Mouton Classics: From Syntax to Cognition. From Phonology to Text [Reprint 2013 ed.] 9783110886009, 9783110173468

Citation preview

Mouton Classics

w DE

G

Mouton Classics From Syntax to Cognition From Phonology to Text Volume 1

Mouton de Gruyter Berlin • New York 2002

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

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

Die Deutsche

Bibliothek

- CIP Einheitsauf

nähme

Mouton classics : from syntax to Cognition ; from phonology to text. Berlin ; New York : de Gruyter

Vol. 1. -2002

© Copyright 2002 by Walter de Gruyter GmbH & Co. KG, 10785 Berlin All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted in any form or by any means, electronic er mechanical, including photocopy, recording or any Information storage and retrieval system, without prior permission in writing from the publisher. Printing and Binding: WB-Druck, Rieden am Forggensee. Cover design: Sigurd Wendland, Berlin. Printed in Germany.

Content

Volume 1 1. Phonology Harry van der Hülst and Nancy A. Ritter. Head-Driven phonology [1] A. Traill. The perception of clicks in !X6ö [2]

3 59

2. Morphology Frans Plank. Paradigm size, morphological typology, and universal economy [3]

75

Richard J. Hayward. The origins of the North Ometo verb agreement systems [4] 95 3. Formai Syntax Kenneth Haie. Gore structures and adjunctions in Warlpiri syntax [5] 117 Richards. Kayne. Unambiguous paths [6]

153

Luigi Rizzi and lan Roberts. Complex Inversion in French [7] . . . .

195

4. Functional Linguistics Edward L. Keenan. Semantic correlates of the ergative/absolutive distinction [8]

227

Simon C. Dik. How to build a natural language user [9]

255

Simon C. Dik. Preview of Functional Grammar [10]

269

Greville G. Corbett and Norman M. Fräser. Default genders [11] . 297 5. Historical Linguistics Werner Winter. The distribution of short and long vowels in stems of the type Lith. Ssti: vesti: mesti and OGS jasti: vesti: mesti in Baltic and Slavic languages [12]

343

Bernard Comrie. Morphology and word order reconstruction: Problems and prospects [13]

359

vi

Contents

Volume 2 6. Language and Cognition Adele E. Goldberg. The inherent semantics of argument structure: The case of the EngUsh ditransitive construction [4]

375

Ronald W. Langacker. Reference-point constructions [15]

413

Suzanne Kemmer and Arie Verhagen. The grammar of causatives and the conceptual structure of events [16]

451

7. Language Acquisition Melissa Bowerman. Mapping thematic roles onto syntactic functions: are children helped by innate linking rules? [17]

495

8. Discourse and Text Hans-Jürgen Sasse. The thetic/categorical distinction revisited [18]

535

Kees Hengeveld. Cohesion in Functional Grammar [19]

605

Joäo Costa. Focus in situ: Evidence from Portuguese [20]

621

9. Sociology of Language Einar Haugen. Language and ethnicity [21]

665

Michael Meeuwis. Flemish nationalism in the Belgian Congo versus Zairian anti-imperialism: Continuity and discontinuity in language ideological debates [22]

675

Florian Coulmas. Language masters: defying linguistic materialism [23]

719

Miriam Eisenstein Ebsworth and Timothy Ehsworth. The pragmatics and perceptions of multicultural Puerto Ricans [24] . . .

731

10. Semiotics Thure von Uexküll, Werner Geigges and Jörg M. Herrmann. Endosemiosis [25]

771

Paul Ekman and Wallace V. Friesen. The repertoire of nonverbal behavior: Categories, origins, usage, and coding [26]

819

Bibliography (Volume 1 und 2)

869

1. Phonology

6

Head-Driven Phonology

Harry van der Hülst & Nancy A. Ritter

1. Introduction

The fundamental units underlying spoken natural language comprise a vast array of possible sounds. The observable fact that one language differs from anoÄer witii respect to the kinds of sounds used and the combinatory interactions among sounds stems from a language's choice of the possible sounds available in conjimction with a grammar that dictates the syntax of these sounds. At first Observation, the possible combinations of the members of this set of soimds or segments may seem to vary in each language, such that in some languages structures may be quite complicated (for instance in allowüig for complex Clusters of consonants to occur word-initially, finally, or medially) while in other languages such combinatory elaborations are non-existent. Yet, upon closer investigation of cross-linguistic data, there seems to be a number of generalizations that can be made about how sounds are grouped together. Given that such generalizations can be extracted from languages which appear to have rather varied surface structures, Üie next logical step would be to try and motivate such generalizations in a principled way. The question arises, then, as to what kind of theory best captures the basic structure of phonological architecture at the segmental level. While in the study of (morpho-)syntax there is a well-estabUshed concept of a system of principles and parameters, comprising the common core of innate knowledge known as Universal Grammar (UG), the study of phonology is ofren beheved to be different and much less attention has been paid to developing a unifying set of principles and parameters which define the common core and potential Variation in the architecture of phonological representations. This is actually rather surprising given the fact tiiat the focus of attention in phonological theory has been on the representational aspect for so many years (from the mid-seventies to the early nineties). Despite the enormous progress which has been made in specific domains (such as foot structure), no overall parametric theory of phonological structure has emerged in mainstream generative phonology. In most textbooks and general studies, it still appears to be the case that different levels of phonological structure (segmental, syllabic, and

4

Harry van der Hülst and Nancy A. Ritter

[114]

higher up) are of a rather diverse formal character. Although segments, syllables and feet are usually all represented as tree-like objects, the formal properties of these trees (in terms of headedness, recursion, binarity of branching etc.) are taken to be very different by most phonologists. This chapter serves to outline an approach, called Head-Driven Phonology (HDP), in which the central claim is that phonological structure, at all levels of the hierarchical Organization, is best characterized in terms of binary head/dependent relations. This approach incorporates proposals put forward within two closely related theories known as Dependency Phonology (DP; Anderson & Ewen 1987) and Government Phonology (GP; Kaye, Lowenstamm & Vergnaud 1990) and especially delineates the idea that some of the imiversal principles that have been proposed for syntax can also be claimed to relate to the phonological component (and vice versa). In this view then, phonology is not radically different from morpho-syntax. Contrary to the idea that the modularity of grammar entails the expectation that the modules are totally different formally, Anderson & Ewen (1987) introduce the Structural Analogy Hypothesis which expresses the initial expectation that morpho-syntax and phonology appeal to the same (or analogous) structural principles and parameters. Thus, this chapter disavows the claim that universal grammar is only synonymous with Syntax and üistead advocates the hypothesis that the notion of there being a universal grammar extends to the phonological component as well (cf Ritter 1995). The HDP model, then, Claims that phonology is driven by asynunetrical head/dependent relations and that such relations underlie phonological representations and are the key to understanding phonological processes. HDP-relations are manifested in terms of liceming mechanisms, which serve to authorize the imits that comprise phonological representations. In this chapter, we focus on how head-driven phonology is equipped to analyze complex phonotactic pattems in terms of a highly restricted set of maximally binaiy head/dependent relations. We point out here that HDP, like DP and GP, is essentially a non-derivational, monostratal theory which also crucially incorporates the notion of parameters (Kaye 1995, van der Hülst & Ritter 1999, in prep.). Representations are well-formed within the ränge of parameter settings if all the principles and licensing mechanisms are satisfied. The chapter is structured in the following way. In section 2, we preface the discussion of our approach with a characterization of the basic tenets of dependency and govemment phonology. We do not make an attempt in this section to Cover the ftill history of govemment/dependency phonology and also wish to make it clear from the onset that we do not feel committed to presenting these theories in any canonical way. This means that we choose our own wording in order to express our comprehension of the essence and insights of

[115]

Head-Driven phonology

5

these theories. We then offer a description of our general approach which differs from DP and GP in various ways, trying to develop a typology of licensing mechanisms. In section 3, we tum our attention to applying HDP to a case study (drawing on van der Hülst & Ritter 1998): the phonotactic Organization of minor syllables in Kanunu. Finally, in section 4, a brief summary with conclusions is presented.

2. An outline of Head-Driven Phonology

The fundamental contribution of DP is the claim that phonological structure involves head/dependency relations at all levels of Organization (including intrasegmental Organization). Intrasegmental feature classes, segments themselves, syllabic constituents (like onsets and rhymes), syllables themselves, feet, phonological words, and so on are all depicted as headed constituents. In fact, it is explicitly stated (in the form of the structural analogy hypothesis mentioned in the previous section) that the notion of a headed, binary constituent structure defmes what is in common between morpho-syntactic and phonological (often called prosodic) structure. GP takes a very similar perspective regarding both the structural analogy hypothesis and the Organization of phonology proper. A major presentational or methodological difference between DP and GP is that much that is "implied" in DP is stated in a restricted manner and much more explicitly in GP. Additionally, GP seems to be an approach which incorporates the role of universal grammar into the phonological component more clearly than DP does (cf Ritter 1995 for discussion). GP achieves tliis goal by providing a system of principles in conjunction with language-specific parameter settings which together define lexical items in an economical manner requiring minimal computation. In this way, a formal system is achieved for yielding well-formed representations from which the phonologies of individual languages can be construed.

2.1. Principles We formulate the central head/dependency relations as follows: (1)

Head/Dependency Principle: an object is either a head or a dependent; if a dependent, it can only exist if it is in a relationship with a head to which it is adjacent at some level.

6

Harry van der Hülst and Nancy A. Ritter

[116]

Ulis principle, as it will be shown, subsumes all sorts of relations which are manifested in terms of (i) structural constituent relations of govemment (sisterhood), (ii) structural domination relations, and (iii) syntagmatic relations between elements that may or may not belong to dififerent constituents: (2)

HD-relations i. Structural relation of govemment: (H

D)

ii. Structural (paradigmatic) relation of domination:

H I

D iii. Non-structural (syntagmatic) relation: (x

y) or > (the direction of the head-dependent relation may vary)

x)

(y >

(2ü) bears on the melodic content of skeletal positions and on the structural content of higher units. (2iii) will be argued to exist with reference to the content of skeletal positions only. Another principle which we claim is an innate and necessary part of UG is the Binarity Principle. (3)

Binarity Principle: all head/dependent relations are maximally binary.

The two principles in (1) and (3) allow for the presence of either a head alone, or a head and a dependent, but never a dependent alone, nor combinations of more than one head or more than one dependent. With reference to the latter Situation we note especially that the temary branching structures in (4) are mied out (where H = hea^ d = dependent, and HP = constituent):' (4)

a.

* HP / I X d H d

b.

*

HP f X r ^ H dl d2

When, as in constituent stmcture, head/dependency relations are hierarchically layered, some imit that is a head can form a imit which is a dependent at a higher level. This recursion yields the effect of subsuming all material into a layered, hierarchical prosodic stmcture. In order to delimit the expanse of this recursive stmcture from continuing ad infmitum, it has been claimed that there is an Upper limit to the number of domains or layers (Nespor & Vogel 1986).

[117]

Head-Driven phonology

7

Starting from the smallest units, we adhere to the view that the prosodic hierarchy has at least the foUowing layers: (5)

X°

nucleus

foot

word

phrase

Observe that we do not include the 'syllable' in this hierarchy. The reasons for this will be discussed below. We do not wish to commit ourselves here to a view regarding levels of Organization above the phonological phrase, conceming ourselves in this chapter with structures that occur at the lexical level. It follows from our two principles that a (phonological) word has the following maximal structure:^ (6)

N

Word level Foot level

N N

N N

Helsloot (1993, 1995) proposes that the word is a bounded constituent in precisely this sense, presenting an argimient based on verse in Italian.^ We would now like to propose that the word in this sense equals what in GP is called the non-amlytic domain (Kaye 1995), and what others have called the 'word-level' (cf. Chomsky & Halle 1968, Borowsky 1994). The object which we define as the overall head of the word domain, is the nucleus which projects as head at all levels up to the topmost node of the domain (shown as emboldened in 6). This (ultimate) head is the 'strongest', most viable head in the word, and, apparently as a result of this, a condition on its content exists, namely that its content is free, i.e. with no restrictions on its segmental material. Being free implies that there are no restrictions on what this head can dominate, in the sense that any vocalic segment can appear in this position. It thus becomes the Site for maximal contrast (cf Dresher & van der Hülst 1995, 1998). It also follows from being free that the possibility of this ultimate head (UH) being empty never exists since, as will be discussed in section 2.4 below, the justification of an empty position is controlled by licensing mechanisms. Other nucleic heads, however, while heads in themselves, are directly or indirectly dependent on the UH, thus allowing for the possibility (but not the necessity) of the content of these other nucleic heads to be restricted or to be empty. Heads of feet, though typically strong, do not necessarily show all contrasts. When the foot is not the strongest foot in the word, as our case study in section 3 will show, heads of structurally dependent feet can, in fact, be very restricted, showing no possibility for contrast at all (cf. also Ritter 1998a).

8

Harry van der Hülst and Nancy A. Ritter

[118]

It follows from the above that words of more than four syllables must consist of more than one phonological word.'* A simplex word like hippopotamus, with five syllables, forms, if we are right, two phonological words (as in 7):^ (7)

N" 0°

N°

N" 0°

N®

N" 0°

N"^

N" 0°

N"

N® 0 °

N" 0°

m

m

t

(Here, and henceforth, '[0]' undemeath a skeletal point or zero-level position indicates that the position does not dominate content and is inaudible.) Our approach postulates feet that do not bear stress (such as the final foot in 7) which implies that we do not assume a one-to-one relationship between being the head of a foot and bearing stress. The phenomenon of stress is a phonetic interpretation of metrical structure and it is consistent to say that, for example, only contentful heads of branching feet bear stress. The foot status of monosyllabic feet is evidenced by facts conceming the aspiration of stops and the occurrence of /h/ (italicized), which both have been claimed to be dependent on initial position in the foot (examples from Davis 1999):® (8)

tir

a. W

wi Ta

b.

W

ne ra

pe hu

sau kee ma ra

zon

We realize that our view on (lexical) prosodic structure needs more justification than we can give here and we refer the reader to van der Hülst & Ritter (in prep.) for that. The Claim made here suggests that, in general, analyses appealing to "unbounded iteration of foot structure" cannot be correct imless we assume that the

[119]

Head-Driven phonology

9

resulting structure (as in the case of hippopotamus) contains more than one phonological word. Occasionally, the metrical literature has appealed to such intermediate structures. The analysis that Hayes (1995) proposes for Passamaquoddy groups syllables into feet, feet into 'cola' (singular colon), and cola into a word. In our tiieory, the cola are prosodic words and the resulting structure is a prosodic phrase; cf. van der Hülst (1997). Another consequence of our view is that neither 'temary feet' nor 'unboimded feet' can exist. This raises interesting issues with respect to the prosodic structure of words in languages with so-called temary rhythms and socalled unbounded word accent. Here we will simply assume that temary feet can be analyzed in terms of combinations of two feet (i.e. as prosodic words, as in the example above). Van der Hülst (1997) applies this analysis to temary systems of Äe iambic type. Rifkin (1999) argues for prosodic word status of dactylic feet. With respect to one variety of unbounded systems, lexical accent systems, Revithiadou (1999) has shown that there is, in fact, no need for imbounded constituents since lexical accents (for example in Russian) appear in positions which only allow the formation of well-formed phonologicd words. Yoshida (1995) proposes a bounded approach to another system that has been claimed to be unboimded, lexical accent stmcture in Standard Japanese (cf Ritter 1998a).

