The First Glot International State-of-the-Article Book: The Latest in Linguistics [Reprint 2014 ed.] 9783110822861, 9783110169546

Citation preview

The First Glot International State-of-the-Article Book

W G DE

Studies in Generative Grammar 48

Editors

Harry van der Hulst Jan Köster Henk van Riemsdijk

Mouton de Gruyter Berlin · New York

The First Glot International State-of-the-Article Book The Latest in Linguistics

edited by

Lisa Cheng Rint Sybesma

Mouton de Gruyter Berlin • New York

2000

Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter G m b H & Co. KG, Berlin.

The series Studies in Generative Grammar was formerly published by Foris Publications Holland.

© 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

The first glot international state-of-the-article book : the latest in linguistics / edited by Lisa Cheng, Rint Sybesma. p. cm. - (Studies in generative grammar ; 48) Includes bibliographical references. ISBN 311-016954-1 (cloth : alk. paper) - ISBN 311-016953-3 (pbk. : alk. paper) 1. Linguistics. 2. Linguistics-Bibliography. I. Cheng, Lisa Lai Shen. II. Sybesma, R. P. Ε. III. Series. P121 .F55 2000 410-dc21 00-058924

Die Deutsche Bibliothek —

CIP-Einheitsaufnahme

The first glot international state-of-the-article book : the latest in linguistics / ed. by Lisa Cheng ; Rint Sybesma. - Berlin ; New York : Mouton de Gruyter, 2000 (Studies in generative grammar ; 48) ISBN 3-11-016954-1

© Copyright 2000 by Walter de Gruyter G m b H & Co. KG, 10785 Berlin. All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced 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. Printing: Werner Hildebrand, Berlin. Binding: Lüderitz & Bauer GmbH, Berlin. Printed in Germany.

To Teun Hoekstra (1953-1998) in memory

Table of Contents

Preface

ix

The function of Functional Categories Teun Hoekstra

1

Control in GB and Minimalism Norbert Hornstein

27

Ergativity: A perspective on recent work Alana Johns

47

When verb phrases go missing Kyle Johnson

75

Topic and focus Henriette de Swart & Helen de Hoop

105

Recent research on the acquisition of L2 competence: Morphosyntax and argument structure Lynn Eubank & Alan Juffs

131

Model Theoretic Syntax Tom Cornell & James Rogers

171

The rise of Optimality Theory Luigi Burzio

199

Acquisition of phonology Paula Fikkert

221

viii

Table of Contents

Tone: an overview San Duanmu

251

The phonology-morphology interface Geert Booij

287

Metrical phonology Harry van der Hulst

307

Generative metrics Curt Rice

329

One phonology or two? Sign language and phonological theory Wendy Sandler

349

Semantic change: An overview Elizabeth Closs Traugott

385

Affiliations and addresses

407

Preface

The articles — State-of-the-Articles — in this book give an overview of 15 different topics in the field of linguistics. Each of them outlines the major developments of the past years in their area. What were the main issues five or ten or fifteen years ago? What hypotheses have been put forth over the years? What solutions have been proposed for what problems and which of those solutions, and which of the problems, are still with us? Reviewing and evaluating a huge body of literature, each paper summarizes the main contributions made, and ends with an overview of the most salient outstanding issues. An important part of each article is an extensive up-to-date bibliography. An earlier version of each contribution to this book appeared as a Stateof-the-Article in the first two volumes oiGlot International (1995-1997). All have been revised; depending on the field and the time that has lapsed since the original conception, more or less radical revisions were made. In all cases the bibliography has been updated. We are extremely sad that Teun Hoekstra is no longer with us and that he never got to see this book appear. We thank all the authors and Guido Vanden Wyngaerd for their cooperation. Lisa Cheng Rint Sybesma

The function of Functional Categories Jeun Hoekstra Over the last decade the concept of Functional Category has found its way into generative grammar. Instead of major constituents being endocentric projections of the major lexical categories (Ν, A, V and P), they are now taken as more complexly organized structures, in which the lexical projection is dominated by a number of functional categories. These host purely grammatical properties in part associated with formal properties of the lexical projections. The current literature abounds with proposals concerning such Functional categories (F-cats, as opposed to L-cats). I shall try to provide an overview of some lines of thought that are relevant to this exciting development.

1. The introduction of F-cats The primary motivation for the introduction of F-cats was the insufficient number of positions provided by simple phrase structure configurations. If a clause is a projection of V, there should be a unique position of V. However, in V2 languages, finite verbs in root-clauses occur in a different position than in embedded clauses. Den Besten (1979) proposed that this results from movement of the finite verb in root-clauses, to a position identified as C (complementizer), which would capture the complementarity of V2 (the finite verb in second position in root contexts) and complementizers (in embedded contexts). What is the nature of C and how does C relate to the rest of the clause? Clearly, C is not a L-cat, but rather hosts closed category items. Under a restricted theory of the internal structure of L-projections, there is no room for this element. It is assumed that C is outside the L-projection of V, and that C is a head, taking the projection it combines with (lexically headed by V) as its complement. If we make the further assumption that phrases are built up in terms of two head-determined relations, viz. head-complement and head-specifier, and we extend this assumption to C, the structure of a clause with a complementizer would then be as in (1): (1)

Spec C [χρ

V...]

where XP is the complement of C. We can then assume a concept of structure preservation: heads move to head-positions, XPs to XP-positions. The major properties of V2 now follow: V moves to C and some XP moves to its Spec, yielding the V2 order.

2

Teun Hoekstra

The same reasoning may be used to argue that XP in (1) is not simply VP, but must contain further internal structure. Pollock (1989), extending proposals by Emonds (1978), applies these considerations to the word order differences between French and English, illustrated in (2): (2)

a. John often kisses Mary/*John kisses often Mary b. *Jean souvent embrasse Marie/embrasse souvent Marie

The fact that the finite verb precedes the adverb in French, but follows it in English can be accounted for if the structure available to both is as in (3), with V-movement taking place in French, but not in English: (3)

Spec

I

i

ADV

V object I

In fact, Pollock shows that within XP in (1) there should be minimally two head positions external to the lexical projection of V, as infinitives may occur on two sides of adverbs in French, but not to the left of pas, the position obligatorily held by finite verbs. He identifies these positions as Tns and Agr, but I shall label them F : and F2. The different positions of the verb visä-vis pas and souvent in (4) can then be expressed by saying that a finite verb must occur in Fj, while infinitives occupy either F2 or the base position. (4)

a. b. c.

Jean ne (lit) pas (*lit) souvent (*lit) un livre ne (*lire) pas (lire) souvent (lire) un livre ...Fj pas F 2 souvent [ VP V object]

On the basis of structure preservation, then, we are led to the conclusion that a clause is not simply a projection of a lexical head V, but rather is an extended projection, in the terms of Grimshaw (1991), here formulated as [F*VP], with F* some number of F-cats associated with V. The same reasoning can be applied to nominal phrases: are these simple NPs, or do we find reason to think that nominal phrases too are of the form [F*NP], with F some set of F-cats associated with N? The literature indicates that the latter conclusion is correct. Early proposals to this effect were made by Szabolcsi (1984). Abney (1987) argues that determiners should be taken as F-cats taking NP as its complement. Ritter (1991) argues that in addition an F-cat Number should be considered part of the nominal structure. Valois (1991), Cinque (1993), Szabolcsi (1994), Longobardi (1994) among many others show the relevance of the postulation of various F-cats internal to nominal phrases. For instance, in Italian we find side by side il vecchio Gianni 'the old John' and Gianni vecchio 'old John', an alternation which could be captured by assuming head movement of the noun Gianni to the position of il, crossing the adjective and thereby

3

The function of Functional Categories

accounting for the absence of a determiner in that derived order. The absence of a determiner in the Hebrew construct state in (5a) can equally be captured if N-to-D movement is called upon in its analysis, as indicated in the structure in (5a'). In the free state construction in (5b), the use of a determiner is required in order to indicate definiteness. (5)

a. a'.

b.

beit ha-mora house the-teacher [ D [ XP .... [ N P

....N ....]....]

ha-bayit §el ha-mora the-house of the teacher

Valois and Cinque show that the order of adjective and noun in English differs from that in French in a way parallel to the different order of adverb and V, cf.: (6)

a. b.

the frequent destruction of Rome la destruction frequente de Rome

The parallel between adj-N vs. N-adj and adv-V vs. V-adv raises a further question: is there a connection between the F-cats in the nominal system and those in the verbal system? We return to this question below. From [F*NP] and [F*VP] it is a little step to generalize to [F*LP], with L any lexical category. And indeed, also for APs and PPs proposals can be found to the effect that they equally involve extended projections. One obvious candidate is an F-cat Agr, in terms of which the agreement on adjectives and prepositions can be expressed. Consider the Hungarian forms in (7): (7)

a.

(en) nek-em I tO-lSG 'to me' (te) nek-ed you to-2SG 'to you'

b.

az (en) haz-am the I house-lSG 'my house' az (te) haz-ad the you house-2SG 'your house'

As the examples indicate both nouns and prepositions show agreement. The structure of the prepositional phrases might be as in (8), with head-movement of Ρ to Agr and XP-movement of the complement of Ρ to the SpecAgr. The structure of the nominal phrase may in this respect be entirely parallel.

4 (8)

Teun Hoekstra [^.p Spec Agr erij -em

t

[χρ P/N compl] nek/haz t-

I

Abney (1987) and Corver (1990) propose that APs also involve a Degree functional category. Koopman (1993) proposes a whole set of F-cats internal to prepositional phrases. So, the conclusion indeed seems to be that all lexical projections involve a set of F-cats. These F-cats can be used to account for language internal word order patterns, for word order differences between languages, as well as for various morpho-syntactic features associated with L-cats. Indeed, there seems to be a kind of consensus for the idea that all linguistic variation is variation in F-cats.

2. A theory of F-cats Although the introduction of F-cats provides us with new insights, gives flexibility in the analysis of variation within and across languages, and allows us to formulate restrictive hypotheses on possible phrase structure configurations, their theoretical status leaves many questions unresolved. To mention a few of these: I. II. III. IV.

does each language have the same set of F-cats? does each category have the same number of F-cats? what is the relationship between the F-cats of the different L-cats? do F-cats form a homogeneous set?

These questions arise because there is as yet no theory of F-cats. While L-cats may be considered as defined and hence limited in terms of the substantive features [±N] and [+V], we have no idea, let alone a restrictive theory, of F-cats. This is illustrated by Pollock's (1989) claim that Fx and F2 in (4c) are Tns and Agr respectively, while Belletti (1990) identifies them rather as Agr and Tns. This illustrates that the postulation of an F-cat may be motivated entirely on the basis of a heuristics derived from considerations of the principles of structure preservation. In Section 1, this was exemplified as far as positions of heads are concerned: on the assumption that Move α may not create head positions, each occurrence of a head where it is not expected, requires the presence of a head position where the displaced head has moved to. If X-bar theory requires that each head projects a Spec-Head-Complement configuration, the presence of such a head also makes available a Specifier (obviously, the XP from which the head is moved constitutes the complement of the head to which the extracted head

The function of Functional Categories

5

has moved). Conversely, if we find an XP in a displaced position, this requires a Specifier position or at least a maximal projection where this XP can have been attached, and hence such a situation equally requires a head for this projection. This may be illustrated with the example in (9): (9)

a. b.

John most probably is not ill [χρ John Χ [γρ most probably [γρ is not [ VP e ill]]]

Given our earlier assumptions, John is in the Spec of some XP, while is in some head position outside of VP, as it occurs to the left of not. What is the phrase structural position of most probably? When we make the assumption that adverbs are adjoined to maximal projections, it must be adjoined to some FP other than the XP hosting John in its specifier, i.e. the structure should minimally contain two F-cats outside VP, as indicated in (9b). This example thus shows how XP (and hence X) is motivated by the requirement that John be in a Specifier, while YP (and hence Y) is motivated by the requirement that there is a projection to which the adverb can adjoin and the head is can be moved to. But, it is not immediately evident what the nature of these F-cats is, and indeed, there are numerous proposals regarding the nature of various head-positions which are argued to exist on the basis of such reasoning from the point of structure preservation. To mention a few examples from the domain of verb or clause related F-cats, there is Subject, Indirect Object and Object Agr, C, Tns, Asp, Mood, Focus, Topic, Negation, Modality, Polarity, Existential, Voice, clitic(s) and so on. It has been proposed that Agr itself should be split up in Person, Number and Gender. This is often aptly referred to as the explosion of F-cats. Evidently, I shall not try to establish whether each of these suggestions has sufficient motivation, nor what the status of each of these is with respect to the questions I started this section out with. I merely mention these examples to illustrate the theoretical vacuum that exists in this respect.

3. Linguistic

variation

3.1. Presence or absence of F-cats Let us turn to linguistic variation. When F-cats were first introduced within the framework of generative grammar, their availability was exploited to account for linguistic variation. There are various ways in which this can be done. An early approach, illustrated by the work of Kuroda (1988), exploited the idea that certain differences between English and Japanese could be accounted for if it was assumed that while English grammar has an F-cat IP, Japanese has not. In this view, UG would make

6

Teun Hoekstra

available a set of F-cats, from which particular languages could select. Presence or absence of F-cat X thus constitutes a parameter of UG, to be set by the language learner on the basis of the evidence for X available in the input. Clearly, such a position quite brutally stipulates variation, specifically if we take our current understanding of F-cats into consideration. Moreover, the evidence available to the child in determining the presence of a F-cats may be of a rather subtle nature, as we illustrated above for English, which threatens the learnability of such parametric options. Apart from these theoretical considerations, the empirical basis for the Kuroda approach has been called into question as well. 3.2. The licensing

role of F-cats

More recent proposals in the literature also cast doubt on the validity of an approach to linguistic variation in terms of absence or presence of F-cats. Specifically, various hypotheses concerning individual F-cats have been put forth concerning their function in the grammar. Such hypotheses are of an entirely different nature than the heuristic reasonings based on structure preservation: while the latter concern the basic phrase structural inventory, the former relate to the role of F-cats in licensing the phrase structural relations they are involved in. Let me illustrate this with two examples. The first concerns the role of Neg, taken as an F-cat. Consider the following data on Italian Neg-doubling (note that I deliberately ignore the discussion on the question whether expressions like in (lOa-c) should be regarded as negative expressions, or rather as negative polarity items): (10)

a. b. c. d. e.

Non e arrivato nessuno NEG is arrived no one Gianni non ha visto nessuno Gianni NEG has seen no one Gianni non ha visto mai il San Marco Gianni NEG has seen never the San Marco Nessuno e arrivato no one is arrived Mai ha visto il San Marco never has seen the San Marco

The negation is doubly expressed in (lOa-c), i.e. both by the element non and another negative expression, nessuno in (lOa-b), mai in (10c). In (lOd-e), the negative element non is absent, while the "second" negative expression holds a position in front of the verb. This pattern of negation can be explained if we assume that non is the head of a F-cat NegP, and that non may be absent if a negative element occurs in its specifier. Zanuttini

The function of Functional Categories

7

(1991) and Haegeman & Zanuttini (1991) suggest that what happens overtly in (lOd-e), may occur at LF in (lOa-c), as a consequence of a condition, the Neg-criterion, which requires negative heads and negative phrases to be in a Spec-Head agreement relation, either at S-structure or at LF. They suggest that the Neg-criterion is part of UG. If so, no language would be allowed to have negative XPs without also having a Neg-head, as a consequence of UG. Assuming that all languages have negative XPs, then, UG in fact imposes Neg as an F-cat for each language. A similar case can be made for another F-cat which features in recent work, viz. the category of AgrO (cf. Chomsky 1989; Kayne 1989). As is wellknown, some languages show overt object agreement, while others do not, which might be used as evidence of a parametric option formulated in terms of absence or presence of the F-cat AgrO. However, the hypothesis has been put forth that AgrO is involved in the Case-licensing of the object, not only in languages manifesting overt object agreement, but also in languages in which AgrO has no morphological manifestation. Structural Case licensing, then, would uniformly be a matter of Spec-head relations in Agr-projections. If correct, the Case-requirement being part of UG, variation between languages cannot be expressed in terms of absence or presence of the relevant F-cat: AgrO is part of the grammar of each natural language by necessity (cf. Lasnik 1993 and references cited there for empirical arguments in favor of AgrO in English). If we now assume that similar licensing roles are associated with each type of F-cat, it will be clear that the grammar of each language has the full set of F-cats prescribed by UG. Whether this strong position can be maintained is a matter of further research. There is a clear danger here of equating presence of a certain meaning with the presence of a F-cat. To give one example, recent research has revealed that indefinite DPs allow for a generic or an existential interpretation, the latter obtaining if the indefinite is, broadly speaking, more deeply embedded in the structure (cf. Kratzer 1988; Diesing 1993). A mechanism of existential closure is called upon to capture this interpretive effect; more specifically VP might be the domain within which indefinites are in the scope of existential closure. This semantic effect may easily be "syntacticized" by postulating a F-cat "existential", dominating VP, and attracting indefinites in its Spec at LF or S-structure (cf. Kondrashova 1994). While such an option may turn out to be correct, it seems to me that it should be motivated on grounds other than the interpretive effect itself which it seeks to capture. Suppose that a theoiy can be formulated as to which licensing roles UG has relegated to F-cats, and that these licensing roles are uniformly required for each language, so that linguistic variation cannot be a matter of choice of F-cats. If it is correct, as has been proposed by various people, that linguistic

8

Teun Hoekstra

variation apart from the arbitrary pairing of form and meaning of lexical elements resides in the F-cats, all parametric options must be reduced to properties of their F-cats, other than absence or presence. We should ask ourselves, then, along which lines properties of F-cats may vary. Given the computational format of X-bar theory, there can be no more than three dimensions of variation: (a) the dimension of the head-complement relation; (b) the dimension of the inherent properties of the F-head itself; (c) the Specifier. Proposals of each type can be found in the literature and it may be instructive to consider some of these. I shall not go into possible variation with respect to the Specifier; in as far as there might be variation in this respect, this would be a matter of its relation to head-properties. 3.3. Variation in complement

selection

With respect to the head-complement relation, variation is conceivable in terms of categorial selection. Zanuttini (1991), Ouhalla (1990) argue that languages may vary with respect to the relative order of Agr, Τ and Neg. Similarly, while Pollock (1989) argues that Τ dominates Agr, Belletti (1990) argues for the opposite order. Much of the motivation for such proposals derives from the mirror principle (Baker 1985), according to which the order of affixes should reflect the syntactic hierarchical structure. Consider the examples in (11): (11)

a. b.

Les enfants arriv-er-ont demain the children arrive-TNS-AGR tomorrow Ad-y-segh Moha iharkusn dudsha will(TNS)-3MS(AGR)-buy Moha shows tomorrow

French Berber

As these examples show, the affix order of Tns-Agr on V French is the mirror image of the order in Berber: while Agr is peripheral in French, Tns is peripheral in Berber. Under the mirror principle, this would require that Agr dominates Tns in French, while Tns dominates Agr in Berber. Ouhalla suggests that this parametric difference tends to correlate with VSO (Tns higher) versus SVO (Agr higher). This correlation would be explained if we make the following two assumptions: (a) the subject uniformly occupies the SpecAgr-position; (b) the verb moves up to the highest F-cat. V-raising to Tns in Berber then takes the verb across the subject, yielding VSO order. An even more dramatic difference holds between Berber and Turkish, as illustrated in (12) (taken from Ouhalla 1990): (12)

a.

John elma-lar-i

ser-me-di-0

John apple-PL-ACC like-NEG-PAST-AGR

Turkish

The function of Functional b.

9

Categories

Ur-ad-y-xdel Mohand dudsha NEG-will-3SG-arrive Mohand tomorrow

Berber

The order of the functional morphemes Neg, Tns, Agr in relation to the stem appears to be the inverse. Again, this could be captured in terms of differences in complement selection, as proposed by Ouhalla (1990). However, such an approach seems unlikely if the licensing role of F-cats is taken into consideration. We would expect that the relation between Neg and Tns is inherently given by the functional roles these play (cf. Laka 1990). Similarly, if AgrS is involved in checking the nominative Case of its Specifier, and this Case is made available by Tns, one would assume their relative order to be fixed. If that is correct, the variation in order of the relevant morphemes requires a different approach. Consider the overt expression of negation. Languages differ in this regard: some languages have an adverbial element, comparable to French pas, while others have a clitic-like element, sitting on the finite verb, like French ne. Still others express negation with a specific negative verb (cf. Arabic lays-at), while in still others, we find a morpheme internal to the verb complex, as in Turkish. Such morpho-syntactic differences can be explained in a different way than in terms of complement selection, e.g. the difference between ne and pas comes about by the finite verb skipping pas which holds the SpecNegP, while taking along the head ne in successive head movement. The Turkish case may involve a negative morpheme in the verb, which is checked by passing through a Neg-head, which itself has no morphological manifestation. The Berber case, finally, might involve an independent negative verb, itself inflected for Tns and Agr, but without raising of the lexical verb. At this point, this can be no more than a suggestion, but the fact that the verb stem follows the functional morphemes might be indicative of the correctness of this assumption, if it is correct, as Kayne (1994) argues, that heads uniquely adjoin to the left.

3.4. Variation in formal

properties

Along with the perhaps optimistic assumption that all languages share the same set of F-cats, we thus make the even further assumption that selectional relationships among them are fixed by UG. Let us then move on to possible variation in terms of properties of the F-cats themselves. As indicated in the previous section, it is important to distinguish between presence of an F-cat (negation) and the way in which it is realized. Does a certain overt element instantiate the head-position? In the discussion of negative doubling in Italian above, we already indicated that a head may be empty or overt, depending on whether or not its specifier is filled. This is

10

Teun Hoekstra

not only true for Negation; the same situation obtains with interrogative complements. Consider the example in (13): (13)

a. b.

John wondered *(whether/if) he should read this book John wondered which book (*whether/if) he should read

As (13) illustrates, a ίϋΛ-phrase in SpecCP and an overt [+wh] complementizer are in complementary distribution in English (although, of course, it may be misguided to call whether a head: rather, it may be in SpecCP, thereby identifying the u;&-features of the head under Spec-head agreement (cf. Kayne 1991). The strict complementarity then concerns the availability of Spec, an analysis which certainly complicates the issues. Whether should, under this perspective, be considered a kind of expletive, similar to do in do-support). And again, we find variation in this respect between languages. In Dutch dialects (or idiolects), there is no such strict complementarity, i.e. presence of the Dutch counterpart of whether / if in (13b) is optional. Another point of variation concerns the obligatoriness of S-structure movement of an available w/i-phrase: movement is obligatory in Dutch and English, but spoken French allows ιιΛ-in-situ at S-structure, with the requirement that the w/i-complementizer be overt: (14)

a. b.

II se he himself II se he himself

demande ask demande ask

*(si) Jean a lu quel livre whether Jean has read which book quel livre (*si) Jean a lu which book whether Jean has read

The complementarity can be understood in the following way: the features of some F-cat X must be identified, either through lexicalization of the head, or through a Spec-Head agreement relation. Another point of variation was illustrated with verb placement in French and English. While maintaining that the same F-cat positions are available in these languages, Pollock suggested that French finite verbs move further up than the verb does in English. He appeals to a notion weak vs. strong, saying that the F x position in (4c) above is strong in French, but weak in English. The precise nature of the weak/strong distinction is less relevant in the present connection. Pollock appeals to a notion of opacity for thetamarking created by attachment of a theta-assigning category to a weak F-cat. This allows him to account for the fact that auxiliaries, which do not assign theta-roles, (optionally) move to weak F-cats, as illustrated in (15): (15)

a. b.

John has often kissed Mary n'avoir pas eu unejeunesse heureuse, ... NEG'have not had a childhood happy

The function of Functional Categories

11

It is to be expected that the notion of weak/strong is related to the same notion which is relevant in determining the overt realization of F-cats such as Neg and Comp discussed above, i.e. that strength relates to (phonological) visibility. Chomsky (1992, p. 44) makes this explicit, saying that Arabic SVO vs. VSO correlates with strong versus weak agreement. The idea here is that if AgrS is strong, it requires its specifier to be filled, thus attracting the subject, resulting in SVO-order. Evidence that this may indeed be the case is put forth by Platzack & Holmberg's (1989) investigation of V-movement in the Scandinavian languages. Translating their results into our current terminology, we can say that V-movement to the topmost F-category takes place if there is an overt manifestation of person agreement. It is the overtness of person agreement which apparently contributes to the strength of the relevant F-cat, probably Agr. Note that this squares well with Kayne's (1990) analysis of English s in the present tense as reflecting number, rather than person. There would then be a correlation between lack of V-raising as in mainland Scandinavian and English, and absence of person agreement in the inflectional system. If there is such a correlation between morpho(phono)logical richness and movement, however, a strong typological prediction is made: the overtness of agreement etc. in V-final languages should be less than it is in V-initial languages. Comparison of the Berber and Turkish examples above does not immediately support this prediction. The historical development of English from OV to VO does not speak in favour of the prediction either, but clearly a wider range of facts should be taken into consideration. It should be noted that the two types of cases we have mentioned so far are fundamentally distinct: while identification of the head-features via the specifier allows the head to remain empty in the case of Neg and C, overt visibility of head-features triggers movement to the specifier in the case of Agr. It is unclear why these cases should behave in such opposite ways. I return briefly to this question below. Another proposal concerning variation in head-properties concerns the notion of government. The alternation in Arabic in SVO vs. VSO order is captured (by Koopman & Sportiche 1991 among others) by a choice regarding the way Case is assigned: under government to the right, or under Spec-Head agreement, the latter option yielding SVO order with rich agreement, the former VSO order with default agreement. Rich agreement thus is a result of movement of the subject rather than a trigger, as under Chomsky's (1992) suggestion mentioned above. Chomsky in effect suggests that there is no such variation, and that Case checking rather is uniformly a matter of Spec-Head agreement. Related to this is the notion of directionality: while certain categories are

12

Teun Hoekstra

head initial, others are head final. The difference between the Scandinavian languages and the Germanic OV-languages is said to be explainable in these terms. Usually, one would say that the VP in Dutch/German is headfinal, while in Scandinavian it is head initial. If all parametric options are associated with F-cats, there is no directionality internal to lexical projections. Rather, the SOV-order in Dutch/German must follow from some F-cat selecting a complement to its left (either as a primitive property, or as a consequence of some additional relationship, e.g. Case assignment). Kayne (1994) argues that even this approach is misguided. Instead, he suggests that the X-bar schema is not only fixed with respect to the hierarchical ordering imposed, but also in its linear dimension, which, universally, is Spec-Head-Complement. Word order differences are then to be regarded as resulting from differences in the extent to which heads and XPs have moved up at S-structure (which is then equivalent to saying to the left). Assuming, for example, the direct object in both English and Dutch to be in the same position, the verb must have moved to a higher F-cat in English than it has in Dutch. This is consistent with Chomsky's (1992) minimalist program. This also assumes that languages are alike with respect to the F-cats they have, and that differences in word order are entirely determined by whether certain movements occur at S-structure (or, in his terminology, before Spell Out) or at LF. This itself is determined by strength of F-cats in the sense explained above. The resulting picture is that there is no room for variation either in terms of absence or presence of F-cats, nor in terms of selection, nor in terms of directionality.

4. Similarities

and differences between F-cats

Let us turn to the other questions mentioned at the beginning of Section 2. These have hardly begun to be addressed. Only the parallelism between the nominal system and the verbal system has been the subject of some discussion, most notably in Szabolcsi's work. We certainly find various types of parallels. The functional role of Tns and D for instance is claimed to be parallel in that Tns saturates the event-role of the verb, just like D saturates the R(eference)-role of the noun (cf. Higginbotham 1985; Longobardi 1994). Szabolcsi (1994) notes a parallelism between D and C in that the SpecDP appears to function like an operator type position in Hungarian: both SpecCP and SpecDP function as an escape hatch for A-bar movement. Just like clauses have C and Τ with an intricate relationship between them, nominal projections might have two D-type positions, with an Agr-head in between, as in (16):

The function of Functional Categories (16)

C D

Agr Agr

Τ D

13

V.... N....

We might in fact conjecture that the difference between Ν and V (and similarly for other categories) is a matter of functional superstructure, yielding a perspective that might be called functional determination of lexical categories. Obviously we are far from the possibility of seriously discussing the points of contact between the functional superstructures of the distinct lexical categories. Should aspect be considered the verbal analogue of nominal gender? Should directionality in prepositional phrases be considered the analogue of aspect or tense (cf. the difference between in and into and the infinitival tense marker to)? Is er in Dutch PPs like er in (therein=in it) in a Spec of some F-cat comparable to C (cf. Van Riemsdijk 1978)? These are all interesting questions to be pursued in further research, but it is a healthy sign that the theoretical framework invites us to raise such intricate and precise questions. Let us finally turn briefly to the last question, viz. whether the set of F-cats is homogeneous. This issue again has not received much attention in the literature, but we might distinguish between several types of F-cats in terms of a distinction between Α-type F-cats and A-bar-type F-cats, analogous to the distinction between XP-positions. Such a distinction was proposed by Roberts (1991) in the context of long head movement. In brief, the head movement constraint (HMC, cf. Travis 1984) requires that a head moves to the first head dominating it. Roberts appeals to Rizzi's (1990) revision of the ECP in terms of relativized minimality, which relativizes the concept of intervener in terms of the type of relationship to be established: X constitutes an intervener between Y and Ζ if X is of the same type as Y, and if relevant structural relationships hold. According to Rizzi, relevant types are A-positions, Α-bar positions and head positions. Roberts' proposal is that heads may equally be divided in A- and Α-bar types. The question that arises is which F-cats should be considered Α-heads and which A-bar heads. Li (1990) suggests that Α-heads assign theta-roles, while other heads are Α-bar. In this way, all F-cats would presumably be considered Α-bar heads. Li argues that this is the case for at least C, I and Neg. Roberts' idea is, however, that the A/A-bar distinction may differentiate between F-cats. A rather fundamental distinction in the domain of XPs is the distinction between operator positions and non-operator positions: SpecCP, the most obvious Α-bar position is such a typical operator position, while e.g. SpecIP typically is not (for discussion on the A-or-A-bar status of SpecCP, see Rizzi 1990; Vikner and Schwartz 1992; Zwart 1993; for a similar discussion on the status of SpecIP, see Diesing 1990 and Rögnvaldsson

14

Teun Hoekstra

and Thrainsson 1990). It would seem natural to extend this distinctive property to the domain of heads, certainly in view of the pairing of operator-head features and operators which the Neg-criterion mentioned above and the analogous ιυΛ-criterion, suggests. Consider the case of so-called Long Head Movement (LHM) discussed by Rivero (1991) in Bulgarian (and several other languages). The relevant facts are given in (17): (17)

a.

Az sum procel knigata I have+lSG read book+the b. Procel sum knigata read have+lSG book+the c. Az ne sum procel knigata I NEG have+lSG read book+the d. *ne procel sum knigata e. *procel ne sum knigata

Bulgarian is a pro-drop language. In (17b), the non-finite verb is moved across a finite auxiliary, an instance of LHM. (17d-e) show that the presence of Neg blocks such LHM. Assuming, as does Rivero, that the nonfinite verb is moved to C, we have the schema in (18): (18)

C ...AUX^ ... V f i n

The non-finite verb may be said to move to an Α-bar head, across an Α-head, without violating relativized minimality. If Neg intervenes, however, such LHM is blocked by relativized minimality, Neg being an intervening Α-bar head in the sense of relativized minimality. Again, whether such a distinction is relevant requires further study. I note here that the difference we encountered above between the licitness of an empty head if the Spec is filled, vs. the triggering of movement to the Spec if the head is strong in terms of, say, Person agreement, might be related to the A/A-bar distinction. Heads which may be left empty in function of the overt presence of a constituent in its Specifier, may be A-bar heads, which in the absence of such overt movement to their Specifier serve as scope indicators. Α-heads, on the other hand, do not indicate scope, but check off agreeing inflectional features, as in the case of subject agreement on the verb in its relation to the nominative Case of the subject. This brings us back to the question of potential differences in nature and number of F-cats associated with the different L-cats. It would seem that some of the F-cats found in the literature are not related to inflectional properties of an L-head (specifically the verb), but rather to sentence level properties such as focus, negation, and illocutionaiy force (question,

The function of Functional Categories

15

imperative etc.). Borrowing a term from Chomsky (1992), these F-cats may be called non-L-related (or Α-bar heads), contrary to those F-cats that are related to inherent properties of L-cats themselves, which are called L-related (or Α-heads). The "operator"-like F-cats may then be regarded as associated not to V per se, but to the "minimal" clause, which happens to be lexically headed by V. The set of L-related F-cats of V may then be limited and comparable to the set of L-related F-cats of N.

5. The role of F-cats in language acquisition Since all of syntax is basically regulated through (properties of) F-cats in the grammar, a very central question concerns the status of F-cats in the process of language acquisition. Pretty much the same questions arise in this area as in language variation (discussed in section three). So, according to one position, held by Radford (1990) and several other researchers, the initial state of the language acquisition process is characterized by the absence of F-cats. This would virtually mean that there is no syntax at the earliest stages, as indeed Lebeaux (1988) and Radford (1990) maintain. Under this view F-cats become available in one of two ways. Either they are subject to maturation — the strongest position in this regard is developed by Tsimpli (1996), who proposes that UG contains a separate functional module that matures around the age of 2 years — or alternatively, the acquisition of F-cats is input-driven by a process of lexical learning. A particular F-cat is postulated in the child's grammar only after it has been learned on the basis of the input. If no evidence is available in the input, the F-cat will never become part of the grammar. This implies that grammars may vary in terms of absence or presence of F-cats. This view has long been defended by Clahsen and his associates (cf. e.g. Clahsen & Penke 1992), and more recently receives support from adherents of the minimal projection idea (cf. Grimshaw 1997; Speas 1994). Within these frameworks the acquisition of F-cats becomes a rather language-particular enterprise. I will refer to all of these approaches, which differ from each other in various ways, as the Initial No F-cats position (No-Fcats, for short). Opposed to these views is the strong continuity approach, which holds that the child's grammar comes fully equipped with all relevant F-cats. Obviously, this view is compatible with or identical to the view that adult grammars do not vary from each other in terms of the number, selection and role of different F-cats. This strong view, which is often referred to in the language acquisition debate as the 'Full Clause Hypothesis' (FCH) finds its defenders among Wexler (1994), Boser, Lust, Santelmann & Whitman (1992), Hyams (1996) and Rizzi (1994) among others.

16

Teun Hoekstra

It would obviously take us too far afield to look at all the relevant discussion that has taken place in this connection. The debate is extensive and interesting. Let me briefly review some of the kinds of evidence that is brought to bear on the debate between adherents of the No-F-cats position and the FCH. For proponents of the No-Fcat position, not every occurrence in child utterances of morphemes that would instantiate an F-cat in the target grammar necessarily counts as a counterexample to their position. One might claim, for example, that the complex form is learned as an unanalyzed unit. Within this approach the criterion level for attributing functional knowledge to the child can be set quite high. For example, Clahsen (following a tradition started in Brown 1973) requires 90% correct usage of forms in a paradigm before attributing the child with knowledge of an AGR category. It is also possible to argue that the phrase structural position of a particular morpheme, while a F-cat head in the target grammar, is different in the child's grammar. This is what Radford argues is the case for w/i-phrases and negation, for example. So, since there is no CP category in the child's grammar, wA-phrases are assumed to be adjoined to VP, as is not, which equally doesn't instantiate a F-cat NEG. So, while the No F-cat position in principle predicts that there should be no expression of F-cats in early grammar, the theory can protect itself against counterexamples with a number of reasonable auxiliary assumptions. Turning now to the FCH, we expect the full range of F-cats to be available. Hence, the absence of F-cat material that is required in the target grammar constitutes refuting evidence in principle. But, of course, the FCH also has ways of protecting itself against too easy a refutation. For instance, claiming that there is an Agr category in the child's grammar does not imply that the child will evidence the full range of the paradigm in its usage of Agr. And there are more sophisticated ways of accommodating differences between adult and child grammar, without giving up the FCH. So, while the empirical predictions of the two hypotheses are clear, and are both clearly refuted at first sight, the level of argumentation has to run somewhat deeper than sheer observation of morphemes in child utterances. A relevant point to make here is that one has to be careful with observations: divergence with adult grammar catches the eye more easily than compliance with it. In order to get to correct assessments one should not therefore work with isolated examples, but rather do careful quantitative studies over corpora of child language. Fortunately, a lot of data is currently available through Childes (MacWhinney & Snow 1985), and many careful quantitative studies have been and are being performed, especially by adherents of the FCH. These studies have brought to light various cases in which the syntactic correlates of a particular F-cats are systematically found, while the morphological manifestation of the relevant F-cat may be

17

The function of Functional Categories

quite irregular or sporadic. Let me give two examples that are typical in this regard. The first concerns the raising of finite verbs across negation pas in French, discussed above. In the adult grammar, finite verbs move across pas, while non-finite verbs do not. Pierce (1992) did a quantitative study of the data of two French children, and finds a consistency between finiteness and position as displayed in Table 1. The numbers clearly show that they behave in accordance with adult grammar, even though their use of the agreement paradigm is very limited at the time. Table 1. Finiteness and position of negation in French (Pierce 1992) finite verb

non-finite verb

neg V

11

77

Vneg

185

2

Nathalie and Daniel Corpora, Lightbown (1977) On the basis of this data we have to conclude that the children's grammar has the category Agr (or whatever is the proper characterization of the relevant F-cat), and hence the syntactic effects of this F-cat. The fact that there is not a 90% correct usage of the full paradigm is completely irrelevant. The child knows the difference between a finite form and a non-finite form. We see more or less the same thing in the second example I want to discuss: finite verb movement to C in Germanic. As is well known, the Germanic languages (except English) display the verb second phenomenon: the finite verb of the main clause is found in second position. Table 2 shows that children are aware of this V2-requirement, even though they also allow independent sentences without finite verbs, so-called root infinitives (RIs). Clearly, then, the F-cats responsible for the V2-syndrome are present in the child's grammar. Table 2. Finiteness and position of verb in Andreas (Poeppel & Wexler 1993) finite verb

non-finite verb

V2-position

197

6

Final position

11

37

Andreas corpus: Wagner (1985); CHILDES, MacWhinney & Snow (1985) As noted above, this table also demonstrates another property of child language: the use of RIs. These RIs have often been taken as evidence for

18

Teun Hoekstra

the No-Fcat position, the idea being that RIs are the result of absence of Tense and Agreement. However, it is not at all evident that RIs can be taken to support the No-Fcats position. First, the RI-stage does not constitute a stage prior to the development of finite verbs. Rather, RIs and finite verbs occur side-by-side, as Table 2 shows. Second, the RIs themselves cannot be taken as forms without F-cats, since there is minimally the infinitival morpheme. Third, the RI-phenomenon shows a large number of systematic properties that might be expected on the basis of its infinitival status, for example its position following negation in French and its sentence final position in the V2 languages (cf. tables 1 and 2). Finally, the RI-phenomenon is not available in all child languages: whether it occurs or not depends on the morphosyntax of the target grammar. The reader is referred to Hoekstra & Hyams (1998) for extensive discussion of this point. An important result that emerges from these various studies is that there is no such thing as a "universal child language". Under the No-Fcat position, however, one does expect a universal child language, namely a language determined by the absence of F-cats. Recall that under the standard view, variation between languages results from properties of F-cats. And indeed, child languages vary from each other precisely in ways that are expected on the FCH. Children appear to converge on the morphosyntactic properties of their target language very early, and without many mistakes (Hoekstra & Hyams refer to this as Early Morphosyntactic Convergence, EMC). Of course, we also observe differences between the child's language, hence grammar, and the adult's. Within the camp of the proponents of the FCH there are two strategies to explain these differences. One strategy is Rizzi's theory of truncation. The fundamental difference between adult grammar and child grammar is that the adult is subject to the CP-axiom, which says that every sentence must be a CR This axiom is subject to maturation. As long as it has not matured, a sentence may project to any lower projection in the clausal hierarchy. For example, on this view, the RI-phenomenon results if a sentence is a Full Clause, which has been truncated under TR This has various consequences, which I shall not discuss (see Rizzi 1994; Haegeman 1994). The alternative is an approach in terms of optional underspecification of F-cats, advocated by Hyams (1996). According to this strategy, differences between adult and child language comes about as a result of optional underspecification of particular F-cats. According to Hyams, this is not a matter of grammar, but a consequence of the interaction between grammar and pragmatics, where the pragmatic requirements imposed on the adult language exclude the use of certain grammatical options which are available to the child.. The study of the role of F-cats in language acquisition is fascinating, as

19

The function of Functional Categories

language acquisition data may shed light on the kinds of theoretical questions that came up in Section 3. It is precisely in this context that many linguists interested in language variation have been drawn to the field of acquisition.

6. F-cats in the Minimalist

Program

The introduction of F-cats was a major step in the development of the Minimalist Program (Chomsky 1992). It would take us too far afield to provide a detailed overview of the program in all its aspects. I shall therefore attempt to discuss those features of the Minimalist Program that are clearly dependent on the existence of F-cats in the grammar. The basic assumption of the Minimalist Program is that there is a unique computational system that generates from the lexicon pairs of PF (Phonetic Form) and LF (Logical Form), which are the interface levels with the acoustic-perceptual systems and the conceptual-intensional systems. These are the only relevant representations, i.e. the former concepts of D-structure and S-structure no longer exist. D-structure is discarded, as there is no unique level at which all lexical insertion has taken place. Rather, the computational system builds up structures in a gradual manner, using the operations of Merge and Move interspersed, until the level of LF is reached. The traditional level of S-structure is equally unavailable: at some point the derivation branches off to PF, by the operation Spell-Out. Languages may vary in terms of where exactly SpellOut takes place. To illustrate this, let us look again at the different positions of the verb in English and French that we considered before (cf. the discussion of (2)-(4) above). This difference can be explained by assuming that the English verb occupies the same position as the French verb does at LF, but that the step to AgrS is taken in French before SpellOut, so that its effects are visible in PF, while the verb moves to AgrS in English after Spell-Out. This conception is essentially similar to the former distinction between syntactic (or pre-S-structure) movement and LF-movement. Every movement is driven by the requirements imposed by inflectional features. Lexical items are drawn from the lexicon in fully inflected form, with a featural specification. F-cats come into play as the sites where these features are checked. If a feature is strong, it needs to be checked before Spell-Out, and hence induces pre-Spell-Out movement, while movement is postponed until after Spell-Out is the feature is weak. This follows from the principle of Procrastinate, one of the economy considerations of the minimalist

20

Teun Hoekstra

program, which is based on the assumption that LF-movement is cheaper than pre-Spell-Out movement. A question which has always been lurking, but becomes even more glaring within the context of the minimalist program is: What is the relationship between F-cats and features? Does every feature correspond to an F-cat, or can different features comprise a single F-cat? So, could we have a F-cat Infi, in which Tense, Case, and Agr-features are all checked, or do we need a separate F-cat for each of these (classes of) features? Do we have a F-cat for Person agreement, one for Number agreement, and one for gender agreement, so that each agreement feature corresponds to its own F-cat (cf. Johnson 1990; Shlonsky 1990)? We saw earlier that considerations of structure preservation, or adherence to particular X-bar assumptions force us to postulate a fair number of F-cats, but these structural considerations might in fact be given up. So, we saw earlier that uniqueness of a specifier may require the postulation of an F-cat, but Chomsky (1995, ch.4) suggests that a F-cat may take multiple specifiers, and hence can check features of more than one phrase at the same time. This move, while certainly motivated, does not help much in answering this question, or tightening the theory of F-cats. What we see, then, is a shift of attention from F-cats to features. And with this shift come further shifts. So, Chomsky makes the further natural assumption that if movement of categories is driven by the need for features to be checked, the cheapest way for these features to be checked is in fact that these features move by themselves, rather than through piedpiping of the category that they belong. This idea allows for yet a different approach to the positional difference between French and English verbs: in both languages, the agreement features of the verb are checked: in English, the features have moved independently, in French they have pied-piped the verb along. Under this formulation, there is no need to distinguish between levels of application of movement: all movement can be done before SpellOut. What used to be LF-movement, is now pure feature movement. Roberts (1998) makes interesting use of this idea in explaining why English auxiliary verbs are typically found in more leftward positions than corresponding lexical verbs. The assumption is that in the case of lexical verbs only the formal features of V may move, stranding the verb at Spell-Out, while auxiliary verbs, which consist solely of formal features, move as a category. Another interesting application of the feature movement theory is presented by Lasnik (1998) in the analysis of the iftere-construction. As (number) agreement in ί/iere-constructions is determined by the associate, it has long been assumed that the associate moves to there at the level of LF, i.e. after Spell-Out. As Lasnik points out, a problem with this assumption is, that the scope properties of the associate do not change as a result

The function of Functional Categories

21

of this alleged movement: in there seems to be a unicorn in the garden the scope of a unicorn is necessarily narrow; it does not get the wide scope possibility that a unicorn has in a unicorn seems to be in the garden, even though at the LF-interface it occupies the same hierarchical position. Lasnik proposes that this scope problem is avoided if instead of the category a unicorn moving at LF, only the formal features of a unicorn move in order to check the agreement on the verb. This feature movement can take place before Spell-Out, as it doesn't result in any reordering effect in any case. The more we can appeal to feature movement instead of LF movement of a full category, the less significant LF movement becomes in general. Indeed, quite recently we have seen a move away from LF movement for instance in Kayne's (1998) work on negative expressions, and in Koopman & Szabolcsi's (1998) analysis of non-finite complementation structures in Hungarian and Germanic. What these analyses have in common is that they assign far more intricate structures to various constructions than had previously been thought of. In these structures, typically, various F-cats are postulated that trigger various movements. They are required to provide the structural space needed for the particular analyses, but the nature of these postulated F-cats remains unclear. While it is a conceptually desirable move to narrow down the distance between Spell-Out and the LF-interface, the fact that the analyses depend on the presence of F-cats whose nature is unclear, severely weakens these proposals. In this respect, then, we are in the same position as we were some ten years ago, when considerations of structure preservation forced us to postulate F-cats, without really knowing what these postulated categories stood for.

7. Conclusion The concept of F-cats has contributed to the theory of phrase structure: it allows us to formulate much more restricted hypotheses on the internal structure of phrases, viz. to restrict these to the Spec-Head-Complementizer format. Moreover, it invites us to narrow the gap between semantics and syntax in terms of specific syntactic representations of semantic interpretations. Finally, it yields a clear perspective on the ways in which to capture linguistic variation. The study of the properties of F-cats constitutes a fascinating area of research, therefore, most certainly against the background of the restrictive claims that are being put forth in the current literature. While at first sight the current literature may make a bewildering impression due to the large variety of proposed F-cats, the goals of the enterprise are clear and promising.

22

Teun Hoekstra

A Functional Categories

Bibliography

Abney, S. (1987). The English noun phrase in its sentential aspect. Doctoral dissertation, MIT. Baker, M. (1985). The mirror principle and morphosyntactic explanation. Linguistic Inquiry 16, 373-415. Belletti, A. (1990). Generalized verb movement: Aspects of verb syntax. Torino: Rosenberg. Besten, Η. den (1979). On the interaction of root transformations and lexical deletive rules. [Published in: On the formal syntax of the Westgermania, edited by W. Abraham, 47-132. Amsterdam: John Benjamins, 1983.] Boser, K., B. Lust, L. Santelmann & J. Whitman (1992). The syntax of CP and V2 in early child German. Proceedings ofNELS 22, 51-65. Brown, R. (1973). A first language. Boston: Harvard University Press. Chomsky, N. (1989). Some notes on economy of derivation and representation. MIT Working Papers in Linguistics 10, edited by I. Laka & A. Mahajan, 43-74. Chomsky, N. (1992). A minimalist program for linguistic theory. MIT Occasional Papers in Linguistics 1. Chomsky, N. (1995). The minimalist program. Cambridge, Mass.: MIT Press. Cinque, G. (1993). On the evidence for partial Ν-movement in Romance NP. Manuscript, University of Venice. Clahsen, Η. & M. Penke (1992). The acquisition of agreement morphology and its syntactic consequences. The acquisition of verb placement. Functional categories and V2 phenomena in language acquisition, edited by J. Meisel, 181-224. Dordrecht: Kluwer. Corver, Norbert (1990). The syntax of left-branch extractions. Doctoral dissertation, Tilburg University. Diesing, M. (1990). Verb movement and the subject position in Yiddish. Natural Language and Linguistic Theory 8, 41-79. Diesing, M. (1993). Indefinites. Cambridge, Mass.: MIT Press. Emonds, J. (1978). The verbal complex V-V in French. Linguistic Inquiry 9,151-175. Fassi Fehri, A. (1993). Issues in the structure of Arabic clauses and words. Dordrecht: Kluwer. Grimshaw, J. (1991). Extended projections. Manuscript, Rutgers University. Grimshaw, J. (1997). Projection, heads, and optimality. Linguistic Inquiry 28,373-422. Haegeman, L. (1994). Root infinitives, tense and truncated structures. Language Acquisition 4, 205-255.

The function of Functional Categories

23

Haegeman, Liliane & Rafaella Zanuttini (1991). Negative heads and the Neg criterion. The Linguistic Review 8, 233-251. Higginbotham, J. (1985). On semantics. Linguistic Inquiry 16, 547-593. Hoekstra, Τ. & N. Hyams (1998). Aspects of root infinitives. Lingua 106, 81-112.

