to further investigation, but I consider the evidence discussed in this paper as a little step in favour of the "derivational" status of the suffix -mente.

Notes 1. Cf. Bauer (1983: 8 4 - 8 5 ) and Bybee (1985: 225). 2. Cf. Siegel (1977). The reported observation would also be a violation of the "atom condition" of Williams (1981). 3. Cf. Mutarello (1987). 4. In Scalise (1983) the difference of status between -mente in Latin and in Italian is discussed.

References Bauer, Laurie 1983 English word-formation. (Cambridge: Cambridge University Press). Bybee, Joan 1985 Morphology. (Amsterdam: Benjamins).

98

Sergio Scalise

Mutarello, R. 1987 Alcuni aspetti delta prefissazione in italiano. [Unpublished thesis, Universita di Venezia.] Scalise, Sergio 1983 Morfologia lessicale. (Padova: CLESP). 1984 Generative morphology. (Dordrecht: Foris). 1988 "Inflection and derivation", Linguistics 26: 561 — 581. Scalise, Sergio —Irene Zannier 1982/83 "Restrizioni sulle regole di formazione di parola: la condizione di adiacenza e la condizione atomo", Quaderni Patavini di Linguistica 3: 159 — 210. Scalise, Sergio —F. Bevilacqua —A. Buoso —G. Piantini in press "II suffisso -mente", in: Raccolta di saggi in memoria di A. Limentani (Venezia). Siegel, Dorothy 1977 "The adjacency condition and the theory of morphology", ΝELS 7: 189-197. Williams, Edwin 1981 "On the notions 'lexically related' and 'head of a word'", Linguistic Inquiry 12: 145-274.

English compounds in Italian: the question of the head* Irene Vogel

1. Introduction In this paper, I will examine, and propose an account of, the fact that when Italian borrows a particular type of compound from English, typically only one of its members is retained (e.g., night club —• night). I will first describe the relevant data and then consider several hypotheses as to how to account for them. While the phenomenon in question represents a relatively small area of Italian morphology, it will be shown that it nevertheless involves a more general issue, the relationship between compound structure and basic sentential word order.

2. English compounds in Italian Italian began borrowing compounds from English regularly around the middle of the last century. Zolli (1976) cites, for example, the items in (1), where it will be noticed that the form used in Italian consists of only one member of the English compound. (la) b)

waterproof coat —* (It.) waterproof (1868) smoking jacket —> (It.) smoking (1888)

While (1 a) is no longer in use, (1 b) is very much alive. Other forms heard commonly include the following: (2a) b) c) d)

night club scotch tape water closet plaid blanket

—• night —• scotch —> water —• plaid

* I would like to thank Wolfgang Dressier, the anonymous reviewers, and the participants in the Krems meeting for their helpful comments.

100

Irene Vogel

A number of the items in (1) and (2) also appear in other European languages (e. g., smoking in French, Dutch, German). In such cases, it is not always clear whether each language borrowed the word from English independently or whether one language borrowed it first and other languages borrowed it from this one rather than from English. As Haugen (1988: 8) points out, there are serious difficulties "in teasing out the English models, their origins in speech or writing, their tortuous ways of reaching the European public, and the current results in the form of local replicas". While there may be some doubt as to the origin of a number of the earlier borrowings in Italian which also appear in shortened form in other languages, Italian differs from a number of other languages which have borrowed many of the same items. Of the fifteen languages and their treatment of English loanwords reported in the studies in Filipovic (1982), none shows as consistent a pattern of shortening of compounds as Italian. Instead, the full form is usually retained, though sporadic instances of shortening are also mentioned (e.g., Albanian strip 'comic strip' (Mehmeti 1982); Dutch keeper 'goal keeper' (Gerritsen 1982); Finnish klosetti 'water closet' (Oresnik 1982)). In Italian, on the other hand, the reducing of English compounds is quite productive and continues to contribute new forms to Italian such as those in (3) (T. Cravens, personal communication). (3 a) b)

Pink Floyd Rolling Stones

—• Pink —» Rolling

It should be noted, furthermore, that the shortening is not restrictecd to "popular culture" items among speakers who may not have a good command of English. In fact, (il) British is often heard as the shortened form of British Council among those who have occasion to speak of this organization (W. Dressier, personal communication). Finally, still further support for the productivity of the reduction rule comes from my own observations of the same process being applied by an Italian child who spontaneously produced the shortened forms in (4), having heard the full forms. (4 a) b) c) d)

self service pop corn snack bar beauty case

—• self —> pop —*• snack —» beauty

What is interesting about the way in which Italian borrows and shortens the type of compounds in question is not only that it reduces the com-

English compounds

in Italian

101

pound to a single word, but that the word that is retained is precisely the one native speakers of English would not choose if they were to shorten the same compounds. Compare the English examples and their Italian equivalents in (5). (5a)

night club:

b)

smoking jacket:

c)

scotch tape:

Let's go to the club. Andiamo al night. Give me the jacket, please. Dammi lo smoking, per piacere. I need the tape. Ho bisogno dello scotch.

The question that must be addressed now is why Italian reduces the compounds in the opposite way from the way in which they are reduced in English. This is not to say, of course, that all English compounds borrowed by Italian are shortened. Some are retained in their full form (e.g., T-shirt; pickup 'record player'). Others are used in either their full form or in the shortened form (e.g., blue jeansjjeans, basketball/basket). Still others have undergone translation, either retaining the English word order, particularly in earlier borrowings (e.g., banconota 'banknote', gentiluomo 'gentleman' (originally from French). While these patterns are rather sporadic in contemporary Italian, the one illustrated in (1) —(4) is productive and it is this one that will be the focus of the rest of this investigation.

3. Compounds in English and Italian The structure of the compounds under investigation here is either Noun + Noun (N + N) or Adjective 4- Noun (A + Ν). Ν + Ν and A + Ν compounds are common in English and in both cases the entire compound is always a noun, as illustrated in (6), in conformity with the "Is a" Condition stated in (7). (6a) b)

N + N: A + N:

[[water]» [lily] Ν ]n [ [blue]^ [cheese] N ]N

(7)

"Is a" Condition (cf. Allen 1978: 105) [[X] [Y] ] z : Ζ is a Y

102

Irene Vogel

Thus, water lily "is a" type of lily and blue cheese "is a" type of cheese. This follows from the fact that the head of complex words in English is, with few exceptions, on the right (cf. Williams 1981). That is, the category label and certain other features of the right member of a compound percolate up to the node that dominates it and its sister, as shown below. (8 a) watery b) bluet,

cheeseN

Italian, like English, has both Ν + Ν and A + Ν compounds, as illustrated in (9). (9a)

Ν + N:

b)

A + N:

croce via capo stazione alto piano giallo limone

'cross road' 'station master' 'plateau' (lit.: 'high level') 'lemon yellow'

Unlike English compounds, however, Italian compounds do not always have the head on the right, as shown below (cf. Scalise 1984). (10a) croceN

via^ N.

b) altoi

piano N

(11a) capoN

stazioneN

giallof

limoneN

b)

The compounds in (10) have the same structure as English compounds; those in (11) have the opposite structure. That is, while the features for syntactic category, as well as number and gender, percolate from the right

English compounds

in Italian

103

member of the compound in (10), they percolate from the left member in (11). As it turns out, however, only (11) represents a productive pattern, a point we will return to below.

4. Three hypotheses about Italian compound borrowing In analyzing the Italian data, we will consider the following hypotheses regarding the shortening phenomenon: HI: It is due to some basic linguistic strategy. H2: It is due to some aspect of English compounds. H3: It is due to some aspect of Italian compounds. To maintain HI, we would have to argue that there is something particularly salient about the first member of compounds, regardless of the characteristics of either the source language or the borrowing language. A basic strategy in this case might be something like "delete everything but the first word". This, however, would run counter to findings in other areas of linguistic behavior which indicate that it is the rightmost, not the leftmost, part of a string that is most salient. For example, it has been demonstrated in free-recall experiments that there is a strong recency effect in which it is the last part of the material presented that is the best retained (cf. Klatzky 1975). In addition, it has been observed that the ends of words tend to be more salient to children learning language than the beginnings. Slobin's (1973: 191) Operating Principle A for language acquisition, "Pay attention to the ends of words", expresses this strategy and accounts, for example, for the fact that children tend to learn suffixes before preposed elements with similar linguistic functions. A strategy that retains the first element of a compound solely on the basis of its position would therefore seem to lack independent motivation and, at best, be ad hoc. According to the second hypothesis, it is the nature of the English compounds themselves that is responsible for the Italian borrowing pattern. Both syntactically and semantically the first element, whether it is a noun or an adjective, in some way modifies the head of the compound. A more detailed examination of the semantic and syntactic relations between the two words, however, is not particularly revealing. Despite various attempts to categorize the relations between the members of

104

Irene Vogel

compounds (e.g., Levi 1978; Warren 1984a, 1984b), according to Selkirk (1982: 25), for the type of compounds under consideration "the range of possible semantic relations between the head and nonhead is so broad and ill defined as to defy any attempt to characterize all or even a majority of the cases". More generally, it is unlikely that the syntax and/ or semantics of one language would influence the borrowing patterns of another one since this would mean that the speakers of the borrowing language know the relevant structures of the source language. In fact, while we can assume that the first person or group to introduce an English compound into Italian knows the meanings of its component words and the relation between them, it would be unrealistic to make the same assumption about the majority of Italian speakers who subsequently incorporate the item into their vocabulary. The possibility that the phonology of the source language influences the borrowing pattern, however, is not as easy to dismiss since speakers of the borrowing language can be sensitive to certain aspects of the sounds they hear without knowing anything about the structure of the source language. Of relevance to the phenomenon under investigation is the often made observation that the main stress of noun compounds in English is typically on the leftmost member (e.g., Chomsky —Halle 1968). We might thus hypothesize that Italian retains the first word of a compound because this is the one bearing the main stress. Of the examples given above in (1) —(4), however, it turns out that only half follow this pattern, as can be seen below. (12 a)

b)

smoking jacket night club wäter closet pop corn snäck bar beäuty case waterproof coat scotch täpe plaid blanket Pink Floyd Rolling Stones self service

Thus, the phonology of the English compounds does not seem to be crucial either. At this point, then, we must also reject our second hypothesis, (H2), since there does not seem to be anything in the phono-

English compounds

in Italian

105

logical or in the syntactic and semantic structure of English compounds that could account for the Italian data. Finally, let us consider the third hypothesis. While it is true that Italian has noun compounds with the head on the right like English, it was shown above that there are also compounds with the head on the left. In fact, it is the latter type that is more common and represents the productive pattern in contemporary Italian. Of the almost 3,000 items listed as compounds by Tollemache (1945), approximately 560 are of the form NA, NN or AN. Approximately 68% of these have their head on the left, while only 32% have their head on the right.1 New ("loose") compounds when they are created in contemporary Italian typically have the form shown below, where Ή ' = head and 'M' = modifier. (13 a) 'ferry boat' valigiaH

armadioM

'wardrobe-valise' In contrast to these "loose compounds", the ones in which the head is on the right are generally older, lexicalized forms (e.g., gran cassa 'bass drum' (lit.: large box), mezzanotte 'midnight') or forms that have been translated from another language, maintaining the original word order (e.g., croce via 'cross roads') (cf. Tollemache 1945). From these observations, we see that what underlies the retention of the leftmost element of English compounds is the structure of Italian itself. That is, in its borrowing pattern, Italian is, in fact, respecting its own contemporary compound structure where the head is on the left. Hypothesis 3 is thus confirmed. If we now look beyond the structure of compounds to the basic word order of Italian, we find further support for H3. That is, the compound patterns are actually part of a more general phenomenon. From a typological point of view, it has been observed (cf. Greenberg 1963) that the unmarked pattern in SVO languages is to have modifiers follow the head. For example, in noun phrases, adjectives typically follow the noun, as do genitive constructions, as illustrated in (14) and (15), respectively. (14) (15)

Ν + A: Ν + Gen:

un uomo alto 'a tall man' (lit.: 'a man tall') il libro di Clelia 'the book of Clelia'

106

Irene Vogel

Since Italian is an SVO language, compounds with the head on the left are precisely what we would expect. The other cases are mostly translations from other languages with different word order patterns or from earlier stages in the development of Italian when the basic word order was not strictly SVO.2 Thus, we can conclude that in retaining the leftmost member of English compounds, Italian is behaving in a way consistent with other aspects of the language. That is, it is retaining the element that is in the crucial position according to its own structure. The fact that the crucial element (i.e., head) is on the right in English is irrelevant for Italian, as are other semantic and phonological characteristics.

5. Further considerations If the above analysis is correct, this should allow us to predict how a language will borrow compounds in those cases in which only a part is consistently retained through the operation of a rule. What is needed is information about (a) the internal structure of compounds in the borrowing language, and (b) the basic word order in that language. Specifically, what is predicted is that the position of the head in the borrowing language will determined that the member of the borrowed compound that occupies this position will be the one that is retained. Data from other languages, and in particular ones with different word orders than Italian are needed, of course, to evaluate the validity of the analysis beyond Italian. As we have already seen, however, there may also be data which at first glance tend to obscure the patterns of compound formation and borrowing. We must, therefore, take several additional factors into consideration, the most important of which is productivity. The proposal advanced here only deals with cases where there is a distinct rule in operation; it does not apply to sporadic instances of the reduction of borrowed compounds. In the case of Italian, it was seen that there are compounds with the head on the right as well as the left, so we might expect these to have an effect on the borrowing pattern too. On further examination, however, it was found that those with the head on the right typically had sources other than an Italian word formation rule (e.g., croce via 'cross roads', gentiluomo 'gentleman' translated from the analogous English or French compounds).

English compounds

in Italian

107

At first glance several other forms also seem to complicate the situation in Italian: (16 a) b)

blue jeans plum cake

—• jeans —• cake

While these items appear to contradict our generalization that it is the leftmost element of the English compounds that is retained, further investigation shows that they do not, in fact, constitute counterexamples to the above analysis. Jeans was borrowed from English already in its shortened form, and we can hypothesize the same for (16b), though I have never encountered an Italian who actually uses the short form cake (cited in Zolli 1976) rather than plum cake. Thus, it is often essential to know something about the way in which specific words enter a language. Other sources of borrowings may also provide relevant data. For example, in Italian, we find additional compounds of German origin which are shortened in the same way English compounds are: (17 a) b)

Volkswagen Blitzkrieg

—• ν ο Iks —• blitz

While blitz most likely entered Italian through English, already in its shortened form, volks appears to be a form Italian developed on its own. We thus have some confirmation of our prediction since the shortening phenomenon also appears to work the same way on compounds from at least one other language, though in general there are far fewer borrowings from other languages in Italian. Finally, the extent to which borrowed forms are integrated into the borrowing language must be taken into consideration. In the case of Italian, let us consider the small set of recently formed items illustrated in (18). (18 a) scuola bus b) fiera district

'school bus' 'fair district'

In these items, the head is on the right, contrary to the claim made above that the productive compound formation rule in Italian places the head on the left. These forms, however, are not felt to be Italian, and their low degree of integration appears to be determined not only by the fact that they contain elements that are, indeed, foreign, but also precisely because of the order of the words (i. e., the head is on the right). Related to this, we find some speakers who refer to gli Stones and i Floyd (rather than using the more common i Rolling and i Pink), and we also occa-

108

Irene

Vogel

sionally encounter the shortened form club, instead of the more regular form night. In the case of club, the "non-Italianness" is also usually marked phonetically. That is, we often hear [kleb], which seems to indicate that the speaker knows the English"u" is not pronounced [u], as it would be in Italian. That [ε] is not the correct English vowel either is not relevant here. What all these cases have in common is the fact that they reflect an attempt on the part of speakers to exhibit their knowledge of English by using forms that do not conform to the patterns of Italian. As long as such items are felt to be "exotic" and external to the system of Italian, they do not constitute counterexamples to the proposal advanced here. That is, they are not Italian forms and therefore do not need to conform to the linguistic patterns of Italian. To the contrary, they are intentionally used to signal knowledge of a different language.

6. Conclusions In this paper, I have investigated the way in which English compounds are borrowed into Italian. Three hypotheses were considered in accounting for the observed pattern whereby compounds, when borrowed, are reduced to only their first element. It was argued that neither a general strategy such as "keep the leftmost element" nor a strategy based on the syntactic, semantic, or phonological properties of the English compounds themselves could account for the data. Instead, it was proposed that the pattern observed in Italian borrowing depends on the structure of native Italian compounds and this, in turn, is related to more general characteristics of word order in Italian sentences. On the basis of the analysis of Italian, it was suggested, furthermore, that it might be possible to predict borrowing patterns involving shortening of compounds in other languages as well. Finally, several additional factors and apparent counterexamples to the proposed analysis were examined and the importance of such issues as productivity and degree of integration of borrowings were considered.

English compounds in Italian

109

Notes 1. This pattern may, in fact, be fairly common in other Romance languages as well. See Bauer (1978: 41), where it is reported that French, too, has a strong preference for (at least NN) compounds to have their head on the left, in contrast with Danish and English. 2. This is, in fact, not an isolated phenomenon. Changes in compound structure related to changes in basic word order have also been noted, for example, in Irish (cf. Ahlqvist 1985).

References Ahlqvist, Anders 1985 "The ordering of nominal compounds in Irish", in: J. Fisiak (ed.) Historical semantics — historical word formation (Berlin: de Gruyter), 1—9. Allen, Margaret 1978 Morphological investigations. [Ph. D. dissertation. University of Connecticut], Bauer, Laurie 1978 The grammar of nominal compounding. (Odense: Odense University Press). Chomsky, Noam —Morris Halle 1968 The sound pattern of English. (New York: Harper and Row). Filipovic, Rudolf (ed.) 1982 The English element in European languages. (Zagreb: Zagreb University Press). Gerritsen, Johan 1982 "English influence on Dutch", in: R. Filipovic (ed.), 1 5 4 - 1 7 9 . Greenberg, Joseph H. 1963 "Some universals of grammar with particular reference to the order of meaningful elements", in: J. H. Greenberg (ed.) Universals of language, (Cambridge, Mass.: M I T Press), 6 1 - 1 1 3 . Haugen, Einar 1988 "The influence of English: a transatlantic perspective", Folia Linguistica 22: 3-9. Klatzky, Roberta L. 1975/1980 Human memory. Structures and processes. (San Francisco: Freeman). Levi, Judith N. 1978 The syntax and semantics of complex nominals. (New York: Academic Press). Mehmeti, Ismail 1982 "A morphological and semantic analysis of the adaptation of anglicisms in Albanian", in: R. Filipovic (ed.) 28 — 56. Oresnik, Brigitta 1982 "On the adaptation of English loanwords into Finnish", in: R. Filipovic (ed.), 1 8 0 - 2 1 2 . Scalise, Sergio 1984 Generative morphology. (Dordrecht: Foris). Selkirk, Elizabeth O. 1982 The syntax of words. (Cambridge, Mass.: M I T Press). Slobin, Dan I. 1973 "Cognitive prerequisites for the acquisition of grammar", in: Ch. A. Ferguson—D. I. Slobin (eds.) Studies of child language development. (New York: Holt, Rinehart & Winston), 1 7 5 - 2 0 8 .

110

Irene Vogel

Tollemache, Federico 1945 Le parole composte nella lingua italiana. (Roma: Edizioni Rores di Nicola Ruffolo). Warren, Beatrice 1984a "Covert connectors", Quaderni di Semantica 5.2: 331—349. 1984b "The functions of modifiers of nouns", Quaderni di Semantica 5.1: 111 —123. Williams, Edwin 1981 "On the notions 'lexically related' and 'head of a word'", Linguistic Inquiry 12: 245-274. Zolli, Paolo 1976 Le parole straniere. (Bologna: Zanichelli).

The importance of combining forms* Beatrice Warren

1. Introduction Consider the morphemic and semantic analyses of the noun alcoholic. A morphemic analysis yields the two morphemes alcohol and ic; the semantic analysis could be said to yield three elements: "person"-"addicted to""alcohol". We can match the semantic element "alcohol" with the morpheme-form alcohol·, -ic, I would argue, is a morpheme-form devoid of lexical meaning; it does, however, serve to indicate a change from nonpersonal to personal reference in this case, and so indirectly it can be coupled with the element "person". The element "addicted to" is not formally expressed. It is a semantic element which, I suggest, was originally constructed and not retrieved from the mental lexicon. I have elsewhere (Warren 1988) outlined how the construction of unexpressed relational meanings within words and phrases can be accomplished. I will therefore not dwell on this aspect of the semantic analysis here. Let me only point out that meaning construction of this kind involves linguistic and extra-linguistic knowledge as well as the application of pragmatic rules of interpretation and also that the outcome of this meaning construction is only predictable provided we have the particular extralinguistic knowledge required. It need not have been "addicted to", but could have been, say, "selling", "producing", "appointed to prevent the imbibing of", etc. For some reason, however, "addicted to" is the relation which established itself and which is now a part of the meaning of the noun alcoholic and as such retrievable from the mental lexicon, although it would be difficult to assign it any special morpheme-form.

* As the reader will detect, the ambiguity of the title is intended: either interpretation being apposite. I would here also like to thank most heartily Gunilla Malmborn and Gabriele Stein — the former for pointing out inadvertencies in the original manuscript and above all for helping me excerpt examples; the latter for generously sending me some relevant articles.

112

Beatrice

Warren

Consider next the morphemic and semantic analyses of workaholic. According to word-formationists, its morphemic analysis yields: work plus a, which represents a linking phoneme, plus holic\ the semantic analysis yields "person"-"addicted to"-"work". The semantic element "work" can be matched with the morpheme-form work, which is a straightforward content morpheme; the two semantic elements "person" and "addicted to" can be linked with the rest of the phonemes of the word, i.e., (a)holic, thus making this sequence of phonemes into another straightforward content morpheme. Coming across the word workaholic for the first time, we have to associate with -(a)holic the semantic elements "person"-"addicted to". To do so, we crucially depend on knowing the meaning of alcoholic. The above claim, i.e., that -aholic is a straightforward morpheme is as a matter of fact infelicitous, since -aholic is a combining form and combining forms are morphemes of a rather special kind. Being neither proper roots, nor proper affixes, they upset the morphologists' neat subdivision of morphemes into roots and affixes. How then do wordformationists deal with combining forms? As we will see in the next section, standard works on word formation yield somewhat heterogeneous descriptions and definitions.

2. Survey of definitions of combining forms Adams (1973) discusses combining forms in a chapter entitled "Neoclassical compounds". She points out that the OED reserves the term "combining form" for first elements in neo-classical compounds and calls second elements endings. Thus, micro- would be an example of a combining form, whereas -scope would exemplify an ending. However, she herself does not make this terminological distinction. (See Adams 1973: 129, where she refers to -graph, -cide, -meter, and -scope as combining forms.) Also, in the supplements of the 1970s, the editors of the OED seem to have abandoned this distinction themselves. Quirk et al. (1985) claim that combining forms are almost obligatorily initial, giving psycho- in psychotherapy as a model example of a combining form. Semantically, combining forms are said to be like the first constituents of compounds, but in other respects they are like prefixes. That is to say, they tend to be bound morphemes and they tend to be unstressed.

The importance

of combining forms

113

Typically they are neo-classical. A vowel (usually -o- but often -/-) is inserted as a link between the combining form and the base. They point out that there are certain bases which are particularly common, viz. -meter, -graphy, -gram, and -logy. Bauer (1983) does not hesitate to apply the term to first as well as last elements, distinguishing simply between initial and final combining forms. Unlike prefixes, initial combining forms almost invariably end in a vowel. If they do not, the linking vowel -o- may be added. The crucial difference between initial combining forms and prefixes, according to Bauer, is that the former may combine with final combining forms whereas prefixes may not. Consequently, hyper- in hyperactive could be considered a prefix, but must be considered a combining form in hypertrophy. Similarly, the crucial difference between suffixes and final combining forms is that — unlike suffixes — final combining forms may co-occur with initial ones. In fact, Bauer (1983: 272) suggests that they invariably do. Marchand (1969) claims that the term "combining form" is usually employed for prepositive elements which derive from Latin or Greek stems of full words, whereas prepositive elements which go back to Latin or Greek prefixes, prepositions, or particles are termed prefixes. He does not think that this distinction is justified, however (Marchand 1969: 3.1.5). Nevertheless, he himself distinguishes between prefixes (bound morphemes that are prefixed to full English words) and prepositive elements which occur in compounding on a Neo-Latin basis, such as astro-, electro-, galato-, hepato-, osteo-, which he thinks are of "a purely dictionary interest". As far as final elements are concerned, Marchand seems to make the following distinctions: suffixes, semi-suffixes, and terminal elements. As examples of semi-suffixes he gives -like, -worthy, -monger, -way/ways, -wise, -wort, -wright. These are like full words in that their word character is still recognizable. They are like suffixes in that some of them are used only as second words of compounds (Marchand 1969: 4.80). Suffixes Marchand defines as elements which are not independent words and which are tacked on to full English words or to allomorphs of English words. Last elements which are not combined with full words or allomorphs of full words, such as, for instance, -scope in galvanoscope, he considers not to be proper suffixes; they are simply terminal elements and excluded from consideration in his book. Nevertheless, Marchand (1969: 4.1.8.1—4.1.8.2) does discuss the creation of last elements such as -athon, -burger, -furter, -rama and calls them suffixes, which seems inconsistent since they often do not have English words or allomorphs of English words as bases.

114

Beatrice

Warren

Hansen et al. (1985) make a simple distinction between proper suffixes on the one hand and semi-suffixes alias suffixoids on the other. Consequently, no distinction is made between, for example, -teria and -monger — both are semi-suffixes alias suffixoids. In parallel fashion, they distinguish between proper prefixes on the one hand and semi-prefixes or prefixoids, alias combining forms, on the other, keeping to the tradition that the term "combining form" should be restricted to the first element. As a conclusion of this survey, let me quote the entry under "combining form" in the Second Barnhart Dictionary of New English. Since this is already in condensed form, I quote it in full: "Combining Forms. This term is usually restricted to forms that occur in COMPOUNDS 1 and DERIVATIVES, such as semi- and -naut; sometimes, of course, the form coincides with a free-standing word, such as graph as contrasted with telegraph·, and even a freestanding word may be used as a combining form, such as -PERSON in place of -man. Many of our combining forms are borrowings from Latin or Greek and are so well established in English that they are freely used to form new words. Even the simplest list of words beginning with AGRI-, bio-, INDUSTRIO-, micro-, astro-, PETRO-, SEXO-, illustrates the range and productivity of such forms. Combining forms are particularly frequent and important in the creation of new TECHNICAL TERMS. Like other linguistic forms, these word components are subject to change and extension of meaning. For instance, the development of astronautics in the 1960's expanded the meanings of the old combining forms of astro-, and cosmo- to include outer space and space travel. More recently, the energy crisis has caused the form PETRO- to take on an extended meaning related to the oil industry, while -ATHON was applied to any prolonged activity resembling a marathon. Similar changes are noted under the entries SYN-, FLEXI-, -ORIENTED, and -WATCHER. Other examples of the use of combining forms taken out of older words, such as -GATE in WATERGATE, are cited in the separate notes on ABSTRACTED FORMS and NONCE WORDS."