2.2. Licensing mechanisms The principles of head/dependency and binarity defmed above work together and are manifested in a variety of forms of licensing, exemplified in (2). We will now discuss these types of licensing in more detail.

2.2.1. The structural relation of govemment (sisterhood) Stmctural relations of sisterhood are the framework for hierarchical constituent stmcture. In this type of head-dependent relation known as 'govemment', a head is able to project a constituent by virtue of its potential to form a goveming domain with an adjacent sister dependent. Anticipating a discussion of the types of rhyme stmcture that govemment phonology allows, (9) gives two branching configurations. (9a) is an illustration of sisterhood, i.e. the fu-st node that dominates the head also dominates the dependent. Let us call this minimal c-command. Example (9b) illustrates a context in which minimal c-command does not obtain since the intermediate node H' does not dominate both the head (H) and the dependent (D). Whether c-command is met or not, however, there

10

Harry van der Hülst and Nancy A. Ritter

[120]

is a head-dependency relation in both cases which satisfies both the principles of head/dependency and binarity in tiiat, in each of the two cases, the dependent unit is in an adjacent binary relationship with its head: (9)a.

b.

GP allows both structures in (9) as possible rhyme structures. (10a) characterizes long vowels, while (10b) is a rhyme constituent which contains a short vowel and a closing consonant: (10) a.

b.

R 1

R

1

N 1

N f X

\ X

X

In section 2.2.1.5, we will argue that we can do without the stractures in (9,10b)^ and thus that minimal c-command is a necessary condition on branching structures. The structural governing relation which only allows binary relations between sisters ultimately yields a universal structural Schema of the onset (O) and rhyme (R) units, the foot, the word, etc., and in so doing, specifies the notion of the well-formedness of each of these constituents and how they relate to one another. Example (11) illustrates the hierarchical architecture, based on this notion of govemment relations, up to the word level:

(11)

O" 0° (0°)

N^ (W)

(Here and henceforth we use < to represent a head/dependent relationship.) The special relationship between O" and N" is discussed in section 2.2.1.2. We will fu-st Start with the lowest level, the zero-level of the hierarchy. With respect to the higher levels of the foot and word, we have ah-eady established in

[121]

Head-Driven phonology

11

section 2.1 that temary and imbounded structures are disallowed, and we have also suggested how apparent occurrences of such structures can be analyzed without appealing to non-binary constituents. In the following discussion, we will make it clear where our views deviate from GP.

2.2.1.1. Theskeleton Begiiming, then, from the terminal points, it should be noted that in our approach we notalionally replace the traditional notion of skeleton (a string of socalled x-positions) with 'zero' positions. The zero-level units (such as N°) are visible to the phonology and as such are subject to principles of headedness, binarity and a third principle, the Empty Category Principle (ECP), which we discuss in section 2.4.4. This zero-level unit dominates segmental expressions (mediated, in our approach, by a 'root node' which dominates phonological elements) and also encodes the notion of timing in order to differentiate between long and short segments (cf section 2.2.1.6 on long vowels). Constituents which do not have zero-level terminal heads merely serve as representative cognitive placeholders in the structural Schema but do not contribute to any interpretation per se. They could just as readily be left out of the representation. For instance, words which begin with a vowel on the surface and also phonologically behave as vowel-initial will have a structure in which the initial onset constituent may be representationally present but its zero-level head will not be present, as in the French word ami 'fiiend' (masc. sing.) illustrated in (12) below (cf Tranel 1987): (12)

O"

N" I N»

O" I 0°

N" I N°

a

m

i

(Here, and elsewhere, we suppress the intermediate N'-level, which we will get rid of entirely in section 2.2.1.5.) Since the initial onset constituent contains no zero-level position, i.e. no head, it lacks the component which the computational system recognizes as necessary to make an object visible to the phonological system. Consequently, the lexical item in (12) is phonologically construed as vowel-initial. In cases, however, where an initial onset is claimed to be phonologically relevant, although it lacks any segmental content and appears silent on the surface, the onset constituent is headed by a zero-level position (O"). Constituents which project from a zero-level head are subject to the aforementioned principles, since being at the zero-level identifies these objects as being phonologically

12

Harry van der Hülst and Nancy A. Ritter

[122]

relevant to enter into licensing relations. Example (13) gives an illustrative representation of the phenomenon that has been described as h-aspire in French, in which apparent vowel-initial words act as if they begin with a consonant, as with haricot 'bean' below. (13)

O"

N"

O"

N"

O"

N"

qO

nO

qO

^

qO

^

[0]

a

r

k

0

i

Thus a contrast with respect to the presence or absence of the zero-level head is claimed to exist in the case of onsets which accounts for the variance in phonological relevancy of onset constituents which appear silent on the surface. Whereas nuclei can co-occur with onsets that lack a zero-level position, it is not Our Position that the reverse Situation is possible. In fact, we claim that onsets cannot occur with nuclei that lack a zero-level head. We derive this from the fact, discussed in the next section below, that onsets can only be prosodically anchored by being adjoined to a nucleus that is itself phonologically viable (i.e. has a zero-level head). This stems from the assumption that prosodic anchoring is a necessaiy condition for the interpretability of onsets and that prosodic viability implies the existence of phonological visibility. Since a nucleus lacking a zero-level position itself would not be visible to the phonology and thus could not be a potential recipient of phonetic interpretation, such a nucleus could not occur and therefore could not be an anchor for an onset. As a result, nuclei (as opposed to onsets) always have zero-level head positions.*

2.2.1.2. How do onsets fit in? GP represents syllabic constituents like onsets and rhymes in sequence on a Single tier or plane, as in (11). These two units do not form a syllabic constituent which GP rejects (cf Brockhaus, this volume). In this section, we present the somewhat different view that is adopted in HDP. Since onsets can be missing at the zero-level, they are not calculated, whether present or not, as being phonologically relevant in creating the prosodic structure. As a result, onsets seem to somehow stand outside of the prosodic hierarchy. In our approach, we represent this by also rejecting the syllable as a constituent in the prosodic structure. Lexically, there is a head-dependency relation between an onset and a nucleus, which we will refer to as 'anchoring' (or more neutrally: adjunction). This adjunction relation could be considered as creating an Interface level between two planes, similar to the way Semitic rootand-pattem languages operate. In the case of the onset-nuclear interface, the

[123]

Head-Driven phonology

13

nuclear plane is the core which bears the notion of prosody, and the onset plane serves to demarcate prosodic peaks and carries with it some notion of semantic content. Example (14) illustrates this concept: (14)

Anchoring/Adjunction

We assume, with govemment phonology, that every nucleus must be preceded by an onset (and vice versa)', but we place the two units on different planes to more clearly express that onsets do not partake in the prosodic hierarchy. We advance an Observation here regarding reduction processes that, to our mind, has not been made in the literature before, and which strongly supports the idea that onsets do not form part of the prosodic hierarchy. It is well-known that vowels may reduce in weak metrical positions. One aspect of reduction is shortening or monophthongization. English is a clear case in point. However, reduction never seems to involve the change from branching onsets to simple onsets. In a pair like photograph - photography we note that the vowel of /graef reduces m the second form, while die düster /gr/ remains in tact. Reduction fails to effect onsets, because reduction is triggered by occurring in strong or weak positions which are defined in terms of the prosodic hierarchy. Onsets do not partake in the prosodic hierarchy.'"

2.2.1.3. Coda

liceming

Govemment phonology, as we have seen in (10), repeated here as (15), makes a distinction between branching nuclei and branching rhymes: (15) a. N

The 'coda' is analyzed as a specifier at the rhymal level. It is argued in Kaye, Lowenstamm & Vergnaud (1990) and Kaye (1990, 1995), that the 'coda' is not a constituent in itself because it cannot brauch (i.e. there seems to be no empiri-

14

Harry van der Hülst and Nancy A. Ritter

[ 124]

cal need for postulating branching codas)". Since this coda position is removed by one projection (namely N) from the nuclear head and thus not an inunediate sister of the nuclear head, the coda position is not minimally c-commanded by the nuclear head, as pointed out in section 2.2.1. Because tiie coda position is not minimally c-commanded, one could conclude that the relevant structure is simply ill-formed (a position that we will take in section 2.2.1.5). In GP, a milder position is adopted: the coda position is allowed, but it needs extra hcensing. This is the way in which we view what Kaye (1990) calls Coda Licensing. According to this condition, a coda can only occur if followed by an onset which iicenses' it: (16)

R

O

N

r \

X

X

(coda licensing)'^ As Kaye (1990) points out, a consequence of coda licensing (i.e. the requirement that all codas be followed by an onset) is that closed syllables cannot occur word-finally. Yet, lexical items often have final consonants. In this context the consonants cannot be licensed by a following onset, but there is another way of licensing them if it is assumed that the consonants form onsets which are anchored to a following 'empty' nucleus.'^ The word cat would be represented in govemment phonology as in (17): (17)

0 11

N 11

0 11

N 11

X

X

X

X

k

ae

t

[0]

In section 2.2.1.7, we will tum to the question as to whether coda licensing will be adopted in our approach. First, in the next section, we discuss another Ucensüig principle that has been adopted in govemment phonology. A final point with respect to coda licensing must be made. This principle ensures that onsets take priority over codas, since a string VCV, to be in agreement with coda licensing caimot be parsed as VC.V, because, in that case, the coda is not licensed by a following onset. Coda licensing also ensures that onsets are maximized (cf the so-called Maximal Onset Principle) because, as pointed out in Kaye, Lowenstamm & Vergnaud (1990) and Kaye (1990), an

[125]

Head-Driven phonology

15

onset can only license a preceding coda if the onset consonant is 'less sonorous' than the preceding coda (cf. Harris 1990 for a formalization of the notion 'sonorant' in terms of complexity). A string VbrV can thus never be parsed as Vb.rV because an /r/ can never coda-license a /b/. Such a string will be parsed as V.brV in a language that permits branching onsets, and as V.b[0].rV (i.e. with an empty nucleus after the /hl) if no branching onsets are permitted 2.2.1.4. Government licensing Before we tum to Government Licensing, it will be necessary to say a few words about another central mechanism of GP, viz. Proper Government (PG). We have seen that, in GP, nuclei are allowed to be empty, i.e. contain a skeletal point that does not contain any content. In order to avoid a proliferation of silent nuclei, GP demands that they be licensed. This demand is stated in the form of the so-called Empty Category Principle: An empty nucleus must be licensed in order to be inaudible.''* There are several hcensing mechanisms pertaining to empty nuclei, which we discuss further in section 2.4. The most ünportant of these is proper govemment. In essence, proper govemment obtains if the empty nucleus has a neighboring nucleus that is audible, i.e. either contentful or an empty unlicensed nucleus (and thus phonetically interpreted). In most GP accounts, PG is a right-to-left (iambic) relation:'^ (18)

R

O

N

N

[0]

R

P

(proper govenmient) PG can be 'blocked' if there is an intervening consonant düster (either a codaonset sequence or a branching onset). This means that such consotiant Clusters cannot be preceded by an empty silent nucleus; cf Kaye, Lowenstamm & Vergnaud (1990) for details. In section 2.4.4 we show how PG is integrated into HDP. With this background, we tum to Charette (1990). Charette (1990) proposes a licensing principle that involves a relationship between some unit that licenses another unit to govem a dependent. The central Observation is that in some cases it appears that inaudible nuclei cannot follow a consonant düster, albeit coda-onset Clusters or branching onsets:

16

Harry van der Hülst and Nancy A. Ritter

(19)a. * R

O

R

b. * O

R N

N

XXX

XXX [a] 1 p

[126]

[0]

(coda licensing)

p 1[0] (govemment)

In the relevant languages, empty nuclei, when followed by an audible nucleus (i.e. when properly govemed), are permitted. Yet in the configiirations in (19), Äe empty nucleus must be audible even when followed by an audible nucleus that govems it. Why would this be so? Observe that, according to the theory of GP, in both (19a) and (19b) the /pZ-segment must govem another consonant, the govenunent relation being indicated. In (19a) the /p/ must coda-license the preceding rhymal coda, while in (19b) the onset head must govem its dependent. The proposal that Charette makes is that consonants that must govem a preceding (as in 19a) or following consonant (as in 19b) need to receive licensing power from a following nucleus, and in order to do this, a nucleus must be filled. She refers to this phenomenon as 'license-to-govem' or 'govemmentlicensing': (20) a.

R

O

R N

XXX

XXX

a 1 p

p 1 a

O

N

p

i/[0]

17

< (govemment licensing)'^ In sections 2.2.1.6 (note 26), we show that the 'local' (or direct) form of govemment licensing in (20a) is needed. For the 'nonlocal' (or indirect) form in (20b) we propose an alternative in sections 2.2.1.7 and 2.2.3. In this chapter, we accept Yoshida's accoimt of CSS (even though it is also nonlocal). In van der Hülst and Ritter (in prep.) we consider local alternatives to Yoshida's analysis.

2.2.1.5. Is the nucleus-rhyme distinction necessaty? Allowing the two structures in (15) raises the question of potentially combining the presence of a branching nucleus and a coda, thus creating a 'superheavy' syllable, as in (22). Govemment phonology wishes to ban such a constmct (assimiing that it is imnecessary for the analysis of languages) and does this by pointing out (referring to the notion of 'strict locality' of govemment) that in such a case, the head would not be adjacent to the coda dependent: (22)

This stmcture is one that might come to mind for rhymes that contain a long vowel and a closing consonant, a so-called 'superheavy' rhyme. We tum to the GP analysis of superheavy rhymes (both W C and VCC) in section 2.2.1.6. Another requirement that govenmient phonology proposes on O/R stmctures is that of strict directionality: the head always govems to the right. This rales out a nimiber of further stractures: (23) a.

*HP

b.

18

Harry van der Hülst and Nancy A. Ritter

[128]

Such structures are, in fact, disallowed not only for rhymes but also for onsets. Structures as in (23) have occasionally been proposed for onsets consisting of three consonants (e.g. spr in English); cf. van de Weijer (1994) for discussion. In Our mind, positing a structural position to accommodate special Clusters (such as those involving initial /s/) seems to miss their special character. Thus, we believe that allowing such temary structures will only and undesirably extend the descriptive power of the theory leading to a loss of explanatory power.'* Let US now focus on the two structures in (15) that are allowed in govemment phonology. The consequence of banning the temary structure in (22), results in the fact that the structures in (15), stand in complementary distribution to one another with respect to the representation of the nuclear complement. In section 2.2.1, we announced that we would hke to disallow the structure in (15b). We will now develop this idea that no intermediate bar-level exists.'' Head-driven phonology, then, will only acknowledge the structures in (24):^° (24)

a.