Hyams, N. (1996). The underspecification of functional categories in early grammar. Generative perspectives on language acquisition, edited by H. Clahsen, 91-128. Amsterdam: John Benjamins. Johnson, K. (1990). The syntax of inflectional paradigms. Manuscript, University of Wisconsin-Madison. Kayne, R.S. (1989). Facets of Romance past participle agreement. Dialect variation and the theory of grammar, edited by P. Benincä, 85-103. Dordrecht: Foris Publications. Kayne, R. S. (1989). Notes on English agreement. CIEFL Bulletin 1,41-67. Kayne, R. S. (1991). Romance clitics, verb movement, and PRO. Linguistic Inquiry 22, 647-686. Kayne, R.S. (1994). The antisymmetry of syntax. Cambridge, Mass.: MIT-press. Kayne, R.S. (1998). Overt vs. covert movement. Syntax 1,128-191. Kondrashova, N. (1994). Semantic functional projections? Ξ Ρ: evidence from Russian. Paper presented at WECOL, UCLA. Koopman, H. (1993). The structure of Dutch PPs. Manuscript, UCLA. Koopman, H. & D. Sportiche (1991). The position of subjects. Lingua 85, 211-258. Koopman, H. & A. Szabolcsi (1998). Verbal complexes. Manuscript, UCLA. Kratzer, A. (1988). Stage-level and individual-level predicates. Manuscript, University of Massachusetts, Amherst. Kuroda, I. (1988). Whether we agree or not. A comparative syntax of English and Japanese. Linguisticae Investigationes 12,1-47. Laka. I. (1990). Negation in syntax: On the nature of functional categories and projections. Doctoral dissertation, MIT. Lasnik, H. (1993). Lectures on minimalist syntax. Manuscript, University of Connecticut. Lasnik, H. (1998). Levels of representation and the elements of anaphora. Atomism and binding, edited by H. Bennis, P. Pica & J. Rooryck, 251-268. Dordrecht: Foris Publications. Lebeaux, D. (1988). Language acquisition and the form of the grammar. Doctoral dissertation, University of Massachusetts, Amherst. Lightbown, P. (1977). Consistency and variation in the acquisition of French. Doctoral dissertation, Columbia University. Li, Y.-F. (1990). X°-binding and verb incorporation. Linguistic Inquiry 21, 399-426.

24

Teun Hoekstra

Longobardi, G. (1994). Reference and proper names. Linguistic Inquiry 25, 609-665. MacWhinney, Β. & C. Snow (1985). The child language exchange system. Journal of Child Language 12, 271-296. Ouhalla, J. (1990). Sentential negation, relativized minimality and the aspectual status of auxiliaries. The Linguistic Review 7, 183-231. Pierce, A. (1992). Language acquisition and syntactic theory: a comparative analysis of French and English child grammars. Dordrecht: Kluwer. Platzack, C. & A. Holmberg (1989). The role of AGR and finiteness. Working Papers in Scandinavian Syntax 43. Poeppel, D. & K. Wexler (1993). The full competence hypothesis of clause structure in early German. Language 69,1-33. Pollock, J.-Y. (1989). Verb movement, Universal grammar and the structure of IP. Linguistic Inquiry 20, 365-424. Radford, A. (1990). Syntactic theory and the acquisition of English syntax. Oxford: Basil Blackwell. Riemsdijk, H. van (1978). A case study in syntactic markedness. Lisse: Peter de Ridder Press. Ritter, Ε. (1991). Two functional categories in noun phrases: evidence from Modern Hebrew. Perspectives on phrase structure: Heads and licensing [Syntax and semantics 25], edited by S. Rothstein, 37-62. New York: Academic Press. Rivero, M.-L. (1991). Long head movement and negation: Serbo-Croation vs. Slovak and Czech. The Linguistic Review 8, 319-352. Rizzi, L. (1990). Relativized minimality. Cambridge, Mass.: MIT-press. Rizzi, L. (1994). Some notes on linguistic theory and language development: the case of root infinitives. Language Acquisition 3, 371-393. Roberts, I. (1991). Excorporation and minimality. Linguistic Inquiry 22, 209-218. Roberts, I. (1998). Have/be raising, move F, and Procrastinate. Linguistics Inquiry 29,113-126. Rögnvaldsson, Ε. & Η. Thräinsson (1990). On Icelandic word order once more. Modern Icelandic syntax [Syntax and Semantics 24], edited by J. Maling & A. Zaenen. New York: Academic Press. Shlonsky, U. (1990). The hierarchical representation of subject agreement. Manuscript, University of Haifa. Speas, M. (1994). Null arguments in a theory of economy of projections. UMass Occasional Papers in Linguistics 17,179-209. Suppes, P. (1973). The semantics of children's language. Archives de Psychologie 42, 207-279. Szabolcsi, A. (1984). The possessor that run away from home. The Linguistic Review 3, 89-102.

The function of Functional Categories

25

Szabolcsi, A. (1994). The noun phrase. The syntactic structure of Hungarian [Syntax and semantics 27], edited by F. Kiefer & Κ. Kiss, 197-274. New York: Academic Press. Tsimpli, I. (1996). The prefunctional stage of First Language Acquisition. A crosslinguistic Study. New York: Garland. Travis, L. (1984). Parameters and effects on word order. Doctoral dissertation, MIT. Valois, D. (1991). The internal syntax ofDPs. Doctoral dissertation, UCLA. Vikner, S. & B. Schwartz (1992). The verb always leaves IP in V2 clauses. Manuscript, University of Geneva and Boston University. Wagner, K., (1985). How much do children say in a day. Journal of Child Language 12, 475-487. Wexler, K. (1994). Optional infinitives, verb movement and the economy of derivation in child grammar. Verb movement, edited by D. Lightfoot & N. Hornstein, 305-350. Cambridge University Press. Zanuttini, R. (1991). Syntactic properties of sentential negation. Doctoral dissertation, University of Pennsylvania. Zwart, J.-W. (1993). Dutch syntax. Doctoral dissertation, Groningen University.

Control in GB and Minimalism Norbert Hornstein

1.

Introduction

From the earliest work in generative grammar, control structures have been treated in a theoretically sui generis manner. Rosenbaum (1967) distinguishes control from raising, assigning the former to the category of deletion processes while relegating the latter to movement. This partition survives the move from the standard theory to GB, albeit with a change in technical garb. In control configurations, deletion under identity yields to construal of a base generated empty category PRO. PRO is a contentless unindexed NP, viz. [ NP e], a base generated denizen of D-structure whose antecedent is determined at LF by the control module. This treatment differentiates PRO from NP-t(race) in two ways. First PRO is base generated whereas NP-f is a transformational by-product. Second, an NP-f's indexation is fixed under applications of Move while that of PRO is a function of the construal rules in the control module. This said, at LF, raising and control structures are rather similar, in that both involve the binding of structurally identical empty categories. Where the two structures differ is that control is an inter-chain relation (i.e. between two chains) while raising is an intra-chain relation (i.e. between two parts of a single chain). This in turn is the theoretical reflex of the central empirical distinction between raising and control, viz. that the binder of NP-ί is not independently θ-marked while the antecedent of PRO is. (1) illustrates these points. (1)

a. b.

John; seemed [[NP e\ to kiss a koala] Johnj hoped [[NP e\ to kiss a koala]

28

Norbert Hornstein

In (la), John is interpreted as a "koala-kisser" while in (lb) it has this role in addition to having the "hoper" role. John and the empty category in (la) form two "links" of a single chain. John and the empty category in (lb), in contrast, each constitute their own chains and the indicated indexation relates them anaphorically. With the standard assumption that chains are the bearers of θ-roles, this results in the observed facts. What undergirds this last venerable assumption? D-structure. In fact, the classical distinction between raising and control follows seamlessly from the assumption that D-structure exists. Consider the reasoning. D-structure has two distinctive properties: it is input to the transformational component and the locus of thematic discharge. Thus, prior to "displacement" operations (i.e. transformations) that rearrange phrase markers, words/morphemes are assembled into D-structure phrase markers by being lexically inserted into the available thematic positions. After lexical insertion, transformations apply, mapping D-structure phrase markers into others. Given the requirements of D-structure, transformations cannot relate θ-positions (via movement) as all θ-positions have been filled by lexical insertion in forming the D-structure phrase marker. Consequently, movement between θ-positions is impossible and so control relations cannot be the observed manifestations of Move. Furthermore, if D-structure has only θ-positions filled (in addition to all such positions being filled), then raising structures must be products of movement as only movement will be able to relate a θ to a ηοη-θ position. In sum, the classical vision of D-structure as the representation of pure GF-Θ, i.e. the phrase marker where all and only theta information is grammatically rendered, forces the theoretical distinction between raising and control. The theta criterion further buttresses this view of D-structure, in particular the idea that all thematic information is discharged via lexical insertion. The relevant feature of the theta criterion is the demand that there be a bi-unique relation between Θ-roles and chains, in particular that every chain bear at most one θ-role. This effectively prohibits all movement from one θ-position to another. But if movement into θ-positions is forbidden, yet all thematic roles must be discharged, then the only alternative is to fill each θ-position via lexical insertion. The step from the theta criterion to the postulation of PRO in control structures is a short one. This picture of control survives essentially intact to the present day. Chomsky (1995, Chapter 4), for example, maintains that only trivial chains (i.e. 1-member chains) can receive or assign θ-roles. In effect, θ-roles can only be discharged via Merge not Move. This restricts θ-role assignment to simple 1-member chains, or, to put this negatively, prohibits the assignment of θ-roles to chains of more than one member. This is functionally

Control in GB and

29

Minimalism

equivalent to the Standard Theory/GB requirement that θ-roles only be discharged at DS via lexical insertion. In what follows, I review the standard wisdom on control focusing on the distribution and interpretation of PRO. This review spans both the classical GB accounts and more recent minimalist approaches. I include a discussion of some current work of my own that takes as its starting point the denial that thematic structure is grammatically realized exclusively via lexical insertion or, in more current idiom, during the merger of trivial chains.

2. Control's basic

properties

Control structures come in two varieties; local and long distance. (2)

a. b.

John hopes to win John hopes that winning will be fun

In the earliest treatments, two distinct rules were involved in the derivation of (2a, b). Equi NP Deletion applied in (2a) while Super Equi was responsible for (2b). Equi interacted closely with the Minimal Distance Principle in determining the antecedent of the controlled clause, John in (2a). GB has honored essentially the same distinction. Williams (1980) argues for a distinction between obligatory and non-obligatory control configurations. The former conform to the earlier Equi structures while the latter shadow Super Equi configurations. This distinction is adopted in much subsequent work including Köster (1984), Lebeaux (1985), and Manzini (1983) among others. The relevant GB structures of (2) are provided in (3). (3)

a. b.

John hopes [PRO to win] John hopes [that [[PRO winning] will be fun]]

Obligatory and non-obligatory control configurations differ in several important ways. Note, for example, that (3b) has at least two interpretations; one paraphraseable as 'John hopes that his own winning will be fun' and one as 'John hopes that one's winning will be fun'. This ambiguity is absent in (3a) with the second "arbitrary" reading missing. We return to a full description of the interpretive differences between obligatroy and nonobligatory control structures below. In light of (3), control structures pose two questions: (i) where can PRO appear and (ii) how is PRO interpreted. The GB answer to the first question is that PRO appears in ungoverned positions, in effect, the subject position of non-finite clauses. In (3a) we find PRO in the subject position of an

30

Norbert Hornstein

infinitive clause and in (3b) it appears in the subject of a clause initial gerund. PRO seems unable to appear anywhere else. (4)

a. b. c.

*John hopes (that) PRO will win *John saw PRO *John talked to PRO

It is reasonable to assume that in the examples in (4) PRO is governed by various lexical heads. In (4a) it is the finite morpheme in Infi, in (4b) the verb saw and in (4c) the preposition to. If one further assumes that nonfinite clauses do not contain lexical heads (at least of the relevant sort), then the presence of PRO in (3a, b) correlates with the absence of head government (henceforth simply "government") of these SpecIP positions. Thus, the distribution of PRO conforms to the descriptive generalization (5). (5)

PRO can only appear in ungoverned positions

Note that (5) need not prohibit generating PRO in a governed position. This is fortunate as PRO can be base generated in object position so long as it moves to an ungoverned position by S-structure (SS). (5) is a generalization that holds at S-structure or later. (6)

John hopes [PRO; to be recognized

Is (5) an axiom or does it follow from more basic principles? There are several GB attempts to reduce (5) to other grammatical principles. Bouchard (1984), for example, proposed reducing the distribution of PRO to Case Theory (see also Vanden Wyngaerd 1994 for a related appproach based on agreement). He observed that the contexts that disallow PRO are generally also contexts in which Case marking applies. This point is clearly illustrated in the examples in (4). Lasnik (1992), however, notes a problem for the proposal that Case can fully account for the distribution of PRO. There are several non-Case marking contexts where PRO is nonetheless disallowed. (7)

a. b. c.

*John believes sincerely Mary/PRO to be clever *it is likely [John/PRO to solve the problem] *My belief [Hany/PRO to be intelligent]

It would appear then, that a Case theoretic approach to the distribution of PRO is too weak. A second approach, the one that became the GB standard, links the distribution of PRO to the Binding Theory. Chomsky (1981) proposed that PRO was a pronominal anaphor, i.e [+P,+A], subject to both Principles A and Β of the Binding Theory. Principle A requires an anaphor to be bound in its domain, Principle Β that a pronoun be free (i.e. not bound). Given

Control in GB and Minimalism

31

that a single expression cannot be both bound and free in its domain, this entails that PRO never has a domain. A way for PRO to fail to have a domain is if it is ungoverned given the definition of domain in (8). (8)

α is the domain for an expression Ε iff α is the smallest phrase in which Ε is governed and α has a subject

Note that if PROs are pronominal anaphors subject simultaneously to Principles A and Β of the Binding Theory then we have derived the descriptive generalization (5) from more general principles. Note too, that given that the Binding Theory applies at S-structure or later, then we have also derived the fact that (5) is a generalization that holds later than D-structure. Nonetheless, there are problems with the PRO theorem approach to the distribution of PRO. First, as noted by Bouchard (1984), it appears that if we advert to subjects in the definition of domains, then we need not invoke government. Consider, for example, an alternative definition of domain. (9)

α is the domain for an expression Ε iff α is the smallest domain in which α has an accessible subject.

(10)

β is an accessible subject for Ε iff β is a SUBJECT and β is distinct from Ε

(11)

SUBJECTS are [ΝΡ,ΙΡ], [ΝΡ,ΝΡ] and Agr

(9M11) in conjunction with standard clauses of the Binding Theory handle the standard binding data. For example, they cover at least as much empirical territority as Chomsky (1986). However, they do so without adverting to the notion "governor". As such, it is unclear that adding this notion to the definition of domain is truly necessary. But if it isn't, the PRO theorem exploits a feature of the Binding Theory that is unneccesary except for deriving the distribution of PRO. This leaves the PRO theorem on the same conceptual footing as the bare assertion that (5) obtains. A second problem is that it is not always the Case that pronouns and anaphors are in complementary distribution. This means that they do not always have identical domains. Consider examples such as (12). (12)

a. b.

The men saw their mothers The men saw each other's mothers

Here we appear to have a pronoun {their) and an anaphor (each other) both licit in the same SpecNP/DP position. By the logic of the Binding Theory one might expect PRO to be licit here as it could be free in its guise as a pronoun yet bound in its anaphoric aspect. Nonetheless PRO is prohibited here.

32

Norbert Hornstein

(13)

*The men saw PRO mothers

Chomsky (1986) provides a technical escape for cases such as this. The cludge is to require that an expression have a unique domain. The reasoning from (12) to (13) first evaluates PRO qua pronoun with respect to Principle Β and then evaluates PRO qua anaphor to see if it respects Principle A. The domain relevant for the Principle Β evaluation is different from the domain exploited to determine is Principle A is met. Chomsky (1986) proposes that we disallow this domain shifting when the binding status of a single expression is being considered. In effect he proposes that a single expression Ε must have a unique domain, as in (8) above. Thus a PRO cannot be evaluated with respect to two different domains as (13) does. Chomsky's proposal allows the Binding Theory to block (13) while preserving the assumption that PRO is a pronominal anaphor. However, the solution circumvents a problem unique to PRO. Thus, this fix to the PRO theorem once again questions whether the distribution of PRO actually derives from independently motivated principles of the Binding Theory. A third problem with the PRO theorem is evident when one considers the interpretation of obligatory and non-obligatory control structures. The PRO theorem relies on the assumption that every PRO is simultaneously +P and +A. However, obligatory control and non-obligatory control PROs differ rather drammatically in their interpretive properties. Furthermore, these differences make sense if obligatory control PROs are anaphors while nonobligatory control PROs are pronouns. In short, the interpretive data point to the conclusion that PRO is ambiguous. A PRO theorem approach to PROs distribution cannot accept this ambiguity thesis. To handle the interpretive differences evident in obligatory and non-obligatory control structures requires a rather complex control module. Consider the following paradigm illustrating the interpretive properties of obligatory control structures. This paradigm derives from Williams (1980), Lebeaux (1985), Higginbotham (1992) and Fodor (1975). (14)

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

*It was expected PRO to shave himself *John thinks that it was expected PRO to shave himself *John's campaign expects PRO to shave himself John expects PRO to win and Bill does too (=Bill win) *Johni told Maryj PROi+j to leave together The unfortunate expects PRO to get a medal Only Churchill remembers PRO giving the BST speech

(14a) shows that an obligatory control PRO must have an antecedent. (14b) indicates that this antecedent must be local and (14c) indicates that it must c-command the PRO. (14d) shows that this PRO cannot have split anteced-

Control in GB and Minimalism

33

ents. PRO in (14f) only has the 'de se' interpretation in that the unfortunate believes of himself that he will be a medal recipient. (14g) has the paraphrase (15a) not (15b). On this reading only Churchill could have this memory for Churchill was the sole person to give the speech. The two different readings follow on the assumption that obligatory control PRO must have a c-commanding antecedent. This requires only Churchill to be the binder. The unavailable reading has Churchill as the antecedent. This is possible in (15b) where the pronoun can have a non-c-commanding antecedent. (15)

a. b.

Only Churchill remembers himself giving the BST speech Only Churchill remembers that he gave the BST speech

PRO in non-obligatory control environments contrasts in every respect with the obligatory control cases. (16)

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

It was believed that PRO shaving was important Johnj thinks that it is believed that PRO; shaving himself is important Clinton'Sj campaign believes that PRO; keeping his sex life under control is necessary for electoral success John thinks that PRO getting his resume in order is crucial and Bill does too The unfortunate believes that PRO getting a medal would be boring Only Churchill remembers that PRO giving the BST speech was momentous

(16a) indicates that non-obligatory control PRO does not require an antecedent. (16b) demonstrates that if it does have an antecedent it need not be local. (16c) shows that the antecedent need not c-command this PRO. (16d) contrasts with (14d) in permitting a strict reading of the elided VP, i.e. the reading in which it is John's resume which is at issue. (16e) can have a non 'de se' interpretation and (16f) is consistent with many people other than Churchill recalling that the BST speech was a big deal. Note that each nonobligatory control reading contrasts with those available in the obligatory control examples in (14). The cases in (14) and (16) contrast in one further interesting way; the former can be paraphrased with PRO replaced by a reflexive while the interpretive doubles of (16) replace PRO with pronouns. (17) illustrates this with the counterparts of (14c) and (16c). (17)

a. b.

* John's; campaign expects himself; to shave himself Clinton's; campaign believes that his; keeping his sex life under control is crucial for electoral success

In short, the differences in obligatory and non-obligatory control structures duplicate, where applicable, what one finds with locally bound anaphors

34

Norbert Hornstein

versus pronouns. This makes sense if PRO is actually ambiguous — an anaphoric expression in obligatory control configurations and pronominal in non-obligatory control structures rather than simultaneously a pronoun and an anaphor as the PRO theorem requires. This, then, speaks against reducing the distribution of PRO to the Binding Theory by way of the PRO theorem. The upshot of these considerations is the following. The favored GB account for the distribution of PRO in terms of the PRO theorem requires assuming that all PROs are simultaneously pronouns and anaphors. This, however, fits poorly with PROs interpretive properties which indicate that PROs are ambiguous. Obligatory control PROs function like anaphors while non-obligatoiy control PROs are interpreted as pronouns. In effect, the standard story about the distribution of PRO and the account of how PROs are interpreted pull strongly in opposite directions. A virtue of recent minimalist musings on PRO is that they alleviate this theoretical tension by ousting the PRO theorem. We turn to this next.

3. Minimalist

accounts

Minimalism fits poorly with the PRO theorem given the latter's reliance on the core notion of government. The minimalist program as outlined in Chomsky (1995) eschews relations like government and favors notions based on more fundamental notions relating to phrasal structure; relations like Spec-head and head-complement for example. In addition to the theoretical undesirability of the PRO theorem, Chomsky & Lasnik (1993) argue that it is empirically insufficient. Consider (18). (18)

a. b.

We never expected [PRO to be found t] *We never expected PRO to appear to t that Bill left

If movement is last resort and PRO must be ungoverned then the threat of being governed should suffice to license PRO's movement in (18a). But if this suffices in (18a) why is it insufficient in (18b)? Chomsky & Lasnik argue that both cases fall together if we assume that PRO has a Case that requires checking. In effect, they urge a return to Bouchard's original intuition that the distribution of PRO is the province of Case Theory. They add a twist, however. They assume that PRO has a sui generis Case, dubbed "null Case", that can only be checked by the Infi of non-finite control clauses. PRO (and PRO alone) carries null Case, a Case that gets checked in the SpecIP of the standard control clauses. Given this proposal, (18b) violates greed (or feature mismatch) as null Case marked PRO moves from a Case position to another Case position thereby violating last resort.

Control in GB and

Minimalism

35

Observe that this approach to the distribution of PRO has two virtues. First, government is replaced by Case checking, the latter being the paradigmatic instance of a Spec—head relation in minimalism. Second, this approach permits the natural account of the interpretive facts noted above: PRO is ambiguous, obligatory control PRO is an anaphor, non-obligatory control PRO a pronoun. Note that the ability to make this distinction between types of PRO is not exclusive to minimalism. The distinction is already implicit in Chomsky & Lasnik (1977) and is explicit in Williams and Lebeaux among others. However, it is not compatible with accounts that based the distribution of PRO on the PRO theorem. By abandoning this latter approach minimalist proposals can account for the properties noted in (14) and (16) in the way earlier non-PRO theorem based accounts could and did. The most fully worked out version of this null Case approach to obligatory control is Martin (1996). There are two broad assumptions that he utilizes in implementing the proposal. Following Stowell (1982), he argues that control infinitives differ from raising infinitives in that the former have tensed Infls. Tense quite generally checks Case and infinitive tense checks null Case. As only PRO carries null Case, only PRO can appear in the SpecIP position of tensed infinitives. In effect, Martin provides motivation for the assumption that some infinitives can check Case by assimilating them to the class of clauses that uncontroversially do so, finite clauses. Both finite IPs and control infinitives are +Tense and so it is not surprising that their respective SpecIPs are Case positions. For some criticism of this assumption see Brugger 1996. Martin's second assumption is that obligatory control PRO is akin to the romance clitic se (see Raposo & Uriagereka 1996) and that obligatory control is actually a form of clitic climbing. The idea is that subject control structures have LF configurations like (19a) and object control structures look like (19b). (19)

a. b.

[ TP John [ r [ T PRO-T] hopes [TP [ DP i PR0 pro] to win]]] [ TP They; [ r Τ [ vP t{ [ VP persuaded [XP Bebeto [ x . [ x PRO X] [TP [DP i P R 0 pro] to play for Flamengo]]]]]]

PRO is a clitic specifier of a pro headed DP. It raises and cliticizes to T° in subject control structures and to some Agr-like X° in object control structures. Crucially, the clitic PRO has a very impoverished feature content. This is similar to what Raposo & Uriagereka propose for Romance se. Specifically, the PRO head of the chain is featurally non-distinct from other nominals. This makes it impossible to distinguish the PRO headed clitic chain from other nominals in its minimal domain. The result is that the PRO chain "collapses" and the PRO chain and the chain of the controller

36

Norbert

Hornstein

form a single LF chain with two thematic roles. The fused chain, Martin proposes, yields the strict anaphoricity noted in the paradigm displayed in (14) and (15). Thus, in (19a), the chain headed by John and the one headed by PRO form a single collapsed Α-chain at LF as PRO is featurally nondistinct from John, a nominal element in the same minimal domain. In (19b) the PRO chain and the one headed by Bebeto fuse to form a single composed Α-chain. It is possible to trace the interpretive properties of obligatory control structures to this structural fact. Consider a couple of examples. Recall that obligatory control structures do not permit split antecedents. The present analysis provides an account of this. Such structures result from the fusing of two chains, one headed by PRO due to PRO's being indistinguishable from other nominals in the same domain. In the case of split antecedents the three relevant chains could not collapse, as the heads of all three chains would not be in the same minimal domain. At most two would be. For example the structure of (20a) is (20b) with Mary and PRO being in the the same minimal domain; the configuration for the chains to collapse. Bill, however, is outside this minimal domain and so its chains would not collapse with the other two. As control is a matter of forming a single composed chain, split control is not available in obligatory control configurations. (20)

a. *John; told Maryj PROi+j to leave together b. [TP John; [ r Τ [ vP ti [yp told [χρ Maiy [x-[x PRO X] [TP [DP i PR0 pro] to leave together]]]]]]

Similar reasoning accounts for the c-command condition on antecedents in obligatory control configurations. (21a) is unacceptable because John is not in the same minimal domain as PRO. As such, a collapsed chain between John and PRO is not available and the indicated indexing is unavailable. (21)

a. b.

John's; mother hopes PRO, to wash himself [TP [John's mother][r [T PRO-T] hopes [TP [DP i PR0 pro] wash himself]]]

This analysis has a further payoff. It reduces the locality effects characteristic of obligatory control to the "shortest move" or "minimal link" constraint. Given that obligatory control results from clitic like movement, we expect it to be subject to locality restrictions typical of such movement. For example, the Minimal Distance Principle, first noted by Rosenbaum (1967), can now be reduced to "shortest move". Consider (19b) once again. They cannot be the relevant controller for cliticizing PRO to T° skips over the potential landing site offered by X. This instance of the Minimal Distance Principle can be reduced to the shortest move condition in the minimalist program if excorporation is prevented.

Control in GB and Minimalism

37

Two ideas embodied in Martin's analysis - that obligatory control involves movement and that chains can have multiple theta roles (at least at LF) - characterize two other current minimalist approaches to control as well. The differences regard what moves and to where. Manzini & Roussou (1997), propose eliminating PRO from the grammar entirely. To accomplish this they propose that θ-roles are actually relations to aspectual heads. Building on work of Tenny (1994) and Borer (1993), they propose that θ-roles be reduced to aspectual function and that aspectual properties of clauses be discharged by different sorts of aspectual heads. On this view, θ-roles are properties that DPs have in virtue of being in Spec-Head configurations with designated aspectual heads. This assimilates θ-relations to feature checking relations more generally, in contrast to Chomsky (1995) who radically distinguishes θ-relations from the checking relations that morphological features enter into. One important consequence of this view of θ-roles is that it dispenses with the Theta Criterion. Just as a DP can enter into multiple checking relations to check morphological features, a DP can also enter into checking relations with multiple aspectual heads. Manzini & Roussou's second assumption is that there is no A-movement of DPs. Rather DPs are generated in their Case positions and aspectual features raise to θ-mark them. This accounts for the fact that PRO positions are phonetically null as there is nothing there at all, according to this proposal. With these two assumptions in hand, control can be seen as the LF raising of the features of an embedded aspectual head to a matrix Inflection. By so moving, the embedded aspectual head θ-marks the matrix generated DP. Consider a sample derivation. (22)

John Infi [Asp0 tried [to [Asp1 leave]]]

The structure in (22) is generated in overt syntax. After Spell Out, F-movement applies raising the features ofAsp0 to the matrix Infi. Manzini & Roussou follow Chomsky 1995 in assuming that traces cannot block movement across them. Thus, the trace ofAsp0 does not block movement of the F-features of Asp1 to the matrix Infi. As bearing a θ-role just is being in a specified Spec-Head configuration with the relevant aspectual features, this pair of movements allows John in (22) to bear two roles. Though the details of the Manzini & Roussou analysis differ greatly from Martin's, they share the advantage of reducing many locality properties of obligatory control, e.g. the Minimal Distance Principle, to locality restrictions on X°-movement. Furthermore, both theories allow a DP or the chain it heads to bear multiple roles. Hence to the degree that this is responsible for the particular properties that obligatory control enjoys, both can account

38

Norbert

Hornstein

for these in roughly the same way. The principle difference between Manzini & Roussou versus Martin centers on their respective attitudes to null Case and theta discharge. Manzini & Roussou dispense with null Case by assuming that Α-movement does not exist in the grammar. Though it is unclear how this prohibition is to be enforced given general minimalist assumptions - how is Α-movement to be prevented? - it suffices to remove null Case and PRO from the grammar. Second, pace Martin and Chomsky (1995), they assume that θ-roles can be discharged via movement, i.e. that non-trivial chains can assign θ-roles. This is a more radical assumption as it questions the view that θ-discharge is a "base" property, an assumption retained in Chomsky (1995). Hornstein (1996a) adopts a position concerning θ-discharge akin to Manzini & Roussou's. It proposes reducing obligatory control to A-movement and identifies PRO as an intermediate NP-f. The assumptions required to make this proposal operable given a minimalist background are the following. (23)

a. b. c. d.

Theta roles are features on verbs Greed is enlightened self interest A D/NP "receives" a theta role by checking a theta feature of a verbal/predicative phrase that it merges with There is no upper bound on the number of theta roles a chain can have

(23a-d) are required if obligatory control is to be reduced to A-movement given standard minimalist assumptions. What differentiates raising from control is that the former moves a DP to a ηοη-θ position while the former moves it to a θ-position. To be compatible with the least effort notions characteristic of the minimalist program, greed in particular, movement to a theta position must be kosher. Once the assumptions in (23) are adopted, little more needs to be stated in order to derive the basic facts about the distribution of PRO and its interpretation in obligatory control structures. Null Case can be dispensed with as PRO is identified here as just another NP-f. In fact, null Case is incompatible with this proposal given that movement from a Case marked position is generally illicit. The distribution of PRO reduces to those positions without Case. As for the interpretation of PRO, the basic properties in (14) follow from the fact that obligatory control PRO is derived via Α-movement. The requirement that there be an antecedent follows from the definition of Move. That the antecedent must be local and c-command the PRO follows from the general properties of chains and movement. The fact that only sloppy readings are possible under VP ellipsis parallels what happens in raising constructions where only a sloppy reading is available

Control in GB and Minimalism

39

as well. The other features of obligatory control follow as well and the interested reader is referred to Hornstein (1996a) for discussion. Let's consider a sample derivation to help fix the mechanics of the proposal. The basic structure of a subject control sentence is (24). (24)

a. b.

John hopes to win [ IP Johnj 1° [yp hopes [IP ti to [yp ti win]]]]

John merges in the Spec of win thereby getting a theta role. It then raises to SpecIP to check the Extended Projection Principle feature of the embedded clause. Thereafter, John raises to Spec of hopes getting a second theta role. Last of all, it raises to SpecIP where it checks the Extended Projection Principle feature of the matrix 1°, as well as checking its own Case features and the Case and agreement features of the matrix inflection. Object control structures are bit more complex. Consider a sample derivation. (25)

a. b.

John persuaded Harry to leave [IP2 Johnj [I0 past [ypg Harry; [ypg t} v+persuaded [VP2 t{ persuaded [IP1 [to [ypj t{ leave]]]]]]]

(25b) starts by merging leave and Harry thereby providing Harry with a theta role. Harry then raises to check the embedded Extended Projection Principle features of the lower clause. Observe that this violates procrastinate. Next the IP merges with persuade and then, with another violation of procrastinate, Harry raises to merge in the Spec of persuade and receives its second θ-role. The VP so formed then merges with υ and John is merged to the Speci; position receiving a θ-role. John then raises checking its Case features and checking the Extended Projection Principle, Case and agreement features of the matrix clause. At LF, Harry moves to the outer Specie (or AgrO) to check Case. This derivation gives the correct results but at the cost of two violations of procrastinate. What permits these violations? Without these violations Harry would not be in a position to check its Case features. For example, if John were merged in the SpecIPl then it would block movement of Harry to the outer SpecVP3 at LF and so its Case could not be checked. So too if John were merged in SpecVP2 instead of raising Harry. In either case we would get a minimality/shortest move violation. Thus, procrastinate is violated to permit convergence. As convergence trumps procrastinate, this licenses the violations indicated and the derivation is licit. Two points are worth noting. First, like the theory in Martin (1996) and Manzini & Roussou (1997) obligatory control involves movement, though in this case it is XP, rather than X°, movement. Like them, this is crucially exploited in reducing the Minimal Distance Principle to shortest move

40

Norbert

Hornstein

considerations. Second, this analysis crucially relies on the assumption that traces are relavant to minimality, in contrast to the assumptions outlined in Chomsky (1995) and Manzini & Roussou. This second point has interesting implications for the status of chains and traces within the minimalist program, but this is not the place to pursue these consequences (see Nunes 1995; Hornstein 1996b) for some discussion. The three approaches to obligatory control have many features in common yet exploit very different machinery to attain rather similar effects. At the most general level all three exploit the intuition that obligatory control involves movement and that obligatory control structures violate the most straightforward interpretation of theta criterion, at least at LF. All three theories also reject the standard GB PRO theorem analysis for the distribution of PRO. The theories also diverge in important ways. Martin claims that raising is different from control in that structrures of control involve a PRO that is absent in the latter. Both Manzini & Roussou and Hornstein do without PRO. Martin treats theta roles as different in kind from features like Case. An uncomposed chain can have but a single θ-role. Both Hornstein and Manzini & Roussou abandon this requirement on θ-discharge and allow a chain to be multiply θ-marked via movement. Future research will likely center on dissecting the technical mechanisms that each approach exploits to see how to extend them to other kinds of structures, e.g. control into adjuncts, control in noun complement constructions, and control into weak islands. Some of these issues are broached in the work referred to. However, behind the technical differences there is a convergence of opinion that the basic interpretive properties of obligatory control require rethinking the theta criterion and the structure of chains.

4. Non obligatory

control

Section 3 deals with current approaches to the properties of obligatory control structures. What of non-obligatory control? As noted in (16), interpretively, non-obligatory control structures pattern as if the relevant empty element is a pronoun. Even the arbitrary reading characteristic of non-obligatory control structures can be accommodated on the assumption that the pronoun one has a null counterpart (see Chomsky 1986). Thus, it seems reasonable to assume that these structures have a null pro in the relevant Spec position. What about the distribution of pro? It turns out that non-obligatory control readings are largely restricted to positions from which movement is

Control in GB and Minimalism

41

not allowed, such as complex subjects or ΐϋΛ-islands (see (16) and examples in Manzini 1983). Furthermore, like PRO, pro is restricted to ungoverned SpecIPs. However, it is also clear that the distribution of obligatory and non-obligatory control structures is not free. Were this the case we would have no evidence for the existence of obligatory control structures. The reason is that the interpretive properties of obligatory control PRO are a subset of those manifested by pro in non-obligatory control configurations. For example, non-obligatoiy control pro can have a local c-commanding antecedent but it doesn't need one. Obligatory control PRO differs from non-obligatory control pro in being illicit unless locally bound. Therefore, if obligatory and non-obligatory control structures were in free distribution the non-obligatory control option would render invisible those properties diagnostic of obligatory control. In short, were pro able to go wherever PRO goes we would never detect the presence of the latter. How then to accommodate the distribution of pro? One possibility is that pro constitutes the "elsewhere" case (see Hornstein 1996a). In other words, pro appears when obligatory control is not licit. In terms of the analyses in Section 3, this amounts to saying that pro is licensed just in case movement is not possible. This comports well with the observation that non-obligatory control occurs when the pro/PRO is within islands. In a minimalist context, the distribution of pro looks similar to the problem posed by do-support. The problem is how to state the fact that one only uses do when there is no other licit way to converge. Arnold (1995) has noted that a necessary condition for stating this is not to count do as part of the array and to make it a morpheme whose use is costly (see Chomsky 1995, Chapter 2). Consider why this is so. If a necessary condition for comparing derivations with regard to economy is that they share a common array, then (26a, b) are prima facie non-comparable if do is part of the array. (26)

a. John left b. *John did leave (unstressed do)

This makes it impossible to state the fact that do can only be used as a last resort. If, however, do does not count as part of the array and its use is costly, then the convergence of (26a) blocks the convergence of (26b). We can extend this logic to non-obligatory control structures if we assume for pro what Arnold proposes for do. The use of pro is costly and so can only be used if a derivation cannot converge without it. As movement is permitted where islands don't obtain, obligatory control will always trump non-obligatory control where permitted. However, islands prevent movement and so 'pro' can be used to "save" an otherwise illicit derivation. This covers the basic observed facts concerning the distribution of nonobligatory control configurations. See Aoun, Choueiri & Hornstein (1998)

42

Norbert Hornstein

and Hornstein (1998) for a more general discussion of the distibution of pronouns, including pro.

5. Conclusion Control structures have been a focus of research from the earliest days of generative grammar. Minimalism has led to a fundamental rethinking of these constructions and how to account for their properties. Though the owl of Minerva generally spreads its wings at dusk and this is quite definitely still the early dawn for minimalism, current proposals seem novel and encouraging. There is a reasonable hope that in the near future both the distribution and interpretation of control structures will be analyzable in terms of independently motivated properties of Universal Grammar rather than requiring modules and principles specifically designed to accommodate the properties of control structures alone.

A Control

Bibliography

Andrews, Avery (1990). Case structures and control in modern Icelandic. Modern Icelandic syntax, edited by Joan Maling & Annie Zaenen, 187-234. [Syntax and Semantics 24.] San Diego: Academic Press. Arnold, M. (1995). Case, periphrastic do and the loss of verb movement in English. Doctoral dissertation, University of Maryland, College Park. Aoun, Josef, L. Choueiri & Norbert Hornstein (1998). Resumption and last resort. A minimalist proposal. Manuscript, USC, Los Angeles and University of Maryland, College Park. Bach, Emmon (1979). Control in Montague grammar. Linguistic Inquiry 10, 515-531. Bennis, Hans & TeunHoekstra (1989). PRO and the Binding Theory. Linguistics in the Netherlands 1989, edited by Hans Bennis & Ans van Kemenade, 11-20. Dordrecht: Foris Publications. Bouchard, Denis (1984). On the content of empty categories. Dordrecht: Foris Publications. Bouchard, Denis (1985). PRO, pronominal or anaphor. Linguistic Inquiry 16,471-477. Bresnan, Joan (1982). Control and complementation. The mental representation of grammatical relations, edited by Joan Bresnan, 282-390. Cambridge, Mass.: MIT Press. Brody, Michael (1982). On deletion and local control. MIT Working Papers in Linguistics 4, 5-14.

Control in GB and Minimalism,

43

Brody, Michael (1985). On the complementary distribution of empty categories, Linguistic Inquiry 16, 505-546. Brugger, G. (1996). The representation ofpresent perfect types. Manuscript, UCLA. Burzio, Luigi (1986). Italian syntax. Dordrecht: Reidel. Chierchia, Gennaro (1984). Topics in the syntax and semantics of infinitives and gerunds. Doctoral dissertation, University of Massachusetts, Amherst. Chierchia, Gennaro (1990). Anaphora and attitudes de se. Semantics and contextual expressions, edited by Renate Bartsch, 1-32. Dordrecht: Foris Publications. Chierchia, Gennaro & Pauline Jacobson (1986). Local and long distance control. Proceedings ofNELS 16, 57-74. GLSA, University of Massachusetts, Amherst. Chomsky, Noam (1981) Lectures on government and binding. Dordrecht: Foris Publications. Chomsky, Noam (1982) Some concepts and consequences of the theory of government and binding. Cambridge, Mass.: MIT Press. Chomsky, Noam (1986). Knowledge of language. New York: Praeger. Chomsky, Noam (1995). The minimalsit program. Cambridge, Mass.: MIT Press. Chomsky, Noam & Howard Lasnik (1993). The theory of prnciples and parameters. [Reprinted in Chomsky 1995.] Clark, R. (1985). Boundaries and the treatment of control. Doctoral dissertation, UCLA. Epstein, Samuel (1984) Quantifier-pro and the LF-representation of proarb. Linguistic Inquiry 15, 499-505. Farkas, D. (1988). On obligatory control. Linguistics and Philosophy 11, 27-58. Franks, Stephen & Norbert Hornstein (1992). Secondary predication in Russian and proper government of PRO. Control and grammar, edited by Larson et al., 1-50. Fodor, J. (1975). The language of thought. New York: Crowell. Grinder, John (1970) Super equi-NP deletion. Papers from the 6th regional meeting, Chicago Linguistic Society, 297-317. Chicago Linguistic Society, Chicago, Illinois. Higginbotham, James (1992). Reference and control. Control and grammar, edited by Larson et al., 79-108. Hoekstra, Teun & Hans Bennis (1989). A representational theory of empty categories. Linguistics in the Netherlands 1989, edited by Hans Bennis & Ans van Kemenade, 91-99. Dordrecht: Foris Publications.

44

Norbert Hornstein

Hornstein, Norbert (1996a). On control. Manuscript, University of Maryland, College Park. Hornstein, Norbert (1996b). Movement and chains. Manuscript, University of Maryland, College Park. Hornstein, Norbert (1998). On pronouns. Manuscript, University of Maryland, College Park. Hornstein, Norbert & David Lightfoot (1987). Predication and PRO. Language 63, 23-52. Huang, James C.-T. (1989). Pro-drop on Chinese. A generalized control theory. The null subject parameter, edited by Osvaldo Jaeggli & Ken Safir, 185-214. Dordrecht: Kluwer. Jacobson, Pauline (1992). Raising without movement. Control and grammar, edited by Larson et al., 149-194. Kayne, Richard (1991). Romance clitics, verb movement, and PRO. Linguistic Inquiry 22, 647-686. Kimball, John (1971). Super Equi-NP deletion as dative deletion. Papers from the 7th regional meeting, Chicago Linguistic Society, 142-148. Chicago Linguistic Society, Chicago, Illinois. Koster, Jan (1984). On binding and control. Linguistic Inquiry 15,417-459. Larson, Richard (1988). Promise and the theory of control. Linguistic Inquiry 19, 335-391. Larson, Richard, Sabine Iatridou, Utpal Lahiri & James Higginbotham, editors (1992). Control and grammar. Dordrecht: Kluwer. Lasnik, Howard (1992). Two notes on control and binding. Control and grammar, edited by Larson et al. Lebeaux, David (1984). Anaphoric binding and the definition of PRO. Proceedings ofNELS 14, 253-274. GLSA, University of Massachusetts, Amherst. Lebeaux, David (1985). Locality and anaphoric binding. The Linguistic Review 4, 343-363. Manzini, M.-Rita (1983). On control and control theory. Linguistic Inquiry 14, 421-446. Manzini, M.-Rita & Anna Roussou (1997). A minimalist theory of movement and control. Manuscript, Firenze/UC London and Bangor. Martin, R. (1996). A minimalist theory of PRO and control. Doctoral dissertation, University of Connecticut, Storrs. Mohanan, K.P. (1983). Functional and anaphoric control. Linguistic Inquiry 14, 641-674. Nunes, Jairo (1995). The copy theory of movement and linearization of chains in the Minimalist Program. Doctoral dissertation, University of Maryland, College Park.

Control in GB and

Minimalism

45

Postal, Paul (1970). On (»referential complement subject deletion. Linguistic Inquiry 1, 439-500. Raposo, Eduardo & Juan Uriagereka (1996). Indefinite se. Natural Language and Linguistic Theory 14, 749-810. Reuland, Eric (1981). Empty subjects, Case and agreement and the grammar of Dutch. Binding and filtering, edited by Frank Heny, 159-190. London: Croom Helm. Roberts, Ian (1987). The representation of implicit and dethematized subjects. Dordrecht: Foris Publications. Rooryck, Johan (1987). Les verbes a contröle. Doctoral dissertation, K.U. Leuven. Rosenbaum, P. (1967). The grammar of English predicate complement constructions. Cambridge, Mass.: MIT Press. Sigurösson, Halldor (1991). Icelandic Case-marked PRO and the licensing of lexical arguments. Natural Language and Linguistic Theory 9, 327-363. Stowell, Tim (1982). The tense of infinitives. Linguistic Inquiry 13, 561-570. Tenny, Carol (1994). Aspectual roles and the syntax-semantics interface. Dordrecht: Kluwer. Terzi, Arhonto (1992). PRO in finite clauses: a study of the inflectional heads of the Balkan languages. Doctoral dissertation, New York University. Thompson, Sandra (1973). On subjectless gerunds in English. Foundations of Language 9, 374-383. Thräinsson, Hoskuldur (1979). On complementation in Icelandic. New York: Garland. Wasow, Thomas & Thomas Roeper (1972). On the subject of gerunds. Foundations of Language 8, 44-61. Williams, Edwin (1980). Predication. Linguistic Inquiry 11, 203-238. Williams, Edwin (1985). PRO and the subject of NP. Natural Language and Linguistic Theory 3, 297-315. Williams, Edwin (1987). Implicit arguments, the Binding Theory, and control. Natural Language and Linguistic Theory 5,151-180. Wyngaerd, Guido Vanden (1994). Prolegomena. Distribution and reference of infinitival subjects. Berlin: Mouton de Gruyter.

Ergativity A perspective on recent work Alana Johns

1. Introduction Ergativity can be defined as a property of a language where the subject of the intransitive verb has something in common (e.g. Case, agreement) with the object of the transitive verb. In contrast, the subject of the transitive verb does not share this property. An example from the Inuit language is given in (1). (A list of abbreviations and terms used in examples in this article can be found before the bibliography.) (1)

a.

anguti-up tuktu

taku-jaa

transitive

man-ERG caribou-ABS See-TRANS.PART.3S/3S

b.

'The man saw the caribou' tuktu niri-juq

intransitive

caribou-ABS eat-iNTR.PART.3s

'The caribou is eating' In the transitive example in (la) the agent anguti 'man' is marked with ergative -(u)p Case (in this language termed relative), and the object tuktu 'caribou' is marked with absolutive Case. In the intransitive (lb), the subject of clause tuktu 'caribou' is marked with absolutive Case. A s is standard, absolutive Case is phonetically null. The two examples in (1) are said to display ergativity according to the above definition, since absolutive Case is shared by both the object of the transitive in (la) and the subject of the intransitive in (lb), but is not found on the subject of the transitive verb in (la). The definition of ergativity given above assumes a traditional definition of subject, object, and transitivity. Within the theory of generative grammar, no

48

Alana Johns

such elements exist axiomatically, but are based on structural concepts. The generative grammarian must ask certain basic questions. What do we mean when we say that anguti 'man' is the SUBJECT of a transitive verb in (la) and that tuktu 'caribou' is the OBJECT of the transitive verb? In what sense is the verb takujaa, a verb which is morphologically marked with the features of two arguments TRANSITIVE? This article will be limited to a discussion of these and related issues concerning ergativity within, roughly speaking, generative grammar; however the reader is referred to Dixon (1994) and references therein for a guide to wealth of insights and data from other points of view, which, unfortunately, for reasons of space, we will not examine here. Without an a priori notion of subject, object, and transitive as primitives, a number of interesting questions arise concerning examples such as those in (1). Does the fact that, in both examples, tuktu 'caribou' is marked with the same Case indicate that the structural position is the same in both instances? If so, what is that position — SpecTP, SpecVP or something else? If the two absolutive NPs do not have the same syntactic position, how do we explain the fact that different structural positions may receive the same Case, i.e. what is the relation between Case and configuration? Another question is whether ergativity is one phenomenon universally, e.g. a parameter, or whether there are different types of ergativity? If there are different types of ergativity, is there a finite number of types? For example, it has been observed (Anderson 1976; Dixon 1979,1994) that some ergative languages treat both absolutive marked NPs as the same grammatical function, e.g. restrict relativization to arguments which take absolutive Case. Such languages are said to show syntactic ergativity. In contrast, in the majority of ergative languages, there are some syntactic differences between the absolutive NP associated with the intransitive verb and the absolutive NP associated with the transitive verb. In such languages, where absolutive Case does not reflect a unique grammatical function, the language is described as having morphological (i.e. superficial) ergativity. In spite of their morphology, such languages appear to operate on an underlying nominative-accusative basis, i.e. the subject of the intransitive verb and the subject of the transitive verb group together in contrast with the object of the transitive verb. Whether or not the distinction between syntactic and morphological ergativity is binary is an open question. It may be that there is a continuum of ergativity (i.e. microparameters rather than a macroparameter). Where syntax and morphology diverge, the issue of which syntactic properties operate independently of the ergative morphology (see Section 4 below) may provide some clues to our understanding of ergativity.

49

Ergativity 2. Early

issues

The earliest discussions of ergativity in generative grammar are those of Hale (1970), Anderson (1976) and Chung (1977). These works deal with a perennial issue within ergativity — the relation of the transitive clause in an ergative/absolutive system (which I will call here the ergative clause) to the passive. In both Australian and Polynesian languages, there is evidence that the ergative and passive clauses are historically related. The similarity between a passive clause and an ergative clause is readily perceived. Both have a patient in the unmarked Case, the same Case as that assigned to single arguments of intransitive verbs; both have the agent assigned a marked Case (ergative or oblique). The main differences between the two clause types is that the passive is non-active while the ergative clause is active, and that the ergative NP is central to the syntactic structure of the clause, while the oblique NP in the passive clause is typically optional, if not actually prohibited. The question now becomes how a language changes a passive into an ergative clause (see Estival & Myhill 1988), and whether or not there is any evidence that passive is synchronically involved in the ergative clause. Hale (1970) entertains three hypotheses. The first of these is that there exist a) pseudo-ergative languages where the agent of the transitive verb (although it is the deep subject) can only appear in the surface within VP, and the passive transformation must take place, moving the object from within VP to subject position. The other two types of ergative languages have undergone some sort of historical reanalysis. The less common is b) passive-ergative languages where the deep structure rules are reconfigured such that the agent is generated within the VP and the object is generated in subject position directly. The final type of reanalysis (and the more common) is that of c) active-ergative languages, where the agent is basegenerated in subject position and given ergative Case, while the object gets nominative Case within the VP (as does the subject of an intransitive verb). Passive does not take place in this type of language. Under Hale's classification, types b) and c) are true ergative languages, since either i) active or ii) passive structure is absent from the grammar. Thus only type a) will have a dual notion of subject (deep and surface). This approach to ergativity is quite interesting for the variety of paths to ergativity which it explores, seen later in the heterogeneous approach of Bittner & Hale (1996a, b). Chung (1977) argues that a passive-to-ergative reanalysis can take place gradually. This would seem to imply a compositional process, whereby different components or features may be affected independently (to some degree). During this historical change, where an oblique agent becomes a

50

Alana Johns

syntactic subject, more superficial rules (e.g. Relativization, Clefting, Question Movement) are affected earlier in a change than "deeper" rules (e.g. Raising, Equi). The fact that different types of syntactic rule types pick out different notions of subject is a longstanding issue in the discussion of ergative languages. Anderson (1976, 1977, 1992) points out that the majority of ergativity languages display morphological rather than syntactic ergativity, thus the more familiar notion of subject predominates. He therefore considers ergativity to be essentially a superficial phenomenon. He argues that while historically many of the superficially ergative languages developed their ergative clauses from passive (and some later become perfective), or possessive constructions, synchronic analyses have reanalysed the ergative morphology as markers independent of their historical source, e.g. special marking on transitives, or perfective marking.