The importance

of combining forms

115

3. The corpus of combining forms It is obvious from this survey that there are certain word components which linguists intuitively feel are neither affixes nor roots. It is also obvious that there is as yet no uniform terminology for these elements, nor any generally valid description of them. 2 For example, some linguists insist on their classical origin; others include in the category native morphemes too; some emphasize the ability of these word components to occur in words without proper roots; others ignore this interesting aspect of their character. The present paper is an attempt to improve on the descriptive as well as explanatory adequacy of the account of these affixlike morphemes. To do so, I have excerpted from Mort (1987) words containing elements which on intuitive grounds I considered good candidates for being formatives of this kind. The excerption yielded 81 different types of likely combining forms. These were of course first divided into first and last elements. However, both in the case of initial and final combining forms, my examples naturally divided themselves into two further groups: one consisting of elements which are allomorphic variants of some other word, e.g., astro-, which is a variant of Lat. astrum, and -drome, which is a variant of Gk. dromos; and one consisting of elements which represent parts of other words, e.g., cyber-, which we recognize as a part of the word cybernetics and -aholic, which is not a variant of alcoholic, but a part of it. Among the final combining forms, there was also a third group of forms, viz., elements which from a purely formal point of view are not new morphemes, but which have novel meanings. Gate, for example, is no new morpheme-form, but as a combining form it has a novel meaning, i.e., "political scandal involving a cover-up". Altogether, there are then five subgroups as illustrated in Table 1. For additional examples, see lists in the Appendix. Glancing through the Appendix, we will notice, among other things, that my examples support the claim that initial combining forms end in a vowel, more precisely /i/, /o/, or /a/. (Of these, /o/ and probably /i/ — although my examples happen not to confirm this — can often be looked as linking phonemes.) In other words, initial combining forms differ from final ones in that the former have a characteristic phonetic shape, whereas the latter are amorphous in this respect. We will also notice differences between the groups, in particular between Group I, on the one hand, and Groups II and III, on the other. With a few exceptions, notably debtno- and smello-, both the first and

116

Beatrice Warren

Table 1. The five subgroups of combining forms. Initial

Final combining forms

Group I: Allomorphs of model words (in this case (Lat. astrum and Gr. dromos)

astro- in astrodome "stadium with translucent domed roof'

-drome in alpinodrome "place for climbing contests"

Group II: Truncated forms of model words (in this case cybernetics and alcoholic)

Cyber- in cyberphobia "fear of computers"

-aholic in spendaholic "compulsive spendthrift"

Group III: Parts of model words, which happen to be established morphemeforms

-gate in Yuppiegate "scandal involving five Yuppies"

last elements in Group I are of classical origin and are established combining forms. The majority of the examples in Group I can in fact be looked upon as classical loans which have been phonetically adapted so as to form smooth components of English words. Among the examples of the two Groups II and Group III, however, we find non-classical elements and also novel morphemes, i.e., novel either from a purely semantic point of view, as exemplified by -gate, or from a formal as well as from a semantic point of view, as exemplified by -zak in newzak, "repeated news coverage of the same event" modelled on muzak. It follows that the formation of a certain kind of combining forms is a source of truly novel morphemes, which is something morphologists seem to have overlooked. My main interest in combining forms is the way in which these novel morphemes can be created. My suggestion is that they have been formed by a process which Jespersen (1950: 384 ff) called secretion. 3 What this process involves will be the topic of the next section.

4. Secretion Secretion could be said to involve a rearrangement of semantic elements: from having been associated with no particular part of a word-form, certain elements become connected with a certain section of a word. I

The importance

of combining forms

117

tried to demonstrate the nature of this process by discussing the combining form -(a)holic in the introductory section, -aholic, I claimed, is a content morpheme meaning "person addicted to", which is a meaning we derive from alcoholic, although we do not associate "person addicted to" with the phoneme sequence -holic in alcoholic. Linguists look upon secretion either as the result of some abbreviation (i.e., applied to my example, workaholic would be the result of a contraction of work alcoholic) or as the result of some folk-etymological misdivision of morphemes (i.e., alcoholic would mistakenly be believed to consist of alco + holic). The suggestion that secreted combining forms are the results of contractions of two syntagmatically related words is clearly implausible in some cases. To think of prequel as a contraction of *presequel or newzak as a contraction of *news muzak is simply unreasonable, for instance. In other cases, "the contraction theory" appears more plausible, at least at first sight. Wamography "literature or films glorifying war and violence", for example, could well be a contraction of war pornography. On second thought, however, we will be reminded that -nography in wamography does not mean "films, literature designed to cause undesirable excitement about sex" and we will discover that wamography and war pornography are in fact not semantically equivalent, the literal meaning of war pornography being "pornography issued during wartime". We must conclude that in forming wamography, we have not only meddled with the form of the word pornography, we have also meddled with its meaning. Therefore, to consider -nography as simply an abbreviation of pornography is an oversimplification which is not innocuous. The suggestion that secreted combining forms may be the outcome of uninformed morphemic analyses is an explanation that I accept in some cases, but again not in all. For instance, it is easy to accept the suggestion by Marchand (1969: 4.1.7) that lemonade was mistakenly analysed as lemon meaning "lemon" and -ade meaning "drink", although -ade in fact was no English morpheme at all. I find it considerably less plausible that people could have mistaken alco- for a morpheme meaning "alcohol" and holic for a morpheme meaning "person" or "person addicted to" or that pornography should be taken to consist of the morphemes pormeaning "sex" and -nography meaning "film, literature". Instead I suggest that the coiner of, say, wamography does not mistake -nography for a morpheme, he makes it into one. That is to say, in secreting new morphemes we save so much of a word as we feel is necessary for it (i) to provoke the right associations and (ii) to have the

118

Beatrice

Warren

phonological characteristics of a morpheme suitable to be a word component. We simply discard the rest. Note, however, that the discarded part requires replacement by some established content morpheme which adds some specifying feature of meaning. Warnography, for example, could then be said to have been formed by secreting -nography from pornography provoking the following association "films, literature, etc., designed to cause undesirable excitement about ...". By adding the word war, we get the gloss "films, literature, etc., designed to cause undesirable excitement about warfare". As I have already pointed out, the elements in Group III (i. e., -boom, -fare, -gate, etc., see p. 130 in the Appendix and Table 2) are alike in that from a formal point of view, they are not new morphemes. However, I claimed, as combining forms they express novel meanings. This is a claim that needs to be qualified. It is true of -fare "social benefit", -gate "scandal", -wagon "trend", -wave "programme", but it is not true of -boom "sudden great increase". The correct claim is instead that as a combining form -boom is restricted to a particular sense, viz., the sense it has in the combination baby boom. Similarly, -speak and -meter, for example, have the senses they have in newspeak and hydrometer (see Table 2). In other words, the meanings of these combining forms are not novel but particularized. However, this particularization is derived from, and depends on, a model word. Also, it is my contention that from a semantic point of view, the morphemes in this group are all bound. Table 2. The meanings of the combining forms in Group III. "Suffix"

Meaning-particularization

-boom -fare -gate

"sudden great increase (in) " "social benefits (involving) " "the scandal involving some cover-up (which occurred at or had to do with) " "past or imagined future disaster (involving) " "instrument (for) measuring " "a person (who achieves) " "jargon (occurring in/having to do with) " "sudden popular trend (consisting in) " "computer programme (stored in) "

-mare -meter -smith -speak -wagon -ware

(The dotted line signifies that some complementing semantic feature is obligatory. Brackets surround relational meanings, which, as is shown in Section 6, are in principle variable.)

The importance

of combining forms

119

5. Secretion and abbreviation In Section 3, I mentioned that it is in the two Groups II and in Group III that we find examples of secreted combining forms. It should be pointed out, however, that I do not think of all the combining forms in these groups as secreted. More precisely, I suggest that all the combining forms in Group II of last elements are secreted, whereas — with one possible exception — the combining forms of Group II of first elements are not secreted but simply abbreviated. (The possible exception is cyber-, which expresses the meaning "machine/computer" and may do so through transferance of a semantic element originally associated with no particular section of cybernetics.) Among the elements in Group III, there are two combining forms that I consider to be abbreviated, viz., -wagon and -mare. These will be discussed presently. What then is the difference between secretion and abbreviation? The combining form -burger in cheeseburger and fishburger nicely illustrates the difference. In cheeseburger, -burger is an abbreviated combining form; in fishburger, it is a secreted form. Cheeseburger means "hamburger with cheese"; fishburger does not mean "hamburger with fish", but "fried patty made of fish served in a bun". Secretion is a process in which certain semantic elements in a linguistic unit are kept and others discarded. Abbreviation is a process in which all the semantic elements are kept, although the form of the unit is made shorter. Secretion is like metaphorization in that the interpreter is asked to disregard certain semantic elements of a word or a phrase. It differs from metaphorization in that the form of the word or the phrase is not kept intact. This can be summarized (and, admittedly, oversimplified) as below: FORM

CONTENT

RESULT

Secretion

shortened

shortened

incomplete unit

Abbreviation

shortened

intact

complete unit

Metaphorization

intact

shortened 4augmented

complete unit

Note that linguistic units which have undergone an abbreviation or a metaphorization process remain complete units; linguistic units which have undergone a secretion process become, I maintain, semantically incomplete units. 4 The reason for this is that the discarded semantic elements require overt replacement. Compare in this respect the meta-

120

Beatrice Warren

phorization process. When we use a word in a metaphorical sense, we expect the interpreter to work out — without overt manifestation — not only which features are inapplicable, but also which new features of meaning we wish to express or highlight. Now consider -wagon and -mare. Branwagon is glossed by Mort as "the increasing trend towards eating foods considered healthy (e.g. bran and fresh vegetables)". Wagon in branwagon obviously derives from bandwagon in the phrase jump on the bandwagon "to join the most popular movement without due consideration". It is included among the examples in Group III, since it has a particularized sense, which can only be worked out provided one is familiar with the model word and its occurrence in the idiom cited above. However, since the removal of band from bandwagon does not involve any semantic modification, it must be seen merely as an abbreviated, not as a secreted form. This is confirmed by the fact that bran bandwagon is semantically equivalent to and semantically as felicitous as branwagon. For similar reasons, -mare in nukemare must be considered an abbreviated rather than a secreted form.

6. Interpreting combining-form composites I have suggested above that secretion can be a fairly straightforward exploitation of shared lexical knowledge. That is to say, finding that we need a word for the notion "person addicted to work", which is like the notion "person addicted to alcohol", except for the semantic element "alcohol", we make use of the word for this latter notion by keeping so much of it that the notion "person addicted to" will be provoked and indicate the semantic modification by replacing the discarded element with the morpheme-form for "work". Constructing the meanings of words containing secreted combining forms may, however, require more mental processing than the above suggests. First of all, relational meanings are not necessarily carried over as in workaholic but may have to be constructed afresh. Consider, for example, Radfemspeak "jargon of radical feminists", computerspeak "jargon for computers", and cataloguespeak "jargon occurring in catalogues". Moreover, just as in the case of so many compounds and derivatives, we will also find that the meaning of the whole often depends on some particular extra-linguistic knowledge assumed to be shared and assumed

The importance of combining forms

121

to be prompted by the morphemes in question. Spendaholic, for example, suggests more than simply "person addicted to spending". Mort, realizing, as it were, the close connection between reference and meaning, feels obliged to add to his definition of this word the following explanatory comment: "The behaviour of a spendaholic — is an important social phenomenon. It often results from lack of security. It sometimes afflicts young people and often those who have been deserted by their spouse. It is not restricted to those two typical groups, of course." In order to understand words with secreted combining forms for the first time, we need to (i) recognize and be familiar with the model word — its form, meaning and referents — and (ii) be able to adjust its meaning, which then requires — as I have frequently pointed out — discarding some features of meaning and adding others. We are guided in this by the added content morpheme and usually by our knowledge of the referents of the model word and the new composite. That is to say, of all the facts that we know about the referent(s) of the model word, we choose some as features of meaning. Which ones depends on the nature of the referent of the new composite. In this way, we can explain how the concrete noun gate comes to mean "scandal". We can also explain why, in specifying the meaning of a combining form as part of langue, we often find a set of possible features rather than a stable set of necessary features. Bauer (1983: 272), for example, specifies the meaning of -naut as (adventurer, explorer) (travelling) (especially in space) (in an enclosed space), adding — not quite correctly — that any combination of the parenthesized elements is possible. The implication is that when -naut is part of parole — that is, when it occurs in combination with another morpheme in an actual context — then one particular combination of features will suggest itself. I accept this with some modifications. Firstly, obviously not all contexts are equally effective in singling out one and only one combination. Also, certain types of features invite greater uncertainty as to their appropriateness than others, viz., typically features which are not empirically confirmable. For example, in selecting features of meaning for -zak in newzak, we will, I believe, be more certain that "something recorded" and "transmitted with great frequency" are intended features of meaning than the feature "regrettably". Semanticists tend to give these empirically unconfirmable features of meaning special status by referring to them as connotations. It is my impression that words containing combining forms tend to be particularly rich in connotations. Consider, for example, -speak, which I believe owes its popularity to its negative connotation. Finally, fuzziness of meaning may also

122

Beatrice

Warren

be caused by the fact that often there is more than one model word possible. In principle, as soon as a composite containing a combining form has been established, it may in turn serve as a model word. As an example of this, let us consider -naut again. Originally, it was secreted in the sense "explorer of unknown territories" from argonaut, referring to a sailor in the legendary Argo (see Adams 1973: 185), to form aeronaut, which refers to people travelling in balloons. Aeronaut in turn served as a model word for astronaut, cosmonaut among others, which explains the existence of the features "especially in space" and "in an enclosed space" in Bauer's definition.

7. Root, affix, or neither? The question which will be pursued in this section is: are combining forms a special kind of root or affix or are they in fact neither? Let us first consider similarities and differences between roots and combining forms. Combining forms and roots are both morphemes with lexical content. Roots may be free or bound. Combining forms, I have argued, are invariably bound. It follows that a free morpheme must be a root, since — again by definition — it cannot be an affix either. The difficulty is therefore to distinguish between bound roots and combining forms. Granted that we know which elements are suffixes, this is possible. It follows from the following: a bound root must combine with a suffix to make a word; a single combining form combined with a suffix does not make a word. Consequently, a bound morpheme in word-final position is either a suffix or a combining form, but not a root. A single bound morpheme followed by a suffix is a root. If it is followed by anything else, it is either a prefix or a combining form. In this connection, let us deviate from the topic of the section and compare ordinary compounds and composites containing combining forms. An initial combining form may co-occur with a final one (e.g., neurology) or with a free morpheme (neurophilosophy). A final combining form may consequently be found together with an initial combining form, but also with a free morpheme (feature-itis)5 and very exceptionally with a prefix [prequel). Whatever type of combination, the first element is the modifier and the last the head. 6 This is also true of regular compounds. It may appear then that the only difference between ordinary compounds

The importance

of combining forms

123

and combining-form compounds is that in the former the components are free morphemes, whereas in the latter, first or last or both elements are bound. There is, however, another important difference. Compounding is normally a hyponymy-creating device, i.e., steamship is a hyponym of ship, tablespoon a hyponym of spoon, etc. Composites containing a secreted final combining form are co-hyponyms rather than hyponyms of their heads. Warnography is not a hyponym but a co-hyponym of pornography. Similarly, spendaholic is not a type of alcoholic but a type of addict. Here we witness one of the consequences of secretion. Having established that roots and combining forms are different, let us return to the subject of the section and consider whether it is possible to distinguish between affixes and combining forms. Distinguishing between suffixes and combining forms is normally not problematic. Suffixes may have derivational or inflectional functions or serve to modify the content of the head {kitchenette = "small kitchen"). Final combining forms do not serve any of these functions. They are lexical elements, invariably functioning as heads. Distinguishing between prefixes and initial combining forms, however, may sometimes be problematic, which is evidenced by the fact that not infrequently we find inconsistencies among word-formationists and lexicographers as to which individual morphemes should be considered affixes and which combining forms. For example, Marchand (1969: 3.36) classes neo- as a prefix, whereas Quirk et al. (1985: 1.29) identify it as a combining form or, conversely, Marchand (1969: 3.50.1) classes pseudo- as a combining form, whereas Quirk et al. (1985: 1.23) call it a prefix. Prefixes, like initial combining forms, tend to be modifiers. Both tend to have lexical content. These are important similarities. There are, however, some differences: prefixes do not have to end in /i/, /o/, or /a/, although they may and quite often do. More importantly, prefixes, like all affixes, have, or have had, productive force. Initial combining forms, like all combining forms, need not have productive force. Indeed, they may be nonce formations. Expressed differently, prefixes are modifiers of a well-established kind, belonging to a more or less closed set. Combining forms are open-set items. They are lexical in nature and may have rather specific meanings. Finally, initial combining forms, like all combining forms and unlike affixes, can be connected with a "model word". These last two differences between affixes and combining forms inspire the following hypothesis: As a rule, affixes were originally independent words — hence no allomorphic status — with some auxiliary function in a syntagm. This

124

Beatrice

Warren

auxiliary function, if it turned out to be repeatedly useful, triggered off a grammaticalization process, involving phonetic as well as semantic erosion to different degrees — hence the once free morphemes become bound morphemes with productive force and often with little lexical content. The auxiliary functions that these elements served would be derivational, inflectional, or to modify the lexical content of some head. In the former two cases, semantic erosion would follow naturally; in the latter case complete semantic erosion would be prevented by the fact that an element which serves to modify the meaning of some other element must carry some meaning; it cannot be purely functional. Combining forms, on the other hand, as a rule, started off as incomplete variants or bits of words — hence their allomorphic status. These elements had to be combined with some other unit to make an independent unit. In other words, combining forms were originally bound, lexical elements, better equipped to resist grammaticalization. However, they would not be immune to this process. Hence we find combining forms, particularly among initial combining forms, which have developed features characteristic of affixes. Pseudo- and neo- may serve as examples. These are productive, bound, initial modifiers with prefix-type meanings. Therefore, from a synchronic point of view, it will seem unjustified to class, say, semi- and pre- as prefixes, but pseudo- and neo- as combining forms. My position then is that combining forms are different from affixes in that they are open-set dictionary elements, whereas affixes are closedset elements that have undergone a grammaticalization process. Since, however, combining forms are not precluded from undergoing this process, we may come across morphemes, which from a diachronic point of view are combining forms, but from a synchronic point of view are affixes. How such morphemes should be classed is necessarily arbitrary. Summing up this section, combining forms are neither roots, nor affixes. Nevertheless, in individual cases, it may not be possible to decide unequivocally whether a particular morpheme should be considered an affix or a combining form.

8. Summary The present study reveals that there are two main ways in which socalled combining forms may be formed: by phonetic modification of some existing morpheme or by secretion. It is possible to distinguish between

The importance

of combining forms

125

two types of phonetic modification: minor ones such as we witness in alpino- (from alpinus), circo- (from circum), smello- (from smelt) and -scope (from scopium) and proper clippings as exemplified by eco- (from ecology) and -wagon (from bandwagon). Secretion is not simply an abbreviation process but is seen as a process which makes it possible to create new morphemes (see examples among the last elements in Group II in the Appendix), or new meanings for established morpheme-forms or to transfer certain meaning-particularizations (see examples in Group III: Table 2). Note that secreted forms tend to be final elements. That is to say, it appears that secretion serves above all as a co-hyponym-creating word-formation device. The examples also show that it is not justified to apply the term "combining form" to morphemes of classical origin only, although among non-secreted elements, classical morphemes predominate. Finally, to restrict the term "combining form" only to first elements seems unwarranted, although, admittedly, there are definite differences between first and final combining forms. Initial combining forms are modifiers, final combining forms are heads. Initial combining forms have a characteristic phonetic "shape", ending in a vowel; as just noted, they tend not to be secreted. Final combining forms are phonetically amorphous and are either abbreviated or secreted. In most cases, it is easy to identify combining forms. There are, however, morphemes which may qualify as combining forms and affixes equally well. This happens when a morpheme formed in the manner characteristic of combining forms develops grammatical features. The most important difference between combining forms and affixes is the fact that combining forms are open-set items which may be formed on the spur of the moment. The products of compounding and affixation are novel combinations of existing morphemes; the products of conversion are semantic modifications of existing morphemes; the products of clipping, acronym-formation, and blending are abbreviations, i.e., phonetic modifications, of existing morphemes. In forming composites containing combining forms, we are not restricted to existing morpheme-forms. The accounts of word formation I am familiar with neglect to make clear this crucial difference between the process of forming composites involving combining forms and other word formation processes. The real reason for my interest in combining forms is, however, that once again we find support for the idea steadily gaining ground among semanticists, viz., knowing what other people know and knowing how they are likely to process this knowledge are essential parts of meaning formation, which of course is an essential part of word formation.

126

Beatrice Warren

Appendix

List A: First elements "Prefix"

Containing noun

Model word

alpinodrome "place for climbing contests" (balloon) angioplasty "technique for treating angina" elect roan tennogram "chart of electricity in antennae" aquatube "water slide" astrodome "stadium with translucent domed roof' biocomputer, biohazardous, bioholonics, biomotor et al. chronobiology et al. "the study of biological rhythm in the human body" circotherm( oven ) "oven with fan circulating heat" debtnocrat "banker dealing with international debts" electroantennogram, et al. "tracing of electricity in antennae" palaeoe thnobo tany gigadisc, gigaflops "disc capable of storing large amounts of data" heterosexism, (heterosexist) "discrimination agabinst homosexual«" alS iatroblast "embryonic doctor" iridology, (iridologist) "technique of diagnosis involving iris" lithotripter "device for pulverizing kidney stones" mega-blockbuster, megabrand, megabuck, mega-city et al.

Lat. alpinus

Group I alpinoangioantennolaquaastro-

biochrono-

circodebtno-

electroethnoIgiga-

hetero-

iatroiridolithomega- "great"

Gk. angeion ?E antenna Lat. aqua Gk./Lat. astrum

Gk. bios Gk. chronos

Lat. circum Ε debt

Lat. electricus Gk. ethnos ?Gk. Gigas

Gk. heteros

Gk. iatros Gk. iris Gk. lithos Gk. megas

The importance of combining forms

'Prefix'

Containing noun

Model word

mega- "million'

megaflops "a million flops" microburst, microfloppy, microgravity multitask, multi-user, multi-processor neurophilosophy et al. "philosophy of the mind-brain" optoelectronic(s) "electronics involving interaction of light and electronic signals" palaeoethnobotany, palaeotopomorphologist petrocurrency "currency of nation depending on oil production" phytolith, phytotherapist "minerals in plants", "therapist using plant extracts" polymyositis "inflammation of striated muscle" psychographic (s), psychoneuroimmunology radioglaciology "the study of glaciers by radar" smellometer "instrument for measuring body odor" videosomatography "technique involving video pictures for improving man-machine interface' sonochemistry "chemistry using ultrasonics" spiroplasma "spirally shaped micro-organisms" sulphoxidizer "person who can break down sulphur" technobattle, technofear, technofreak, technophobia et al. telebanking, telecommute, teleshopping palaeotopomorphology "study of land to discover traces of primitive cultures" ureteroscope "instrument for viewing urinary tracts"

Gk. megas

micromult ineuroopto-

palaeopetro-

phyto-

polypsychoradiosmello-

Isomato-

sonospirosulph(o)-

technoteletopo-

uretero-

127

Gk. mikros Lat. multus Gk. neuron Gk. optikos

Gk. palaios E/Lat./Gk. petroleum

Gk. phyton

Gk. polys Gk. psyche Lat. radius Ε smell Gk. soma

Lat. sonus Lat. spira E/Lat. sulphur

Gk. techne Gk. telos Gk. topos

Gk./Lat. ureter

128

Beatrice

Warren

"Prefix"

Containing noun

Model word

Iviru-

virustasis "stopping of growth of viruses"

?Lat. virus

cyberphobia "fear of computers" ecosocialism "socialism concerned with ecology" econospeak "the jargon of economists" Eurofighter, Eurο feebleness, Eurosclerosis, Eurowimp et al. pluro-communism "communism and pluralism combined" vegeboom "sudden popularity of vegetarianism"

cybernetics

Group II cyberecoeconoEuroIpluro-

vege-

ecology economics Europe(o) pluralism

vegetarianism

List B: Last elements "Suffix"

Containing noun

Model word

iatroblast "embryonic doctor" yobbocracy "rule by louts and thugs" debtnocrat See debtnoalpinodrome "place for climbing contests" lumpectomy "surgical removal of cancer" electroantennogram et al. "chart of electricity in antennae" psychographic (s) "market research of attitudes" videosomatography See somato-

Gk. blastos

Group I -blast -cracy -erat -drome -ectomy -gram -graphic -graphy

?aristocracy ?bureaucrat Gk. dromos Gk. -ektomia Gk. gramma Gk. graphos Gk. graphia

The importance of combining forms

"Suffix'

Containing noun

-lith

phytolith Gk. lithos "mineral particle from plants" biologue Gk. logos "biographical radio programme" chronobiology et al. Gk. -logia "study of biological rhythm in the human body" mergermania Gk./Lat. mania "craze for mergers" cyber phobia et al. Gk./Lat. phobia "fear of computers" angioplasty Gk. -plastia See angioureteroscope Lat. scopium "instrument for viewing urinary tracts" virustasis Gk. stasis See viru-

-logue -logy

-mania -phobia -plasty -scope

-stasis

Model word

Group II -(a)holic -(a) thon -erati

-gram

-itis

-nik

-nography

-tro -rrhoea

spendaholic "compulsive spendthrift" bikeathon, duckathon, pedalathon, readathon, swimathon slopperati, glitterati "the deliberately untidy rich"; "jetsetters", respectively Tarzangram, potatogram "greeting delivered by costumed person"; "message on a potato" feature-itis "excessive occurrence of feature articles" Wappnik "journalist refusing to cross picket line in Wapping to go to work" warnography "literature or films glorifying war and violence" outro "ending of song or musical number" processorrhoea "excessive flow of words induced by word processor"

alcoholic marathon lliterati

telegram

Ίarthritis, etc.

Irefusenik

pornography

intro diarrhoea

129

130

Beatrice Warren

"Suffix"

Containing noun

Model word

-quel

prequel "the story of the events leading up to those in an existing work" newzak "repeated news coverage of the same events"

sequel

-zak

muzak "soft music played continuously in stores, etc."

Group III

-boom -fare

-gate

-mare

-meter -smith -speak Ί-wagon

-ware

vegeboom "sudden popularity of vegetarianism" workfare "social benefits in return for unpaid work" Yuppiegate, Westlandgate "scandal involving five Yuppies, Westland helicopters", resp. nukemare "fear of disaster induced by nuclear materials" magnometer, smellometer "instrument measuring yang and body odour", resp. leaksmith "person leaking information" cataloguespeak, computer speak, econospeak etc. branwagon "trend to eat health foods, such as bran" firmware "computer programme stored in a read only memory"

baby boom welfare

Watergate

nightmare

barometer, etc.

?blacksmith, etc. newspeak band wagon

software

Comments (i) In excerpting examples, problems of inclusion always arise. Admittedly, some of the examples in the Appendix are not indisputable examples of combining forms. These have been marked with question marks. Similarly, some composites which rightly should have been included may have been excluded. Only combining forms part of novel composites have been considered; that is to say, from, for

The importance of combining forms

131

example, palaeotopomorphologists, which consists of four combining forms, morpho- and -logist have not been extracted, being parts of an established composite. (ii) For reasons of space, the glosses are brief. The interested reader is referred to Mort (1987), who supplies full explanations. (iii) Linguists normally think of -o- and -i- as linking vowels added to the initial combining form, whereas -a- in -(a)holic, for instance, is considered a linking phoneme added to the final combining form. Although I cannot quite justify this practice, I have adhered to it here.