O"

I

b.

O"

c.

N"

I

We stipulate (as in govemment phonology) that the direction of the govemment head/dependent relation is non-parametric, universally left to right (left-headed) within the most minimal of constituent domains, i.e. at the zero-level within the onset constituent (cf 24b) and within the nuclear constituent (cf 24d). At the adjunction site of O" and N", where N" is the head, the licensing relation is a right-headed one, following from the claim (cf section 2.2.1.2) that prosodic structure (composed of nuclei only) is the foundation on which the onset is anchored.^' Thus we do not recognize a single-bar intermediate level as a constituent which can fimction as a head and enter itito a head/dependent relation. Only 'zero level' heads and maximal projections can fimction as heads, suggesting that only units of the same bar-level can enter into a head/dependent relation. An immediate advantage of adopting the position that minimal c-command is necessary is that this move rules out the onset structure in (25). GP, by allowing the analogous structure for rhymes, cannot explain why the same type of structure is not used for onsets:

[129]

(25)

Head-Driven phonology

19

O"

We realize that our proposal needs further support, which we will give by showing here and elsewhere (van der Hülst «fe Ritter, in prep.) that the resulting restrictive theory can account for all the relevant facts that we are aware of.

2.2.1.6. Do we med two structures for long vowels? We now address the question as to whether the structure in (24d) represents both long vowels and short vowels followed by another segment. It has been argued, for example in Yoshida (1990) that long vowels can (and in some cases clearly must)^^ be represented as bi-nuclear structures (with an empty intervening onset as in 21). If we use (24d) for long vowels, this means that we have two structures in the theory for long vowels. We therefore propose to represent all vowels as in (26): (26)

0" 11 x k

N" 11 X

0"

N" 11 X

a > (spreading)

The vowel melody of the first nucleus 'spreads' to the second nucleus in the phonetic interpretation of this structure. We propose, then, to restrict the branching nucleus option in (24d) to vowel-consonant sequences, where the 2nd member of such sequences includes glides (or approximants), liquids, homorganic nasals^^ and left halves of geminates.^" This move also seems supported by the typological facts. The presence of long vowels does not imply the presence of closed syllables, and vice versa. If both are the consequence of parametrically allowing branching nuclei, then this mutual implication would be the expected result. We refer to van der Hülst & Ritter (in prep.) for further extensive discussion of long vowels (for consonants; cf section 2.2.3). Here we limit ourselves to some remarks about the distribution of long vowels.^^ It has been observed that long vowels fail to occur in 'closed syllables' either totally (type I), or only medially (type II). As shown in Kaye (1990), a language like Yawelmani belongs to type 1: long vowels occiu" only 'in open syllables'. A

20

Harry van der Hülst and Nancy A. Ritter

[ 130]

further restriction obtains, however, in that even in open syllables, long vowels are prohibited if word-final. The prohibition on the occurrence of long vowels can lead to altemations involving vowel shortening when underlyingly long vowels end up being in a 'closed syllable'. The shortening process has been termed Closed Syllable Shortening (cf. section 2.2.1.4). Arabic dialects have been mentioned as typical examples of type II: word-finally, long vowels followed by a consonant (creating 'superheavy' syllables) are allowed, but wordmedially, long vowels can occur in open syllables only. The evidence regarding long vowels in final open syllables is less clear in type II languages. This distribution of W C is usually paralleled by the distribution of VCC in that VCC either does not occur at all (type I), or only finally (type II). Government phonology accounts for these facts as follows. Adopting a binuclear representation of long vowels, we represent in (27a) a string that results in a siuface superheavy syllable W C . We then adopt the proposal in Yoshida (1992), based on Charette (1990), discussed in section 2.2.1.4, which claims tiiat in languages that bar W C altogether, the head of a long vowel (Ni), which is the nucleus dominating the vowel melody, must be licensed to govern the following empty nucleus (N2) by a following contentfiil nucleus (N3). Thus, if N3 is not audible, as in (27a), the structure is ill-formed. Given this requirement of being licensed-to-govem, word-final long vowels are also prohibited. Extending the argument to VCC 'rhymes', (27b) shows an ill-formed structure because N2, being inaudible, is not properly govemed by a following audible nucleus:^® N, 11

0

N2 11

0 11

N3 11

X

X

X

-->

P

[0]

X

a —

n

-

(govemment licensing not possible) N2 11

0

N3 11

X

X

X

m

p [0] - H (PG not possible) < -

To account for languages of type II, which allow W C and VCC finally, in which case N3 is at the end of the word, we must postulate that word-final

[131]

Head-Driven phonology

21

empty nuclei (unlike medial empty nuclei) do have the power to govemmentlicense, and to properly govem; cf. section 2.2.1.4 and note 16. Word-final long vowels are still prohibited because there simply is no following nucleus to govemment-Iicense the head of the long vowel.

2.2.1.7. Do M)e need coda licensing? Having abandoned the difference in (15), we must review the evidence for coda licensing, i.e. the claim that a 'coda' consonant must be followed by an onset, barring the occurrence of domain-final consonants. A first Observation we make is that the licensing relationship between a coda and following onset could be generalized by stating that each dependent (whether in the nucleus or in the onset) must be licensed by a following audible head. We have seen in section 2.2.1.4 that Charette (1990) proposes a principle (license-to-govem) which, among others, demands that an onset dependent can only occur when its head is licensed-to-govem its dependent by a following nuclear head (cf 20). We argue in section 2.2.3 that the required result in this case can be derived fi-om a (syntagmatic) licensing relation (as shown in 28a) that subsumes the coda licensing relation (displayed in 28b).This means that all zero-level positions are now formally iinked', as shown in (28c), either in terms of left-headed govemment or in terms of right-headed interconstituent syntagmatic licensing (
== k o s t ' .

The first sentence at NOUN should be read as saying that the locative Singular consists of the stem followed by an -e ending. A path enclosed in double quotes in a DATR sentence is used to retrieve the specified value (in this case, < s t e m > ) for the item (in this case, a lexical item) inheriting from it. If we wanted to find the nominative

[61]

Default genders

303

plural of tosr, we would inherit the sentence == "" _ i . Before going any further we would have to find out what the of kost'is. Since the answer is stated in (2) to be kost\ the nominative plural must be h)stl. If we wanted to know the locative Singular of kosf, we would never inherit the definition of locative Singular at the NOUN node because it is overridden at N _ I I I , from which kost' inherits first, thereby blocking the more general ending. The definition of locative singular at N _ I I I establishes an asymmetric identity between the locative singular form of an N _ I I I noun and its dative Singular == ""). It is Worth noting some salient features of this approach. First it is declarative; we do not give underlying forms, and then transform them into other forms. We make a set of (partly conflicting) statements, and by embedding them in a network we specify the relations which hold between them. Since the approach is declarative we do not have ordering of rules. Second, defaults may be seen as hierarchically related: we make default statements about nominals, about nouns, about the N_0 class, and so on. Third, since Computer Interpreters are available for the DATR language, it is possible to check that an inheritance network expressed in DATR captures the intended generalizations. All our Network Morphology analyses have been checked in this fashion. More detailed introductions to default inheritance can be found in Gazdar (1987), and Daelemans, De Smedt and Gazdar (1992). Briscoe, de Paiva and Copestake (1993) and Calder (1994) are good places to look for recent formally explicit research in linguistic defaults. Introductions to DATR can be found in Gazdar (1990; forthcoming).

4. Defaults in a Single language We now retum to our central topic, namely gender. Russian has three genders, masculine, feminine, and neuter. We could enter a value for gender in the lexical entry of each noun. However, this would miss generalizations at two levels: first, the gender of Russian nouns does not appear to be random, and second, more generally, we have argued that languages never have to specify gender for the majority of

304

Greville G. Corbett and Norman M. Fräser

[62]

nouns. We have claimed that the gender of the overwhelming majority of nouns can always be predicted, either from semantic information which must, in any case, be stored in the lexical entry, or from semantic infonnation supplemented by formal information, which may be morphological or phonological (Corbett 1991: 68). 4.1. Defaultsfor gender assignment In Russian, the gender assignment rules refer to semantic and to morphological information: Semantic assignment rules for gender 1. Sex-differentiable nouns denoting males (humans and higher animals) are masculine: for example, Student '(male) Student'; 2. Sex-differentiable nouns denoting females are feminine: for example, uateVnica '(female) teacher' Nouns which are sex-differentiable are those denoting beings whose sex matters to humans (that is, other humans and domesticated animals) and where the difference is striking (as in the case of Hons). Very few exceptions are left by these rules, but there are many nouns which are not covered by them. Those remaining are distributed Over the three genders and their distribution is accounted for by the morphological assignment rules. The major morphological assignment rules are as foUows: Morphological assignment rules for gender 1. nouns of declensional class I are masculine; 2. nouns of declensional classes n and IE are feminine; 3. nouns of declensional class IV are neuter. As we shall consider in more detail later, there is some overlap between the two sets of rules: nouns which denote males are often in declensional class I. We might try to do away with the semantic assignment rules. However, the two sets of rules can make conflicting assignments, and when they do it is the semantic rules which dominate. The crucial case is mu^ina 'man' (and similar nouns), which ought to be masculine according to its semantics (male), but

[63]

Default genders

305

feminine according to its moq)hology (declensional class II). In fact, it belongs in the masculine gender.5 There are also nouns which do not decline, and to which the above morphological assignment rules do not apply. However, their lack of declension is itself a matter of morphology, and this interacts with the semantic feature of animacy to allow gender assignment. We treat indeclinable nouns as having their own declensional class (V); nouns of this class may be subject to the normal semantic assignment rules. Failing this, they are masculine if animate and neuter if not.6 In Our Network Morphology account (the detail of which can be found in Fräser and Corbett 1995), every noun inherits from the NOUN node. Since the following path equation is specified at NOUN, every noun inherits it unless it is overridden by a more specific equation declared in the lexical entry: (3) NOUN: < s y n g e n d e r > == GENDER:< "" >

In Order to find a value for the path , the path is evaluated (i.e., the sex of the noun's denotatum is retrieved) and then a path consisting of only the sex is evaluated at the node GENDER (shown below): (4) GENDER: == masc < f e m a l e > == fem < u n d i f f e r e n t i a t e d > == "".

The Interpretation of this fragment of D ATR is straightforward in the case of sex-differentiated nouns. If the denotatum of the noun is male then the gender is masculine, and if female then the gender is feminine. 7 In these two instances there is a simple default, which holds of nouns in general. However, if the sex is undifferentiated, it is necessary to consider additional criteria conceming form, and this is done by evaluating the path . Formal gender is defined for each of the five main declensional classes. Thus, for example, the following equation is located at the node for declensional class n nouns, N_II: (5)

N_II: < f o r m a l _ g e n d e r > == fem

306

Greville G. Corbett and Norman M. Fräser

[64]

This is used in the evaluation of the path (the function of the initial mor will be speit out below). This is a lower level default, which holds just for class II nouns. The effect of these defaults is that a class II noun denoting a male (such as muzcina 'man') will have masculine gender; a class II noun denoting a female (such as ucitel'nica -female teacher') will have feminine gender; and a class II noun denoting a non-sex-differentiated denotatum (such as komnata 'room') will have feminine gender, the default gender for class II nouns. As we noted above, there is an additional complication with declensional class V, where animacy must also be taken into consideration. Some equations from the node N_V, from which all class V nouns inherit, are given below. (6)

N_V: == < "" > == masc == neut

In Order to assign a formal gender for a given class V noun it is necessary to evaluate the path for that noun and use it to select masculine gender if the noun denotes an animate and neuter gender if it denotes an inanimate. Our analysis so far accounts for the gender of the vast majority of Russian nouns. It is an advance on Corbett (1982) in that it is formally explicit, and since it is encoded in DATR we can see that the right predictions are indeed made. 4.2. Defaults for inflectional class To illustrate our theme of defaults we will look briefly at the question of declensional class. So far we have assumed that the declensional class of each noun wiU be specified in its lexical entry. This seems to be missing a generalization, since there are common, though by no means exceptionless, correspondences between meaning, gender, and declensional class. Indeed, some earlier analyses attempt to predict declensional class from other Information which was specified (for discussion see Corbett 1982, Fräser and Corbett 1995). For many nouns, it is the case that declensional class is pre-

[65]

Default genders

307

dictable from semantic or formal information. The semantic correspondences are as follows: Semantic assignment rulesfor declensional class 1. Sex-differentiable nouns denoting males (humans and higher atiimals) are of declensional class I: for example, Student '(male) Student'; 2. Sex-differentiable nouns denoting females are of declensional class II: for example, uateVnica '(female) teacher' There are substantial numbers of nouns whose declensional class must be specified to override rule one. These are nouns like muzcina 'man', which denote males but which decline according to declensional class n. (Lazova 1974: 942-943 puts the figure at 273, but the number involved is actually larger because there are many hypocoristics of this type, like Sasa 'Sasha' which do not appear in dictionaries lüce Lazova's.) There are fewer instances of overrides to rule 2, but we find a small number of nouns like svekrov' 'mother-in-law' in declensional class HI. Formal assignment rule for declensional class 1. Nouns whose stem ends in a vowel are of declensional class V. The effect of this rule is to make nouns whose stem ends in a vowel indeclinable (class V are the indeclinables). Our rule follows Worth (1966), though as he points out the idea goes back much further. A noun Uke taksi 'taxi' is entered as such in the lexicon and this guarantees its indeclinability. These generalizations can be formalized fairly easily. Consider the foUowing DATR fragment, which is positioned at the NOUN node: (7)

NOUN: == "" == DECLENSION: < "" "" >

The first equation indicates that in order to find a value (or values) for the path (and its extensions), it is necessary to evaluate the < d e c l e n s i o n a l _ c l a s s > path. The second equation is somewhat more complex: a value can be retrieved for the < d e c l e n s i o n a l _ c l a s s > path by evaluating a path consisting of the value of

308

Greville G. Corbett and Norman M. Fräser

[66]

the path < i n f l_root f inal> followed by the value of the path . Paths beginning make available information about the inflectional root (i.e., the stem) of a word. The path is used to störe infonnation conceming the final segment of the inflectional root, specifically whether it is a consonant or a vowel. We may assume that this Information would, in reality, be supplied by a phonological component, though for convenience we simulate this crudely by means of the foUowing equation recorded at the NOUN node.^ (8)

NOUN: == consonant

This generalization holds for all Russian nouns, except certain indecHnable nouns of relatively recent foreign origin. In the case of these exceptions the default generalization is overridden in the lexical entries. We have already noted how values for the path are supplied either in the lexical entries or by means of a default assignment of undifferentiated sex. Thus, where declensional class is predictable, it is found by evaluating a path at the D E C L E N S I O N node. This path consists of Information on the final segment of the inflectional root of the word followed by the sex of the word's denotatum. The DECLENSION node is given below: (9)

DECLENSION: === N_I: == N_II: == N_V:.