3. Recent

analyses

Beginning with Marantz (1984), there has been a constant and growing interest in ergativity within generative grammar. This interest is based no doubt on the challenges which an ergative systems poses for Case and agreement analyses. Another reason for the renewed interest is that languages with complex morphology are now being analysed in more detail than was possible in earlier versions of the theory, where rules rather than morphemes (or features), were the prevailing means of description. The reason that rules do not lend themselves to morphology is that the rules not only had to capture the syntactic generalization, but had to introduce the morphemes as well, thus making languages with morphology appear more complex than languages with no or little morphology. To those that work with morphological languages, the reverse seems to be true; languages with complex morphology signal subtle and general facts about the language. The introduction of thematic roles and functional categories to the theory made the connection between morphology and the grammar more salient. Finally, the notion of covert vs. overt morphology and feature strength (Chomsky 1995) has focussed even more on morphology (see also Halle & Marantz 1993). At present there are many, many versions of ergativity within generative grammar (see Manning 1996 and Ura 1998 for very good discussions of some of the issues). These analyses are concerned with synchronic aspects of ergative systems, and the majority of them assume that within universal grammar ergativity is determined by few, possibly only one, property (i.e. parameters). In Johns (1992), I argue that ergativity is epiphenomenal in

51

Ergativity

Inuktitut. Implicit in that claim is the view that ergativity is a label which we apply to a configuration of surface properties, but that the grammatical properties which underlie what we call ergativity may be numerous and unrelated. From this perspective, there is little reason to study ergativity in general, any more than there is reason to try to explain, for example, all V2 languages, although naturally some V2 languages will have quite similar analyses. The important thing is to account for observed V2 properties, not to develop a theory of V2. In the following discussion, I will survey a number of analyses of ergativity, with a view to identifying those properties which have been proposed as making ergative languages different. These proposals often cut across the above-mentioned distinction between syntactic and morphological ergativity, since not all analyses deal with, or even allow, the distinction.

3.1. Thematic differences With the introduction of the idea that thematic roles were assigned to certain positions at D-structure, it became possible to conceive of the question as to whether all languages assigned thematic roles in the same fashion. Marantz (1984) proposes that syntactically ergative languages in fact assign thematic roles in a mirror image to the way that other languages assign thematic roles. Thus where most languages assign the agent role to the subject of a transitive clause, and the patient role to the object, Marantz claims that syntactically ergative languages assign the patient role to the subject and the agent role to the object. In some ways, this proposal resembles the ergative as passive analysis, except that the agent role is the direct object of the verb, rather than an adjunct. Marantz also proposes that morphologically ergative languages are like nominativeaccusative languages with special Case assignment properties. As evidence for both types of ergative language, Marantz analyses Central Arctic Eskimo as syntactically ergative and Greenlandic Eskimo as morphologically ergative. Johns (1984, 1987) argues that both types of Eskimo are the same syntactically, thus weakening the evidence for a distinction along these lines. Johns further argues on the grounds of learnability that having agent roles assigned to direct objects should not be permitted within universal grammar. Levin (1983) provides a more detailed investigation of Marantz's proposal, showing that of the three ergative languages she investigates, Warlpiri, Dyirbal and Basque, only Dyirbal is syntactically ergative (i.e. patientsubject, agent-object), while the other two are morphologically ergative. Basque, in addition to being ergative, poses a challenge in that intransitive verbs do not all assign absolutive Case to their subjects, as does the Inuit

52

Alana Johns

language (see example ( l b ) above). Instead, some Basque intransitive verbs assign absolutive Case to their subjects and others assign ergative Case to their subjects, as shown in (2), examples from Levin (1983). (2)

a.

Mirenek

ni

ikusi nau 1S.N0R-UKAN-3S.N0RK

Miren-NORK I-NOR see

b.

'Miren saw me' Ni etorri naiz I-NOR come

c.

Ί came' Jonek

lS.NOR-IZAN

jäten

du

Jon-NORK eat-PRES 3S.NOR-UKAN-3S.NORK 'Jon eats'

d.

Nik

lan

egin dut

I-NORK Work-NDEF/NOR do

3S.N0R-UKAN-lS.N0RK

Ί worked' As we can see in (2a), in Basque the agent in a transitive verb gets a special Case NORK (ergative), while the patient of a transitive verb gets the NOR (absolutive) Case. Example (2b) shows the typical properties of an ergative language, the subject of the intransitive verb being assigned the same Case, in this instance NOR, as the patient in (2a). In contrast, the subject of the intransitive verbs in (2c) patterns with the agent argument of (2a), i.e. gets assigned NORK. Levin proposes that Basque assigns Case on the basis of D-structure relations. As a result, an unaccusative verb like (2b) is assigned a Case which indicates it is a D-structure object. Verbs which pattern like (2c,d) mark that the subject is a D-structure subject. Moreover there is some evidence that these latter verbs are not intransitive, e.g. the double agreement. Hale & Keyser (1993) propose that unergative verbs universally involve incorporation of a nominal object at the level of lexical structure. Laka (1993), following traditional grammars, argues for an analysis where Basque differs in that the object does not incorporate, thus explaining why (2d) patterns like the transitive (2a) (see Nash 1995, 1996 for a similar analysis of Georgian unergatives). In contrast, Austin & Lopez (1995) discuss arguments against non-incorporation. Another proposal involving thematic roles is that of Manning (1994, 1996), who, working within an LFG approach, argues for a level of argument structure (a-structure), having a type of subject. Α-structure operates in parallel to the syntax (gr-structure), which has its own type of subject. This allows for an unfamiliar alignment between arguments and grammatical functions, similar to that proposed by Marantz for syntactic ergativity, but at the same time keeps both notions of subject active, thus allowing different syntactic processes to pick out different subjects.

53

Ergativity

Nash (1996) argues that ergative languages are different from nominative/accusative languages in that in the former, the agent of a transitive clause is merged VP internally, while in the latter, it is merged as part of a higher projection headed by a functional category (e.g. v) (see also Nash 1995 for an analysis of Georgian ergativity). Thematic explanations, to the degree which they are successful, have been associated with languages with morphological ergativity. 3.2. Case

differences

In an ergative Case system where the absolutive Case is both more common than the ergative Case (found on two, rather than just one, grammatical functions), and is usually phonetically null, it is an obvious question to ask whether or not absolutive Case is in fact nominative Case. For those who think that absolutive is indeed nominative, the analysis must become more structural, since the patient of the transitive verb is assigned absolutive (see 3.3 below). For those who think that ergative Case is equivalent to nominative, a different sort of analysis is required. Within such analyses the intransitive absolutive becomes anomalous, recalling Levin's (1983) analysis of Basque unaccusatives. Levin & Massam (1985) focus on Case properties and the notion of morphological ergativity, which they refer to as surface ergativity, and leave syntactic ergativity aside. They assume structural positions to be standard across languages but Case assignment not to be. For Levin & Massam the structure of a transitive clause in both ergative/absolutive and nominative/accusative languages is the same, i.e. agent in SpecIP and patient as complement of V. The difference lies in the fact that whereas in a nominative-accusative language, the default Case (nominative) is assigned by I, in an ergative-absolutive language the default Case (absolutive) is assigned by V. A further assumption is that the default Case is obligatorily assigned. Under this view then, ergative and nominative are similar in that they are both assigned to SpecIP, while absolutive and accusative are similar in that they are both assigned to NP-inside-VP. Yet from another point of view, absolutive and nominative are similar in that they both are default. Ergative and accusative are similar in that they are both non-default. Levin & Massam propose that the non-default Case of any system (i.e. accusative and ergative) must be assigned under theta-government, i.e. assignment of these Cases is not purely structural. This innovation, although not without problems with respect to exceptional Case marking (as the authors point out), is one of the first attempts within generative grammar to formalize the marked nature of the accusative or ergative as

54

Alana Johns

compared with nominative or absolutive (see also Marantz 1991, and Bittner & Hale's notion of Case competitor below). A later version of the Case approach within an early minimalist framework can be found in Bobaljik (1992, 1993) who proposes that all ergative languages are morphologically ergative, i.e. that syntactic ergativity does not exist. Like Levin & Massam (1985), Bobaljik claims that languages will pick a default Case, the Obligatory Case Parameter, whereby a language will obligatorily assign one structural Case, either nominative or absolutive. One of the more interesting aspects of this work is the introduction of non-finite clauses into the discussion of ergative languages. Bobaljik argues that if he is correct and the ergative NP is checked for Case by the higher agreement element Agrl, then it is the ergative argument and not the absolutive one which will not receive Case or trigger agreement in nonfinite clauses, just as it is the nominative NP which does not receive Case or trigger agreement in non-finite clauses in nominative-accusative languages. Thus in control structures (see Section 4 below), we would expect the absence of ergative Case. The accuracy and significance of this prediction are a matter of some dispute (see Manning 1994, Wharram 1996 and Johns & Smallwood 1999). Massam (1994) argues against functional categories as the source of Case, proposing instead that Case is assigned to both arguments within the VP on the basis of semantic properties (aspectual or thematic). Absolutive Case is obligatory in ergative languages ä la Levin & Massam. Mahajan (1994) provides a Case account of ergativity, where the ergative Case on the subject results from the non-incorporation of an oblique Ρ which is the source of auxiliary have (be + P) in more familiar languages (following Kayne 1993). This explains the fact that ergative case is often found where other languages use have, e.g. perfective (see Section 5). Austin & Löpez (1995) propose that languages differ as to the position of the obligatory unmarked Case assigner: a) highest spec (= nominativeaccusative) b) middle spec VP (= ergative/absolutive Inuit type) or c) lower spec VP (= ergative absolutive Basque/Georgian type). Ergative languages with type c) Case marking are predicted to display split intransitive marking (see 2b, c above), to mark indirect objects with dative Case only, and not to have antipassive. 3.2.1. Woolford (1997) Woolford (1997) utilizes the full range of Case system possibilities, arguing (a) that ergative is always lexical (inherent) Case, (b) that objects can get Case through lexical Case or one of two structural positions (within or just above VP). She points out that three-way Case systems have been proposed,

55

Ergativity

where subjects of intransitive verbs are assigned nominative Case, subjects of transitive verbs are assigned ergative Case and objects of transitive verbs are assigned objective Case (see Blake 1994 and Bittner & Hale 1996a). Further Woolford proposes that Nez Perce and Kalkatungu exhibit a four-way system, where, in addition to objective Case, objects of transitive verbs can be assigned accusative Case. Case assignment depends on where the assigning/checking takes place in the tree, but it also depends on general systemic properties. It is likely that a rigorous minimalist version of this approach could build in some these properties, e.g. that an object with accusative Case can never cooccur with an ergative subject. The claim that ergative is always a lexical Case, similar to that of dative subjects, (also made by others, including Mahajan 1989 and Nash 1996) has merit in the sense that ergative Case is much more restricted semantically than nominative or absolutive. Indeed an often overlooked fact about Inuktitut is that ergative Case and agreement are only one manner by which to license the external argument of a verb taking two arguments (see Section 7). The Case approach advocated by Woolford appears somewhat rich. There are numerous positions in which various Cases are assigned, and there is also the distinction between lexical and structural Case. Both structural and lexical Case can be assigned to the same position in at least one instance. While objective Case is assigned outside of VP, accusative structural Case is assigned by the V to its object and accusative lexical Case is assigned to themes, presumably also in object of V positions. Finally, there is at least one generalization (see above) which is required independently to explain Case combination restrictions. 3.3. Structural

differences

Changes in generative grammar made available other analyses of ergativity. In particular, a reexamination of phrase structure led to accounts where the structure of the transitive clause in ergative languages could be analysed as being somehow different from that in non-ergative languages. As we will see, there have been numerous analyses along these lines. What these accounts have in common is a reappraisal of what constitutes a transitive clause in these languages. As we shall see, in many of these accounts the "transitive" verb does not in fact have a complement at S-structure (see also Massam 1992 for Niuean). Under this revised view, we are dealing with two arguments united by a single predicate structure; however the structure is not that of the traditional (SpecIP; NP-inside-VP) of generative analysis. Bok-Bennema & Groos (1984) and Bok-Bennema (1991) propose that a

56

Alana Johns

transitive verb in a syntactically ergative language is unable to assign Case to its object (see also Payne 1980). An example of such a language is the Inuit language. This property is similar to but distinct from passive. One means of assigning Case to the object is to assign it oblique Case, as in an antipassive (see 7. below); another strategy is to move the object to a position where it will get absolutive Case from I. In the Case of the "transitive" clause, Bok-Bennema (1991) argues that the patient in absolutive Case adjoins to I' below the ergative Case marked element. Ergative (relative) Case is assigned to the agent NP in SpecIP by a feature ^genitive) in the transitive mood. The fact that only transitive verbs assign ergative Case (and not intransitives, which assign absolutive Case) must be stipulated for the transitive I. 3.3.1. Functional categories and structure The advent of functional categories allowed even further elaboration of structure. In the following discussion I will refer to the generic "transitive" tree in (3), and discuss the proposed values for the functional categories, and some of the reasoning behind the analyses. (3)

ZP

ZP = TP (Murasugi), CP (Bittner and Hale) AgrlP (Bobaljik), AgrS (Campana) AgrPv (Johns)

Ζ' YP

YP = TrP (Murasugi), IP (Bittner and Hale) Agr2P (Bobaljik), AgrO (Campana) Y' AgrPN (Johns) VP NP1

V' NP2

This tree structure can form the basis of comparison for the analyses of Johns (1992), Campana (1992), Murasugi (1992a), Bobaljik (1993), and Bittner & Hale (1996a, b). For all these analyses, with the exception of Johns (1992), who deals with thematic roles at the level of argument structure, the patient argument (NP2) is generated as the complement of V and the agent argument (NP1) is generated in SpecVP position, as shown in (3). For all analyses there are two functional projections above the VP. Differences between analyses lie

57

Ergativity

in the nature of the functional projections, and which N P (1 or 2) (if any) moves to which Spec position (SpecZP or SpecYP). 3.3.2. Johns (1992) Johns (1992) analyses ergativity for Canadian dialects of the Inuit language, referred to as Inuktitut. Under the assumption that ergativity is simply a patterning which receives a label, a number of possible sources of this patterning are potentially possible. In the Inuit ergative clause, a passive participle on the verb prevents the verb from assigning Case (see also Bok-Bennema & Groos 1984). The participle being nominal in category allows a nominal specifier (SpecYP). The head Y is a nominal functional category (Agr N ), which could just as well be D. Above Y P is a predicative functional category ZP, the head of which is Ζ (Agr v ). The nominal nature of Y accounts for why ergative Case in the Inuit language is the same as possessive Case, and why the order of agreement is agent agreement (Agr N ) internal to patient agreement (Agr v ). In addition, it is claimed that relative clauses consist only of the lower level of the structure (YP). The final feature of the analysis is that head movement of V through Y to Ζ forces SpecYP to adjoin to ZP thus explaining word order and binding properties in the language (see 4.1 below). The fact that the position and properties of the participle change through movement between Y and Ζ is a synchronic account of the passive to ergative issue and has implications for how diachronic change may take place (recall Hale 1970 argues passives are only a diachronic source in some Australian languages). 3.3.3. Campana (1992) Campana (1992) proposes a more general theory of ergativity (which particularly applies to Mayan, Dyirbal and Chamorro) in which Y is AgrO and Ζ is AgrS. Agents NPs move to SpecYP to have their Case checked while patients adjoin to ZP. Campana develops a theory of movement in these languages designed to capture the well-known restriction that ergative NPs cannot undergo W H movement from transitive clauses (recall Chung 1977), as shown in the Mam examples (from England, cited by Campana) in (4). (4)

a.

alkyee-qa xhi tzaj who-PL

b.

'Whom did Jose grab?' *alkyee saj who

t-tzyu-7n

Cheep

rec DEP/3PA DIR 3sE-grab-DS Jose

t-tzyu-7n

kab' xiinaq

REC DEP/3SA/DIR 3sE-grab-DS two man

'Who grabbed the men?'

58

Alana Johns

According to Campana, the problem in (4b) is that at LF, (where he consider Case checking to take place), the absolutive argument has adjoined to ZP, thus preventing the moved ergative N P in COMP from antecedentgoverning its trace. 3.3.4. Murasugi (1992a, 1997) For Murasugi (1992a), working within an early version of the minimalist framework (see Chomsky 1995), ZP is T P and Y P is a TrP (transitive phrase), which perhaps corresponds to υ within the current minimalism framework (Chomsky 1995). NP1 moves to the specifier position of TrP, while NP2 moves to the specifier position of TP (although not until LF). The motivation for using TrP rather than simply AgrOP stems from the fact that a number of ergative languages exhibit morphemes signalling transitivity, not just agreement forms. For instance, as Jelinek (1993, p. 22) points out, Straits Salish transitive verbs display a transitivizing suffix following the root, as shown in (5) (Straits Salish). (5)

leg-t-ogal·

=la

=sxw

see-TR-lP.ACC =PAST =2s.NOM

Tou saw us With this analysis, Murasugi explains the properties of the transitive clause discussed above (relative clauses, agreement order, etc.). The motivation for N P 1 moving to SpecTrP rather than SpecTP is based either on the strength of the head TrP (strong in ergative languages; weak in accusative languages) as in Murasugi (1992a, b), or on the claim (Murasugi 1997a) that the verb in syntactically ergative languages does not assign Case to its object, with the result that shortest move determines that the first available N P (1) moves to SpecTrP. An interesting area of discussion in Murasugi (1992a) concerns non-finite clauses in ergative languages (Mayan, Lezgian, the Inuit language, Abkhaz and Dyirbal). She argues that C can exceptionally have a feature +finite which can check the Case of an N P in SpecTP in a -tense clause, such that some ergative languages exceptionally Case mark the absolutive (intransitive subject/transitive object). This marking parallels that of Portuguese inflected infinitives in accusative languages. The examples in (6) are Lezgian (cited Murasugi 1992a, p.c. from Haspelmath) (6)

a.

didedi-z ktab stold-a xa-na k'an-zawa mother-DAT book-NOM table-INESS be-NONFlN want-PRES 'Mother wants the book to be on the table'

59

Ergativity b.

didedi-z gagadi ktab qacu-na k'an-zawa mother-DAT boy-ERG book-NOM buy-NONFIN want-PRES 'Mother wants the boy to buy the book'

Here it is claimed that both the object in (6b) and the subject in (6a) have their Case checked by C. While both the analysis of the facts and the claims resulting from that analysis are disputed (c.f. Bobaljik 1993; Wharram 1996), these constructions deserve further research and discussion within ergative languages. Murasugi (1997) continues her research into non-finite clauses in ergative languages in an analysis which explains the restriction that the ergative argument cannot be relativized (see 3.3.2 and 3.3.3) on the basis that it is a reduced (participial) clause. Ura (1998) provides the most recent analysis along these lines in the Minimalism framework, using multiple feature checking in an Agr-less model. 3.3.5. Bittner & Hale (1996a and b) In an important and influential pair of articles, Bittner & Hale (1996a, b) present an extensive formal treatment of Case, agreement and structure, using a wide variety examples from many languages. The range of theoretical and empirical discussion makes these articles somewhat challenging to read. Bittner & Hale's account could be classified here either as a structural account, or as a Case account (see Section 3.2 above), since one of the most prominent features of their analysis is the claim that syntactic Case results from structural configuration. This treatment is quite complex, and Bittner & Hale (1996a) is devoted entirely to the theoretical underpinnings, while Bittner & Hale (1996b) illustrates in detail how this approach explains a number of types of ergativity. The basic idea is that both types of ergative languages (syntactic and morphological) will have a VP structure identical to that of nonergative languages. Ergative languages in general are unable to assign Case to their object (following Bok-Bennema & Groos 1984). In terms of the generalized structure given in (3) above, Bittner & Hale propose that, in syntactically ergative languages, NP2 moves to SpecYP, with NP1 remaining in situ. In morphologically ergative languages both NP1 and NP2 remain in situ within the VP. Syntactically ergative language are what they term opaque and the problem of Case assignment to the object of the transitive verb must be dealt with external to the VP. As a result, the object must raise into SpecIP to get null (nominative) Case from C. The movement of an NP into SpecIP triggers the assignment of a Case to the NP in SpecVP under a notion of Case competition, which is at the heart of their analysis of Case assignment.

60

Alana Johns

The definition of Case competition is quite technical, but the intuitive notion seems to be that a certain configuration of two NPs will result in one of those NPs getting assigned a marked structural Case — either accusative or genitive/ergative (see also Marantz 1991). This idea might be related to Hale & Keyser (1986) where the notion of constructionally determined linguistic properties is briefly mentioned, and to later versions of this notion, most notably Hale and Keyser (1993) who propose that thematic roles are the products of structure. It is also related to Burzio's generalization where accusative Case and external argument are crucially dependent on one another (see also Nash 1996 for a similar analysis based on the presence of something like υ). Morphologically ergative languages, such as Warlpiri, in Bittner & Hale's account are what they term transparent in that the arguments within VP are licensed in situ (see also Massam 1994 and Nash 1996), and the ergative Case assignment to the subject of a transitive verb results from Case competition with the object. This entails that morphologically ergative languages display nominative-accusative syntax, since the arguments remain where they were generated, subject of transitive in SpecVP, transitive object in NP, VP. Transparency is achieved via coindexing of C, V and I (I to C movement, and then binding of V by I). Since these are the crucial factors that distinguish Warlpiri from the Inuit language, it is important that independent evidence be provided. The main overt difference between the two languages is that Warlpiri has auxiliaries (which they attribute to I to C movement), and independent pronouns, as in (7a) (Bittner & Hale 1996b, p. 554, (35b)). In (7b), I provide an example from the Inuit language. (7)

a.

Ngajulu-rlu ka-rna-ngku nyuntu nya-nyi me-ERG Ί s e e you'

b.

PRS-1SG-2SG y o u

taku-vait see-TRANS.IND. Ί see you'

Warlpiri

see-NPST

Inuktitut

ls/2s

Bittner & Hale discuss further points of comparison, showing how the differing mechanisms underlying ergativity (object raising vs. transparency) in each sample language (the Inuit language vs. Warlpiri) leads to different syntactic properties. Differences lie in (a) structural Case assignment, (b) pronominal agreement, (c) minimal scope options (see Section 6 below), and (d) binding relations (see Section 4 below). In the area of structural Case assignment, Warlpiri antipassives differ from similar constructions in the Inuit language, in that the transparency of the VP in Warlpiri allows the object to function as a Case competitor, thus creating the circumstances (somewhat mysterious) for ergative Case

61

Ergativity

to be assigned to the transitive subject, as in (8a) ((45b) in Bittner & Hale 1996b), while in the Inuit language, the transitive subject will get assigned absolutive Case, as in (8b) (their (28b)). (8)

a.

Ngajulu-riu ka-rna-rla me-ERG

b.

jurlarda-ku wapaZ-paka-rni Warlpiri

PRS-1SG-3SG honey-DAT APASSv-chop-NPST

Ί am chopping for honey Juuna miiqqa-nik paari-n/up-p-u-q

W. Greenlandic

J u u n a child-PL.INS look.after-APASS-IND-(-TR)-3SG

'Juuna is looking after the children' With respect to agreement, the Inuit language and Warlpiri are shown to differ in that only the latter picks out a nominative-accusative version of subject. Bittner & Hale claim that a positive feature of their analysis, is the fact that they avoid "reducing" possessed nominals and transitive clauses to one another. The parallels between these structures have long been the subject of much debate in the linguistics of Eskimo-Aleut (see references cited in Bittner & Hale 1996b, footnote 20). Nevertheless, built into their analysis are numerous stipulations that achieve the same end, e.g. that the passive head in the Inuit language is N, and that the passive head can be reanalysed (only diachronically for them) as the head I (Bittner & Hale 1996a, p. 33). Further, they stipulate that the nodes I and D appear in structurally parallel configurations, and that they both assign ergative Case.

4. Control and binding As discussed in Section 3.3, syntactic properties in some ergative languages led to analyses where the patient role is found in a position higher than the agent. An outstanding problem, however, for these structural accounts of syntactic ergativity is the fact that, at the same time, most of these languages have grammatical properties where the agent appears to be higher or more prominent than the patient. Control and binding are areas where this is the case. Within an analysis based on some version of morphological ergativity, this problem does not exist, since the syntax will resemble that of nominative accusative languages. Control is a complicated issue in ergative languages, primarily because a PRO/pro analysis is often used as an argument rather than as something to be argued for. To what extent these language have infinitivals and whether or not certain predicates are associated with embedded clauses with PRO subjects needs further research. As mentioned above, Murasugi (1992a) takes an initiative in examining these constructions in detail. For

62

Alana Johns

analyses of ergativity featuring non-finiteness as an issue, see Bobaljik (1992,1993), Bittner (1994), Bittner & Hale (1996a, b) Wharram (1996) and Johns & Smallwood (1999).

4.1. Binding In languages which display both syntactic ergativity and switch reference phenomena (called obviation by Bittner & Hale), the ergative subject and not the patient of the transitive clause (which is in a higher position) is the antecedent for binding, as shown in (9) (Inuktitut). For an analysis of this phenomena, see Bittner (1994), and for an analysis and overview discussion of a number of other analyses, see Manning (1996). (9)

a.

anguti-up irni-nt

taku-jaa

man-ERG son-3REFL.ABS see-TR.PART.3s/3s

b.

c.

'The manj saw his; son' anguti-up irni-α taku-jaa man-ERG son-3s.ABS see-TR.PART.3s/3S 'The man; saw hisj son' *irni-mi anguti taku-jaa son-3REFL-ERG man.ABS see-TR.PART.3s/3s

'His son; saw the man;' We see here that binding facts suggest that the ergative argument is higher than the absolutive argument in a transitive clause, even though a number of the analyses we have just seen have the absolutive argument move to a position higher than the ergative argument (e.g. those of Johns, Campana, Murasugi and Bittner & Hale) There are a number of positions in response to this issue. Johns (1992) claims that the agent N P in an Inuit transitive clause adjoins to the highest projection because of the movement of the lexical element with which it is linked to a higher position. This also accounts for the unmarked SOV ordering in transitive clauses. Bittner (1994) and Bittner & Hale (1996a, b) account for the facts in (9) through a formalism of coindexation and subject path, which achieves the desired results (and leaves unexplained SOV ordering). Manning (1994, 1996) argues that there exists two levels of structure: grammatical and argument. While at the level of grammatical structure the absolutive is the subject, at the level of argument structure, the agent and the theme pattern together. Binding works on the latter level, thus accounting for the facts in (9). Manning also points out that his analysis is quite similar to that of Bittner (1994) in that the subject path account reconstructs into the VP which is equivalent to an argument structure account.

63

Ergativity

For Bok-Bennema (1991) and Bobaljik (1993), who have analyses where the ergative argument is higher in the tree than the absolutive, the facts in (9) follow.

5. Split

systems

One property that has been noted by researchers in ergativity is that no language seems to be wholly ergative (see Dixon 1994). All ergative languages display some syntactic properties that work on a nominativeaccusative basis, e.g. control and/or binding (see Section 4 above), or else they display nominative-accusative marking in a sub-section of the grammar, e.g. the pronominal system (Dyirbal). A language may alternate between nominative-accusative and ergativeabsolutive based on some factor, e.g. person features, aspectual class, etc. Thus the heterogeneous nature of ergativity discussed in Bittner & Hale (1996a, b) is truly widespread. While much research has focussed on the languages which display a strong degree of ergative syntax, such as the Inuit language and Dyirbal, equally interesting are the many languages within which ergativity appears to be linked to some other property of the grammar. Some generalizations exist; for example, it seems that when one language displays both nominative/accusative and absolutive/ergative systems based on an aspectual split, the absolutive/ergative system, not the nominative/accusative, is associated with the past or perfective, e.g. in Hindi (see Mahajan 1994). This association remains to be explained, although it naturally is linked with the issue of passive and ergativity, since passive and perfective are also related. A final area is that of split intransitive systems (sometimes called split S). These are systems where some subjects of intransitive pattern like the object of a transitive verb, while other intransitive subjects pattern like the subject of a transitive verb. In this way, the intransitive split resembles the unaccusative/unergative dichotomy, although the verb class distinction often appears to be based on lexical semantic factors, such as inherent control of the action. As Dixon (1994) points out, there also exists what he calls fluid-S systems, where the choice of intransitive verb class is determined by situation, not lexical class. For example, the marking on an intransitive verb marking might reflect absence or presence of volition. Falk (1997), in a theoretical approach which aims to incorporate much of the typological findings (notably Dixon 1994), discusses the importance of animacy or definiteness in split systems, claiming that these are discourse factors, and that they determine the absence of accusative Case just as

Alana Johns

64

similar factors determine the presence or absence of accusative in languages like Turkish and Hebrew. It seems likely that the alternations of split ergativity will provide insights into what types of factors underlie various systems. For a discussion of split ergativity, see Dixon (1994), Campana (1992), Massam (1992), Phillips (1993), Jelinek (1993), Mahajan (1994), Austin & Löpez (1995), Nash (1996), Bittner & Hale (1996a), Woolford (1997) and Falk (1997) and Ura (1998). Split systems, like all ergative systems, call out for careful analysis within the confines of generative grammar. Bobaljik (1998) convincingly argues that what appears to be a split agreement pattern in the agreement systems of Chukotko-Kamchatkan languages is actually nominativeaccusative, i.e. that the "ergative" properties of the agreement system result from independent grammatical factors.

6. Scope, specificity, etc. One of the more interesting claims in Bittner (1994), and Bittner & Hale (1996b) is that the raising of object argument to a higher position in syntactically ergative languages, as in (10) (West Greenlandic, Bittner & Hale 1996b, (67)), predicts scope readings which are significantly different from those in nominative-accusative or morphologically ergative languages. (10)

Ullumi Juuna-p atuartuq ataasiq uqaluqatigi-nngi-la-a today Juuna-ERG student one talk.to-NEG-IND-3SG.3SG There is one student that Juuna did not talk to today'

The object of the verb atuartuq ataasiq 'one student' (which is in absolutive Case) has only wide scope interpretation with respect to the negator -nngiinside the verb. Thus the missing interpretation here is the one where Juuna did not talk to a (any) student. The restriction to wide scope for objects of transitive verbs (in absolutive Case) contrasts with the ergative N P (which they show through other examples can take both wide and narrow scope). They claim that the restriction on the absolutive argument shown in (10) parallels that of English subjects, which, as discussed in Bittner (1994), normally take only wide scope (English One student didn't talk to Juuna cannot mean 'No student talked to Juuna' while the English Junna didn't find a book can mean either that there is a book that Juuna didn't find, or that Juuna didn't find any book). Warlpiri, with the object getting Case in situ, does not have the wide scope restriction for objects. These facts are challenging to collect in other dialects of the Inuit language (thus attesting to Bittner's fieldwork skills). In particular, for

Ergativity

65

Canadian dialects, I have not yet been able tofindthe narrow scope reading claimed for ergative NPs (e.g where no student talked to Juuna), but get wide scope readings for both arguments. The problem of comparison is compounded by the fact that West Greenlandic obligatorily uses the indicative mood in all declarative clauses, a mood which in more westerly dialects is restricted variously to questions,first/secondperson, or existentials (Johns 1995). Working within a minimalist framework, Manga (1996a, b) proposes a somewhat different account, wherein specificity is the relevant factor in the movement of the object to the SpecTP. Manga focusses on differences between Bittner & Hale's Case-through-structure account, as compared with her own Case-checking account. Manga compares analyses of ergativity, but, given that the specificity claim is the most salient feature in her analysis (and seems correct to me at least), it is unfortunate that she does not make an in-depth comparison with the scope account. Manning (1996, pp. 84-98) provides an insightful overview of the merits of a scope vs. a specificity account for the Inuit language, concluding that neither account fully succeeds (see also de Hoop 1992). A provocative and recent account using this type of evidence is Dechaine & Manfredi (1998). Based on the work of Longobardi (1994), and Bittner & Hale (1996a, b), D£chaine and Manfredi argue, using only the interpretation of bare nouns as evidence, that Igbo exhibits an abstract ergative/ absolutive Case system. Igbo does not have overt Case, and has an SVO ordering. This contrasts with Mahajan (1994) and Nash (1996), who adopt the position that ergative languages cannot be SVO, and Dixon (1994), who states that no language relying solely on constituent order to indicate syntactic function has been attested which is both SVO and ergative. Nevertheless, Dechaine and Manfredi provide examples which show that the bare Ν object of a transitive verb, and the bare Ν intransitive subject of a derived intransitive verb (there are no underived intransitives in Igbo) are interpreted as generics, i.e. pattern together. This is shown by the italicized N's in (11). In contrast, the bare Ν subject of a transitive verb is existential, as shown by the non-italicized Ν in (11a). (11)

a.

b.

Ökä ta-ra 6kha Igbo, Transitive rat chew-T maize 'Some rat is corn-fed / Some rats are corn-fed' Oke ä-ta-a-la okhä ähun Igbo, Intransitive rat AGR-chew-DEL-PERF maize that 'That corn is rat-chewed'

While a wider range of examples from Igbo would make this analysis more convincing, it demonstrates that the term ergativity can be extended into

Alana Johns

66

areas beyond its traditional usage. Whether all asymmetries along these lines can only be explained as the result of Case configurations remains to be seen (especially since the examples in (11) are aspectually distinct). In this instance, the Case explanation would be that the object 'maize' in (11a) remains in situ, but that Case is not assigned to it by the verb. Instead, absolutive Case is assigned to the object by movement of the V to C. Since this language would be transparent according to Bittner & Hale's analysis (see Section 3.3.5), the scope facts will be differentfromthose in Inuktitut, and the object should only have narrow scope (like nominative-accusative objects) when additional factors make it existential. Evidence provided by Dechaine and Manfredi suggests that it does, but the same evidence would be consistent with a non-ergative analysis. No examples are given which would show whether or not the ergative subject also has narrow scope, which Bittner & Hale claim to be true of ergative languages (see also above).

7.

Antipassives

Related to the issue of scope and specificity, but forming a research area in its own right, is the topic of antipassive, those constructions often found in languages displaying ergativity. Antipassives are constructions wherein the object appears to be in something like an oblique phrase, while the subject is linked to an intransitivized verb. Like passive, the construction normally involves additional morphology, as in (12) (Dyirbal, from Dixon 1994 p. 13). (12)

quma

bural-rja-nyu

father-ABS see-ANTIPASS-NONFUT

yabu-gu mother-DAT

'Father saw mother' These constructions, originally treated within Relational Grammar, have been compared to passive, since both involve the displacement of the direct object from its canonical position. Baker (1988) analyses antipassive as a type of noun incorporation, where the antipassive morpheme is the incorporated object. Jensen & Johns (1989) propose that antipassive in the Inuit language involves affixation of a morpheme which is both an agentive nominalizer (like -er in English), and a predicate. Bittner & Hale (1996a), following Baker, analyse the antipassive as a base-generated adjoined nominal element N, with the exception of the Inuit language, which they analyse as incorporating (see Wharram 1998 for a semantic account linking Inuit antipassive with noun incorporation). They extend their analysis of antipassive to account for accusative Case. According to them, accusative Case is very similar to the oblique Case of antipassive. They propose that some languages, e.g. English, have an adjoined D rather than the antipassive

67

Ergativity

Ν. The core idea is that adjoined D or Ν serves as a Case competitor for assigning marked structural Case to either the object or oblique argument. Thus they maintain the it-takes-two-to-tango approach for possessives, ergative, and accusative/oblique objects. The importance of the antipassive construction within an examination of ergativity lies in the fact that it allows the language to have an alternative means of licensing two arguments of the same predicate within a clause. Bittner (1987) shows that the antipassive construction in the Inuit language has the object in a position of narrow scope (this equates to nonspecific in Manga 1996), while Bittner (1994) states that either scope is allowed, but that the narrow one is preferred where an alternate transitive (i.e. ergative) form is also grammatical. Manning (1996, p. 94) discusses the fact that the antipassive construction is sometimes obligatory in an ergative language, e.g. in the Inuit language where relativization is restricted to arguments in absolutive Case (see Section 3.3.2 and 3.3.3). As a result of this restriction, a relativized agent must obligatorily appear as the absolutive subject of an antipassive. Under these circumstances, it appears that the object may have wide or narrow scope. The nature and properties of antipassive morphemes themselves are clearly deserving of further research. Bittner (1987) suggests that in the Inuit language they have aspectual properties, and I believe she is correct. Such an analysis would form part of explanation for why so many activity predicates do not require overt morphology (although they are often analysed as having zero antipassive morphemes). 8. Discussion In Section 3 above, I put forward my view that ergativity per se is too typological (i.e. superficial) to have an underlying explanation, but is only a label for a certain configuration of properties, which themselves may be fundamental or, alternatively, superficial, within an individual grammar. Dixon (1994, p. 219) states that he "would suggest that there is no necessary connection between ergative characteristics and any other linguistic feature." I suggest that this viewpoint is supported by the numerous analyses based on as many languages extant. In other words, I think there is little value in studying ergativity as a thing in itself. More productive directions of research are issues such as the nature of features and structure, as discussed in Chomsky (1995) and Halle & Marantz. (1993), or the theory of Case and agreement as presented in Bittner & Hale (1996a). In spite of what I have just said, there do appear to be some areas of agreement surrounding the topic of ergativity. A number of linguists seem

68

Alana Johns

to agree that the verb in the ergative clause does not assign Case to the object in some languages. Related to this is the frequent view that there are two functional projections (or their equivalents) above VP, as shown by the generic tree given in (3) above (although this has become standard for nominative-accusative languages as well). There also seems to be a consensus that a nominal property is involved in the ergative patterning of some languages at least. Thus Bittner & Hale rely on an adjoined Ν to trigger accusative/oblique Case, and state that, in some languages, I is a reanalysed passive Ν (Bittner & Hale 1996a, p. 33). In (Johns 1992), I propose that the participle varies between a nominal and active verb interpretation based on structural position. Another area where there seems to be agreement is the need for a difference between a structural domain and a thematic domain (Guilfoyle, Hung & Travis 1992; Massam 1994; Manning 1994; and Bittner & Hale 1996a, b). Whether two domains, roughly above and below VP, are enough, remains to be seen. Finally there are three issues which I think are deserving of closer attention in the future. The first two are the issue of split systems, as discussed in Section 5, and the nature of non-finite clauses, as discussed in Section 4. The third issue is the relation between diachronic change and synchronic analysis, which was an element of early discussions (see Section 2), but has received little recent attention within generative grammar. As mentioned above, Bobaljik (1998) shows that the alleged split ergativity of Chukotko-Kamchatkan languages can be derived from other properties, and therefore epiphenomenal, but it is also interesting that he does so via an analysis which other linguists propose as diachronic (see Johns 1992 for a similar claim about Inuktitut). Generative grammar has always had the formal abilities to capture alternations within a single derivation, and it is surprising that there are not more SPE versions of synchronic ergative syntax. Related to all these issues is the need for more research and fieldwork on sets of closely related languages and dialects which display ergativity.

Abbreviations Inuit: ERG = ergative; ABS = absolutive; INS = instrumental Case; TRANS = transitive; INTR = intransitive; PART = participial mood; IND = indicative mood; APASS = antipassive; 3S/3S = third singular agent/third singular patient; 3 RE FL = reflexive agreement Basque: NORK = ergative Case; NOR = absolutive Case; UKAN = auxiliary Tiave'; IZAN = a u x i l i a r y 'be'; PRES = p r e s e n t ; NDEF = n o n - d e f i n i t e M a m : p ( L ) = plural; REC = recent past; DEP = dependent aspect; A = absolutive; Ε = ergative; DIR = directional; DS = directional suffix L e z g i a n : DAT = dative; NOM = nominative; INESS = inessive; NONFIN = nonfinite

69

Ergativity Warlpiri: NPST = nonpast Igbo: DEL= delimited; PERF= perfective. Dyirbal: ANTIPASS = antipassive; NONFUT = nonfuture

Acknowledgments Thanks to Elizabeth Cowper and Diane Massam. This work was supported by the Social Sciences and Humanities Research Council of Canada (grant number 410-94-570).

An Ergativity

Bibliography

Anderson, S. (1976). On the notion of subject in ergative languages. In Subject and topic, edited by Charles N. Li, 1-24. New York: Academic Press. Anderson, S. (1977). On mechanisms by which languages become ergative. Subject and topic, edited by Charles N. Li, 317-363. New York: Academic Press. Anderson, S. (1992).A-morphous morphology. Cambridge University Press. Austin, J. & L. Löpez (1995). Nominative, absolutive and dative languages. Proceedings of NELS 25, 1-15. GLSA, University of Massachusetts, Amherst. Baker, M. (1988). Incorporation: A theory of grammatical function changing. Chicago: The University of Chicago Press. Baker, M. (1996). The polysynthesisparameter. New York: Oxford University Press. Bittner, M. (1987). On the semantics of the Greenlandic antipassive and related constructions. International Journal ofAmerican Linguistics 53, 194-231. Bittner, M. (1988). Canonical and noncanonical argument expressions. Doctoral dissertation, University of Texas at Austin. Bittner, M. (1994). Case, scope and binding. Studies in Natural Language and Linguistic Theory. Dordrecht: Kluwer. Bittner, Μ. & K. Hale (1996a). The structural determination of Case and agreement. Linguistic Inquiry 27,1-68. Bittner, Μ. & K. Hale (1996b). Ergativity: Towards a theory of a heterogeneous class. Linguistic Inquiry 27, 531-604. Blake, B.J. Case. Melbourne: Cambridge University Press. Bobaljik, J. (1992). Nominally absolutive is not absolutely nominative. Proceedings ofWCCFLXI, 44-60. Stanford Linguistics Association.

70

Alana Johns

Bobaljik, J. (1993). On ergativity and ergative unergatives. MIT Working Papers in Linguistics 19, 45-88. Bobaljik, J. (1998). Pseudo-ergativity in Chukotko-Kamchatkan agreement systems. L'ergativite [Recherches linguistiques de Vincennes 27], edited by Lea Nash, 21-44. Bok-Bennema, R. & A. Groos (1984). Ergativiteit. GLOT 7,1-49. Bok-Bennema, R. (1991). Case and agreement in Inuit. Dordrecht: Foris publications. Campana. M. (1992). A movement theory of ergativity. Doctoral dissertation, McGill University. Chomsky, N. (1995). The minimalist program. Cambridge, Mass. MIT Press. Chung, S. (1976). Case marking and grammatical relations in Polynesian. Doctoral dissertation, Harvard University. Chung, S. (1977). On the gradual nature of syntactic change. Mechanisms of syntactic change, edited by Charles N. Li, 3-55. Austin: University of Texas Press. D6chaine, R.-M. & V. Manfredi (1998). SVO ergativity and abstract ergativity. L'ergativite [Recherches linguistiques de Vincennes 27], edited by Lea Nash, 71-94. De Hoop, H. (1992). Case configuration and noun phrase interpretation. Doctoral dissertation, Groningen. Dixon, R.M. W. (1979). Ergativity. Language 55, 59-138. Dixon, R.M. W. (1994). Ergativity. Cambridge University Press. Estival, D. & J. Myhill (1988). Formal and functional aspects of the development from passive to ergative systems. Passive and voice, edited by M. Shibatani, 441—491. Amsterdam: John Benjamins. Falk, Y.N. (1997). Case typology and Case Theory. Manuscript, Hebrew University of Jerusalem. Fortescue, M. (1995). The historical source and typological position of ergativity in Eskimo languages. Etudes I Inuit I Studies 19, 61-75. Guilfoyle, Ε., H. Hung, & L. Travis (1992). SPEC of IP and SPEC of VP: Two subjects in Austronesian Languages. Natural Language and Linguistic Theory 10, 375-424. Hale, K. (1970). The passive and ergative in language change: The Australian case. Pacific studies in honour of Arthur Capell, edited by S.A. Wurm & D.C. Laycock, 757-781. Sydney: A.H. & A.W. Reed. Hale, K. & S.J. Keyser (1986). Some transitivity alternations in English. Lexicon Working Papers 7. Center for Cognitive Science, MIT.

Ergativity

71

Hale, Κ. & S.J. Keyser (1993). On argument structure and the lexical expression of syntactic relations. The view from Building 20, edited by K. Hale & S.J. Keyser, 53-109. Cambridge, Mass.: MIT Press. Halle, M. & A. Marantz (1993). Distributed morphology. The view from Building 20, edited by K. Hale & S.J. Keyser, 111-176. Cambridge, Mass.: MIT Press. Jelinek, E. (1984). Empty categories, Case and nonconfigurationality. Natural Language and Linguistic Theory 2, 39-76. Jelinek, E. (1993). Ergative splits and argument type. Papers on Case and agreement I [MIT Working Papers in Linguistics 18], 15-42. Jensen, J.T. & A. Johns (1989). The morphosyntax of Eskimo causatives. Theoretical perspectives on Native American languages, edited by D. Gerdts & Κ. Michelson, 209-229. Albany, NY: SUNY Press. Johns, A. (1984). On the nature of grammatical relations and Case assignment in Eskimo. Cahiers linguistiques d'Ottawa 12, 49-87. Johns, A. (1987). Transitivity and grammatical relations in Inuktitut. Doctoral dissertation, University of Ottawa. Johns, A. (1992). Deriving ergativity. Linguistic Inquiry 23, 57-87. Johns, A. (1995). On some mood alternations in Labrador Inuttut. Grammatical relations: Theoretical approaches to empirical questions, edited by C. Burgess, K. Dziwirek, & D. Gerdts, 131-151. Stanford: CSLI Publications. Johns, A. & C. Smallwood (1999). On (non-)finiteness in Inuktitut. Toronto Working Papers in Linguistics 17. The Hague: Thesus - Holland Academic Graphics. Johnson, M. (1980). Ergativity in Inuktitut (Eskimo) in Montague Grammar and in Relational Grammar. Bloomington: Indiana University Linguistics Club. Kayne, R. (1993). Toward a modular theory of auxiliary selection. Studia Linguistica 47, 3-31. Kibrik, A.E. (1985). Toward a typology of ergativity. Grammar inside and outside the clause, edited by J. Nichols & A. Woodbury, 268-323. Cambridge University Press. Laka, I. (1993). Unergative that assign ergative. Papers on Case and agreement I [MIT Working Papers in Linguistics 18], 149-172. Levin, B. (1983). On the nature of ergativity. Doctoral dissertation, MIT. Levin, B. (1987). The middle construction and ergativity. Lingua 71,17-31. Levin, J. & D. Massam (1985). Surface ergativity: Case/theta relations reexamined. Proceedings of NELS 15, 286-301. GLSA, University of Massachusetts, Amherst. Longobardi, G. (1994) Reference and proper names. Linguistic Inquiry 25, 609-665.

72

Alana Johns

Mahajan, A. (1989). Agreement and agreement phrases, in MIT Working Papers in Linguistics 10, 217-252. Mahajan, A. (1994). The ergativity parameter: have-be alternation, word order and split ergativity. Proceedings of NELS 24, 317-331. GLSA, University of Massachusetts, Amherst. Manga, L. (1996a). Specificity in Inuktitut and syntactic representation. Etudes /Inuit/Studies 20, 63-85. Manga, L. (1996b). An explanation for ergative versus accusative languages: An examination ofInuktitut. Doctoral dissertation, University of Ottawa. Manning, C. (1994). Understanding ergativity. Doctoral dissertation, Stanford University. Manning, C. (1996). Ergativity: Argument structure and grammatical Relations. Stanford: CSLI Publications. Marantz, A. (1984). On the nature of grammatical relations. Cambridge, Mass.: MIT Press. Marantz, A. (1991). Case and licensing. Paper presented at East Coast Conference on Linguistics 1. Massam, D. (1992). Direct and indirect object: The Niuean Case system. Cahiers de Linguistique de l'UQAM 1, 69-80. Massam, D. (1994). Case without functional projections. Proceedings of the 1994 Annual Conference of the Canadian Linguistic Association, 369-379. Toronto Working Papers in Linguistcs, University of Toronto. Merlan, F. (1985). Split intransitivity: Functional oppositions in intransitive inflections. Grammar inside and outside the clause, edited by J. Nichols & A. Woodbury, 324-362. Cambridge University Press. Murasugi, K. (1992a). Crossing and nested paths: NP movement in accusative and ergative languages. Doctoral dissertation, MIT. Murasugi, K. (1992b). Possessive and transitive clauses in Inuktitut. Proceedings of the 1992 Annual Conference of the Canadian Linguistic Association. Department of Linguistics, University of Toronto, 215-226. Murasugi, K. (1995a). Lexical Case and NP raising. Grammatical relations: Theoretical approaches to empirical questions, edited by C. Burgess, K. Dziwirek, & D. Gerdts, 309-320. Stanford: CSLI Publications. Murasugi, K. (1995b). Review of Case, scope and binding, by Maria Bittner. Journal of Linguistics 31,157-181. Murasugi, K. (1997a). Nested paths in syntactically ergative languages. Theoretical issues at the morphology-syntax interface, edited by A. Mendikoetxea & M. Uribe-Etxebarria, 325-358. [Supplements of the International Journal of Basque Linguistics and Philology.] University of the Basque Country, and Gipuzkoako Foru Aldundia, Donostia. Murasugi, K. (1997b). Relative restrictions on relative Clauses. Proceedings of the NELS 27, 273-286. GLSA, University of Massachusetts, Amherst.

