Paradigms: The Economy of Inflection [Reprint 2010 ed.] 9783110889109, 9783110127614

Citation preview

Paradigms

Empirical Approaches to Language Typology 9

Editors

Georg Bossong Bernard Comrie

Mouton de Gruyter Berlin · New York

Paradigms The Economy of Inflection

edited by Frans Plank

Mouton de Gruyter Berlin · New York

1991

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

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

Library of Congress Cataloging in Publication Data

Paradigms : the economy of inflection / edited by Frans Plank. p. cm. — (Empirical approaches to language typology : 9) Includes bibliographical references and indexes. ISBN 3-11-012761-X (cloth : acid-free paper) : 1. Grammar, Comparative and general — Inflection. I. Plank, Frans. II. Series. P251.P37 1991 415-dc20 91-33626 CIP

Die Deutsche Bibliothek — Cataloging in Publication Data

Paradigms : the economy of inflection / ed. by Frans Plank. Berlin ; New York : Mouton de Gruyter, 1991 (Empirical approaches to language typology ; 9) ISBN 3-11-012761-X NE: Plank, Frans [Hrsg.]; GT

© Copyright 1991 by Walter de Gruyter & Co., D-1000 Berlin 30. All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Typesetting and Printing: Arthur Collignon GmbH, Berlin. Binding: Dieter Mikolai, Berlin. Printed in Germany.

Contents Contributors Preface Of abundance and scantiness in inflection: A typological prelude Frans Plank

vii ix

1

Form and function in identifying cases Bernard Comrie

41

Paradigm size, possible syncretism, and the use of adpositions with cases in flective languages Silvia Luraghi

57

Pragmatic disguise in pronominal-affix paradigms Jeffrey Heath

75

Geometric representation of paradigms in a modular theory of grammar Katherine McCreight — Catherine V. Chvany

91

Systematic versus accidental phonological identity Arnold M. Zwicky

113

Syncretism and the paradigmatic patterning of grammatical meaning Jadranka Gvozdanovic 133 Rasmus Rask's dilemma Frans Plank

161

The assessment of paradigm stability: Some Indo-European case studies Robert Coleman 197 Inflection classes: Two questions with one answer Andrew Carstairs-McCarthy

213

Organising principles for nominal paradigms in Daghestanian languages: Comparative and typological observations Aleksandr E. Kibrik 255

vi

Contents

The geometry of verb paradigms in Teso-Turkana Gerrit J. Dimmendaal

275

Author Index

307

Subject Index

311

Language Index

315

Contributors Andrew Carstairs—McCarthy, Department of English, University of Canterbury, Christchurch 1, New Zealand. Catherine K. Chvany, Department of Foreign Languages and Literatures, Massachusetts Institute of Technology, Cambridge, MA 02139, USA. Robert Coleman, Emmanuel College, Cambridge CB2 3AP, England. Bernard Comrie, Department of Linguistics, University of Southern California, Los Angeles, CA 90089-1693, USA. Gerrit ]. Dimmendaal, Afrikaanse Taalkunde, Rijksuniversiteit te Leiden, Postbus 9515, NL-2300 RA Leiden, The Netherlands. Jadranka Gvo^danovic, Slavisch Seminarium, Universiteit van Amsterdam, Spuistraat 210, NL-1012 VT Amsterdam, The Netherlands. Jeffrey Heath, Program in Linguistics, University of Michigan, 1076 Frieze Building, Ann Arbor, Ml 48109, USA. Aleksandr E. Kibrik, Kafedra strukturnoj i prikladnoj lingvistiki, Filologicesky fakultet, Moskovskij Universitet, Moskva, USSR. Silvia Luraghi, Oriental Institute, University of Chicago, 1155 East 58th Street, Chicago, IL 60637, USA. Katherine McCreight, Department of Foreign Languages and Literatures, Massachusetts Institute of Technology, Cambridge, MA 02139, USA. Frans Plank, Philosophische Fakultät, Universität Konstanz, Postfach 5560, D-7750 Konstanz l, Federal Republic of Germany. Arnold M. Zwicky, Department of Linguistics, Ohio State University, 1841 Millikin Road, Columbus, OH 43210-1229, USA.