Notes 1. Capitalization signifies that the particular word or form has an entry of its own in the dictionary. 2. I am not the first to have noticed inconsistencies in the treatment of combining forms. See Stein (1977 and 1984). 3. Barnhart et al. (1980), however, use the phrase "abstracted form" for secreted elements. 4. The fact that combining forms are bound morphemes does not prevent them from occasionally being converted into proper nouns. Consider, for example, itis, which can be used as a noun in the sense "bodily condition or disease". 5. Bauer's claim that final combining forms are restricted to the company of initial combining forms is not correct. 6. There is an exception to this rule, viz., the so-called dvandva compounds exemplified by Franco-German, Anglo-Polish, Sino-Italian. These compounds consist of two heads rather than a modifier and a head.

References Adams, Valerie 1973 An introduction to Modern English word-formation. (London: Longman). Bauer, Laurie 1983 English word-formation. (Cambridge: Cambridge University Press). Barnhart, Clarence L., et al. 1980 The second Barnhart dictionary of New English. (Bronxville, Ν. Y.: Barnhart Books). Hansen, Barbara — Klaus Hansen —Albrecht Neubert — Manfred Schentke 1985 Englische Lexikologie. (Leipzig: Enzyklopädie). Jespersen, Otto 1950 Language — its nature, development and origin. (London: Allen & Unwin). Marchand, Hans 1969 The categories and types of present-day English word-formation. (München: Beck). Mort, S. (ed.) 1987 Longman Guardian new words. (Harlow: Longman). Quirk, Randolph, et al. 1985 A comprehensive grammar of the English language. (London/New York: Longman).

132

Beatrice Warren

Stein, Gabriele 1977 "English combining forms", Linguistica 9 (= Tartu riikliku ülikooli toimetised 437), 140-147. 1984 "Word-formation in Dr. Johnson's dictionary of the English language", Journal of the Dictionary Society of North America 6: 35 — 112. Warren, Beatrice 1988 "Ambiguity and vagueness in adjectives", Studia Linguistica 42: 122 — 177.

Compounding and inflection Wiecher Zwanenburg

0. The general idea behind this paper is the hypothesis that in the unmarked case languages organize their word structure hierarchically in such a way that inflection is peripheral to compounding. I want to discuss here the apparently exceptional relationship between compounding and inflection in the Romance languages, and more particularly in French. In French, we find cases where inflection seems to be inside compounding, as in (la), next to the expected cases where inflection is peripheral, as in (lb): (la) b)

des secretaires-generaux 'secretaries-general' des basses-cours 'poultry yards' des en-tetes 'letter headings' des apres-midi{s) 'afternoons'

I will argue that the cases of (la) are only apparent exceptions to the above-mentioned generalization. Before discussing this point in Section 2, I will consider in Section 1 the possibility to distinguish in a satisfactory way between derivational and inflectional affixation. This possibility, questioned by quite a number of linguists, is of course a necessary condition for the above-mentioned generalization to make sense. 1. As to derivational and inflectional affixation, one of those who have recently argued that it is doubtful whether one can distinguish the two in a principled way is Selkirk (1982: 69 — 77). She argues convincingly that the two types of affixation must be accounted for by the same morphological (sub)module. They obey quite similar or maybe identical principles of morphological X-bar structure and show quite similar and in many cases identical phonological behavior. But this does not mean that there is not a principled distinction between the two types of affixation with respect to syntax. Anderson (1982) proposes the following criterion, which seems to me to be adequate:

134 (2)

Wiecher Zwanenburg

Inflectional morphology is what is relevant to the syntax, with the exclusion of properties of lexical insertion per se and concomitant principles of subcategorization etc.

Anderson concludes from this that derivational and inflectional affixation must be treated in different modules, derivation in the lexicon before lexical insertion at S-structure, and inflection in an extended phonological module after lexical insertion. But this runs counter to Selkirk's important insight that there must be some (sub)module where it is possible to generalize over derivational and inflectional affixation as to morphological X-bar structure and phonological behavior. In my view, one can do justice to Anderson's insights within the general framework of Selkirk's model. That is, both derivational and inflectional affixation are accounted for in one and the same morphological module within the lexicon by comparable and in many cases identical rules. At S-structure, the necessary syntactic information is available for lexical insertion of the desired inflectional form of a word, so there is no need to have an extended phonological module take care of inflection. But within the morphology, the affixation rules operate at least at two different levels, first for derivation at a syntax-independent level, and second for inflection at a level dependent on syntactic information. The distinction between the two levels must be stipulated within morphology. According to Di Sciullo — Williams (1987: 25) this distinction need not be stipulated, but follows from the position of heads in words. For an inflectional affix to communicate by percolation its syntactically relevant information to the word, it must be in head position, that is in the rightmost position, and thus outside derivational affixes. This excludes a form like *nounshood instead of nounhoods. But then Di Sciullo-Williams (1987: 2 5 - 2 7 ) relativize the notion head, defining per feature F the head of a word with respect to F as the rightmost element of that word marked for F. This accounts for cases like the French diminutive suffix -ot, forming nouns from nouns, adjectives from adjectives, and verbs from verbs respectively (frer-ot, pal-ot, and touss-ot-er): -ot is unmarked for the feature category, and thus the respective bases are heads with respect to category. But now *nounshood is not excluded any more. In order to save their analysis, Di Sciullo —Williams (1987: 27 — 28) assume a principle saying that if a feature is defined for a category, then all members of that category are marked for that feature. Thus noun in nounhoods will be marked for (singular) number and the plural affix must

Compounding

and inflection

135

be in rightmost position, as in (3a). The same holds for a compound like choir boys in (3b). But Di Sciullo — Williams' principle is unable to explain the cases where inflection is outside a derivational affix unmarked for category. This is what we find for example in French touss-ot-ait in (3c). We have seen that -ot is unmarked for category. Thus nothing would predict the agrammaticality of *touss-ait-ot. (3a) b)

c)

nounhoods j *nounshood choir boys / *choirs boy

touss-ot-ait 'coughed dryly', literally 'cough-little-ed' / *touss-&\\-ot

So I feel that the hierarchical relation between derivation and inflection should be stipulated. And this should be done along the lines suggested by Anderson, but within Selkirk's general framework.

2. Having thus established the possibility and the necessity to distinguish between derivation and inflection, let us look more closely into the apparently problematic relationship between compounding and inflection in French. My examples will all concern nominal number inflection. They are from written French, including where possible not so frequent cases where orthographic differences between singular and plural are parallelled by differences in pronunciation. Consider first the following cases, all problematic like those of (la): (4a)

Ν

—*•

NN:

b)

Ν —• N N :

c)

Ν —• N A :

d)

Ν —• A N :

e)

A —* A A:

oiseaux-mouches 'humming-birds' amiraux-poetes 'admirals who are poets' timbres-poste 'postage stamps' (Eufs-mayonnaise 'eggs with mayonnaise' coffres-forts 'safes' secritaires-generaux 'secretaries-general' basses-cours 'poultry yards' beaux-arts 'fine arts' sourds-muets 'deaf-mute (pi)' verbaux-nominaux 'verbal-nominal (pi)'

In all these cases, the plural affix is on the two words forming the compound or on the first one only. This is contrary to the hypothesized generalization, which would predict it to be on the entire compound only. Now, Di Sciullo-Williams (1987: 7 8 - 8 8 ) argue that examples like the ones in (4) are syntactic expressions. Insofar as they are idiomatic,

136

Wie eher Zwanenburg

they will be listed in the lexicon and as such be "listemes". Better-known examples of such expressions are the much-discussed idiomatic verbal expressions of the type to take to task or French mettre ä profit 'to exploit'. If this interpretation is correct, the cases of (4) are not problematic at all for the generalization mentioned in the beginning. We are dealing with phrases instead of compounds, and it is only to be expected that inflection is marked on the words of these phrases. This means that not only does Di Sciullo — Williams' analysis help to explain our data concerning the position of inflectional affixes, but at the same time our data provide independent evidence for their analysis. Now consider the other kind of French compounds, illustrated in (lb). In fact we can distinguish two types, as in (5a) and (5b). We have seen that they are in principle unproblematic for our generalization, because inflection, if present at all, is on the entire word and not on one of its parts, and thus it is outside compounding: (5a) b)

Ν —> V N: des essuie-glace 'screen wipers', literally 'wipe-ice' des couvre-chefs 'head-dresses' Ν —*• PP: des en-tetes 'letter headings', literally 'at-heads' des apres-midi(s) 'afternoons'

This would conclude my argument, but for the fact that compounds like those in (5) confront us with problems which are indirectly linked to the topic of this paper. So let us examine the cases of (5a) and (5b) in that order before concluding. Di Sciullo —Williams argue that the type of (5a) allows for other complements than nouns, as illustrated in (6). On the basis of these examples they argue that this type must in fact be analyzed as in (7). (6a) b) c) d)

N - -> V N : N - ^VA: N - -> V Adv: N - -•VP:

(7)

N - -> VP

essuie-glace 'screen wiper' gagne-petit 'poorly paid person' couche-tard 'night-reveller' frappe-devant 'sledge hammer'

This leads them to admit more generally, for the periphery of the grammar, nonmorphological word-creating rules reanalyzing phrases such as VP as words. Unlike the cases of (4) and like ordinary words created by morphological rules, syntax may only refer to these words as a whole and cannot "look into them".

Compounding

and inflection

137

Di Sciullo —Williams then argue that in fact there are more possibilities for this type of words, as shown in (8). And they conclude that the right generalization is as in (9). (8a) b) c) d) e)

des des des des des

trompe-l'oeil 'trompe l'ceil paintings' boit-sans-soif 'boozers' bons-ä-rien 'good-for-nothings' hommes-de-paille 'straw-men' hors-la-loi 'outlaws'

(9)

Ν —• XP

Now, this would constitute a problem for our generalization: in (8c) and (8d), unlike the other cases, inflection is not on the entire compoundlike word but on a part of it only, contrary to our generalization. But it is easy to see that precisely this fact allows us to interpret the cases of (8c) and (8d) as syntactic expressions with idiosyncratic meaning or listemes, like those in (4). Thus, such expressions turn out to be more widespread than Di Sciullo — Williams suggest. In this way we are led to maintain rule (7) instead of (9), but generalized over nouns and adjectives to account for adjectival cases like (10). The rule will then take the form of (11). (10)

un enfant brise-tout 'a child that breaks everything'

(11)

A/N —> VP

This will take care of all of Di Sciullo — Williams' examples except (8e). Now, it turns out that (8e) represents the second case of (5) to be discussed, that of Ν —> PP in (5b). Consequently, we need a second nonmorphological word-creating rule of the form Ν —> PP, but this rule can be generalized over nouns and verbs, as the examples in (12) show. I will discuss elsewhere in more detail the generality of the cases of (12) and their prepositional character, as well as the occurrence of PP in adjectives. The rule will have provisionally the form of (13). (12a) b)

N: V:

en-tete en-chain-(er)

(13)

N/V —> PP

'to enchain'

This leaves us with a question for future research concerning the status of nonmorphological word-structure created by rules such as (11) and

138

Wiecher Zwanenburg

(13). The examples given so far suggest that its hierarchical relationship with inflection is the same as that of compound rules: inflection is outside. But then Di Sciullo — Williams do not discuss the not infrequent case of VP nouns where VP contains a plural object, like casse-noisettes 'nutcracker', which have the same form in the plural. This fact does not impair the reasoning above about the ungrammaticality of *nounshood and * choirs boy as plural forms of nounhood and choir boy. The question to be examined is to what extent syntax-like nonmorphological word structure admits inflected elements. And more generally, what are the restrictions on such a structure other than nonreferentiality discussed by Di Sciullo —Williams. 3. I conclude that the French facts do not force us to abandon the hypothesis that in the unmarked case languages organize their word structure hierarchically in such a way that inflection is peripheral to compounding. In order to reach this conclusion I have followed Di Sciullo — Williams (1987) in analyzing French compounds partly as listed phrases, partly as cases of marked nonmorphological word structure. My treatment has given independent evidence for this analysis. I have argued against Di Sciullo — Williams that the distinction between derivational and inflectional affixation must be stipulated. Moreover I have argued that their rule (9) Ν —• XP must be replaced by (11) N/A —• VP and (13) N/V —• PP.

References Anderson, Stephen R. 1982 "Where's morphology?", Linguistic Inquiry 13: 571 —612. Di Sciullo, Anna-Maria —Edwin Williams 1987 On the definition of word. (Cambridge, MA: MIT Press). Selkirk, Elizabeth O. 1982 The syntax of words. (Cambridge, MA: MIT Press).

Topic 3: Inflectional morphology and clitics

Arguments against the passive as a universal morphological category Paul Kent Andersen

The purpose of this paper is to argue that what is generally regarded as a "passive" in the literature does not in fact constitute a single, universal/prototypical grammatical category, but on the contrary, that there is evidence for the existence of several, universal/prototypical categories that should replace the single category "passive". Due to the limitations imposed on this paper, the following discussion will center around only one of the many aspects of the "passive", namely "passive" morphology. Accordingly, the basic thesis that will be defended here is that the varieties of verbal morphemes employed in constructions generally regarded as "passives" in the literature are in fact instances of a number of distinct, prototypical, morphological categories. Consequently, there is preliminary evidence that "passive" constructions employing these respective morphological categories are themselves instances of distinct, prototypical constructions.

1. Introduction For quite some time now it has been recognized that constructions generally regarded as "passives" differ in morphological, syntactic, semantic, and even discourse features; in fact, there is very little that is shared by all "passive" constructions, cf., among others, the conclusions reached by Siewierska (1984: 259). Thus we are faced with the problem that there is no single, generally accepted definition of the "passive" that does justice to all instances of "passives" attested in the literature. This is particularly evident in the most recent collection of papers concerning the "passive" and Voice edited by Shibatani (1988): here we find that the definitional properties of the "passive" are regarded by some authors, e.g., Rude and Cooreman, as discourse features, by others as syntactic

142

Paul Kent

Andersen

properties, e.g., Siewierska and Kimenyi, and by others still, e.g., Dezsö, as the "passive" morpheme. Of course, one of the easier ways out of this problem is to define the "passive" in a rigid manner and then to claim that instances not covered by the definition are not "passives". Since, moreover, "theoretical consideration of passive constructions has naturally focused on English" (Langacker — Munro 1975: 789), it therefore comes as no surprise to find that the English "passive" generally ends up either as the expression par excellence of "the passive prototype" — cf., e.g., We shall refer to passives like (lb), John was slapped, as 'basic passives' Our justification for calling such passives 'basic' is that they are the most widespread across the world's languages (Keenan 1985: 247)1 and The characterization in 40 applies to prototypical passives like English Many soldiers were killed, or its Japanese equivalent Takusan no heitai ga koros-are-ta (Shibatani 1985: 837)2 or positioned at one extreme along "the passive continuum", cf. If the correlations (6), (7), and (8) above indeed hold, then one excepts one major continuum in the typology of passivization, with the English type representing one extreme of the continuum... . The Ute extreme of the scale would have just the opposite properties (Givon 1981: 171). Had theoretical considerations been focused on a language other than English, we may have found that an entirely different prototypical definition of the "passive" would have been proposed and that the English "passive" would have been regarded as an "alternative", 3 "pseudo-passive", "peculiar passive", or the like. In the present paper an alternative approach to the problems of the "passive" will be taken, i.e., instead of looking for one single "passive", the possibility of the existence of a number of distinct "passives" will be investigated.

2. Passive morphology: preliminary remarks The point of departure for the following morphological discussion will be the questioning of the following fundamental axiom underlying most discussions of passive syntax and morphology. "Passive constructions"

Arguments

against the passive

143

are defined in terms of specific characteristics (varying from author to author) including the stipulation that the verb/predicate contain a "passive morpheme", 4 whereas when it comes to concrete examples of "passive constructions" the relevant verbal morphology — i.e., morpheme — is isolated and invariably glossed as PASSIVE. In other words, a construction is a "passive" because it contains (among others) a "passive" morpheme and a morpheme is a "passive" because it is employed in a "passive" construction. That this axiom is blatantly circular appears to pose no serious problems for the majority of authors. 5 The question that we will now pursue is whether there is any empirical evidence for recognizing a specific morpheme as a "passive" other than the fact that it is employed in a "passive" construction. Even if we would assume that specific verbal morphology is a definitional characteristic of "passive" constructions, does this automatically mean then that all instances of these specific verbal morphemes are necessarily representatives of one and the same prototypical morphological category? It is now my contention that the recent typological studies of verbal morphology by Bybee (1985) and Dahl (1985) will offer us a working framework in which to answer these questions.

3. Verbal categories Bybee (1985) and Dahl (1985) have each devoted an entire monograph to the investigation of verbal categories based upon two independent samples of languages especially suited for typological research. Although the results that were reached do differ in various aspects, they both demonstrate the existence of a number of prototypical verbal categories. But because they both concentrate on Tense, Aspect and Mood, they unfortunately have very little to say about "passive" Voice; both do, nevertheless, agree that such a morphological category exists. So, for example, according to her principles of relevance6 and generality,7 Bybee establishes the following hierarchical order of verbal morphemes relevant to the verb root: VERB-Valence-Voice-Aspect-Tense-Mood-number/ person/gender Agreement. 8 Since the "passive" is regarded as an instance of the category Voice, we see that within this framework the "passive" not only has a specific place in this hierarchy of verbal categories, it also has a specific function (in terms of relevance and generality) that distin-

144

Paul Kent

Andersen

guishes it from the other categories. In this respect it is worth noting the definitions of the categories for Valence, Voice, and Aspect employed by Bybee in her investigation: Valence refers to differences in the number or role of arguments that the verb stem can take. Voice indicates the perspective from which the situation described by the verb stem is viewed. Aspect refers to the way the internal temporal constituency of the situation is viewed (Bybee 1985: 28).9 Thus we have reason to believe that should there be a single, universal morphological category for "passive voice", it will exhibit not only a specific place along the hierarchy of verbal categories, but also a specific function with regards to Bybee's principles of relevance and generality. Dahl, on the other hand, gives the following reason why he has chosen not to investigate the "passive" with his method based upon a comprehensive questionnaire: In addition to the combinatorial explosion, there are additional reasons why it is hard to study e.g. passive constructions with a methodology of this type. In some languages, passives may be very marginal or even not occur at all; other languages may have several constructions that could be labelled passive, with more or less subtle differences in conditions of use. This means that by simply giving an English sentence in the passive you cannot guarantee that what comes out in the translation is the passive construction you are looking for, if it is a passive at all (Dahl 1985: 47). In the following investigation we will assume that it is indeed possible to classify the distinctive verbal morphology employed in constructions generally regarded as "passives" according to the principles established by Bybee and Dahl without recourse to the circular axiom mentioned above. Our first indication that the morphemes employed in "passive" constructions are not all instances of one and the same verbal category can be seen from the fact that morphemes isolated and glossed as "passives" generally perform functions other than that of appearing in "passive constructions", i.e., "passive" morphemes are generally employed in constructions other than the "passive". 10 A closer look at these other constructions will reveal the fact that they are themselves not identical, but rather that they do define certain independent sets of construction types. Thus it is not the case that all of the other functions of the "passive" are the result of the mere extension of a single, basic "passive" meaning in all of these instances.

Arguments

against the passive

145

4. Agreement The first class of morphemes to be discussed is the set of "middle" inflectional endings of Classical Greek and other Indo-European languages. The reason for starting with the "middle" should be rather obvious: our modern term "passive" derives via Latin from the Greek grammatical tradition, in which the term pathos (which was translated into Latin as passivum) designates the morphological set of personal endings generally referred to now as the "middle" in contrast to the corresponding "active" ones.11 So, for example in (la) below from Classical Greek the "middle" form of the verb is employed (among others) in "passive" constructions (TV represents the thematic vowel) and contrasts with the corresponding "active" form in (lb): (la)

b)

lou-o-metha wash-TV-1 PL/MIDDLE 'We are being w a s h e d 2 Ιού-ο-men wash-TV-1 PL/ACTIVE 'We are washing.'

This now brings us to the actual morphological category employed in (la). First of all, the categories generally referred to as "active" and "middle" cannot be separated from the categories for person and number, thus we have evidence that the categories for "active" and "middle" are expressions of the category Agreement. These morphemes for Agreement (along with the TV) are furthermore placed after the morphemes for the categories Tense, Aspect, and Mood, 13 thus giving evidence that the morphological categories expressed by the "active" and "middle" Agreement markers are not instances of the category "Voice" within Bybee's framework. Second of all, we can determine that a number of other construction types also employ the same "middle" inflectional endings, cf., e.g., Barber (1975) and Haspelmath (1987). These include the reflexive, reciprocal, anticausative, numerous so-called "middle constructions" (see the passage from Klaiman quoted below) and the "passive". 14 From this list of construction types we can furthermore determine that the common functional denominator of all of these constructions is the following:15 "The implications of the middle (when it is in opposition with the active) are that the 'action' or 'state' affects the subject of the verb or his interests" (Lyons 1968: 373). So, for example, Klaiman (1982: 408) gives the following functions of the category "middle":

146

Paul Kent

Andersen

a) Plain Middle: results of action accrue to Subject b) Reciprocal Middle: referents of plural Subject do action to one another c) Reflexive Middle: Subject performs action to self d) Deponent Middle: action denotes physical/mental disposition of Subject e) Nucleonic Middle: object of action belongs to, moves into, or moves from sphere of Subject f) Plain Passive: Subject does nothing, is affected in consequence of action By characterizing the function of the "middle" inflectional category in this way we can now see that this morphological category is not as relevant to the verb as the categories Valence, Aspect, Tense, and Mood are: the difference between the "active" and "middle" Agreement categories here does not directly affect or modify the semantic content of the verb. Moreover, the categories for "active" and/or "middle" are necessarily applicable to all verb stems (there are no finite verb forms not conjugated for one or the other category) and thus — according to Bybee (1985: 16 f.) — the semantic function of these categories must be rather general and have only minimal semantic content. Thus it comes as no surprise to find that they are placed farther away from the verb root than the other categories. This now brings us to the semantic characterization of the "passive" constructions that employ the "middle" inflectional endings as the distinctive, morphological feature. Since the "middle" inflectional endings are expressions for the category Agreement, we can determine that the fundamental function is referential in nature, cf.: My basic thesis is that agreement is referential in nature. It helps identify or reidentify referents. It does this by giving information on grammatical properties of its referent and, thus, of the NP representing it if one is around. The functions of agreement in the marking of syntactic relations derive from this primary function (Lehmann 1988: 55). Recall, furthermore, that according to Jakobson (1971: 134) "PERSON characterizes the participants of the narrated event with reference to the participants of the speech event". If, however, we assume that there is not only a speech event and narrated event, but also a narrative discourse — as distinguished, e.g., in Dahl (1985: 112). "I define a narrative

Arguments

against the passive

147

discourse as one where the speaker relates a series of real or fictive events in the order they are supposed to have taken place" — we will be in the position to better characterize the actual function of the "middle" morpheme in this type of "passive" construction: one participant of the narrated event — i.e., the "Actor" — does not participate in the narrative discourse (/speech act) within its respective narrated context; in the corresponding "active" the "Actor" participates not only in the narrated event, but also in the narrated discourse (/speech act). Thus we see that the narrated event is not altered at all in this type of "passive" construction. To sum up, in Classical Greek and other languages the specific morphological means employed in the "passive" can be identified as an inflectional category for person/number "Agreement". The basic/ prototypical semantic function of the "middle" is furthermore to indicate that the "subject" is "affected" by the verbal action. The specific function of this morphological category when employed in the "passive" can be characterized as indicating that the "Actor" participates in the narrative event without, however, participating in the narrative discourse (/speech act). Since reflexive affixes in numerous languages throughout the world exhibit almost 16 the very same functions as the "middle", there is some indication that they may also represent the same prototypical verbal category. 17

5. Perfect Let us now turn to the type of "passive" found in languages like English, cf., e.g.: (2)

John was seen (by Mary).

Our very first problem will be to isolate the specific morpheme employed in this construction that distinguishes it from a corresponding "active" construction. Unfortunately, it is a general consensus among various linguists (especially those working within formal frameworks) that the distinctive morphology in "passives" such as (2) above is the so-called "passive participle" — or more accurately, the suffix -en which forms the "passive participle". 18 Nevertheless, it should be rather obvious that this form of the verb is not in any way restricted to "passive" constructions, cf.:

148

Paul Kent

Andersen

(3)

John has seen Mary.

(4)

The glass is unfortunately broken. 19

What is morphologically distinctive of this type of "passive" construction is the combination of a copula (or other auxiliary) and the "past participle", i.e., a periphrastic construction. Having isolated the specific morphological form, we can now attempt to specify the morphological category expressed by this periphrastic construction. In a recent investigation of this type of "passive" in English, German, and Russian, Beedham (1982) has come to the conclusion that the "passive" can be characterized as a construction describing "an event leading to a state", cf.: It is postulated ... that passives mean the portrayal of a state as the result of a preceding action... . It is the contention of the present work that the passive is an aspect, like in English the perfect and the progressive... . The very similar restrictions of co-occurrence that the passive and the resultative perfect display suggest that they are semantically similar, thus confirming that the passive describes an event leading to a state (Beedham 1982: 147). With respect to verbal categories it therefore appears that this type of "passive" represents the category "Perfect" or some closely related category (or variant thereof)· 20 Although Bybee (1985: 159 ff.) considers the "Perfect" — she prefers the term "anterior" in this context — to be a special instance of "Tense", Dahl's characterization of the "Perfect" (PFCT) is more relevant to our discussion, cf.: "PFCT is rather consistently marked periphrastically... . Typically, constructions involving a copula or some auxiliary together with some past participle or similar form of the verb are used" (Dahl 1985: 129). Dahl (1985: 133 ff.) furthermore makes the distinction between the Perfect of result and Resultative constructions and remarks that "there also seems to be a high correlation between passive voice and resultative constructions: indeed, in many languages, resultative constructions are only found in the passive voice or some functional equivalent of it..." (1985: 135). The difference between the Perfect of result and Resultative constructions — according to Dahl (1985: 134 f.) — is that in the Perfect of result there is a focus on the event, whereas in the Resultative construction the focus is on the (resulting) state. Although Dahl apparently assumes that "passives" are instances of Resultative constructions, there is reason to believe that at

Arguments

against the passive

149

least in some languages the "passive" could in fact represent a Perfect of result, cf. the following examples from Swedish: (5)

När hon kom ut pä gatan, upptäckte hon, att bilen var stulen. 'When she came out onto the street, she discovered that the car was stolen.'

(6)

När hon kom ut pä gatan, upptäckte hon, att bilen blev stulen. 'When she came out onto the street, she discovered that the car was being stolen.'

(7)

Bilen blev stulen, medan hon var inne pä banken. 'The caf was stolen while she was in the bank.'

(8)

* Bilen var stulen, medan hon var inne pä banken.