Ergativity

73

Nash, L. (1996). The internal ergative subject hypothesis. Proceedings of NELS 26,195-209. GLSA, University of Massachusetts, Amherst. Nash, L. (1995). Portee argumentale et marquage casuel dans les langues SOVet dans les langues ergatives. Doctoral dissertation, University Paris 8. Novak, E. (1996). Transforming the images. Berlin: Mouton de Gruyter. Payne, J. (1980). The decay of ergativity in Pamir languages. Lingua 51, 147-186. Philips, C. (1993). Conditions on agreement in Yimas. Papers on Case and agreement I [MIT Working Papers in Linguistics 18], 173-213. Philips, C. (1995). Ergative subjects. Grammatical relations: Theoretical approaches to empirical questions, edited by C. Burgess, K. Dziwirek, & D. Gerdts, 341-357. Stanford: CSLI Publications. Plank, F. (1979). Ergativity: Towards a theory of grammatical relations. London: Academic Press. Roberts, T. (1994). Subject and topic in St'dt'imcets (Lillooet Salish). M.A. thesis, University of British Columbia. Sadock, J. (1996). Reflexive reference in West Greenlandic. Contemporary Linguistics, 1. University of Chicago. Silverstein, M. (1976). Hierarchy of features and ergativity. Grammatical categories in Australian languages, edited by R.M. W. Dixon, 112-171. Canberra: Australian Institute of Aboriginal Studies. Ura, H. (1998). Ergativity and Checking Theory. Manuscript, Osaka University. Wharram, D. (1996). In the event of an event: A Minimalist account of 'subjects'. M. A. thesis, Memorial University of Newfoundland. Wharram, D. (1998). Indefinites and the antipassive in Inuit:A, er, morphological theory of scope. Manuscript, University of Connecticut. Woodbury, A. (1977). Greenlandic Eskimo, ergativity, and Relational Grammar. Grammatical Relations [Syntax and Semantics 8], edited by P. Cole & J . M. Sadock, 307-336. New York: Academic Press. Woodbury, A. (1981). Study of the Chevak dialect of Central Yup'ik Eskimo. Doctoral dissertation, University of California, Berkeley. Woodbury, A. (1985). Noun phrase, nominal sentence and clause in Central Alaskan Yupik Eskimo. Grammar inside and outside the clause, edited by J. Nichols & A. Woodbury, 61-88. Cambridge University Press. Woolford, E. (1997). Four-way Case systems: Ergative, nominative, objective and accusative. Natural Language and Linguistic Theory 15, 181-227.

When verb phrases go missing Kyle Johnson

1. Introduction English speakers are equipped with a kind of anaphora that allows for the missing predicate in (2) to be recovered from the bracketed material in (1).

(1)

Holly Golightly won't [eat rutabagas],

(2)

Fred won't, either.

A very similar, perhaps identical, phenomenon is found in (3) and (4). (3)

a. b.

Mag Wildwood will read Fred's [story], Joe Bell will read Holly's.

(4)

a. b.

Jos6 Ybarra-Jaeger asks that [we go to the meeting], Sally Tomato will ask when.

Whereas in (2) a VP has gone missing, in (3b) an NP is absent and in (4b) an IP disappears. These are cases of ellipses, each with names that trace back to the construction-based orientation of yesteryear: "VP Ellipsis," "N Deletion," and "Sluicing." Though they are distinguishable on the basis of the syntactic category that has elided, the pragmatic and syntactic conditions under which they occur are close enough to believe that they answer to the same process. This, at any rate, has become a common practice. In (1M3), then, we have evidence of a process which renders phrases of a certain sort silent and recovers their content. But this process is choosey; it never allows DPs or adverb phrases to partake (in English), and it only lets VPs, NPs (which is what the Ns of Ν Deletion become under the DP Hypothesis) and IPs elide when conditions are just so. (5) illustrates the unelidability of adverb phrases and DPs.

76 (5)

Kyle Johnson

a. Jose worded his letters carefully. b. *Sally also worded his letters. c. *Sally also worded carefully.

And (6) illustrates some of the ways in which VP, NP and IP elision is finicky. (6)

a. *?Even though Doc Golightly urged us to [buy a house], I couldn't figure out how to. b. Til read Madame Spanella's long [story] only if you read Fred's short. c. *Although I can't tell you when Lulumae Barnes left, I can tell you whether.

(And the same could be demonstrated for adjective phrases and PPs, though there are enough uncertainties with these cases that I'd rather dodge them here. See Epstein 1984.) It's not too hard to see this choosiness as stemming from properties of the syntax of the phenomena. That it is sensitive to morphosyntactic category, for example, suggests a syntactic force. And the fact that the anaphora which the ungrammatical ellipses in (6a, b) are trying to express can be rendered with an overt anaphor, as in (7), suggests, more particularly, that conditions on empty categories are involved. (7)

a. b.

Even though Doc Golightly urged us to [buy a house], I couldn't figure out how to do it. I'll read Madame Spanella's long [story] only if you read Fred's short one.

This is the direction explored by Zagona (1988a, Chapter 4, 1988b) and Lobeck (1987a, b, 1995), who formulate a condition on ellipses sites on the basis of these facts. Their proposals have much in common with the Empty Category Principle (ECP), which similarly governs where "traces," may be distributed. We might also see the language variation that ellipsis's finickiness displays as indicative of its syntactic roots. A commonplace conception of this sort of variation traces it back to whatever the relevant syntactic constraint(s) are, either by way of differences in the constraint itself, or through language parochial properties with which the constraint interacts. A successful application of such a program to ellipsis would have to explain why VP Ellipsis is exceedingly difficult to find among the world's languages, whereas Sluicing is very common; and why NP Ellipsis, not as rare as VP Ellipsis, nonetheless shows tremendous cross-linguistic variation. Part of our task is then to understand this syntactic component to the

When verb phrases go missing

77

operation. We look for the principle or principles that determine what can be the targets of ellipsis. It is hoped that these principles will go some way towards deriving the apparent sensitivity to category reviewed above. Perhaps, that is, the principles that determine the distribution of phrases based on their categorial membership dovetail with the principles that govern the distribution of ellipsis sites to derive the apparent dependency on category that ellipsis has. One way to converge on what these principles are is to compare ellipsis with other phenomena which are known to make use of the principles under suspicion. We might hope to link one of the properties of ellipsis to principles which are believed to govern that property in the other phenomena. That ellipsis creates a phonetically null region in a phrase-marker, for example, is shared by the null argument (or PRO Drop) phenomenon so common outside English. It is also shared by movement operations, which create traces in their wake. Null arguments are widely believed to be licensed by virtue of their proximity to heads of certain sorts; and traces are thought to be licensed by their proximity to the moved item (and sometimes also by virtue of their relationship to some nearby head). So one hypothesis is that these (or perhaps related) licensing conditions determine what the range of targets for ellipsis are as well. (8) a. b.

The sites of ellipsis are all and only those which meet the licensing conditions on: null arguments, or traces.

The ECP-like approach in Zagona and Lobeck, mentioned above, might be seen as expressing the hypothesis in (8b). (Though Lobeck 1995 can be seen as pursuing the thesis in (8a).)

2. Matching

the licensing condition on Ellipsis with its content

An issue that connects with the syntactic conditions on ellipsis is what the phonetically null region is made up of. It is natural, for example, to think of the ellipsis site as holding a null proform, if the licensing condition turns out to be like that for null arguments. And it is natural to think of the ellipsis site as holding a trace, in the event that its licensing condition more nearly matches that of traces. Under the first scenario, we might also try to see the anaphoric properties that ellipsis invokes as making use of the same mechanisms that pronouns in general do. That is, we might speculate that the reason the syntactic conditions on ellipsis are the same (if they are) as those on null arguments is because there is a kind of null proform in

78

Kyle Johnson

such cases. This position has as its latest champion, Dan Hardt (see Hardt 1993); but a similar hypothesis can be found in Schachter (1977), Partee & Bach (1984) and Chao (1987). A popular alternative is to see the ellipsis site as being derivationally related to a full syntactic version of the phrase. So, for example, the surface representation of (2) could be seen as related to the fuller Sam won't eat rutabagas either through either a deletion process (that removes the VP by way of its anaphoric connection to a previously occurring VP) or a reconstruction process (that forms from the surface representation an LF into which the understood VP is copied). Sag (1980, originally 1976) advocates the deletion approach, and Williams (1977a) sponsors the reconstruction method. Let's not worry about judging between these two, but consider instead how these derivational approaches differ from the proform idea. Of course, on the derivational view of VP Ellipsis it is not obvious why either of the licensing conditions in (8) should be true. One possibility, perhaps, would be to reinterpret traces as silent "copies" of the term that has moved. This is the kind of connection that Ross envisioned, with his typology of copying and chopping rules, and meshes well too with Chomsky's recent construal in these terms of the reconstruction effects that movement evokes. In any case, let's postpone too this issue. 2.1. For the derivational

view

The difference between the derivational and proform approaches to VP ellipsis engaged much of the early literature on anaphora and ellipsis. (Hankamer & Sag 1976 have an enlightening review.) Grinder & Postal (1971), for example, argue that VP ellipsis is a form of "Identity of Sense" anaphora, a dependency that obtains when the anaphor recovers the semantic content of its antecedent, rather than its antecedent's referent. Thus an antecedent and an Identity of Sense anaphor do not independently express ways of referring to the same entity, but instead constitute expressions with the same denotation. One of their interesting arguments comesfromwhat they dub the "Missing Antecedent" phenomenon, illustrated by (9). (9)

My uncle doesn't have a spouse but your aunt does and he is lying on thefloor.(Grinder & Postal 1971, p. 278 (17a))

In (9), the VP ellipsis in the but clause introduces the indefinite antecedent to he (= a spouse). That is, the ellipsis site in (9) recycles the semantic material of its antecedent and thereby introduces a referent that the antecedent didn't. Identity of Sense anaphora in general, and ellipsis in particular, has a straightforward expression under the derivational approach.

When verb phrases go missing

79

Suppose in such cases that there is a level of representation where an Identity of Sense anaphor (= ellipsis site) is made up of a syntactic representation, and it's this representation that is matched against the antecedent. Thus, the Identity of Sense anaphor recycles the linguistic content of its antecedent — in the case of (9) it's constituted of have a spouse, the very words that make up its antecedent—thereby re'invoking its denotation. The proform approach, on the other hand, will have to overcome the fact that the Missing Antecedent phenomenon is not present in the transparent pronominal anaphora of (10), as Bresnan (1971) observes. (10)

*My uncle didn't buy anything for Christmas, but my aunt did it for him, and it was bright red. (compare: My uncle didn't buy anything for Christmas, but my aunt did, and it was bright red.) (Bresnan 1971, p. 591, (9))

(Though note, as Bresnan does, that it is possible to find an antecedent for it in (10) through simple deduction. This, perhaps, explains the improvement (10) enjoys with thought; see Postal 1972 where this effect is elevated to a challenge to Bresnan's interpretation of the (9)/(10) contrast.) Hence, the Missing Antecedent phenomenon makes the proform approach look doubtful, but fits well the derivational interpretation of ellipsis. There is a slightly different case which, like the Missing Antecedent phenomenon, suggests that an ellipsis site is made up of linguistic material recovered in the antecedent. This case is found in contexts of extraction, where the ellipsis contains a variable bound to the extracted item. (11), drawn from Fiengo & May (1994), which contains a thorough and valuable discussion of VP ellipsis, illustrates. (11)

a. b.

I know which book Max read, and which book Oscar didn't. This is the book of which Bill approves, and this is the one of which he doesn't. (Fiengo & May 1994, p. 229, (99a, c))

In these cases too the ellipsis site seems to have internal parts: in (11a), a variable bound by which book, and in (lib), a variable bound to which. And, as with the Missing Antecedent phenomenon, obvious pronouns do not seem to tolerate the same kind of internal parts; compare (11) with (12). (12)

a. *I know which book Jose didn't read for class, and which book Lulumae did it for him. (compare: I know that Jose didn't read this book for class, but that Lulumae did it for him.)

80

Kyle Johnson b. *This is the book which O.J. Berman reviewed, and this is the one which Fred won't do it. (compare: O.J. Berman reviewed this book but Fred won't do it.)

Just as in (9), then, the ellipses in (11) recover their syntactic form (=[yp read t] and [VP reviewed £]) from their antecedents. By contrast, pronouns have no syntactic form beyond the lexical item they constitute, and (12) therefore results in a violation of the ban against vacuous quantification. Thus, as in the Missing Antecedent cases, we are led to the conclusion that an ellipsis site should not be seen as consisting of a hidden pronoun. (This argument for the derivational approach can be found in Tancredi 1992, and goes back at least to Chao 1987.) That the elided material in cases such as (11) does contain syntactic material, and more particularly a trace bound to the Α-moved item, is strengthened by Haik's (1987) discovery that various island effects hold into the ellipsis site. (13)

a.

I know which book Mag read, and which book Mr. Yunioshi said that you hadn't. b. *?I know which book Mag read, and which book Mr. Yunioshi asked why you hadn't. ? c. * I know which book Mag read, and which book Mr. Yunioshi read my report that you hadn't. d. *?I know which book Mag read, and which book Mr. Yunioshi was discussed after I had.

The difference between (13a) and the others is the familiar bounding constraints, whatever they may be, which govern how far a wh-phrase may move: descriptively speaking, (13b) is an instance of Chomsky's wh Island Constraint, (13c) exemplifies Ross's Complex NP Constraint, and (13d) is a result of the Adjunct Condition. In so far as these conditions are ones that hold of an A-moved term and its trace, we have reason to believe that there is a trace in the ellipses of these examples. Cases of Antecedent Contained Deletion, introduced by Bouton (1970), probably illustrate a similar point. In such cases, the antecedent VP appears to contain the ellipsis, as in (14). (14)

Dulles [suspected everyone who Angleton did [e]].

(Spies figure largely in the examples of Antecedent Contained Deletion because of the spook-ridden and influential May (1985) and Fiengo & May (1994).) This is a situation that VP Ellipsis permits, but that overt forms of VP anaphora do not.

81

When verb phrases go missing (15)

a. *Dulles suspected everyone who Angleton did it. b. *Dulles suspected everyone who Angleton did so.

Of course, this suggests again that VP Ellipsis should not be seen as the silent version of a proform. The ungrammaticality of the proform anaphora in (15) is no doubt related to the similar difficulties in (16). (16)

a. *Dulles bought [a portrait of i t ^ . b. *Dulles praised [the picture of a portrait of itjj.

There is something that prevents the pronouns in (16)frombeing referentially dependent on the argument containing them; and presumably this same force is at play in (15). It is not immediately clear, however, how (14) can arise even if it is not an instance of proform anaphora. On the derivational approach to VP Ellipsis, for instance, there is no antecedent with the proper form for the ellipsis in (14), as can be seen in (17). (17)

Dulles [suspected everyone who Angleton did [auopoctcd every one who Angleton did ...]].

Even if we could figure out how to fill in "...," the struck-out VP in (17) is not the one that is understood to be elided in (14). A popular approach to this problem on derivational theories of VP Ellipsis is to view Antecedent Contained Deletions as being licensed by movement of the argument which contains the ellipsis. May (1985), for instance, follows Sag's (1976) suggestion that the argument containing the elision scopes out of the antecedent VP. He argues that there is a level of syntactic representation projected from the surface in which arguments occupy the position at which their scopes are computed. In this level of Logical Form (LF), the object in (14) can be scoped out of the VP which is to serve as antecedent, yielding (18). (18)

[everyone who Angleton did [e]]j [Dulles [suspected t^].

And in (18), the VP is now of nearly the right form to serve as antecedent, as (19) illustrates. (19)

[everyone who Angleton did [ouopcctcd

[Dulles [suspected

The only difficulty with (19) is that the form of the verb is not appropriate for the ellipsis site; let's set this aside for the moment. Hornstein (1994, 1995) offers a variant of this account which credits movement of the object out of the antecedent VP not to QR, but to "Object Shift," the name given to a kind of short Scrambling — an instance of

82

Kyle Johnson

Argument Movement — found in the Scandinavian languages. (See Holmberg 1986, Deprez 1989 and Vikner 1995.) On Hornstein's interpretation, Object Shift brings direct and indirect objects into the Specifiers of functional projections (which license them, perhaps through Case marking) at LF. These functional projections are thought to lie between the surface position of the subject and the VP. Hence, the LF Hornstein would give to (14) might be as in (20). (20)

[Dulles [ F P [everyone who Angleton did [auapcctcd [F° [ V P suspected *,]]]]. (understand "F°" to be the functional head that licenses objects)

How these two variants should be compared will depend on whether "Object Shift" is identified with QR. An interesting consequence of these approaches is that they manufacture in the elided VP a trace, which in turn forces the environment in which the elision arises to have a binder for that trace. In (14), the relative pronoun binds the trace. In fact, a general feature of Antecedent Contained Deletions is that they involve relative clauses. An Antecedent Contained Deletion cannot survive in other kinds of complex NPs: (21)

*I [told a rumor that Mag did [told α rumor that Mag told a rumor that...]].

The that-c\ause in (21) cannot be a complement to rumor. That is, the content of the rumor I told cannot be that Mag told a rumor, instead, the ί/ιαί-clause says that Mag and I told the same rumor. So, this fact about Antecedent Contained Deletion, like that in (11), suggests that an elided VP can have enough of the form of a VP to hold a trace. Indeed, it is in the context of Antecedent Contained Deletion that Ha'ik demonstrated island effects: (22)

*John met everyone that Peter wondered when he could [e]. (Ha'ik 1987, p. 511, (18))

These facts suggest, therefore, not only that the VP elided in Antecedent Contained Deletion contexts has a trace in it, but that it must have that trace. Like the examples in (11), this supports the derivational view of ellipsis. An unfortunate feature of this last argument for the derivational approach to VP Ellipsis is that it rests on locating a phonetically empty term — the trace — within an ellipsis. But as we have all no doubt had occasion to discover, finding the location of invisible things is not trivial. Could we have mislocated it? Might the trace in (19) and (11) in fact be outside the ellipsis, as indicated in (23).

When verb phrases go missing (23)

a. b.

83

I know which book Max read, and which bookj Oscar didn't [e]i j. Dulles [suspected everyone who! Angleton did [e] i ; ].

This would be possible if VP Ellipsis, or some other ellipsis phenomenon, could elide portions of VPs, leaving remnants in the positions that the traces in (23) occupy. Interestingly, that does seem possible: (24)

a. b.

While O.J. Berman read Fred, he didn't [e] Dickens, Sally suspected Joe, but he didn't [e] Holly.

Levin (1986) dubs these "Pseudogaps," and argues that they involve a process different from VP Ellipsis. Her reason for distinguishing the two turns on the observation that there are certain environments where VP Ellipsis is permitted, but Pseudogapping is not. In fronted adverbials, for instance, VP Ellipsis but not Pseudogapping is licensed (an observation that goes back to Sag's dissertation). (25)

a. Although Holly doesn't [e], Doc eats rutabagas. b. * Although Mag doesn't [e] eggplants, Sally eats rutabagas.

It is possible, then, that we have mistaken VP Ellipsis for Pseudogapping in (11) and (14). This is what Jacobson (1992) and Hardt (1993, Section 2.5), who credits Lappin & McCord (1990), propose. Hardt leans on the fact that cases like (11) and (14) are prohibited in fronted adverbials, and so appear more like Pseudogapping than VP Ellipsis. (26)

a. *?Although I don't know which book Sam did [e], I do know which book Sally read, b. *?Near everyone Angleton did [e], Dulles stood. (compare: Dulles stood near everyone Angleton did.)

If successful, this reanalysis would remove (11) and (14) as support for the derivational view. However, while it may well be possible to give some of these cases a Pseudogapping source, Hai'k (1987), Kennedy (1997) and Tomioka (1997) give compelling reasons for not reducing all such cases to Pseudogapping. Pseudogapping is subject to restrictions on the remnants which are not always reflected in the appropriate way in VP Ellipsis. Pseudogapping, for instance, cannot elide part of a prepositional phrase, as Ha'ik (1987) and Kennedy (1997) observe, nor may it remove part of a noun phrase, as Tomioka (1997) points out. Illustrative examples are in (27). (27)

a. *Sally will stand near Mag, but he won't [e] Holly. b. *While Holly didn't discuss a report about every boy, she did [e] eveiy girl.

But elided VPs may (appear to) hold traces in these positions; (28) illustrates.

84 (28)

Kyle Johnson a. b. c.

?

I know which woman FRED will stand near, but I don't know which woman YOU will [e]. Sally will stand near eveiy woman that you will [e], I know which woman HOLLY will discuss a report about, but I don't know which woman YOU will [e]. Holly discussed a report about every boy that Berman had [e].

(The relative badness of (28a) reflects the difficulty in moving u;A-phrases out of adjuncts, and contrasts with (27a).) These examples at least, then, must not be Pseudogaps; and it is reasonable to conclude, therefore, that VP ellipsis can host a trace. ((28b) and (28d) also speak against Hornstein's account of Antecedent Contained Deletion which, recall, assimilates it to Object Shift. Object Shift is unable to strand a preposition, and this is what would be required of it in (28b); nor is it able to move noun phrases out of other noun phrases, as is necessary in (28d). (Examples illustrating the facts I report here about Object Shift can be found in Holmberg 1986.) This is Kennedy's 1997 point — that Object Shift is not the source of Antecedent Contained Deletion — but his arguments cut against a Pseudogapping source for them as well.) Moreover, despite the mismatches in licensing environments that Levin catalogues, there is, I think, considerable reason to believe that Pseudogapping is a special instance of VP Ellipsis. The reason is that the terms which can serve as remnants in Pseudogaps are limited in a way that suggests they have moved from the elided VP. That Pseudogapping is VP Ellipsis from which a remnant has moved is a thesis that Jayaseelan (1990) and Lasnik (1995a) have advanced in recent years. Jayaseelan argues that Pseudogapping is the product of eliding a VP after Heavy NP Shift has occurred, which would explain, for instance, why Pseudogapping cannot strand a preposition, since Heavy NP Shift cannot do so either. (29)

*Sam stood near yesterday every one of the women we'd been discussing.

Lasnik, by contrast, argues that Object Shift is responsible for bringing the remnant out of the elided VP. Object Shift, at least as it is found in the Scandinavian languages, shares with Heavy NP Shift an inability to strand prepositions, as noted above. But Object Shift differs from Heavy NP Shift in being able to move pronouns and the first object of a double object construction, things Heavy NP Shift cannot do, as (30) indicates. (30)

a. b.

*Truman visited yesterday you. *Truman told the story Rusty.

When verb phrases go missing

85

Because pronouns and the first object of a double object construction can be the remnants in Pseudogapping (as in (31)), Lasnik concludes that Object Shift makes a better candidate than does Heavy NP Shift for the mechanism which evacuates remnants from a Pseudogapped VP. (31)

a. b.

While Truman didn't visit me, he did [e] you. While Truman didn't tell me a story, he did [e] Rusty.

Actually, however, even Object Shift does not have quite the right properties to be the source of the constraints on Pseudogapping remnants. Object Shift is unable to move prepositional phrases, for instance; whereas prepositional phrases, though somewhat strained, may be the remnants of Pseudogapping. (32)

a. b.

While Jose won't talk about Mag, he might [e] about Holly, Although Doc might tell it to you, he won't [e] to me.

Moreover, Levin notes that Pseudogapping seems able to strand remnants which are buried quite deeply in the elided VP; some of her examples are in (33). (33)

a.

Then the police started to pick up soloists — like they did [e] you. I'm sure I would like him to eat fruit more than I would [e] cookies. I think you need to show yourself more than you do [e] anyone else. (Levin 1986, pp. 15-16, (4), (7) & (13))

b. c.

While Levin suggests that cases like these are possible only in comparatives, raising the spectre that they may result from whatever it is that creates (the otherwise peculiar) elisions in these constructions, I think they are not impossible in other contexts. Consider, for example, (34). (34)

a. b. c.

?

While I wouldn't like him to eat cookies, I would [e] fruit. While I think you need to examine yourself, you don't [e] anyone else. While Truman doesn't want to visit every city, he does [e] Barcelona.

?

In (33) and (34), the remnants are the objects of a clause embedded within the elided VP. If Object Shift is the process that brings remnants out of the elision, it would have had to move them clear out of the clause they start in. But Object Shift, at least as it is represented in the Scandinavian languages, cannot move an object that far. For these reasons, then, Object Shift is

86

Kyle Johnson

not quite therightmechanism to blame for bringing remnants out of a Pseudogapped VP. Still, the argument that Jayaseelan and Lasnik offer on behalf of reducing Pseudogapping to VP Ellipsis is, I believe, sound. It requires, however, an analogy to the kind of Scrambling that Dutch hosts in its middle field, rather than to Heavy NP or Object Shift. In Dutch, it is possible to scramble objects, whether they be pronouns or larger, and prepositional phrase complements leftwards past adverbs and the like. This Scrambling, like Heavy NP Shift and Object Shift, is also unable to strand prepositions (The exception to this is the Scrambling of "R-pronouns," discussed in van Riemsdijk 1982; but modern English doesn't have this phenomenon (except in Sluicing contexts).) Thus, Scrambling has the constraints on Pseudogapped remnants that Jayaseelan and Lasnik singled out as indicative of movement. Further, Scrambling is able to span longdistances, bringing objects out of an embedded clause and beyond the VP which that clause is embedded in, as in (35). (35)

...dat Jan Marie heeft geprobeerd [£ te küssen]. ...that Jan Mary has tried to kiss '...that John has tried to kiss Mary.'

It therefore has the power to generate the large Pseudogaps in (33) and (34); simply let VP Ellipsis be fed by long-distance Scrambling, as shown in (36). (36)

...FP

Barcelona

F

VP

want

elide

to visit t^

Scrambling, then, shows the features we have encountered so far in Pseudogapping: it can relocate PPs and DPs, though not if it entails stranding a preposition, and it can do so over long distances. Moreover, long-distance Scrambling in Dutch and the span that large Pseudogaps may have are subject to hauntingly similar constraints. Long-distance Scrambling is restricted to certain kinds of non-finite complement clauses; it is blocked from adjunct clauses, as in (37a), and from finite complement clauses, as in (37b).

When verb phrases go missing (37)

87

a. *...dat Jan het boekzijn vader gelezen heeft .. .that John the book his father read has [om t te plezieren]. C° to please '...that John has read the book to please his father.' b. *...dat Jan de Krant beweert [dat Sam t leest], .. .that John the paper claimed that Sam read '...that John claimed that Sam read the paper.'

And similarly the remnants of a Pseudogap cannot be embedded within an adjunct clause, as in (38a), nor may they be found within a finite complement clause, as in (38b). (38)

a. *While Rusty might leave in order to please Mag, he won't [e] his father. b. *While Doc might claim that O.J. Berman had read his book, he wouldn't [e] the paper.

And finally, the class of terms that Scrambling may affect probably matches the range of terms that may remain after Pseudogapping. For example, verbal particles typically make bad remnants for Pseudogaps, as in (39); and they also resist Scrambling, as in (40). (39) (40)

a. *While Perry might switch the TV OFF, he won't [e] ON. b. *I'll turn the radio DOWN, but I won't [e] UP. *...dat Jan de TV uit steeds zet. .. .that Jan the TV out all the time puts '...that Jan turns the TV off all the time.' ...dat Jan de TV steeds uit zet. ...that Jan the TV all the time out puts '...that Jan turns the TV off all the time.' (Zwart 1993, 321)

There are other relevant cases to look at, but in outline it looks like the pattern of remnants left by Pseudogaps matches those that are able to Scramble out of VPs in Dutch. If Pseudogapping is VP Ellipsis, this finds an explanation: remnants should be just those phrases able to move out of the elided VP. Of course, this leaves the large problem of understanding how English avails itself of Scrambling in these contexts when it is otherwise unable to. (Lasnik 1995a offers a suggestion on this point, but not one that easily accommodates to the long-distance cases in (33) and (34).) So we have two reasons for doubting that Pseudogapping can rescue us from the conclusion that examples like (11) and cases of Antecedent Contained Deletion reveal traces in elided VPs. First, there are some

88

Kyle Johnson

examples of these kinds for which a Pseudogapping source is very dubious (namely: (28)). And second, it is not clear that Pseudogaps are anything more than elided VPs with traces in them to begin with; the evidence reviewed above raises the suspicion that the remnants in Pseudogaps have Scrambled from an elided VP. So far as I know, Pseudogapping is the only way of analyzing cases like (11) and Antecedent Contained Deletions which avoids the conclusion that elided VPs can contain a trace. To the extent that it fails, then, confidence in the derivational account of VP Ellipsis strengthens. There is other evidence on behalf of the derivational approach to VP Ellipsis, beyond that provided by the Missing Antecedent phenomenon and the evidence, just reviewed, that elided VPs can contain a trace. Fiengo & May, for example, discuss a number of ways in which anaphoric connections are expressed in elided VPs which suggests that an elided VP has the syntactic structure of an overt VP. 3. Against the derivational

view

There are also facts, however, which have been taken to weigh against a derivational view of VP Ellipsis. Interestingly, though, the derivational approach must be wedded to an additional assumption for these facts to be seen as problematic. The additional assumption is that the form of the elided VP must match, more or less perfectly, the form of its antecedent VP. This is because these problematic facts all have the following characteristic: they are cases where the antecedent VP could not be copied without change into the ellipsis site. These will then be problems for any theory that holds both (41i) and (41ii). (41)

i. ii.

An ellipsis site consists of a silent version of the phrase understood to be there, and The silent phrase in an ellipsis site is syntactically identical to some overt phrase.

This is not so onerous a wedding, as most derivational approaches do embrace some sort of syntactic identity condition on antecedent and elided VP. But, as might be expected, a rabid derivationalist will point to (41ii) as the source of trouble before pointing to (41i). Let's look at some of these cases, and examine how far into (41) they cut. Some of them involve the presence of variables within an ellipsis site, just the sort of case which was taken in the previous section to support the derivational view. Hardt (1993, Section 2.4), for instance, points out that

When verb phrases go missing

89

cases such as (42) should be expected to be ungrammatical under a derivational view. (42)

a. b. c.

China is a country ίΛαί Joe wants to visit t, and he will too, if he gets enough money, (from Webber 1978) This is just the kind of thing that Harris could have suggested t. And in fact, he did. Harry is someone they would like to send t to the Olympics. And they will [e] too, if they can finance it. (Hardt 1993, pp. 15-16, (21H22))

If the antecedent VPs recycle their trace into the ellipsis site, the results should be on a par with (43). (43)

a. b. c.

.. .he will [visit t\ too, if he gets the money. And in fact, he (did) [suggested t\. And they will [send t to the Olympics].

The examples in (43) are ungrammatical, of course, at least in part because they carry an unbound trace. (The form of the main verb in (43b) is also a source of ungrammatically; we return to this.) But the examples in (42) are just fine. So if VP ellipsis is simply a way of disguising an otherwise normal VP, and antecedent VPs show us what elided VPs look like, why is there a contrast between (42) and (43)? A variety of answers appear possible. One, of course, would be to abandon the derivational approach altogether and adopt the analogy to pronominal anaphora which Hardt sponsors. But then the differences between pronominal anaphora and ellipsis reviewed above will have to be explained. This might be done, perhaps, by imagining that we simply chose the wrong pronouns to compare ellipsis to in the discussion above. Perhaps, for example, we should have analogized to the VP anaphor do so, which not only appears to be licensed in contexts like (42) — witness (44a) — but also invokes the Missing Antecedent effect, as can be seen from (44b). (44)

a. b.

China is a country that Joe wants to visit, and he will do so too, if he gets enough money, Jerry wouldn't read a book by Babel, but Meryl has done so and it was pretty good.

Unfortunately, however, do so anaphora does not seem able to rescue violations of the No Vacuous Quantification law in the same way that VP Ellipsis can; (45) contrasts with (11).

90 (45)

Kyle Johnson a. b.

*I know which book Max read, and which book Oscar hasn't done so. *This is the book of which Bill approves, and this is the one of which he can't do so.

So however it is that do so has the abilities that it does, it still fails to have the ones needed to subsume VP Ellipsis. Unless some other proform can be found whose properties match those of ellipsis, this answer to the problem does not seem promising. Another possibility would be to maintain that VP Ellipsis hides a normal VP, and imagine that in (42) the moved phrase in the antecedent clause binds, somehow, both the variable in the antecedent VP and in the elided VP. That is, we might try to see in (42) the syntax of an across-the-board movement, maybe along the lines sketched in (46). (46)

a. b. c.

...a country [that [[Joe wants to visit t ], and [he will vioit t too]], if he gets enough money. ...the kind of thing [that [[Harris could have suggested t]. And [in fact, he did ouggcat [cejoe] Substitution processes Stopping: the change of fricatives and affricates into stops vinegar [bidu] Prevocalic voicing: the voicing of obstruents before vowels pocket [bat] Final devoicing: the devoicing of final voiced obstruents, knob - » [nap] Fronting: the production more towards the front of mouth duck —> [dat] Gliding: the changing of a liquid into a glide rock [wat]

Smith (1973) formulates these rules — which he called "realisation" rules — in an SPE framework, and assumes that they are simplified and ultimately unlearned in the course of development; Stampe (1973) calls them "natural" rules, which have to be suppressed in the course of acquisition. Spencer (1986) reanalyses Smith's data in a non-linear phonological framework. Iverson & Wheeler (1987) analyse many of the assimilation processes using non-linear phonological tools. A non-linear framework allows us to formulate the rules much more elegantly. However, even in a

228

Paula Fikkert

non-linear phonological framework, where representations are enriched and the number of rules severely limited — only spreading (assimilation) and delinking (deletion) rules are allowed — the problem mentioned above remains. Although, for instance, stopping can now be elegantly described as the delinking of the feature [continuant], and consonant harmony as the spreading of one or more features from one consonant to another (as we will seen in 3.2.1), in the formulation of the rule reference still has to be made to an underlying representation that resembles the adult target form. These works have been criticised because the rules do not seem "psychologically real": it is hard to believe that a child, having an underlying representation which resembles the adult form — based on the fact that the child's perception is far more advanced than his or her production — subsequently changes it to create a new impoverished form. Nevertheless, this is often implicitly assumed. If the input form is the underlying form and resembles the adult target form we have to conclude that the rules are performance rules and do not reflect competence. Another problem with formulating rules to express the relationship between adult and child forms is that rules can only operate on input or adult forms, while many phenomena seem to be better accounted for by assuming constraints on the output, the child's forms. For example, if in a particular position not only fricatives are changed into stops, but also other types of consonants, such as liquids and nasals, we could still try to formulate a rule, but this will result in a collection of ad hoc statements (cf. Menn 1978). By constraining possible output forms the relationship between adult and child forms can be expressed more accurately. This idea has found support in recent literature (cf. Macken 1992; Levelt 1994; Fikkert 1994a, b; Demuth 1995a, b; Demuth & Fee 1995); it is now often assumed that children have certain canonical forms or templates onto which the adult forms are mapped. Since these canonical forms or templates are constrained in certain ways, the child's production form often differs from the adult target form. Development means getting rid of constraints and/or elaborating templates so that the child forms resembles the adult target more and more. How this may proceed will be shown in 3.3 for segmental processes and in 4 for suprasegmental processes. To summarise, all approaches assume an input form that is more or less identical to the adult target form, and an output form — the child's production. They differ, however, in the way they formulate the relationship between input and output forms. In recent work attention has been shifted to explaining this relationship on the basis of a child's developing phonological system, rather than merely describing it by formulating a rule or process. One segmental "rule" that has been topic of much debate lately is consonant harmony (cf. Levelt 1994).

229

Acquisition of phonology 3.2.1. "Consonant

harmony"

Consonant harmony (CH) is the process by which consonants in the word become more similar. This usually only affects primary place of articulation features. The process is relatively often attested in child language, but is hardly found in adult languages, where it always involves secondary place of articulation features, never primary. CH is usually defined as an "assimilation-at-a-distance" process (Vihman 1978). Features from one consonant spread to a non-adjacent consonant. A well-known example is presented in Menn (1978): [gAk] for duck. In non-linear phonology CH is accounted for by spreading the features of one consonant to a consonant not specified for place of articulation (Stemberger & Stoel-Gammon 1991). Coronals are usually assumed to be underspecified for place and are therefore prone to adopt features spreading from other consonants. This feature-filling process can be represented as in (3a). A problem arises, however, when the vowel is also specified for place, since now the spreading results in crossing association lines, as shown in (3b). Of course, this problem does not arise if we assume that consonants and vowels have different sets of place features (e.g. Stemberger & StoelGammon 1991). However, evidence from consonant-vowel interactions points towards a shared set of features for consonant and vowels (cf. Lahiri & Evers 1991). McDonough & Myers (1991) provide a different solution to the problem in (3b) by assuming that vowels and consonants are on different planes (planar segregation), a view shared, for instance, by Macken (1992,1995). In this view, the two consonants are adjacent and there is no intervening vowel that causes association lines to cross. This account is schematised in (3c): (3)

a.

C H AS FEATURE SPREADING FROM A SPECIFIED TO AN UNSPECIFIED SEGMENT I

d

Λ

k

dorsal

•

g

A

k

dorsal

Paula Fikkert

230

b.

C H AS FEATURE SPREADING FROM A SPECIFIED TO AN UNSPECIFIED SEGMENT I I

d

0

k

Λ

I

Place

*g

I

Place

Place

dorsal dorsal C.

k

Λ

I

I

Place

I

Place

dorsal dorsal

C H AS FEATURE SPREADING ASSUMING PLANAR SEGREGATION

dorsal

dorsal

Place

Place k _

d

0

->

I

Place

g

k

Place

Place

I

I

dorsal

dorsal

Although this seems an elegant account of the process, examination of the full vocabulary of a child reveals certain problems, as argued by Levelt (1994). First, planar segregation presupposes that the order of consonants and vowels is entirely predictable. As long as the child only has CV syllables, this is the case. When the child has VC, CV and CVC words, this statement is no longer valid. At this point the order of consonants and vowels in a word has to be learned and planar segregation can no longer be assumed. Second, if spreading is feature-filling, that is, if spreading is always from a specified (i.e. labial or dorsal) to an underspecified (i.e. coronal) consonant, the forms in (4a) are expected, but not those in (4b): APPARENT CASES OF C H

(4) a.

b.

(from Levelt

brood

/bRo:t/

Thread'

poes

/pus/

'cat'

-»

bed vis

/bet/ /vis/

'bed' 'fish'

- » [det] —> [dis]

1994)

[bo:p] [puf]

Further evidence against the account presented by McDonough & Myers (1991) comes from other apparent cases of consonant harmony. As Levelt points out, in Dutch words like /s\un/ are often produced as [pum], which appears to involve [labial] spreading. However, the only labial element in the target word is the vowel. These cases can only be accounted for by assuming that the vowel spreads its place features to the consonants. Levelt therefore investigated all cases of consonant harmony in the

231

Acquisition of phonology

Fikkert/Levelt database (12 children were recorded at two-week intervals for one year) and discovered that most of them could be reanalysed as consonant-vowel interactions. The forms in (4a) have a labial vowel and labial consonants, the forms in (4b) have a coronal vowel and coronal consonants. In other words, the whole word seems to have one place specification. Menn (1978) and Iverson & Wheeler (1987) also propose accounts in which features are specified for whole words, but they implicitly assume either planar segregation or different features for vowels and consonants. CV-interaction does not explain all consonant harmony cases. Words like zeep 'soap' /ze:p/, produced as [pe:p], are not accounted for. Although, taken in isolation, these words may be odd, they can be readily understood by taking into consideration not only whole words, but also whole vocabularies at certain points in time, as we will see now. 3.3. Considering

the whole

lexicon

Waterson (1971) observed that all early production forms of her son fitted into one of five basic word structures, also called "prosodies" or "canonical forms". Furthermore, she noted that these early production forms often did not have a straightforward relationship with the adult forms: the relationship could not be expressed by any of the rules or processes described in 3.2. Nevertheless, on closer inspection, adult and child forms had certain features in common, although the distribution of these features in the word might be completely different. She accounted for those phenomena by assuming that what is perceived best is produced earliest, and that the schemata of these early production forms or prosodies facilitate both the production of other forms and the acquisition of new forms, through pattern recognition. Development takes place when the child perceives more phonetic detail, which differentiates new prosodies, until the final state is reached in which each word has its own prosody. Although Waterson's analysis may account for the initial stages, it has been convincingly shown that incomplete perception at best accounts for a small subset of the production data and that in most cases the child can perceive differences that he or she cannot produce (Smith 1973; Macken 1980; Dinnsen & Barlow 1998). Recently, the focus of explanation has shifted towards output constraints. Macken (1992) noticed that many words are built according to the same recipe: labial consonant — vowel — coronal consonant — vowel, so that a Spanish word like sopa 'soup' is produced as [pota]. Levelt (1994) makes the same observation for Dutch at a particular stage in the development. Usually, this stage is preceded by one in which children only have words

232

Paula Fikkert

that are either completely labial or completely coronal (as shown in (4)), that is, one place specification per word. Gradually, more differentiations are made. In the first "mixed" forms, labials are always attached or aligned to the left edge of the word, explaining why zeep can become [pe:p] and sopa [pota]. Similarly, when dorsals are produced by the child, they are first obligatorily attached to the right edge of words, explaining why a Dutch word like kip 'chicken' /kip/ is produced as [tEk] or [pEk], Alignment constraints are also proposed by Velleman (1996). Thus, as the child's phonological system develops, features are aligned to word edges, rather than to the whole word. Later, these alignment constraints are gradually relaxed, so that features can be attached to any segment in the word. As a result the child is able to expand the set of word forms, until each word has its own form. Work like this shows that it is not sufficient to look at features or segments in isolation, but that one needs to take whole words into account. Furthermore, it is also important to consider a child's whole vocabulary at certain stages to gain a deeper understanding of (a) how segment inventories and vocabularies develop and (b) why processes such as those mentioned in 3.2 take place. This shows once more the importance of longitudinal databases. Work from a holistic point of view has only just begun, and much more research is needed.

4. Acquisition of suprasegmental

phonology

Although research on the acquisition of suprasegmental phonology is not abundant, its development has been similar to research on the acquisition of segmental phonology. In the seventies, a major goal was the identification of the main differences between adult forms and child forms, by formulating a set of rules or processes such as those given in (5): (5)

(Ingram 1976) cat -» [koe] blanket ['bake] banana ['ncenoe]

LIST OF SYLLABLE STRUCTURE PROCESSES

1. 2. 3.

Final consonant deletion Cluster reduction Unstressed syllable deletion

Again, these processes or rules are at best a description of the relationship between adult target forms and children's production forms, and provide no insight into why children's forms differ from adult forms. With the emergence of non-linear phonology these rules were subsequently reanalysed in a non-linear framework. The relationship between input (adult) and output (child) forms was often described as the result of mapping the adult target onto the child's template (cf. Iverson & Wheeler 1987; Fee 1995; Fikkert

233

Acquisition of phonology

1994). If the child's template cannot contain the whole segmental string of the adult target, this results in simplifications, as illustrated in (6): (6)

M A P P I N G OF ADULT TARGET ONTO CHILD'S WORD TEMPLATE

Ft = Wd

a. b. c.

bΘ

C [k] [b] 1 [n]

V [aa] t [a] η [ae]

C

V

[k] [n]

[Θ] [ae]

(6a) and (6b) depict final consonant deletion; (6b) shows in addition cluster simplification; and (6c) illustrates unstressed syllable deletion. The representation in (6) provides a graphic description of the processes, but still leaves many questions unanswered. For example, what determines the shape of the child's template and why is the mapping the way it is. Why is the [bl] cluster reduced to [b]? Why is the initial unstressed syllable in (6c) deleted and not the unstressed final syllable? Moreover, (6) does not tell us anything about how the child forms develop towards the adult target forms. Insight into these questions can be gained by carefully examining longitudinal acquisition data within a formal linguistic theory, together with a theory of acquisition. If there is an innate Universal Grammar (UG) which contains universal principles and parameters, with default values for each parameter, than UG predicts the initial stage in acquisition: all parameters have the default value. The language learner has to look for evidence in the input data (the language of the environment) to change a parameter from the unmarked default value to the marked value. If such evidence is encountered, the parameter is set to the marked value; if not, it remains in the default value. The acquisition process continues until all parameters have the setting required for the language that the child is learning. Formal linguistic theory tells us something about the initial state (all parameters have the default value) and the final state of acquisition (all parameters are fixed as required for the target language), but does not make specific predictions about the intermediate stages, although it drastically reduces the number of possible grammars a child can come up with. Insight into the acquisition process and the intermediate stages can be gained from a careful study of longitudinal acquisition data.

234

Paula

4.1. Syllable

Fikkert

structure

The acquisition of syllable structure has hardly been studied. Although the statements that children (a) start with CV syllables, (b) reduce consonant clusters, and (c) often delete final consonants are commonplace in the literature, claims on further development are hard to find. With respect to onsets the following development has been found for Dutch children (Fikkert 1994a): after a stage in which onsets are obligatorily present in the child's production forms — resulting in default CV syllables, even when the target syllable is onsetless — onsetless output forms appear (interestingly this development is not found for Portuguese (Freitas 1997)). Finally, complex onsets are produced. Characteristic of Dutch children's first complex onsets is that the two members of the onset differ maximally in sonority: it consists preferably of a stop plus a glide (Jakobson's principle of maximal contrast). Furthermore, three stages can be distinguished in the acquisition of obstruent-sonorant clusters: (a) at the first stage obstruent-sonorant clusters are simplified to single obstruents (again creating a maximal contrast, here between onset and nucleus); (b) at the next (optional) stage they are simplified to single sonorants (acquiring more subtle contrasts); and (c) finally, they are produced as obstruentsonorant clusters. A striking finding is that, while most children start with obstruent-sonorant clusters, some children first have /s/-obstruent clusters. Apparently, these involve two different, unrelated parameters. This is confirmed by Freitas' (1997) longitudinal study on the acquisition of Portuguese syllable structure in which she recorded six Portuguese children for at least one year: while Portuguese has both obstruent-sonorant and /s/obstruent clusters, children first acquire the /s/-obstruent clusters. Why and how these differences arise has to be the subject of further research. As for the development of rhymes in Dutch children's speech, Fikkert (1994a, b) distinguishes five stages. First, only open syllables are allowed, where vowel length is non-distinctive, again resulting in the default CV syllable with a simple onset and a simple rhyme. Second, branching rhymes, i.e. rhymes consisting of a nucleus and a coda (an obstruent), appear (maximal contrast between the vowel and following consonant). Third, branching nuclei occur, consisting of a long vowel or a short vowel plus a sonorant consonant (acquiring more subtle contrasts). Fourth, extrasyllabic positions are acquired, allowing syllables ending in a long vowel plus a consonant, or a short vowel plus a sonorant-obstruent cluster. Finally, syllables ending in two or more obstruents appear in the child's output forms. Interestingly, not all these stages are confirmed for English. Salidas & Johnson (1997) report that their subject can control vowel length from the onset of production. Of course, one factor that may be crucial in

235

Acquisition of phonology

explaining this difference is that vowel length in English contributes to syllable weight, whereas it does not in Dutch, making the vowel length distinction less salient for the Dutch language learner. Cross-linguistic considerations are important because they show that children do not just follow a strict path from unmarked to more marked. Rather, the phonological system of the target language as reflected in the input largely determines the way in which acquisition proceeds. Differences in the acquisition patterns are therefore to be expected but should fall within a limited range.

4.2. Word stress Until fairly recently, the literature on the acquisition of stress mainly focused on the following two questions: (1) whether children learn stress lexically or by rule; and (2) whether children are biased towards a particular foot type. Hochberg (1988a, b) argues that children do indeed learn stress rules, while Klein (1984) concludes that there is lexical primacy during the early stages of learning word stress. Allen & Hawkins (1978, 1980) found that English children are biased towards a trochaic pattern, with initial unstressed syllables often being deleted to fit this pattern. Hochberg (1988a, b), however, concludes that children approach the task of stress learning without a bias towards any particular stress type. The issue of stress acquisition has recently been addressed in the literature from a learnability perspective, without looking at actual acquisition data (Dresher & Kaye 1990; Gillis, Durieux, Daelemans & van den Bosch 1992); others base their work on psycholinguistic experiments (Echols 1987, 1988; Echols & Newport 1992; Gerken 1992a, b; 1994); and yet others analyse longitudinal data from children's development (Fikkert 1994a, b; Fee 1992; Demuth 1995a, b). Echols (1987, 1988) and Echols & Newport (1992) demonstrate that children are most likely to retain the stressed and final syllables of adult target words. They claim that these syllables are most salient and therefore best perceived by the child, following Waterson's (1971, 1989) principle of "what is best perceived is best produced". They make no claims about the child's own stress system. Gerken (1992a, b, 1994) shows that an account based solely on perception does not explain the facts and that children seem to have a preference for trochaic words. This is confirmed by the longitudinal study carried out by Fikkert (1994a), in which it is shown that iambic and trochaic target words are treated differently by children in that the former are more prone to truncation and show more stress errors, thus confirming Allen & Hawkins' observations. Fikkert shows further that, by

236

Paula

Fikkert

studying the child's production forms more carefully, a clear developmental pattern appears, as illustrated in (7): (7)

DEVELOPMENT OF DISYLLABIC TARGET WORDS Adult

a. b.

target

Stage 1

baby'baby"/'beibi-y ['beibii] gitaar 'guitar' /xii'tair/ ['ta:]

Stage 2

Stage 3

Stage 4

[Hbexbii] ['beibii] ['beibii] ['siita:] ['hii'tau] [^ii'tau]

The target in (7a) contains one foot; the one in (7b) more than one foot, as shown in (8): (8)

FOOT TEMPLATES OF TARGET WORDS

Wd

Wd (F)

ba

by

gi

taar

The child's forms at stage 1 all contain a single quantity-insensitive trochaic foot. At this stage the child maps the segmental content of the final (stressed) foot of the target words onto his or her own foot, as shown in (6c). The child's forms at stage 2 still contain exactly one foot, but the monosyllabic forms of stage 1 are now disyllabic. The transition from stage 1 to stage 2 may be triggered by the fact that the child's output in (7b) and the adult input forms display a mismatch in the number of syllables. None of the stress parameters are changed: since there are no stress mismatches the child has not (yet) encountered evidence that triggers the setting of a stress parameter from the default to the marked value. As a result the child forms are disyllabic, with initial stress for both initially and finally stressed target words at stage 2. When these new output forms are compared with the input forms, the mismatch in the number of syllables can be seen to be solved; however, now a stress mismatch exists. The existence of words with the same number of syllables but different stress patterns may trigger the setting of the quantity-sensitivity parameter to the marked value quantity-sensitive, since in a quantity-insensitive system words with the same number of syllables should have the same stress pattern. At stage 3 every closed syllable is considered heavy and forms a foot on its own. Moreover, the data show that the string of segments is fully parsed into feet and the main stress parameter is still not relevant: the child produces both feet with the same degree of stress.