The first two paths encode the semantic assignment rules for declensional class. The first element of each path is consonant so these paths potentially apply to the bulk of Russian nouns. The second elements of the paths narrow down on two subsets of these. The first path assigns core nouns (typical native nouns) denoting males to class N_I; the second assigns core nouns denoting females to class N _ l l . The third path picks out those nouns whose inflectional root ends with a vowel and assigns them to class N_V, the class for indeclinables. In DATR, every Symbol which begins with a dollar sign ($) is a variable. The variable $sex is defined to ränge over all possible values for the path, namely male, f emale

[67]

Default genders

309

and u n d i f f e r e n t i a t e d . One group of nouns is not covered by the equations at this node, namely those whose stem ends in a consonant and which have non-sex-differentiated denotata. Such nouns must typically specify declensional class in their lexical entries. However, as we shall see shortly (in the discussion leading to (10)), a substantial proportion of them can be assigned to a declensional class by the setting of a default. It is Worth contrasting declensional class assignment with gender assignment. A clear difference is the outcome when semantic and formal rules make different predictions. For gender assignment, quite generally, it is the semantic rule which takes precedence. As we noted earlier, with nouns like muzcina 'man', which ought to be masculine according to its semantics (male), but feminine according to its morphology (declensional class II), it is the semantic rule which "wins" and the noun is masculine. When we find a similar clash in declensional class assignment, in Russian at least,9 the formal factor dominates. Thus where attase 'attache' denotes a male and so "should" be in declensional class I, it nevertheless ends in a vowel and so "should" be in declensional class V. It is actually in declensional class V. Conversely, as far as gender is concerned, where it denotes a male and so "should" be masculine, it is nevertheless in class V and so "should" be neuter; as already noted, semantic factors take precedence in gender assignment and so it is masculine. A second difference between gender assignment and declensionalclass assignment, is that the gender assignment rules make a prediction for every noun (which has to be overridden in an extremely small number of instances), while the declensional-class assignment rules leave large numbers of nouns with no prediction. A non-sexdifferentiable noun with a stem ending in a soft consonant could decline according to class I, II, HI or IV.io However, assignment to the four classes is by no means equally likely. Consider data on the numbers of nouns in each class, to the nearest fifty (derived from Lazova 1974, especially pp. 942-943), given in Table 2 (next page). These raw statistics give a rough picture of what is going on. The figures do not take account of the effect of derivational morphology; for instance, over 4,300 of the nouns in class III have the suffix -ost' which forms abstract nouns from adjectives (star-yj 'old', star-ost' 'old age'). If the suffix is labelled as belonging to class III, then the number of distinct members of the class is substantially reduced. Similarly nominalizations in -anie/-enie (like razrusenie 'destruc-

310

Greville G. Corbett and Norman M. Fräser

[68]

tion', derived from razrmit^ 'destroy') inflate the figure for IV (see Schupbach 1984 for discussion). Table 2. Number of nouns in the diffeient declensional classes Class Example No of nouns

I

n

m

IV

V

zakon

komnata

kost'

vino

taksi

'law'

'room'

•bone'

'wine'

'taxi'

20,850

16,050

5,150

11,050

45011

Table 2 shows that I and n have substantially more members than the other three. We can therefore set a default which will put nouns in the majority declensional class I. Clearly this will be overridden in many instances, but equally it will simplify a very substantial number of lexical entries. This is a considerable simplification, and one which can be justified in three interrelated ways. First, there is the numerical preponderance of class I. Second, it is the declensional class which takes the majority of borrowings. There is no need to appeal to any markedness considerations to explain this, it is simply that class I has no ending in the citation form, the nominative Singular, so that foreign words ending in a consonant are typically borrowed into this class. The third argument is most important for our purposes. If the declensional class is I, and masculine is the gender associated by default with this class, then we make masculine the default gender for nouns, without specifying it directly. Sources vary as to the numerical preponderance of the genders. All agree that the neuter is easily the weakest; Lazova (1974: 942-943) has only marginally more masculines than feminines, while Muaiik (1971: 196-197) calculates that the masculine has 46% to the feminine 41% (on a sample of 33,952 nouns) and Zaliznjak's dictionary has 46% masculine and 38% feminine (sample 47,030: see üola and Mustajoki 1989: 9); the masculine is growing fastest, from the assignment of borrowings (see the sources reported in Corbett 1991:78). The only change required in our DATR theory to include all of these interacting defaults is given below, as a revision to the fragment shown above as (9).

[69]

Default genders

311

(10) DEF_DECL: == N_I: DECLENSION: == DEF_DECL == N _ I I : < >

== N_V:. Here, the variable $ p h o n o l o g i c a l _ t Y p e ranges over c o n s o -

nant and vowel, and $sex ranges over male, female and und i f f e r e n t i a t e d , as before. In all cases other than the two explicitly described in the paths at the D E C L E N S I O N node, the maximaUy general path specified at the D E F _ D E C L node is inherited and, thereby, class I becomes the default declensional class for nouns.

4.3. A higher-level default W e argued above that if class I is the default declensional class, and masculine is the gender associated by default with this class, then we make masculine the default gender. While this result fits with the intuitions of some investigators, it seems unsatisfactory to others. If masculine is the default gender for Russian, then we would expect it to appear, for instance, in examples like the foUowing: (11)

Byl-o

was-NEUT.SG

xolodn-o

COld-NEUT.SG

'It was cold.' Here there is no overt subject, but the verb and adjective must still take a particular agreement form and they take not the masculine but the neuter. The resolution of this apparent paradox is that we are claiming only that the masculine is the default gender for noms. Taking a broa der view, we would claim that there is a default for gender at a higher level than the nodes relating directly to nouns. This higher default is necessary for items other than nominals which may head syntactic constituents with which gender agreement is required. The Situation arises if, say, an Infinitive phrase stands in subject position (and there is a past tense verb), or there is an inteqection or other quoted material. Here we normally find the neuter.

312

Greville G. Corbett and Norman M. Fräser

[70]

Thus in an analysis of the füll lexicon, at a high level, perhaps at the level of WORD, the default gender is the neuter; at the lower level—that of nouns—it is the masculine. Of course, the lower default is much the more important: gender is a central category for Russian nouns, while for the various items (such as infinitives) covered by the higher default, it is of much less importance. This hierarchy of defaults would thus allow us to capture the Intuition that in one sense the masculine is the default gender, while in another it is the neuter. Slavists will recognise, however, that to say that the top level default is the neuter is an oversimplification. Before returning to see why this is so, we should first consider more carefully the nature of defaults, as applied to linguistic phenomena.

5. Types of default Consider the foUowing Situation. Mary and John both work for a firm in Clacton. Mary is the personnel manager and has her office in Clacton. OccasionaUy, when there are problems or training courses she spends the day at head office in Truro. By default, then, Mary works in the office at Clacton. John is a salesman. He normally spends Mondays in the south of England, Tuesdays in Wales, and Wednesdays and Thursdays in the north. If, however, clients cannot see him, or his car is unserviceable, or there is a department meeting, he goes to the Office in Clacton. Fridays he usually spends exhausted in bed, but during school holidays he goes to the office. By default, then, John also works in the office in Clacton. Intuitively the two cases are rather different. Mary is "normally" at the Office, John is not. And yet at a higher level of abstraction it is true to say that the office is the default workplace for both. It is these two types of default, both reasonable uses of the term, which have led to the differences in usage in the literature, both generally and specifically in relation to gender. In our analysis of the gender system of Arapesh (Fräser and Corbett (forthcoming) based on Aronoff (1992, 1994: 97-103) and like him following Fortune (1942)) we distinguish these two types of default which, though conceptually related, are nonetheless formally

Default genders

[71]

313

distinct. In the first type, the default accounts for the cases when "everything goes right" (as in Mary working in the office). We shall call instances of this type "normal case defaults". In the second use of the term, a default is something which applies when the normal system breaks down, when "something goes wrong" (as in John working in the office). We shall call instances of this type "exceptional case defaults". There is a common conceptual core running through both usages of the term "default": the default is the last thing you get to. However, a normal case default is retrieved after failing to find any more specific value; an exceptional case default is retrieved after findrng too much Information - Information which blocks normal retrieval and causes a backstop value to be accessed instead. One form of default is concemed with typicality, the other with exceptionality. It is therefore particularly important that conceptual and terminological confusion be avoided by proper definition of terms.12 One of the payoffs of working in a formally explicit framework such as Network Morphology is that it lays bare the differences between these otherwise confusable notions. In instances where normal case defaults apply, lexical entries are characterized by their brevity. Because a word is fairly typical, many of its parts can be left underspecified, to be filled in by default inheritance. On the other hand, there is an inverse correlation between radically underspecified lexical entries and exceptional case defaults. An exceptional case default is unlikely to apply unless the lexical entry includes some idiosyncratic Information. With all this in mind, let us retum to example (11), repeated here for convenience: (11)

Byl-o was-NEUT.SG 'It was cold.'

xolodn-o COld-NEUT.SG

There are various circumstances where similar forms are required: when there is no agreement Controller as in (11), when the Controller is a clause or Infinitive phrase and so on. It is reasonable therefore to invoke a default here. The circumstances are, roughly speaking, all those where agreement is not controlled by a prototypical noun phrase (one headed by a noun or pronoun). This can be seen as things

314

[72]

Greville G. Corbett and Norman M. Fräser

going wrong (agreement is normally controlled by a prototypical noun phrase), and so we have an exceptional case default. Consider now the forms used. In the case of the verb it is straightforwardly a neuter Singular form. But the adjective is more complex. Russian adjectives have two forms available for predicate use, the long form and the short form. The short form is being lost is most uses; however, in examples like (11), the short form is required. Thus it is not sufficient to say that these forms are neuters. For these and further reasons discussed at greater length elsewhere (Corbett 1980) we need to distinguish "neutral" agreement forms, as required for agreement with non-prototypical Controllers, from other agreement f o r m s . 13 In Russian, by default, these neutral forms are the same as the neuter. A clear case of the "neutral form" having special properties is found in Romanian, where it varies according to the particular agreement target (data from Donka Parkas, see Corbett 1991: 213-214). (12)

e

evident

cä

a

venit,

is

clear.MASC.SG

that

has

come

§i asta

o

and

this.FEM.SG

§tie knows

toatä all

it.FEM.SG

lumea the.world

'It is clear that s/he came and everyone knows this.' Here we have a clause as subject (some might prefer to say there is no subject); the predicative adjective, which has to mark agreement, is masculine (the feminine evidenta, "is unacceptable). Asta 'this' can stand for 'that s/he came' or 'it is clear that s/he came'. What concems US is that it must be feminine (the masculine a, "sta is unacceptable). Thus the form used for neutral agreement in Romanian varies according to the type of target involved. The next example includes attributive modifiers: (13)

Un

bum

putemic

a.MASC.SG.

"boom"

strong.MASC.SG

[73]

Default genders

a has

fost been

315

auzit heard.MASC.SG

'A loud boom was heard.' Here un 'a' is masculine, like the agreeing predicate. We now try the relative pronoun: (14)

a admis has admitted nu not

e is

cä that

a has

venit, come

ceea ce which.FEM.SG

surprinzätor surprising.MASC.SG

'S/he admitted that s/he came, which is not surprising.' Ceea ce is a complex relative, the first part of which shows feminine gender agreement. We thus have masculine agreement for attributive modifiers and the predicate, and feminine for the relative pronoun and the demonstrative (which replaces the personal pronoun here). However, though ceea ce is feminine, its predicate surprinzätor is masculine (*surprinzätoare feminine). This shows that ceea ce is a remarkable neutral form: though morphologically feminine, it must carry a feature to distinguish it from ordinary feminines. The reason for Ulis is the fact that its antecedent is a clause, and the evidence for the special feature is that ceea ce controls masculine predicate agreement (as clauses do). Let us try the demonstrative in a similar environment: (15)

asta e uluitor this.FEM is amazing.MASC 'this is amazing'

Here asta refers to a Situation not a specific object. While it is morphologically feminine, its predicate is masculine. Thus asta too is a special neutral form, since it controls a different agreement from the asta which can stand for a noun of feminine gender.

316

Greville G. Corbett and Norman M. Fräser

[74]

6. Possible examples of defaults Given that the notion of default appears valuable, we now sketch infonnally the areas of gender where this notion might apply.

6.1. Gender assignment Perhaps the most straightforward examples of defaults are found in gender assignment systems of the semantic type. A clear instance is found in Diyari, an Australian language which had about a dozen speakers at tiie last report, living near Lake Eyre in the north of the State of South Australia. One gender is for "all animates whose reference is distinctly female, for example, women, girls, bitches, doe kangaroos etc."; the other is for "all others, that is, male animates, non-female animates, non-sexed animates and all inanimates" (Austin 1981: 60). By default in Diyari nouns are masculine. The converse system, in which nouns denoting males are singled out as masculine and all others are feminine, occurs in Kala Lagaw Ya, the language of the westem Torres Straits Islands (Bani 1987). Here by default notms are feminine. (Note, however, that the moon is also masculine, as is generally the case in the languages of Australia.) These are obvious cases of normal case defaults.