In (5) the focus is on the resulting state and hence represents a prime example of a Resultative construction. In (6) and (7), on the other hand, the focus is on the event itself, thus representing examples of the Perfect of result. Notice that in Swedish two different constructions are employed in these two instances: in (5) the auxiliary is vara "to be" and in (6) and (7) it is bli "to become". 21 We can therefore conclude that the specific morphological category employed in these types of "passive" constructions is a variant of the "Perfect", i.e., Perfect of result and/or Resultative; hence it is placed in an entirely different position along Bybee's hierarchy of verbal categories and exhibits an entirely different function than the morphological means employed in the previous type of "passive" constructions. Needless to say, the consequence of this is that the semantic structure of this type of "passive" differs considerably from that of the previous type: here the narrated event itself is altered to present a state resulting from a preceding action, whereas in the "middle"/"reflexive" type of "passive", the narrated event is not changed in any such manner. Moreover, the specific construction types in addition to the "passive" that employ this type of verbal morphology likewise differ from the range of construction types that employ "middle" or "reflexive" morphology: certainly, this periphrastic means cannot be employed in and of itself to express reciprocals, anticausatives, "plain middles", "deponent middles", "nucleonic middles", etc., just as the "middle" and "reflexive" cannot be employed to express resultati ves/states.

150

Paul Kent

Andersen

6. Valence The next types of "passive" constructions that we will discuss employ morphemes representative of the category Valence. Here there are basically two different types of morphemes that can be distinguished: (a) causatives in which the valence is increased and (b) "detransitives" in which the valence is reduced. Let us therefore start with an investigation of "passive" constructions that employ causative morphology or morphemes that have developed from causatives. Although the use of causative morphology in the "passive" is not as well-known nor as widely discussed in the literature as other types of "passives", in recent years there has been a growing literature on the subject. So, for example, in his discussion of the development of "passives" from causatives Haspelmath (1988: 34 ff.) makes reference to the following languages that exhibit this development: Hungarian (see also Dezso (1988: 306, 316 f.)), Greenlandic (Inuit), Kafa, Arawak, Wolof, Chinese (see also Hashimoto (1988)), Gujarati, and various Altaic languages. To this list we can add Korean (Keenan 1985: 262), Finnish (Lehtinen 1984: 14 ff.), and perhaps many more. It should be perfectly clear now that this type of morpheme represents a completely different verbal category than those discussed previously (i.e., the "middles/reflexives" and "Perfects"): not only are these morphemes placed in positions different from the others discussed above with respect to Bybee's hierarchical order of verbal categories, thus giving evidence of a basic difference in relevance and generality, they also exhibit completely different functions as can be seen from the fact that the construction types that employ this type of morpheme are not the same construction types that employ the other types of "passive" morphemes. With regard to the "get passive" in English and similar constructions, for example, we can determine the specific semantic function of the causative morpheme when employed in the "passive", cf.: The choice of a form like The window got broken over The window was broken seems to imply that the window somehow brought the catastrophe onto itself — if only by being in the wrong place at the wrong time. Its presence, as it were, catalyses the action performed by a quite separate agent (Barber 1977: 22). Whether this is true of all "passives" with causative morphology is not at all clear. Notice, though, that there are — semantically speaking — a number of different types of causatives; each will presumably entail a slightly different semantic analysis.

Arguments

against the passive

151

Since the "passive" is traditionally characterized as entailing detransitivization it should also come as no surprise to find that morphemes that decrease Valence are employed in "passive" constructions. The following example from Basque may in fact exhibit one such instance, should the morpheme -a actually perform the function of decreasing the Valence of the verb resulting in the auxiliary agreeing with one core constituent less than the corresponding active. (9a)

b)

Gizon-a-k txakurr-a-0 maluskatu zuan man-the-ERG dog-the-ABS beat AUX(3SgSubj3SgObj) 'The man beat the dog.' Gizon-a-k

txakurr-a-0

maluskatua

zan

man-the-INSTR dog-the-ABS beat(PASS) AUX(3SgSubj) 'The dog was beaten by the man.' (cf. Keenan 1985: 248 f.) Note also the discussion in Givon (1988) in which it is argued that the "passive" suffix -ka in Ute was also associated with detransitivization. Again, we can determine that such morphemes that decrease Valence represent yet another morphological category distinct from all of those discussed above not only with respect to the actual function expressed but also by the specific types of construction types that also employ this same morpheme.

7. Aspect When it comes to Aspect we can recall the generalization drawn by Keenan (1985: 267): "If a language has any passives it has ones which can be used to cover the perfective range of meaning." Accordingly, all languages that have at least one "passive" construction at their disposal have a "passive" with perfective Aspect. We should note, though, that there is a great deal of confusion in the literature with respect to the categories perfective Aspect and Perfect: certainly, the "passive" in English is to be regarded as an instance of the Perfect and not of the perfective Aspect. It therefore appears that this confusion is incorporated into Keenan's generalization here. This is not to say that perfective should be replaced by Perfect in Keenan's generalization: surely there may be "passives" that do indeed employ verbal morphology representative of

152

Paul Kent Andersen

the category perfective Aspect. In fact DeLancey (1981), who does differentiate between perfective Aspect and Perfect, states that "Perfective views an event from its terminal point" (1981: 647) and proposes that in the "passive" there is a "terminal view point with respect to the transitive vector" (1981: 647). Thus we see that according to DeLancey the "passive" is an expression of terminal viewpoint and hence perfective Aspect.22 Should the specific verbal morphology that is employed in such "passive" constructions actually represent the category perfective Aspect, then we would be able to conclude that this category is again distinct from the other categories discussed above, cf.: A PFV verb will typically denote a single event, seen as an unanalysed whole, with a well-defined result or end-state, located in the past. More often than not, the event will be punctual, or at least, it will be seen as a single transition from one state to its opposite, the duration of which can be disregarded (Dahl 1985: 78). We would furthermore expect to find that if these morphemes are employed in constructions other than the "passive", then these construction types would differ from those construction types that employ other verbal categories.

8. Voice Up to this point in our discussion we have shown that the morphemes distinctive of many "passive" constructions do not represent instances of the verbal category Voice within Bybee's (1985) framework. This should not, however, be taken as evidence that a category "passive" Voice does not exist; certainly Bybee must have found enough instances of morphemes for the verbal category Voice (that are placed between morphemes for Valence and Aspect along her hierarchy of verbal categories and that do furthermore exhibit a distinctive function in terms of relevance and generality that is situated between those of Valence and Aspect) to warrant her postulation of this category in this particular hierarchical position. Unfortunately, Bybee does not give any examples or references to languages where such a situation occurs. Presumably, there were many languages in her sample that exhibit a morpheme for "passive" Voice in addition to morphemes for Valence and Aspect that is indeed placed

Arguments

against the passive

153

between these other morphemes. Since I have no direct access to, or knowledge of, these languages, I have not yet been able to investigate these morphemes; thus, we are not yet in the position to give a characterization of them. It should be evident, however, that such morphemes for "passive" Voice will indeed exhibit a function entirely different from those discussed above.

9. Summary and conclusions The basic assumption of this paper has been that it is possible first of all to isolate and identify the verbal morphology that is distinctive of constructions generally claimed to be "passives" in the literature, and second of all to classify the specific morphemes as instances of prototypical verbal categories in accordance with the methods proposed by Bybee (1985) and Dahl (1985). The preliminary results presented here give sufficient evidence that the morphemes traditionally regarded as "passives" do not represent instances of a single, prototypical morphological category, but rather that they represent instances of the following distinct categories: (i) (ii) (iii) (iv) (v) (vi) (vii)

Agreement Perfect of result Resultative increase in Valence (causative) decrease in Valence ("detransitive") perfective Aspect "passive" Voice

Each of these prototypical verbal categories furthermore has its own distinct function, thus giving evidence that the constructions employing these verbal categories are themselves distinct and not merely variants of a single, prototypical construction type. Semantically, these functions can be interpreted as various processes by which a narrated event is transformed into an appropriate discourse event — or state — within the respective discourse context. Since the basic function of Agreement is referential in nature (cf. Lehmann 1988), we see that the specific function of Agreement has to do with identifying participants of the narrated event with respect to the

154

Paul Kent Andersen

discourse event within the respective discourse context without changing the narrated event in any way. In "passive" constructions employing "middle" or "reflexive" Agreement morphology we furthermore can conclude that one participant — i.e., the "Actor" — of the narrated event does not participate in the discourse event within the respective discourse context. In the case of the two types of Perfect we see that the verbal category has the effect of changing the narrated event into the state that results from the narrated event. The Perfect of result places the focus on the event itself, whereas the Resultative places the focus on the resulting state. In the case of an increase in Valence we find that the narrated event is expanded by the addition of a Participant, or more to the point by embedding the narrated event into another event, thus giving evidence for event-embedding. In the case of a decrease in Valence we have evidence that the narrated event is changed by removing one of the Participants; hence we have an instance of event-splitting in the sense of Reichenbach (1947: 268). In the case of the perfective Aspect we see that the narrated event is changed to another, more compact and perfective event. Unfortunately, the present investigation has not led to any concrete conclusions concerning the semantic function of morphemes for "passive" Voice.

Notes 1. Notice that although this particular construction may be widespread in the world's languages syntactically speaking, the specific morphology found in this construction is certainly restricted to a rather small number of languages. 2. Shibatani's characterization 40 of the passive prototype contains the following morphological property: Active = P; Passive = P[ +passive] (837). As I will argue in the course of this paper, there is no evidence for a single, prototypical passive morpheme, i.e., P[ +passive], 3. Iconoclastic though it may be, in actual fact the type of "passive" attested in English should be classified as either a stative or a resultative according to Nedjalkov and Jaxontov's (1988: 6) definition: "... the stative expresses a state of a thing without any implication of its origin, while the resultative expresses both a state and the preceding action it has resulted from". So, for example, in the sentence "When I came at five, the door was shut, but I do not know when it was shut" the first occurrence of the "passive" was shut is a stative, whereas the second is a resultative. Note, however, that not all passives can be classified as states or resultatives (especially those discussed in Sections 4, 6, 7, and 8 below). For a further discussion see Section 5 below; note, though, the difference in terminology.

Arguments against the passive

155

4. Cf., among others, Keenan (1985: 245): "Thus consider how, as field workers, we can tell if a sentence in a language is passive or not. What is it about passives that makes them observably distinct in surface from basic actives? ... Thus we cannot recognize a passive in terms of its NPS being marked or positioned in the sentence in ways different from those used in basic actives... . In fact the only way we know that (4b) above is passive is by the presence of a specifically passive suffix, -aki, on the verb. And this observation turns out to be general across languages. That is, in general in a language, what is distinctive about the observable form of passives is localized within the predicate or verb phrase (understood broadly enough to cover auxiliary verbs)." See also the relevant discussion in Haspelmath (1988: 3 — 11). A notable exception to this is Siewierska (1984: 256). 5. Notice that some authors have even attempted to motivate this axiom, cf., e.g., Haspelmath (1988: 1): "My view that the nature of passive morphology is inseparable from the nature of passive syntax..." and (1988: 51, fn. 3): "And that, more generally, whenever a syntactic construction is associated with morphological marking, the nature of the morphology is inseparable from the nature of the syntactic construction." Should the results of this paper be accepted, Haspelmath would then be forced to accept the existence not of one single "passive" construction, but rather a number of distinct "passive" constructions. 6. Cf. "A meaning element is relevant to another meaning element if the semantic content of the first directly affects or modifies the semantic content of the second" (Bybee 1985: 13). 7. Cf. "By definition, an inflectional category must be applicable to all stems of the appropriate semantic and syntactic category and must obligatorily occur in the appropriate syntactic context. In order for a morphological process to be so general, it must have only minimal semantic content. If a semantic element has high content, i.e., is very specific, it simply will not be applicable to a large number of stems" (Bybee 1985: 16 f.). 8. Bybee's hierarchy of verbal categories is reminiscent of, and certainly related to, the distinction made by Foley —Van Valin (1984) between different, structured layers of the clause and specifically between different operators of the various layers. 9. It should be pointed out that Bybee's (1985: 20) characterization of Voice is based upon Barber (1975), which in turn represents the characterization of only one single type of "passive" construction, i.e., those discussed under the heading "Agreement" below. Should the results of this study be accepted, the characterization of Voice will necessarily need modification. 10. Cf., e.g., Shibatani (1985: 825): "Cross-linguistic evidence that passives are often related to other constructions is overwhelming. In fact, it is far more deep-rooted and widespread than has been indicated by casual observations found in many descriptive materials". See also Keenan (1985: 253): "Moreover, the same types of formal morphological means used in deriving basic passives are often used to derive VPS which are not passives". 11. See Andersen (1989) for a detailed discussion. 12. As will be discussed below, the Classical Greek clause louo-metha can also mean 'We are washing ourselves' as well as 'We are washing each other'. 13. Notice, in this respect, the following generalization already mentioned in Greenberg's discussion of implicational universals: "Universal 30. If the verb has categories of person-number or if it has categories of gender, it always has tense-mode categories" (Greenberg 1966: 93). 14. Since there are numerous references in the literature of other "passive" morphemes that are also employed in reflexive — possibly even reciprocal and anticausative — constructions, these construction types may prove to be not indicative of this type of morphological category. Notice, though, that these other "passive" morphemes have

156

15.

16.

17.

18.

19. 20. 21. 22.

Paul Kent Andersen not been identified as instances of any particular verbal category (other than the "passive"): a detailed study of these morphemes may reveal the fact that they too are instances of this prototypical verbal category. At any rate, what is distinctive of the "middle" is its employment in a number of so-called (intransitive and transitive) "middle constructions". Although there has been much discussion of this notion of "affectedness" as the primary function of the "middle" in recent years, it should be noted that this very function has been recognized in some linguistic circles for more than a century. In fact, the very term päthos that was employed for this morphological category in the Greek grammatical theory has the meaning "affection", cf. Andersen (1989). The characterization of the function of the "middle" as "affectedness" should also be regarded as a very useful, but still only a prototypical characterization, because there are some exceptions. So, for instance, there are examples of the "middle" employed in impersonal "passives" in which there is no "subject" to be affected. This applies not only to impersonal "passives" of intransitive verbs such as Latin venitur, curritur, itur (cf. Lehmann 1985: 247 f.), but also to impersonal "passives" of transitive verbs that therefore result in "non-promotional passive" constructions, cf. the two examples from Plautus discussed in Wackernagel (1950: 146). Furthermore, there are also other means of expressing the fact that the "subject" is affected by the verb, cf. the type Latin pudet me and German Mir ist heiß with "active" morphology. Notice though that in a few languages the reflexive markers are placed closer to the verb root than categories for Tense and Agreement; they therefore can be regarded as less prototypical in such instances. Should the semantic characterization of this type of "passive" construction given above be valid, I see absolutely no reason not to regard constructions with active morphology that express this very same semantic structure as "passives", cf. Keenan (1985: 247): "It appears however that languages without passives have somewhat more grammaticized means for expressing functional equivalents of basic passives. Perhaps the most common means is to use an active sentence with an "impersonal" third plural subject. By impersonal here we mean simply that the third plural element is not understood to refer to any specific group of individuals." The only reason for not regarding these constructions as "passives" is by reference to the fundamental — yet totally circular — axiom discussed above. In fact Keenan (1985: 255) even states that "[o]ne might expect, then, to find passives VPS which are identical to the transitive verbs they are derived from." Moreover, there are numerous instances in which such third-person plural markers do in fact develop into morphemes for the "passive", cf., e.g., Shibatani (1985: 845 f.) and Haspelmath (1988: 37). Cf., e.g., the discussion in Bresnan (1982: 81 - 8 2 , fn. 5), Hoekstra (1984: 129 ff.), Sells (1985: 55), Riemsdijk - Williams (1986: 248 f.), Jaeggli (1986: 590 ff.) and McCloskey (1988: 44 ff.). A notable exception to this rule is Haider (1986). This particular construction contrasts with "passives" such as The glass is regularly broken by vandals', the morphology of both predicates is, however, identical. According to Nedjalkov — Jaxontov (1988: 6 ff.) the category in question here is the resultative, or more specifically the objective resultative. Note also the distinction in German between the "Zustandspassiv" with the verb sein "to be" and the "Vorgangspassiv" with the verb werden "to become". Whether terminal view point does indeed presuppose perfective Aspect is certainly debatable.

Arguments against the passive

157

References Andersen, Paul Kent 1989 "Remarks on the origin of the term 'passive'" Lingua 79: 1 — 16. Barber, E. J. W. 1975 "Voice — beyond the passive", Berkeley Linguistic Society 1: 16 — 24. Beedham, Christopher 1982 The passive aspect in English, German and Russian. (Tübingen: Narr). Bresnan, Joan 1982 "The passive in lexical theory", in: J. Bresnan (ed.) The mental representation of grammatical relations (Cambridge, MA: MIT Press), 3 — 86. Bybee, Joan L. 1985 Morphology. A study of the relation between meaning and form. (Amsterdam: Benjamins). Dahl, Osten 1985 Tense and aspect systems. (Oxford: Blackwell). DeLancey, Scott 1981 "An interpretation of split ergativity and related patterns", Language 57: 626-657. Dezsö, Läszlö 1988 "Passiveness in Hungarian: with reference to Russian passive", in: M. Shibatani (ed.), 291-328. Foley, William A. - Robert D. Van Valin, Jr. 1984 Functional syntax and universal grammar. (Cambridge: Cambridge University Press). Givon, Talmy 1981 "Typology and functional domains", Studies in Language 5: 163 — 193. 1988 "Tale of two passives in Ute", in: M. Shibatani (ed.), 417 - 440. Greenberg, Joseph H. 1966 "Some universale of grammar with particular reference to the order of meaningful elements", in: J. H. Greenberg (ed.) Universals of language (Cambridge, MA: MIT Press), 7 3 - 1 1 3 . Haider, Hubert 1986 "Fehlende Argumente: Vom Passiv zu kohärenten Infinitiven", Linguistische Berichte 101: 3 - 3 3 . Hashimoto, Mantaro J. 1988 "The structure and typology of the Chinese passive construction", in: Μ. Shibatani (ed.), 329-354. Haspelmath, Martin 1987 "Transitivity alternations of the anticausative type". (Köln: Institut für Sprachwissenschaft, Universität Köln; Arbeitspapier Nr. 5, Neue Folge.) 1988 "Passive morphology: a crosslinguistic and diachronic study" [ms.]. Hoekstra, Teun 1984 Transitivity. Grammatical relations in Government-Binding Theory. (Dordrecht: Foris). Jaeggli, Oswaldo A. 1986 "Passive", Linguistic Inquiry 17: 587-662. Jakobson, Roman 1971/1957 "Shifters, verbal categories and the Russian verb", Selected Writings II (The Hague: Mouton), 130-147. Keenan, Edward L. 1985 "Passive in the world's languages", in: T. Shopen (ed.) Language typology and syntactic description (Cambridge: Cambridge University Press), vol. 1: 243-281.

158

Paul Kent Andersen

Klaiman, Μ. H. 1982 "Affectiveness and the voice system of Japanese: satisfaction guaranteed or your money back", Berkeley Linguistic Society 8: 267 — 281. Langacker, Ronald W. — Pamela Munro 1975 "Passives and their meaning", Language 51: 789 — 830. Lehmann, Christian 1985 "Ergative and active traits in Latin", in: F. Plank (ed.) Relational typology (Berlin: Mouton), 243-255. 1988 "On the function of agreement", in: M. Barlow —Ch. A. Ferguson (eds.) Agreement in natural language. Approaches, theories, descriptions (Stanford: Center for the study of language and information), 55 — 87. Lehtinen, Tapani 1984 Itämerensuomen passiivin alkuperasta [On the origin of the Baltic-Finnish passive] (Helsinki: Suomalaisen Kirjallisuuden Seura). Lyons, John 1968 Introduction to theoretical linguistics. (Cambridge: Cambridge University Press). McCloskey, James 1988 "Syntactic theory", in: F. J. Newmeyer (ed.) Linguistics: The Cambridge Survey I. Linguistic theory: Foundations (Cambridge: Cambridge University Press), 1 8 - 5 9 . Nedjalkov, Vladimir P.— Sergej Ε. Jaxontov 1988 "The typology of resultative constructions", in: V. P. Nedjalkov (ed.) Typology of resultative constructions (Amsterdam: Benjamins), 3 — 62. Reichenbach, Hans 1947 Elements of symbolic logic. (New York: Dover) [Reprint], Riemsdijk, Henk van —Edwin Williams 1986 Introduction to the theory of grammar. (Cambridge, MA: MIT Press). Sells, Peter 1985 Lectures on contemporary syntactic theories (Stanford: Center for the study of language and information). Shibatani, Masayoshi 1985 "Passives and related constructions: a prototype analysis", Language 61: 821-848. Shibatani, Masayoshi (ed.) 1988 Passive and voice. (Amsterdam: Benjamins). Siewierska, Anna 1984 The passive: a comparative linguistic analysis. (London: Croom Helm). Wackernagel, Jacob 1950 Vorlesungen über Syntax mit besonderer Berücksichtigung von Griechisch, Lateinisch und Deutsch l 2 . (Basel: Birkhäuser).

The empty morpheme entailment* Robert Beard

1. Separationist morphologies The view that affixation and other morphological operations are separate from those of lexical and inflectional derivation has been articulated recently by Beard (1981, 1987, 1988), M o o r t g a t - v a n der Hulst (1981), Pounder (in press), Szymanek (1985), Wood (1985) and for inflection alone by Anderson (1982) and Matthews (1972). On this view, the addition and adjustment of lexical categories in lexical bases are carried out by abstract morpholexical (for word formation) and morphosyntactic (for inflection) derivation rules. Morphology comprises autonomous phonological "spelling-out" operations such as affixation, reduplication, revocalization, or autosegmental recombination. (1)

Lexeme

Lexical feature inventory I /p/

Operations on Lexemes

Grammatical Component Lexicon Syntax

Morphology Phonology

where /P/ = the phonological matrix of the lexical formant. This model, the "separation hypothesis", implies that language is based on two discrete types of basic elements, lexemes and (grammatical) morphemes, each with its own related set of operations located in different parts of the grammar. * Comments from Andrew Carstairs, Wolfgang Dressier, Yves-Charles Morin, Glanville Price, and several other participants in the conference have significantly strengthened the revision of this paper. My deepest gratitude to all of them.

160

Robert Beard

It does not take issue with the claim that lexical morphemes are signs; lexemes are undisputedly signs, direct associations of sound and meaning. The separation hypothesis is distinguished by its definition of grammatical morphemes as semantically empty phonological operations and not lexical objects.

2. The empty morpheme entailment If morphology consists of semantically empty operations modifying the phonological formant (only) of lexemes, such information as gender, agency, diminutive specification is not brought to a lexical derivation by an affix from the lexicon as lexical morpheme theories assume. This paper examines this prediction against the evidence of affix synonymy. It shows that the evidence of affix synonymy militates against the view that the gender of new lexical (L-)derivatives in Indo-European languages originates specifically in suffixes. Lexical morpheme theories (e.g., Lieber 1981, Selkirk 1982, Di Sciullo —Williams 1987), assume that affixes are lexemes, direct associations of sound and meaning stored in the lexicon; hence they predict affix synonymy as a regular aspect of the synonymy common among lexical items. These theories predict that affixes combine with stems in essentially the same fashion that lexical bases compound. (2a) b)

die [Fleisch [brühe]Ff.m]Fem die [Schön [heit]Pem]Fem

'meat broth' 'beauty'

(2) represents a typical German compound and L-derivation in which the rightmost constituent seems to transmit its category information to the newly derived word. If the lexical morpheme assumption is valid and suffixes project their category features in the same way as lexemes do: (3a) b) c)

The affix must be present to project or percolate its features. Affix synonymy should be approximate, not absolute. Synonymous affixes should, ceteris paribus, vary in their subcategorization.

(a) "Zero lexemes" are not found among compounds like those of (4) — (5) such that the rightmost element of a compound must be omitted without concomitant loss of meaning, e.g. das Fleisch 'meat' —• *die

The empty morpheme entailment

161

Fleisch 'meat (beef) broth'. Colloquial forms like eine Weiß (= Weißwurst), eine Brat (= Bratwurst) must be performative wordplay since the head noun is stylistically optional, not grammatically omitted, (b) Lexical synonymy is approximate, not exact. (4)

die Fleischbrühe die Kalbsbrühe

die Fleischbouillon (die Kalbsbouillon)

die Bohnensuppe die Kalbssuppe

Suppe is distinguished from Brühe and Bouillon by its heavier ingredients. Brühe also means 'gravy, sauce' besides 'bouillon'. Finally, (c) while some synonyms of such semantic fields are sporadically associated with one gender, such associations are only occasional (5)·1 (5)

die Vorspeise die Schultasche

das Zwischenessen der Postsack

das Nachtmahl der Sportbeutel

As Bloomfield (1933: 280) put it, "[t]here seems to be no practical criterion by which the gender of a noun in German, French, or Latin could be determined". The separation hypothesis predicts something quite different. Since Lderivation and affixation are independent, (a) derivation may operate without affixation, thus generating zero-marked forms. Since separation allows several affixes to mark an identical L-derivation, (b) it predicts absolute synonymy among such forms. Finally, in cases of derivational synonymy, we should (c) find instances of derivationally determined gender which does not vary with suffix selection. Since derivation determines meaning and affixes are semantically empty, in most instances derivation should determine gender, number, and diminution and these categories should remain consistent across suffix markers. To test the predictions of the empty morpheme entailment of the separation hypothesis against the lexical morpheme hypothesis, this paper examines two Indo-European L-derivations which are marked in each language by several ostensibly synonymous suffixes. The purpose is to discover whether suffixes or abstract derivation rules determine gender among synonymous L-derivations.

162

3.

Robert

Beard

Two Indo-European derivational classes with systematic synonymy

3.1. The western Indo-European deadjectival nominalizations The first evidence that the separation hypothesis makes the correct predictions is found in the abstract deadjectival nominals throughout the western Indo-European languages. The recategorization of (6) (there is no semantics) is not only absolutely the same for each language but across languages as well. (6)

Italian

German

Russian

French

la la la la la

die die die die die

mjagk-ost' [F] slep-ot-a [F] tis-in-a [F] bel-izn-a [F] sin' [F]

la la la la la

fals-itä facond-ia lucent-ezza brav-ura stupid-aggine

Krank-heit Fähig-keit Naiv- ität Finster-nis Tief-e

moit-eur differen-ce modern-ite faibl-esse just-ice

For the lexical morpheme position, the examples of (6) should be derivationally identical with, or at least similar to, (4) —(5). But the synonymy here is absolute, not the approximate synonymy of lexical items. Moreover, all the nouns of (6) bear feminine gender regardless of the suffix used and regardless of its productivity. Now, if the gender of an Lderivation is determined by that of the suffix, no reason exists for the consistent association of femininity in (6). The consistency which we find suggests that the rule, Adj —> N, itself, not the suffixes marking it, provides the femininity. 2 One might say that the lexical representations of these suffixes coincidentally associate feminine gender with nominalization and are fortuitously synonymous. (7)

-esse + Noun + Fem Adj_ _

-eur + Noun + Fem Adj_

-ite + Noun + Fem Adj _

-ce +Noun -l-Fem Adj - _

-ice +Noun + Fem Adj_ _

This exit is blocked two ways: (a) French -eur and German -nis are not always feminine and (b) gender is also assigned when no affix at all is added. German -nis is an unproductive nominalization suffix. French -eur is highly productive as a masculine agentive marker.