237

Acquisition of phonology

When comparing his or her output forms with the input forms the child may detect that not all feet in the language have the same amount of stress, which may trigger the setting of the main stress parameter at stage 4. Now, the child's representation of the target words in (7) is adult-like. This account demonstrates that a close study of child data reveals the principled and systematic nature of development. The child builds up his or her grammar step by step. The transitions from one stage to the next can be understood as (a) the setting of one or more parameters from the default (unmarked) value to the marked; and/or (b) the extension of the child's template. Although metrical theory might not predict exactly what the intermediate stages are, the attested stages can easily be accounted for within the theory. It might be the case that the study of the acquisition of other stress systems will reveal different patterns, but the theory severely reduces the number of possible intermediate grammars. Also, it predicts that the initial stages are more or less equivalent, and independent of the language being acquired. Again, it is an empirical question whether this is true, and more research based on detailed longitudinal databases is required. Very recently, the model described above has triggered a series of studies, most of them carried out for the acquisition of English (cf. Demuth & Fee 1995; Kehoe & Stoel-Gammon 1997; Salidis & Johnson 1997). In those studies not all stages proposed by Fikkert (1994a) are confirmed. Invariably the assumption is made that Dutch and English prosodic word structure is largely similar. However, these historically closely related languages have several important differences: they differ at least with regard to syllable weight and length and with regard to extrametricality. The different input for Dutch and English children predicts differences in acquisition. It is therefore all the more regrettable that the comparison has not focused on differences in target systems. Hopefully, this will be undertaken in the near future. What additionally complicates the comparison of these studies with Fikkert's study is the fact that most of them used a new phonological framework — optimality theory — to account for and describe the results.

5. Optimality theoretic accounts of

acquisition

Optimality theory (OT) differs in a number of aspects from previous phonological theories. First, there are no rules or derivations. Instead, given an input, all possible output structures are generated. A languagespecific ranking of universal (innate; but see Boersma 1998) constraints (i.e. the grammar) selects the optimal candidate among all output candidates. The constraints are violable, but only minimally so: violation of highly

238

Paula Fikkert

ranked constraints eliminates the candidates that violate them. The optimal candidate is the one most harmonic, i.e. the one that incurs the fewest violations of highly ranked constraints. Second, while non-linear phonology cares about the structure of input representations, OT only pays attention to constraints on the output. Although the output optimally corresponds to the input, the input is not in focus. Third, there are basically two types of constraints: markedness (or well-formedness) constraints and faithfulness constraints. The latter type of constraints serves to minimise the differences between input and output. Markedness constraints eliminate marked output and favour unmarked structures. The two types generally are in conflict. The ranking of the constraints resolves this conflict: if faithfulness constraints outrank markedness constraints, input and output forms are minimally distinct; on the other hand, if markedness constraints outrank faithfulness constraints, the output structures may differ substantially from the input structures; the output structures will then reflect "the emergence of the unmarked" (McCarthy & Prince 1994). The acquisition task consists of detecting the language-specific ranking of universal constraints, on the basis of the adult target forms. The early child productions differ significantly from the forms they hear (input forms). To account for the discrepancy between the input form (i.e. the adult output form) and the child's output form, faithfulness constraints are generally ranked low (and are, therefore, allowed to be violated), while markedness constraints are ranked high (preferably unviolated), leading to the "emergence of the unmarked" (cf. Gnanadesikan 1996; Demuth 1996; Pater 1997; Goad 1998). As the child develops, the output forms become more and more faithful to the input forms: this is accounted for by reranking of the constraints: faithfulness constraints are promoted, markedness constraints demoted. Thus, the acquisition process involves the reranking of constraints. What triggers change is not discussed in the literature so far. OT allows for elegant accounts of phenomena that show the interaction of prosodic and segmental phenomena, such as the alignment of certain segmental features to certain edges of prosodic domains; that is, the theory makes it possible to formally and accurately express output constraints on child production forms that previously have been known as "templates", "recipes", "mould", "canonical forms", "prosodies" etc. These were mostly assumed to help the child structure lexical representations (cf. Waterson 1971,1987; Macken 1978), but did not necessarily assume that the child's underlying representation was fully specified and/or adult-like. This is the underlying assumption in most work in OT. How children acquire these underlying (input) representations has not (yet) been studied, but is not a trivial issue in the theory. To me it was refreshing to see that very recently

Acquisition of phonology

239

Dinnsen & Barlow (1998) address this question explicitly in their discussion of chain shifts in acquisition. Their conclusion is that even in an OT framework the process of acquisition (or change for that matter), can only be understood if one assumes underspecified input representations, which reflects constraints on input representations. Of course the field is very young and active, and future research will undoubtedly involve issues of whether OT can do without constraints on inputs. The answer to this question has obvious consequences for the way acquisition issues will be dealt with in OT.

6. Concluding remarks The question of how learning is accomplished in the presence of incomplete and contradictory input can be studied purely from a formal theoretical point of view, without looking at actual data. This is often referred to as the logical problem of acquisition. An important characteristic of any theory of grammar should be that it is learnable. Therefore, any theory should also provide an account of the acquisition process (cf. Dresher & Kaye 1990; Gillis, Durieux, Daelemans & van den Bosch 1992; Pulleyblank & Türkei 1996; Tesar & Smolensky 1998). I have shown in this article that research into the acquisition of phonology is ideally not only based on formal theories of phonology, but also on analyses of longitudinal data from child language, in which the complete set of data at different stages of development is taken into account. Different phonological theories, of course, make different predictions concerning the specific details of acquisition. Acquisition studies should help decide on which theory is better suited to account for the attested variation and uniformity in children's grammars. To conclude, although the first studies of acquisition of phonology date from some time ago, progress has been very slow, partly because the field is interdisciplinary, partly because theoretical frameworks change before they have been fully tested for acquisition, and partly because the study of the actual acquisition process is very time consuming. Nevertheless, by combining the efforts of theoretical phonologists, psycholinguists and researchers studying child language, we may hope to find an answer to the question of how phonology is acquired, which part of phonology is innate, and which part has to be learned.

240

Paula Fikkert

An Acquisition of Phonology

Bibliography

Allen, G. D. & S. Hawkins (1978). The development of phonological rhythm. Syllables and segments, edited by Bell & Hooper, 173-185. Allen, G.D. & S. Hawkins (1980). Phonological rhythm: definition and development. Child phonology: production, edited by Yeni-Komshian et al„ 227-256. Archibald, J., editor (1995). Phonological acquisition and phonological theory. Hillsdale, New Jersey: Lawrence Erlbaum. Barlow, J. (1997). A constraint-based approach of syllable onsets: evidence from developing systems. Doctoral dissertation, Indiana University, Bloomington. Beers, M. (1995). The phonology of normally developing and languageimpaired children. Doctoral Dissertation, University of Amsterdam. Beers, M. (1996). Acquisition of Dutch phonological contrasts within the framework of feature geometry theory. Proceedings of the UBC International Conference on Phonological Acquisition, edited by Bernardt et al., 28-41. Bell, A. & J.B. Hooper, editors (1978). Syllables and segments. New York: Elsevier-North Holland. Berg, T. (1992). Phonological harmony as a processing problem. Journal of Child Language 19, 225-257. Berman, R. A. (1977). Natural phonological processes at the one-word stage. Lingua 43,1-21. Bernardt, B. (1994). The prosodic tier and phonological disorders. First and second language phonology, edited by M. Yavas, 149-172. San Diego: Singular Publishing Group. Bernardt, B., J. Gilbert & D. Ingram, editors (1996). Proceedings of the UBC International Conference on Phonological Acquisition. Somerville, Mass.: Cascadilla Press. Bernardt, B. & J. Stemberger (1998). Handbook of phonological development. From the perspective of constraint-based nonlineair phonology. San Diego: Academic Press. Blache, S.E. (1978). The acquisition of distinctive features. Baltimore: University Park Press. Boersma, P. (1998). Functional phonology: formalizing the interactions between articulatory and perceptual drives. Doctoral dissertation, IFOTT/University of Amsterdam. [LOT International Series 11.] The Hague: Holland Academic Graphics. Branigan, G. (1976). Syllabic structure and the acquisition of consonants: the great conspiracy in word formation. Journal of Psycholinguistic Research 5,117-133.

Acquisition of phonology

241

Brown, R. (1973). A first language: the early stages. Cambridge, Mass.: Harvard University Press. Chomsky, Ν. & M. Halle (1968). The sound pattern of English. New York: Harper & Row. Crystal, D. (1986). Prosodic development. Language acquisition. Studies in first language development, edited by Fletcher & Garman, 174-197. Demuth, K. (1993). Issues in the acquisition of the Sesotho tonal system. Journal of Child Language 20, 275-301. Demuth, K. (1995). Markedness and the development of prosodic structure. Proceedings of the NELS 25, edited by J. Beckman, 13-26. GLSA, University of Massachusetts, Amherst. Demuth, K. (1996a). The prosodic structure of early words. From signal to syntax: bootstrapping from speech to grammar in early acquisition, edited by J. Morgan & K. Demuth, 171-184. Hillsdale, N.J.: Lawrence Erlbaum. Demuth, K. (1996b). Alignment, stress, and parsing in early phonological words. Proceedings of the UBC International Conference on Phonological Acquisition, edited by Bernardt et al., 113-125. Demuth, K. (to appear). Prosodic constraints and morphological development. Phonological, syntactic and neurophysiological aspects of early language acquisition, edited by J. Weissenborn & Β. Hoehle. Amsterdam: John Benjamins. Demuth, K. & J. Fee (1995). Minimal words in early phonological development. Manuscript, Brown University & Dalhousie University. Dinnsen, D.A. & J. A. Barlow (1998). On the characterization of a chain shift in normal and delayed phonological acquisition. Journal of Child Language 25, 61-94. Donegan, P. & D. Stampe (1979). The study of natural phonology. Current approaches to phonological theory, edited by D.A. Dinnsen, 126-173. Bloomington: Indiana University Press. Dresher, B.E. (1991). Cues and parameters in phonology. CLS 27. Dresher, Β. E. (1992). A learning model for a parametric theory in phonology. Formal grammar: theory and implementation, edited by R. Levine, 290-317. Oxford: Oxford University Press. Dresher, B.E. (1999). Charting the learning path: cues to parameter setting. Linguistic Inquiry 30, 27-67. Dresher, B.E. & Kaye, J.D. (1990). A computational learning model for metrical theory. Cognition 34,137-195. Echols, C.H. (1987). A perceptually-based model of children's firstwords.Doctoral Dissertation, University of Illinois. Echols, C.H. (1988). The role of stress, position and intonation in the representation and identification of early words. Papers and Reports on Child Language Development 2, 39-46.

242

Paula Fikkert

Echols, C. & Ε. Newport (1992). The role of stress and position in determining first words. Language Acquisition 2,189-220. Elbers, L. & J. Ton (1985). Pay pen monologues: the interplay of words and babble in the first word period. Journal of Child Language 12,551-565. Fee, J. (1991). Underspecification, parameters, and the acquisition of vowels. Unpublished doctoral dissertation, University of British Columbia, Vancouver. Fee, J. (1992). Exploring the minimal word in early phonological acquisition. Proceedings of the 1992Annual Conference of the Canadian Linguistic Association. Toronto Working Papers in Linguistics. Fee, J. (1995). Segments and syllables in early language acquisition. Phonological acquisition and phonological theory, edited by Archibald, 43-61. Fee, J. & D. Ingram (1982). Reduplication as a strategy of phonological development. Journal of Child Language 9, 41-54. Ferguson, C. A. (1975). Fricatives in child language acquisition. Proceedings of the 8th International Congress of Linguistics, edited by L. Heilmann, 647-664. Bologna: Societa editrice il Mulino. Ferguson, C.A. (1976). New directions in phonological theory: language acquisition and universale research. Current issues in linguistic theory, edited by R. W. Cole, 247-299. Bloomington: Indiana University Press. Ferguson, C.A. (1983). Reduplication in child phonology. Journal of Child Language 10, 239-243. Ferguson, C.A. & C. Farwell (1975). Words and sounds in early language acquisition. Language 51, 419-439. Ferguson, C.A. & O.K. Garnica (1975). Theories of phonological development. Foundations of language development, edited by E.H. Lenneberg & Ε. Lenneberg, 153-180. New York: Academic Press. Ferguson, C. Α., L. Menn & C. Stoel-Gammon, editors (1992). Phonological development: models, research, implications. Timonium, MD.: York Press. Fikkert, P. (1994a). On the acquisition of prosodic structure. Doctoral Dissertation, HIL/Leiden University. [HIL Dissertations 6.] The Hague: Holland Academic Graphics. Fikkert, P. (1994b). On the acquisition of rhyme structure in Dutch. Linguistics in the Netherlands 1994, edited by R. Bok-Bennema & C. Cremers, 37-48. Amsterdam: John Benjamins. Fikkert, P. (1995a). Models of acquisition: how to acquire stress. Proceedings of the NELS 25, edited by J. Beckman, 27-42. GLSA, University of Massachusetts, Amherst. Fikkert, P. (1995b). The role of negative evidence in the acquisition of phonology. Parameters, edited by M. Verrips & F. Wijnen, 33-52 [Am-

Acquisition of phonology

243

sterdam Series in Child Language Development 4.] University of Amsterdam. Fikkert, P. (1998). Acquisition of Dutch phonology. The acquisition of Dutch, edited by S. Gillis & A. de Houwer, 163-222. Amsterdam: John Benjamins. Fikkert, P. (to appear). Compounds: triggering acquisition of prosodic structure. Phonological, syntactic and neurophysiological aspects of early language acquisition, edited by J. Weissenborn & Β. Hoehle. Amsterdam: John Benjamins. Fikkert, P. & M.J. Freitas (1997). Acquisition of syllable structure constraints: evidence from Dutch and Portuguese. Language acquisition: Knowledge representation and processing. Proceedings of GALA 97, 217-222. Edinburgh: Edinburgh University Press. Fletcher, P. & M. Garman, editors (1986). Language acquisition. Studies in first language development. Cambridge: Cambridge University Press. Freitas, M. J. (1997). Aquisiqäo da estrutura sildbica do Portugues Europeu. Doctoral dissertation, University of Lisboa. Fudge, E.C. (1969). Syllables. Journal of Linguistics 5, 253-287. Gennari, S & K. Demuth (1997). Syllable omission in the acquisition of Spanish. Proceedings of the 21th Annual Boston University Conference on Language Development, edited by A. Stringfellow, A. D. Cahana-Amitay, E. Hughes & A. Zukowski, 182-193. Somerville, Mass.: Cascadilla Press. Gerken, L.A. (1992). 'Raffes and 'nanas: a metrical template account of children's weak syllable omissions. Manuscript, State University of New York at Buffalo. Gerken, L.A. (1994a). Young children's representation of prosodic phonology: evidence from English-speakers' weak syllable productions. Journal of Memory and Language 33,19-38. Gerken, L.A. (1994b). A metrical template account of children's weak syllable omissions from multisyllabic words. Journal of Child Language 21, 565-584. Gerken, L.A. (1996). Prosodic structure in young children's language production. Language 72, 683-712. Gillis, S., G. Durieux, W. Daelemans & A. van den Bosch (1992). Exploring artificial learning algorithms: learning to stress Dutch simplex words. Antwerp Papers in Linguistics 71. Gnanadesikan, A. (1996). Child phonology in optimality theory: ranking markedness and faithfulness constraints. Proceedings of the 20th Annual Boston University Conference on Language Development, edited by A. Stringfellow, A.D. Cahana-Amitay, E. Hughes & A. Zukowski, 237-248. Somerville, Mass.: Cascadilla Press.

244

Paula Fikkert

Goad, Η. (1989). Language acquisition and the Obligatory Contour Principle. Stanford University Papers and Reports on Child Language Development 28,115-122. Goad, H. (1998). Consonant harmony in child language: an optimalitytheoretic Account. Generative studies in the acquisition of phonology, edited by S.-J. Hannahs & M. Young-Scholten. Goad, H. & D. Ingram (1987). Individual variation and its relevance to a theory of phonological acquisition. Journal of Child Language 14, 419-432. Gregoire, A. (1937, 1947). L'apprentissage du langage. Vol. 1 (1937); Vol. 2 (1947). Liege, Paris: Droz. Hale, M. & C. Reiss (1998). Formal and empirical arguments concerning phonological acquisition. Linguistic Inquiry 29, 656-683. Hammond, M. (1990). Parameters of metrical theory and learnability. Logical issues in language acquisition, edited by Roca, 47-62. Hannahs, S.-J. & M. Young-Scholten, editors, (1998). Generative studies in the acquisition of phonology. Amsterdam: John Benjamins. Hochberg, J.G. (1987). The acquisition of word stress rules in Spanish. Papers and Reports on Child Language Development 26, 56-63. Hochberg, J.G. (1988a). First steps in the acquisition of Spanish stress. Journal of Child Language 15, 273-292. Hochberg, J.G. (1988b). Learning Spanish stress: developmental and theoretical perspectives. Language 64, 638-706. Ingram, D. (1974). Phonological rules in young children. Journal of Child Language 1,29-64. Ingram, D. (1976/1989). Phonological disability in children. London: Edward Arnold. Ingram, D. (1978). The role of the syllable in phonological development. Syllables and segments, edited by Bell & Hooper, 143-156. Ingram, D. (1981). Procedures for the phonological analysis of children's language. Baltimore: University Park Press. Ingram, D. (1986). Phonological development: production. Language acquisition. Studies in first language development, edited by Fletcher & Garman, 223-239. Ingram, D. (1988). Jakobson revisited: some evidence from the acquisition of Polish. Lingua 75, 55-82. Ingram, D. (1989). First language acquisition. Method, description and explanation. Cambridge: Cambridge University Press. Iverson, G. & D. Wheeler (1987). Hierarchical structures in child phonology. Lingua 73, 243-257.

Acquisition of phonology

245

Jakobson, R. (1939/62). Why 'mama' and 'papa'? In R. Jakobson, Selected writings I: phonological studies. The Hague: Mouton 1939, expanded edition reprinted in 1962. Jakobson, R. (1941/68). Child language, aphasia and phonological universals. The Hague: Mouton. Jusczyk, P. W. (1992). Developing phonological categories from the speech signal. Phonological development: models, research, implications, edited by Ferguson et al., 17-64. Jusczyk, P. W. (1997). The discovery of spoken language. Cambridge, Mass.: The MIT Press. Kehoe, M. & C. Stoel-Gammon (1996). Acquisition of stress: an investigation of rhythmic processes in English-speaking children's multisyllabic word productions. Proceedings of the 20th Annual Boston University Conference on Language Development, edited by A. Stringfellow, A. D. Cahana-Amitay, E. Hughes & A. Zukowski. Somerville, Mass.: Cascadilla Press. Kehoe, M. & C. Stoel-Gammon (1997). The acquisition of prosodic structure: an investigation of current accounts of children's prosodic development. Language 73,113-144. Kiparsky, P. & L. Menn (1977). On the acquisition of phonology. Language learning and thought, edited by J. Macnamara, 47-78. New York: Academic Press. Klein, H.B. (1984). Learning to stress: a case study. Journal of Child Language 11, 375-390. Leopold, W.F. (1939-49). Speech development of a bilingual child. A linguist's record. [4 volumes: 1939, 1947, 1949,1949.] Evanston: Northwestern University Press. Levelt, C.C. (1993). Consonant harmony: a reanalysis in terms of vowelconsonant interaction. The acquisition of Dutch, edited by M. Verrips & F. Wijnen, 37—49. [Amsterdam Series in Child Language Development, vol 1.] University of Amsterdam. Levelt, C.C. (1994). On the acquisition of a place. Doctoral Dissertation, HIL/Leiden University. [HIL Dissertations 8.] The Hague: Holland Academic Graphics. Levelt, C.C. (1996). Consonant-vowel interactions in child language. Proceedings of the UBC International Conference on Phonological Acquisition, edited by Bernardt et al., 229-239. Lleo, C. (1990). Homonymy and reduplication: on the extended availability of two strategies in phonological acquisition. Journal of Child Language 17,267-278. Locke, J. (1983). Phonological acquisition and change. New York: Academic Press.

246

Paula Fikkert

Lohuis-Weber, H. & W. Zonneveld (1996). Phonological acquisition and Dutch word prosody. Language Acquisition 5, 245-283. Macken, Μ. (1978). Permitted complexity in phonological development: one child's acquisition of Spanish consonants. Lingua 44, 219-253. Macken, Μ. (1979). Developmental reorganization of phonology: a hierarchy of basic units of acquisition. Lingua 49,11-49. Macken, Μ. (1980). The child's lexical representation: the 'puzzle — puddle — pickle' evidence. Journal of Linguistics 16, 1-17. Macken, Μ. (1986). Phonological development: a cross-linguistic perspective. Language acquisition. Studies in first language development, edited by Fletcher & Garman, 251-268. Macken, Μ. (1987). Representation, rules and overgeneralization in phonology. Mechanisms of language acquisition, edited by B. MacWhinney, 367-397. Hillsdale, New Jersey: Lawrence Erlbaum. Macken, Μ. (1992). Where's phonology? Phonological development: Models, research, implications, edited by Ferguson et al, 249-269. Macken, Μ. (1995). Phonological acquisition. The handbook of phonological theory, edited by J. A. Goldsmith, 671-696. Oxford: Blackwell. Macken, Μ. & C.A. Ferguson (1983). Cognitive aspects of phonological development: model, evidence and issues. Children's language, 4, edited by K.E. Nelson, 255-282. Hillsdale, New Jersey: Lawrence Erlbaum. MacWhinney, B. (1995). The CHILDES project: tools for analyzing talk. [2nd edition.] Hillsdale, New Jersey: Lawrence Erlbaum. Matthei, E. (1989). Crossing boundaries: more evidence for phonological constraints on early multi-word utterances. Journal of Child Language 16, 41-54. McCarthy, J. & A. Prince (1993). Prosodic morphology I: constraint interaction and satisfaction. Manuscript, University of Massachusetts, Amherst & Rutgers University, New Brunswick, NJ. McCarthy, J. & A. Prince (1994). The emergence of the unmarked: optimality in prosodic morphology. Proceedings of the NELS 24, edited by M. Gonzalez, 333-379. GLSA, University of Massachusetts, Amherst. McDonough, J. & S. Myers (1991). Consonant harmony and planar segregation in child language. Manuscript, UCLA & University of Texas at Austin. Menn, L. (1971). Phonotactic rules in beginning speech: a study in the development of English discourse. Lingua 27, 225-251. Menn, L. (1977). An autosegmental approach to child phonology — first explorations. Harvard studies in phonology I, edited by G.N Clements, 315—334. Linguistics Department, Harvard University. Menn, L. (1978). Phonological units in beginning speech. Syllables and segments, edited by Bell & Hooper, 157-172.

Acquisition of phonology

247

Menn, L. (1980). Phonological theory and child phonology. Child phonology: production, edited by Yeni-Komshian et al., 23-41. Menn, L. (1983). Development of articulator^ phonetic, and phonological capabilities. Language production, 2, edited by B. Butterworth, 1-50. London: Academic Press. Menn, L. & E. Matthei (1992). The "two-lexicon" account of child phonology: looking back, looking ahead .Phonological development: models, research, implications, edited by Ferguson et al., 211-247. Menn, L. & C. Stoel-Gammon (1995). Phonological development. The handbook of child language, edited by P. Fletcher & B. MacWhinney, 367-397. 335-360. Oxford: Blackwell. Mohanan, K.P. (1992). Emergence of complexity in phonological development. Phonological development: models, research, implications, edited by Ferguson et al., 635-662. Morgan, J. L. & K. Demuth, editors (1996). Signal to Syntax: bootstrapping from speech to grammar in early acquisition. Hillsdale, New Jersey: Lawrence Erlbaum. Moskowitz, A. I. (1970). The two-year-old stage in the acquisition of English phonology. Language 46, 426-441. Nouveau, D. (1994). Language acquisition, metrical theory and optimality. Doctoral dissertation, Utrecht University. Olmsted, D. L. (1971). Out of the mouth of babes: earliest stages in language learning. The Hague: Mouton. Pater, J. (1997). Minimal violation and phonological development. Language Acquisition 6, 201-253. Pater, J. & J. Paradis (1996). IVuncation without templates in child phonology. Proceedings of the 20th Annual Boston University Conference on Language Development, edited by A. Stringfellow, A.D. CahanaAmitay, E. Hughes & A. Zukowski, 540-551. Somerville, Mass.: Cascadilla Press. Peters, A.M. (1977). Language learning strategies: does the whole equal the sum of the parts? Language 53, 560-573. Peters, A. M. (1983). The units of language acquisition. Cambridge University Press. Preyer, W. (1889). The mind of the child. New York: Appleton. [Translation of the orginal German edition from 1882.] Priestly, T. (1977). One idiosyncratic strategy in the acquisition of phonology. Journal of Child Language 4, 45-65. Prince, A. & P. Smolensky (1993). Optimality theory: constraint interaction in generative grammar. Manuscript, Rutgers University, New Brunswick, NJ & University of Colorado at Boulder.

248

Paula Fikkert

Pulleyblank, D. & W. Türkei (1996). Optimality theory and learning algorithms: the representation of recurrent featural asymmetries. Current trend in phonology: models and methods, J. Durand & Β. Laks, 653-684. European Studies Research Institute, University of Salford, Salford, Manchester, England. Rice, K. & R Avery (1995). Variability in a deterministic model of language acquisition: a theory of segmental elaboration. Phonological acquisition and phonological theory, edited by Archibald, 23-42. Roca, I. M., editor (1990). Logical issues in language acquisition. Dordrecht: Foris Publications. Salidis, J. & J. Johnson (1997). The production of minimal words: a longitudinal case study of phonological development. Language Acquisition 6, 1-36. Schwartz, R. & L.B. Leonard (1982). Do children pick and choose? An examination of phonological selection and avoidance in early lexical acquisition. Journal of Child Language 9, 319-336. Schwartz, R., L.B. Leonard, D.M. Frome Loeb & L. Swanson (1987). Attempted sounds are sometimes not: an expanded view of phonological selection and avoidance. Journal of Child Language 14, 411-418. Shibamoto, J.S. & D.L. Olmsted (1978). Lexical and syllabic patterns in phonological acquisition. Journal of Child Language 5, 417-457. Simonsen, H. G. (1990). Barns fonologi: system ofvariasjon hos tre Norske og ett Samoisk Barn. Doctoral Dissertation, University of Oslo. Smith, Ν. V. (1973). The acquisition of phonology: a case study. Cambridge University Press. Smolensky, P. (1996). On the comprehension/production dilemma in child language. Linguistic Inquiry 27, 720-731. Spencer, A. (1986). Towards a theory of phonological development. Lingua 68, 3-38. Spencer, A. (1989). Issues in the theory of phonological acquisition. NIAS report, Faculty of Languages, Polytechnic of Central London. Stampe, D. (1973). A dissertation on natural phonology. Doctoral Dissertation, University of Chicago. Stemberger, J. P. (1988). Between word processes in child phonology. Journal of Child Language 15, 39-62. Stemberger, J. P. (1991). Radical underspecification in language production. Phonology 8, 73-112. Stemberger, J. P. (1992). A connectionist view of child phonology: phonological processing without phonological processes. Phonological development: models, research, implications, edited by Ferguson et al., 165-189. Stemberger, J. P. (1993). Glottal transparancy. Phonology 10,107-138.

Acquisition of phonology

249

Stemberger, J. P. & C. Stoel-Gammon (1991). The underspecification of coronals: evidence from language acquisition and performance errors. The special status of coronals. Internal and external evidence, C. Paradis & J.-F. Prunet, 181-199. San Diego: Academic Press. Stern, C. & W. Stern (1907). Die Kindersprache. Leipzig: Barth. Stoel-Gammon, C. (1983). Constraints on consonant-vowel sequences in early words. Journal of Child Language 10, 455-457. Stoel-Gammon, C. & J. A. Cooper (1984). Patterns of early lexical and phonological development. Journal of Child Language 11, 247-271. Stoel-Gammon, C. & J. P. Stemberger (1994). Consonant harmony and phonological underspecification in child speech. First and second language phonology, edited by M. Yavas, 63-80. San Diego: Singular Publishing Group. Studdert-Kennedy, M. & E.W. Goodell (1992). Gestures, features and segments in early child speech. Haskins Laboratories Status Report on Speech Research, SR-111/112, 1-14. Templin, M. (1957). Certain language skills in children. [University of Minnesota Institute of Child Welfare Monograph Series 26.] Minneapolis: University of Minnesota Press. Tesar, B. & P. Smolensky (1998). Learnability in optimality theory. Linguistic Inquiry 29, 229-268. Tse, J.K.-P. (1978). Tone acquisition in Cantonese: a longitudinal case study. Journal of Child Language 5,191-204. Velleman, S. (1996). Metathesis highlights feature-by-position constraints. Proceedings of the UBC International Conference on Phonological Acquisition, edited by Bernardt et al. Velten, Η. (1943). The growth of phonemic and lexical patterns in infant speech. Language 19, 281-292. Vihman, M. (1978). Consonant harmony: its scope and function in child language. Universals of human language, Vol. 2: Phonology, edited by J.H. Greenberg, 281-334. Stanford: Stanford University Press. Vihman, M. (1981). Phonology and the development of the lexicon: evidence from children's errors. Journal of Child Language 8, 239-264. Vihman, M. (1996). Phonological development. The origins of language in the child. Oxford: Blackwell. Vihman, M. & S. L. Velleman (1989). Phonological reorganization: a case study. Language and Speech 32, 149-170. Vihman, M., S.L. Velleman & L. McCune (1993). How abstract is child phonology? Towards an intergration of linguistic and psychological approaches. First and second language phonology, edited by M. Yavas, 9-44. San Diego: Singular Publishing Group.

250

Paula Fikkert

Waterson, N. (1971). Child phonology: a prosodic view. Journal of Linguistics 7,179-211. Waterson, N. (1987). Prosodic phonology: the theory and its application to language acquisition and speech processing. Newcastle upon Tyne: Grevatt & Grevatt. Wijnen, F., Ε. Krikhaar & Ε. den Os (1994). The (non)realization of unstressed elements in children's utterances: evidence for a rhythmic constraint. Journal of Child Language 18, 59-83. Yeni-Komshian, G.H., J.F. Kavanagh & C.A. Ferguson, editors (1980). Child phonology: production. New York: Academic Press.

Tone An overview San Duanmu

1.

Introduction

This article surveys developments and achievements in research on tone in the past thirty years, primarily in the generative framework, and outstanding issues that pertain to current research. I begin with a discussion of what tone languages are and point out some similarities between tone and intonation. 1.1. Tone languages and non-tone

languages

It is not hard to identify some languages that are clearly tone languages. A good example is Standard Chinese (also called Mandarin), which has four tones on full syllables, shown in (1). (1)

CONTRASTIVE TONES IN STANDARD CHINESE (MANDARIN)

[mal] [ma2] 'mother' 'hemp'

[ma3] "horse'

[ma4] 'to scold'

The numbers 1-4 refer to the four tones: tone 1 is a high, tone 2 is a rise, tone 3 is a low (or low-rise in final position), and tone 4 is a fall. The four words in (1) are phonologically identical in all respects except tone (I forgo the issue of duration for the moment). In other words, tone (or pitch contour) is lexically contrastive in Standard Chinese (for other functions of tone, see Section 3.14). Most tone languages are found in Asia and Africa. In addition, tone had been reported in some European and Native American languages. It is not hard, either, to identify some languages that are non-tone

252

San Duanmu

languages. For example, English is typically considered a non-tone language. When the word [kaet] 'cat' is said with different tones (or pitch contours), additional meanings are added, such as surprise, doubt, disbelief, and so on; however, the word still refers to the same animal. Thus, tone is not lexically contrastive in English. However, difficulties arise when one tries to classify languages as either tonal or non-tonal. A well-known example is Japanese, in which every word has a fixed tone pattern. This can be seen in (2), taken from McCawley (1978, p. 113), where syllables are separated by a hyphen, and where Η is a high tone and L is a low tone. (2)

PITCH-ACCENT IN JAPANESE

ßa-ki-ga ka-Äi-ga ka-ki-ga

H-L-L L-H-L L-H-H

'oyster' 'fence' 'persimmon'

The examples show that tone is lexically contrastive in Japanese, as it is in Chinese. On the other hand, as noted by McCawley (1965) (see also Block 1946; Haraguchi 1977), the tone pattern of a Japanese word is predictable if we posit an abstract "accent" position for each word, shown by italics in (2). This analysis accounts for the fact that, for a given word form, there are only as many possible tonal patterns as there are syllables (ignoring "unaccented" words). For example, the form in (2) has three syllables, so it has three possible tone patterns. It should be noted that, unlike stress, accent is not necessarily accompanied by greater duration or amplitude. Apart from its effect on pitch, accent is hardly felt by native Japanese speakers; I will return to this point below. Since tone (or pitch) is predictable from accent in Japanese, it is often called a "pitch-accent" language. But English also has pitch contours. In addition, the contours are quite predictable (e.g. Goldsmith 1976; Liberman 1975; Pierrehumbert 1980). For this reason, English has been called a tone language, too. For example, Goldsmith suggests that in neutral intonation English words have the tone pattern (M)HL, where Μ is a mid tone and Η is linked to the stressed syllable. Some examples are shown in (3). (3)

TONE IN ENGLISH

Chi-ca-go Bos-ton

M-H-L H-L

A-me-ri-ca Ca-na-da

M-H-L-L H-L-L

Analyses like (2) and (3) can be extended to African languages (e.g., Goldsmith 1982; Kenstowicz 1987; Sietsema 1989; Kenstowicz & Kisseberth 1990), American Indian languages (e.g. Hinton 1991), and European tone or pitch-accent languages, such as Lithuanian (Halle & Vergnaud 1987) and Serbo-Croatian (Inkelas & Zee 1988). Such analyses raise serious

253

Tone

questions for a typology of tone languages, as noted by Hyman (1978) and McCawley (1978). In particular, the distinction between tone and non-tone languages is no longer obvious. In addition, if English and Japanese are tone languages, one wonders what language is not. 1.2. Tone and

intonation

The pitch contour in English is often called intonation. Indeed, all "nontone languages" have intonation (I will return to the question of whether tone languages can have intonation). Although both tone and intonation are characterized by pitch (for a discussion on the difference between pitch and the fundamental frequency, see Beckman 1986), one may still wonder whether they are different. For example, are tone and intonation produced by the same articulatory mechanism? Are tone and intonation perceptually the same? Should tone and intonation be represented with the same phonological features? Phonetically, there is no evidence that tone and intonation are made with different articulatory mechanisms (cf. Zemlin 1981). In addition, as far as listeners are concerned, tone (in Chinese) and intonation (in English) are interchangeable. I will cite two examples. First, when English words are borrowed into Cantonese Chinese, a stressed English syllable usually becomes a high toned Cantonese syllable. This is shown in (4), from Chao (1980, p. 42), where stress in English is indicated by italics. (4)

T O N E IN BORROWING

English

Cantonese H-L

Oakland Pacific

[ok-l0n] M-H-L [pha-si-wik]

Obviously, Cantonese speakers hear the high pitch on a stressed English syllable as a high tone. Second, Standard Chinese has a rule that changes tone 3 (low) to tone 2 (rise) before tone 3, or 3 2/ 3. Cheng (1968) noted that in code switching, tone 3 changes to tone 2 before an unstressed English syllable as well, but not before a stressed one. This is shown in (5), where Chinese is transcribed in Pinyin. (5)

TONE IN CODE SWITCHING BETWEEN C H I N E S E AND ENGLISH

a. b.

xiao2 professor (*xiao3 professor) 'small professor' xiao3 lecturer (*xiao2 lecturer) 'small lecturer'

254

San Duanmu

The word xiao3 'small' is underlyingly tone 3. It changes to tone 2 in (5a) but not in (5b). The reason is that in (5a) the following syllable pro- is unstressed and has a low tone (although Goldsmith marks it as M, see Section 3.1); this is perceived by Standard Chinese speakers as tone 3 (low) and as a result the rule 3->2 is triggered. In (5b), the following syllable tecis stressed and has a high tone, which does not trigger tone change. Finally, let us see whether there are reasons to posit different phonological features for tone and intonation. Traditionally, tone refers to the pitch contour of a syllable or a word, and intonation refers to the pitch contour of a phrase or an utterance. But as Goldsmith (1981) pointed out, intonation in English consists of a sequence of word tones, similar to what one finds in a "tone" language. In addition, tonal behavior in English is similar to that in other tone languages. For example, when the word tone (M)HL falls on 'Japan', we get M-HL, where HL on the second syllable is realized as a fall; this is what one finds in tone languages (see below). We have seen, then, that tone and intonation display certain similarities. This point will recur in the rest of this article.

2. Developments

in the past 30 years

In early works of generative phonology (1950s and 1960s), there was very little mention of tone. This is probably because a main issue then was distinctive features, and tone did not seem to fit in easily. In particular, distinctive features work well for the representation of segments, but tone has traditionally been thought to be a "suprasegmental" property, a property that belongs to units larger than the segment (e.g. Pike 1948; Firth 1957). Another difficulty is that some tones seem to require "trajectory" features, such as "rise" or "fall", whereas regular segmental features are all "static", such as [round] or [voice], but not [increasing round] or [diminishing voice]. The lack of discussion on tone or intonation left a striking gap in the monumental Sound pattern of English (SPE\ see Chomsky & Halle 1968, p. 329). Wang (1967) made the first serious attempt to represent tone with distinctive features. He posited seven binary features to represent all tones. However, out of the seven features, four were trajectory features: [contour], [rise], [fall], and [convex]. In addition, with seven features many possible tones are predicted, of which only 13 were actually found. Perhaps because of this, Wang's work was not quickly adopted by other generative phonologists. A year after SPE appeared, Woo (1969) proposed a theory of tone features that had great influence on later works. Woo's theory has two parts. First, all contour tones are combinations of level tones. Second, tone

255

Tone

bearing units are segments in the rime and each tone bearing unit can carry just one tone. Woo's theory is illustrated in (6). Following Chao (1930) and Wang (1967), Woo posited five levels, but for illustration I will focus on two, Η and L; TBU stands for tone bearing unit. (6)

Woo's (1969) THEORY OF TONE FEATURES Traditional term Tone features TBUs needed High Η 1 Low L 1 Rise LH 2 Fall HL 2 Fall-rise HLH 3 Rise-fall LHL 3

In Woo's analysis, a fall is HL, a rise is LH, and a fall-rise is HLH. This is similar to treating the diphthong [ai] as a combination of [a] and [i] and the triphthong [auo] as a combination of [a], [u], and [a]. In addition, a short/light (monomoraic) syllable has one rime segment, so it can only carry a level tone. A simple contour tone (rise or fall) consists of two level tones and must fall on a long/heavy (bimoraic) syllable, which has two rime segments. A complex contour tone (fall-rise or rise-fall) consists of three level tones and must fall on a syllable that has three rime segments (trimoraic). For example, in Standard Chinese, all full syllables are long. In particular, tones 1, 2, 4, and the low version of tone 3 have two rime segments and are HH, LH, LL, and HL respectively; thus, [ma] in (1) should be [maa]. In addition, the low-rise version of tone 3 has three rime segments and is LLH. Woo's work was significant in two ways. First, it proposed that tone could be represented with "static" features after all. Second, Woo's work was based on Chinese, where most contour tones are found. If Chinese tones do not require trajectory features, tones in other languages are unlikely to either. The idea that contour tones are made of level tones was essentially adopted by subsequent generative phonologists. However, the idea that there is a direct relation between the length of a syllable and the number of tones it can carry was often ignored, although it is, in my view, essentially correct. I will return to it below. The next development came with the works of Leben & Williams . Leben (1971,1973) suggested that tones should be represented on an independent tier, parallel to the segment tier. Williams (1976) further suggested that the two tiers are coordinated by "tone mapping rules", which later became known as "association lines" and "association conventions". This line of research culminated in the influential work of Goldsmith (1976), through which the term "autosegmental phonology" became known

256

San Duanmu

(other equivalent terms are three-dimensional phonology, nonlinear phonology, and multi-tiered phonology). In the following decade, a large number of autosegmental studies ensued. Among works on tone, most dealt with African languages (e.g. Clements 1978; Clements & Ford 1979; Odden 1981; Hyman 1981; Pulleyblank 1983; Clements & Goldsmith 1984; Cassimjee 1987; Myers 1987, and many others). Main issues of interest included the association conventions, the representation of multiple tone levels, default tones, floating tones, downdrift and downstep, prelinked tones, accent marking, the OCP (Obligatory Contour Principle) effect, and the interaction between tone and morphosyntactic structure. Downdrift (a Η after a L is lower in pitch than the preceding H) and downstep (a Η is lower in pitch than a Η immediately preceding it) were later found to be the same phenomenon after it was realized that a downstepped Η always had a floating L right in front of it. Generative works have also been done on European tone (or pitch-accent) languages, such as Swedish, Norwegian, Serbo-Croatian, and Lithuanian (e.g. Bruce & Garding 1978; Garding 1979; Fretheim 1981; Withgott & Halvorsen 1988; Inkelas & Zee 1988; Halle & Vergnaud 1987). There were fewer generative studies on Asian tones in the 1970s and 1980s, with most of these focusing on Chinese languages. Main works included those of Yip (1980), Wright (1983), Chan (1985), Shih (1986), and Chen (1987). Primary issues of interest were the representation of contour tones, the representation of tone sandhi, and domains of tone sandhi. The Chinese dialects that received the most attention were Standard Chinese, in particular its tone 3 sandhi rule, Tianjin, Southern Min dialects, and Wu dialects. Besides "real" tone languages, tone-related work has also been done on other languages, such as English, Japanese, Dutch, and German (Liberman 1975; Bing 1979; Ladd 1980; Pierrehumbert 1980; Poser 1984; Beckman & Pierrehumbert 1986; Gussenhoven 1988; F6ry 1989, among others). In particular, Liberman (1975) has argued that English intonation can be analyzed in terms of a stress system and a tone system, where the tone system consists of a single sequence of Hs and Ls. This approach, now a standard in generative analyses of "intonation" languages, suggests that tone and intonation are essentially the same phenomenon. Parallel to the generative study on tone, four other developments should be mentioned. First, there was an effort to find universale of tone through cross-linguistic survey (e.g. Fromkin 1972; Hyman & Schuh 1974; and Maddieson 1978). Second, there were works on tonogenesis in Asian languages, i.e., the emergence of tone as the result of the loss of consonant features (e.g. Haudricourt 1954; Maran 1973 and Matisoff 1973). Third, some phonetic models on tone were proposed (e.g. Halle & Stevens 1971; Hombert 1978). Finally, there is a continuous flow of descriptive works by

257

Tone

linguists in China, mostly written in Chinese. Of special interest are experimental studies on Standard Chinese (e.g. Lin, Yan & Sun 1984; Lin 1985; Yan & Lin 1988; He & Jin 1992; Wang & Wang 1993), a series of works on Tibetan (e.g. Hu 1980; Hu, Qu & Lin 1982; Qu 1981), and a large number of studies on the Wu dialects, especially the work of Xu, Tang, You, Qian, Shi & Shen (1988) on Shanghai.

3. Current issues relating

to tone

Despite the tremendous progress in tone research in the past three decades, many issues remain quite open. Consequently, in discussing current issues, I will often speculate on what I think is likely to be the right solution. 3.1. Levels of tone There have been many discussions on how many tone levels are needed to describe all languages (e.g. Chao 1930; Wang 1967; Halle & Stevens 1971; Anderson 1978; Yip 1980; Clements 1983; Huang 1985; Hyman 1986; van der Hulst & Snider 1993), but there is still no agreement. Phonetically, pitch is the primary perceptual correlate of tone, and in real speech there can be many pitch levels. But if we focus on phonemic levels, i.e., those that are distinctive, the number of levels is quite small. In particular, if downstep or downdrift belongs to phonetic implementation, which need not be represented with separate levels, far fewer phonemic levels are needed. In African languages, two phonemic levels, Η and L, are often sufficient, although many more have been posited (I return to "depressed" tones later). The same is true for English. For example, Goldsmith (1981) posited (M)HL as the neutral English word tone, where syllables before stress have a higher pitch level than syllables after stress. But if we assume downstep (or what Beckman & Pierrehumbert 1986 call "catathesis"), by which both Η and L descend gradually, there is no need to posit M, and the English word tone can simply be (L)HL. For this reason, Liberman (1975) and Pierrehumbert (1980) only assumed two levels for English, Η and L. In Asian languages, three or four contrastive levels are quite common; five contrastive levels have also been reported (Chang 1953; Shi, Shi & Liao 1987). The widely used system of Chao (1930), adopted by the International Phonetic Association (1989), posits five levels. Anderson (1978) also posits five levels, primarily because of Chang's (1953) report that in certain MiaoYao languages five levels are found on isolated monosyllables. If tones are to be represented by distinctive features, one needs to convert multiple levels into binary features. It is possible to translate any

San Duanmu

258

multi-level system into a binary system; the challenge is to find justification for the translation. For example, English vowels have three heights, for which SPE uses two binary features, [high] and [low]. The analysis is justified by the fact that there are rules that refer to high and mid vowels, captured by [-low], and rules that refer to mid and low vowels, captured by [-high]. However, many Asian tone language have few tone sandhi rules, especially those that involve tone split and tone spreading, which might provide insight into the composition of the tones. An important tone model was suggested by Yip (1980, 1993), in which four levels are posited. The four levels are divided into two "Registers", [+upper] and [-upper] (not to be confused with a more loosely used notion of "register", which simply means a tone level). Each Register is divided into two levels, [+raised] and [-raised], which I will write as [+H] and [ - H ] and call them the "Pitch" feature (in contrast to "pitch", which refers to phonetic pitch height). Thus, the four tone levels are represented by two binary features, as shown in (7). (7)

YIP'S ( 1 9 8 0 ) MODEL OF TONE FEATURES

Register +upper -upper

Pitch +H -H +H -H

I II III IV

The strongest argument for Yip's model, in my view, is that Register is independently related to the voicing of the onset consonant. In particular, in those Chinese languages that have retained onset voicing in obstruents, [+upper] Register invariably goes with a voiceless onset and [-upper] Register invariably goes with a voiced onset. The same effect is found in many African languages, where a voiced obstruent consonant lowers the pitch of a neighboring tone and makes it [-upper] Register. I will return to this point below. An apparent shortcoming of Yip's model is that it cannot represent more than four levels. But reports of five (or more) levels are few and sketchy. There are, in addition, ways to accommodate five levels in Yip's system; for example, if one level is toneless, then the rest can be represented by the features in (7). If, however, more levels are needed, one can perhaps posit three tones in each Register, giving a total of six levels. Clear data that bear on this issue are still wanting. Multiple tone levels also raise a question in perception. Since speakers differ both in pitch height and in pitch range, tone levels must be interpreted in relative terms. Now given (7), when one hears a higher tone followed by a lower tone, there are six possibilities: III-IV, II-III, II-IV, I-II, I-III, and

259

Tone

I-IV. How does the listener decide which is correct? This question is addressed in the next section. 3.2. Pitch and

Register

For all tones, the primary phonetic correlate is pitch. However, tones in [-upper] Register are usually accompanied by a special voice quality, sometimes referred to as breathy or murmured. This is true not only of Chinese languages (e.g. Cao & Maddieson 1992), but also of African languages, such as Zulu (e.g. Laughren 1984 and references therein). Although the role of voice quality is often ignored in tonal literature, it is possible that voice quality is the main perceptual cue for distinguishing [+upper] and [-upper] Registers. If this is correct, the task for tonal perception will be much simpler than previously thought: a binary distinction for Register based on voice quality, and a binary distinction for Pitch based on pitch level. 3.3. Articulation

of tone

There have been many studies on the articulatoiy characteristics of tone, but the precise relation between the articulation of tone and tone features remains unclear. Zemlin (1981) surveyed previous works on articulatory pitch control and identified two major mechanisms: the cricothyroid muscles and the vocalis muscles. The former controls the elongation and the thickness of the vocal cords; the latter controls the "isometric tension" of the vocal cords. Duanmu (1990) speculated that these two mechanisms correspond to the two features of tone: the cricothyroid activity relates to Pitch, and the vocalis activity relates to Register, with a secondary effect on pitch (for a slightly different proposal, see Bao 1990). In this view, [+upper] is related to stiffer vocalis muscles, and [-upper] is related to slacker vocalis muscles. It is worth noting that [+upper] is related to a voiceless consonant and [-upper] to a voiced one, and that in the model of Halle & Stevens (1971) voiceless consonants have [+stiff vocal cords] and voiced consonants have [+slack vocal cords]. In other words, the effect of consonant voicing on Register can be seen as the spreading of the feature [stiff/slack vocal cords] from the consonant to the vowel (independent of the Pitch feature the vowel already has). Such proposals make specific claims about the connection between phonological features and phonetics and remain to be verified by experimental work.

260

San Duanmu

3.4. Contour tones Woo's (1969) proposal that contour tones are clusters of level tones has been widely adopted by African phonologists, but it has met considerable skepticism from Chinese phonologists (e.g. Yip 1989; Bao 1990; Chan 1991). Contour tones pose two problems for distinctive feature theory. First, if contour tones are basic units, they require trajectory features such as [rise] and [fall], or a modified version of it, as shown in (8), after Yip (1989), where TBU is the tone bearing unit. (8)

YIP'S ( 1 9 8 9 ) MODEL OF CONTOUR TONE UNITS

rise

TBU

L

H

fall

TBU

H

L

Duanmu (1994) argued that there is no compelling evidence for contour tone units. Instead, all contour tones are clusters of level tones. But even if this is true, there remains another problem. Since Leben (1971) and Williams (1976), it has been commonly assumed that two (or more) tones can be linked to a single short vowel, which creates a "short contour tone". Now if a short rise has the feature [-H, +H], then the vowel carrying it has two values of the same feature [H]. A segment that contains two (or more) values of the same feature is a "contour segment". If phonological theory admits contour segments, many possible segments are predicted, most of which are not found. Anderson (1976) noted this problem and suggested that perhaps contour features can be allowed for suprasegmental features but not for segmental features. However, the distinction between segmental and suprasegmental features is unclear. In addition, as will be discussed below, there are reasons to consider tone a segmental feature, too. Duanmu (1994) argued that in Chinese languages a contour tone always falls on a long syllable, as suggested by Woo (1969). Each rime segment, therefore, can take just one tone, either Η or L (ignoring Register). Duanmu (1994) also argued that, based on available evidence, "short contour tones" in African languages always occur on lengthened syllables. For example, the widely cited short contour tones in Mende (Leben 1973; Goldsmith 1976, and subsequent works) occur on lengthened vowels (Aginsky 1935; Ward 1944). The same is true in English. For example, in neutral intonation, "black' is HL (fall) and 'blackbird' is H-L (H on the first syllable and L on the second). It appears that the former has a short contour tone. But as Goldsmith (1981, p. 299) pointed out, and as is well known phonetically, [ae] is lengthened in 'black' but not in 'blackbird'. In other words, real cases of

261

Tone

short contour tones (contour tones on vowels that are not lengthened) are yet to be seen. This is not surprising: in a perception experiment by Greenberg & Zee (1979), it was found that a rising pitch contour will be heard as a level tone if the vowel is short (80 ms or less). There is, therefore, no compelling reason to posit contour tone features in distinctive feature theory (for arguments against other contour features, see references cited in Duanmu 1994).