Preface Recent years have seen several individual attempts to shed new light on inflectional paradigms, or also to recall and reappraise relevant insights of the past. This volume is a collective effort to document the kind of (nonpartisan) work currently being done in this area, and thus to consolidate the renaissance of one of the most venerable notions in morphology. What we especially hope to encourage is a more intensive search for generalisations about the structure of paradigms, in the various kinds of languages which have any, which in the past has been hampered by an inclination to regard paradigms, those perennially popular representational devices, as repositories of the particular (almost like dictionaries). This collection grew out of a workshop I organised at the Annual Meeting of the Societas Linguistica Europaea, held at Freiburg im Breisgau, 13 — 15 July 1988. The twelve papers presented on this occasion, dealing with various aspects of the structure of paradigms, proved too heterogeneous to be usefully assembled in a single volume. It seemed preferable to focus on a more limited range of questions, and ones that would generally be considered to be fundamental for Comparative Paradigmatics. In light of the emphasis of much ongoing research and of a strong tradition in morphological typology, I suggested that patterns of the dearth and the profusion of inflections, i. e. of deviations from the ideal norm of "one meaning — one form", to which languages called "agglutinative" represent the closest actual approximations, should be the leitmotif of prospective contributions. What was sought, thus, were language-particular and comparative, synchronic and diachronic, descriptive and explanatory accounts of patterns of formal non-distinctness in paradigms and of coexisting paradigms (declensions and conjugations) and subparadigms, reflecting on the implications of such patterns for the structure of paradigms and their representation. Earlier versions of Carstairs', Coleman's, and Plank's papers had been prepared for the workshop at Freiburg. The other contributions were solicited afterwards and were completed during 1989, some earlier, some later. Revolutions, removals, and other vicissitudes of academic life prevented a few further prospective contributions from materialising in time. Thanks are due to Bernard Comrie for expert and prompt comments on the entire manuscript, and to Wolfgang Schellinger for help with the proofs. Paradigms in the form of lists are almost certain to miss out on important interconnections between the items listed. So are collections of articles, and here, for technical reasons, the arrangement is inevitably sequential. It is up to the reader to make the extra connections. February 1990

Frans Plank

Of abundance and scantiness in inflection: A typological prelude Frans Plank

1. The differences between el and manus Tables 1 and 2 set out the inflected forms of the nouns for 'hand' in Turkish and Latin, el and manus, and give the meaning of each form at the head of the appropriate column and at the left-hand side of the appropriate row. What are the significant structural properties of such sets of inflected words as represented in these paradigms? It would seem rather insignificant that both paradigms are two-dimensional matrices, for they could easily have been laid out differently — one- or three-dimensionally, for example. As long as all inflected forms are enumerated and the meaning of each is properly identified, the mode of their arrangement might seem purely conventional. More significant are perhaps some points on which our two paradigms differ. There is first of all, an obvious difference in size: the Turkish paradigm is much larger than the Latin one, containing 84 as opposed to 12 members. While el and manus both inflect for Number and Case, with an equal number of terms realising these categories in the two languages (two Numbers, six Cases), el in addition inflects for Possession, distinguishing Possessors by Person (1st, 2nd, 3rd) and Number (Singular and Plural) — for which purpose Latin has separate pronouns. If the Turkish paradigm in Table 1 is divided into subparadigms according to this additional distinction, the subparadigm of Unpossessed forms (i. e. the leftmost column) is as large as its Latin counterpart. Both can in turn be divided, again with identical results, into two Number subparadigms for Case and six Case subparadigms for Number. As to the inflectional categories shared by el and manus, the terms realising them are identical, except that el has a Locative while manus has a Vocative. Identical term names, however, do not always imply identical uses of the respective forms. The Turkish Nominative (also known as Absolute), for example, also marks direct objects as long as they are indefinite, in which function Latin nouns are in the Accusative; or the Turkish Singular, perhaps more appropriately characterised as numerically neutral, is the usual form of nouns accompanied by numerals, where Latin nouns are in the Plural.