The empty morpheme entailment

(8a)

b)

DAS Erzeug-nis Bekennt-nis Verhält-nis LE lis-eur fum-eur chant-eur gag-eur

163

'product(ion)' 'confession' 'relation' 'reader' 'smoker' 'singer' 'bettor'

Should -eur and -nis be treated as homonyms, at least four problems arise: (a) an enormous, unjustified proliferation of affixes, (b) contradiction of phonological evidence, (c) circularity in the argument, and (d) no basis for choosing between homonymy and polysemy. Affix proliferation occasioned by theoretical homophony like (9) can hardly be exaggerated; besides the masculine and feminine uses of French -eur, it also marks adjectives: migrat-eur 'migratory'. If -eur represents a lexical morpheme, it represents three. Beard (1985) discusses in detail the extent of such multi-exponence across languages. Phonologically, almost all such "homonymous" suffixes of (8) are unitary and historically always have been. Without phonological support, the argument for homonymy becomes circular. Differences in meaning may not be used as an argument for postulating two independent lexical items. An argument designed to explain two unexpected meanings of an ostensibly single formant cannot postulate two formants from the assumption of the independence of the meanings. Beard (1988: 31—34) discusses in detail the impossibility of choosing between homophony and polysemy as a theoretical solution to this quandary. Theories based on the separation hypothesis posit at most one derivation rule for each of -eur's three functions (denominal adjective, deverbal agentive, deadjectival nominal) and one suffix -eur with complex conditions on insertion which allow it to be inserted in a variety of conditions. + Adjective + Masculine

(9)

0 —*

-eur j


for — unlike the German noun — in the English noun the plural is definitly not a canonical "Kennform" but is in general regularly assigned by a paradigm structure condition. Thus,

214

Wolfgang Ullrich Wurzel

the irregular plural form of house practically represents a combination of types III (suppletive stem) and I (unsystematic "Kennform"). This shows that irregularity is a gradual phenomenon. There are no problems in choosing the correct forms for lexicon representations of types II to V if the principle of disjunctive ordering is presupposed. 13 Let us take the example Stadt: If [ — PI] is demanded by the morphosyntactic representation, only //Jtat/ BF /N can be chosen, since //Jtait/BF/N is characterized as [+P1] in the lexicon and incompatible with the morphosyntactic representation. However, if [+P1] is demanded, both forms are compatible, since the canonical base forms are not specified for their categories. Here, the principle of disjunction causes the more specific form to be chosen, and this is the form with the characterization [ + P1], that is, //Jta:t/ BF /N·

To be certain, the concept of inflectional mechanism presented here still leaves undecided quite a number of questions and it is equally certain that other solutions than those proposed here would be possible in certain points. Nevertheless, this concept on the whole allows a quite plausible explanation both of the functioning and of the changing of inflectional systems. It permits practicable, easy-to-follow delimitations between unmarked and marked as well as between regular and irregular phenomena of inflectional morphology. Based on these distinctions, a differentiated coverage ensues of various "ranges of regularity" to be expected in more complex inflectional systems, including irregularities, which are thus properly integrated into this concept of morphology. Future research will show how viable this concept is.

Notes 1. Compare also recent papers by Anderson and Zwicky, especially Anderson (1986), Zwicky (1985, 1989). The concept of inflectional system advanced here furthermore shows a number of other characteristics common with Anderson's and Zwicky's position. 2. In this field, plausibility considerations are largely dominant. There is not yet any theory of the base form. 3. Compare Wurzel (1984: 51 ff.) where examples of combinations of base form and stem inflection can be found. 4. For the concept of inflection class markedness see Wurzel (1988). 5. Compare Bittner (1985, 1988). 6. In Wunderlich (1985) and Wiese (1986), it is explicitly or implicitly assumed that splural is generally regular for German nouns and that all other plurals, among them the «-plural of feminines ending in -e (the only one possible for such nouns), are irregular. Such an assumption may meet the requirements of the underlying model of

The mechanism of inflection

215

a lexical multilevel morphology but it is incompatible with the facts of German inflectional morphology! 7. For the default principle in morphology see Wurzel (1984: 125 fT.) and Zwicky (1985). 8. If a word has several forms for a specific category, the corresponding possibilities will be specified in the lexicon in a disjunctive way without blocking each other. This applies, e.g., to the German noun Sau 'sow' and its plural forms Säue and Sauen. Since, for words of this type, the η-plural is unmarked, the inflectional entry has to be as i0 1 follows: j ^pjjj · f i r s t c a s e the paradigm structure conditions assign the feature [«/PI], in the second case the features [Uml/Pl] and [n/D.Pl]. 9. Compare Anderson (1986/ 11 fT.) and Wurzel (1989). 10. Compare Anderson (1986: 30 fT.). 11. This applies similarly to words having defective (incomplete) paradigms not discussed here. 12. An interesting relevant example is the German noun Junge 'boy' with the plural form Jungs (beside Jungen and Jungens). Here, the lexicon representation takes the following form: i//jung + e / B F / N I V/jung/ BF / N [ + Pl]j l ' / r , J · Thus, instead of the base form inflection applying elsewhere (compare Junge—Jungen), actually stem inflection occurs in plural formation. 13. For the disjunction principle in inflectional morphology see Kiparsky (1982: 9 fT.), Anderson (1982: 593 ff. and 606) as well as Zwicky (1985: 3 fT.). f

/ p | 1

References Anderson, Stephen R. 1982 "Where's morphology?", Linguistic Inquiry 13: 571 —612. 1986 "Morphological theory", Linguistics: The Cambridge Survey (Cambridge: Cambridge University Press), 146 — 191. Bittner, Andreas 1985 "Das 'Stark-Schwach-Kontinuum' der neuhochdeutschen Verben", Acta Linguistica Hungarica 35: 31 —42. 1988 Starke 'schwache' Verben — schwache 'starke' Verben. Überlegungen zur Struktur des deutschen Verbsystems im Rahmen der natürlichen Morphologie [Dissertation A, Akademie der Wissenschaften der DDR, Berlin]. Kiparsky, Paul 1982 "Lexical morphology and phonology", Linguistics in the Morning Calm (Seoul: Linguistic Society of Korea), 3 — 91. Wiese, Richard 1986 "Schwa and the structure of words in German", Linguistics 24: 697 — 724. Wunderlich, Dieter 1986 "Probleme der Wortstruktur", Zeitschrift für Sprachwissenschaft 5/2: 209-252. Wurzel, Wolfgang U. 1984 Flexionsmorphologie und Natürlichkeit. Ein Beitrag zur morphologischen Theoriebildung (Studia grammatica 21) (Berlin: Akademie Verlag) [In English: Inflectional Morphology and Naturalness (Dordrecht/Boston/Lancaster/ Tokyo: Foris)].

216 1987

1988

1989

Zwicky, 1985 1990

Wolfgang Ullrich Wurzel "Paradigmenstrukturbedingungen: Aufbau und Veränderung von Flexionsparadigmen", in: A. Giacalone-Ramat — O. Carruba —G. Bernini (eds.) Papers from the 7th International Conference on Historical Linguistics (Amsterdam: Benjamins), 629 — 644. "Gedanken zur Flexionsmarkiertheit", in: M. Bierwisch — W. Mötsch —I. Zimmermann (eds.) Syntax, Semantik und Lexikon (Studia grammatica 29) (Berlin: Akademie Verlag), 259-277. "Weshalb morphologische Kategorienmarker nicht als selbständige Einheiten im Wörterbuch des Lexikons repräsentiert sein sollten, dürften und können — Ilse zum Sechzigsten gewidmet", to appear in Linguistische Studien (Z1SW der ADW der DDR, Reihe A), H. 194, 277-298. Arnold M. "How to describe inflection", Proceedings of the 11th Annual Meeting of the Berkeley Linguistic Society, 372 — 386. "Inflectional morphology as a (sub)component of grammar", this volume, 217-236.

Inflectional morphology as a (sub)component of grammar Arnold M. Zwicky

This is a report on a program of modular grammar (which I sometimes refer to as the Interface Program), incorporating morphology of several kinds, syntax, their accompanying semantics, and phonology. I pursue an enterprise of this scope in the belief that an understanding of any one of these domains depends crucially on appreciating how it fits with all of the others. The program is substantive rather than notational in character, and my goal is a framework that encompasses the full range of attested phenomena, so that at the moment at least I am willing to sacrifice theoretical parsimony for the sake of adequacy. Here I touch on a few of the aspects of this framework that concern inflectional morphology, recasting and expanding material in Zwicky (1985, 1986, 1987a, 1988). The ultimate intellectual source of these ideas lies in Sapir's (1921) notion of "grammatical process"; the more recent inspiration comes from Matthews's explorations of the Greco-Roman "word and paradigm" tradition (1972); and in consequence my ideas have an obvious kinship with those of Anderson (1988a, b), especially insofar as both of us assign no fundamental theoretical significance to the morpheme.

1. Lexemes and morphological rules Clarity will be served by making at least the distinction between the concepts of a lexeme (the Welsh lexeme CATH 'cat'), its (inflectional) forms (singular cath, plural cathod), and their (phonological) shapes (the mutation shapes cath, gath, chath, ..., cathod, gathod, chathod, ...). I will cite lexemes with capitalization throughout, forms and shapes in standard orthographies or phonemic transcriptions.

218

Arnold Μ.

Zwicky

Lexemes are complexes of grammatically relevant information, including at least the following: semantics; a syntactic category; a set of syntactic subcategories, each encoding the ability of the lexeme to occur as the head word in a syntactic construction; morphological features, including paradigm classes; a list of forms, each form being a pairing of a set of grammatical category features with one or more phonological shapes, as in the example below; and a list of stems, each stem being a complex of phonological properties, with one stem distinguished as the primary stem. A fragment of the form list for the english lexeme RHYME: {, , 2 1 correct" group (n = 18) Responses Cluster

Phoneme

Irr.

Null

Μ

Ν

Μ

Dupl. 0 0 0 5.6 0 0 0 0 0 0 0 0 0

Ν Μ /η/ /m/ Μ /d/ /g/ Μ Ν

0 0 0 0 0 0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0 0 0 0

0 0 0 11.1 0 5.6 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0 0 0 0

100.0 100.0 100.0 83.3 100.0 94.4 100.0 100.0 100.0 100.0 100.0 100.0 100.0

/f/ /Ρ/ Ν Ν ßl

0 0 0 5.6 0

0 0 0 0 0

0 0 0 0 5.6

100.0 100.0 100.0 94.4 83.3

0 0 0 0 0

0 0 0 0 0

Μ Μ 1)1 β/ /δ/ m

0 0 0 0 0 0

77.8 66.7 83.3 94.4 88.9 100.0

0 0 0 5.6 5.6 0

0 0 5.6 0 0 0

0 0 0 0 0 0

Μ Λ)/ Ν

/»/

Mean percent correct:

Μ Ν Μ Total Mean age

22.2 33.3 11.1 0 5.6 0 98.3 95.6 85.2 94.7 5.78

257

258

Bruce L.

Derwing

Actually, we now have data to suggest that the analysis 1 D is the best one, which posits two lexical variants and one low-level phonological rule: Analysis I D (Hybrid) — iz / [ +sibilant] (PI) = -ζ / Ε R l . [ +obstruent] —> [ — voiced] /

+ obstruent — voiced

This conclusion is based on a recent miniature artificial language study carried out by one of my students (Dennis 1988) on grade-school children, which confirmed an earlier study of mine with adults in showing that, while obstruent-final devoicing was largely automatic in appropriate novel environments (e. g., 92% for /z/), subjects did not know what to do when faced with novel sibilant + sibilant (e.g., /gAc + z/) or homorganic stop + stop clusters (e.g., /bek + g/), where lax-vowel insertion occurred in only 3% —16% of the trials).4 In any event, it seems clear that feature-nodes, as well as morphological analysis, will have to be included in any satisfactory connectionist or analogical account, and, with this understanding, the long-term prospects for analogical models are much better than one might gather from a first reading of the Pinker—Prince (1988) evaluation of the specific Rumelhart — McClelland realization. Moreover, there is no real prospect that the implementation of these design improvements will yield a model that, according to Pinker — Prince (1988: 182) "may be nothing more than an implementation of a symbolic rule-based account". The reason for this is that, as noted by Skousen (1989), there are a number of absolutely fundamental characteristics that distinguish the generative rule-based approach to the description of language from the analogical one, and these differences provide a potentially vast empirical ground on which the relative merits of the two approaches might be systematically and carefully weighed. I will briefly discuss only a few of these here. A key contrast, of course, is that fundamentally different mechanisms are posited to account for productivity in the two cases; thus, (1), while the generative approach requires a problematic (innate) learning strategy for discovering the "simplest" set of rules from the data, the analogical alternative is predicated upon the existence of a large memory capacity to store the data (combined, of course, with a capacity to analogize on

Morphology and the mental lexicon: psycholinguistic

evidence

259

the basis of overt similarities and differences between stimuli, already well known to exist5); (2), while the former must need to know how the rules interact (consider the endless debate in linguistics on questions of rule-ordering and the cycle), the latter needs only to be able to access data quickly, another conditio sine qua non of the real-time language processor; and (3), while the former requires finding the correct rule that applies to a given context, the latter must instead find some appropriate example after which to model behavior. A good illustration of this third contrast is provided by the soundspelling correspondences of English. There are, for example, three common pronunciations of the English digraph ea, namely, as the j\j in bead, the /ε/ in head, or as the /e/ in break. In the spirit of a rule-based account, an attempt would presumably be made to isolate one of these as the "regular" case, relegating the rest to the status of "exceptions" (cf. Pinker-Prince 1988: 137). In a study by K a y - M a r c e l (1981), however, it was shown that the reading of the novel nonce-spelling yead varied as a function of the other items in the set. Thus reading gone, bead, shoe, pour led to the reading /yid/, while gone, head, shoe, pour led instead to the reading /yed/. In other words, the "rule" employed depended on the examples consulted, in conformity with the assumptions of the analogical approach. 6 By the same token, in her investigation of the acquisition of English past-tense forms, Marchman (1988: 10—11) has found (a) that "children are sensitive to several types of regularities that govern the formation of [these forms]" and (b) that "there does not appear to be a common, developmental^ related sequence in how children organize the categories of regular and irregular verbs", leading to the conclusion that acquisition might best be viewed "as a protracted resolution of several competing and interdependent sub-systems". 7 Similarly, I have collected data which show multiple, competing sources of productive morphological regularity, even in adult speakers; in the most notable case, the past-tense treatment of nonce-stems ending in the alveolar stops /t/ or /d/, such as /smit/ and /koyd/, the results for our adult subjects show as much strength for the null pattern of hit, let, cut, put, bid, rid, etc., as for the so-called regular/-id/ pattern of verbs like hated or dreaded, and with plenty of inconsistency and individual variation thrown in, to boot (Derwing — Baker 1974).8 This points to another distinguishing feature (4), namely, that, while the generative rule-based approach implies categories whose boundaries are well-defined, and transitions in predicted behavior that are sharp and precise, the analogical approach implies categories and transitions in

260

Bruce L. Derwing

predicted behavior that are gradual and imprecise. This follows as a corollary of the inherent design feature of connectionist accounts that paths to regularity are non-unique, a property that seems to me to be perfectly compatible with the notion of a "scale of ruleness" (Mötsch 1990) and with the traditional notion of "analogical leveling" in language change (see also Derwing et al. 1988 for a phonological illustration related to the lexicon). Finally, in (5) we find a global macro-difference that perhaps best characterizes the essential contrast between the two approaches: while in a rule-based system usage is a function of the description, in an analogical account usage is the description (Skousen 1989, pp. 3 — 4, who lists many other important differences, as well; see also the discussion by Winograd 1975: 185 — 191 on the distinction between the "declarative" and "procedural" knowledge). Interestingly, on the basis of this last feature, connectionist or analogical accounts also show some promise of providing an explanation for one of the most puzzling notions in linguistics, namely, the concept of tacit knowledge, i. e., "knowledge that" of which a speaker does not have any overt knowledge (Chomsky 1965). Rumelhart (1979: 2 — 3) has put it this way: Perhaps the classical case of using knowledge how (procedural knowledge) to produce knowledge that (factual knowledge) occurs in the domain of grammatical judgements. The knowledge that we have about language seems to be largely embedded in the procedures involved in the production and comprehension of linguistic utterances. This is evidenced by the relative ease with which we perform these tasks when compared with our ability to explicate the knowledge involved in them. Semantic knowledge would appear to be the same. Whereas we can quickly interpret sentences, it is only with the most painstaking effort that we can produce definitions of terms with any generality. Though still in its infancy, the debate between the "rule" and "analogy" perspectives is certain to generate a good deal of both interest and heat in the years to come, judging by the size and scope of the opening volley from Pinker — Prince (1988). At the very least, linguists now know that there is a new kid on the block who is showing a surprising amount of spunk and verve, who has a few bucks to throw around and, though not necessarily from the very best of families, it can at least be said that he comes from one with a lot of "connections". The competition, as usual, will be good for both sides, forcing them not only to analyze their own assumptions more carefully, but also to shore up their barricades with

Morphology

and the menial lexicon: psycholinguistic

evidence

261

more substantial, empirical stuff. And although connectionist accounts in particular still have a long way to go to establish their place as a solid alternative to linguistic rules, perhaps the most exciting thing about them is that they are not only compatible with the notion of a highly redundant lexicon, but actually require one in order to work. From this point of view, at the very least, the approach can be said to have gotten off on the right foot, in the sense that it can be said to have received a good deal of implicit psycholinguistic support at the outset. One final qualification still does need to be mentioned, however, which relates to the fact that the vast bulk of research on which the full-listing hypothesis is based has been research on English, or at least on a small set of languages typologically very close to English. This is a potentially very important limitation, since, in the general scheme of things, English comes quite close to the morphologically "isolating" end of the morphological spectrum. What can we say, then, about the bounds of the lexicon in a language of the agglutinative type, or any other (such as Eskimo) in which morphological richness and productivity reach levels that we are accustomed to in English only in syntax? To take even the relatively familiar case of the Finnish language, for example, it has been estimated by Karlsson (1986) that, under the strong form of the full-listing hypothesis (which would claim that all real and potential word-forms of a language are lexically stored as wholes), "Finns would have mental lexicons that are (tens of) thousands of times larger than English-speaking people's lexicons". In view of what he regards as the implausibility of such a conclusion, Karlsson goes on to propose a modified version of the full-listing hypothesis, suggesting that only the "prototypical forms" are lexically stored, viz., only those forms "which speakers have heard, actively used, and which make sense both pragmatically and semantically" (Karlsson 1986: 28). This makes perfectly good sense. (Why create and store words, after all,· that have never been used or heard?) Even for English, this idea would allow for the possibility that some inflected forms are listed separately in the lexicon, while others are not, a situation that might help explain some of the inconsistent experimental results on this particular point (see, for example, the discussion of inflectional affixes in Aitchison 1987: 109 — 116 and the new data provided by Stemberger-MacWhinney 1988).

262

Bruce L. Derwing

4. Conclusions In sum, after a long period of neglect, rapid progress has been made in recent years in the psycholinguistic investigation of morphology, both in terms of the character of the mental lexicon and the ways in which morphological productivity might be accounted for. As would be expected from such a surge of research activity, new and even potentially revolutionary theoretical ideas have also begun to emerge. At least as often as not, these new discoveries have flown in the face of received knowledge in linguistics (such as support for the full-listing hypothesis, as contrasted to a notable lack of support for anything like the generative idea of a morpheme-invariant underlying form); we have also seen some of the pivotal rule proposals considerably weakened, such as English lax-vowel insertion (or even completely undermined, as in the case of the phonological analysis of English vowel-shift; see Wang —Derwing 1986); finally, the challenge has even been raised that rule-based systems may be fundamentally misguided. Considering the overall scorecard to date, perhaps the time is ripe to rethink some of our most fundamental and cherished theoretical assumptions, now that evidence is accumulating from psycholinguistic research that they may be dead wrong. To be sure, progress is not always painless, but it is very exciting nonetheless.

Notes 1. The issue as to how these "stages" were defined, and the justification for them, goes well beyond the scope of this paper, but is dealt with in some detail in Baker — Derwing (1982) and Derwing-Baker (1986). 2. Except the /z/-stems, whose special status can be explained (see Derwing—Baker (1979: 212).

3. Linguists often use the word rouges to illustrate this pattern in introductory courses, but this is a highly marked plural of a mass noun which children are unlikely to know. The most likely analog in many dialects is the plural form of garage (itself rather rare a usage), but this is a /j/-stem in the dialect tested. 4. Note that analysis 1 D illustrates the notion of "phonologically conditioned suppletion", as developed by Carstairs (1990). 5. See Derwing (1973) and especially (1976) for extensive illustrations of the analogydrawing capabilities of the normal child. 6. While few, if any, generativists would actually want to incorporate sound-spelling relations within the same purview of phonological/morphological description, Skousen shows that the analogical approach is equally appropriate to either domain. 7. Cf. Derwing (1979) and Derwing — Baker (1979) on the "competition" among alternative formal patterns, and Bates — MacWhinney (1987) and MacWhinney (1987) for a more general "competition model". 8. Grammaticality/acceptability judgements show a similar indeterminacy, as Ross (1979), in particular, has emphasized.

Morphology and the mental lexicon: psycholinguistic evidence

263

References Aitchison, Jean 1987 Words in the mind. (Oxford: Basil Blackwell). Baker, William J. —Bruce L. Derwing 1982 "Response coincidence analysis as evidence for language acquisition strategies", Applied Psycholinguistics 3: 193 — 221. Bates, Elizabeth — Brian MacWhinney 1987 "Competition, variation, and language learning", in: B. MacWhinney (ed.) Mechanisms of language acquisition (Hillsdale, N. J./London: Lawrence Erlbaum), 157-193. Berko, Jean 1958 "The child's learning of English morphology", Word 14: 150-177. Butterworth, Brian 1983 "Lexical representation", in: B. Butterworth (ed.) Language production (New York: Academic Press), vol. 2: 257 — 294. Carstairs, Andrew 1990 "Phonologically conditioned suppletion", this volume, 17 — 23. Chomsky, Noam 1965 Aspects of the theory of syntax. (Cambridge, Mass.: MIT Press). Cutler, Anne 1983 "Lexical complexity and sentence processing", in: G. B. Flores d'Arcais — R. J. Jarvella (eds.) The process of language understanding (New York: Wiley), 4 3 - 7 9 . Dennis, Diane 1988 Rule governed behaviour in English inflectional morphology [M. Sc. Thesis, University of Alberta]. Derwing, Bruce L. 1973 Transformational grammar as a theory of language acquisition: a study in the empirical, conceptual and methodological foundations of contemporary linguistics. (London: Cambridge University Press). 1976 "What kind of rules can children leam?", in: W. von Raffler-Engel — Y. Lebrun (eds.) Baby talk and infant speech (Amsterdam: Swets & Zeitlinger), 6 8 - 7 8 . 1979 "Psycholinguistic evidence and linguistic theory", in: G. D. Prideaux (ed.) Perspectives in experimental linguistics (Amsterdam: Benjamins), 113 — 138. 1987 "A cross-linguistic experimental investigation of syllable structure. Part I: background and methodology", in: S. Delancey —R. M. Tomlin (eds.) Proceedings of the Third Annual Meeting of the Pacific Linguistics Conference (Department of Linguistics, University of Oregon, Eugene, OR), 93 — 102. Derwing, Bruce L. — William J. Baker 1974 Rule learning and the English inflections (Final report to the Canada Council, File No. S72-0332). 1979 "Recent research on the acquisition of English morphology", in: P.J. Fletcher—Μ. Garman (eds.) Language acquisition: Studies in first language development (Cambridge: Cambridge University Press), 209 — 233. 1980 "Rule learning and the English inflections (with special emphasis on the plural)", in: G. D. Prideaux — B. L. Derwing —W.J. Baker (eds.) Experimental linguistics: integration of theories and applications (Ghent: StoryScientia), 247-272. 1986 "On assessing morphological development", in: P. J. Fletcher—Μ. Garman (eds.) Language acquisition: Studies in first language development (Cambridge: Cambridge University Press), 2nd ed., 326 — 338.

264

Bruce L. Derwing

Derwing, Bruce L. — Maureen L. Dow —Terrance M. Nearey 1988 "Experimenting with syllable structure", in: J. Powers —K. de Jong (eds.), Proceedings of the Fifth Eastern States Conference on Linguistics (Department of Linguistics, Ohio State University, Columbus, OH), 83 — 94. Diller, Karl C. 1978 The language teaching controversy. (Rowley, MA: Newbury House). Henderson, Leslie 1985 "Issues in the modelling of pronunciation assembly in normal reading", in: Κ. E. Patterson —J. C. Marshall —M. Coltheart (eds.) Surface dyslexia (London: Erlbaum), 459 - 508. Karlsson, Fred 1986 "Frequency considerations in morphology", Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung 39: 19 — 28. Kay, Janice —Anthony Marcel 1981 "One process, not two, in reading aloud: lexical analogies do the work of non-lexical rules", Quarterly Journal of Experimental Psychology 33 A: 397-413. MacWhinney, Brian 1987 "The competition model", in: B. MacWhinney (ed.) Mechanisms of language acquisition (Hillsdale, N. J./London: Lawrence Erlbaum), 249 — 308. Marchman, Virginia A. 1988 "Rules and regularities in the acquisition of the English past tense", CRL Newsletter (Center for Research in Language, University of California, San Diego, La Jolla, CA) 2: 4 - 1 3 . Marslen-Wilson, William D. — Lorraine K.Tyler 1980 "The temporal structure of spoken language understanding", Cognition 8: 1-71. Mötsch, Wolfgang 1990 "Problems of word structure theories", this volume, 79 — 85. Murrell, Graham A.— John Morton 1974 "Word recognition and morphemic structure", Journal of Experimental Psychology 102: 9 6 3 - 9 6 8 . Pinker, Steven —Alan Prince 1988 "On language and connectionism: analysis of a parallel distributed processing model of language acquisition", Cognition 28: 73 — 193. Ross, John R. 1979 "Where's English?", in: C. J. F i l l m o r e - D . Kempler-W. S.-Y. Wang (eds.) Individual differences in language ability and language behavior (New York: Academic Press), 127 — 163. Rumelhart, D. E. 1979 Analogical processes and procedural representations (CHIP 81) (La Jolla, CA: Center for Human Information Processing, University of California at San Diego). Rumelhart, David E. —James L. McClelland — the PDP Research Group 1986 "On learning the past tenses of English verbs", Parallel distributed processing: explorations in the microstructures of cognition vol. 2: Psychological and biological models (Cambridge, MA: KIT Press), 216 — 271. Skousen, Royal 1989 Analogical modeling of language. (Dordrecht: Kluwer). Stanners, Robert F.—James J. Neiser —Scott Painton 1979 "Memory representation for prefixed verbs", Journal of Verbal Learning and Verbal Behavior 18: 733-743.