3.5. The tone bearing unit Four kinds of entities have been proposed as the tone bearing unit. They are described in (9) and illustrated with the syllables [man] and [ma] in (10), where {} delimits a tone bearing unit. (9)

PROPOSALS OF TONE BEARING UNITS

a. b. c. d. (10)

The The The The

entire syllable (or the voiced part of it) rime portion of the syllable (but not the onset portion) mora (including the onset) moraic segment (the segment in the rime)

ILLUSTRATION OF DIFFERENT PROPOSALS OF TONE BEARING

UNIT

a. b. c. d.

man (man) {an} {ma}, {n} {a}, {n}

ma {ma} {a} {ma} {a}

(9a) was assumed by Chao (1930), Firth & Rogers (1937), and Wang (1967), among others. (9b) was proposed by Howie (1974). (9c) was proposed by Hyman (1985). (9d) was proposed by Woo (1969) and is consistent with the moraic theory of Hayes (1989), where the onset consonant is not dominated by a mora. There is no evidence that the onset can be a tone bearing unit by itself. Rather, the question is whether it can carry part of the tone when it is voiced, such as [m] in [man]. (9a) and (9c) assume that it does. (9b) and (9d) assume that it does not. The first assumption is often made but never argued for, as far as I am aware. In contrast, Howie (1974) has argued for the second assumption and shown that the pitch contour of a voiced onset (sonorant consonants or glides) is often irregular, and that the expected pitch contour does not start until the rime portion. If Howie is correct, the remaining choice is between (9b) and (9d). According to (9b), both [man] and [ma] have one tone bearing unit. According to (9d), [man] has two tone bearing units and [ma] has one.

262

San Duanmu

There is no question that (9d) is true in some languages, such as Luganda and Japanese. (9b) is often thought to be true in Chinese languages (e.g. Chao 1930; Firth & Rogers 1937; Wang 1967; Yip 1989; Chan 1991). It may appear that the choice between (9b) and (9d) is a matter of language variation. But as Duanmu (1994) argued, the tone bearing unit in Chinese is also (9d), as suggested by Woo (1969) (see also Shih 1997, who argued that the tone bearing unit in Standard Chinese is not the syllable but the moraic unit). If this is correct, there may be no cross-language variation in the tone bearing unit. Instead, the tone bearing unit is a moraic segment in all languages. 3.6. Tonogenesis and consonant-tone

interaction

Tonogenesis refers to the emergence of tone in a previously non-tone language based on (earlier) consonantal contrasts. The term is first introduced by MatisofF (1970), although the phenomenon has been studied in earlier works. Tonogenesis is commonly found in Asian languages. A wellknown example is Vietnamese (Haudricourt 1954; MatisofF 1973), in which the loss of coda consonants created an initial set of three tonal contrasts, and a subsequent loss of onset voicing split the contrasts into six. In other languages, such as Lhasa Tibetan (Hu 1980; Duanmu 1992), the loss of onset voicing created a two-way tonal contrast. Some researchers believe that Chinese also evolved from a non-tone language through a similar process (e.g. Baxter 1992, p. 7). Tonogenesis in Native American languages is somewhat controversial (see de Jong & McDonough 1993 for a recent discussion). I am not aware of studies on tonogenesis in African languages (but see depressor consonants below). The effect of coda consonants on tone is not well understood (see Maran 1973 for some discussion). Some linguists believe that coda consonants gave rise to four tones in Chinese (cf. Baxter 1992 and references therein). A final fricative (e.g. [s] or [h]) gave rise to a falling tone. A final glottal stop gave rise to a rising tone. A syllable with no obstruent coda gave rise to a level tone. A syllable that ended in [p], [t], or [k] gave rise to a fourth category, which is characterized by its shortness. However, since all cases of tonogenesis from coda consonants happened long ago, and since there is no language that is undergoing or has recently undergone such a process, its exact nature is yet to be determined. The effect of onset voicing on tone is better understood and can be generalized as "voiceless-high" and "voiced-low": a voiceless obstruent consonant leads to a higher tone and a voiced obstruent consonant leads to a lower tone. The featural link the between voicing and tone is briefly discussed in Section 3.3 and will be further illustrated below. Sonorant

263

Tone

onsets may behave either like voiceless obstruents or like voiced obstruents. In Underspecification Theory (cf. Steriade 1987; Archangeli 1988), this is expected, since sonorants can be underlyingly unspecified for voice. (11a) shows an example from Lhasa Tibetan, which acquired tone relatively recently (Hu 1980); the voiceless [k] is aspirated, which is not transcribed here, (lib) shows an example from Shanghai Chinese, which still retains onset voicing in addition to different tones (Xu, Tang, You, Qian, Shi & Shen 1988). (11)

TONOGENESIS IN TIBETAN AND SHANGHAI

a.

b.

Historical Tibetan [ko] -> lie' [go] -> •hear' Shanghai Chinese [se] high rise 'umbrella' [ze] low rise 'wealth'

Lhasa Tibetan [ko] high [ko]

low

In (11a), a historically voiceless onset gave rise to a syllable with a high tone, and a historically voiced onset gave rise to a syllable with a low tone. In Shanghai Chinese, there are two rising tones on "smooth" syllables (those without a glottalized rime). As seen in (lib), the high rise occurs with a voiceless onset and the low rise occurs with a voiced onset. Apparent exceptions to voiceless-high and voiced-low have been observed (e.g. Kingston & Solnit 1988). However, such exceptions often result from subsequent tone change. For example, the historical Chinese tone Ting" has split into Yin Ping and Yang Ping in Northern dialects (also called Mandarin dialects), all of which have lost onset voicing. Yin Ping occurs with historically voiceless onsets. In Beijing Mandarin, Yin Ping is a high level tone, but in Tianjin Mandarin, Yin Ping is a low level tone. It is reasonable to assume that Yin Ping in Beijing Mandarin is closer to the value right after Ping split than Yin Ping in Tianjin Mandarin. In other words, at the time of tone split, voiceless-high and voiced-low seems to be a valid correlation. The effect of onset voicing on tone is not purely a split between a higher pitch set and a lower pitch set. As far as facts are clear, the split is always accompanied by a difference in voice quality: clear in the upper set and murmured in the lower set (cf. sections 3.2 and 3.3). For this reason, voiceless-high and voiced-low can also be called Register-genesis or Register-split. The split of an existing tone by onset voicing can be viewed in various

264

San Duanmu

ways. A possible analysis of (lib) is shown in (12) (see Hombert, Ohala & Ewan 1979 for a different proposal and Maddieson 1984 for criticisms of Hombert, Ohala & Ewan 1979). (12)

FEATURAL ANALYSIS OF REGISTER-SPLIT IN SHANGHAI A.

B. LH

I I

SEE

LH

I

-4 S E E

C. LH

I

I

ZEE

I

LH

I

[+slack]

I

-»ZEE

Ψ

[+slack]

LH

I

I

I

SEE

V

[+slack]

A monosyllable is long in isolation (Duanmu 1994). Before split, the vowel carries LH (a rise). Following the theory of underspecification, let us assume that voicelessness and [+upper] are unspecified, as in (12a). In (12b), the voiced onset has the feature [+slack (vocal cords)] (after Halle & Stevens 1971). Next [+slack] spreads to the vowel, causing a split in tone: (12a) has [+upper] by default, and (12b) has [-upper] = [+slack], spread from the onset. The latter is characterized by a murmured voice quality, and accompanied by a lowered pitch level owing to the slackness of the vocal cords. If [+slack] is further delinked from the onset, which often happens in Shanghai, (12b) will become (12c), where the Register on the vowel solely carries the contrast that used to be carried by the voicing of the onset. The Tibetan case can be analyzed in a similar way. One possibility is shown in (13). As in Shanghai, monosyllables are likely to be long in Tibetan, but for illustration I have ignored vowel length. In addition, on monosyllables Η is usually realized as HL and L is usually realized as LH, which need not concern us here (see Duanmu 1992). (13)

a. b.

FEATURAL ANALYSIS OF TONOGENESIS IN TIBETAN

ko

Η I ko

ko ko go I I [+slack] [+slack] [+slack] The default Register is [+upper], which is found in (13a). The spread of [+slack] creates [-upper] Register in (13b). The Register difference between the two words further induces a Pitch difference: [+upper] induces Η and [-upper] induces L. The fact that certain features tend to co-occur is not

Tone

265

unusual and has been attributed to "enhancement" (Stevens & Keyser 1989). For example, voiceless consonants tend to be aspirated and voiced consonants tend to be unaspirated, and back vowels tend to be rounded and front vowels tend to be unrounded. In fact, as in English, [k] is aspirated in Tibetan (Hu 1980), apparently because aspiration and voicelessness enhance each other. It has been reported that in some Mon-Khmer languages, the loss of onset voicing has lead to a split in the voice quality of the vowel, without an obvious split in pitch levels (cf. Jenner, Thompson & Starosta 1976). Voiceless onsets has lead to clear voice (the so-called "first register") and voiced onsets have lead to murmured voice (the so-called "second register"). Such cases can be analyzed in (14), using [ko] and [go] as hypothetical examples. (14)

FEATURAL ANALYSIS OF REGISTER-SPLIT IN MON-KHMER

a. b.

ko go ko I I [+slack] [+slack]

Again, the default Register is [+upper], as in (14a). The spread of [+slack] creates [-upper] Register in (14b). The difference between Tibetan and Mon-Khmer is that in Tibetan the Register-split has further induced a Pitch difference, whereas in Mon-Khmer it has not. The effect of consonants on tone has also been observed in African languages, such as Zulu, Xhosa, Ewe, Nupe, Ngizim, Mijikenda, and Nguni (e.g. Hyman & Schuh 1974; Laughren 1984; Cassimjee & Kisseberth 1992). For example, a voiced obstruent, known as a "depressor" consonant, is found to lower the pitch level of a neighboring tone. The lowered tone can be seen to be in the [-upper] Register and analyzed in the same way as in (12), where [+slack] spreads from the consonant to the vowel.

3.7. Tone and stress Since McCawley's (1965) pitch-accent analysis of Japanese, many other languages have been analyzed in the same way. However, since "accent" lacks phonetic correlates, it cannot be checked directly with the native speaker, and as Pulleyblank (1983) pointed out, sometimes accent is used merely as an idiosyncratic diacritic marker. Pulleyblank further pointed out that in many cases, accent is simply a placeholder for a lexical tone, and if we assume lexically linked tones, there is no need to assume accent. For example, the accent locations in Japanese can simply be seen as locations to which Η is linked in the lexicon (Poser 1984). Similarly, Blevins (1993)

San Duanmu

266

argues that Lithuanian tone can be analyzed by assuming lexically linked Η instead of accent. Pulleyblank's proposal raises the question of whether stress (or accent) and tone can both occur in the same language. This issue has generated considerable discussion. There is little doubt that in languages like English, both stress and tonal patterns are determinable (at least by trained linguists). But perhaps English is not a "real" tone language. Indeed, I am not aware of any "real" tone language in which native speakers or trained linguists have a clear judgment for phonetic stress. For example, in Chinese languages, native speakers do not agree on stress locations on full syllables, nor do Chinese linguists agree (cf. Chao 1968, p. 38 for Mandarin and Selkirk & Shen 1990, p. 315 for Shanghai). Similarly, in African languages, Luganda has been at the center of the debate, yet there is no agreement on whether it has accent. Some linguists consider it to have tone only, some consider it to have accent (from which tone is predicted), and some consider it to require both tone and accent (Hyman & Katamba 1993, and references therein). There is, in fact, a good reason why stress is hard to feel in tone languages. The perceptual study of Fry (1958) shows that pitch is the most important cue for stress. Since pitch variation is not freely available in tone languages, it is natural that their speakers find it hard to tell stress. In metrical phonology, both stress and accent refer to the head of the prosodic unit "foot" (I ignore higher levels of stress). It is likely that prosodic structures are present in all languages (e.g. Selkirk 1981; McCarthy & Prince 1986). For example, Poser (1990) shows that despite the lack of phonetic stress, Japanese has foot structure. Similarly, Shih (1986) and Yip (1992,1994) have argued that, despite the lack of agreement on stress, Chinese languages also have foot structure. Nevertheless, as Pulleyblank pointed out, it is rarely shown whether the purported accent in tone languages is motivated on independent metrical grounds. In addition, if accent can be justified independently, one would like to know how it interacts with tone. In Chinese languages stress is determinable on independent metrical grounds (Ao 1992, 1993; Duanmu 1993, 1995). In addition, there is a specific relation between tone and stress, which I will call the Tone-Stress Principle and state in (15). (15)

TONE-STRESS PRINCIPLE

A stressed syllable is accompanied by an underlying tone pattern. An unstressed syllable is not accompanied by an underlying tone pattern.

267

Tone

A stressed syllable need not carry the entire tone pattern by itself. For example, in Goldsmith's analysis of English, a stressed syllable in neutral intonation is accompanied by (M)HL. When it is the only syllable, it carries HL, and when it is surrounded by unstressed syllables, it carries just H. According to (15), when a syllable loses stress, it will lose its underlying tone pattern. This agrees with the fact that in fast speech, where more syllables are destressed, fewer tone domains occur. I will illustrate (15) with some data from Shanghai Chinese, which has two tone patterns on monosyllables, LH and HL (ignoring Register). Polysyllabic domains also have two patterns, [L-H-L...L] if the initial syllable is underlyingly LH, and [H-L...L] if the initial syllable is underlyingly HL (I ignore a more restricted third pattern). Thus, [L-H-L...L] is an expansion of LH over the first two syllables, with additional syllables receiving a default L. Similarly, [H-L...L] is an expansion of HL. In normal careful speech, polysyllabic foreign words form disyllabic tone domains (with a trisyllabic final domain if there is an odd final syllable), indicated by parentheses in (16), where underlying tones are shown above surface tones (syllables are separated by a hyphen; [z] is syllabic in [sz]). The initial syllable of each domain is generally longer than the noninitial syllable(s), which is not indicated here (see Zhu 1995; Duanmu 1999). (16)

(Duanmu 1995) HL LH LH LH LH (H LXL Η L) ka- Ii- fo1- jii- ya 'California'

TONE DOMAIN FORMATION IN SHANGHAI CHINESE

LH LH HL LH LH LH (L Η)( Η L )(L H ) d3e'-kha'- sz- lu- va'-k h a' 'Czechoslovakia'

The underlying tone pattern of a syllable in a foreign name is that of the character used to represent the syllable. In hyper-articulated speech, where every syllable is stressed, every syllable can surface with its underlying tone pattern. (16) shows binary foot formation. In addition, it suggests that stress is initial in each foot. This is confirmed by a comparison between [1 2] and [2 1] compounds (numbers indicate the number of syllables in a word). In normal careful speech, a [1 2] compound forms one domain, as in (17a), but a [2 1] compound forms two, as in (17b). (17)

TONE DOMAINS IN [ 1 2 ] v s . [ 2 1 ] COMPOUNDS IN SHANGHAI

a.

(sä fe-ga) b. raw tomato "raw tomato'

(fe-ga)(thä) tomato soup 'tomato soup'

The difference between [1 2] and [2 1] is predicted if stress is left-headed, so that in [1 2] there is a stress clash, namely, two stresses occurring on adjacent syllables, as shown in (18a). In metrical phonology, stress clash

268

San Duanmu

should be avoided; this can be achieved by deleting stress from the second word, which gives rise to a trisyllabic foot. In contrast, there is no stress clash in [2 1], as shown in (18b), therefore both feet remain. LEFT-HEADED METRICAL STRUCTURES IN [ 1 2 ] v s . [2 1]

(18)

a.

χ χ

b.

(x)(x χ) Stress clash

χ

χ

(χ x)(x) No stress clash

In many other Asian tone languages, such as Standard Chinese, Cantonese Chinese, and Thai, consecutive full syllables do not form disyllabic domains. According to Duanmu (1993), the reason is that in those languages every full syllable is heavy and forms a bimoraic foot, so that it is able to retain its underlying tones. In contrast, in languages like Shanghai all syllables are underlyingly light (which can become heavy under stress, as happens when a syllable occurs in isolation), therefore a foot usually contains two (or more) syllables. In other words, the Tone-Stress Principle applies to both kinds of Asian tone languages. In English (e.g. Pierrehumbert 1980), Norwegian (e.g. Withgott & Halvorsen 1988), and Dutch (e.g. Gussenhoven 1988), a tone pattern accompanies each stressed syllable. This is similar to the Tone-Stress Principle. It remains to be seen whether the Tone-Stress Principle applies to African tone languages, once accent is motivated on independent metrical evidence (for some discussion, see Kenstowicz 1987; Hyman 1987,1989; Sietsema 1989; Kisseberth & Cassimjee 1992; Hyman & Katamba 1993). The Tone-Stress Principle is not the only source of tone assignment. An obvious exception is boundary tones, which occur at the edges of certain phonological units (Liberman 1975; Pierrehumbert 1980; Beckman & Pierrehumbert 1986). Another exception is the "associative tone" in some African languages, which is a floating tone that occurs between two nominals in a genitive relation (e.g. Odden 1980; Williamson 1986). It is unclear whether the associative tone is a boundary tone, and I will leave the issue open. 3.8. Typology of tone languages I mentioned in Section 1 that it is difficult to establish a typology of tone and non-tone languages, and that in some sense all languages are tonal. Beach (1924, p. 84) made a similar remark. However, there are obvious differences among languages like English, Japanese, and Chinese, which ought to be captured. Pike (1948, p. 3) offered a narrower definition of tone languages, according to which every syllable in a tone language can carry a contrastive tone.

269

Tone

By this definition, Japanese is not a tone language, because only the accented syllable carries a contrastive tone. Similarly, Swedish and Norwegian are not tone languages, because only the stressed syllable carries a contrastive tone (Pike 1948, p. 14). However, by the same definition, Chinese languages will not be tone languages, because, as discussed in Section 3.7, only stressed Chinese syllables are accompanied by a contrastive tone pattern. Consequently, differences among languages like English, Japanese, and Chinese are again left unaccounted for. In fact, if the ToneStress Principle is correct, no language would qualify as a tone language by Pike's definition. Goldsmith (1981) suggested that all languages have tone, but there is a distinction between "tone" and "accentual" languages in terms of tone linking. In a tone language, such as Mende, tones are linked to syllables (or tone bearing units) in a word from left to right. In contrast, in an accentual language, such as Japanese and English, a designated tone is linked to an accented syllable first (such as the Η in the English (M)HL pattern, shown in (3)), and other tones are linked afterwards. However, this proposal implies that there is no accent or stress in a tone language (if there is, the question will be why the accent does not attract a designated tone), which is incorrect. In addition, the difference between Japanese, whose word tones are fairly rigid, and English, whose word tones are more flexible, is still left unaccounted for. Extending Goldsmith's idea, I suggest that all languages have both stress (accent) and tone (see Section 3.7). In addition, each stressed (or accented) word is associated with a tone pattern. Languages differ in how many tone patterns they have and whether the tone patterns are lexically contrastive. (19) shows how the three traditional typological categories are interpreted in the present system. (19)

a. b. c.

INTERPRETATION OF TONE LANGUAGE TYPOLOGY

Tone language (e.g. Chinese) A stressed word can be linked to one of two or more tone patterns that are lexically contrastive. Pitch accent language (e.g. Japanese) Every stressed word is linked to the same tone pattern. Non-tone language (e.g. English) A stressed word can be linked to one of two or more tone patterns that are not lexically contrastive (depending on intonational meaning).

In Standard Chinese, a stressed syllable can take one of four contrastive tone patterns, as seen in (1). In Japanese, there is one tone pattern, (L)H(L), for all stressed words, as seen in (2). In English, a stressed word

270

San Duanmu

can take several tone patterns. Two examples are shown in (20), taken from Bolinger (1986, p. 24), where stress is indicated by * and where the contours Bolinger drew were converted into tone features. (20)

TWO PATTERNS OF TONE (INTONATION) IN ENGLISH

a. b.

L H-L She only *

Η wants

Η L-H She only

L wants *

#

L to

*

HL help. *

Η to

LH help. *

According to Bolinger, (20a) "is more like a reprimand or 'telling" rather than 'persuading"", and (20b) "might be used to soothe someone who has misunderstood the woman's motives". The first two stressed syllables are linked to a Η in (20a) and a L in (20b). The last stressed syllable is linked to HL in (20a) and LH in (20b). According to Pierrehumbert (1980), English has seven tone patterns for a word with a stress (boundary tones aside). However, whatever its tone pattern, the lexical meaning of an English word remains unchanged. The categories in (18) may not apply in all cases. For example, Chonnam Korean has two word tone patterns, HHL and LHL, predictable from the underlying voicing of the initial segment (Jun 1990). If voicing is taken to be contrastive in Chonnam Korean, then its tone patterns are not, so Chonnam Korean is not (18a). And since there are two tone patterns, Chonnam Korean is not (18b) either. Finally, the choice of the tone patterns does not depend on intonational meaning, so Chonnam Korean is not (18c). Languages like Lithuanian present another kind of problem. In Lithuanian, a long syllable can carry one of two contrastive tones, a rise or a fall, shown in (21). (21)

TONE IN LITHUANIAN HL-L

[viiras] 'man'

LH-L

[viinas] 'wine'

The data can be analyzed in three ways, shown in (22), where W represents a long vowel. (22)

a.

T H R E E ANALYSES OF LITHUANIAN TONE

Stress analysis W Η

W Η

Tone

271 b.

Tone analysis W

c.

W

Η Η Stress and tone analysis *

W

HL

*

W

LH

In (22a), suggested by Halle & Vergnaud (1987), an accent is assigned to either the first or the second slot of a long vowel. Then, Η is assigned to the accented slot and L is assigned to other vowel slots. However, (22a) assumes that accent can occur on the second mora of a heavy syllable, which is metrically controversial (Kager 1993; Hayes 1995). In (22b), suggested by Blevins (1993), Η is lexically linked to either the first or the second slot of a long vowel. A default L can then be assigned to other vowel slots. As Blevins pointed out, (22b) avoids the use of accent and so it appears to be more economical. On the other hand, (22b) implies that Lithuanian has no metrical system, which is doubtful. Besides, (22b) does not simplify the theory overall, since in languages like Chinese and English, both stress and tone must be posited anyway (see Section 3.7). (22c) assumes that Lithuanian has accent and two lexical tone patterns, LH and HL. Which analysis is correct depends on (a) whether Lithuanian has accent and (b) whether accent can fall on the second mora of a heavy syllable. If (22a) is correct, Lithuanian is a pitch-accent language. If (22c) is correct, Lithuanian is a tone language. If (22b) is correct, Lithuanian belongs to none of the categories in (19). Whether all languages can be subsumed under a single toneaccent theory is an interesting question that invites further discussion.

3.9. Tone domains The term "tone domain" has been used to refer to two different things. First, it refers to a stretch of sounds over which an expected contour occurs. For example, in Shanghai Chinese the pattern [HL] can occur on one or more syllables, as in (23). A monosyllabic expression is clearly lengthened, which is shown in (23). The initial syllable of a polysyllabic domain is also longer than noninitial syllables (cf. Zhu 1995), which is not shown in (23) (see Duanmu 1999 for a more detailed account of syllable length in Shanghai). (23)

T O N E DOMAIN IN SHANGHAI

HL see 'three'

Η L se pe 'three cups'

L Η L se du pe 'three big cups'

San Duanmu

272

Domains like those in (23) have one stress, therefore they are a stress domain, or a foot. A second meaning of tone domain refers to a stretch of sounds that are affected by a tone rule. For example, in Xiamen (Chen 1 9 8 7 ) , every full syllable has two lexical tone patterns, one used in the final position (A) and the other in nonfinal positions (B). An example is given in (24), following Chen's transcription, where the pitch values of A and Β tones need not concern us. (24)

T O N E DOMAIN IN XIAMEN AL

A2

pang hongfly kite 'to fly a kite'

A3

(Chen

1987)

B1 B2 A3

ts'e

A domain like (24) has been called a tone domain because every nonfinal syllable takes its Β tones. But a domain like (24) is not a single stress domain. In (24) all the three syllables are stressed and bear their own lexical tones, whether it is the A or the Β copy (the syllable ts'e has a long vowel, which is not shown in the transcription). Both A and Β copies are lexically specified and totally unrelated to what tones other syllables have. Both kinds of domains interact with morphology and syntax in intricate ways. For examples from Chinese languages, see Shih ( 1 9 8 6 , 1 9 9 7 ) , Chen ( 1 9 8 7 , 1 9 9 6 ) , Selkirk & Shen ( 1 9 9 0 ) , Zhang ( 1 9 9 2 ) , and Duanmu ( 1 9 9 5 ) . For examples from African languages, see Clements ( 1 9 7 8 ) and Odden ( 1 9 8 7 ) . 3.10. Is tone a prosodic

feature?

Tone has traditionally been considered a prosodic feature, which belongs to a unit larger than a segment (e.g. Chao 1930; Pike 1948; Firth 1957; and Wang 1 9 6 7 ) . There are several reasons for this view. First, a tone pattern usually remains constant independent of its carrier (see (3) and ( 2 3 ) ) . Second, tone can survive vowel deletion and relocate to another vowel. Third, tone is quite free to move or spread from one syllable to another; in contrast, segmental features are usually not so mobile. On the other hand, there are reasons to consider tone a segmental feature. First, although a tone pattern can extend over several syllables, each moraic segment carries just one tone, Η or L. In this regard, a tone feature ([+H] or [-H]) is like a segmental feature, which ultimately resides in a segment and which cannot occur twice within a segment (see sections 3.4 and 3.5). Second, a suprasegmental feature is usually thought to be different from a segmental one. However, in many Asian languages, tones came from consonant features. For example, in (12) [+slack] is a voicing feature on the consonant and Register feature on the vowel. If we are to

Tone

273

treat a feature consistently, we ought to consider [+slack] to be a segmental feature in both cases. Third, tone is not unique in surviving segment deletion. Features like [+nasal], [-back], and [+round] are also able to survive segment deletion. Finally, let us consider the mobility of tone. First, it will be noted that tone is not unique in moving or spreading. Vowel features [back], [high], [low], and [round], and consonant features like [nasal] and [retroflex], have all been found to spread in harmony processes. In the current view of feature theory (e.g. Sagey 1986; Halle 1995), every feature lies on an independent tier. A feature F can move or spread when there is nothing in its way, that is, when there is no specification of F in the intervening segments. Tone is more mobile only because unstressed syllables are not specified for tone (cf. the Tone-Stress Principle in (15)), as a result of which tone can spread onto and through them. It is worth noting that in tone spreading it is usually the Pitch feature that spreads, and not the Register feature. This follows if Register is the same as voicing (sections 3.1, 3.2, 3.3, and 3.6), which is usually specified in consonants. Since most syllables have a consonant, Register rarely spreads out of a syllable. We have seen then that tone features are not fundamentally different from features like [nasal], [round], and [back], which are usually considered segmental features. Of course, one can define suprasegmental features by their behavior in a given context. For example, one can define a suprasegmental feature as one that spreads across syllables. Thus, [nasal] is a suprasegmental feature when it spreads, but a segmental feature when it does not. However, by this definition, tones are also segmental features when they do not spread, and since in most Asian languages tones do not spread, one would conclude that Asian tones are mostly segmental. But the usefulness of this definition is not obvious: as discussed in Section 3.7 and 3.9, whether a tone pattern spreads or not depends on stress. When there are many stressed syllables, tones appear to be inactive, and when there are fewer stressed syllables, tones appear to be mobile. A definition of suprasegmental features based on their mobility would therefore be quite superfluous. In the phonemic tradition, tone has sometimes been called a phoneme. This is primarily because lexical tones are contrastive and they did not appear to be part of the segments. In the present view, tones reside on segments and tone features are essentially similar to other segmental features. There is, therefore, no motivation to call tones phonemes, just as there is no motivation to call [nasal] or [round] phonemes.

274

San Duanmu 3.11. Categorical

and gradient

values

There has been some debate about the categorical vs. gradient nature of linguistic features and tone bears on this issue. In real speech, pitch levels are infinitely variable. However, no language uses more than a small number of contrastive tone levels. In particular, as discussed in Section 3.1, there may be only four contrastive tone levels, and if Register is cued by voice quality (sections 3.2 and 3.3), there may be just two levels that are cued by pitch, Η and L. In any case, contrastive tone levels are strikingly small in number, especially in view of the fact that in singing any ordinary person can use at least a dozen or so music notes at ease. The paucity of contrastive tone levels supports the view that linguistic features are categorical in nature. 3.12. Tone and pitch

values

In real speech the same tone feature can occur at different pitch levels. This is due to several factors. First, Η has a higher pitch with greater stress than it does with lesser stress. Second, the same tone will gradually descend in pitch owing to the downdrift effect. Downdrift was first observed in African languages but was later found in other languages, too, such as English, Japanese, Korean, Chinese, and German (Pierrehumbert 1980; Beckman & Pierrehumbert 1986; Jun 1990; He & Jin 1992; F£ry 1993). Third, segmental features, such as voicing and aspiration, can also affect pitch contour. To interpret the pitch contours of a language, therefore, it is necessary not only to know its tone patterns but also its prosodic structure, including various levels of stress that are sensitive to phrasal structure and pragmatic focus (e.g. Gussenhoven 1984; Bolinger 1985; McLemore 1991; Selkirk 1995, among others). This is a big challenge, not only because phrasal stress and pragmatic focus are less understood than word stress, but also because stress in tone languages is not as obvious as stress in non-tone languages. 3.13. Tone and intonation

in the same

language

According to Liberman (1975) and Pierrehumbert (1980), among others, intonation patterns in English can be represented by a linear sequence of Hs and Ls. The sequence is made by a proper choice of tone patterns for each accentual phrase plus a proper choice of boundary tones and "phrase accents". However, in languages like Chinese, word tones are lexically determined, and there is little flexibility in varying the sequence of Hs and Ls independently (I ignore tonal Register here). This raises the question of

Tone

275

how languages like Chinese express intonational meaning, such as statement, doubt, surprise, query, command, etc. Chao (1933) suggests that many functions of intonation in other languages are fulfilled in Chinese by the use of particles. When particles are not used, tone and intonation can be combined through "addition". There are two cases. The first is "successive addition", where a rise or a fall is added to the end of an utterance; this is similar to a boundary Η or L in English. Two examples in Standard Chinese are shown in (25) (cf. Chao 1933, p. 131). Strictly speaking, the resulting syllable must be lengthened to three moras in order to carry the three tones (cf. Section 3.5). For simplicity the lengthening is not shown. (25) CHINESE

SUCCESSIVE ADDITION OF TONE AND INTONATION IN STANDARD

Tone LH + nan 'difficult' Tone HL + mai 'sell'

Intonation L ->

LHL nan 'affirmation' 'Surely difficult!' Intonation Η -> HLH mai 'question' 'Sell?'

[nan] 'difficult' has a rising tone (LH). The boundary intonation L carries an affirmation meaning. When the two are added, the result is a rise-fall (LHL). Similarly, [mai] 'sell' has a falling tone (HL). The boundary intonation Η carries a question meaning. When the two are added, the result is a fall-rise (HLH). The second case is "simultaneous addition", where intonation is superimposed on word tones, as a result of which the pitch range of an utterance is raised, lowered, expanded, or compressed, while the word tones remains distinctive. Simultaneous addition is supported by the phonetic studies of Shen (1989) and He & Jin (1992), who found that the question intonation raises the pitch height of the entire utterance, without changing the distinctiveness of word tones. If simultaneous addition occurs, tone and intonation must be different entities. This raises a serious problem. First, to represent tone and intonation in Chinese, a single sequence of Hs and Ls (ignoring Register) is not enough. Second, there is the potential question of whether English pitch contours reflect tone, or intonation, or both. If they reflect tone, one wonders why English has no intonation. If they reflect intonation, it means that, like tone, intonation can be represented by a linear sequence of Hs and Ls, and then there is the question of what relation there is between the

276

San Duanmu

intonational Hs and Ls and the tonal Hs and Ls. If they reflect both, and if Liberman (1975) and Pierrehumbert (1980) are essentially correct, then one wonders why a single sequence of Hs and Ls can represent both tone and intonation in English but just tone in Chinese. The solution to the problem is not obvious. But let us consider two proposals. First, it is possible that languages can use one of two ways to indicate an intonation: either by a boundary tone, or by changing the pitch level of an entire utterance. In the first way, intonation and word tones are in a linear sequence (Chao's "successive addition"). English uses this method throughout, but Chinese uses it only sometimes. In the second way, intonation and tone are superimposed on each other (Chao's "simultaneous addition"). English does not use this method, but Chinese sometimes does. As a result, English tone and intonation can be represented by a single sequence of Hs and Ls, whereas Chinese tone and intonation can be represented this way only sometimes. This proposal essentially admits simultaneous addition at least in some languages and calls for a representation that goes beyond a single sequence of Hs and Ls. For example, one may assume two levels of representation, a lower level of Hs and Ls for word tones, and a higher level of Hs, Ls, rises, falls, etc., for phrase and sentence intonation. The second proposal attempts to preserve a single sequence of tones, and account for their undulations in other ways. A motivation for this proposal is that, if [+H] and [-H] are values of a distinctive feature, there ought to be just one sequence of them, since there is only one sequence of values of any other distinctive feature. It is worth noting that Chao (1933) mentioned only two cases of simultaneous addition in Chinese, "raised level of pitch" and "lowered level of pitch" (pitch range widening and narrowing were found to be correlated to pitch raising and lowering). In He & Jin (1992), six intonational meanings were designed (statement, expecting confirmation, question, simple request, command, and exclamation), but again just two cases of simultaneous addition were observed: raised or lowered. If these are the only cases of simultaneous addition (besides the neutral intonation), there is a possible solution. The raised level may have an accented final boundary H, which (whether it is actually produced or not) is targeted at a higher pitch level (as any Η with a main accent does) and so may have prevented downstep in the preceding syllables. Similarly, the lowered level may have an accented final boundary L, which is targeted at a lower pitch level and so may have accelerated downstep in the preceding syllables. At this point, this proposal is rather speculative. Whether it is a correct account of simultaneous addition will be left open.

Tone

277

3.14. Functions of tone A basic function of tone is lexical distinctiveness. However, tone can serve other functions, such as syntactic distinction and attitudinal meaning. For example, in Igbo, a floating Η can indicate the associative structure "noun of noun" (Williamsom 1986). In Chinese, a boundary Η can indicate a question, and a boundary L can indicate an affirmation or a "protesting statement" (Chao 1933, pp. 131-133). In English, attitudinal meanings can also be expressed by word tones (pitch-accents) and boundary tones (e.g. Liberman 1975; Pierrehumbert 1980). The fact that tone has multiple functions is again not unique among distinctive features. For example, palatalization ([-back, +high]) is lexically distinctive but can also indicate the mimetic structure in Japanese (Mester & Ito 1989).

4.

Conclusions

Research on tone in the past three decades has tremendously increased our understanding of the subject. It also triggered developments in many other areas of phonology. For example, the rise of autosegmental phonology, resulting from tone research, has brought about progress in such areas as harmony processes, templatic morphology, syllable structure, feature geometry, and metrical phonology. Works on tone domains have also propelled research on the syntax-phonology interface. For example, Clements's (1978) work on Ewe and Chen's (1987, written in 1985) work on Xiamen have lead to the influential end-based theory of Selkirk (1986) on the syntax to phonology mapping. Works on tonogenesis and tone features call for a reconsideration of the traditional distinction between segmental and suprasegmental features. Finally, works on contour tones have direct implications for the status of contour features in distinctive feature theory. As seen in Section 3, many issues remain unresolved and will continue to concern researchers for some time. In addition, although I have not discussed the large body of phonetic literature on tone, it is clear that tone is a principal area in which phonetic and phonological researches come together.

Acknowledgments Thanks to Pam Beddor, Lesley Milroy, James Myers, Bonny Sands, and Rint Sybesma for their comments.

278

San Duanmu

A Tone Bibliography Aginsky, Ε. (1935). A grammar of the Mende language. [Language Dissertations 20.] Philadelphia: Linguistic Society of America. Anderson, S.R. (1976). Nasal consonants and the internal structure of segments. Language 52, 326-344. Anderson, S.R. (1978). Tone features. Tone: a linguistic survey, edited by Fromkin, 133-175. Ao, B.X.P. (1992). Metrical constituents as domains of tone sandhi in Nantong Chinese. Paper presented at CLS 28, Chicago. Ao, B.X.P. (1993). Phonetics and phonology of Nantong Chinese. Doctoral dissertation, Ohio State University, Columbus. Archangeli, D. (1988). Aspects of underspecification theory. Phonology 5, 183-207. Bao, Z.M. (1990). On the nature of Tone. Doctoral dissertation, MIT. Baxter, W. H. (1992). A handbook of Old Chinese phonology. Berlin: Mouton de Gruyter. Beach, D.M. (1924). The science of tonetics and its application to Bantu languages. Bantu Studies, 2nd series, vol. 2, 75-106. Beckman, M. (1986). Stress and non-stress accent. Dordrecht: Foris Publications. Beckman, M. & J. Pierrehumbert (1986). Intonation structure in Japanese and English. Phonology Yearbook 3, 255-309. Bing, J. (1979). Aspects of English prosody. Doctoral dissertation, University of Massachusetts Amherst. [Distributed by the Indiana University Linguistics Club, Bloomington, Indiana.] Blevins, J. (1993). A tonal analysis of Lithuanian nominal accent. Language 69, 237-273. Bloch, Β. (1946). Studies in colloquial Japanese II: Syntax. Language 22, 200-248. Bolinger, D. (1985). Two views of accent. Journal of Linguistics 21,79-123. Bolinger, D. (1986). Intonation and its parts. Stanford, CA: Stanford University Press. Bruce, G. & E. Garding (1978). A prosodic typology for Swedish dialects. Nordic prosody: Papers from a symposium, edited by E. Garding, G. Bruce & R. Bannert, 219-228. Lund University. Cao, J.F. & I. Maddieson (1992). An exploration of phonation types in Wu Dialects of Chinese. Journal of Phonetics 20, 77-92. Cassimjee, F. (1987). An autosegmental analysis ofVenda tonology. Doctoral dissertation, University of Illinios, Urbana-Champaign.

Tone

279

Cassimjee, F. & C. Kisseberth (1992). The tonology of depressor consonants: Evidence from Mijikenda and Nguni. Proceedings of the Eighteenth Annual Meeting ofthe Berkeley Linguistics Society February 14-18,1992: Special Session on the Typology of Tone Languages, edited by L.A. Buszard-Welcher, J. Evans, D. Peterson, L. Wee & W. Weigel, 26-40. Chan, Μ. Κ. M. (1985). Fuzhou phonology: A non-linear analysis of tone and stress. Doctoral dissertation, University of Washington, Seattle. Chan, Μ. Κ. M. (1991). Contour-tone spreading and tone sandhi in Danyang Chinese. Phonology 8, 237-259. Chang, K. (1953). On the tone system of the Miao-Yao languages. Language 29, 374-378. Chao, Y.-R. (1930). A system of tone letters. Le maitre phonetique 45, 24-27. Chao, Y.-R. (1933). Tone and intonation in Chinese. Bulletin of Institute of History and Philology, Academia Sinica 4.2,121-134. Chao, Y.-R. (1968). A grammar of spoken Chinese. Berkeley: University of California Press. Chao, Y.-R. (1980). Chinese tones and English stress. The melody of language, edited by L.R. Waugh & C.H. van Schooneveld, 41—44. Baltimore: University Park Press. Chen, M. (1986). The paradox of Tianjin tone sandhi. Proceedings of the Chicago Linguistic Society 22, 98-114. Chen, M. (1987). The syntax of Xiamen tone sandhi. Phonology Yearbook 4, 109-149. Chen, M. (1991). An overview of tone sandhi phenomena across Chinese dialects. Languages and dialects of China, edited by W. S.-Y. Wang, 113-158. [Journal of Chinese Linguistics monograph series 3.] Chen, M. (1996). Tone sandhi. Manuscript, University of California San Diego. Cheng, C.-C. (1968). English stress and Chinese tones in Chinese sentences. Phonetica 18, 77-88. Chomsky, Ν. & M. Halle (1968). The sound pattern of English. New York: Harper and Row. Clark, M. (1990). The tonal system oflgbo. Dordrecht: Foris Publications. Clements, G.N. (1978). Tone and syntax in Ewe. Elements of tone, stress, and intonation, edited by D.J. Napoli, 21-99. Washington, DC: Georgetown University Press. Clements, G.N. (1983). The hierarchical representation of tone features. Current approaches to African linguistics vol. 1, edited by I. R. Dihoff, 145-176. Dordrecht: Foris Publications.

280

San Duanmu

Clements, G.N. & K. Ford (1979). Kikuyu tone shift and its synchronic consequences. Linguistic Inquiry 10,179-210. Clements, G.N. & J. Goldsmith, editors (1984). Autosegmental studies in Bantu tone. Dordrecht: Foris Publications. de Jong, K. & J. McDonough (1993). Tone and tonogenesis in Navajo. UCLA Working Papers in Phonetics 84,165-182. Duanmu, S. (1990). A formal study of syllable, tone, stress and domain in Chinese languages. Doctoral dissertation, MIT. Duanmu, S. (1992). An autosegmental analysis of tone in four Tibetan languages. Linguistics of the Tibeto-Burman Area 15, 65-91. Duanmu, S. (1993). Rime length, stress, and association domains. Journal of East Asian Linguistics 2,1-44. Duanmu, S. (1994). Against contour tone units. Linguistic Inquiry 25, 555-608. Duanmu, S. (1995). Metrical and tonal phonology of compounds in two Chinese dialects. Language 71, 225-259. Duanmu, S. (1999). Metrical Structure and Tone: Evidence from Mandarin and Shanghai. Journal of East Asian Linguistics 8,1-38. Fery, C. (1989). Prosodic and tonal structure of Standard German. Doctoral dissertation, University of Konstanz. Fery, C. (1993). German intonational patterns. [Linguistische Arbeiten 285.] Tübingen: Max Niemeyer Verlag. Firth, J. R. (1957). Papers in linguistics 1934-1957. London: Oxford University Press. Firth, J.R. & B.B. Rogers (1937). The structure of the Chinese monosyllable in a Hunanese dialect (Changsha). Bulletin of the School of Oriental Studies viii.4. Reprinted in Papers in linguistics 1934-1957 by Firth, 76-91. Fretheim, T. (1981). Nordic prosody II. Trondheim, Norway: Tapir. Fromkin, V.A. (1972). Tone features and tone rules. Studies in African Linguistics 3, 47-76. Fromkin, V. Α., editor (1978). Tone: A linguistic survey. New York: Academic Press. Fry, D.B. (1958). Experiments in the perception of stress. Language and Speech 1,126-152. Gandour, J. (1974). On the representation of tone in Siamese. UCLA Working Papers in Phonetics 27,118-146. Garding, E. (1979). Sentence intonation in Swedish. Phonetica 36, 207-15. Goldsmith, J. (1976). Autosegmental phonology. Doctoral dissertation, MIT. [Reproduced by the Indiana University Linguistics Club, Bloomington, Indiana, 1981.]

Tone

281

Goldsmith, J. (1981). English as a tone language. Phonology in the 1980's, edited by D. L. Goyvaerts, 287-308. Ghent, Belgium: E. Story-Scientia. Goldsmith, J. (1982). Accent systems. The structure of phonological representations (Part I), edited by H. van der Hulst & N. Smith, 47-63. [Linguistic Models 2.] Dordrecht: Foris Publications. Greenberg, S. & E. Zee (1979). On the perception of contour tones. UCLA Working Papers in Phonetics 45, 150-64. Gussenhoven, C. (1984). On the grammar and semantics ofsentence accents. Dordrecht: Foris Publications. Gussenhoven, C. (1988). Adequacy in intonation analysis: the case of Dutch. Autosegmental studies on pitch accent, edited by van der Hulst & Smith, 95-122. Halle, M. (1995). Feature geometry and feature spreading. Linguistic Inquiry 26, 1-46. Halle, Μ. & K. Stevens (1971). A note on laryngeal features. RLE Quarterly Progress Report 101,198-213, MIT. Halle, M. & J.-R. Vergnaud (1980). Three dimensional phonology. Journal of Linguistic Research 1, 83-105. Halle, M. & J.-R. Vergnaud (1987). A essay on stress. Cambridge, Mass.: MIT Press. Haraguchi, S. (1977). The tone pattern of Japanese: An autosegmental theory of tonology. Tokyo: Kaitakusha. 't Hart, J. & R. Collier (1975). Integrating different levels of intonation analysis. Journal of Phonetics 3, 235-55. Haudricourt, A.G. (1954). De l'origine des tons en vietnamien. Journal Asiatique 242, 68-82. Hayes, B. (1989). Compensatory lengthening in moraic phonology. Linguistic Inquiry 20, 253-306. Hayes, B. (1995). Metrical stress theory: Principles and case studies. Chicago: University of Chicago Press. He, Y. & S. Jin (1992). Beijinghua yudiao de shiyan tansuo [Intonations of the Beijing Dialect: An experimental exploration]. Yuyan Jiaoxue Yu Yanjiu 1992.2, 71-96. Herbert, R.K. (1975). Reanalyzing prenasalized consonants. Studies in African Linguistics 6.2,105-123. Hinton, L. (1991). An accentual analysis of tone in Chalcatongo Mixtec. Papers from the American Indian Languages Conferences ..., edited by J.E. Redden, 173-82. Hombert, J.-M. (1978). Consonant types, vowel quality, and tone. Tone: A linguistic survey, edited by Fromkin, 77-111. Hombert, J.-M., J. Ohala & W.G. Ewan (1979). Phonetic explanation for the development of tones. Language 55, 37-58.

282

San

Duanmu

Howie, J. (1974). On the domain of tone in Mandarin: Some acoustical evidence. Phonetica 30,129-148. Hu, T. (1980). Zangyu (Lasa hua) shengdiao yanju [A study of tone in Lhasa Tibetan]. Minzu Yuwen 1, 22- 36. Hu, Τ., A.T. Qu & L.H. Lin (1982). Zangyu (Lasa hua) shengdiao shiyan [Experiments on tone in Lhasa Tibetan], Yuyan Yanjiu 1982.1,18-38. Huang, J. C.-T. (1985). The autosegmental and metrical nature of tone terracing. African linguistics: Essays in memory ofM. W.K. Semikenke, edited by D. Goyvaerts, 209-238. Amsterdam: Benjamins. van der Hulst, Η. & N. Smith, editors (1988). Autosegmental studies on pitch accent. [Linguistic Models 11.] Dordrecht: Foris Publications. van der Hulst, Η. & K. Snider, editors (1993). The phonology of tone: The representation of tonal register. [Linguistic Models 17.] Berlin: Mouton de Gruyter. Hyman, L. (1978). Tone and/or accent. Elements of tone, stress, and intonation, edited by D. J. Napoli, 1-20. Washington, DC: Georgetown University Press. Hyman, L. (1981). An autosegmental accentual account of Luganda tone. Studies in African Linguistics, supp. 8 , 64-68. Hyman, L. (1985). A theory of phonological weight. Dordrecht: Foris Publications. Hyman, L. (1986). The representation of multiple tone heights. The phonological representation of suprasegmentale: Studies on African languages offered to John M. Steward on his 60th birthday, edited by K. Bogers, H. van der Hulst & M. Mous, 109-152. Dordrecht: Foris Publications. Hyman, L. (1987). Prosodic domains in Kukuya. Natural Language & Linguistic Theory 5, 311-334. Hyman, L. (1989). Accent in Bantu: An appraisal. Studies in the Linguistic Sciences 19, 115-134. Hyman, L. & F. Katamba (1993). A new approach to tone in Luganda. Language 69, 34-67. Hyman, L. & R. Schuh (1974). Universals of tone rules: Evidence from West Africa. Linguistic Inquiry 5, 81-115. Inkelas, S. & D. Zee (1988). Serbo-Croatian pitch accent: The interaction of tone, stress, and intonation. Language 64, 227-248. International Phonetic Assciation (1989). Report on the 1989 Kiel Convention. Journal of the International Phonetic Association 19, 67-80. Jenner, P. Ν., L. C. Thompson & S. Starosta, editors (1976). Austroasiatic studies Part I, University of Hawaii Press. Jun, S.A. (1990). The accentual pattern and prosody of the Chonnam dialect of Korean. Ohio State University Working Papers in Linguistics 38,121-140.

Tone

283

Jun, S.A. (1993). The phonetics and phonology of Korean prosody. Doctoral dissertation, Ohio State University, Columbus. Kager, R. (1993). Alternatives to the Iambic-Trochee Law. Natural Language & Linguistic Theory 11, 381-432. Kenstowicz, M. (1987). Tone and accent in Kizigua—a Bantu language. Certamen Phonologicum, edited by P.-M. Bertinetto & M. Loporcaro, 177-188. Torino: Rosenberg & Sellier. Kenstowicz, M. & C. Kisseberth (1990). Chizigula tonology: The word and beyond. The phonology-syntax connection, edited by S. Inkelas & D. Zee, 163-194. Chicago: University of Chicago Press. Kim, M. (1996). The tonal system of accentual languages. Doctoral dissertation, University of Chicago. Kingston, J. & D. Solnit (1988). The tones of consonants. Manuscript, University of Massachusetts Amherst and University of Michigan, Ann Arber. Kisseberth, C. & F. Cassimjee (1992). Tone and metrical structure in Shinjazidja. Paper presented at CLS 28, Chicago. Ladd, D.R. (1980). The Structure of intonational meaning: Evidence from English. Bloomington: Indiana University Press. Laughren, M. (1984). Tone in Zulu nouns. Autosegmental studies in Bantu tone, edited by Clements & Goldsmith, 183-234. Leben, W. (1971). Suprasegmental and Segmental Representation of Tone. Studies in African Linguistics, supp. 2, 183-200. Leben, W. (1973). Suprasegmental phonology. Doctoral dissertation, MIT. Liberman, M. (1975). The intonational system of English. Doctoral dissertation, MIT. Lin, M. C., J.Z. Yan & G.H. Sun (1984). Beijinghua lianzizu zhengchang zhongyin de chubu shiyan [Preliminaiy experiments on the normal stress in Beijing disyllables]. Fangyan 1984.1, 57-73. Lin, T. (1985). Tantao Beijing hua qingsheng xingzhi de chubu shiyan [Preliminary experiments in the exploration of the nature of Mandarin neutral tone]. Beijingyuyin shiyan lu [Peking University working papers in phonetics], edited by T. Lin & L.J. Wang, 1-26. Beijing: Peking University Press. McCarthy, J. & A. Prince (1986). Prosodic morphology. Manuscript, University of Massachusetts, Amherst and Brandeis University. McCawley, J. (1965). The accentual system of modern standard Japanese. Doctoral dissertation, MIT. McCawley, J. (1978). What is a tone Language? Tone: A linguistic survey, edited by Fromkin, 113-131. McLemore, C. A. (1991). The pragmatic interpretation ofEnglish intonation: Sorority speech. Doctoral dissertation, University of Texas, Austin.