6.2. Gender agreement Here we shall discuss three broad types of problem, which are all caused by agreement Controllers other than straightforward noun phrases. The problems arise because if a particular target type can mark agreement in gender then in many languages it must. The first type of problem is that there are constructions in which the target has to agree in gender with a Controller which is not specified for gender. The obvious examples here are those of the tj^e we have already discussed, namely the "neutral" agreement which results from agreement with non-prototypical noun phrases. The second type of problem involves cases where the choice of gender agreement in the normal way would force greater specificity than is possible (or perhaps desirable) for the Speaker. A Speaker may wish

[75]

Default genders

317

to refer to a child but be unable to select gender agreement based on sex. Finally there are complex noun phrases which are overspecified for gender. 6.2.1. Non-prototypical Controllers As noted earlier, if an agreement target can agree then typically it must agree, even if the agreement Controller lacks the appropriate features. We termed this enforced agreement "neutral agreement". The idea of "neutral agreement" can be seen as a development of Jespersen"s "conceptional neuter" (1924: 241-243); the wider term is preferred since the phenomenon is found in gender systems other than the Indo-European type, to which Jespersen restricted himself. The ränge of non-prototypical Controllers varies from language to language. It may include clauses, Infinitive phrases, nominalizations, interjections, and other quoted phrases, noun phrases in particular cases (for example, subject noun phrases in an oblique case), dummy Clements, and certain null elements (see Klajn 1984-85: 351) for examples of non-prototypical Controllers from various Indo-European languages). Languages may solve the problem of agreement with nonprototypical Controllers by pressing one of the regulär gender/number forms into Service. The form may be termed the "neutral agreement form" or the "default agreement form". The first problem which arises is the motivation for the selection of a particular form. In some languages a semantic account is plausible; thus in Russian, almost all nouns denoting humans are in the masculine and feminine genders, while inanimates are distributed across all three genders. The use of the neuter for default agreement could be understood as the selection of the gender which is most appropriate in semantic terms (thereby avoiding the semantic clash of neutral with human, which would arise with the other genders). Similar arguments can be given for languages as diverse as the Algonquian language Menominee, the Omotic language Zayse, the Niger-Kordofanian language Eula, the Mon-Khmer language Khasi, the North East Caucasian languages Archi and Khinalug and the Australian language Ungarinjin (see Corbett 1991: 206-207) The fact that a semantic explanation works for examples which are so diverse genetically might make us expect it to apply without

318

Greville G. Corbett and Norman M. Fräser

[76]

exception. Nevertheless there are languages for which the semantic criterion fails. In some we find that a different criterion is at work, but we leave others as perplexing problems. Lak is a language which, though related to Archi and Khinalug, differs from them in an interesting way. Again there are four genders, and nouns are assigned to them as foUows: I - male humans, II - some female humans (older females), HI remaining female humans, most other animates, some inanimates, IV - a very few animates, some inanimates. Again we would expect gender IV to be used for neutral agreement, as in Archi and Khinalug; in fact gender III is used (Kibrik 1979: 13): (16)

g,anisa

d-uc'an

she n-come 'she can come'

b-uqlHaj

ni-can

b-u-r

III-AUX.3RD

The gender II agreement marker d- on the Infinitive d-uc'an is controlled by a deleted subject noun phrase g^a, coreferential with g^anisa. The point of interest is the agreement on the modal and on the auxiliary; both have a gender HI prefix, since they are controlled by the infinitive. Thus gender HI is used for neutral agreement and there is no evident explanation in terms of the semantics of the Controller genders as to why this should be (gender IV would be expected). However, as Aleksandr Kibrik points out (personal communication), an explanation is available if we take into account the pattem of the target gender forms, given in Table 3. Table 3. Gender agreement in Lak SINGULAR

PLURAL

gender I (male humans)

0/w

b/w

gender II (some female hmnans - older)

d/r

b/w

gender III (most remaining animates, some inanimates)

b/w

b/w

gender IV (lesidue - largely inanimates)

d/r

d/r

Forms before the slash are prefixal, those after it are internal or suffixal. As can be seen in Table 3, the forms for gender IV and gender n are identical in the Singular. The use of the gender IV marker d/r for neutral agreement is ruled out by the fact that this marker is

[77]

Default genders

319

also the marker for gender n, which contains only nouns denoting females. This leaves b/w, the gender i n agreement form as the only possibility for neutral agreement. This particular syncretism is not found in Archi and Khinalug, and so in those languages the gender IV forms are available for neutral agreement. Thus far we have found an explanation for the choice of neutral form in several languages, according to the semantics of the Controller genders (with additional complications in the case of Lak). We now tum to cases which are initially surprising and for which the semantics of the Controller genders appears to provide no help. But in these cases we find an explanation available in terms of the gender agreement forms. For example, the East Cushitic language Bayso has two genders, masculine and feminine, and uses the masculine agreement forms for neutral agreement (Corbett and Hayward 1987: lln): (1-7)

ihaaddo person

boc,aano to beat

ka - meelan - ya PARTICLE.MASC - bad - C0P.3RD.SG.MASC

To beat people is bad.' In the last item, ka- is the associative particle in the masculine form and -ya is the clitic copula, third Singular masculine. In Qafar, another Cushitic language, it is the feminine which is used. This can be shown with the so-called 'm-nominalization'. The clitic element -m attaches to a wide ränge of items, and the resulting form takes feminine agreement, despite ending in a consonant and so being of a phonological shape associated with masculines: (18)

gaddäli

kinnim

ydt

wealthy.man

is. 3RD.SG + m

to.me

celta seems.3RD.SG.FEM

'He seems to me to be a wealthy man.'

320

Greville G. Corbett and Norman M. Fräser

[78]

Here the -m is attached to the phrase 'is a wealthy man' and this new dement is a non-prototypical Controller; as a result, the verb celta stands in the neutral form, which is the feminine (Hayward and Corbett 1988:266-268). There are no obvious differences in the semantics of gender in the two languages to explain the different choice of neutral form (R. J. Hayward, personal communication). However, if we look at the morphology of agreeing forms there is a ready explanation. Table 4 gives the forms of the associative particle for the two of the four numbers of Bayso which are significant for differentiating gender, and the verbal agreement markers of Qafar which are found in agreement with simple noun phrases. Table 4. Pattems of syncretism in Bayso and Qafar Bayso

MASCUUNE FEMININE

Qafar

SINGULAR

PLURAL

ka

ka

ta

ka

SINGULAR y/0 t

PLURAL t t

There is considerable syncretism: one form Covers three of the four theoretical slots, being used for the Singular of one gender and the plural of both. And in each case it is this form, the one with the wider ränge of use, which is the neutral agreement form. Thus the neutral form is here aligned with a default inflectional form. It is Worth clarifying the criteria we have employed so far. In the first languages discussed it was the semantics of the Controller genders—the genders into which nouns are divided—which provided the reason for the choice of neutral agreement form. In these languages there is a gender for abstracts or inanimates, or at least one containing few if any nouns denoting humans, which appeared an appropriate gender for non-prototypical Controllers to attach themselves to. In the case of Bayso and Qafar, however, there is no obvious choice in terms of the semantics of the Controller genders, since humans and other animates are found in both of the available genders, as are inanimates. Here a second criterion comes into play: the choice of the neutral form can be understood in terms of the morphology of the agreeing elements (target gender forms). Lak can be Seen as an intermediate type, in which the semantics of the Controller

[79]

Default genders

321

genders and the pattern of syncretism of the target gender forms interact to determine the neutral agreement form. Some instances are raore difficult, as in the Bantu language Chichewa: (19)

"aaa"

a-na-mv-

eka

"aaa" 1 - FAST - hear - PASSIVE 'an "aaah" was heard' Reraarkably, class 1 agreement (as for humans) is used here, as the agreement marker prefixed on the verb suggests; to demonstrate this conclusively requires an example with object agreement (Corbett and Mtenje 1987:14): (20)

a - na-mumva l.SUBJ - FAST - l.OBJ

- hear 'he heard a crying sound'

"rmyo"

crying sound

The combination of a- as subject agreement marker in (19) and -muas object agreement marker in (20) means that we are dealing with class 1 markers. What makes the Chichewa case particularly surprising is that there is a special agreement form for infinitives, and this is not used for neutral agreement. Though neutral agreement forms may appear (in some languages) to be identical to some other form, they are usually odd in some ways. Thus they typically appear identical to Singular markers, but they lack plural counterparts. This can be illustrated by conjoining. Moreover certain target types may be avoided. And as we saw in Romanian, the form to be used can vary according to the target type. Some languages have unique neutral agreement forms (examples are Spanish, Portuguese, the Surselvan dialect of Romansh, Uixainian and the Sele Fara dialect of Slovene). However, no language has yet been found with a füll set of unique neutral forms: regulär gender/number forms are used for some targets. An interesting development occurs when neutral forms are used when the Controller is an apparently straightforward noun phrase. This phenomenon is well attested in Scandinavian languages (for references see Corbett 1991: 216 and for extensive discussion see Källström 1993: 188-246 and Hedlund 1992: 95-111). Our example is.Norwegian, taken from Faarlund (1977). Norwegian predicative

322

Greville G. Corbett and Norman M. Fräser

[80]

adjectives distinguish Singular from plural and, in the Singular, neuter from common: (21)

Ein a

ny new

utanriksminister foreign secretary

sä

dumt.

so

stupid.NEUT.SG

ville would

ikkje not

vere be

'A new foreign secretary would not be a bad idea.' The Interpretation is that having a new foreign secretary would not be a bad idea. If the adjective were in die common form dum, then it would agree directly with the subject noun phrase and the Interpretation would be less complimentary. Default agreement then may be required in a ränge of circumstances: agreement with a clause as subject, with an absent subject, and so on. Note that the ränge of circumstances varies from language to language. In Russian, as in many other Indo-European languages, an Infinitive phrase requires default agreement, while in Chichewa, and many other Bantu languages, there is a special gender for infinitives and the default is not invoked here. We have seen too that the default form overlaps to a greater or lesser extent with "normal" forms, and that the gender form with which it overlaps can be successfully predicted in some languages at least. In these instances a careful analysis would be required to determine which are the normal case defaults and which are the exceptional case defaults. Before moving on to our second major type of agreement default, we wiU consider two further types of case which might be expected to fit easily under neutral/default agreement but which in fact do not. The first concems missing antecedents of pronouns, and is a problem raised by Tasmowski-De Ryck and Verluyten (1981; 1982). Consider the foUowing French example. John is trying to get a large table into the boot of his car. Mary says: (22)

Tu You entrer enter

n'arriveras NEG.will.manage dans in

la the

jamais never voiture car

'You'll never get it into the car.'

ä to

la itFEM

faire make

[81]

Default genders

323

Here the pronoun must be in the feminine form la (*le is unacceptable); table 'table' is feminine in French. However, if John were trying to get a desk into bis car, then the pronoun would be masculine (bureau 'desk' is masculine). The pronoun must be in the appropriate gender; since there is no antecedent present, the question is the source of the gender of the pronoun. These data have been one of the stimulants to an interesting debate on the nature of deixis and anaphora (see Corbett 1991: 244 for extensive references). Most agree that the gender of the pronoun in examples like (22) is determined by that of the default description of the referent in question. Normally this default description wül be the basic level term, a notion which comes from Rosch (see, for example, Rosch 1978, and for discussion see Pulman 1983: 83-106). The basic level is the appropriate level. In (22) above, the basic level term for the object in question is table, which is feminine, hence the use of la. The fact that it is a piece of fumiture {meuble, masculine) and indeed an object {pbjet, masculine) does not affect the gender since neither are basic level terms (Bosch 1987: 73). Hence we have another, rather different, sort of default, and one which need not coincide with the neutral agreement type. The second type concems citation forms and other uses where there is no possible antecedent. For instance, if speakers wish to discuss a particular word, in a language where items of that class always agree in gender, then they will have to use one of the gender forms. A specific instance of this general problem is the use of numerals in abstract counting. It would surely be a reasonable prediction, that in such instances the neutral form (as for, say, impersonal sentences) would be used. Reasonable but not necessarily true. If counting in the abstract, a Russian Speaker says dvadcat' odin 'twenty-one.MASC', even though the gender-differentiating odin 'one' has a neuter form (odno). We might also have expected that older defaults would be preserved in fossilized form in complex numerals like odinnadcat' 'eleven' and dvenadcat' 'twelve'. But again the Situation is more complex: in odinnadcat' 'eleven' the masculine odin is preserved, while in dvenadcat' 'twelve' we apparently have the feminine dve, though the form dates from a period when dve was the form for neuter as well as feminine (see Suprun 1969: 51-52 for an argument for dvenadcat' based on the number of examples of use with feminine noims (including weights and measures) and Comrie 1992: 768 for discussion).

324

Greville G. Corbett and Norman M. Fräser

[82]

6.2.2. Reference problems Even if the agreement Controller is a noun phase headed by a noun er pronoun, there may still be problems involving gender agreement, caused by reference difficulties. These have mainly been investigated relative to human referents, though there can be similar problems (usually in larger gender systems) with non-humans. There are at least three types of problem (it is not even clear whether all the different types have yet been identified): the gender required may be unknown, unclear or mixed. Suppose we ask Who said that?, in a language which requires agreement in gender on the verb. In this first type we cannot determine the gender, since we cannot identify the referent of who; thus the gender required is unknown. Similarly when we ask What was that? we may have theoretically possible referents of more than one gender. As a variant of this type we may have a noun, like English manager or friend, which can be used of a person of either sex. Again we may not know the sex of the referent. Second there are cases where the gender required is unclear because the referent is non-specific; here there are various possibilities: I f a patient wishes to change doctors, he/he or she/they should advise the receptionist. A third area of difficulty here is agreement with a noun denoting a group of referents which would separately be referred to with nouns of different genders. The most obvious examples involve humans of both sexes (villagers, athletes). Here again the sex cannot be uniquely determined, but if the language distinguishes gender in the plural, then clearly one form must be selected for agreement purposes. We shall see that there are two main approaches to dealing with these problems. First, one of the possible alternative agreement forms may be used by Convention—an obvious type of default. If the "reasonable possibilities" are genders A and B, then either A or B is chosen. The second possibility is for an "evasive" form to be used. If the "reasonable possibilities" are genders A and B, then gender C is chosen. It is often assumed that in a Single language, all problem types are dealt with in the same way (for example, it may be stated or implied that a particular gender is the unmarked or default one and so used in all these cases). But in fact languages may handle the three parts of the problem differently. This is an area where there has been a good

[83]

Default genders

325

deal of research on one small part of the topic but where much of the problem is only poorly understood. Consider first the case where the appropriate gender is unknown. Suppose we have a language in which there is at least a masculine gender (containing nouns denoting males, and other nouns) and a feminine gender (for females and other nouns). For the problem cases above, one set of target gender forms, say the masculine set, could be used by Convention. This Situation is found in many IndoEuropean languages. Let us take Russian examples: (23) Kto eto sdelal? who this did.MASC •Who did this?' The Speaker does not know the sex of the person responsible, but the masculine is used. Surprisingly, even in a setting in which the person must be one of a group of women, masculine agreement is still normal. Though the literature might suggest otherwise, it is not the case that the masculine is always used. In the Nilotic language Maasai, we find the feminine used for questions when the person involved could be male or female (Tucker & Mpaayei 1955: 27). Retuming to Russian, nouns which can denote a male or a female, like vrac 'doctor', take masculine agreements if the sex is not known. In Archi, however, we find an "evasive" form. Archi has four genders, I and n for humans, male and female, III and IV less clearly defined semantically but with the larger animates in III and most abstracts in IV. In Archi, nouns like lo 'chüd', adam 'person', c'ohor 'thief, misgin 'poor person' take gender IV agreements in the Singular if the sex of the referent is unimportant or unknown (BQbrik 1972: 126). Archi shows a particularly clear example of an evasive form, since gender IV does not contain any nouns denoting h u m a n s . 14 The second type of problem, non-specific referents, has created a considerable literature, but generally with reference to a small number of languages. Languages like Russian use the masculine here (see, for instance, Doleschal 1993: 39). The third problem involves mixed groups. Usually mixed groups of humans are investigated, but in large gender systems there could be analogous problems with inanimates. Given, however, a mixed group of humans, in Serbo-Croat we find the masculine plural oni 'they' in such cases. We may take the problem back into derivational

326

Greville G. Corbett and Norman M. Fräser

[84]

morphology: Amerikanac (masculine) is a male American, while Amerikanka (feminine) is a female American in Serbo-Croat. To refer to Americans in general, the plural of the masculine noun is used, that is, Amerikanci, and it takes masculine plural agreements. This instance of the way in which gender is assigned to nouns denoting mixed groups links directly to the analysis of agreements used with conjoined noun phrases, which we consider in the next section. The "opposite" system is found in the Khoisan language Dama, spoken in northem Namibia; here mixed groups of people are referred to using the feminine pronouns (John Payne personal communication). Here too "evasive" forms are an alternative strategy. Polish for instance uses the neuter Singular. This usage is described by Gotteri (1984), who took up the term "evasive" foUowing a suggestion by Doroszewski. An example of the Polish neuter in evasive use is the foUowing: (24)

Ktör-es one-NEUT winn-e guilty-NEUT

z from

matzonköw spouses

zarzucanej imputed

mu it.DAT

jest is zbrodni crime

'One of the spouses is guilty of the crime he or she has been accused of.' Maizonkowie is masculine personal and means 'husband and wife'; when either the husband or the wife is potentially the referent, then the evasive neuter is used. The neuter cannot be used in all the situations we have considered; in most the masculine is used (for examples see Herbert and Nykiel-Herbert 1986: 67). Most interestingly, the evasive neuter seems to be used in the sort of contexts which also preclude the use of generic he in English, that is where there are implied disjuncts, one of which is specifically female. Weiss (1993) also discusses the Polish Situation. The evasive neuter is found in Serbo-Croat too, but in rather limited use. Though a füll analysis would be required in order to be certain, it would appear that in this section, the examples that use one of the expected genders should be treated as normal case defaults, while the evasive forms are exceptional case defaults.