Morphology and the mental lexicon: psycholinguistic evidence

265

Stanners, R. F. —James J. Neiser — William P. Hernon —Roger Hall 1979 "Memory representation for morphologically related words", Journal of Verbal Learning and Verbal Behavior 18: 399—412. Stemberger, Joseph R — Brian Mac Whinney 1988 "Are inflected forms stored in the lexicon?", in: M. Hammond — Μ. Noonan (eds.), Theoretical morphology: Approaches in modern linguistics (San Diego: Academic Press), 101 — 116. Vennemann, Theo 1974 "Words and syllables in natural generative grammar", in: A. Bruck et al. (eds.) Papers from the parasession on natural phonology (Chicago: CLS), 346-374. 1988 "The rule dependence of syllable structure", in: C. Duncan-Rose — T. Vennemann (eds.) On language: Rhetorica, phonologica, syntactica: a Festschrift for Robert P. Stockwell, from his friends and colleagues (London: Rontledge), 257-283. Wang, H. Samuel —Bruce L. Derwing 1986 "More on vowel shift: the back vowel question", Phonology Yearbook 3: 99-116. Winograd, Terry 1975 "Frame representations and the declarative/procedural controversy", in: D. G. Bobrow —A. Collins (eds.) Representations and understanding: studies in cognitive science (New York: Academic Press), 185 — 210.

Rule-creating creativity: analogy as a synchronic morphological process* Jaap van Marie

1. Introduction In this paper, I will deal with the following two topics: first, the importance of the analysis of existing words for the study of morphological creativity and, in relation to that, second, the role of analogy in the process of the creation of new complex words. By stressing these points, this paper joins in with the tradition established by (among others) Humboldt, Paul, and Saussure, according to which existing words, both simplex and complex, are subjected to a constant, never-ending inspection on the part of the speakers of the language. In this conception, the lexicon must, in the words of Humboldt, not be equated with "eine fertig daliegende Masse" since it is — even if we leave the coining of new words outside consideration — "ein fortgehendes Erzeugniss und Wiedererzeugniss des wortbildenden Vermögens" (Humboldt 1836: 109-110). A recent rephrasing of this view can be found in Mötsch (1987: 24): "The creation of new words (...) presupposes rules. But rules need not have an existence of their own. We may conceive of rules as the result of a process of analysis operating on similarity of items of the vocabulary". Motsch's views come close to the stand taken in van Marie — Koefoed (1980) and van Marie (1985) where it was argued that the speakers of a language have the capacity to construct rules on the basis of the existing words, a capacity we called "rule-creating creativity". This ability, then, directly bears upon analogy as a synchronic morphological force. In the following, I hope to make clear that this never-ending inspection of the words in the lexicon may have different results and may result in * I am indebted to Geert Booij and Caroline Smits for their critical comments on an earlier draft of this paper.

268

Jaap van Marie

different coining-devices. In my opinion, at least the following three phenomena should be distinguished: (1) the incidental actuation of existing derivational patterns, (2) the coming-into-existence of new derivational patterns, and (3) non-derivational coining, particularly affix substitution. (Note that for reasons of space limitations the third phenomenon will not be discussed; but cf. van Marie 1985: chap. 8.) From this it follows that analogy, even within the realm of derivational morphology, is no homogeneous concept. Yet, however different these phenomena may be, they have one thing in common: their starting point. All three processes have their origin in the analysis of existing, complex words in the lexicon. The following attempt to come to grips with analogy as a synchronic force is based on a large-scale experimental investigation of morphological creativity and the different forms it may adopt. This investigation involves the "feminization" of neutral personal names in Dutch by children at the age of 11 to 17, which has resulted in a corpus of over 5,000 terms for females. In modern Dutch, there is a whole series of such categories of female personal names based on nouns, cf. the following list of morphological categories (not exhaustive): (a)

-in: ap-in, boer-in, cf. aap 'ape, monkey' and boer 'farmer' -es: prinses, cf. prins 'prince' -se: kasteleinse, cf. kastelein 'innkeeper' -e: leerlinge, cf. leerling 'pupil' -ster: wandelaarster, cf. wandelaar 'walker, promenader'

(b)

-(t)rice: conductrice, cf. conducteur 'conductor' -(t)euse: controleuse, cf. controleur 'controller'

(The suffixes in (a) are added to a base, whereas the suffixes in (b) have the same position as the suffixes of their neutral counterparts.) What is interesting about these categories of female personal names is that most of them are decidedly non-productive. Only the categories in -e and -ster are productive, though the derivational base of both categories is of rather a restricted type (van Marie 1985, 1986). As the semantic properties of these categories are very transparent, they constitute an excellent starting point for an experimental investigation of morphological creativity as the speakers are not hindered by semantic opacity in their attempts to find the most suitable female counterpart of the given neutral personal names.

Rule-creating

creativity

269

2. The incidental actuation of existing derivational patterns The never-ending inspection of existing words on the part of the speakers of a language may result in the setting-up of completely incidental rules in order to actuate a normally extinct word-type (van Marie —Koefoed 1980; van Marie 1985). These rules underlie the, often stylistically marked, ad hoc formations which form part of a pattern whose characteristic precisely is that it "normally" never gives rise to newly coined words. In relation to these incidental rules several remarks are due. First, these rules are of the "normal" additive type, i.e., they add an affix to a base (of a complexity lower by one degree), which means that they do not involve affix substitution. Put differently, as far as their formal properties are concerned, these rules are completely identical with the "permanent rules" which account for the actuation of the productive word-types. From this it follows that, as far as their formal and semantic characteristics are concerned, these ad hoc formations may completely conform to their existing counterparts (from which the incidental rules by means of which they are formed are deduced). Consider, e.g., vrekkin, which in the tests turned out to be by far the most popular female counterpart of vrek 'miser'. As to both its formal and semantic characteristics, this formation cannot be distinguished from its actual congeners such as held-in {held 'hero') or vriend-in (vriend 'friend'). Note, however, that the ad hoc derivatives may also differ from their actual counterparts, in consequence of the fact that the incidental rules are often of a rather straightforward, "unsophisticated" type. Specifically, all kinds of details characteristic of the actual words may be glossed over in the process of rule creation. Consider the following cases which illustrate the point I wish to make. Ad hoc formations very frequently differ from their existing counterparts as to their formal make-up. As far as existing words are concerned, -in (as a pure Germanic suffix) hardly ever figures after non-native stems. In the experimental data, however, formations in which -in is added to a non-native base do occur. Consider giraff-in as the female counterpart of giraffe 'id.', in which -in is added to a non-native simplex base, a phenomenon which has no parallel in the actual words of Dutch. Compare also monteur-in (cf. monteur 'mechanic'), in which -in figures after the non-native suffix -eur, a combination of affixes which does not occur in the actual words of modern Dutch (and which, by the way, sounds rather odd indeed).

270

Jaap van Marie

As mentioned above, what these ad hoc formations show is that the incidental rules appear to be of a rather straightforward, oversimplified type. All kinds of additional "details", particularly those relating to the conditions on the base, may easily be glossed over in the process of rule creation. Note that in a framework which involves rule creation, this can be explained in a natural manner, since it is even hard to imagine how it would be possible to detect all these details. As a first guess it does not seem too far-fetched to start from the idea that it is one (or very few) word-pair(s) to which the analysis of the actual words preceding the creation of an incidental rule is restricted. That is, on the whole it will be only one or very few word-pair(s) which are taken into account. From this it follows directly that it is simply impossible to discover all the details of the extinct word-type, which means that the deduced incidental rules are of the straightforward, "degenerate" type more or less by definition.

3. New derivational patterns Let me next discuss the second manifestation of analogy as a synchronic force. The experimental study of morphological creativity makes perfectly clear that the inspection of the actual words in the lexicon may lead to new derivational patterns which can be characterized as short-cuts in the derivational system and which lead to (informally) affix clustering. In this paper I will confine myself to those short-cuts which involve the reinterpretation of a derivative of the second degree as a derivative of the first degree (cf. van Marie 1984). Clearly, this type of reinterpretation is nothing but the direct — synchronic — reflex of the constant inspection on the part of the speakers of a language to which the words in the lexicon are subjected. In order to come to grips with this type of reinterpretation, consider a formation like dans-er-es. In grammars of modern Dutch, it is common practice to regard this word as the derivative in -es of danser 'dancer', which in its turn is the derivative in -er of dans-en 'to dance'. That is, danseres is usually considered a derivative of the second degree. However, as was already pointed out by Paul (1920: 245) in many cases it is questionable whether analyses of this type are justified, cf.: Sehr häufig ist der Fall, dass eine Ableitung aus einer Ableitung in direkte Beziehung zum Grundworte gesetzt wird, wodurch dann auch

Rule-creating creativity

271

wirkliche direkte Ableitungen veranlasst werden mit Verschmelzung von zwei Suffixen zu einem. The net result of this is clear: a formation like danseres can directly be related to dans-en, resulting in the coming into existence of the new, "secondary" suffix -eres. This example suggests that we should distinguish synchronic mechanisms of reinterpretation: speakers of a language have the ability to place the actual words in different configurations, i.e., they may establish "new" derivational relationships — relationships which are independent of fixed derivational patterns. Note, moreover, that these mechanisms operate on existing complex words. It is my impression that this type of reinterpretation does not take place at random but that it is particularly prominent in specific types of complex words, i.e., those which, in one way or another, involve a nontranspositional process. If this impression is correct, this finding is important in that it makes clear that the synchronic mechanisms underlying this type of reinterpretation are, at least in part, governed by grammatical factors. On the basis of my above remarks, the synchronic mechanisms bringing about the synchronic short-cuts in the derivational system can be defined as follows: (a)

[[[BASE]X A]x B]y => [[BASE], AB]y

(b)

[[[BASE]X C]y D]y => [[BASE]X CD]y

(c)

[[[BASE]X E]x F]x

[[BASE]X EF]X

Examples: groenigheid, the abstract noun in -heid derived from groenig 'greenish', which in its turn is based on groen 'green', exemplifies (a); danseres, which illustrates type (b), is discussed above; lokettiste, the female counterpart in -e of lokettist 'ticket-clerk', which in its turn is derived from loket 'ticket window, ticket office', is an exemple of (c). From the above it follows that reinterpretation of words which are formed by means of combination of a transpositional and a non-transpositional process (i.e., (a) and (b)) have the same effect: in both cases there arises a new transpositional suffix. All new suffixes, transpositional or not, consist of the "condensed" combination of the former suffixes (which may remain in the language as independent elements but this need not be the case).

272

Jaap van Marie

As said before, the experimental study of morphological creativity makes perfectly clear that these short-cuts in the derivational system are far from rare, as many newly coined terms for female persons contain such a secondary, condensed affix. See below for some examples: * -aire (probably < neutral -air + female -e), cf. montaire ( = female monteur 'mechanic') * -ares ( < neutral -aar + female -es), cf. nazares ( = female nazaat 'descendant') * -(e) res ( < neutral -er + female -es), cf. posteres ( = female post (bode) 'postman') * -(e)rin ( < neutral -er + female -in), cf. girafferin (= female giraffe 'id.') * -eure ( < neutral -eur + female -e), cf. collegeure {— female collega 'colleague') * -iere ( < neutral -ier + female -e), cf. makeliere ( = female makelaar 'broker') *

-iste ( < neutral -ist 4- female -e), cf. emiriste ( = female emir 'id.')

* -(l)inge (probably < neutral -ling -f female -e), cf. emiringe female emir 'id.') As can be observed in other types of creatively coined words, in many cases phonological adaptation of either stem or affix is involved. This means that there is some experimentation on the part of the speakers of the language in order to arrive at words which "sound good".

4. Conclusions In the preceding I have discussed two mechanisms which I consider to bear upon analogy as a synchronic force. In my view this discussion justifies two conclusions. First, what these mechanisms have in common

Rule-creating creativity

273

is that they take existing complex words as their starting point. I am, of course, aware of the fact that the notion "existing complex word" is a much debated one. Several linguists have doubted the relevance of this notion to competence theories of word-formation. The consequence of this stand is clear: such theories may provide insight in several respects, but they are unfit to deal with analogy and with the type of wordformation upon which analogy bears. To put this differently, within the framework of theories which deny the notion "existing complex word" all phenomena discussed here must be viewed as part of performance. Secondly, even on the basis of the above we cannot but ascertain that analogy is a rather heterogeneous phenomenon. In my opinion it is the joint operation of these two factors which underlies the fact that so many morphological theories have not succeeded in coming to grips with analogy as a synchronic process.

References Humboldt, Wilhelm von 1836 Über die Verschiedenheit des menschlichen Sprachbaues und ihren Einfluss auf die geistige Entwicklung des Menschengeschlechts. (Berlin: Schneider). Marie, Jaap van 1984 "Morfologische veranderingen in breder perspectief' [Morphological changes in a broader perspective], in: Vorm en funktie en tekst en taal (Leiden: Brill), 1 3 1 - 1 5 3 . 1985 On the paradigmatic dimension of morphological creativity. (Dordrecht: Foris). 1986 "The domain hypothesis: the study of rival morphological processes", Linguistics 24: 601 - 6 2 7 . Marie, Jaap van —Geert A. T. Koefoed 1980 "Over Humboldtiaanse taalveranderingen, morfologie en de creativiteit van taal" [On Humboldtian linguistic changes, morphology, and linguistic creativity], Spektator 10: 1 1 1 - 1 4 7 . Mötsch, Wolfgang 1987 "On inactivity, productivity and analogy in derivational processes" [Paper presented at the round table on word structure theories, 14th International Congress of Linguists, Berlin, GDR], Paul, Hermann 1920 Prinzipien der Sprachgeschichte5. (Halle: Niemeyer).

Topic 6: Typology and non-Indo-European morphologies

Sapir's approach to typology and current issues in morphology* Stephen R. Anderson

The Sixth International Phonology Meeting began in July, 1988 with a commemoration of the 50th anniversary of the death of Trubetzkoy. By the time this paper appears, we will have passed another, similar anniversary: that of the death of Edward Sapir in 1939. I hope it is not inappropriate to devote a paper read at that meeting to the work of Sapir, who was one of the few North American linguists who had a substantial acquaintance with — and an influence on — the work of Trubetzkoy. In any case, among the major figures of our discipline, Sapir surely merits a place on the podium in any discussion of the typology of morphological systems. In his little book Language (Sapir 1921), he presented one of the most detailed typological schemas ever proposed for the word structure of natural languages, basing his discussion primarily on non-Indo-European material. He also believed deeply in the importance of such an analysis. For example, much of his proposed classification of the languages of North and South America rested not on evidence of the traditional comparative sort, but rather on his perception of typological similarities among languages. While he certainly did not believe that such structural similarities were a sure guide to genetic relationship, he did feel that once the core features of a language's structural type had been identified, these were the properties that were most likely to remain (and to have remained) stable over time. As such, they could be a guide even to rather remote relationships, in the absence of other evidence. "Typology" for Sapir seems to have been pretty much the same thing as "morphological typology", and he would thus have been an ideal person to discuss the topic of "Typology and non-Indo* The research reported in this paper was supported in part by grant number BNS84 — 18277 from the US National Science Foundation, and the paper was written while the author was supported in part by a fellowship from the John Simon Guggenheim Foundation. The support of both of these organizations is gratefully acknowledged.

278

Stephen R.

Anderson

European morphologies". Since he is not available, though, I will at least try in the present paper to take my inspiration from his text. It is not immediately obvious just how to address this topic in a way that makes contact with the current work of those dealing in morphological theory. While everyone talks about languages as belonging to some or another particular type, and of such and such a feature as "typologically significant", I know of no serious attempt since Sapir's in 1921 to articulate a comprehensive view of just what the range of linguistic types might be in the area of word structure. Even today, when people talk about the type to which a language belongs morphologically, they are as likely as not to intend by this one of the nineteenth century labels "agglutinating", "inflectional", "polysynthetic", and the like: labels that Sapir had hoped to improve on. Perhaps my doubts about the vigor with which typological studies are in fact currently being pursued can be taken to be confirmed by the fact that apparently no one at all (apart from the plenary speakers) proposed a paper on this topic for the Krems Morphology Meeting, and there were no section or discussion papers on the program of that meeting that were devoted to the topic. Of course, everyone can agree that a valid classificatory scheme for natural languages — one that succeeded in isolating the essential and largely unchanging features of individual languages, as Sapir wished — would indeed be a worthwhile accomplishment. But I submit that such a high-minded goal has not in fact been achieved in the area of word structure, by Sapir or by anyone else; and that it is in fact very unlikely that it ever will be. Words have the form they do because of the syntactic structures into which they are organized; because of the phonological exigencies of the language; and especially (following the remarks of Hagege 1990) because of the history of the language with respect both to grammar and to sound. While it is certainly possible (and indeed interesting) to study the properties of morphological systems, I think it is unlikely that these systems per se, standing as they do in a position where they are subject to so many other influences, will turn out to have genuinely autonomous properties that will reveal the true "genius" of a language. In the earlier history of typologizing, though, at the time when a linguistic type was of necessity a morphological type (because neither syntax nor phonology in the modern sense had really been discovered yet), there was a tendency to see the goals of typological study in relation to a supposed teleology in language change. Some early (and not particularly lamented) typologists, who took the linguistic structure of Greek,

Sapir's approach to

typology

279

Latin, and/or Sanskrit to represent the acme of human intellectual organization, saw the role of typology as providing an account of the evolutionary stages by which the spirit of mankind had risen out of the primeval slime of isolating-ness to pass through the stages of agglutination and polysynthesis to arrive at the ultimate goal of inflection. Sapir compared this approach to that of the "zoologist that sees in the organic world a huge conspiracy to evolve the race horse or the Jersey cow" (Sapir 1921:124), and there is no particular reason to soften that judgment now. Especially in light of the fact that attested changes go in all directions among the classical morphological types, there is no reason at all to imagine that such a typology will serve as a guide to the orientation of linguistic change, even if it does turn out to serve some other purpose. But if it is indeed so unlikely that typology will show us the path of a stadial evolution of languages, why do we bother to do it at all, and what does a typology really amount to? While one sometimes gets the impression that a typology is simply a taxonomy — a classificatory scheme intended to provide a convenient set of labels for roughly equal subclasses of languages — that obviously cannot be correct; or at least if it is, it describes an enquiry of singularly little intellectual interest. To see this, suppose someone were to propose the following classification: (a) those languages in which the root meaning 'tongue' begins with a [ -I- Coronal] segment are to be called "iconic"; (b) those languages in which the root for 'tongue' begins with a [ — Coronal] segment are to be called "anti-iconic"; and (c) those languages in which there is more than one root for 'tongue', belonging to different classes, or in which there is no basic root for this notion at all, are to be called "transitional". Now in fact I have not tabulated the languages of the world that would belong to each of these classes, but suppose for the moment that such a count revealed roughly equal numbers of languages in each set. Still, no one would (or should!) take this proposal seriously. But why not? Presumably, what makes such a proposal blatantly silly is the fact that nothing at all follows from the phonological composition of the root for 'tongue' in a language, except for something about the way sentences sound which involve talk about tongues. The early typologists at least had the idea that something would follow from determining whether a language was "isolating", "agglutinative", etc.: namely, some predictions about what was likely to happen to it in the next round of linguistic change. If we now have serious reservations about the utility of that terminology, it is presumably because we are pretty sure that those consequences do not, indeed, follow: that is, that the traditional terms

280

Stephen R. Anderson

do not really characterize a teleology inherent in language development. Our doubts do not spring from problems about how to apply the labels (although this may in fact be a far from trivial matter as many writers, including Hagege 1990, have made clear). So we can conclude that the parameters of a typology ought to be ones from which something follows: that is, they ought to identify groups of properties that co-vary with one another, so that knowing how one thing works entails knowing about others as well, as a direct consequence of whatever it is that motivates the typological labels. This is certainly the sense that typology (at least good typology ...) has had in syntax. For instance, the alacrity with which the syntactician strives to know immediately, for any language under discussion, whether it is SVO, SOV, or VSO (or perhaps one of the other possibilities brought forcefully to our attention by Derbyshire —Pullum 1981), comes from the fact that at least since the work of Greenberg (1963), we have felt confident that knowing the relative order of the main constituents of a simple transitive main clause will allow us to predict a whole host of other things, ranging from the relative order of modifiers and their heads all the way down (for some, at least) to whether the language is more likely to have a rule of vowel harmony or one of palatalization. But the best typological work in syntax certainly does not stop at identifying an inventory of correlated properties. When we find a number of things that appear to go together in a "typologically significant" way, inquiring minds want to know: Why? And the form that the answer to this question most generally takes is the development of a framework for grammatical description that will provide a relatively limited number of dimensions along which variation is possible in the system of individual languages, with each dimension allowing only a small number of possibilities. Such a framework should then make a number of individual consequences follow from the setting of each of these "parameters" in a given way. Sticking to the example of word order, the X-theory of phrase structure is a descriptive schema for the internal organization of syntactic units as all being in a sort of harmony with one another, subject only to a few options, such as "heads are phrase-final" (vs. "heads are phraseinitial"). If this is indeed true, the X-theory of phrase structure can be regarded as attempting a genuinely explanatory account of how and why particular sub-facts correlate in the way they do within individual languages. When we have achieved such a result, we are entitled to say we have (at least part of) a substantial typology, where the range of language

Sapir's approach to typology

281

types is then the same as the range of available parameter settings. Insofar as there are not many possible settings, there will not be very many possible types; and insofar as the languages avail themselves in roughly equal numbers of the available options, they will distribute in roughly equal fashion among the types. We can then use the parameter values themselves as convenient labels for the resulting typological categories. It makes excellent sense, however, to ask just how likely it is that a situation comparable to that found in syntax will in fact turn out to obtain in morphology. How plausible is it, that is, that the system of word-formation rules in a language will turn out to be globally parametrized in the way syntactic structure seems to be, as implied in any serious search for a morphologically based typology of languages? Some efforts have been made along these lines by, for instance, those who think of morphology as the syntax of words and who focus on a program of extending essentially syntactic insights to domains within the word. Regardless of how one feels about the ultimate productivity of this enterprise on its own terms, it still leaves most of the more mechanical side of word formation — getting the segments right and getting them in the right places — largely untouched. When one turns to the details of how words are composed of phonological material, my own impression is that the number of substantial correlations within a language that could give rise to a productive parametrization of the sort sketched above is actually very limited. Much of the reason for this (as stressed by Hagege 1990) is the fact that word structure is not really an autonomous domain in the same way syntax and phonology are. Much more of what one finds in morphological systems is the result of the interplay of other areas of grammar, and especially of diachronic change; and it is in these other domains that the basic parameters of a language's structure are presumably to be sought. However interesting it may be to study, much of morphology may in some sense be an epiphenomenon. To shift the emphasis of the discussion somewhat, though, and despite the rather pessimistic attitude suggested above toward the potential significance of any typology of morphological systems, the search for such a typology has an important role to play; and indeed, when put into proper perspective, it constitutes another way of posing the most fundamental questions there are about word structure. This is because, beside the search for overall correlations among phenomena that might yield a minimal specification of the range of languages in the world, there is another methodology available for pursuing typological studies. This is

282

Stephen R. Anderson

to attempt to ensure that whatever the account that is given of some area of linguistic structure (morphology, for instance), it is one that will be adequate to accommodate all of the systems that might be encountered. One can do "typology", that is, in the sense of exploring the full range of diversity in the languages of the world, and without necessarily feeling that the effort has been a failure if that diversity is not ultimately reducible to some small number of binary- or ternary-valued parameters. This is actually a perfectly respectable goal to take along in exploring the typology of morphological systems (among other parts of language), even though it is not what most people think of as the goal of typology. But now notice that on this reading, the development of a genuinely adequate typology is a project which is essentially co-extensive with the development of a generally applicable morphological theory. That is, a theoretical descriptive framework which is really adequate to all of the world's languages, can be seen as constituting in itself the most central kind of a "typology", even if its descriptive dimensions do not reduce to some small number of parametric possibilities. Of course, where correlations are to be found, they ought properly to be incorporated into the theory; and if enough are found, maybe it will be possible once again to interpret the theory as a set of labels for language types. But this is only one of many possible outcomes, and in the meantime there seems no particularly good reason to distinguish fundamentally between linguistic typology and linguistic theory, construed as a general enterprise. Let us, then, turn to the most fully developed notion of word-structure typology in the traditional literature, taking this to be (at least the outline of) a theory of the range of constructs that are necessary in principle for the description of the morphological systems of all possible languages. As I suggested at the outset of this paper, the system outlined in Sapir's Language goes as far as anything in this direction, and it is that framework that I would like to examine briefly here. Following a tradition represented also by his teacher Franz Boas, Sapir starts from the notion that the word structure system of a language is composed of two kinds of object: a set of grammatical processes and a set of concepts expressed by the application of those processes. There must also be, of course, a basic word stock that serves as the foundation for morphological elaboration. Now the important thing about these two sets is the fact that they are in principle quite independent of one another: The question of form in language presents itself under two aspects. We may either consider the formal methods employed by a language,

Sapir's approach to typology

283

its 'grammatical processes,' or we may ascertain the distribution of concepts with reference to formal expression. What are the formal patterns of the language? And what types of concepts make up the content of these formal patterns? The two points of view are quite distinct. (Sapir 1921: 57) The independence of the two sorts of object means that the relation between processes and concepts is (at least in the general case) many-tomany, with the same process associated with multiple concepts and/or the same concept expressed by more than one process. For example, in English the single process "suffix /z/" (more formally, /Χ/—>/Χ + ζ/) may be invoked to express plurality, third singular present in verbs, possessive, etc.; while the same concept, e.g., 'plural' may be expressed in a number of different ways ("suffix /z/", "umlaut stem /au/ to /ai/", /Xum/—»/Xa/, etc.). This is a notion of morphological structure which has also been adopted in the more recent literature in one form or another by, among others, Robert Beard, Brian Joseph, Arnold Zwicky, and myself (to mention only some of the authors represented in the present volume). As these and other writers have stressed, it differs in quite fundamental ways from views based on the traditional conception of the "morpheme", an exaggeratedly minimal Saussurean sign composed of the unity of a constituent of form with a constituent of meaning. As opposed to the oneto-one relation between elements of form and of content that ought to obtain if words were really composed exhaustively of such "morphemes", the picture presented by Sapir (and its successors in the current literature) predicts the rather more complex picture that seems actually to characterize much of natural language (and of which, of course, the simple oneto-one association is merely a special, limiting case). Reasons to believe in such a position have been developed elsewhere (see, e.g., Anderson 1988); the present remarks are intended only to point out that this was also Sapir's view. With respect to the typology of morphological systems, Sapir proposed to replace the traditional one-dimensional classifications (such as that into isolating, agglutinating, inflecting, etc., languages) with a number of logically independent classificatory parameters. Given his view that morphology is made up of a set of processes and a set of concepts, it is quite natural that the basic division within his scheme is between one dimension relating to the type(s) of processes a language employs and another referring to the sorts of concepts it expresses morphologically. A third dimension, referring to the degree of internal complexity of words, cor-

284

Stephen R.

Anderson

responds to the traditional scale that runs from "isolating" to "polysynthetic". Sapir actually has little to say about this last dimension (which for him runs from "analytic" languages through "synthetic" to "polysynthetic" ones), and since it is less obvious how it might correspond to a distinct aspect of the grammar, I will ignore it here. Our present focus is on Sapir's typologies of processes and of concepts, and on the presuppositions about morphological structure which appear to underlie them. The types of process that Sapir recognizes can be described in comparatively brief terms, since the inventory he gives is a rather familiar one. He notes that concepts can be expressed by any of the following techniques: (la) b) c) d) e) f)

Word order Composition (compounding of stems) Affixation (including prefixation, suffixation, and infixation) Internal modification (vocalic or consonantal ablaut, consonant mutation, etc.) Reduplication Variations in accent (pitch, stress, etc.)