284

San Duanmu

Maddieson, I. (1978). Universals of tone. Uni versals of Human Language vol. 2: Phonology, edited by J. Η. Greenberg, 335-365. Stanford, CA: Stanford University Press. Maddieson, I. (1984). The effects on FO of a voicing distinction in sonorants and their implications for a theory of tonogenesis. Journal of Phonetics 12, 9-15. Maran, L. R. (1973). On becoming a tone language: A Tibeto-Burman model of tonogenesis. Consonant types and tone, edited by L. Hyman, 97—114. [Southern California Occasional Papers in Linguistics 1.] University of Southern California, Los Angeles. Matisoff, J. A. (1970). Glottal dissimilation and the Lahu high- rising tone: A tonogenetic case-study. Journal of the American Oriental Society 90, 13—44. Matisoff, J. A. (1973). Tonogenesis in Southeast Asia. Consonant types and tone, edited by L. Hyman, 71-95. [Southern California Occasional Papers in Linguistics 1.] University of Southern California, Los Angeles. Mester, R. A. & J. Ito (1989). Feature predictability and underspecification: Palatal prosody in Japanese minetics. Language 65, 258-293. Myers, S. (1987). Tone and the structure of words in Shona. Doctoral dissertation, University of Massachusetts, Amherst. Newman, P. (1986). Contour tones as phonemic primes in Grebo. Phonological representations of suprasegmentale, edited by K. Bogers, H. van der Hulst & M. Mous, 175-193. Dordrecht: Foris Publications. Odden, D. (1980). Associative tone in Shona. Journal ofLinguistic Research 1.2, 37-51. Odden, D. (1981). Problems in tone assignment in Shona. Doctoral dissertation, University of Illinois, Urbana-Champaign. Odden, D. (1987). Kimatumbi phrasal phonology. Phonology 4,13-36. Odden, D. (1995). Tone: African languages. The handbook of phonological theory, edited by J. Goldsmith, 444-475. Cambridge, Mass.: Blackwell. Pierrehumbert, J. (1980). The phonetics and phonology of English intonation. Doctoral dissertation, MIT. Pike, K. (1948). Tone languages. Ann Arbor: University of Michigan Press. Poser, W. (1984). The phonetics and phonology of tone and intonation in Japanese. Doctoral dissertation, MIT. Poser, W. (1990). Evidence for foot structure in Japanese. Language 66, 78-105. Pulleyblank, D. (1983). Tone in lexical phonology. Doctoral dissertation, MIT. Pulleyblank, D. (1986). Tone in lexical phonology. Dordrecht: Reidel. Qu, A. T. (1981). Zangyu de shengdiao ji qi fazhan [The Tibetan tone and its development]. Yuyan Yanjiu 7,177-194.

Tone

285

Redden, J. Ε., editor (1991). Papers from the American Indian Languages Conferences Held, at the University of California, Santa Cruz, July and August 1991. [Occasional Papers on Linguistics 16.] Southern Illinois University. Sagey, E. (1986). The representation of features and relations in nonlinear phonology. Doctoral dissertation, MIT. Selkirk, E. (1981). On prosodic structure and its relation to syntactic structure. Nordic Prosody II, edited by T. Fretheim, 111-140. Trondheim, Norway: Tapir. Selkirk, E. (1995). Sentence prosody: intonation, stress, and phrasing. The handbook of phonological theory, edited by J. Goldsmith, 550-569. Cambridge, Mass.: Blackwell. Selkirk, Ε. & T. Shen (1990). Prosodic domains in Shanghai Chinese. The phonology-syntax connection, edited by S. Inkelas & D. Zee, 313-337. CSLI, Stanford University, Stanford, Calif. Distributed by University of Chicago Press. Shen, X. N. S. (1989). Interplay of the four citation tones and intonation in Mandarin Chinese. Journal of Chinese Linguistics 17, 61-73. Shi, F., L. Shi & R. R. Liao (1987). An experimental analysis of the five level tones of the Gaoba Dong language. Journal of Chinese Linguistics 15, 335-361. Shih, C. L. (1986). The prosodic domain of tone sandhi in Chinese. Doctoral dissertation, University of California, San Diego. Shih, C.L. (1997). Mandarin third tone sandhi and prosodic structure. Studies in Chinese phonology, edited by J. L. Wang & N. Smith, 81-123. Berlin: Mouton de Gruyter. Sietsema, B.M. (1989). Metrical dependencies in tone assignment. Doctoral dissertation, MIT. Steriade, D. (1987). Redundant values. Papers from the 23rd Annual Regional Meeting of the Chicago Linguistic Society, Part 2: Parasession on Autosegmental and Metrical Phonology, 339-362. Stevens, K. & S.J. Keyser (1989). Primary features and their enhancement in consonants. Language 65, 81-106. Tsay, J. S. (1991). Tone alternation in Taiwanese. Arizona Phonology Conference, volume 4, 76-87. University of Arizona, Tucson. Wang, J. & L.J. Wang (1993). Putonghua duo yinjie ci yinjie shi chang fenbu moshi [The types of relative lengths of syllables in polysyllabic words in Putonghua], Zhongguo Yuwen 1993.2 (233), 112-116. Wang, W.S.-Y. (1967). Phonological features of tone. International Journal of American Linguistics 33, 93-105.

286

San Duanmu

Ward, I. C. (1944). A phonetic introduction to Mende. An introduction to the study of Mende, Κ. H. Crosby, 1-7. Cambridge, England: Heffer and Sons. Williams, E. (1976). Underlying tone in Margi and Igbo. Linguistic Inquiry 7, 436-468. Williamson, K. (1986). The Igbo associative and specific constructions. Phonological representations ofsuprasegmentals, edited by K. Bogers, H. van der Hulst & M. Mous, 195-208. Dordrecht: Foris Publications. Withgott, Μ. & P.-K. Halvorsen (1988). Phonetic and phonological considerations bearing on the representation of East Norwegian accent. Autosegmental Studies on Pitch Accent, edited by van der Hulst & Smith, 279-94. Woo, N. (1969). Prosody and phonology. Doctoral dissertation, MIT. Wright, M. (1983). A metrical approach to tone sandhi in Chinese dialects. Doctoral dissertation, University of Massachusetts, Amherst. Xu, B.H., Z.Z. Tang, R.J. You, N.R. Qian, R.J. Shi & Y.M. Shen (1988). Shanghai Shiqii fangyan zhi [Urban Shanghai dialects]. Shanghai: Shanghai Jiaoyu Chubanshe. Yan, J.Z. & M.C. Lin (1988). Beijinghua sanzizu zhongyin de shengxue biaoixian [Acoustic characteristics of the stress in Beijing trisyllables]. Fanyan 1988.3, 227-237. Yip, M. (1980). Tonal phonology of Chinese. Doctoral dissertation, MIT. Yip, M. (1989). Contour tones. Phonology 6,149-174. Yip, M. (1992). Prosodic morphology in four Chinese dialects. Journal of East Asian Linguistics 1,1-35. Yip, M. (1993). Tonal register in East Asian languages. The phonology of tone, edited by van der Hulst & Snider, 245-268. Yip, M. (1994). Isolated uses of prosodic categories. Perspectives in phonology, edited by J. Cole & C. Kisseberth, 293-311. [CSLI Lecture Notes No.51.] Center for the Study of Language and Information, Standford University. Zemlin, W. R. (1981). Speech and hearing science: Anatomy and physiology. Second edition. Englewood Cliffs, N.J.: Prentice-Hall. Zhang, H.M. (1992). Topics in Chinese phrasal phonology. Doctoral dissertation, University of California, San Diego. Zhu, X.N. (1995). Shanghai tonetics. Doctoral dissertation, Australian National University, Canberra.

The phonology-morphology interface Geert Booij

1.

Introduction

The study of the interface of phonology and morphology has been one of the central topics of research of generative phonologists in the eighties. The basic issue is: how does the morphological structure of words affect the application of the rules of phonology? The best known theory of that kind of interaction is the theory of Lexical Phonology, a theory formulated within a rules-and-derivations framework. The problem often encountered in discussions and evaluations of this theory is that 'Lexical Phonology' does not stand for one specific hypothesis but rather for a cluster of related, but not necessarily logically interdependent theoretical claims. My aim in this article is to first discuss what the basic hypothesis of Lexical Phonology is, and to provide arguments for it. Subsequently, I will review a number of related, but logically independent hypotheses concerning the organization of the lexicon such as level ordering. I will argue that the theory of Lexical Phonology derives from some very simple basic assumptions concerning rule application. I hasten to add that, in the study of this interface, one might also investigate, inversely, how morphology makes use of phonological categories. This is the research objective of "Prosodic Morphology" (McCarthy & Prince 1990, 1993), which requires a separate state-of-the-art article.

288

Geert Booij 2. The basic

claim

The basic hypothesis of Lexical Phonology which was independently arrived at by Pesetsky and Booij, and which was incorporated into a more elaborate theory by Kiparsky (Pesetsky 1979; Booij 1981; Kiparsky 1982) is the following: (1)

Morphology and phonology apply in tandem

To put it simply, you take a word, and apply the applicable phonological rules right away; you may then apply a morphological rule to that word, which creates a new domain of application for the phonological rules of the language, the second cycle, which in its turn can be input for another morphological operation that creates a third cycle, and so on. In other words, hypothesis (1) predicts that (2)

a. b.

the phonological rules of a language that apply to words apply cyclically, at least in principle; the morphological rules of a language may refer to derived (predictable) phonological properties of their input words.

The cyclic application of phonological rules in its turn predicts that (3)

phonological and morphological rules may make use of phonological information that is no longer present in the phonetic form of words.

Below, I will summarize the evidence for these claims. Hypothesis in (1) is in fact a consequence of an even simpler idea, namely the following minimal assumption (cf. also Kaye 1992): (4)

Apply a rule when possible

This principle predicts, for instance, that the rule of word stress of a language applies immediately to a given word, before it is subject to morphological operations. Thus, cyclicity of stress assignment follows from principle (4). Of course, principle (4) does not predict that all phonological rules of a language apply in the lexicon. Phonological rules can be specified as to the domain in which they apply. Rules that apply within the domain of the syllable, the foot, or the prosodic word can already apply in the lexicon because these prosodic categories are already available during the construction of words (Booij 1988, 1992; Inkelas 1989, 1993). On the other hand, many rules have domains larger than the word, e.g. the phonological phrase. Such rules are by definition postlexical (post-syntactic) rules since their applicability depends on the availability of domains created on the basis of syntactic structure.

The phonology-morphology

289

interface

Secondly, principle (4) may also be restricted by specific constraints which force a rule to apply post-cyclically, even though the rule has the word as its domain. In particular, the cyclic application of rules has been argued to be subject to the condition of Strict Cyclicity that says that rules only apply cyclically in derived environments, at least when they apply in a structure-changing fashion (Kiparsky 1982). So we get three stages at which phonological rules can apply (Booij & Rubach 1987,1991; Borowsky 1993; Hargus 1993; Mohanan 1995): (5)

cyclic level word level postlexical level

Note that this reconstruction of the basic idea of Lexical Phonology implies that a rule can apply at more than one level, as has been argued in Mohanan & Mohanan (1984) and Kiparsky (1985). This can be illustrated by the rule of Prevocalic Schwa-deletion in Dutch. This rule obligatorily deletes a schwa before an adjacent vowel within the domain of the prosodic word. Since the sequence schwa-vowel is a derived environment, the rule can apply cyclically, and there are independent arguments why cyclic application of this rule is necessary (see Booij 1981; Booij & Rubach 1984). Interestingly, new possibilities for the application of the rule are created by the syntax: Dutch has a number of vowel-initial enclitics (weak forms of pronouns and determiners) that can be incorporated phonologically into the preceding prosodic word. Thus, we get new schwa-vowel sequences at the postlexical level that are subject to the rule of Prevocalic Schwa-deletion: (6)

Ik zette het Ik haalde hem

Ί put it' Ί fetched him'

/zeta at/ /halda am/

[zetot] [haldom]

This view of Lexical Phonology (also advocated in Kiparsky 1985) goes against the older one that the three levels mentioned above imply that the phonological rules of a language are divided into three ordered blocks corresponding to these levels. Moreover, it also implies that there is no need to provide each rule with a label as to the block to which it belongs. The view outlined above implies that syllabification of words is a cyclic process, and there is ample empirical evidence for this claim, the hypothesis of continuous syllabification (Booij & Rubach 1984; Itö 1986,1989; Hayes & Abad 1989; Rubach & Booij 1990). Syllabification rules also nicely illustrate that rules are not necessarily confined to one level, since they reapply at the cyclic level, the word level, and the postlexical level. For instance, after insertion or deletion of a vowel by a rule of phonology at the post-lexical level, syllabic structure has to be redetermined. In other words, syllabification rules are indeed "anywhere rules", but this needs not be stipulated.

290

Geert Booij

The classic example of cyclic rule application is the Main Stress Rule for English. It concerns pairs of deverbal nouns like cömpensätion (derived from compensate) versus cöndensätion (derived from condinse). The second syllable of condensation still bears some degree of stress since on the previous cycle (condinse) it had main stress, and therefore is does not allow for vowel reduction. However, Halle & Vergnaud (1987, p. 104) argue that this cyclicity effect is a language-specific property of English, to be expressed by a rule of Stress Copy that copies line 1 asterisks (line 1 is the foot level) assigned on preceding cycles. Normally, the stress of previous cycles is erased, and the stress pattern of the new cycle is determined completely anew. Yet, this does not mean that cyclicity of stress assignment is no longer motivated empirically since the application of certain word formation rules (for instance, suffixation of the deverbal -al) is dependent on the stress pattern of the input words. There are quite a number of cases in the literature in which a morphological operation requires information concerning the prosodic structure or the stress patterns of its input words (but see Odden 1990 for critical comments). A clear example of the role of cyclic stress assignment in the operation of phonological rules can be found in Dutch. In complex words created by the addition of a vowel-initial non-native suffix, the last vowel of the input word is lengthened unless it bears the main stress ofthat word on the first cycle (Booij 1995): (7)

alcoh[o]l 'alcohol' profess[o]r 'professor'

alcoh[0]lisch 'alcoholic, A' alcoh[o]list 'alcoholic, N' profess[o]räal 'professorial'

tromp[e]t 'trumpet' fag[o]t "bassoon' t[o]n 'ton' bl[6]k 'block'

tromp[e]ttist 'trumpet player' fag[o]ttist 'bassoon player' t[o]nnage 'tonnage' bl[o]kk£er 'to block'

but:

A nice near-minimal pair is formed by: £ir[d]/i 'tyrant'- tir[a]nniseer 'to tyrannize' versus cdn[o\n 'canon'- can[o]niseer 'to canonize'. It is the difference in stress pattern between the two base words that explains why vowel lengthening only occurs in the second verb in -iseer. Note that on the second cycle, the stress pattern is redetermined, and therefore vowel lengthening must apply on the second cycle on the basis of the stress pattern derived on the first cycle. For instance, in both alcoholist and trompettist main stress is on the final syllable. Yet, vowel lengthening only takes place in alcoholist, reflecting the difference in location of stress between alcohol and trompit. This also implies that on the second cycle the lengthening rule has to precede the reapplication of the Main Stress Rule.

The phonology-morphology

interface

291

As stated in (2b), the principle "Apply when possible" also implies that the application of morphological rules may be preceded by the application of phonological rules. A classic case is that of reduplication which may copy part of an input word that clearly has already undergone phonology, as in Bloomfield's (1933) example of internal reduplication in Tagalog pa-mumutul, derived frompamutul that in its turn derives from pang-putul. Another kind of evidence for cyclic rule application is that morphological operations may impose phonological restrictions on their input words which pertain to derived phonological properties, in particular the number of syllables of the input words and their stress properties. A simple example is English comparative formation which is only possible for monosyllabic words, and for disyllabic words with a second light syllable such as happy. Furthermore, the affixation of English deverbal -al is subject to the condition that the last syllable of the verbal stem bears stress: try-trial, but organize — *organizal. Also, the German prefix ge- that is used in the formation of past participles requires the following stem to begin with a syllable with main stress. The relevance of the prosodic structure of the input of morphological operations is particularly clear in the framework of Prosodic Morphology. The basic claim of this theory is that morphological operations such as reduplication and infixation, which are formalized in terms of templates and prosodic circumscription, are "defined in terms of the authentic units of prosody: mora, syllable, foot, prosodic word" (McCarthy & Prince 1993, p. 1), which clearly implies that these prosodic units must be present in the input words. In the case of reduplication, one might avoid this kind of interaction by representing reduplication as an abstract prosodic template attached to a stem which is then filled in after the application of phonology to the stem (Odden 1993). In case of infixation, however, this is no feasible solution since the location of the infix depends on the prosodic structure of the stem, which must therefore be available when infixation applies.

3. The word level In the original model as proposed by Kiparsky (1982) it was assumed that all lexical (applications of) rules are cyclic. However, it appeared that certain lexical rules should apply after all morphology has been done. The earliest level at which they can apply is the word level, i.e. after all morphology has been performed. For example, the rule of Final Devoicing in Dutch that devoices obstruents in codas should apply after morphology (Booij 1981). Otherwise, wrong phonetic forms would be derived, such as [heltin] instead of the correct [heldin] for heldin 'heroine' with the morpho-

292

Geert Booij

logical structure [[held]in], derived from held /held/ 'hero'. An example of a word level rule from English is the rule of 1-velarization: l's are alveolar in onsets, and velarized in codas: (8)

onset: coda:

[l]ap, Yo[l]anda, vigi[l]ant pa[l], ca[l], vigi[l]

As shown by the pair vigil — vigilant the rule must be postcyclic, otherwise vigilant would inherit a velarized [1] from its base word vigil. The word level is not to be seen as a particular part of the lexical component of the grammar. The rules involved just apply when possible. Since their domains are the lower category prosodic domains syllable, foot and prosodic word, which are available before syntax, the rules can apply right away, i.e. before syntax. The only constraint they are subject to and that has to be encoded on the relevant rules is that they cannot apply until after the end of the morphological operations. The main effect of applying rules at the word level instead of the cyclic level is that the phonological generalizations involved will be true at the surface level, and hence they form transparent generalizations. In the case of the rule of Final Devoicing in German and Dutch, moreover, cyclic application of this rule would make the underlying distinction between voiced and voiceless obstruents in morpheme-final position completely useless, since these obstruents always occur in coda position on the first cycle, and hence the voiced obstruents would never get the chance to surface. A concomitant correct prediction is that word level rules are not bled by resyllabification effects at the postlexical (= post-syntactic) level. For instance, in French the lexical rule of Closed Syllable Adjustment that determines the height of mid vowels applies lexically to closed syllables, although these syllables may be open at the postlexical level due to resyllabification Cenchainement") (Booij 1984). Similarly, the Dutch enclitic -ie /i1 induces resyllabification of the final consonant of its host, thus shifting it to an onset position, but in case that consonant is an obstruent, it is realized as a voiceless consonant, although it stands in onset position, as is correctly predicted by the theory. To give an example, the phonetic form of vond-ie 'found he' is [vonti], and not *[vondi], although the syllabification pattern is (νοη)σ(ίϊ)σ (Booij 1985, 1996). In German, Final Devoicing also has to apply before postlexical cliticization, except in (lexicalized) expressions such as hab ich, Tiave I', hab es "have it' which have two possible pronunciations: [hap/bip], [hap/bos] (Wiese 1996). One may wonder how the application of rules that should only apply cyclically are blocked from applying at the word level. A possible solution is the idea that rules apply as soon as possible, but can be turned off at a later level. For instance, the Polish rule of Palatalization discussed below,

The phonology-morphology

interface

293

at least if it is a rule, should be blocked from applying morpheme-internally at the word level (Rubach 1984). Otherwise, we would also palatalize the first /s/ of servis+e 'service-LOC SG', after all, and the desirable effect of Strict Cyclicity would be undone. In the case of Polish Palatalization, this problem does not arise at all if, as proposed by Kiparsky (1993), Strict Cyclicity is seen as the effect of structure-building rules applying to underspecified representations. For instance, the first /s/ of servis will be fully specified as [+anterior], whereas the second /s/ will be unspecified for [anterior]. When this underspecified segment occurs before a [-ant] segment, it will be [-ant], and otherwise, by default, [+ant]. However, in other cases such as Finnish Consonant Gradation it remains necessary to restrict application of the rule to phonological words available at the lexical level, hence this rule must be turned off at the postlexical level, unlike Prevocalic Schwa-deletion in Dutch. Another general constraint on the lexical application of rules is that they should be obligatory. This reflects the classical distinction in European structuralist phonology between word phonology and sentence phonology. Take the case of the Dutch rule of/n/-deletion that is optional and deletes /n/'s in rhymes consisting of a schwa plus /n/ at the end of a morpheme: (9)

wapen 'weapon' wapen-en 'weapons' molen 'mill' molen-aar 'miller'

[wapan] or [wapa] [wapanan] or [wapana] [molan] or [mole] [molana:r]

Apart from its optionality it could have been a word level rule. However, it is clear that it should apply post-lexically since it must be bled by cliticization and the ensuing resyllabification: (10)

zetten het 'put it' /zetan at/ phonetic form (ζε)σ(ίθ)σ(ηθΐ)σ

If we allowed for deletion of the /n/ at the word level, we would expect the phonetic form [zetat] to be possible through /n/-deletion followed by Prevocalic Schwa-deletion. However, this phonetic form is completely out as realization of zetten het, although it is a normal realization of zette het, with the verb in the past singular form.

4. Properties of lexical rules In the course of the development of the theory of Lexical Phonology, a number of specific distinguishing properties have been proposed for lexical rules, summarized in Kaisse & Shaw (1985):

294

Geert Booij

a. b. c. d. e.

Lexical rules apply within words only are cyclic are subject to Strict Cyclicity are structure-preserving may have exceptions

We may also add the property of obligatoriness to this list, but clearly the inverse does not hold for postlexical rules, which might be obligatory as well. We should also note that these defining properties are not to be taken as defining properties of the rules themselves, but rather of the way in which they apply, since the same rule may apply both lexically and postlexically. Moreover, the fact that lexical rules apply within words only is simply a matter of the domain specification that each rule requires. Furthermore, as we saw above, cyclicity is not an inherent property of rules, but follows from principle (4): "Apply a rule when possible'. Therefore, the question is rather whether the following statements are true (12)

a. b.

The application of phonological rules at the cyclic level is subject to Strict Cyclicity Phonological rules that apply lexically are structure-preserving

Cyclicity and strict cyclicity must be distinguished: one can also envisage a theory in which rules apply cyclically without Strict Cyclicity being in force. Classic cases in favour of Strict Cyclicity are the English Trisyllabic Laxing Rule that accounts for alternations like sane-sanity and the Polish rule of Palatalization of consonants before front vowels and glides discussed above. The analysis of Trisyllabic Laxing and Palatalization in terms of Strict Cyclicity is not without problems, as Szpyra (1989, pp. 69-83) noted. Quite a number of exceptions to these rules remain unexplained. For instance, the following complex English words are exceptions to Trisyllabic Laxing: (13)

hyphenate, momentary, dangerous, vaporous, alienate, motorist

In the case of Polish Palatalization it appears that this rule does not apply before the following suffixes (Szpyra 1989, p. 79): (14)

-e FEM PL, -ego GEN SG, -emu DAT SG, -ej DAT SG, -em INSTR SG, -e NOM PL

However, these criticisms do not imply that Strict Cyclicity is incorrect, but only that this concept may not suffice to account for the phenomena under discussion here. For instance, the rules may have lexical exceptions, or may be conditioned by specific morphemes. Moreover, there are other cases in

The phonology-morphology interface

295

the literature where Strict Cyclicity appears to make the right predictions (cf. Clark 1990). Evidence for the necessity of Strict Cyclicity can also be found in the realm of syllabification. As argued by Guerssel (1986), Kenstowicz & Rubach (1987) and others the distinction between (high) vowels and glides is not a difference in feature specification, but a difference in prosodic positions: vocoids are vowels when in nuclear position, and glides otherwise. It is usually predictable whether a vocoid will surface as a vowel or as a glide. However, in some languages the vowel-glide distinction may be distinctive in certain positions. In those cases vocoids that unpredictably surface as nuclei will be specified as such in the lexicon, for instance by assigning them the node Nucleus lexically. This lexical specification has to block the Gliding rule, which is part of the syllabification algorithm, and assigns vocoids in that position to a non-nuclear position. Otherwise, the lexical specification would not even make sense. That is, Strict Cyclicity will block such structure-changing applications of the syllabification algorithm, as required. Note, however, that the Strict Cyclicity effect is obtained automatically if syllabification rules are seen as rules that only build structure. Thus, they will respect existing lexically given prosodic structure. As argued in Rubach & Booij (1990), the only structure changing operation involved in syllabification is the Coda Erasure rule which removes the coda node of the last syllable of a base word when a suffix is attached, after which the structure building rules reapply. The same remark applies to languages that require lexical (i.e. unpredictable) marking of stress in some words. Such lexical specifications make only sense if they will not be undone by the stress rule(s) of the language. An example of such a language is Greek (Malikouti-Drachman & Drachman 1992, p. 203). The Strict Cyclicity effect follows from the fact that stress rules are metrical structure building rules that respect metrical structure already provided in lexical representations. Structure Preservation is the hypothesis that lexical rules do not introduce types of segment that do not occur in the underlying segment inventory of the language (cf. Myers 1991). For instance, the distribution of the aspirated stops in English is completely predictable, and hence there are no underlying aspirated stops in English. Consequently, it is predicted that the rule of aspiration applies postlexically. Kiparsky (1985, p. 92) explicitly hypothesized that all lexical rules are structure-preserving, including the word level rules. However, certain evidence suggests that this claim cannot be completely upheld: certain lexical rules are not structure-preserving. To begin with, Dutch has a rule of Pre-r-lengthening which lengthens vowels before Irl in the same prosodic

296

Geert Booij

word. This rule introduces non-distinctive properties, it creates segments that do not occur at the underlying level: (15)

biet /bit/ [bit] 'beet' bier /bir/ [bi:r] Tjeer' eet /et/ [et] 'eat' eer /er/ [e:r] "honour'

Yet, the rule appears to be a lexical rule since it has exceptions: the past tense forms of a number of ablauting verbs do not participate in the lengthening, e.g. wierp 'threw', hielp "helped',ftedier/'spoiled'.Suppose now that we assume that it is only the cyclic level that is subject to the constraint of Structure Preservation. Kipar sky (1985, p. 135, note 3) suggested the possibility that the marking conditions that express generalizations concerning underlying segments may be turned off at a later level in the lexicon. This solves the problem, at least for Dutch, since the rule is clearly a rule that should not and cannot apply cyclically: the environment in which it applies is never a derived environment. The same problem occurs for German, as pointed out by Hall (1989). The dorsal fricative consonant in German is a velar [x] after a back vowel ("Achlaut"), and a palatal fricative [5] ("Ich-laut") elsewhere. In other words, the distribution of [x] versus [9] is predictable by rule. The rule that specifies the velarity of the fricative after back vowels is therefore not structurepreserving since there is no underlying opposition [x] versus [5] (the other allophone is introduced by a default rule). Yet, the rule must be lexical since it only applies within morphemes; the palatal [9] does occur after a back vowel as the first consonant of the diminutive suffix -chen, hence minimal pairs like (16)

Kuchen 'cakes' [ku:xan] versus Kuh+chen 'little cow' [ku:pan]

(cf. also Macfarland & Pierrehumbert 1991 and Iverson & Salmons 1992). German R-vocalization also introduces a segment that does not belong to the underlying segment inventory (Hall 1993). Yet, it must apply at the word level, because it is not affected by resyllabification. This supports the idea that Structure Preservation is not a condition on the application of rules at the word level.

5. Level

ordering

A typical feature of many analyses in the framework of Lexical Phonology is that they make use of levels and level ordering, to such an extent that Lexical Phonology is sometimes equated with level ordering. The idea of

The phonology-morphology interface

297

level ordering is that the lexicon consist of two or more ordered levels or strata. Morphological rules are assigned to specific strata, and phonological rules can also be assigned to one or more of these strata. The classical hypothesis of level ordering for English is there is a level 1 at which stressshifting affixes are attached to stems, and at which the Main Stress Rule of English applies, and a level 2 for the attachment of stress-neutral affixes, where the Main Stress Rule no longer applies. The morphological prediction then is that stress-neutral affixes are peripheral to stressshifting affixes. For English different numbers of lexical levels have been proposed: two (Siegel 1974), three (Kiparsky 1985), and four (Halle & Mohanan 1985). Other languages for which levels have been claimed to be relevant are: (17)

Dutch (3 levels, Paulissen & Zonneveld 1988, 2 levels, Kager 1989) Malayalam (4 levels, Mohanan & Mohanan 1984) Dakota (3 levels, Shaw 1985) Mandarin (4 levels, Packard 1990, but see Sproat & Shih 1993) Icelandic (2 levels, Kiparsky 1985) German (3 levels, Wiese 1988) Igbo (3 levels, Clark 1990) Italian (3 levels, Scalise 1984) Basque (2 levels, Hualde 1991)

The basic idea behind level ordering is that the phonological behaviour of affixes correlates with their morphological behaviour with respect to affix ordering. From the beginning it was clear that there is not always a perfect correlation between the phonological and the morphological behaviour of word formation processes. For instance, Mohanan & Mohanan (1984) had to allow for loops to earlier strata in order to derive the correct phonetic forms of certain types of complex word, which undermines the basic idea. A number of types of objection to level ordering can be found in the literature: it makes incorrect predictions (i.e. there is counterevidence), it is a superfluous mechanism because the facts already follow from other principles, and it creates conceptual problems. The classic example of empirical counterevidence for level ordering in English is the word ungrammaticality, in which first the stress-neutral prefix un- has been added to the adjective grammatical, and subsequently the stress-shifting suffix -ity. Therefore, ungrammaticality was called a bracketing paradox since in this case morphology requires a structure different from phonology. Similar examples are extrametricality and governmental, and words in -ation derived from words ending in the stress-

298

Geert Booij

neutral suffix -ize such as legalize-ation, divinize-ation, palatalize-ation (Szpyra 1989). The second type of objection is that the required affix ordering already follows from independently established principles (Fabb 1988; Booij 1987, 1989). For instance, in languages such as English, Dutch and German, [-native] suffixes can only be added to [-native] (underived or derived) stems. We do not find *yellowity besides yellowness, whereas both absurdity and absurdness are possible words of English. This makes the correct prediction that [+native] suffixes are always peripheral to [-native] suffixes. Since [-native] suffixes are often stress-shifting, and [+native] suffixes stress-neutral, the observation that stress-neutral suffixes are peripheral to stress-shifting suffixes follows from the constraint with respect to the feature [native]. Interestingly, there are crucial cases that show what the correct generalization is: Dutch has three [+native], but stress-shifting suffixes. The generalization in terms of the feature [native] rightly predicts that such stress-shifting suffixes can be external to stress-neutral suffixes: molen-aar-es 'female miller', ler-aar-is 'female teacher', etc.. Another example of the use of level ordering for morphological purposes is the following. It has been observed both for English and Dutch that the irregular formation of the past tense by means of Ablaut does not occur in verbs with ablauting roots that are derived from compounds: they take the regular past tense suffix: (18)

stand-stood / grandstand-grandstanded

Kiparsky (1982) accounted for this difference by locating the Ablaut rule at level 1, and conversion of nominal compounds to verbs at a later level, at which also regular past tense formation takes place. However, level ordering is superfluous here if we assume a feature [+Ablaut] that triggers the irregular past tense formation is not inherited by the complex verbs since they are conversion of nominal compounds, and features can only percolate across a proper path of nodes of the same syntactic category. In other words, the verb grandstand does not get the feature [+Ablaut], and hence it is subject to regular past tense formation (Booij 1989). Level ordering has also been used for phonological purposes. For instance, it has been proposed that in Italian derivation is level 1, and compounding is level 2 because certain phonological rules apply to derived words, but not to compounds. However, once we realize that compounds consist of more than one prosodic word, this difference in application follows straightforwardly from the domain assignment of the relevant rules: they apply within the domain of the prosodic word (Vogel 1991; Peperkamp 1997). The same point can be made with respect to Basque: levels are made

The phonology-morphology

interface

299

superfluous once the domain of application of the rules is specified properly. For instance, Basque has a rule of Low Vowel Assimilation (Hualde 1991, p. 23): (19)

V

[-low] / [+high] C0 —

In the Baztan dialect this rule applies morpheme-internally, in some derived words (Derivation I), in inflected words, and in host+clitic combinations, but not in compounds and certain types of derived words (Derivation II). Therefore, Hualde (1991, pp. 38-39) concludes to the following organization of Baztan Basque: (20)

Stratum I: inflection and derivation I Stratum II: derivation II and composition Postlexical: cliticization (back to Stratum I)

The rule of Low Vowel Assimilation is then assigned to Stratum I. Hualde argues that this solution is better than making use of a distinction between two types of boundary, + versus #, as was done in standard linear generative phonology, because the level approach eliminates the possibility of unrestricted or arbitrary use of boundary symbols. Note, however, that the ordering proposed implies that inflection is ordered before compounding and derivation II, and that a loop is necessary from a postlexical stage back to Stratum I. The correct application of Low Vowel Assimilation can be achieved without level ordering, by assigning this rule the prosodic word (ω) as its domain of application. Inflectional suffixes, and derivational suffixes of class I can be interpreted as cohering suffixes, i.e. they form one prosodic word with the preceding material. Compounds, on the other hand, consist of more than one prosodic word. The suffixes of class II will be qualified as non-cohering suffixes which form a prosodic word of their own (cf. Booij & Rubach 1984 for cross-linguistic evidence for this approach). That the attachment of clitics induces application of Low Vowel Assimilation follows from the assumption that cliticization means that a function word is prosodically incorporated into the preceding prosodic word, its host. Thus, at the post-lexical level, cliticization creates new domains of application for rules that apply with the prosodic word. Independent evidence for this assumption in the case of Basque is that clitics undergo other rules that are usually confined to words. For instance, in Baztan Basque the initial /b/ of the determiner bat deletes intervocalically, i.e. after a host word ending in a vowel, whereas this rule does not apply across word boundaries (Hualde 1991, p. 29). Note, again, that this implies that a rule can apply both lexically and postlexically, as was already argued above. In sum, an analysis in which rules are properly specified for their domain of application in terms of

300

Geert Booij

prosodic categories, can make level ordering for phonological purposes, and the concomitant complication of a loop, superfluous. Moreover, it avoids the awkward morphological consequence that inflection is ordered before certain types of word formation (derivation II and compounding). The phonological use of level ordering has also been invoked when languages have lexicons with different historical layers with their own phonological generalizations, as is the case for English, Dutch, and Malayalam. In Germanic languages such as English and Dutch we find a GreekLatin-Romance stratum, and in Malayalam there is a Sanskrit stratum and a Dravidian stratum. Phonological rules that apply to only one layer of the lexicon can be seen as rules that are conditioned by a diacritic feature that refers to the relevant stratum. For instance, in Dutch and English the Latinate prefixes in- and con- exhibit a more radical form of place assimilation than the negative prefix on-/un-: before liquids there is total assimilation in the case of the Latinate prefixes: il-, ir-, cor-, col-. The relevant generalization can be made by means of reference to the feature [-Native], a features that is needed anyway for the morphology of English and Dutch. Clearly, this type of assimilation is a process borrowed from Latin. There is also a conceptual problem with level ordering in word-based morphology: word formation rules may take existing words listed in the lexicon as their inputs since clearly idiosyncratic properties of words recur in complex words derived from them. This implies a loop in the organization of the lexicon, and thus the morphological component may add an affix of level η to an already existing word listed in the lexicon with the structure *base-affix of level n + l \ thus making level ordering empty (Booij 1987; Aronoff 1988). One might try to solve these problems by interpreting a difference in levels as a difference in constituency. This has been proposed by Goldsmith (1990, p. 241), who proposed three hierarchically ordered constituents for Bantu languages: root, stem, word, which parallels Clark's (1990) distinction of three levels for Igbo. Similarly, Golston (1996) defines different domains of rule application by making use of alignment conditions that refer to the edges of different kinds of constituent. Another alternative is pursued in Inkelas, Orgun & Zoll (1997), where the suffixes of Turkish are divided into subsets, each with their own "co-phonology".

6.

Conclusions

In this article I identified the theoretical core of Lexical Phonology as the following minimal assumption: "Apply a rule when possible". The application of

The phonology-morphology

interface

301

rules is further determined by its domain of application, and the principles of Strict Cyclicity and Structure Preservation. With these minimal assumptions it is possible to correctly predict the way in which phonology and morphology appear to interact. It is not necessary to assign a rule to a specific level of the grammar (although it may be the case that certain rules have to be turned off at a later level of derivation). This implies that, in principle, rules can apply at more than one level, for which we saw empirical evidence. Lexical Phonology in its classical form makes use of rules and derivations. Therefore, one might wonder how these insights relate to constraintbased theories of phonology. The answer is that they are not necessarily incompatible, since constraint-based models could allow for constraint evaluation at more than one level, and also in a cyclic fashion. These issues are discussed in Kenstowicz (1994), Booij (1997), Peperkamp (1997), and McCarthy (1998).

Acknowledgments This article is partially based on the article "Lexical Phonology: a Review" by the same author that appeared in Lingua e Stile 29, 525-556 (1994).

A Lexical Phonology

Bibliography

Aronoff, M. (1988). Review of A.-M. di Sciullo & E. Williams, On the definition of word, Language 64, 766-770. Bloomfield, L. (1933). Language. London: Allen & Unwin. Booij, G.E. (1981). Rule ordering, rule application and the organization of grammars. Phonologica 1980, edited by W.U. Dressier, O. Pfeiffer & J. R. Rennison, 45-56. Innsbruck: Institut für Sprachwissenschaft. Booij, G.E. (1984). French C/0 alternations, extrasyllabicity and Lexical Phonology. The Linguistic Review 3,181-207. Booij, G. E. (1985). Lexical phonology, final devoicing, and subject pronouns in Dutch. Linguistics in the Netherlands 1985, edited by H. Bennis & F. Beukema, 21-26. Dordrecht: Foris Publications. Booij, G.E. (1987). Lexical Phonology and the organization of the morphological component. Rules and the lexicon, edited by E. Gussmann, 43-65. Lublin: Catholic University of Lublin.

302

Geert Booij

Booij, G. E. (1988). On the relation between Prosodic Phonology and Lexical Phonology. Certamen phonologicum, edited by P.M. Bertinetto & M. Loporcaro, 63-76. Torino: Rosenberg & Sellier. Booij, G.E. (1989). Complex verbs and the theory of level ordering. Yearbook of Morphology 1989, edited by G.E. Booij & J. van Marie, 21-30. Dordrecht: Foris Publications. Booij, G. E. (1992). Lexical Phonology and Prosodic Phonology. Phonologica 1988, edited by W.U. Dressler, H. Luschützky & J.R. Rennison, 49-62. Cambridge University Press. Booij, G.E. (1995). The phonology of Dutch. Oxford: Clarendon Press. Booij, G.E. (1996). Cliticization as prosodic integration: the case of Dutch. The Linguistic Review 13, 219-242. Booij, G.E. (1997) Non-derivational phonology meets lexical phonology. Derivations and constraints in phonology, edited by I. Roca, 261-288. Oxford: Clarendon Press. Booij, G.E. & J. Rubach (1984). Morphological and prosodic domains in Lexical Phonology. Phonology Yearbook 1,1-27. Booij, G.E. & J. Rubach (1987). Postcyclic versus postlexical rules in Lexical Phonology. Linguistic Inquiry 18,1—44. Booij, G.E. & J. Rubach (1991). Lexical Phonology. The Oxford Encyclopedia of Linguistics, Vol 2, edited by W. Bright, 327-330. Oxford: Oxford University Press. Borowsky, T. (1993). On the word level. Studies in Lexical Phonology [.Phonetics & Phonology 4], edited by S. Hargus & E. Kaisse, 199-234. San Diego: Academic Press. Clark, M. (1990). The tonal system oflgbo. Dordrecht: Foris Publications. Fabb, N. (1988). English suffixation is constrained only by selectional restrictions. Natural Language and Linguistic Theory 6, 527-539. Goldsmith, J. (1990). Autosegmentaland metrical phonology. London: Basil Blackwell. Golston, Ch. (1996). Prosodic constraints on roots, stems, and words. Interfaces in phonology, edited by U. Kleinhenz, 172-193. Berlin: Akademieverlag. Guerssel, M. (1986). Glides in Berber and syllabicity. Linguistic Inquiry 17, 1-12.

Hall, T. A. (1989). Lexical Phonology and the distribution of German [9] and [x]. Phonology 6,1-17. Hall, T. A. (1993). The phonology of German /R/. Phonology 10, 83-106. Halle, M. & K.P. Mohanan (1985). Segmental phonology of modern English. Linguistic Inquiry 16, 57—116.

The phonology-morphology interface

303

Halle, M. & J.-R. Vergnaud (1987). An essay on stress. Cambridge, Mass.: MIT Press. Hargus, S. (1993). Modeling the phonology-morphology interface. Studies in Lexical Phonology [Phonetics & Phonology 4], edited by S. Hargus & E. Kaisse, 45-74. San Diego: Academic Press. Hayes, Β. & M. Abad (1989). Reduplication and syllabification in Ilokano. Lingua 77, 331-374 Hualde, J.I. (1991). Basque phonology. London: Routledge. Inkelas, S. (1989). Prosodic constituency in the lexicon. Doctoral dissertation, Stanford University. Inkelas, S. (1993). Deriving cyclicity. Studies in Lexical Phonology [Phonetics & Phonology 4], edited by S. Hargus & E. Kaisse, 75-110. San Diego: Academic Press. Inkelas, S., 0. Orgun & Ch. Zoll (1997). The implications of lexical exceptions for the nature of grammar. Derivations and constraints in phonology, edited by I. Roca, 393-418. Oxford: Clarendon Press. Ito, J. (1986). Syllable theory in prosodic phonology. Doctoral dissertation, University of Massacusetts, Amherst. Ito, J. (1989). A prosodic theory of epenthesis. Natural Language and Linguistic Theory 7, 217-259. Iverson, G.K. & J. Salmons (1992). The place of Structure Preservation in German diminutive formation. Phonology 9,137-144. Kager, R. W.J. (1989). A metrical theory of stress and destressing in Dutch. Dordrecht: Foris Publications. Kaisse, Ε. M. & P. Shaw 1985. On the theory of Lexical Phonology. Phonology Yearbook 2,1-30. Kaye, J. (1992). On the interaction of theories of Lexical Phonology and theories of phonological phenomena. Phonologica 1988, edited by W. U. Dressler, H. Luschützky & J. R. Rennison, 141-155. Cambridge University Press. Kenstowicz, M. (1994). Cyclic versus non-cyclic constraint evaluation. MIT Working Papers in Linguistics 21, 11-42. Kenstowicz, M. & J. Rubach (1987). The phonology of syllabic nuclei in Slovak. Language 63, 463-497. Kiparsky, P. (1982). From Cyclic Phonology to Lexical Phonology. The structure of phonological representations, part 1, edited by H. van der Hulst & N. Smith, 131-175. Dordrecht: Foris Publications. Kiparsky, P. (1984). On the Lexical Phonology of Icelandic. Nordic Prosody III, edited by C. C. Eiert et al., 135-164. Stockholm: Almqvist & Wiksell. Kiparsky, P. (1985). Some consequences of Lexical Phonology. Phonology Yearbook 2, 85-138.

304

Geert Booij

Kiparsky, P. (1993). Blocking in non-derived environments. Studies in Lexical Phonology [Phonetics & Phonology 4], edited by S. Hargus & E. Kaisse, 277-314. San Diego: Academic Press. Macfarland, T. & J. Pierrehumbert (1991).On ich-laut, ach-laut, and Structure Preservation. Phonology 8,171-180. Malikouti-Drachman, A. & G. Drachman (1992). Greek clitics and Lexical Phonology. Phonologica 1988, edited by W. U. Dressler, H. Luschützky & J. R. Rennison, 197-206. Cambridge University Press. McCarthy, J.J. (1998). Sympathy and phonological opacity. Manuscript, University of Massachusetts, Amherst. McCarthy, J.J. & A. Prince (1990). Foot and word in Prosodic Morphology. Natural Language and Linguistic Theory 8, 209-283. McCarthy, J.J. & A. Prince (1993). Prosodic Morphology I. Constraint Interaction and Satisfaction. Manuscript, University of Massacusetts, Amherst and Rutgers University. Mohanan, K. P. (1995). The organization of the grammar. The Handbook of Phonological Theory, edited by J. Goldsmith, 24-69. Oxford: Blackwell. Mohanan, K.P. & Mohanan, T. (1984). Lexical Phonology and the consonant system in Malayalam. Linguistic Inquiry 15, 575-602. Myers, S. (1991). Structure Preservation and the Strong Domain Hypothesis, Linguistic Inquiry 22, 379-385. Odden, D. (1990). Phonology and its interaction with syntax and morphology. Studies in the Linguistic Sciences 20, 69-88. Odden, D. (1993). Interaction between modules in Lexical Phonology. Studies in Lexical Phonology [Phonetics & Phonology 4], edited by S. Hargus & E. Kaisse, 111-144. San Diego: Academic Press. Paulissen, D. & W. Zonneveld (1988). Compound verbs and the adequacy of Lexical Morphology. Morphology and Modularity, edited by M. Everaert, A. Evers, M. A. C. Huybregts & M. Trommelen, 281-301. Dordrecht: Foris Publications. Packard, J. (1990). A lexical morphology approach to word formation in Mandarin. Yearbook of Morphology 1990, edited by G.E. Booij & J. van Marie, 21-37. Dordrecht: Foris Publications. Peperkamp, S. (1997). Prosodic words. [HIL Dissertation Series 34.] The Hague: Holland Academic Graphics. Rubach, J. (1984). Cyclic and Lexical Phonology. Dordrecht: Foris Publications. Rubach, J. (1993). The Lexical Phonology of Slovak. Oxford: Clarendon Press. Rubach, J. & G.E. Booij (1990). Syllable structure assignment in Polish. Phonology 7,121-157.

The phonology-morphology interface

305

Scalise, S. (1984). Generative morphology. Dordrecht: Foris Publications. Shaw, P.A. (1985). Modularisation and substantive constraints in Dakota Lexical Phonology. Phonology Yearbook 2,173-202. Siegel, D. (1974). Topics in English morphology. New York: Garland. Sproat, R. & C. Shih (1993). Why Mandarin morphology is not stratumordered. Yearbook of Morphology 1993, edited by G. E. Booij & J. van Marie, 185-217. Dordrecht: Kluwer. Szpyra, J. (1989). The phonology-morphology interface. Cycles, levels and words. London: Routledge. Vogel, I. (1991). Level ordering in Italian Lexical Phonology? Certamen phonologicum II, edited by P.M. Bertinetto & M. Loporcaro, 81-101. Torino: Rosenberg & Sellier. Wiese, R. (1988). Silbische und lexikalische Phonologie. Studien zum Chinesischen und Deutschen. Tübingen: Niemeyer. Wiese, R. (1996). The phonology of German. Oxford: Clarendon Press.

Metrical phonology Harry van der Hulst

1.

Introduction

This article intends to inform the reader about Metrical Phonology, i.e. the theory about word level stress. I will discuss its original motivation and scope, how it extended its empirical domain, which changes it underwent and what the current situation is. Since we deal with a theory that was introduced twenty years ago, it would be easy to write a whole book dealing with these matters. In the few pages that I have here I will therefore have to limit myself to a selection of proposals (and references) that, in my view, are representative of crucial developments in the theory. I will not present a strict historical overview, but rather organize the discussion around various aspects of the theory, such as the central principles and parameters, the notation and the most important developments. For reasons of space I will not discuss the interplay between stress and morphology and refer to Revithiadou (1998). The stress theory discussed in this article, then, is essentially designed for fairly straightforward rhythmic aspects of word stress, applying within a domain that is either morphologically underived, or derived in terms of affixes which create morphological structure that is essentially invisible to the stress rules. Also, I limit my attention to the word domain; as a consequence, I will not deal with prosodic structure at a higher level, nor with the syntax- phonology connection (cf. Selkirk 1982; Nespor & Vogel 1986; Inkelas & Zee 1990), nor with the metrical approach to poetic meter (cf. Hayes 1983; Rice, this volume). I will not discuss the phonetics of stress (or the distinction between stressand pitch-accent languages), nor the role of stress vis-ä-vis intonation (cf. van der Hulst 1999, Dogil & Williams 1999, and Gussenhoven & Bruce

308

Harry van der Hülst

1999). The reader must also be warned for the fact that this overview reveals some of my own special interests. Finally, I will be brief on the Optimality Theory treatment of stress systems, and concentrate on the standard parametric approach. Burzio (this volume) will bring the reader up to date with the OT-way of analyzing stress systems.