[85]

Default genders

327

6.2.3. Gender resolution Our last type involves Controllers with too many gender specifications, namely those consisting of conjoined noun phrases. Languages with gender agreement in the plural require rules to determine the form to be employed with conjoined noun phrases, and these rules are termed "resolution rules". As an example, we take Tsova-Tush, sometimes called Bats or Batsbi, which is a Nakh language (a subgroup of North-East Caucasian). It has some 3,000 speakers in Zemo Alvani in Kakheti (in eastem Georgia); all are bilüigual with Georgian, and chüdren are no longer leaming the language. The data come from Dee Ann Holisky (forthcoming, and unpublished field notes).i5 Tsova-Tush has five main genders, and the target agreement forms make up a complex crossed system (shown in Figure 3; next page). Nouns which take v - b (v in ihe singular and in the plural) denote male humans and those which take j - d denote female humans. Thus far gender is semantic, and predictable. The basis for the remaining three major genders has not been established. There are just over 20 nouns which do not fit into the five main genders (Holisky forthcoming: see Corbett 1991: 171-172 for discussion of their status). They belong to three further classes: agreements b-j are taken by 15 nouns, all of which denote body parts (such as 'leg' and 'throat'), a further four nouns, also denoting body parts (this time all paired, like 'ear' and 'hand') take d - j, while three nouns take b - b (they are borag 'knit Slipper', c'ekam 'boot', and kakam 'wool cut in autumn'). Singular

plural

V

b

j

j

d

—

^

Figure 3. The main genders in Tsova-Tush

d

328

Greville G. Corbett and Norman M. Fräser

[86]

The gender resolution examples are straightforward, though unusual.16 If all conjuncts are headed by nouns denoting male humans, then the male human form (b-) is used (Holisky forthcoming, example (74)): (24)

mit'o-E p'et'o-E he"vazar-i Mito-and Peto-and 2SG.GEN brothers-INT 'Are Mito and Feto your brothers?'

b-a AG-be

Conjoining is marked by -E on each conjunct; variant forms are for phonological reasons (Holisky forthcoming: footnote 4). E is a reduced vowel; h is a voiceless pharyngeal fricative and e° is a nasalized vowel. INT indicates the question marker, and AG is for agreement marker. (Holisky (fieldnotes) reports checking various instances with conjuncts referring to male human and female human; in these mixed cases the d- form was used.) In all other circumstances (when not all conjuncts are headed by nouns denoting male humans) the d- agreement form is used (the form which is appropriate for female hiunans, and for two of the remaining major genders). There is a large number of possible combinations: female human plus female human, female human plus other, and the various types of noun not denoting humans, including those from the smaller genders. Here are just a few examples. The first has two nouns from the largest gender, d - d (Holisky fieldnotes): (25)

kotama-E mamala-E d-a chicken-and rooster-and AG-be 'a chicken and a rooster are: It is a chicken and a rooster.'

The next example has two nouns from the large j - j gender (Holisky field notes; in (26) g represents a voiced velar fricative, and E is again a reduced vowel): (26)

goba-E ezo-E d-a fence-and yard-and AG-be 'a fence and a yard are: It is a fence and a yard.'

In the next example (Holisky forthcoming example (26b)) we have two nouns from the small genders b - j and b - b:

[87]

(27)

Default genders

329

kok'a-E borga-E d-a leg-and knit slipper-and AG-be 'a leg and a knit slipper are: It is a leg and a knit slipper.'

The resolution rules are again based on semantics: 1. By default, the gender d- form is used. 2. This is overridden if aU conjuncts are headed by nouns which denote male humans; in this case the male human form (b-)'xsused. For a typology of resolution rules, see Corbett (forthcoming). We might expect resolution rules to be examples of normal case defaults. However, there are systems of some complexity in which it is at least not obvious that they can be handled in this way.

7. Correlation of defaults It is important to note that grammatical defaults of different types may or may not line up together (and when there are only two possibilities, as with two-gender systems, then the coincidence cannot be assumed to be of any great significance). Let us retum to TsovaTush. We saw that the default form in gender resolution is the dgender. Of the numerals in Tsova-Tush, just the numeral 4 (and numbers which contain it) agree. When counting in the abstract, it takes the d- agreement form (Holisky forthcoming, section 2.8). The dgender is the largest one in terms of the number of nouns included and so a reasonable candidate for the default for gender assignment (section 2.1.1.), and verbal nouns, which might be thought of as nonprototypical nouns, are assigned to this gender. Consider finally what happens when the appropriate gender is unknown: (28)

vux d-a? what AG-is 'What is it?"

Again the d- gender form is used. Thus we have a clear example of several defaults lining up.

330

Greville G. Corbett and Norman M. Fräser

[88]

An equally impressive case is Arapesh, as described by Aronoff (1992, 1994: 97-103, following Fortune 1942). The same default gender is used for null heads (1992: 26), interrogatives, including the interrogative for humans when sex is not known (1992: 27), for agreement with conjoined noun phrases when the head nouns are of different gender (1992: 27). Nouns which fall outside the regulär gender assignment rules are assigned to this same default gender (1992: 28) as are "words that designate persons in a sex-neutral fashion" (1992: 29). These examples from large gender systems might lead us to expect that gender defaults regularly line up. However, there are other examples where this is certainly not the case. Moreover, we can show this even in relatively small systems. If we retum to the Russian data discussed in detail earlier, we recall that at one level the default gender is neuter, and at another it is masculine. Then for nouns of particular inflectional classes the default is feminine. Russian of course has three genders, but even in a two-gender system the defaults need not line up. Recall that in Kala Lagaw Ya nouns are assigned by default to the feminine gender (only nouns denoting males are masculine). However, for a Single human of unknown sex, the masculine is used (Alpher 1987: 173).

8. Conclusion The notion of default is of considerable potential value in the analysis of gender systems; indeed it could be so valuable that it is important to be clear in its use. When we attempt to cover a complete system we find that the defaults at different levels may well be different. Just as they may line up fairly consistently, they may also diverge, even in small systems. Since they can diverge, we need more descriptions of complete systems, so that we can attempt a typology of configurations of defaults in gender systems. Notes 1.

Addresses for coirespondence: Greville Corbett, Department of Linguistic and International Studies, University of Surrey, Guildford, Surrey, GU2

[89]

2. 3.

4.

Default genders

331

5XH, UK. Nonnan Fräser, Vcwalis Ltd, Chaston House, Mill Court, Cambridge CB2 5LD,UK. email: [email protected]. [email protected]. This research was supported in part by the Economic and Social Research Council (grants R000233633 and R0(K)236063) and the Leverhuhne Trust (grant F.242M); this support is gratefully acknowledged. Some of the data in this paper have been presented previously: sections 3 and 4 use material from Fräser and Corbett ^995) whUe section 6 takes data from Corbett (1991, chapter 7). What is new here is the central attention given to the notion of default, and the attempt to understand familiar data in this light. In addition, in the sections based on earlier work, new references to recent work have been added. A version of this paper was read at the Workshop 'Approaches to Gender' held at the Forschungsschwerpunkt Allgemeine Sprachwissenschaft, Typologie und Universalienforschung, Berlin 26-28 May, 1994. We are grateM to those present for interested discussion, and especially to Wolfgsmg Wurzel and David Zubin for helpfid suggestions at the meeting, and to Osten Dahl, Ursula Doleschal, Gerald Ga/dar and Andrew Hippisley for subsequent useful comments. For further discussions of markedness see Schwartz (1980) and Greenberg (1988). The selection of DATR as the language of formal expression for Network Morphology is no accident. The theoretical framework emerged out of our exploratory work with DATR, infonned by our earlier experience of default inheritance in Artificial Intelligence and in Word Grammar (Fräser and Hudson 1992). DATR offers a conceptually straightforward syntax and a well-defmed semantics. In addition, the existence of a number of Computer Interpreters for the language allow theories encoded in DATR to be validated automatically. The following automatic phonological correspondences are assumed: a) IM is retracted to its allophone [i] after non-back hard (unpalatalized) consonants. Thus the nominative plural form /zakoni/ will be realized with [i] but /kost'i/ retains [i] since [t'] is soft. b) All consonants which can be palatalized are automatically palatalized before /e/. Thus the locative Singular of /zakon/, namely /zakone/, will be realised with a palatalized [n^. If the consonant is already palatalized as in genitive plural tost'-ej/, it simply remains palatalized. Some consonants are always hard (/§, z, cf), and remain so before Id. On the other hand, /c7 and /sc7 are always soft (palatalized), and naturally remain so before /e/. We have chosen to mark softening redundantly for greater clarity in this instance. In addition, the gutdirals /k, g, x/ are palatalized before N, so that the genitive form /knigi/, from /kniga/ 'book', will be realized with palatalized [gT (which then demands the front allophone [i]). c) There are complex pattems of reduction of vowels in unstressed position, which can safely be omitted from the transcription since our focus is on morphology. In particular, the unstressed /o/ ending of nouns with soft stems such as /pol'o/ 'field' (orthographically pole) is realized as either [a] or [i]. Although this /o/ is never realized as anything approximating to a mid

332

5. 6. 7.

8. 9.

10.

11.

12.

13. 14.

Greville G. Corbett and Norman M. Fräser

[90]

rounded back vowel, positing /o/ is justified by the stressed [6] which occursin [v'inö] 'wine' and [p'it'jö] 'drink(ing)'. For an informative sketch of Kassian phonology, see Timberlake (1993: 828-832). We do not examine the complexities of hybrid nouns here, for which see Corbett (1991:39, 231-232), and for into-esting data see Weiss (1991) and Doleschal (1992; 1993:40-46,138-144). There is a small number of exceptions, for which see Smimova (1979). We are not concemed here with the special case of acronyms where derivational infonnation may be relevant. A more extended account, to include cases where nouns may head noun phrases referring to either sex, would require us to State the first equaüon as: GENDER: == masc. Instances of this kind would be correctly assigned masculine gender so long as the indetenninate sex of the referent were idenüfied by an explicit marker such as either. There is no reason why such a phonological component could not also be fonnalized using DATR, as the work of Reinhard and Gibbon (1991) and Gibbon (1992) demonstrates. While the claim relating to gender assigment is a general one, we have not investigated declensional-class assigimient beyond Russian. It is a plausible hypothesis that in declensional class assignment generally, formal factors wUl take precedence over semantic. If the stem ends in a hard (not palatalized) consonant, then type III can be eliminated. We have no rule to predict membership of class III; stems are by default "hard" (they end in a hard consonant). The HI class is relatively small (see discussion after Table 2). This figure includes the small number of nouns (32 in Lazova's table) which are indeclinable but which do not end in a vowel. They are typically borrpwings. Borrowings like miss 'miss', which denote femdes but end in a consonant are indeclinable. These require an exceptional marker just to indicate that they are indeclinable: the semantic gender assignment rule will then correctly assign them to the feminine gender. Rare borrowings which are indeclinable, consonant final, do not denote an animate, and are masculine or feminine, require two irregulär markers (for declensional class V, and for gender). These marginal cases tend to be integrated over time or lost. An important pq)er which includes consideration of what we would call an exceptional case default is Marcus—^Brinkmann—Clahsen—Wiese— Woest—^Pinker (1993) where -s is discussed as the default plural of German. They quote Van Dam (1940) who called it the Notpluralendung 'emergency plural ending'. See also Clahsen—^Rothweiler—Woest—Marcus (1992). For discussion of the forms used in English for "Situation reference" see Fraurud (1992). Another evasive device is to use the plural, if gender is not distinguished there. Thus Alamblak (a Sepik Hill language of Papua New Guinea) distinguishes masculine and feminine in the Singular, but not in the dual or

[91]

Default genders

333

plural. If "the Speaker is either unable or unwilling lo indicate the gender of an object" (Bruce 1984: 98) then tbe third plural is used: (i)

15.

yen-m

heawrahtm

indom

yamtn

child-3PL she.will.bear.them anodier month.in 'She will bear a child in anotber month.' We are very grateful to Dee Ann Holisky for making her notes available to US.

16.

There is, however, an interesting complication with conjoined plurals, which would take the same target gender form (either j or b). There are a couple of cases of an Informant offering this form (the plural which each plural conjunct would take individuaUy). For discussion of the same phenomenon in Chichewa and in Serbo-Croat see Corbett and Mtenje (1987: 2026). It is argued that in these languages gender resolution is triggered only if there is a clash of genders or if another resolution rule (number or person) operates - since if one type of resolution rule operates, aU available resolution rules must operate.

References Alpher, Barry 1987

"Feminine as the unmarked grammatical gender: buffalo girls are no fools", Australian Journal ofLinguistics

Aronoff, Mark 1992 1994

7:169-187.

"Noun classes in Arapesh", in: Geert Booij—^Jaap van Marie (eds.). YearbookofMorphology 1991. Dordrechf Kluwer. 21-32. Morphology by itself: Sterns and inflectional classes. (Linguistic

Inquiry Monograph 22). Cambridge, MA: MIT Press. Austin, Peter 1981

A grammar ofDiyari,

South Australia. Cambridge: Cambridge

University Press. Bani, Ephraim 1987

"Garka a ipika: masculine and feminine grammatical gender in KalüLagimYa.",

Australian Journal ofLinguistics,

7:189-201.