This catalog of possibilities is not a particularly revolutionary one, and in fact the main point to note about it is the fact that affixation is treated as just one kind of process, on a par with other changes of a nonaffixal character. Morpheme-based views of morphology, in contrast, tend to consider affixes as the only really legitimate kind of morpheme, with everything else described as some sort of morphologically conditioned phonological concomitant of a (possibly null) affix. A contemporary account of the formal processes that play a role in morphological structure would look quite like Sapir's, though the distinct status of some of his types within morphological theory can perhaps be questioned: we might well omit "word order" for instance, since that is not really an aspect of morphology. Reduplication can probably be subsumed at least partly under the heading of affixation, following the now familiar line of McCarthy (1981), Marantz (1982), and others. 1 The category of variation in accent probably involves some instances of affixation and some of internal modification; in Sapir's days accentual phenomena (especially pitch) were treated as somewhat more exotic and less connected with the segmental phonology than we would accept today. On the other hand, we might well add to Sapir's list the possibility of circumfixes (simultaneous prefixation and suffixation corresponding to a single unit of morphological form). This particular kind of 'discontinuous

Sapir's approach to

typology

285

affixation' is particularly well developed in some Indonesian and Philippine languages. For example, in Indonesian ke-...-an seems to constitute a unit, as indicated by the statement that the prefix ke- "is seldom used, except in conjunction with the suffix -an" (MacDonald 1976: 63). Thus, kebisaan 'capability' is derived directly from bisa 'be able' rather than by prefixation of ke- to a hypothetical *bisaan, or by suffixation of -an to an equally hypothetical *kebisa. This circumfix can in fact be applied to more complex bases, as in the case of ketidakmampuan 'impotence', derived from tidak mampu 'not be able' (cf. mampu 'be able'). Its component parts are only marginally and unproductively attested as independent affixes, and not with these roots. The same circumfix also appears in formations like kehujanan 'be caught in the rain' (cf. hujan 'rain') and kelihatan 'be visible' (cf. lihat 'see') where the verbal status of the derivatives is completely incompatible with any claim that the prefix and the suffix represent independent processes that happen to be cumulated in these forms. 2 Among the non-affixational possibilities, it is worth noting further that there are at least a few languages in which metathesis or reordering of segments serves as the marker of some grammatical information. An example is the aspectual category in several of the Salish languages illustrated by the Clallam contrast between ckwu-t 'shoot' and cukw-t 'shooting', based formally on the interchange of the root vowel and the consonant preceding it. Some controversy has surrounded the issue of whether metathesis is involved in such examples as a primitive, but it is argued in Anderson (1988) that at least some of the Salish languages (notably Saanich: see Montier 1986) must be so analyzed. A particularly interesting possibility from a formal point of view is that of marking a grammatical category not by the presence of some overt formative but rather by the (anti-iconic: see Dressier 1985, 1987) deletion of phonological material from the base. As in the case of morphologically motivated metathesis, the number of clear examples of this phenomenon are quite limited, but it must apparently be recognized in some instances. Apart from examples cited in Anderson (1988), a case of this sort has recently been described in some detail by Martin (1988), on the basis of work by Broadwell (1987). Languages of the Southwestern Muskogean group show a process of "syllable (or rhyme) dissociation" illustrated below: (2a)

[Alabama] i. balaa-ka 'lie down (sg.)'; bal-ka 'lie down (pi.)' ii. batat-li 'hit once'; bat-li 'hit repeatedly' iii. kolof-li 'cut once'; kol-li 'cut repeatedly'

286

b)

c)

Stephen R. Anderson

[Choctaw] i. bonot-li 'roll up (sg. obj.)'; bon-li 'roll up (pl. obj.)' ii. bakaaf-li 'split (sg. obj.)'; bak-li 'split (pl. obj.)' [Koasati] i. atakaa-li 'hang (sg.)'; atak-li 'hang (pl.)' ii. lataf-ka-n 'kick (sg.)'; lat-ka-n 'kick (pl.)'

Martin shows that the operation involved in these forms is one which derives the plural (or iterated, etc.) form from the singular by dissociating the final syllable from the prosodic pattern and then re-associating its onset consonant melody into the preceding rhyme (if possible, consistent with the syllabic structures permitted in the language). Such a process would have been called a "subtractive morph" in Hockett's (1947) guide to the structuralist morphological zoo; what is important for our purposes is that it has no apparent (coherent) reformulation as the addition of an affix. With the modifications suggested above, then, a list3 like Sapir's can serve as a description of the grammatical processes available to individual languages. Each particular word-formational process invokes one of them (or perhaps more than one at a time, as in the case of German suffixes accompanied by stem umlaut). As the basis of a typology, such a list can be used to characterize the subset of what is formally possible that is actually instantiated in a particular system. Where a language has no complex words at all, of course, it requires no word-formation processes, and this can be regarded as the limiting case which is identified by the term "isolating". Where all of the rules involve affixation, pure and simple, with no other changes, this is the defining characteristic of an "agglutinating" language. If at least some of the word-formation rules of a language involve non-affixational processes (internal change, metathesis, subtraction, accent shift, etc.), this is the basis for calling the morphology of that language "symbolic" (to use Sapir's term). Finally, the phenomenon that was classically supposed to make IndoEuropean languages of the ancient type so special was the presence of complex affixation: affixes that are accompanied by some morphologically motivated phonological change. Languages with such internally complex processes are called "inflecting" in an earlier terminology, one that seems sufficiently misleading to require replacement, as for instance by Sapir's term "fusional". The defining characteristic of this type is primarily the fact that segmentation into neat, discrete morphemes is rendered more difficult by the extent to which other effects accompany

Sapir's approach to

typology

287

or obscure affixation. The overvaluation of this supposed linguistic "type" seems to be an instance of making a virtue of necessity, and there is no particular reason to recognize it as distinct. Of course, the degree of internal complexity of those word formation rules that cumulate two or more processes as a unitary indication of some category is evidently a possible dimension of the formal structure of a morphological system. This gives us a set of categories (isolating, agglutinating, symbolic, and fusional) that can be laid out (somewhat uncomfortably) on a sort of scale. But does this serve as a "typology" for language? It would, if languages were homogeneous, such that for instance the presence of a single "symbolic" process entailed the exclusivity of this formal type of word formation rule. But as everyone knows, that is not the way the facts actually do turn out. Of course a single rule in a single language does have a particular type within this scheme, but languages as wholes tend to involve greater or lesser mixtures of different sorts of processes. In the rare case where all of a language's morphology is limited to a single formal type, as for instance where Turkish is typically claimed to be limited to suffixation, we can call such a language, e.g., "agglutinating" as a whole. Most languages will not be eligible for such comprehensive labels, however. In general, it is individual rules and not entire languages that can meaningfully be called agglutinating, symbolic, or fusional in any precise, categorial sense. There are clearly generalizations to be made over the sets of wordformation processes that constitute the morphological systems of (entire) languages, but these are only very rarely of the sort captured by labels like "agglutinating", etc. Much more common is the highly local sort of generalization typified by the fact that English verb inflection is based entirely on the rules "suffix /z/" and "suffix /d/". In a given language there are typically some rather specific, language-particular statements to be made about the range of its grammatical processes, perhaps in the form of some sort of redundancy rule over morphological rules or "metarule" as has occasionally been suggested.4 The rule classes identified in such generalizations, however, tend both to be much more specific than those delimited in (1) and also to involve overlap among the classical formal types. Indeed, it is questionable whether a label like "agglutinating" actually captures a significant reality in the cases where it can be consistently applied: surely it is at least as important that all of the morphological material of Turkish is borne by suffixes as that the rules are all ones of affixation.

288

Stephen R. Anderson

If these observations are correct, there is no reason to expect any (suitably modified) version of the rule typology in (1) to serve as the basis of a typology for languages, as implied by the traditional labels. This is confirmed by the fact that when Sapir, who attempts to use such labels as one dimension of his typology, actually gets around to classifying languages on the basis of their morphological "technique", the values he assigns tend to be expressions like "fusional-agglutinative, with a symbolic tinge". Such a formulation may reflect well the overall statistical balance of different process types in a language, but it is not at all the sort of categorial description we expect of a typology, and its actual significance is far from clear. We conclude that it is important to study the range of grammatical processes found in the languages of the world, so as to understand how morphological rules operate; but that this effort is unlikely to pay dividends of the sort imagined by classical typologies. Passing from the study of process types to that of concepts, we reach what can be argued to be the most interesting and innovative part of the discussion. Sapir notes first of all that there is a range of concepts which can be formally expressed in language, and suggests that these fall into a small number of categories, as in: (3 a) b) c) d)

Basic [radical] concepts Derivational concepts Concrete-relational concepts Pure-relational concepts

For a fuller discussion of just how these are delineated by Sapir, the reader must of course consult the original text, but the essential content of the categories in (3) is as follow. The "basic" concepts of (3 a) are those involved in the fundamental semantics of basic lexical roots, which characterize the meanings of the language's elementary word stock (e.g., the properties representing the basic meaning of 'cat'). The "derivational" concepts of (3 b), in contrast, are those introduced by a set of semantic functions which operate on lexical items to yield new lexical stems from existing ones (e.g., the semantic correlate of the diminutive process which maps German Katze 'cat' into Kätzchen 'kitten'). The "pure-relational" concepts of (3d) are a set of intrinsically a-semantic features having purely grammatical significance, but nonetheless formally represented in the shapes of individual words as a reflection of syntactic structure (e.g., the role of the nominative case in Latin to mark the subject); while the "mixed" or "concrete relational" concepts of (3 c) are those that function syntactically, like the pure-relational concepts, but which can be projected

Sapir's approach to typology

289

in some positions from the semantics of some part of the structure (e.g., grammatical number in English, which is grammatically relevant as the basis of the agreement relation between subjects and verbs, but which is a function of the semantics of the subject NP). Sapir then observes that the members of these categories are not all equally necessary to the structure of language in general. In order to say anything at all, of course, the speaker of a language must have at hand some meaningful words, and so every language must express "basic" or "radical" concepts. Furthermore, these words must be capable of being put together syntactically, and the representation of syntactic structure is the business of the "relational" concepts: indicating subject-hood and object-hood, status as modifier vs. modified, etc. Much of the structure indicated by pure relational concepts may be reflected by word order alone, and so is of no particular interest to a theory of word structure. Recall, however, that Sapir himself includes word order as a "grammatical process": in consequence, he can claim that every language must of necessity have some formal expression of his pure-relational concepts, since otherwise it would have no syntactic structure. Abstracting away from word order, then, we can see that "pure-relational" concepts will be represented morphologically exactly where information of a purely syntactic nature is carried by the forms of words. As opposed to "basic" and "pure-relational" notions, however, which must necessarily be formally represented in some fashion in every language, there is no necessity for a language to have any "derivational" concepts at all. It is always possible to represent a complex meaning as a (syntactically formed) combination of two or more basic meanings, like 'little tree', instead of as a derivational function modifying a single basic meaning (as in 'tree-let'). Similarly, it is not necessary for a language to make use of any semantically significant property, like number or (natural) gender, as one of the devices that indicate grammatical structure such as the modifier-modified relation or that between a predicate and its subject. Mixed-relational concepts as a category are thus dispensable to particular languages as well. Sapir thus suggests that we can classify languages according to what elaborations of the basic, irreducible inventory of concept types ("basic" and "pure-relational") they employ. That is, a language may optionally make use of derivational concepts; mixed-relational concepts; both of these; or neither. It might seem that the resulting four classes of language would provide a rather clear-cut categorial distinction, and that any given language would belong to exactly one of the four set types; but in fact

290

Stephen R. Anderson

when Sapir actually applies this classification in concrete cases, it too turns out to be somewhat more of a scale than might be expected. Languages turn out to develop, say, derivational concepts to a greater or a lesser degree. Again, it seems more useful to think of a particular rule as "realizing" or "implementing" a concept of a particular type, rather than of a whole language as doing this. What is particularly interesting about this schema, and what is relevant to current morphological discussion, is the framework within which the distinction between relational and non-relational concepts is introduced. Limitations of time and space preclude a detailed rehearsal of Sapir's analysis here, but its essence is an argument that the grammatical pattern of a sentence is to be described in a way that is completely independent of the degree of morphological complexity (indeed, of the morphological or semantic content) of the words that actually compose it. This is suggested fairly well by the following quotation from Language. After comparing the two sentences The farmer kills the duckling and The man takes the chick, he goes on to observe (Sapir 1921: 85) We feel instinctively, without the slightest attempt at conscious analysis, that the two sentences fit precisely the same pattern, that they are really the same fundamental sentence, differing only in their material trappings. In other words, they express identical relational concepts in an identical manner. We can interpret this, only a little bit anachronistically, as a sort of incipient statement of the lexicalist hypothesis: the claim that the syntax neither manipulates nor has access to the internal structure of words. Rather, it derives a grammatical pattern in which constituency, word order, and the distribution of "relational" concepts are specified — in contemporary terms, an S-structure with no concrete lexical items contained in it. These actual lexical items come from the word stock of the language, which is made up of its basic words together with the results of applying to these any derivational functions the language may employ (and insofar as the rules of the language permit, further derivations from derived stems). Since derivational concepts are integral parts of the lexical items that constitute the "material trappings" of sentences, they clearly must be introduced within whatever we want to call the dictionary or word stock of a language. 5 Relational concepts, on the other hand, are part of the grammatical pattern: in fact, it is precisely the property of being part of the grammatical pattern that makes a concept "relational" in the first

Sapir's approach to typology

291

place. This is true regardless of whether a given relational concept is "pure" (i.e., exclusively of syntactic import) or "mixed" (i.e., related to semantic as well as syntactic properties). We must imagine that the mixed relational concepts are indicated in grammatical structure, and serve to constrain the implementation of that structure in concrete words by requiring that only items whose semantics is consistent with the features present may interpret the relevant positions in the structure. The Janus-like property of mixed relational concepts, which have both a syntactic and a semantic side, leads to the rather interesting consequence that it is not possible to determine the role a given concept plays in the grammar just by knowing its signification. 6 The same concept (in terms of signification alone), that is, may either be relational or not depending on whether it has a role to play in the grammar as well as a meaning. Consider the property 'plural', for instance, whose signification we can agree is clear and more or less identifiable across languages. Sapir compares the role of the "plural" in English with its role in Nootka 7 in a form like that below: (4)

inikw-ihl-

?friinih-

9

is- Ή

fire-in house-pl-dim-def 'the small fires in the house' Sapir argues that in Nootka the plural is a derivational category, as opposed to English, where it is a (mixed) relational one. There are two parts to this argument: "First, the plural element precedes the diminutive in Nootka" (Sapir 1921:105). Now this is only an argument if we assume that concepts of the relational sort form layers of structure that are strictly external to any layers formed by derivational concepts. On the (unchallenged) assumption that the diminutive in Nootka is in fact derivational, this entails a similar derivational status for any concept represented by a structural layer internal to that of the diminutive. Since Sapir's relational concepts are just the ones that correspond to inflectional morphology (as opposed to derivation) in traditional and current writing, the argument thus rests on the traditional observation that inflection comes "outside o f " derivation — an observation made more precise and demonstrated to be a theorem in the morphological system outlined in Anderson (1988). This argument would only be of interest if we could find some independent confirmation of the claim that only the English plural, and not that of Nootka, actually meets the criteria for being a relational category.

292

Stephen R. Anderson

This is in fact provided as the second step in Sapir's argument (1921: 105): What more than anything else cuts off all possibility of comparison between [the plural in English and in Nootka] is this, that in Nootka neither the plural nor the diminutive affix corresponds or refers to anything else in the sentence. In English 'the house-firelets burn' (not 'burns'), in Nootka neither verb, nor adjective, nor anything else in the proposition is the least concerned with the plurality or the diminutiveness of the fire. That is, neither "plural" nor "diminutive" is distributed by a rule of grammatical patterning in Nootka. In English, "plural" is a mixed relational notion, while in Nootka it is a derivational one, not because the category involved means different things in the two languages but rather because English has a syntactic rule of agreement that makes reference to it while Nootka has no such rule. Because of this difference, it is perfectly coherent for the Nootka plural marker to come inside of other clearly derivational material, while this would not be possible in English. The defining property of relational concepts is given rather discursively by Sapir, but this argument (in the context of the rather extensive discussion in chapter 5 of Language) makes it clear that what is meant is the following: (5)

Relational concepts are those that are relevant to the grammatical structure of the sentence, independent of its lexical interpretation.

Let us now sum up Sapir's framework for grammatical description, which has several interesting properties. Formally, the morphology of a language consists of an inventory of processes that can affect the shapes of words so as to express concepts. In particular, morphology does not consist of a collection of form/meaning pairs (minimal signs) that can be concatenated to form words, as in later models taking the morpheme as a primitive notion. Furthermore, the "syntax" (or the "grammar") is a system that describes abstract, neutral grammatical structures without regard to the specific words which will interpret the structural positions within them. The dictionary of a language contains its basic word stock, together with the results of any derivational processes it may utilize. Such derivational processes (better, grammatical processes expressing derivational concepts) apply "in the dictionary" in this sense. In contrast, inflectional grammatical processes (i.e., those distributing and expressing relational concepts) must apply "in the syntax", in the sense that they

Sapir's approach to

typology

293

must apply to a lexically interpreted grammatical structure. Despite this difference in their interaction with the rest of the grammar, a single set of generalizations may characterize the set of "grammatical processes" (or morphological rules) of a language, whether derivational or inflectional. This framework should seem fairly familiar to readers acquainted with the theory which has sometimes been referred to as "extended word and paradigm" morphology, but which has more recently come to be called "a-morphous" morphology (for several reasons, most notably because it tries very hard to get along without morphemes). Naturally, this will seem like a felicitous result to an adherent of this view, since there are many scholars whom it would be a lot worse to have on one's side in a difference of intellectual opinion than Sapir. However, the main point here is the slightly more modest one that such a view of the matter seemed for Sapir to flow rather naturally out of an attempt to explore the limits of the diversity to be found in the word-structure systems of natural languages. He himself said he was talking about the "types of linguistic structure" in the culminating chapter of his discussion of morphological theory. It is fairly clear that when typological investigation is carried out in a genuinely universalizing way, it is not really distinct from the development of a general theory of (morphological) structure. In other words, responsible typology is no different from "doing theory". As a matter more of the sociology of the field than of its substance, people who say they are doing "typology" often are more concerned with catalogs of surface phenomena than with explanatory frameworks, and they have a bias toward descriptions that put everything into a small number of discrete categories; while people who say they are doing "theory" often do most of it on the basis of finding out more and more about a very few languages. Everyone knows that there are more than enough both of abuses and of exceptions on each side, but the goal of this discussion has been to show that the dichotomy is a fundamentally illusory one, since there is no substantial difference between typology and theory when correctly viewed. Of course, if it turns out that the correct descriptive framework admits of only a very few dimensions of variation for languages, with few possible values on each, some will say that we have discovered a typological framework while others will say that we have found the right set of parameters for universal grammar. There is no reason to think that what would make the one set happy shouldn't

294

Stephen R. Anderson

make the others happy too, though. And seeing the typological and the theoretical enterprise as basically the same should provide a worthwhile insight for all concerned.

Notes 1. It is not actually obvious that the affixation analysis solves all the problems involved in reduplication. In particular, the actual mechanics of association of melodic material with the empty skeletal positions analyzed as an affix of reduplication are not completely straightforward. It remains to be seen whether all instances of apparent reduplication can be successfully reduced to the simple addition of an affix which happens to be melodically underspecified. 2. This is because both ke- and -an, in the limited circumstances in which they are independently attested as affixes, produce nouns from various sorts of bases. 3. Naturally, we would like to go beyond listing to provide a more explanatory account of the class of grammatical processes. Along with the apparent seductiveness of the classical morpheme, this seems to be some of what lies behind attempts to reduce all of morphology to affixation. It does not seem possible to accomplish that reduction, unfortunately; but the alternative of saying simply that the class of grammatical processes is delimited by the possibilities of a rich tranformational formalism is unsatisfactory as well. One possibility suggested by Martin (1988) is the claim that the operations available to morphology are exactly those available to the phonology, thus reducing one unsolved problem to another. A complicating factor is the role of historical change in shaping synchronic morphologies: this may have the consequence that actually attested systems contain an unrepresentative (or at least seriously skewed) selection from among the theoretically possible processes. For some discussion of this, see Anderson (1980) and Janda (1984). 4. The notion of "meta-rules" over the morphology was suggested in Anderson (1986). Janda (1982 and elsewhere) has explored in some detail the notion that individual languages make extensive use of a small number of processes. 5. The word "lexicon" has been used in a number of rather distinct senses in the recent literature: as the domain of the "lexical" or "cyclic" rules of the phonology; as the set of surface-grammatical word forms of a language (not necessarily the same as the set of its surface-phonological words); as the set of its uninflected (but possibly complex) stems, perhaps including compounds; etc. It is roughly this last sense of "lexicon" that we would like to invoke here, but given the confusion that exists in the literature about what it means to say something is "in the lexicon" we would simply avoid using the word as much as possible. Aronoff (1988) makes some particularly sharp comments concerning confusions in the use of "lexical" in recent linguistic discourse. 6. Actually, this is already clear from the fact that the same sense can be conveyed either by a basic concept associated with a root or by a derivational function. Since both basic and derivational concepts are localized within the dictionary, however, the indeterminacy involved here is less radical than the point referred to in the text. 7. Although this ethnonym has a securely established status in the linguistic literature, it is actually not an accurate name for the people Sapir referred to (more accurately known as the Tsishaath or Tseshaht — [c'lsa-^ath]) or for their language. Since this paper is not fundamentally concerned with the ethnography of the northwest coast of North America, however, we will perpetuate Sapir's usage here in the spirit of quotation.

Sapir's approach to typology

295

References Anderson, Stephen R. 1980 "On the development of morphology from syntax", in: J. Fisiak (ed.) Historical morphology (The Hague: Mouton), 51 —69. 1986 "Disjunctive ordering in inflectional morphology", Natural Language and Linguistic Theory 4: 1 — 32. 1988 "Morphological theory", in: F.J. Newmeyer (ed.) Linguistics: The Cambridge Survey (Cambridge: Cambridge University Press) I: 146 — 191. Aronoff, Mark 1988 "Two senses of lexical" [Unpublished paper, SUNY, Stony Brook]. Broadwell, George Aaron 1987 "Subtractive morphology in Southwest Muskogean" [Paper presented at the 40th Annual Kentucky Foreign Language Conference]. Derbyshire, Desmond — Geoffrey K. Pullum 1981 "Object-initial languages", UAL 47: 192-214. Dressier, Wolfgang Ulrich 1985 "On the predictiveness of natural morphology", Journal of Linguistics 21: 321 - 3 3 7 . 1987 "Subtraction in a polycentristic theory of natural morphology", in: E. Gussmann (ed.) Rules and the lexicon: studies in word-formation (Lublin: Wydawnictwo Katolickiego Uniwersytetu Lubelskiego), 67 — 77. Greenberg, Joseph H. 1963 "Some universale of grammar with particular reference to the order of meaningful elements", in: J. H. Greenberg (ed.) Universals of language (Cambridge, MA: MIT Press), 7 3 - 1 1 2 . Hagege, Claude 1990 "Do the classical morphological types have clear-cut limits?" this volume 297-308. Hockett, Charles F. 1947 "Problems of morphemic analysis", Language 23: 321 —343. Janda, Richard 1982 "On limiting the form of morphological rules: German umlaut, diacritics, and the 'cluster constraint'", NELS 12: 140-152. 1984 "Why morphological metathesis rules are rare: on the possibility of historical explanation in linguistics", PBLS 10: 8 7 - 1 0 3 . MacDonald, R. Ross 1976 Indonesian reference grammar. (Washington: Georgetown University Press). Marantz, Alec 1982 "Re reduplication", Linguistic Inquiry 13: 435-482. Martin, Jack 1988 "Subtractive morphology as dissociation", Proceedings of the 7 th West Coast Conference on Formal Linguistics. McCarthy, John J. 1981 "A prosodic theory of non-concatenative morphology", Linguistic Inquiry 12: 373-418. Montier, Timothy 1986 An outline of the morphology and phonology of Saanich, North Straits Salish (Occasional Papers in Linguistics 4, University of Montana). Sapir, Edward 1921 Language (New York: Harcourt, Brace & World).

Do the classical morphological types have clear-cut limits? Claude Hagege

0. More or less implicitly, many human languages have been assigned to one of the three, or four, main language types proposed by classical nineteenth century typologists, i.e., fusional, agglutinative, isolating, and incorporating. Only in the twentieth century was attention focused on an objective fact, which by now has become common knowledge among linguists dealing with typology: these classical types are mere abstract constructs, and no language corresponds in its entirety to a given type; rather, most languages evince a certain affinity for one type, but not to the exclusion of others, and some languages seem to present features equally distributed between two or more types. It is now necessary to go one step further. In other words, two directions of research deserve to be explored. First, to what extent does a language belong to one type or another, i.e., what are the details of poly typical complexity? Second, why does such a situation occur, i.e., what are the reasons for polytypical complexity? In order to answer these two questions, I will examine a sample of languages which appear to shed some light on one of the most-debated issues in morphological typology: the exact boundary between the so-called agglutinative and fusional types. Section 1 will present some cases in which two related principles, generally held to be characteristic of the "pure" agglutinative type, are violated: one is the "one form —one meaning" principle, the other is the "no alteration at morpheme boundary" principle. Section 2 will propose an interpretation of these violations: it will show by what process an agglutinative language can acquire fusional features. Finally, we will examine what conclusions can be drawn from these facts as regards the theoretical status of morphology and man's place in nature as a language builder. In other words, my aim here is to test the validity of a theoretical framework (the above-mentioned two principles) and, having shown its inadequacy, to substitute for it another one, whose explanatory power is wider.

298

Claude Hagege

1. Details of poly typical complexity Let us examine, to begin with, three languages traditionally assumed to be agglutinative: Hungarian, Turkish, and Kannarese. 1.1. Hungarian Consider a form häz-a 'his house', and compare it with häz-a-i 'his houses' and with häz-a-i-k 'their houses'. Judging only by these three forms, we could immediately state that there is a perfect application of the "one form —one meaning" principle typical of agglutinative languages, since we can posit k for plural of possessor, i for plural of the possessed object, and a for third person singular (of possessor). However, the same a vowel also appears in häz-a-m 'my house', häz-a-d 'your (sg.) house', häz-a-i-m 'my houses' and häz-a-i-d 'your (sg.) houses', where the possessor is singular, but not third person, as well as in häz-a-i-n-k 'our houses', häza-to-k 'your (pi.) house' and häz-a-i-to-k 'your (pi.) houses', where the possessor is neither singular, nor third person, these forms, along with häz-un-k 'our house' and häz-u-k 'their house', only providing confirmation as to the status of k and i. Consequently, as far as a is concerned, we are left with a dilemma. Either we analyze it, in all the forms where it appears except in häz-a, häz-a-i and häz-a-i-k, as a simple link vowel, and in these three latter forms as a marker of the third person (singular) of the possessor, with such an analysis implying, in a very uneconomical way, that the speaker, within one paradigm, uses two different kinds of a, which have unrelated functions. Or we posit am, ad as markers of the first and second person singular of possessor respectively, un, ato, u as first, second, and third person respectively when there is more than one possessor, these markers being combined with k, the common marker for plural possessor and a...n, resp. a...to as variants for first and second person plural possessor. But since we have already assigned a meaning to i and k and since there is no distinctive function attributable to a in the first and second person singular when there is a single possessed object, nor in the whole paradigm when there is more than one possessed object, we could also say that, for instance, the markers of the first and second person of the possessor are m and d respectively, in which case the one for third person singular is zero. 1 But this conclusion is itself untenable; although it might seem justified to the extent that we take into account the paradigm and its pressures, it makes the simple description of a central feature of Hun-

Do the classical morphological

types have clear-cut limits?