2. The beginning Metrical theory was first developed in the MIT dissertation of Mark Liberman (1975). This thesis primarily deals with the intonational system of English, but Liberman included a new proposal for the representation of English word stress in his work. The proposal owes to ideas put forward by Alan Prince in his 1975 MIT dissertation on Tiberian Hebrew and the theory in its initial form is best known from an article that Liberman & Prince published together in 1977 in Linguistic Inquiry. English word stress had already been extensively analyzed in Noam Chomsky and Morris Halle's The Sound Pattern ofEnglish (1968; henceforth SPE). In the approach offered by Liberman & Prince, the SPE rules were not abandoned. As before, word stress was assigned to vowels by a modified version of the SPE Main Stress Rule and the Compound Stress Rule. The novelty was that, after stress assignment, the string of segments was now fed into an algorithm that parsed it into a constituent structure. This structure was called "metrical" because of the central role of a unit called foot, a term that was borrowed from poetic meter. Liberman & Prince assumed that segments are organized in syllables; SPE had tried to do phonology without syllables, but the reintroduction of this unit in phonology had been argued for by many phonologists, among others Vennemann (1972). Kahn (1976) had proposed a hierarchical representation of syllables and Liberman & Prince adopted this proposal without committing themselves to any particular view on the internal structure of these units. The metrical algorithm that Liberman & Prince introduced added to the syllabified string a layer of bisyllabic constituents, called feet. The resulting tree structure was augmented with the labels "Strong" and "Weak". The S label was assigned to syllables that contained a stressed vowel:

309

Metrical phonology (1)

a. b.

Every sequence of syllables +-, +— etc. forms a metrical tree (i.e. a foot). The feet are organized into a right branching tree: Μ

s

w

s

w

s

w

A +

pa

la

chi

CO

la

1

2

6

-

+ 3 7

-

4

+ 5 8 9

stress feature values

-

6

muscle; Old Japanese fazu 'tip ends of an archery bow where the string is attached' > Modern Japanese hazu 'expectation', and the often brilliant step by step accounts that show that the change is in fact fully motivated. The details of such changes have been taken to show that meaning change is particular to individual words and that there are few generalities to be made. This opinion is no doubt bolstered by such major studies of semantic change such as Breal (1964 [1900]), Bloomfield (1984 [1933], Chapter 24), Stern (1968 [1931]), Ullmann (1957, 1962) which attempted to develop taxonomies of semantic change, and which classified the major semantic changes in terms of disparate pairs like: I.

i. PEJORATION: the tendency to semanticize the more negative connotations of a word, e.g. terms for primates {ape, baboon, gorilla) > brutishness; birds (goose, cuckoo, pigeon, coot, turkey) > foolishness; scavenger birds (buzzard, vulture) > greed; early ME seely blessed, innocent' > late ME silly 'deserving of pity, feeble' > EModE 'ignorant, foolish'; OE stincan 'to smell (Vi)' > ME 'smell obnoxious', early ME 'to smell (Vi)' > later ME 'smell obnoxious', ii. AMELIORATION: the tendency to semanticize more positive connotations: Lat. nescius 'ignorant' > Old French ni(s)ce 'stupid'; borrowed into ME as 'stupid' > 'shy, bashful' > 'fastidious, refined' > 'pleasant, appealing'.

386 II.

III.

Elizabeth Closs Traugott i. RESTRICTION/NARROWING OF MEANING: OE deor 'animal' > 'deer', voyage 'journey' > 'journey by sea'. ii. EXPANSION/GENERALIZATION OF MEANING: Latin armare 'to cover one's shoulders' > arm, target 'small round shield' > 'mark to be shot at, goal'. i. METAPHOR: the mapping of one concept onto another, e.g. space > time as in before 'in front' > 'earlier'. ii. METONYMY/CONTAGION: the association of one meaning with another, e.g. the association of the formula ending the Catholic service, Ite, missa est 'Go now, the meeting is dismissed" > mass (the "meeting" itself), gray power.

Furthermore, research has started from different methodological perspectives, often without adequate consideration of their interdependence (but see Kronasser 1952; Geeraerts 1983). The most important methodologies are: I.

SEMASIOLOGY. Here the focus is on the question: "Given the formmeaning pair ("lexeme") L, what changes did the meaning Μ undergo", i.e. what polysemies do individual lexemes such as before, must, back (of) develop?

(1)

Form L -> Mx > M2 > M3

II.

(2)

III. (3)

ONOMASIOLOGY. This aims at answering the question: "Given a concept C, what lexemes can it be expressed by?", i.e. what are the semantic source domains for expressions of certain concepts such as LATER THAN, OBLIGATION, REAR? (By convention, specific lexemes are lower case, abstract concepts are in capitals). LI L2 L3

^ — ^

C

At a higher level of abstraction, "Given a concept C, what concept(s) is it likely to develop into?", e.g. SPACE > TIME. CI > C2 > C3

Such apparently contradictory taxonomies and disparate methodological approaches have until recently been repeated in most textbooks of historical linguistics that have in other respects taken great pains to incorporate current linguistic theory, and have led to such statements as: "in most

Semantic change

387

cases, semantic change is 'fuzzy', highly irregular, and extremely difficult to predict" (Hock 1991 [1986]). But in recent decades there has also been growing recognition that paths of change are very regular in certain semantic domains, for example relational nouns that refer to spatial relations, and verbs of motion, possession and existence, or modal auxiliary verbs. The domains least subject to regular change appear to be those expressed by non-relational nouns; here innovations are motivated primarily by changes in external referents (e.g. changes in the concept of'town', 'computer'), and only rarely by internal linguistic factors. Interest in regularity in semantic change has grown exponentially, largely tracking advances in functional semantics, pragmatics, and discourse analysis. In particular, attention has been directed toward two mechanisms for semantic change: (a) metaphor, (b) conceptual metonymy, or associations arising out of invited inferencing and use of generalized implicatures. There has also been increasing interest in a process known as subjectification. Metaphor, metonymy and subjectification will be discussed in sections 3-5, after a brief survey of work on semantic change in the first three quarters of the twentieth century. Some unresolved problems will be mentioned in Section 6 and a model for semantic change will be proposed in Section 7.

2. Catalysts for the study of regularity in semantic change A catalyst for early studies of semantic change was the notion of 'lexical field' and semantic taxonomies that developed out of structuralist interest in mutually constraining oppositions (see Ullmann 1964, p. 244). Most studies of semantic change prior to this period had been semasiological. Work in lexical fields shifted attention to conceptual categories. Among the first major studies of a lexical field was Trier's (1931) investigation of the development of terms for intellect in German. He showed that in Middle High German c. 1200 the field of INTELLECT was organized around three terms: wisheit 'courtly, chivalric attainments', list 'non-courtly attainments', and kunst *human wisdom in all its aspects, theological and mundane'. By c.1300, however, feudalism had collapsed, and list was acquiring the pejorative meanings that lead in Modern German to 'cunning, trick', and the field of the intellect was reorganized as wisheit 'religious, mystical experience', kunst 'art', and a new term wizzen 'knowledge'. Culture-specific as this change was, Trier's findings found less acceptance among linguists than the work of the psychologist, Stern, who argued that change within a lexical field could be unidirectional and apparently culture-independent. His example was RAPIDLY > IMMEDIATELY.

388

Elizabeth

Closs

Traugott

Stern's analysis is the precursor of many more recent ones that highlight the role of syntagmatic context in change. He claims that in a process he calls "permutation" and that would now be considered to be a case of the mechanism of inferential metonymy (see Section 4 below), RAPIDLY developed a polysemy of IMMEDIATELY in the context of perfective verbs "denoting the action as a unit" (Stern 1968 [1933], 185-91). Thus OE swifte rapidly > 1300 'immediately'. He argued that this change affected a large set of words (he lists 22) before 1300; words that came to mean 'immediately' or were borrowed with that meaning after 1400, e.g. fleetly (1598), rapidly (1727) did not undergo this change. Unfortunately the time-frames he gave were so small and questionable (dating within the Old English period is notoriously problematic) and the example seemed at the time so language-specific that his conclusions about unidirectionality and cultureindependence were not persuasive. Work on regularity in semantic change received a stronger impetus from anthropology, most especially Berlin & Kay's (1969) cross-linguistic study of color terms. Drawing on the notion of lexical field in linguistics and kinship taxonomies in anthropology, they sought to test the "Sapir-Whorf hypothesis" concerning the arbitrariness of the linguistic sign and of the relation of language to thought. Their conclusion was that the "basic" (oneword) color terms of all languages form a maximal set of eleven perceptual foci, and that there is a partially fixed order in which these foci develop in child language acquisition and in semantic change (Kay 1975, p. 257): [green] ->· [yellow] white '[red]

FOCI

[blue]

[brown]

black [yellow] ->· [green] STAGES I

II

Ilia Illb

IV

V

VI

' purple ' pink orange • grey . VII

They concluded that such paths of development were universal and motivated by non-linguistic perceptual factors, specifically visual perception, and therefore not arbitrary. The details of the hypothesis, which were subsequently challenged and revised, need not concern us here. For studies of semantic change the interest was primarily in the proposal that there were implicational hierarchies of development (see Greenberg 1966) such that one could predict that if a language had a term for BLUE, then it would also have terms for YELLOW and GREEN, and the term for BLUE would have developed later than either YELLOW or BLUE. Being testable

389

Semantic change

and predictive, the hypothesis seemed to have the potential of finding "laws" of semantic change comparable to those in phonology such as Grimm's Law.

3. The mechanism of metaphor Studies followed in rapid succession, largely within anthropological linguistics. These include Derrig (1978), who hypothesized with respect to hues that cross-linguistically WHITE was extended through the mechanism of metaphor > innocence (hence sometimes stupidity), BLACK > evil, ignorance and gloominess, RED > anger and sex, BLUE/GREEN > inexperience and being uneducated, YELLOW > ripeness. She also found with respect to intensity that BRIGHT was associated with understanding, LIGHT with intelligence, DARK with ignorance and opacity, and CLEAR with alertness. The latter findings were connected with work on synaesthesia, the metaphorical extension of terms for the five senses onto other senses and onto social behaviors, language, and so forth (Williams 1976). This phenomenon had already been noted by Breal, who gives as examples of metaphor the mapping of sight onto hearing, e.g. bright note and of taste onto language, e.g. bitter reproach, etc. (Breal 1964 [1900], 130). Probably the most widely studied lexical domain was that of body part terms. For example, using a sample of 118 languages, Brown & Witkowski (1983) undertook an investigation of the relationship between EYE, FACE, SEED, FRUIT. They suggested that in "small-scale societies" EYE has a cultural importance or "salience" far greater than FACE and that forms for the former often have polysemies for the latter, but not vice versa. In languages associated with these societies, EYE is "unmarked" in the sense that it is usually the simple lexical form, whereas terms for FACE may be derived by compounding or derivation from EYE. In such societies "eye and seed are in a sense the centre or core of face and fruit respectively, while face and fruit comprise the periphery of eye and seed. Thus formally speaking, eye and seed are to face and fruit as centre is to periphery or 'figure' to 'ground'" (Brown & Witkowski 1983, p. 76). Furthermore, EYE is sometimes extended to SEED and even FRUIT. In other words, it is not only individual lexical domains that have similar directionalities, but parallel ones may have them as well, given extra-linguistic criteria such as cultural importance. As societies become "large-scale" and urban, they suggested, cultural importance may shift; indeed, FACE and FRUIT tend in such societies to come to be equally or more important than EYE and SEED. The main significance of such studies for semantic change was evidence for unidirectionality, hence predictability.

390

Elizabeth Closs Traugott

Within linguistics, most early work on regularity in semantic change concerned domains similar to those studied earlier by anthropologists and psychologists, most notably space > time (e.g. Traugott 1978), and body part > space (e.g. Kahr 1975), as well as in new domains such as temporal > concessive (Traugott 1982). Like earlier work, it tended to assume that the major mechanism for change was metaphor, construed as the analogical mapping of a more concrete term from a "source" domain onto a more abstract term in the "target" domain. The development of cognitive linguistics (e.g., Lakoff & Johnson 1980 and, from a different perspective, Langacker 1987, 1991) with its focus on the ubiquitous basis of metaphor in experience of the body, and especially in abstract and topological imageschemata based on it, led to the refining of two kinds of analysis, diachronic prototype semantics (Geeraerts 1997), and diachronic metaphorical analysis. Since metaphor involves conceptualizing one area of experience in terms of another and operates "between domains" (Sweetser 1990,19, italics original), processes motivated by it were conceptualized as primarily discontinuous. Since source and target meanings of metaphors constrain each other experientially, it was hypothesized that aspects of the abstract image-schemata are "preserved across metaphorical mappings" (ibid. 59). In a now classic study of the lexical field of intellect, reconceived as involving a mind-as-body metaphor motivated by experience, e.g. vision/tactile acts > intellect, Sweetser showed how verbs of seeing and grasping can come cross-linguistically to be metaphorized as verbs of understanding. She also suggested that in more abstract domains, such as that of modality, there is metaphorical mapping across domains both synchronically and diachronically (Sweetser 1990, Chap. 3). Thus may can be said to operate in three domains: (4)

I. sociophysical world ("content") Kim may go ("content modal of permission") II. mental world (reasoning) Kim may be tired ("epistemic modal") III. world of speaking (speech acting) Kim may be a nice guy, but I don't trust him ("speech-act modal")

Here the image-schemata involved are barriers (Sweetser 1990, p. 61, 70): (5)

I. content: II. reasoning: III. speech acting:

X is not barred by some authority from doing Y I am not barred by my premises from the conclusion Y I do not bar from our conversational world the statement Y

391

Semantic change

Historical evidence for the unidirectional change of content modals > epistemic modals (not epistemic modals > content modals) had already been established (Shepherd 1981; Traugott 1989). The question remained whether evidence for a further change from epistemic into speech act modalities could be found. While evidence in the domain of modal auxiliaries has remained elusive, perhaps in part because of the nature of the data, there is abundant support for the hypothesis in the rise of speech act uses of locutionary verbs (Traugott 1987) and of metatextual discourse markers (see Section 5 below). Meanwhile extensive work was being done on the role of semantic change in grammaticalization. The connection was an obvious one, since crosslinguistically many body part and spatial terms give rise to grammatical prepositions, case markers, complementizers, tense, aspect and modality markers (see e.g., Givon 1975; Fleischman 1982; Lehmann 1995 [1982]; Heine, Claudi & Hünnemeyer 1991; Hopper & Traugott 1993; Svorou 1993; Bybee, Perkins & Pagliuca 1994 and, for a lexicon of grammaticalization in African languages, Heine, Güldemann, Kilian-Hatz, Lessau, Roberg, Schladt & Stolz 1993).

4. The mechanism of metonymy Work on grammaticalization in the US, which itself had grown out of research in syntactic change and analysis of discourse pragmatics (e. g. Givon 1979, papers in Li 1975, 1977), in turn led to historical studies of semantic change of lexemes and constructions not in isolation or paradigmatic relation with other members of a field, but within larger associative and syntagmatic contexts. Much of this work has in recent years been done employing electronic data bases such as the Helsinki Corpus of the History of English (e.g. Rissanen, Kytö & Palander-Collin 1993). Metonymy had for a long time been the poor sister of metaphor. Of metonymy, Ullmann, for example, had said that it is "less interesting than metaphor since it does not discover new relations but arises between words already related to each other ... but... it is an important factor in semantic change" (Ullmann 1962, p. 218). However, Stern had seen its interest for semantic change and had written of the association of mental state with its object or cause, e.g. concern (n.) 'interest > matter that concerns' (Stern 1968 [1931], p. 376). Reenvisioned as a conceptual mechanism operating on pragmatic implicatures and inferences, metonymy provides as rich an explanation as metaphor for semantic change, and in fact in many cases a richer one. Grice's very brief and hesitant hypothesis: "it may not be impossible for what starts life, so to speak, as a conversational implicature

392

Elizabeth Closs Traugott

to become conventionalized" (1989 [1975], p. 39) has been strongly confirmed by work in the last decade. Consider, for example, the relationship of part to whole (known as synecdoche), which is usually considered to be a typical case of metonymy, e.g. keel for ship. In semantic change, it has often been noted that a term for a part will become a term for a whole, but not vice versa, e.g. FINGERNAIL > FINGER > HAND in Australian aboriginal languages. This can be seen as motivated by the unidirectionality of entailments in part-whole relationships: "'part', by definition, entails some idea of 'whole', but a 'whole' entails no notion of'part'" (Wilkins 1996, pp. 275-282). It is not only entailments, but also weaker, defeasible implicatures that can motivate change. That is, suggestions made over and above the semantics of the utterance, motivated especially by the heuristic "Say no more than you must, and mean more thereby" (Horn 1984; Traugott 1989; Traugott & König 1991; Levinson 1995). This heuristic is based on Grice's "Quantity2": "Do not make your contribution more informative than is required". For example, in the modal domain there can be an implicature (represented by +>) from You may go in the permission sense +> 'it is possible you will go' (those who are given permission are likely but not certain to act on the permission). By contrast, there is a stronger implicature from obligation of the type expressed by You must go +> 'it is probable that you will go'. Metaphor and metonymy in their current senses give us two alternative accounts of these shifts from the modality of obligation (called "deontic", or, in Sweetser's terms "content") > the modality of possibility and conclusion ("epistemic"). The first account conceptualizes the change in terms of mapping from one domain to another in terms of factors such as barriers, the second in terms of association of pragmatic meanings with a form in the flow of speech. Neither approach excludes the other: easily comprehended metaphors are consistent with typical associations; both exploit pragmatic meaning. Nevertheless, the two accounts highlight significantly different aspects of language. Metaphors are iconic and analogical; they provide choices within constructions and are therefore paradigmatic. By contrast, implicational metonymies are indexical, associative across constructions, and therefore syntagmatic (Anttila 1989 [1972]).

393

Semantic change •>· Metonymy (syntagmatic, indexical)

•

Metaphore (paradigmatic, iconic)

5.

Subjectification

In the last decade a third type of change has come to be of central interest: subjectification. Although already mentioned in Breal's work (1964 [1900], Chapter 25), subjectivity and subjectification did not come to be of major topic of research until after Benveniste's (1971) work on subjectivity and the distinction between the syntactic subject/'sujet d'enonce" and the "speaking subject'Vsujet d'enonciation" (see also Lyons 1982; Langacker 1990). To take a standard kind of example, the syntactic (surface) subject of Lightning is going to strike is lightning·, the speaking subject is the speaker who is expressing his or her belief about the likelihood that lightning will strike. The historical mechanism of subjectification has been interpreted by Langacker (1990) as the realignment of the viewing arrangement in the speaker's construal of a "scene". One of the best known examples is the development of the motion verb be going to 'be going with the purpose of as in Mary is going to check out the prices > a marker of futurity that can occur in a "raising" construction such as There is going to be a storm, and that can have the form gonna. Of this change Langacker says that "spatial motion on the part of an objectively-construed participant [the subjectECT] is replaced by subjective motion (mental scanning) on the part of the conceptualized (1990, 19). More specifically, the subject which is an argument of the event-structure and is profiled "on-stage" as an intentional agent, is considered to be realigned with the implicit "off-stage" speaker. The focus here is on conceptual content and construals imposed on that content, as expressed at the clause level in constructed examples largely discussed out of context. However, subjectivity is not limited to the domain of the syntactic subject. Indeed, all language is subjective; we cannot say or write anything without expressing our perspective or point of view, "what has been called a speaker's imprint" (Finegan 1995, p. 1). A different interpretation of subjectification, which is based in discourse analysis of (written) data

394

Elizabeth Closs Traugott

embedded in the linguistic context of the text, has been developed in Traugott (1982, 1995, et passim). On this view, subjectification is a phenomenon affecting large domains of the lexicon. If the meaning of a lexical item or construction is grounded in the world of reference, it is likely that over time speakers will develop polysemies grounded in their world, whether reasoning, belief, or metatextual attitude to the discourse. In other words, subjectification is a semasiological development of meanings associated with a meaning-form pair such that the latter comes to mark subjectivity explicitly. While the mechanism of implicational metonymy indexes implicatures, subjectification indexes the relation of these metonymies to speaker attitude, intention, etc. Among examples of subjectification is the development over time of more speaker-based meanings in modal domains. It is generally held that epistemic modals are more subjective than deontic ones (see Lyons 1977; Coates 1983). From a historical perspective, in those languages where polysemous deontic and epistemic modals exist (e.g. English, German, Romance languages, Chinese) the epistemic meaning arises later, as mentioned in Section 3 above. But subjectification is not restricted to this shift in meaning: it occurs within the deontic and epistemic domains themselves, something which cannot be readily accounted for from Langacker's perspective. When epistemic modals develop, they generally do so with only weakly subjective meanings of the type "anyone could conclude". Only later do they develop strongly subjective ones of the type "I conclude", e.g. although epistemic, future-oriented meanings of will can be found in OE (Denison 1993, p. 299), it is not until the mid nineteenth century that we find examples like This will be your luggage, I suppose (1847 Bronte). Other examples of subjectification include the shift from imperative, addressee-oriented let us as in let us go, will you > "hortative" let us go, shall we in the 14thC (Traugott 1995); from sort of' type of' as in I knowe that sorte of men ryght well (1560 Daus) > degree modifier sort of as in It sort o' stirs one up to hear about old times (1833 Hall) in the late 18thC (Tabor 1993). Pejoration and amelioration (see Section 1 above) can be considered to be types of subjectification, as can Stern's example of RAPIDLY > IMMEDIATELY, since IMMEDIATELY is deictic while RAPIDLY is not (see 2.). Particularly interesting changes can be observed in the domain of the development of discourse markers of the type actually, in fact, indeed, anyway, after all (Powell 1992; Traugott In press). Semantically they undergo a shift from adverb of manner and "respect to which" > epistemic modal stance adverb > discourse marker expressing speaker attitude to the relationships between the preceding discourse and what follows, i.e. serving

395

Semantic change

a metatextual function. Syntactically they typically involve a shift from a clause internal adverb (manner, respect) > a sentential adverb > an extraclausal adverb with discourse marker function (the three types of adverb can be abbreviated as VAdv, SAdv, and EAdv respectively). An example is the development of in fact as a adverb of respect meaning 'in practice, in reality' as in But it is evident in fact and experience that there is no such universal Judge (1671 Tillotson, Sermons) > a modal adverb meaning 'really, actually as in Every particle eludes the grasp by a new fraction ... when we endeavour to seize it. But as in fact there must be something which terminates the idea of every finite quantity (1739-40 Hume, Treatise Human Nature) > a discourse marker expressing justification as in "I was wrong," he continued, "in talking of its being broke to you. I should not have used the expression. In fact, it does not concern you — it concerns only myself {1815 Austen, Emma). Similar changes are found in other adverbs: Table 1. Development of discourse markers

actually in fact indeed anyway after all

manner VAdv

epistemic SAdv

metatextual EAdv

15th C 16th C 14th C 11th C 16th C

17th C 18th C 16th C 17th C 17th C

19th C 19th C 17th C 19th C 17th C

An account of such changes from the point of view of metaphor could appeal to three worlds in the ways Sweetser (1990) proposed for modal auxiliaries. On this view, the manner adverb refers to the socio-physical world, the epistemic adverb to the mental world of reasoning, and the discourse marker invokes the speech act world, and the developmental sequence is precisely the one that Sweetser predicted (see Section 3 above). In the case of actually, in fact, indeed, (a) a done thing is evident in the socio-physical world > (b) a done thing is evident in the mental world > (c) a said thing is evident in the world of conversation. However, such an account leaves much underspecified, most particularly the fact that at the modal adverb stage, the adverbs under discussion are adversative, and express rejection of a point of view expressed in a prior utterance by an interlocutor (or implied in the context), while at the metatextual discourse marker stage they express the speaker's perspective on what follows, e.g. that it is a better formulation of what preceded, that it is more precise, etc. An account from the point of view of metonymy can

396

Elizabeth Closs Traugott

specify some of the motivations for these meaning developments: (a) a done thing is evident and by implication highly likely or true > (b) a declarative proposition specially marked as highly likely or true implicates that the speaker believes this position is not expected > (c) an utterance introduced as true despite expectations invites the inference that the speaker believes it is a better, more precise formulation. However, an account from metonymic invited inferencing alone does not motivate the particular set of inferences that repeatedly operate in the case of these and other adverbs, and the variety of examples considered above. By itself neither it nor the metaphor account motivates the shift to speaker-based meanings. It is subjectification that accounts for this directionality.

6. Some unresolved

questions

What motivates the operation of metaphor, metonymy and subjectification with respect to a particular form-meaning pair? Although the answer is not well understood, it seems reasonable to hypothesize that two of the three heuristics of language proposed in Levinson (1995) are the prime motivators. They are the R(elevance)-heuristic: "Say no more than you must, and mean more thereby" which leads to rich interpretations, and the M(anner)-heuristic: "Avoid prolixity", which leads to the interpretation that specially marked, periphrastic expressions warn "marked situation" (these heuristics have their origin in Grice's Quantity 2 cited in Section 4 above, and Manner). Levinson's third heuristic, the Q(uantity)-heuristic, impedes change, since it is of the form: "Say as much as is needed for the occasion", which pairs with "What is not said is not the case" (see Grice's "Quantity 1": "Make your contribution as informative as is required (for the current purposes of the exchange)"). Another question of particular concern is exactly what the status of subjectification is. It can perhaps best be regarded as a specific type of metonymy, one which is associated with and indexical of the speech situation. The motivation for subjectification is that in negotiating interaction speakers strive to signal the discourse management underway, and actively draw on salient implicatures in the process to make their attitudes to the speech situation explicit. This account of subjectification as a diachronic mechanism motivated by the intersubjectivity and ultimately the subjectivity of the speech event gives insight into many of the unidirectionalities of meaning change and into the dominance of subjectification. However, not all semantic change is a case of subjectification (e.g. BODY PART > SPACE), and the constraints on its operation remain to be investigated. One of the puzzles of semantic change has been the fact that old form-

Semantic change

397

meaning pairs are reused with new meanings. From an information-theory perspective, this would hardly seem to lead to improved communication. However, from the point of view of negotiated interaction, it can be seen to serve useful purposes. It invites addressees to enrich what is said and at the same time appears to allow them options of interpretation at least in those contexts where there is ambiguity. Furthermore, it balances the competing pressures on the linguistic system to provide finite means for expressing the multiplicity, indeed infinity, of possible meanings that speakers wish to convey. In this connection, it is of interest in the data for semantic change that new meanings tend to occur in heavily redundant contexts. Indeed, they often continue to be used in such contexts. Redundancy reduces the amount of work the addressee has to do to interpret the speaker's meaning. At the same time it violates the M-heuristic — by being redundant speakers draw attention to their "expressive" intentions. As the discussion has implied, metaphor, metonymy, and subjectification accounts of semantic change are in many practitioners' view speaker- and production- based, and the speakers are considered to be adults as well as children. Speakers are thought of as exploiters of meaning for purposes of communication with addressees. Proponents of metaphor write of speakers using metaphor to problem-solve (e.g. Heine, Claudi & Hünnemeyer 1991). Proponents of metonymy write of "invited inferences", a term that highlights the speaker's active role ("inviting") as well as the hearer's ("inferencing") (e.g. Traugott 1989). In all theories of language change where change is seen as (at least) the discontinuity between interlocutors in the speech event, both speakers producing output and addressees processing input are considered to play a role. There are significant differences, however, in who is considered to do the major work in change. Much theorizing on morphosyntactic and phonological change, by contrast with that in semantic change, has been conceptualized in terms of a model in which children do most of the work, processing input. Drawing on common-sense reasoning ("abduction") and universal principles, children are said to be able to derive a system which is different in certain respects from that of an older generation (see Lightfoot 1991 for extensive discussion of this type of model of language change). However, recent work on language acquisition (e.g. Slobin 1994) and on language change in general (e.g. Hopper & Traugott 1993; Keller 1994) has questioned the significance of the role of children in change, given that children typically do not and in earlier centuries did not have the kind of prestige that motivates the adoption of forms they use. In the case of semantics and pragmatics, small children do not have the kinds of pragmatic skills relevant to semantic change, and acquire them only "through a prolonged developmental process of conversational inferencing"

398

Elizabeth Closs Traugott

(Slobin 1994, 130). Young and older adults, who already have a grammar and negotiate meanings in mutual interaction, would seem to have greater potential to be sources from whom innovations spread, but this needs to be confirmed by studies of language change in progress.

7. Toward a model of semantic

change

In thinking about meaning change, and especially about metonymyinvoking invited inferences it is useful to build on Levinson (1995) and distinguish three levels of meaning relevant to a lexeme: I.

II.

III.

Semantics (coded meaning); this is a convention of language, e.g. in PDE after means 'at a time later than', since means both 'from the time that' and ^cause', i.e. since is semantically polysemous. Utterance-type meaning (generalized conversational implicatures); these are preferred meanings, and conventions of use, but may be canceled, e.g. in PDE after can give rise to an implicature of causality. Thus After the trip to Minnesota she felt very tired +> 'because of the trip she felt very tired'. However, causality is not conventionally associated with after, and is easily canceled: After the trip to Minnesota she felt very tired. It turned out she had been sick for quite some time. Such utterance-type meanings may be pragmatically ambiguous (see Horn 1985; Sweetser 1990), but not semantically ambiguous. Utterance-token meaning (conversational implicatures, that is, meanings arising in context "on the fly").

Historically there is a path from coded meanings to utterance-token meanings to utterance-type meanings to new (and polysemous) coded meanings. Often, utterance-token meanings that are nonce-meanings, or individual, idiosyncratic uses, are introduced redundantly in linguistic contexts so that they can be understood. They are unlikely to survive and play any part in a particular change unless they become salient in a community and a social value is attached to them (e.g. the use of like as an indirect quotation marker (Romaine & Lange 1991)). But after extended use in the conventionalized meaning, they may become semanticized, that is, they may become a well-recognized polysemy, a new coded meaning related to the former one. Of course, at any moment in time, the meaning Μ of a lexeme L is a member of a conceptual domain C. This relationship can be expressed as (ignoring Form):

399

Semantic change (6)

L

Μ I

c

At any moment in time, too, metaphors and metonymies exist in the language, and subjectivities abound. As changes begin to develop, they provide the context and frame of reference within which the mechanisms of metaphor and metonymy, including subjectification, operate to enable reanalysis of meaning. The new polysemy typically serves as a new member of an already extant meaning-type, in other words, it contributes onomasiologically to a semantic domain. For example, if a deontic modal verb like must acquires an epistemic meaning, this is an innovation with respect to that verb, possibly even to the category of verbs, but probably not with respect to the semantics of the language, since epistemicity can be expressed by adverbs, etc. Polysemies may survive for a very long time, and typically additional ones accrete over the centuries. Sometimes an older (or even a newer) one will disappear , e.g. the older sense of agan "have' gave way to the newer one 'owe, have a debt to pay'. Sometimes a polysemy will cease to be understood as semantically connected to its co-polysemes, and split off as a homonym, e.g. ought to is no longer associated with either agan 'have' or 'owe'; still 'motionless' is no longer associated with still 'temporal'. But in principle most older and newer meanings will coexist, although not necessarily in the same relationship to each other as they had over time (Michaelis 1993). The development of a particular form-meaning pair and its relationship to more general conceptual domains can be modeled as in Figure 3:

400

Elizabeth Closs Traugott Utterance-type meaning conventionalizing of implicatures

SP/AD constraints on weighting of implicatures (preferred uses, saliency, relevance, subjectivity, etc.) Utterance-token meaning SP exploits implicatures innovatively in associative stream of speech

Stage I. Coded meaning

L ->

Ml I CI

Stage II. New coded meaning (semanticization) L

Ml

i CI

M2

4 c2

While much remains to be understood about semantic change, work on pragmatics and discourse analysis gives us reason to believe that semantic change not only can be regularly recurring, but indeed must necessarily recur if synchronic processes of inferencing and of strategic interaction are replicated from generation to generation.

Abbreviations C = Concept/conceptual domain; EModE = Early Modern English (c. 1500-1750); L = Lexeme; Μ = Meaning/sememe; ME = Middle English (c. 1150-1500); OE = Old English (c. 600-1150); PDE = Present Day English; means "is realized as"; > means "changes to"; +> means "implicates"; 4- means "is linked by membership to a set"

Acknowledgments Thanks to Richard Dasher and Scott Schwenter for discussion of regularity in semantic change.

401

Semantic change

A Semantic Change

Bibliography

Akatsuka, N. (1997). Negative conditionality, subjectification, and conditional reasoning. On conditionals again, edited by A. Athenasiadou & R. Dirven, 323-354. Amsterdam: John Benjamins. Allen, Κ. & K. Burridge (1991). Euphemism and dysphemism: language used as a shield and weapon. Oxford: Oxford University Press. Anttila, R. (1989 [1972]). Historical and comparative linguistics. Amsterdam: John Benjamins. [2nd ed., Chapter 7.] Benveniste, E. (1971). On subjectivity in language. Problems in general linguistics, 224-246. Coral Gables, Flor.: University of Miami Press (trans. M.E. Meek). [Originally published as: De la subjectivity dans le langage.Problemes deLinguistique Generale, 258-285. Paris: Gallimard, 1958.] Berlin, B. & P. Kay. (1969). Basic color terms: their universality and evolution. Berkeley: University of California Press. Blank, A. (1997). Prinzipien des lexikalischen Bedeutungswandels am Beispiel der romanischen Sprachen. Tübingen: Niemeyer. Bloomfield, L. (1984 [1933]). Language. Chicago: University of Chicago Press. Breal, M. (1964 [1900]). Semantics; studies in the science of meaning. [Translated by H. Cust.] New York: Dover. Brinton, L.J. (1988). The development of English aspectual systems. Cambridge University Press. Brinton, L.J. (1996). Pragmatic markers in English: grammaticalization and discourse function. Berlin: Mouton de Gruyter. Brown, C.H. & S.R. Witkowski. (1983). Polysemy, lexical change and cultural importance. Man 18, 72-89. Bybee, J. L., W. Pagliuca & R. Perkins (1991). Back to the future. Approaches to grammaticalization, edited by Traugott & Heine, Vol. 2,17-58. Bybee, J., R. Perkins & W. Pagliuca (1994). The evolution of grammar: tense, aspect, and modality in the languages of the world. Chicago: University of Chicago Press. Coates, J. (1983). The semantics of the modal auxiliaries. London: Croom Helm. Dasher, R. (1995). Grammaticalization in the system of Japanese predicate honorifics. Doctoral dissertation, Stanford University. Denison, D. (1993). English historical syntax. Oxford: Oxford University Press. Derrig, S. (1978). Metaphor in the color lexicon. Chicago Linguistic Society: Papers from the Parasession on the Lexicon, 85-96. Chicago: Chicago Linguistic Society.

402

Elizabeth Closs Traugott

Ducrot, O. (1983). Operateurs argumentatifs et visee argumentative. Cahiers de Linguistique Franqaise 5, 7-36. Finegan, E. (1995). Subjectivity and subjectivisation: An introduction. Subjectivity and subjectivisation in language, edited by Stein & Wright, 1-15. Fisiak, J., editor (1985). Historical semantics, historical word-formation. Berlin: Mouton de Gruyter. Fleischman, S. (1982). The future in thought and language. Cambridge University Press. Fleischman, S. (1992). Discourse and diachrony: the rise and fall of Old French SI. Internal and external factors in syntactic change, edited by M. Gerritsen & D. Stein, 433-473. Berlin: Mouton de Gruyter. Gamon, D. (1994). On the development of epistemicity in the German modal verbs mögen and müssen. Folia Linguistica Historica 14,125-76. Geeraerts, D. (1983). Reclassifying semantic change. Quadierni di Semantica 4, 217-240. Geeraerts, D. (1997). Diachronie prototype semantics: a contribution to historical lexicology. Oxford: Clarendon Press. Givon, T. (1975). Serial verbs and syntactic change: Niger-Congo. Word order and word order change, edited by Li, 47-112. Givon, T. (1979). On understanding grammar. New York: Academic Press. Geis, Μ. L. & A. M. Zwicky (1971). On invited inferences. Linguistic Inquiry 11, 561-566. Greenberg, J.H. (1966). Some universale of grammar with particular reference to the order of meaningful elements. Universals of language, edited by J.H. Greenberg, 73-113. Cambridge, Mass.: MIT Press. [2nd ed.] Greenberg, J.H. (1978). How does a language acquire gender markers? Universals of human language, edited by Greenberg et al., Vol. 3, 249-295. Greenberg, J.H., C.A. Ferguson & E. Moravcsik, editors. Universals of human language, Stanford: Stanford University Press. [3 Volumes.] Grice, P. (1989 [1975]). Logic and conversation. Studies in the way of words, 22-40. Cambridge, Mass.: Harvard University Press [Originally published in Syntax and Semantics 3. Speech Acts, edited by P. Cole & J. L. Morgan, 41-58. New York: Academic Press.]. Hanson, K. (1987). On subjectivity and the history of epistemic expressions in English. Chicago Linguistic Society 15, 23-52. Heine, B., U. Claudi & F. Hünnemeyer. (1991). Grammaticalization: A conceptual framework. Chicago: University of Chicago Press. Heine, Β., T. Güldemann, C. Kilian-Hatz, D. Lessau, H. Roberg, M. Schladt & Τ. Stolz (1993). Conceptual shift: a lexicon of grammaticalization

Semantic change

403

processes in African languages. Afrikanistische Arbeitspapiere 34/35. University of Cologne. Hock, H.H. (1991 [1986]). Principles of historical linguistics. Berlin: Mouton de Gruyter. [2nd ed.] Hopper, P.J. & E.C. Traugott (1993). Grammaticalization. Cambridge University Press. Horn, L.R. (1984). Toward a new taxonomy for pragmatic inference: Q-based and R-based implicature. Meaning, form, and use in context: linguistic applications, edited by D. Schiffrin, GURT 84, 11-42. Washington, D. C.: Georgetown University Press. Horn, L.R. (1985). Negation and pragmatic ambiguity. Language 61, 121-174. Jucker, Α., editor (1995). Historical pragmatics. Amsterdam: John Benjamins. Kahr, J. C. (1975). Adpositions and locationals: typology and diachronic development. Working Papers on Language Universals 19, 21-54. Stanford University. Kay, P. (1975). Synchronic variability and diachronic change in basic color terms. Language in Society 4, 257—270. Keller, R. (1994). On language change: The invisible hand in language. Translated by B. Nerlich. London: Routledge. Kemmer, S. (1993). The middle voice. Amsterdam: John Benjamins. Kleparski, G.A. (1986). Semantic change and compositional analysis: An inquiry into pejorative developments in English. Regensburg: F. Pustet. König, Ε. (1991). The meaning of focus particles: A comparative perspective. London: Routledge. Kronasser, H. (1952). Handbuch der Semasiologie. Kurze Einführung in die Geschichte, Problematik und Terminologie der Bedeutungslehre. Heidelberg: Winter. Kroon, C.H. C. (1995). Discourse particles in Latin: Α study of nam, enim, gutem, vero and at. [Amsterdam Studies in Classical Philology 4.] Amsterdam: J. C. Gieben. Kytö, M. (1991). Variation and diachrony, with Early American English in focus. Frankfurt-am-Main: Peter Lang. Lakoff, G. & M. Johnson (1980). Metaphors we live by. Chicago: University of Chicago Press. Langacker, R.W. (1987, 1991). Foundations of cognitive grammar. Stanford, Cal.: Stanford University Press. Langacker, R.W. (1990). Subjectification. Cognitive Linguistics 1, 5-38. Lehmann, C. (1995 [1982]). Thoughts on Grammaticalization. Munich: Lincom Europa. [First published as Thoughts on Grammaticalization. A

404

Elizabeth Closs Traugott

Programmatic Sketch. I. Arbeiten des Kölner Universalien-Projekts 48. Köln: Universität zu Köln. Institut für Sprachwissenschaft, 1982.] Lehrer, A. (1985). The influence of semantic fields on semantic change. Historical semantics, historical word-formation, edited by Fisiak, 283-296. Lehti-Eklund, H. (1990). Fran Adverb till markör i text. Studier i Semantisk-syntaktisk Utveckling i Äldre Svenska. Helsingfors: Humanistika Avhandlingar. Levinson, S.C. (1995). Three levels of meaning. Grammar and meaning: Essays in honor of Sir John Lyons, edited by F. R. Palmer, 90-115. Cambridge University Press. Li, C. N., editor (1975). Word order and word order change. Austin: University of Texas Press. Li, C. N., editor (1977). Mechanisms of syntactic change. Austin: University of Texas Press. Lightfoot, D. (1991). How to set parameters: Arguments from language change. Cambridge, Mass.: The MIT Press. Lord, C. (1993). Historical change in serial verb constructions. Amsterdam: John Benjamins. Lüdtke, H. (1985). Diachronic irreversibility in word-formation and semantics. Historical semantics, historical word-formation, edited by Fisiak, 355-366. Lyons, J. (1977). Semantics. Cambridge University Press. [2 Volumes.] Lyons, J. (1982). Deixis and subjectivity. Loquor, ergo sum? Speech, place, and action: Studies in deixis and related topics, edited by R. J. Jarvella & W. Klein, 101-24. New York: Wiley. Matsumoto, Y. (1988). From bound grammatical markers to free discourse markers: History of some Japanese connectives. Proceedings of the Fourteenth Annual Meeting of the Berkeley Linguistics Society, edited by S. Axmeyer, A. Jaisser, & H. Singmaster, 340-351. MacMahon, A.M.S. (1994). Understanding language change. Cambridge University Press. Michaelis, L A. (1993). 'Continuity' within three scalar models: The polysemy of adverbial still. Journal of Semantics 10, 193-237. Mori, J. (1996). Historical change of the Japanese connective datte: its form and functions. Japanese I Korean Linguistics 5, edited by A. Akatsuka, S. Iwasaki, & S. Strauss, 201-218. Nordlinger, R. & E. C. Traugott (1997). Scope and the development of epistemic modality: evidence from ought to. English Language and Linguistics 1, 295-317. Ohori, T., editor (1998). Studies in Japanese grammaticalization: cognitive and discourse perspectives. Tokyo: Kurosio Publ.

Semantic change

405

Onodera, N. 0. (1995). Diachronic analysis of Japanese discourse markers. Historical pragmatics, edited by Jucker, 393-437. Powell, M. J. (1992). The systematic development of correlated interpersonal and metalinguistic uses in stance adverbs. Cognitive Linguistics 3, 75-110. Rissanen, Μ., M. Kytö & Μ. Palander-Collin, editors (1993). Early English in the computer age: Explorations through the Helsinki Corpus. Berlin: Mouton de Gruyter. Romaine, S. & D. Lange (1991). The use of like as a marker of reported speech and thought: A case of grammaticalization in progress. American Speech 66, 227-279. Schwenter, S. A. (1994). "Hot news" and the grammaticalization of perfects. Linguistics 32, 995-1028. Shepherd, S. (1981). From deontic to epistemic: An analysis of modals in the history of English, Creoles, and language acquisition. Papers from the 5th International Conference on Historical Linguistics, edited by A Ahlqvist, 316-123. Amsterdam: John Benjamins. Slobin, D I. (1994). Talking perfectly: Discourse origins of the present perfect. Perspectives on Grammaticalization, edited by W. Pagliuca 119-133. Amsterdam: John Benjamins. Stein, D. & S. Wright, editors (1995). Subjectivity and subjectivisation in language. Cambridge University Press. Stern, G. (1968[1931]). Meaning and change of meaning, with special reference to the English language. Bloomington: Indiana University Press. Sun, C. (1997). Aspectual categories that overlap: a historical and dialectal perspective of the Chinese zhe. Journal of East Asian Linguistics 7, 153-174. Svorou, S. (1993). The Grammar of space. Amsterdam: John Benjamins. Swan, T. (1988). Sentence adverbials in English: a synchronic and diachronic investigation. Oslo: Novus. Sweetser, E.E. (1990). From etymology to pragmatics: Metaphorical and cultural aspects of semantic structure. Cambridge University Press. Tabor, W. (1993). The gradual development of degree modifier sort of and kind of: A corpus proximity model. Chicago Linguistic Society 29, 451—465. Traugott, E.C. (1978). On the expression of spatio-temporal relations in language. Universals of human language, edited by Greenberg et al., Vol 3, 369-400. Traugott, E.C. (1982). From propositional to textual and expressive meanings; Some semantic-pragmatic aspects of grammaticalization. Perspec-

406

Elizabeth Closs Traugott

tives on historical linguistics, edited by W.P. Lehmann & Y. Malkiel, 245-271. Amsterdam: John Benjamins. Traugott, E.C. (1987). Literacy and language change: the special case of speech act verbs. Language, literacy, and culture: Issues of society and schooling, edited by J. Langer, 11-27. Norwood: Ablex. Traugott, E.C. (1989). On the rise of epistemic meanings in English: An example of subjectification in semantic change, Language 57, 33-65. Traugott, E.C. (1995). Subjectification in grammaticalization. Subjectivity and subjectivisation in language edited by Stein & Wright, 31-54. Traugott, Ε. C. (in press). Constructions in grammaticalization. A handbook of historical linguistics, edited by R.D. Janda & B.D. Joseph. Oxford: Blackwell. Traugott, Ε. C. & R. Dasher (in progress). Regularity in semantic change. Traugott, E.C. & B. Heine, editors (1991). Approaches to grammaticalization. Amsterdam: John Benjamins. [2 Volumes.] Traugott, Ε. C. & Ε. König. (1991). The semantics-pragmatics of grammaticalization revisited. Approaches to grammaticalization, edited by Traugott & Heine, Vol. 2,189-218. Trier, J. (1931). Der deutsche Wortschatz im Sinnbezirk des Verstandes. Heidelberg: Winter. Ulimann, S. (1957). The principles of semantics. Oxford: Blackwell. [2nd ed.] Ullmann, S. (1962). Semantics: An introduction to the science of meaning. Oxford: Basil Blackwell. Verhagen, A. (1995). Subjectification, syntax, and communication. Subjectivity and subjectivisation in language edited by Stein & Wright, 103-128. Warren, B. (1992). Sense developments: A contrastive study of the developments of slang senses and novel standard senses in English. Stockholm: Almqvist & Wiksell. Williams, J.M. (1976). Synaesthetic adjectives; A possible law of semantic change. Language 52, 461-78. Wilkins D. P. (1996). Natural tendencies of semantic change and the search for cognates. The comparative method reviewed: Regularity and irregularity in language change, edited by M. Durie & M. Ross, 264-304. Oxford: Oxford University Press.

Affiliations and addresses

Geert Booij HIL/Vrije Universiteit Amsterdam I Department of Linguistics I De Boelelaan 1105 I 1081 HV Amsterdam I The Netherlands [email protected] Luigi Burzio Department of Cognitive Science I Johns Hopkins University I Baltimore, MD 21218-2685 I U.S.A. [email protected] Tom Cornell University of Tübingen I Seminar für Sprachwissenschaft I Kleine Wilhelmstrasse 113 I 72074 Tübingen I Germany cornell@sfs. nphil. uni-tuebingen .de San Duanmu Program in Linguistics I University of Michigan I Ann Arbor, Michigan 48109-1285 I U.S.A. [email protected] Lynn Eubank Division of Linguistics I Department of English I University of North Texas I Denton, TX 76203-3827 I U.S.A. [email protected]

408

Affiliations and addresses

Paula Fikkert Dutch Department I University of Nijmegen I P.O. Box 9103 I 6500 HD Nijmegen I The Netherlands [email protected] Helen de Hoop UiLOTS, Utrecht University I Trans 10 I 3512 JK Utrecht I The Netherlands [email protected] Norbert Hornstein Department of Linguistics I 1401A Marie Mount Hall I University of Maryland at College Park I College Park, MD 20742-7505 I U.S.A. [email protected] Harry van der Hulst HIL/Leiden University I P.O. Box 9515 I 2300 RA Leiden I The Netherlands [email protected] Alana Johns University of Toronto I Department of Linguistics I 130 St. George I Toronto Ont M5S 3H1 I Canada [email protected] Kyle Johnson Linguistics Department I South College I University of Massachusetts I Amherst, MA 01003 I U.S.A. [email protected] Alan JufFs English Language Institute I Department of Linguistics I 2816-CL University of Pittsburgh I Pittsburgh, PA 15260 I U.S.A. [email protected] Curt Rice University of Troms0 I Faculty of Humanities I N-9O37Troms0 I Norway [email protected] James Rogers Department of Computer Science I CS 252 University of Central Florida I Orlando, FL 32816-2362 I U.S.A. [email protected]

409

Affiliations and addresses

Wendy Sandler Sign Language, Linguistics and Cognition Research Laboratory and Department of English Language and Literature I The University of Haifa I 31905 Haifa I Israel [email protected] Henriette de Swart UiL OTS, Utrecht University I Trans 10 I 3512 JK Utrecht I The Netherlands [email protected] Elizabeth Closs Traugott Department of Linguistics 94305-2150 I U.S.A. [email protected]

I Stanford University

I Stanford, CA

Studies in Generative Grammar Edited by Harry van der Hülst, Jan Köster and Henk van Riemsdijk Mouton de Gruyter · Berlin · New York 11 18 19 21 23 25 30 32 35 37 38 39 40 41

42 43 44 45 46 47

Luigi Rizzi, Issues in Italian Syntax. 1982. Sergio Scalise, Generative Phonology. 1986. Joseph E. Emonds, A Unified Theory of Syntactic Categories. 1985. Studies in German Grammar. Edited by Jindrich Toman. 1985. Samuel J. Keyser and Wayne O'Neill, Rule Generalization and Optionality in Language Change. 1985. Pieter Muysken and Henk van Riemsdijk, Features and Projections. 1986. Jan Köster, Domains and Dynasties. The Radical Autonomy of Syntax. 1987. Lars Hellan, Anaphora in Norwegian and the Theory of Grammar. 1988. Günther Grewendorf, Ergativity in German. 1989. Shosuke Haraguchi, A Theory of Stress and Accent. 1991. Reineke Bok-Bennema, Case and Agreement in Inuit. 1991. Case and Other Functional Categories in Finnish Syntax. Edited by Anders Holmberg and Urpo Nikanne. 1993. Carmen Dobrovie-Sorin, The Syntax of Romanian. Comparative Studies in Romance. 1994. Studies on Scrambling. Movement and Non-Movement Approaches to Free Word-Order Phenomena. Edited by Henk van Riemsdijk and Norbert Corver. 1994. Gereon Müller, Α-bar Syntax. A Study in Movement Types. 1995. Susanne Winkler, Focus and Secondary Predication. 1997. The Mathematics of Syntactic Structure. Trees and their Logics. Edited by Hans-Peter Kolb and Uwe Mönnich. 1999. The Syllable. Views and Facts. Edited by Harry van der Hulst and Nancy A. Ritter. 1999. Valentina Bianchi, Consequences of Antisymmetry. Headed Relative Clauses. 1999. Marc van Oostendorp, Phonological Projection. A Theory of Feature Content and Prosodic Structure. 2000.