1''rans Plank

·„. « -s. « «„

·.,« -sS ΟΗ CO

Ο ΡΗ

SS

S S .S

«

Α"

S ^

ο α.Μ

ο α.6C

£ ~Ι ~§

-« -S

ω

ΐ-Μ

Ο Η Ο

'c 'c s; "5

α P

D

«

Of abundance and scantiness in inflection

3

Table 2. Inflection of Classical Latin manus 'hand'

Voc Nom Ace Abl Dat Gen

Sg

PI

manus manus manum manu: manui: manu:s

manu:s manu:s manu:s manibus manibus manuum

The completeness of our two paradigms is a matter of controversy, although for different reasons. For Turkish, one might consider adding a seventh and perhaps even an eighth Case, a Comitative-Instrumental expressed by -lej-jJe and a Benefactive in -finj-jfin. These putative Case suffixes resemble postpositions, He 'with' and ifin 'for', following nouns in syntactic construction, but, although evidently originating from these diachronically, show some signs of being in morphological construction with their nouns: for example, their first vowel may be elided, and the remaining one is subject to vowel harmony. (Unlike other Case suffixes, however, these are unaccented.) Turkish, further, has a suffix -ki which forms nouns (sometimes called pronouns) from Genitives, Locatives, or adverbs, which may then be inflected for Number and Case (e. g. eliminkilerden 'from those belonging to my hand'). The inclusion of all these forms would greatly expand the paradigm of el, and it is not self-evident that -ki is derivational rather than inflectional and should therefore be excluded. For Latin, one might consider adding a Locative as a seventh Case, at least in the Singular subparadigm, not because manus would have a further inflectional form with locative meaning, but on the strength of at least one other noun which otherwise tends to inflect like manus but has (or had in pre-Classical times) a Locative Singular form distinct from its other inflected forms, viz. domus 'house' — domi: 'at home' (later replaced by the Dative, domui:}} For each inflectional category in Table 1 there is one term without overt exponent: the Singular Number, the Nominative Case, and Unpossessed. Among the Possessed forms, the Singular of 1st and 2nd Person Possessors too is without overt exponent for Number, lacking the -i% of the corresponding Plural Possessors. (Matters are somewhat more complex with 3rd Person Possessors; and among the three Persons, none is without exponent.) In Latin, the Plural forms of manus are more substantial than the corresponding Singulars, measured in sound segments or vowel quantity, but the Singular does not lack overt exponence. Nor is there a single Case which could

4

t'rans Plank

plausibly be said to be without exponent. In the Singular subparadigm the Ablative might be taken for such a zero form, but this would entail the complication of how to get rid of the length of its final vowel when -s, -m, and -/: are added for the Vocative/Nominative, Accusative, and Dative. The citation form of this Latin noun, the Nominative Singular, is certainly not stripped of all inflections; its Turkish analogue, the Nominative Unpossessed, is. Morphemic segmentation is unequivocal in the paradigm of el, but not in that of manm. The largest invariant segment in Table 1, el, is easily identified as the stem of this noun, and it uniformly serves as the base of all inflections. Segmentation in Table 2 would seem to yield two such basic forms, manuand man-, with the final vowel that distinguishes them perhaps best analysed as a stem formative, absent in the Ablative/Dative Plural. An alternative analysis, not to be rejected out of hand, would segment thus: man-us, manurn, man-u:, etc., with u as part of most exponents of Number and Case. On this analysis the inflection of Latin manus would be single-stemmed, like that of Turkish el. Even though in Latin it may at times be more difficult to locate the boundary between stem and inflection, the phonological cohesion between these morphemic segments seems about as strong in Latin as it is in Turkish. Phonologically the stem and the inflections of el as well as of manus represent single words, as shown by sandhi phenomena, phonotactics, accent, and the like. In Turkish, where the distinction between (morphologically bound) suffixes and (phonologically bound) enclitics may in general be less clear than in Latin, only the former (though not all of them) are subject to the requirements of vowel harmony, and all inflections included in Table 1 meet this criterion of phonological cohesion. In non-phonological respects, however, the bonds between stems and inflections are weaker in Turkish than in Latin. Thus, while a Latin noun holds on to its inflectional ending when followed by another noun in a close coordinate construction inflecting for the same Number and Case, a Turkish noun may shed its inflections in such circumstances: el ve ayaklanmi (or, less elliptically, ellerimi ve ayaklanmi) 'my hands and feet' (Accusative), manu:s et pede:s (not man(u) et pede:s). Likewise, adjectives on their own in Turkish inflect like nouns, but when an attributive adjective is followed by a noun (which is the regular order) it loses its inflection, so that again only the final inflectable member of the nominal group is inflected: kü^üklerimi 'my small ones', kü$ük ellerimi 'my small hands' (Accusative). In Latin all members of a nominal group which can be inflected must be inflected; thus: parva:s manu:s or manu:s parva:s, not parv manu:s or man(u) parva:s, A further indication of the relative looseness of the morpho-