Bosch, Peter 1987

"Pronouns under control? A Reply to Liliane Tasmowski and Paul Verluyten", Journal ofSemantics 5:65-78. Briscoe, Ted -Valeria de Paiva—Ann Copestake (eds.) 1993

Inheritance,

defaults and the lexicon. Cambridge: Cambridge

University Press. Brown, Dunstan P.—Andrew R. Hippisley 1994 "Conflict in Russian genitive plural assignment a Solution represented in DATR", Journal ofSlavic Linguistics 1 (1): 48-76. Bruce, Les 1984 The Alamblak language of Papua New Guinea (East Sepik) (Pacific Linguistics, Series C, No. 81). Canberra: Department of

334

Greville G. Corbett and Norman M. Fräser

[92]

Linguistics, Research School of Pacific Studies, Australian National University. Calder, Jo 1994

"Feature-value logics: some limits on the role of defaults", in: CJ. R u p p — R o s n e r — R . L . Johnson (eds.) Constraints, Language and Computation. London: Academic Press, 205-222. Clahsen, Harald—^Monika Rothweiler—Andreas Woest—Gary F. Marcus 1992 "Regulär and irregulär inflection in the acquisition of German nounplurals", Cognition, 45: 225-255. Comrie, Bemard 1992 "Balto-Slavonic", in: Jadranka Gvozdanovi6 (ed.), Indo-European Numerais (Trends in Linguistics. Studies and Monographs, 57). Berlm: Mouton de Gruyter. 717-833. Corbett, GrevüleG. 1979 Predicate agreement in Kassian (Birmingham Slavonic Monographs 7). Birmingham: University of Birmingham. 1980 "Neutral agreement", Quinquereme - New Studies in Modern Languages 3:164-170. 1982 "Gender in Russian: an account of gender specification and its relationship to declension", Russian Linguistics 6: 197-232. 1991 Gender. Cambridge: Cambridge University Press, (forthcoming) "Types of typology, illustrated from gender systems", to appear in: Frans Plank (ed.). Noun phrase structure in the languages of Europe. Berlin: Mouton de Gruyter. Corbett, Gtevüle G.—^Norman M. Fräser 1993 "Network morphology: A DATR account of Russian Inflectional Moiphology", Journal of Linguistics, 29:113-142. Corbett, Greville G.—R. J. Hayward 1987 "Gender and numb» in Bayso", Lingua, 73: 1-28. Corbett, Greville G.—Alfred D. Mtenje 1987 "Gender agreement in Chichewa", Studies in African Linguistics, 18.1:1-38. Daelemans, Walter—Koenraad De Smedt—Gerald Gazdar 1992 "Inheritance in natural language processing", Computational Linguistics, 18:205-218. Doleschal, Ursula 1992 "Genus und (Ko-)Referenz", in: Tihnann Reuther (ed.). Slavistische Linguistik 1991: Referate des XVII. Konstanzer Slavistischen Arbeitstreffens, Klagenflirt - St. Georgen/ Längsee, 10.-14.9.1991 (=Slavistische Beiträge 292). München: Otto Sagner. 123-135. 1993 Genus als grammatische und textlinguistische Kategorie: Eine kognitiv-funktionalistische Untersuchung des Russischen. [Unpublished doctoral dissertation, University of Vienna.]

Default genders

[93]

335

Dziwirek, Katarzyna 1990 "Default agreement in Polish", in: Katarzyna Dziwirek—Patrick Fairell—Enapel Majfas-Bikandi (eds.), Grammatical relations: A cross-theoretical perspective. Stanford: Stanford Linguistics Association / CSU. 147-161. Evans, Roger—Gerald Gazdar 1989a. "Inference in DATR", Proceedings ofthe 4th Conference ofthe European Chapter of the Association for Computational Linguistics. Manchester, England. 66-71. 1989b. "The semantics of DATR", in: A. G. Cohn (ed.). Proceedings of the Seventh Conference ofthe Society for the Study ofArtificial Intelligence and Simulation of Behaviour. London: Pitman/ Morgan Kaufinann. 79-87. 1995 DATR: A language for lexical knowledge representation. University of Sussex. (=Cognitive Science Research Paper 382). School of Cognitive and Computing Sciences, University of Sussex. Faarlund, Jan Teqe "Embedded clause reduction and Scandinavian gender agree1977 ment", Journal ofLinguistics 13: 239-257. Fortune, Reo F. Arapesh. (Publications of the American Ethnological Society 19). 1942 New York: J. J. Augustin. [1977] [Reprinted, New York: AMS Press.] Fräser, Norman M.--Greville G. Corbett 1995 "Gender, animacy, and declensional class assignment: a unified account for Russian", in: Geert Booij—^Jaap van Marie (eds.). YearbookofMorphology 1994. Dordrecht: Kluwer. 123-150. Fräser, Norman M.—Greville G. Cortjett (forthcoming) Defaults in Arapesh. To appear in Lingua. Fräser, Norman M.—^Richard A. Hudson 1992 "Inheritance in Word Grammar", Computational Linguistics 18: 133-158. Fraurud, Karl Situation reference (what does 'it' refer to?). GAP Working Paper 1992 24. Hamburg: Fachbereich Informatik, Universität Hamburg. [1992] [Reprinted in: Kari Fraurud, Processing noun phrases in natural discourse. Doctoral dissertation, Department of Linguistics, Stockhohn University.] Gazdar, Gerald "Linguistic applications of defauh inheritance mechanisms", in: 1987 Peter Whitelock—Mary McGee Wood—Harold L. Somers—Rod L. Johnson—^Paul Bennett (eds.), Linguistic theory and Computer

1990

applications. London: Academic Press. 37-67. "An introduction to DATR", in: Roger Evans- -Gerald Gazdar (eds.), The DATR Papers. Cognitive Science Research Paper

336

Greville G. Corbett and Norman M. Fräser

[94]

CSRP 139. School of Cognitive and Computing Sciences, University of Sussex. 1-14. 1992 "Paradigm function morphology in DATR", in: L. J. Cahill— Richard Coates (eds.), Sussex Papers in General and Computational Linguistics (Cognitive Science Research Paper 239). Brighton: University of Sussex. 43-53. (forthcoming) "Ceteris paribus", to appear in: J. A. W. Kamp—C. Rohrer (eds.), Aspects ofcomputational linguistics. Berlin: Springer. Gazdar, Gerald—^Ewan Klein—Geofifrey K. PuUum—^Ivan A. Sag 1985 Generalized Phrase Structure Grammar. Blackwell: Oxford Gibbon, Dafydd 1992 "LEX: a linguistic approach to computational lexica", in: U. Klenk (ed.), Computatio Linguae: Aufsätze zur algorithmischen und quantitativen Analyse der Sprache {=Zeitschriftßr Dialektologie und Linguistik. Beiheft 73). Stuttgart: Franz Steiner Verlag. 32-53. Gotteri, Nigel 1984 "The evasive neuter in Polish", in: F. E. Knowles—^J. I. Press (eds.), Papers in Slavonic linguistics II: 1-8. Bimiingham: Department of Modem Languages, University of Aston in Birmingham. Greenberg, Joseph H. 1966 Language Universals: with special reference to feature hierarchies. The Hague: Mouton. 1988 "The present status of markedness theory: a reply to Scheffler", Journal ofAnthropological Research, 43: 367-374. Hayward, Richard J. 1989 "The notion of 'default gender': a key to interpreting the evolution of certain verb paradigms in East Ometo, and its implications for Omotic", Afrika und Übersee, 72:17-32. Hayward, Richard J.—Greville G. Corbett 1988 "Resolution rules in Qafar", Linguistics 26: 259-279. Hedlund, Cecilia 1992 On participles. Doctoral dissertation, Department of Linguistics, University of Stockholm. Distributed by Akademitryck AB, Edsbruk. Herbert, R. K. —Nykiel-Herbert, B. 1986 "Explorations in linguistic sexism: a contrastive sketch", Papers and Studies in Contrastive Linguistics, 21:47-85. Holisky, Dee Ann (with Rusudan Gagua) (forüicoming) "Tsova-Tush (Batsbi)", to appear in: Rieks Smeets (ed.), Indigenous Languages of the Caucasus III. Delmar, NY: Caravan Books.

[95]

Default genders

337

Ilola, Eeva—Arto Mustajoki 1989 Report on Russian Morphology as it appears in Zaliznyak's Grammatical Dictionary (=Slavica Helsingiensia 7). Helsinki: Department of Slavonic Languages, University of Helsinki. Jackendoff, Ray 1975 "Morphological and semantic regularities in the lexicon", Language, 51: 639-671. Jespersen, Otto 1924 The Philosophy ofGrammar. London: Allen and Unwin. KäUström, Roger 1993 Kongruens i svenskan [Congnience in Swedish] (Nordistica Gothoburgensia 16) Göteborg: Acta Universitatis Gothoburgensis. Kibrik, A. E. 1972 "O formal'nom vydelenii soglasovatel'nyx klassov v aränskom jazyke", [The formal identification of agreement classes in Aichi.]Voprosyjazykoznanija, 1:124-131. 1979 Materialy k tipologii ergativnosti: 2. Lakskij jazyk, 3. Öiragskij jazyk . [Materials for a typology of ergativity: 2. Lak, 3. Chirag.] (Predvaritel'nye publikacii 127). Moskva: Institut russkogo jazykaANSSSR. KJajn, Ivan 1984-85 "On conceptual neuter", Zbornik Matice Srpske za filologiju i lingvistihi (Novi Sad). 27-28: 347-354. Lazova, M. V. (ed.) 1974 Obratnyj slo\ar' russkogo jazyka: okolo 125 000 slov. [Reverse dictionary of Russian: about 125, 000 words.] Moskva: Sovetskaja finciklopedija. Marcus, Gary F.—^Ursula Brinkmann—^Harald Clahsen—Richard Wiese—Andreas Woest—Steven Pinker 1993 German inflection: the exception thatproves the rule. Theorie des Lexikons: Arbeiten des Sonderforschungsbereichs 282, number 40. Düsseldorf: Heinrich-Heine-Universität, Seminar für Allgemeine Sprachwissenschaft. Mucnik, I. F. 1971 Grammaticeskie kategorii glagola i imeni v sovremennom russkom literatumom jazyke. [The grammatical categories of the verb and the noun in contemporary Standard Russian.] Moskva:

Nauka. Pulman, Stephen G. 1983 Word Meaning and Belief. London: Croom Helm. Reinhard, Sabine—^Dafydd Gibbon 1991 "Prosodic inheritance and morphological generalisations", Proceedings, Fifth Conference of the European Chapter of the Association for Computational Linguistics, Berlin 1991: 131-136.

338

Greville G. Corbett and Norman M. Fräser

[96]

Rosch, Eleanor 1978 "Principles of Categori2ation", in: Eleanor Rosch—B. L. Lloyd (eds.), Cognition and Categorization, 27-48. Hillsdale, NJ: Lawrence Erlbaum. Schane, Sanford A. 1970 "Phonological and morphological maikedness", in: Manfred Bierwisch—^Karl Erich Heidolph (eds.), Progress in Linguistics. The Hague: Mouton. 286-294. Schupbach, Richard D. 1984 Lexical Specialization in Russian (UCLA Slavic Studies 8). Columbus, Ohio: Slavica. Schwanz, Linda J. 1980 "Syntactic markedness and frequency of occurrence", in: Thomas A. Perry (ed.), Evidence and Argumentation in Linguistics, 315333. Berlin: de Gruyter. Shieber, Stuart M. 1987 "Separating linguistic analyses from linguistic theories", in: Peter Whitelock—Mary McGee Wood—Harold L. Somers—Rod L. Johnson—Paul Bennett (eds.), Linguistic Theory and Computer Applications, 1-36. London: Academic Press. Smimova, G. A. 1979 "Kategorija roda nesklonjaemyx sua^estvitel'nyx", [The category of gender of indeclinäble nouns.], in: V. P. Grigor'ev (ed.), Lingvistika ipoetika. Moskva: Nauka. 86-105. Suprun, A. E. 1969 Slavjanskie cislitel'nye (stanovlenie asliteVnyx kak osoboj casti reä). [Slavonic numerals: the development of numerals as a separate pail of speech.] Minsk: Belorussian State University. Tasmowski-De Ryck, Liliane—S. Paul Verluyten 1981 "Pragmatically controlied anaphora and linguistic form", Linguistic Inquiry, 12: 153-154. 1982 "Lmguistic control of pronouns", Journal of Semantics 1: 323346. Tünberlake, Alan 1993 "Russian", in: Bemard Comrie—GreviUe G. Corbett (eds.). 77ic Slavonic Languages. London: Rouüedge. 827-886. Tucker, A. N.—J. Tompo ole Mpaayei 1955 A Maasai Gramtnar: With Vocabulary. London: Longmans, Green & Co. Van Dam, J. 1940 Handbuch der deutschen Sprache: Zweiter Band: Wortlehre. Groningen: J. B. Wolter's Uitgevers-Maatschappij N. V. [Quoted from Marcus, Brinkmann, Clahsen, Wiese, Woest, and Pinker (1993).]

[97] Weiss, Daniel 1993

Default genders

339

"How many sexes are there? (reflections on natural and grammatical gender in Polish and Russian)", in: Gerd Hentschel—Roman Laskowski (eds.). Studies in Polish Morphology and Syntax (Specimina philologiae slavicae 99). München: Otto Sagner. 71105.

Worth, Dean S. 1966 "On the stem/ending boundary in Slavic indeclinables", Zbomik za filologiju i lingvistiku, 9:11-16. Zaliznjak, Andrej Anatol'evie 1977 Grammaticeskij slovar' russkogo jazyka: slovoizmenenie. [Grammatical dictionary of Russian: Inflection.] Moskva: Russkij jazyk. Zwicky, Arnold 1986 "The general case: basic form versus default form", in: Vassiliki Nikiforidou—^Mary VanClay—^Mary Niepokuj—^Deborah Feder (eds.), Proceedings ofthe Twelfih Annual Meeting ofthe Berkeley Linguistics Society: February 15-17, 1986. 305-314. Berkeley, California: B.L.S., University of California. 1989 "What's become of derivations? Defaults and invocations", in: Kira Hall—Michael Meacham—Richard Shapiro (eds.), Proceedings of the Fifleenth Annual Meeting of the Berkeley Linguistics Society, February 18-20, 1989, 303-20. Berkeley, California: B.L.S., University of California.