299

garian morphology impossible: any noun, within the limits imposed by meaning compatibilities, can be followed by a third person singular possessor suffix, i.e., a vowel (preceded by j- in some definite cases), which is front or back in harmony with the root vowel(s): asztal-a 'his (her, its) table', ösztön-e 'his instinct', kert-j-e 'his garden', kalap-j-a 'his hat', etc. Who would say that -a or -e here should be interpreted as zero? It appears from the above that blending of affixes and discontinuity of some morphemes, even in a language known as agglutinative, give it fusional features which are far from what we would expect of the type to which most of its morphology assigns it. Not leaving Hungarian, one could also mention other phenomena which violate the "no alteration at morpheme boundaries" principle, held to be characteristic of the "pure" agglutinative type. For instance, if we add a postposition like the inessive suffix -banj-ben after häz-a 'his house' or kert-j-e 'his garden', we get häzä-ban and kert-j-e-ben respectively. In other words, the short vowels a ([a]) and e ([ε]) have been replaced by the long vowels ά ([a:]) and e ([e:]). 1.2. Turkish I will briefly describe here some phenomena which all tend to contradict the widely shared assumption of the "perfect" agglutinative character of Turkish morphology. Consider the following list of alternations (from R. Underhill (1986: 14)): Nominative sebep kelebek *jalyan

—> *jalan

—> jala,

2.2. Palauan This Austronesian language, spoken on the westernmost island of the Carolines, would be assigned to the agglutinative type considering that many forms, in the noun and in the verb systems, are derived through suffixation of clearly identifiable morphemes. However, some of the suffixes are stressed, or enclitic, which results in strong alterations of the root vowels of stems combined with such suffixes. Let us take as an example the noun phrase expressing inherent possession. 6 The possessive suffixes constitute the following paradigm (V = stressed vowel): sg. 1.

-Vk

pi. 1. 1.

inclusive exclusive

-Vö -(m)äm

2.

-Vm

2.

-(m)iu

3.

-VI

3.

-(r)ir

Here is a sample list of some of the main phenomena that can be observed: Noun in isolation

Possessed noun

— Long vowel: short vowel: e.g., bahι: 'country' balu-el 'his country' — Short vowel following a sono-

tsork-ό roxb-äm laxm-ί sifr-ä(h)

'his need' 'their width' 'my bread' 'her book'

When the possessed and/or the possessor are plural, allomorphic variation goes even further, e.g., tsrax-äv tsraxe:-nu tsorxe-xem tsorxe ha-9is

'his needs' 'our needs' 'your (pi.) needs' 'the man's needs'

which, along with tsorex and tsork- above, give altogether six allomorphs.

3. Conclusion Two logically related conclusions can be drawn from the present study. Firstly, polytypical complexity is the result of phonetic evolution. Consequently, this study shows that morphology is not an autonomous domain: word structure cannot be analyzed short of having recourse to historical phonetics, or if it is, the kind of analysis that such a choice implies remains at the quite insufficient stage of mere description, and does not teach us what we can expect to learn about the characteristics of human languages. Secondly, and as a consequence, if we do not try to delve into the reasons for polytypical complexity, then we cannot meet the main requirement which prompts every linguist to do linguistics: to know more about man's nature. Linguistics is a social science, and as such, it has something to teach us about human qualities.12 Speakerslisteners, who are far from being "ideal", as was assumed in Chomsky (1965), build their language, through generations, less unconsciously than

306

Claude Hagege

is generally suggested. They do not care for abstract principles such as the purity of a given morphological type, be it agglutinative or flectional. What they tend to realize is ease of articulation at the lowest cost, and, when possible, regular correspondence between form and meaning. But languages are not frozen systems. They constantly change. As a consequence, what at a given time was a means to simplify communication becomes a burden as time passes. Languages are made of an interplay between various levels. What is convenient for the speaker-listener from the phonetic viewpoint may become a complication from the morphological one, as is evidenced by the Turkish facts presented above and by the fate of Estonian consonant mutation, which today is frozen, unproductive, and devoid of clear motivation. The Hungarian and Kannarese blending of agglutinative and fusional features may also be explained by the freezing of productive processes which, at an earlier stage, were motivated. The speaker-listener leaves a living trace of his activity. The spectacular evolution of Palauan morphology towards a fusional type, which makes it so strikingly different from classical Austronesian languages and gives it a Semitic physiognomy, bears evidence of a cultural (gestural) habit which cannot pass unheeded by any careful observer as soon as he arrives on the archipelago: Palauans speak very fast, they strongly stress stressed vowels, and this, unavoidably, has a devastating effect on the unstressed ones. The study of the limits between morphological types gives clues to the investigation of man and society.

Notes 1. A zero-morpheme is proposed for the third person singular suffix in the "several possessions" paradigm {-(j)aimj-(j)eim, -(j)aid/-(j)eid, -(j)ai/-(j)ei, etc.), in Kiefer (1985: 89). 2. Cf. Aronoff-Sridhar (1984: 7, 9). 3. This principle is violated even in "perfectly agglutinative" languages, like those in the Turkic family: in Azeri, the -am and the -ssn of kalir-am and kdir-sm, respectively Ί come' and ' y ° u come' are blendings of person and number, since the plural has forms which are not analyzable synchronically: -ik and -siniz. Likewise, in Osmanli, for the same verb, we have, in the present, gel-i-yor + -um, -sun, -uz and -sunuz respectively. Furthermore, in Salar and in Saryg-Yugur, Turkic languages spoken in the Chinese province of Kansu, there is no mark for the plural possessive, kak-am being, for example, in Salar, 'my brother' as well as 'our brother'. In the declension of the personal pronouns of these languages, we find inflectional features: in Salar, Τ is men, 'of me' is mi, 'to me' is maga or ma. All these phenomena are not particularly "agglutinative", to say nothing of the «regular existence» of characteristic features considered by Austerlitz (1970) as defining an agglutinative type ("developed participial system, the finite verb as cloture marker at the end of the sentence", etc.).

Do the classical morphological types have clear-cut limits?

307

4. The following data are taken from Hint (1981), and from A l l i k - H e l p - P a k k (1987). 5. The other two processes, early alternation of geminate stops and historically late quantity alternation, are still productive in the modern language as is evidenced by the treatment of recent loanwords (cf. Hint (1987)). 6. All the Palauan material comes from my own fieldwork; the results are published as Hagege (1986). 7. The detail of this process, rule 6a of the ten rules characteristic of Palauan morphophonemics, is presented in Hagege (1986: 30). It is one of the most idiosyncratic and striking features of this language: the unpronounceable CwC group (where w = semi-vowel), which results from the deletion of the root vowel whose stress has been shifted to the suffix, yields a pronounceable word by vocalization of w: taut 'aim' + el 'his' —> *twt-el —• tut-el. 8. For some exceptions to this rule, see Hagege (1986: 23). 9. By full vocalic grade, I do not mean "underlying form", since I reject this notion, especially when it is confused with an historical restitution. An abuse of that kind leads Wilson (1972: 47) and Josephs (1975: 497) to posit, in order to "explain" the i in the possessive inflection of ker 'question' (cited above), a "derivation" such as keri —> keri —» ker, by rules of "stress assignment" and "unstressed vowel deletion". As a consequence, Dressier (1974: 139) mentions Wilson and Josephs' "data" as illustrating his generalization on the vowel types that are the most threatened with deletion by apheresis, syncope, and apocope. As a matter of fact, this is by no means a productive rule in modern Palauan, but only a very old and reconstituted historical process. 10. Cf. Hagege (1986: 22). 11. However, there might be some discussion on whether the Hebrew facts are really mere violations of the "one form —one meaning" principle. 12. The linguistic theory which underlies this view is expounded in Hagege (1988).

References Allik, Jüri — Help, Toomas —Pakk, Heiti 1987 "Psycholinguistic evidence for linguemes: the Estonian rective", Symposium on language universal (Tallinn: Academy of Sciences of the Estonian SSR), 8-13. Aronoff, Mark — Sridhar, S. N. 1984 "Agglutination and composition in Kannada verb morphology", in: Papers from the parasession on lexical semantics (Chicago: Chicago Linguistic Society), 3 - 2 0 . Austerlitz, Robert 1970 "Agglutination in northern Eurasia in perspective", in: R. Jakobson —S. Kawamoto (eds.) Studies in general and oriental linguistics, presented to Shirö Hattori on the occasion of his sixtieth birthday (Tokyo: TEC), 1 —5. Chomsky, Noam 1965 Aspects of the theory of syntax. (Cambridge, MA: MIT Press). Dressier, Wolfgang U. 1974 "Phonologische Prozesstypologie", Linguistica generalia I: Studies in linguistic typology (Praha: Acta Universitatis Carolinae, Philologica 5), 1 3 5 144. Hagege, Claude 1986 La langue palau. Une curiosite typologique. (München: Fink).

308

Claude Hagege

1988

Hint, Mati 1981

Lefon inaugurate de la chaire de theorie linguistique. (Paris: College de France) ( = "Linguistic theory. A contribution to an anthropological project", Diogenes 145, 1989, 1 7 - 3 5 ) .

"Neodnorodnost' sistem ceredovanija stupenej estonskogo jazyka" [The nonuniformity of the gradation systems of Estonian], Sovetskoe finno-ugrovedenie (Tallinn) 17: 247 — 265. 1987 "Loan words and the Estonian grade alternation", in: R. Channon —L. Shockey (eds.) In honor of Ilse Lehiste (Dordrecht: Foris), 415—432. Josephs, Lewis S. —Emesiochel, M. — Tmodrang, M. —Wilson, H. 1975 Palauan reference grammar. (Honolulu: The University Press of Hawaii). Kiefer, Ferenc 1985 "The possessive in Hungarian: a problem for natural morphology", Acta Linguistica Academiae Scientiarum Hungaricae 35: 85 — 116. Underhill, Robert 1986 "Turkish", in: D. I. Slobin —K. Zimmer (eds.) Studies in Turkish linguistics (Amsterdam/Philadelphia: Benjamins) (Typological Studies in Language, 8), 7 - 2 1 . Wilson, Helen 1972 The phonology and syntax of Palauan verb affixes (Honolulu: University of Hawaii Working Papers in Linguistics vol. 4, no. 5).

Index of languages

Alabama 285 Albanian 100 Arabic 299 Arawak 150 Azeri 306 Basque 151 Bengali 3 - 1 5 Breton 62 Chinese 14, 150, 223 Choctaw 286 Clallam 285 Czech 35 f. Danish 108 Dutch 9, 100, 2 6 8 - 2 7 2 English 8 - 1 2 , 14, 19, 2 7 - 3 1 , 37, 6 1 - 6 7 , 8 0 - 8 3 , 87, 9 9 - 1 0 8 , 1 1 1 - 1 3 1 , 134f., 138, 142, 144, 147 f., 150 f., 154, 172, 204 f., 210, 212 f., 2 1 8 - 2 3 5 , 2 3 9 - 2 4 5 , 2 5 1 - 2 6 2 , 283, 287, 289, 291 f. English, Middle 30 Estonian 3 0 1 - 3 0 3 , 306 Finnish 100, 150, 233, 261, 3 0 1 - 3 0 3 French 14, 18 f., 4 6 - 5 2 , 5 5 - 5 7 , 63 f., 66, 100f„ 106, 108, 1 3 3 - 1 3 8 , 161 - 1 6 5 , 168, 1 9 7 - 2 0 2 , 228, 234, 268 Fulfulde 18, 21 Georgian 221 German 5, 9 - 1 1 , 32, 6 9 - 7 6 , 8 0 - 8 3 , 100, 107, 148, 156, 1 6 0 - 1 6 4 , 1 6 6 - 1 6 8 , 2 0 4 - 2 1 5 , 221, 223, 229, 286, 288 Greek, Ancient 71, 1 1 3 - 1 1 6 , 1 2 6 - 1 2 9 , 145, 147, 155, 175 f., 278 Greek, Modern 1 7 1 - 1 8 0 Greenlandic 150, 234 Gujarati 150

Hebrew 305, 307 Hua 221 Hungarian 14, 17, 21, 6 9 - 7 6 , 150, 203, 222, 298 f., 306 Icelandic 205 Indonesian 285 Irish 109 Italian 9, 18, 20 f., 62, 8 7 - 9 7 , 9 9 - 1 0 8 , 162, 166 Japanese 14, 142 Kafa 150 Kalkatungu 18 Kannarese 298, 300 f., 306 Koasati 286 Korean 150 Lapp 301 Latin 36, 72, 97, 1 1 3 - 1 1 6 , 1 2 6 - 1 2 9 , 145, 156, 161, 184 f., 204 f., 209, 222 f., 279, 288 Nootka 291 f. Palauan 3 0 3 - 3 0 8 Pashto 166 f. Persian 299 Polish 36, 164 Provencal, Old 201 Romanian 1 8 3 - 1 9 4 Russian 32, 148, 162, 164, 168, 203, 205, 210, 213 Saanich 285 Salar 306 Sanskrit 18, 21, 279 Saryg-Yugur 306 Serbo-Croatian 36, 1 6 4 - 1 6 6 , 168 Sen 17, 21

310

Index of languages

Slovak 36 Swahili 221 Swedish 149, 230

Ukrainian 164 Ute 142 Vietnamese 62

Tagalog 62 Turkana 17 Turkish 18, 22, 287, 298 f., 306

Welsh 217 Wolof 150

Subject index

ablativus modi 96 absolute neutralization 37 f. abstractness 37 f. accent variation 284 ff. ad hoc formations 269 f. adjacency condition 92 f. affix clustering 270 f. — homonymy 48 — substitution 268 afterthoughts 192 agglutinating 286 f., 297 agreement 145 ff., 153, 292 a-morphous morphology 293 analogical models 255, 258 — word formation 85 analogy 85, 251 ff., 2 6 7 - 2 7 3 analytic (type) 284 anaphor 188 anti-iconic 285 argument structure 61—67, 83 articulated declension 185 ff. aspect 151 f. associativity 43 — 57 atom condition 94 attested form 50 — meaning 44, 46, 52 attribute markers 183 automorphism 26 baby talk 73 f. backgrounded episodes 192 base form inflection 204 — component 54 base rule 54 binding 64 f. borrowing 9 9 - 1 0 8 bound word 232 f. bounding domain 9 bounding node 9 f. categorial markers 212 f. chiasmus 28 circumfix 284 f.

clipping 125 clitics 3 - 1 5 , 171-180, 183-194, 232 ff. coherence 25 combining forms 111 — 131 competence theories (of word formation) 273 complex adjectives 88 ff. compound(ing) 88, 9 9 - 1 0 8 , 133-138, 160 f. connectionism 253, 258 ff. connotative effect 74 control 63 f. conventional component 54 conversion 81 cross-classification of affixes and meanings 50 cross-lexeme referral 226 cyclic rules 35 ff. deadjectival nominalization 162 fT. defaults 221 defective paradigms 198 derivational component 54 derivative of the first/second degree 270 determination of gender 166 f. diachronic change 26, 281, 294 diagram 25 diminutives 72 fT. discontinuous affixes 81 discourse function 71 ff. discourse function strategies 184, 191 — 194 drift 26 dvandva 5 f., 131 dynasty of governors 9 — 12 echo words 7 f. emphasis 3 — 15, 71 empty morphs 20 f. — morpheme entailment 159—168 endoclitics 172 f., 178 f. endophoric function 184, 194 evaluative suffixes 89 excessive 69 ff.

312

Subject index

extended word-and-paradigm model 20, 293 extrinsic rule ordering 37 f. fallback procedure 252 feature-node 258 feminization 268 final-segment strategy 253 f. focalization 188, 190 f., 194 foregrounded events 193 full-listing hypothesis 251, 261 function composition 65 f. function of rules 251 fusional 286, 297 gaps 222 f. government 9 — 12 grammatical concepts 282 f., 288 ff., 294 - basic 288 f., 294 — concrete-relational 288 ff. - derivational 288 ff., 294 — mixed-relational 288 ff. — pure-relational 288 ff. - radical 288 f., 294 grammatical processes (sensu Sapir) 217, 282 ff., 293 hailstone models 20 ff. head 61 f., 82ff., 102-107, 134f. homonymy 46 ff., 52, 163 f. iconicity 25 idiosyncrasy 46, 55, 201, 213 incidental rules 269 ff. incorporated markers 213 f. indexicality 185 inflection-class membership 206 inheritance of argument structure 83 f. integration of foreign compounds 109 f. interactive orientation 71, 74 interface program 172, 217 intermediate stems 199 f. intermorphs 20 ff. intimacy 72 irregular(ity) 205, 210-214 "is a" condition 101 island 9 - 1 2 isolating 284, 286 isomorphism 25

Kennform 207, 211 ff, 225 language of love 73 ff. learnability 37 — 39 left branch condition 11,15 left-dislocated objects 191, 194 lexeme-morpheme based morphology 168 lexical component 54 — decision tasks 254 — insertion 201 — phonology 20 — pragmatics of morphology 69 — semantics of morphology 69 lexical-morpheme hypothesis 264 f. lexicalist hypothesis 228, 290 lexicon-external/internal inflection 197 ff. locality 20 locative nominalization 164 ff. logical form 3, 8 — 11 loose compounds 105 mapping rules 219, 221 f. markedness 24, 79 ff, 204 f. markedness isomorphism 31 markers of discourse prominence 192 meta-rule 287, 294 metaphorization 119f. metathesis 285 middle (voice) 145 minor rule 51, 56 modification 81, 284 modular grammar 217 morpholexical phonology 35 — 39 morphological creativity 267 ff. morphology-free syntax 228 morphopragmatics 69 — 76 morphosemantics 69 morphosyntactic representation 209 — word 233 multiple inflection 207 negation 174 ff. neoclassical compounds 112 nonmorphological word-creating rules 136ff. object clitics 191 ff. one form — one meaning principle 297

Subject index opacity 37 f. orientation of government 9 f. overrides 221 paradigm class 219, 224 f. — structure conditions 206 ff., 225 paragraph topic 193 parasitic formation 198 — 200 particle lexemes 230 ff. passive 141 — 156 past-tense inflections 259 ff. percolation 13, 79, 102 f., 160 perfect 147 ff., 151, 154 perfective 151, 154 phrasal affixes 230 f. plural inflections 252 ff. polysynthetic 284 polytypical complexity 298 ff. possessive markers 188 f., 194 postpositions 10 poverty of the stimulus 38 f. pragmatic reversal 72 — strategies 183-194 predictable form/meaning 44 — 52 prefixation 81 f., 93 ff., 122ff. principles-and-parameters approach 38 f. proclitic case morpheme 188 proclitic definite article 184 productivity 87, 91, 100-108, 251 f., 258 f. prominent discourse entities 194 quantificational domain 9 ff. quantifier raising 10—12 redundancy relations 35 redundancy rules 79, 83 reduplication 7 f., 14, 81, 284, 294 referral rules 223 regular-marked/unmarked 205 f. reinterpretation 270 resultative 148, 153 rhyme analogy 252 — strategy 254 rule coherence 25 — 34 rule-creating creativity 267 — 273 rule features 206 ff. rules vs. analogies 251 ff.

313

Saussurean sign 283 scope assignment 9, 11 secondary suffix 271 secretion 116ff., 125 selector 55 semantic drift 93 ff. — interpretation 44 f., 47 f., 52, 79, 83 f. — presupposition 70 ff. semiosis 28 semi-suffix 113 f. separation hypothesis 159 IT. shortening of compounds 100ff., 107 f., 119 f. slot 219 — calculus 226 — competition 221 speech situation 73 ff. stem inflection 204 — rules 223 ff. storage vs. retrieval 249—251 stratification of the lexical component 43, 52-57 strict cyclicity condition 37, 39 structural coherence 29 structure preservation 37 stylistically marked formations 269 subtractive techniques 81 superlative 70 ff., 89 suppletion 1 7 - 2 2 , 209fT. SVO language 105 f. syllable dissociation 285 f. symbolic 286 synonymy 160 ff. syntactic typology 280 synthetic (type) 284 — compounds 80, 82, 84 systematic identity 223 f. telos of change 30 f. topic framework 187 topicalization 188, 190, 194 truncation 50 ff. typological parameters 280 f. valence 150 f., 154 verbal noun 5 voice 152fT. vowel shift, English 262

314

Subject index

Wackernagel's Law 172 f. weak pronouns 173 — 179 word order 99, 101, 105-109, 280, 284, 289 f. word-structure rules 80 word syntax 79

X-theory 80, 280 zero derivation 228 — lexemes 160 — morphs 222

List of contributors

Paul Kent Andersen Fakultät für Linguistik und Literaturwissenschaft, Universität Bielefeld, D — 4800 Bielefeld, Federal Republic of Germany Stephen R. Anderson Cognitive Science Center, The Johns Hopkins University, Baltimore, Maryland 21218, USA Joseph Bayer Max-Planck-Institute for Psycholinguistics, Wundtlaan 1, Postbus 310, NL —6500 AH Nijmegen, The Netherlands Robert Beard Department of Modern Languages, Literatures and Linguistics, Bucknell University, Lewisburg, Pennsylvania 17837, USA Andrew Car stairs Department of English, University of Canterbury, Christchurch 1, New Zealand Danielle Corbin Universite de Lille III, CNRS, URA DO 428 "SILEX", France Bruce L. Derwing Department of Linguistics, University of Alberta, Edmonton, Alberta T6G OZ1, Canada Anna-Maria Di Sciullo Departement de linguistique, Universite du Quebec a Montreal, Case postale 8888, Succursale "A", Montreal, P.Q. H3C 3P8, Canada Wolfgang U. Dressier Institut für Sprachwissenschaft der Universität Wien, Berggase 11, A - 1 0 9 2 Wien, Austria Claude Hagege College de France, 11 Place Marcelin-Berthelot, 75231 Paris Cedex 05, France Brian D. Joseph Linguistics Department, The Ohio State University, 204 Cunz Hall, 1841 Millikin Road, Columbus, Ohio 43210-1229, USA

316

List of

contributors

Ferenc Kiefer A Magyar Tudomänyos Akademia, Nyelvtudomanyi Intezete, Budapest I., Szenthäromsag utca 2, H —1250 Budapest, Hungary Aditi Lahiri Max-Planck-Institute for Psycholinguistics, Wundtlaan 1, Postbus 310, NL —6500 AH Nijmegen, The Netherlands Maria Manoliu-Manea Department of French and Italian, Sproul Hall, University of California, Davis, California 95616, USA Jaap van Marie Koninklijke Nederlandse Akademie van Wetenschappen, PJ. Meertens-Instituut, Diabetologie, Volkskunde en Naamkunde, Keizersgracht 569-571, Postbus 19888, N L - 1 0 0 0 GW Amsterdam, The Netherlands Willem Meijs Engels Seminarium, Universiteit van Amsterdam, Spuistraat NL —1012 VT Amsterdam, The Netherlands

210,

Yves-Charles Μ or in Departement de linguistique, Universite du Quebec ä Montreal, Case postale 6128, Succursale "A", Montreal, P.Q. H3C 3J7, Canada Wolfgang Mötsch Akademie der Wissenschaften der DDR, Zentralinstitut für Sprachwissenschaft, Prenzlauer Promenade 149 — 152, 1100 Berlin, German Democratic Republic Sergio, Scalise Dipartimento di Italianistica e Filologia Romanza, Universitä di Venezia, Dorsoduro 3246, Ca' Foscari, 1 — 30123 Venezia, Italy Michael Shapiro Department of Slavic Languages, Brown University, Box Ε, Providence, RI 02912, USA Andrew Spencer The Polytechnic of Central London, Faculty of Languages, 9 - 1 8 Euston Centre, London NW1 3ET, Great Britain

List of contributors

317

Irene Vogel Department of Linguistics, College of Arts and Sciences, University of Delaware, 46 E. Delaware Avenue, Newark, Delaware 19716, USA Beatrice Warren Stockholms Universitet, Engelska Institutionen, S —106 91 Stockholm, Sweden Wolfgang U. Wurzel Akademie der Wissenschaften der DDR, Zentralinstitut für Sprachwissenschaft, Prenzlauer Promenade 149 — 152, 1100 Berlin, German Democratic Republic Wiecher Zwanenburg Vakgroep Romaanse Taalen en Kulturen, Rijksuniversiteit te Utrecht, Kromme Nieuwegracht 29, NL —3512 Utrecht, The Netherlands Arnold Μ. Zwicky Department of Linguistics, The Ohio State University, 204 Cunz Hall, 1841 Millikin Road, Columbus, Ohio 43210-1229, USA

m

Eung-Do Cook * Keren Rice

m

Athapaskan Linguistics

m

Current Perspectives on a Language Family

m

1989.15.5 χ 23 cm. VIII, 645 pages. With 1 map. Cloth. ISBN 311011166 7 [Trends in Linguistics. State of the Art Reports 16)

m m

m m

m m

m m

m m

This volume represents an attempt to show the present state of the art in the study of this group of Native American languages spoken in a large part of Alaska, as well as on the California coast and in the American Southwest (including such languages as Navajo, Apache, and Hupa). These languages are characterized by a particularly complex verb morphology. The lengthy introduction by the editors gives a general overview of areas that have been important in the field of Athapaskan studies in the past 25 years, and provides the reader with the context in which the following contributions can be seen. The papers themselves deal with diachronic linguistics, phonology and morphology, syntax, discourse and ethnolinguistics, and have been prepared by leading scholars in the field.

m

m m

mouton de gruyter Berlin · New York

ra

George Horn

im] Lexical-Functional Grammar 1983.14,8 x 22,8 cm. IX, 394 pages. Cloth ISBN 90279 3169 0 (Trends in Linguistics. Studies and Monographs 21)

[m

\m mi

ι

m ΐγγΐ

m IYIJ

[m ΓΤΊ |γγι mi

\m\

[mi

The analysis outlined in this monograph is formulated in the context of the major developments in linguistic theory stemming from the proposal of the so-called Lexicalist Hypothesis by Chomsky. The most significant product of linguistic research during this period has been the development and expansion of the lexical component and consequent reorganization and reformulation of the rules of the transformational-generative model, in which this component has been assigned many of the tasks formerly associated with the syntactic component. More recently, various counterproposals to Chomsky's analysis have been suggested. Perhaps the most important of these was developed by Bresnan, the key feature of which is the virtually complete reduction of the syntactic component. This work is an attempt to extend and reformulate certain of Bresnan's and Chomsky's ideas, combining the basic organization of Chomsky's model, in which lexical and non-lexical operations are clearly distinguished, with a non-syntactic account of bound anaphora, control, and NP movement phenomena. The proposed model provides a framework in which universal generalizations can be captured, and language variation can be accounted for without the complex machinery of Chomsky's current analysis, at the same time maintaining distinctions that are obscured in Bresnan's purely lexical analysis.

mouton de gmyter Berlin · New York · Amsterdam