5. Historical Linguistics

WERNER WINTER

The distribution of short and long vowels in stems of the type Lith. hti: vhti: mksti and OCS jasti: vesti: mesti in Baltic and Slavic languages^ 0. In his discussion of Baltic and Slavic present stems of TeT-yerhs with long stem vowel, Calvert Watkins (1969:31-32) agrees with Jerzy Kurylowicz (1956:305-308) in assuming that the lengthening in, e.g., Lith. b'egu, R begu: Gk. phebomai was a special Balto-Slavic development (to be kept apart from the apparent parallel in Lat. est 'he is eating': edere 'eat') that occurred in root formations formerly, or still in historical times, athematic. In his listing of forms, Kurytowicz follows Antoine Meillet; he eniimerates first items with lengthening throughout: lith. b'egu, begti, Slav. *b^gQ (R begu, OCS bHq) 'run, flee': Gk. phebomai lith. Mu, emi, esti, OCS janil, jasti 'eat': Olnd. ätti, Gk. edomai, Lat. edere lith. sedu, s'esti, OCS sfsti 'sit down': Olnd. sdtsi, Gk. hedos, ON settr Lith. isekti 'engrave', OCS s^kQ, s'i'Sti 'cut': Lat. secäre Short and lengthened forms are noted side-by-side in: \l\Yi.grebiu, gr'ebti 'rake': OCSgrebq, greti 'row' Uth.glebiu, glebti 'embrace': Fghbic 'fit together' (OCSglobiti, as listed by Kurylowicz, appears to be unattested) litli. stiegiu, stiegti (< *stegti) 'thatch': R stog 'haystack': Gk. stegö Lith. treSkiu, trekSti 'crackle', OCS triskati 'strepitum edere': lith. trelSkM 'rattle', OCS (R) troska 'boh of lightning' OCS striig, strikati 'sting': OCS stroka, strikati Finally, Kurylowicz adds: lith. iöku, iökti 'jump, dance': OCS skotg, skofiti 'jump' and concludes (1956:307):

344

Werner Winter

[432]

'II resulte de cette liste, qui n'est probablement pas complete, que le baltoslave a allonge le vocalisme de certains verbes ä racine legere en consonne. Tout porte ä croire que ce sont, en accord avec l'hypothese de Meillet, d'anciens verbes radicaux athematiques.' He then proceeds to offer an explanation in terms of his iaw of polarization'. It appears, however, that once a systematic attempt is made to extend the list of forms showing unexpected vowel lengthening, both Meillet's initial hypothesis and Kurylowicz's extension thereof prove to be insufficient to account for the data on band. 1. It is a time-honored tenet of Indo-European comparative Unguistics that a relatively stable correlation is to be assumed between morphological categories and ablaut grade to be found in them; disruptions do, of course, occur, but they require a special explanation, usually in terms of paradigmatic leveling. In the present context, it is therefore unacceptable to be content with a mere Statement that Baltic and Slavic languages show lengthened grade in old athematic presents, when the evidence from other Indo-European languages clearly indicates that e- and/or zero grade is all that can be expected. In this respect then, both Kurylowicz and (in sHght modification of Kurylowicz's approach) Watkins are right in attempting to offer an explanatory hypothesis. What invalidates these hypotheses, however, is that they do not take into account a number of important facts: (1) Vowel lengthening is not limited to verb stems. (2) Vowel lengthening in Tj eT2 roots depends on the nature of T2. (3) Vowel lengthening is not limited to roots but is found equally well in TjeRT^ bases, again with Tj as a conditioning factor. The paper submitted here will discuss only points (1) and (2); data and arguments for (3) will be presented elsewhere. 2. An inspection of the materials found in Fraenkel (1962), Vasmer (19501958), and Vasmer - TrubaCev (1964-1973) yields the lists included in the sections now following. Only items for which a plausible etymology is available have been introduced. The first group of Usts gives forms with long vowels not properly supported by evidence from outside the Baltic and Slavic domain; the second group enumerates forms without lengthening. Within the groups, the arrangement is exactly alike; this is the reason why there are a few lists with no members at all. Whenever available, both Baltic and Slavic forms are included; closely

[433]

The distribution ofshort and long vowels

345

related items from other Indo-European languages are added, with an eye to maximum information as to the nature of T2.

3. Forms with lengthening in Baltic and Slavic languages: 3.1.7'2 = Balticö,Slavic6: Lith. obelis, Latv. äbele 'apple tree', Lith. öbuolas, Latv. äbuols 'apple': R jabloko 'apple': : OHG aphul 'apple' Lith. grebti 'rake, grab', Latv. \grebt 'grab'],gräbt 'grab': [OOSgreti, grebg 'row', Rgresti, grebu 'grab']: : Oln^.grbhnäti 'grabs' 3.2. 7^2 = Baltic if, Slavic d: Uth. esti, emi, edu, Latv. est, fmu, ^du 'eat': OCS fasti, fami 'eat': : Olnd. ddmi, Lat. edo 'I eat' Uth. pedä, p'Mas 'foot, footstep', Latv. p|c?a, peds 'footstep': - : : Lat.pes, pedis 'foot' Lith. püodas, Latv. puöds 'pot': - : : OHG faz 'container' - : OCSpasri, padq 'fall': : OlaA.padyate 'falls' Lith. sMM, semi, sediiu, Latv. sedet 'sit': OCS sMtti 'sit': : Lat. sedere 'sit' Lith. sesti, sedu, Latv. sest 'sit down': OCS sesti 'sit down': : Gk. hezomai 'I sit down' Lith. üosti, üod'iiu, Latv. uöst 'smeU': OCz. fadati 'explore': : Gk. özein 'smell' Lith. vedaras,vedarai 'entrails', Latv. v§dars 'belly': — :: Olnd. udäram 'belly' lÄÜti.vedis, vedys [v&iis. vedys] 'suitor': OCSnevhta 'bride':: HomericGk. eedm 'bridal presents' 3.3. T2 = BalticSlavic f : lith. b'egti, Latv. begt 'run': OCS b'etati, beig 'flee': : Gk. phebomai 'I am in flight' ü t h . (pa)legti 'lie (lay) down': [OCS leM 'lie down']: : Gk. lekhetai 'lies' Uth. nüogas, Latv. mdgs 'naked': OCS nagü 'naked':: Goth. naqaps 'naked' - : OCS naglü 'sudden': : Goth. anaks 'sudden' - : OCS 'lamb': : Gk. amnös, Lat. agnus 'lamb' Lith. üoga, Latv. udga 'berry': OCS vinjaga 'grape': : Lat. üva 'grape' 3.4. T2 = Lith. 2, Latv. z, Slavic z: [Uth. ai, eJ, Latv. es T ] : OCS jazü T : : Lat. ego T Lith. öikä 'goat', oiys, Latv. äzis 'billy goat': OCS jazino 'skin, leather': : Olnd. ajäs 'billy goat'

346

Werner Winter

[434]

3.5.7'2 = Balticp, Slavicp: No examples found. 3.6. T2 = Baltic t, Slavic t: No examples found. 3.7. T2 = Baltic k, Slavic k: lith. -sekti 'cut, carve': O C S s&S^i, stkg 'cut', [sekyra 'axe']: : Lat. secäre 'cut', secüris 'hatchet' 3.8. T2 = Lith. J, Latv. s, Slavic s: No examples found.

4. Forms without lengthening in Baltic and Slavic languages: 4.1. 72 = Baltic Slavic ö: Lith. abü, Latv. abi 'both': O C S oba 'both': : Olnd. ubhäu, Gk. ämphö 'both' OPr. babo 'bean': Rbob 'beam': : Lat. faba 'bean' Uth. bebras, bäbras, Latv. b§brs 'beaver': O C S bebrü, bobni 'beaver': : O H G bibar 'beaver' lÄth.dabä [dobä], La.tv. daba 'manner': Rdoba 'proper time': : Goth. gadaban 'fit' - : O C S doftm 'good': : Lat./aier'artisan' Lith. debesis 'cloud', Latv. debess 'sky': O C S nebo 'sky': : Gk. nephos 'cloud' Uth..gäbalas, L&tw. gabals 'lump': [Pgabnac 'grab']: : Goth.giban 'give', OIr. gaibim ' I take' [lÄih.. gr'ebti], \jsXv.grebt 'grab': OCSgreti, grebg 'row': : Olndi. grbhnäti 'grabs' - : OCSrebro 'rib': : O H G rippi 'rib' OPr. sebbei 'to oneself: O C S sebi 'to oneself: : Lat. sför'to oneself Lith. sidäbras, Latv. sidrabs 'silver': O C S sirebro 'silver': : Goth. silubr, O H G silabar 'süver' U t h . stäbas, Latv. stabs 'post': O C S stoborije 'colonnade': : O H G Stab 'staff OPr. tebbei 'to thee': O C S tebi 'to thee': : Lat. tibi'io thee' Lith. väbalas, Latv. vabals 'beetle': R veblica 'intestinal worm': : O H G wivil 'beetle' 4.2. T2 = Baltic d, Slavic d: Lith. besti, badyti, Latv. best, badit 'sting': O C S bosti, bodg 'sting': : Lat. fodere'dig',OEbedd'hed'

[435] Lith.

The distribution of Short and long vowels

347

medis 'tree, woods', Latv. me^s 'woods': Rmeia 'border': : Olnd.

mädhyas 'intermediate' Lith. medüs, Latv. medus 'honey': OCS medü'honey': : Olnd. mddhu 'honey' Uth.pädas 'floor; sole', Latv. pads 'floor': ORpodü 'foundation: : (see section 9 . 1 ) - : OCS sedilo 'saddle': : Goth. sitls 'seat' (see section 9.3) - : OCS voda 'water': : Goth. wate 'water' (see section 9 . 4 ) Lith. vädas, Latv. vads 'leader': OCS voevoda 'Commander': : Oh. fedim 'I lead' Lith. vädas 'pawn', vadüoti, Latv. vaduöt 'redeem': - : : Goth. wadi 'pawn' Lith. vecTps, vMis [vedys] 'bridegroom': [OCS nevfera'bride']: : Oln&.vadhüs 'bride' (see section 10.2) Lith. vesti, vedü, Latv. vest, vqdu 'lead': OCS vesti, vedo 'lead':: 0\i. fedim 'I lead' - : OCS vedrü 'bright', R vedro 'fine weather': : OHG wetar 'weather' - : OCS xodü 'walk': : Gk. hodös 'way'; but see section 9.6 4.3. T2 = Baltic^, Slavicf: - : OCS bogü 'god': : Av. baya- 'lord, god', Olnd. bhägas 'purveyor' (see section 9 . 5 ) Uth. dägas 'summer heat', öfx.dagis 'suramar': - : : Goth. dags 'day' Lith. degti, Latv. degt 'bum': OCS "ieM. legg 'burn': : Olnd. dähati 'burns' [UWv.palegti 'lie (lay) down']: OCS leiati, Mg 'lie':: Gk. lekhetai 'is lying' - : OCS llgükü ' l i ^ t ' : : Lat. levis 'light' lith. nägas, Latv. nags 'nail', Lith. nagä 'hoof: OCS noga 'foot': : Olnd.

nakhäm 'nail', Lat. unguis 'nail' Uth. nagutis, OPr. mgutis 'nail': OCS nogütt 'nail': : OHG nagal 'nail' Lith. segti 'attack': OCSpmeJ/i, prisegg 'touch': : Olnd. säfati 'hangs' (see section 9 . 2 ) Uth. rägas, Latv. rags 'horn': OCS rogü ' h o m ' : : MHG ragen 'reach high' (?) Uth. vagä, Latv. vaga 'furrow', Lith. vägis, Latv. vadzis 'wedge': — : : Gk.

ophnis 'plough' 4.4. T2 = Uth. t, Latv. z, Slavic z: Uth. eieras, Latv. §z§rs 'lake': OCS jezero 'lake': : Gk. Akherön (?) Lith. efe 'border', Latv. ela 'border of field': - : : Arm. ezr 'border' Uth. e'iys 'hedgehog': OCS je'it 'hedgehog':: Gk. ekhinos 'hedgehog' Uth. veiti 'transport', Latv. vezums 'cart': OCS vesti, vezg 'transport': : Olnd.

vähati 'transports, leads'

348

Werner Winter

[436]

4.5. r2 = Balticp,Slavicp: lith. lepti, Latv. lept 'become spoiled': - :: Lat. lepidus 'delicate' Uth.säpnas, säpnis, Latv. sapn/s 'dream': OCS sünü 'sleep': : Olnd. svdpnas 'sleep, dream' lith. septihtas, sekmas, Latv. septits 'seventh': - : : Lat. septimus 'seventh' Lith. ISlapias, Latv. slapß 'wet': - :: Gk. klepas 'wet' 4.6. T2 = Baltic t, Slavic t: Lith.flf-, Latv.ar- 'av^ray from': OCS otü 'away from':: Olnd. äti 'beyond' - : OCS ottcl 'father': : Gk. ätta 'dear father' Lith. katräs, Latv. katrs 'which one of two': OCS kotoryi 'who': : Goth. hwapar 'which one of two' - : OCS kotora Tight': : MHG hader 'quarrel' Lith. mätas 'measure', Latv. mats 'name of a measure': Rmetit' 'aim': : Gk. metron 'measure' Lith. mesti, metü, Latv. mest 'throw': OCS mesti, metg 'throw': : Gk. möthos 'turmoÜ' (?) Lith. petys, petis, OPr. pette 'Shoulder': - :: Gk.petännumi 'I spread' Lith. ratä 'oath' [if not a loan word]: OKrota 'oath': Olnd. vratäm 'law, vow' Lith. rätas, Latv. rats 'wheel': - : : Lat. rota 'wheel' Lith. tetervä, Latv. teteris 'grouse': OR teterevX 'grouse': : Gk. teträön name of abird Uth. vetuSas, vetas, Latv. v§cs 'old': OCS vetüxü 'old': : Lat. vetus 'old' 4.7. 72 = Baltic Ä:, Slavic Jt: Lith: akis, Latv. acs 'eye': OCS oko 'eye': : Gk. össe 'eyes' Lith. ek'eti, Latv. ecet 'hanow': — :: OHG egida 'harrow' lÄXh., (i)eknos, IjaXw.aknas 'liver': — :: OlnA. ydkrt 'liver' Lith. läkti, Latv. lakt 'lap up': OCS lokati 'lap up': : Gk. läptein 'lap up' Uth.nflfc/zs, Latv. nakts 'night': OCS nöStl 'night': : Lat. nox 'night' lÄih.pekus, 0?r.pecku 'livestock': — :: Lat. pecü 'livestock' [Lith. rekti, Latv. rekt 'roar']: OCS reM, rekg 'say': : Toch. B reki 'word' Lith. sakai, Latv. sakas 'resin': OCS sokü 'sap': : Gk. opös 'sap', Toch. B sekwe 'pus' Lith. sekti 'dry up': OCS isekngti 'dry up': : Olnd. äsakras 'inexhaustable' Uth. sekti, sakyti, Latv. sactt 'teil': OCS söiiti 'point out': : Gk. ennepein 'say' Lith. sekti, Latv. sekt 'follow': — : : Lat. se