Of abundance and scantiness in inflection

5

logical combination of stems and inflections in Turkish is the ability of these inflections to co-occur with entirely different parts of speech, such as participles and finite verbs when these are the last words of clauses in subject, object, or adverbial function. What follows the stem (plus perhaps stem formative) of manus cannot be further morphemically segmented in any consistent manner. There are, thus, no separate exponents for Number and for Case; these two categories are expressed cumulatively in a single ending instead. The inflected forms of el, by contrast, can be analysed as consisting of separate morpheme for the various categories, strung out in several positional classes after the stem: first come exponents of Number (in fact only Plural, since Singular lacks overt expression), then of Person and of Number of a Possessor (Unpossessed lacks an exponent), then of Case (other than Nominative, which likewise has no exponent). This neat picture is only disturbed by Plural nouns with 3rd Plural Possessor ('their hands'), which coincide with both Singular noun forms ('their hand') and Plural nouns with Singular Possessor ('her/his hands'), and hence appear to lack a separate suffix either for nominal Plural (el-0-ler-i) or for Plural of Possessor (el-ler-0-ι). A further complication, though on the face of it not one suggesting cumulation, is introduced by 3rd Plural Possessed forms: when compared to 3rd Singular Possessed forms, two suffixes seem segmentable for Number (-/er) and Person (-/) of the Possessor — but these are not in the same order as with 1st and 2nd Person Possessor, where Person (-im I-in) precedes Number (-/^) of the Possessor.2 In our examples the length of inflectional morphemes does not exceed two syllables, with Ablative/Dative Plural -thus in Latin perhaps being the longest; and on the assumption that this relative uniformity is not coincidental, one would expect words with separatist inflectional exponents to be potentially longer than their counterparts with cumulative exponents — and the more so, the more categories are involved. Moreover, with the separatist technique and its inclination to leave certain terms unexpressed, different inflected forms of a word are bound to differ in length, measured in syllables. The cumulative technique may produce similar patterns, but should do so less systematically. The potentially greater length of separatistically inflected words is compensated for by greater economy in the supply of exponents. Comparing the Number and Case forms of manus with their analogues in the Unpossessed subparadigm of el (the leftmost column in Table 1), they both number twelve — two Numbers multiplied by six Cases. However, Turkish only needs two Number exponents plus six Case exponents to express all these categorical distinctions; and the sum total of eight can even be reduced by two since Singular Number and Nominative Case lack overt exponents.

6

Frans Plank

To accomplish the same feat Latin would need two times six exponents, the sum total resulting from multiplying rather than adding the terms of the categories cumulated. (In actual fact, manus has only seven distinct exponents, but, as Table 2 shows, these do not suffice to express all categorial distinctions.) For the entire paradigm in Table 1, with 84 cells, Turkish would ideally need only ten distinct exponents, and it is self-evident that our enumeration of forms in this table is highly redundant. It would suffice to give the exponent of each term only once: Number: PI -ler (Sg -0); Case: Ace -/, Gen -in, Dat -e, Loc -de, Abl -den (Nom -0); Person of Possessor: l -im, 2 -in, 3 -/' (UnPoss -0); Number of Possessor: PI -i% (Sg -0); and to specify the relative linear order in which these categories are to be expressed after a stem. Isolated complications marring the tidiness of such systems, such as those with 3rd Plural Possessors noted above, would demand additional instructions overriding the regular meaning-form assignments (e. g. 3plPoss -leri, P13plPoss -leri}. Owing to the cumulative technique, there is no way of reducing Table 2 for Latin analogously: here the more complex structuring of the set of inflectional exponents into subparadigms for each of the cumulated categories is not redundant but essential. Tables 1 and 2 both show instances of single forms expressing more than one meaning (or meaning combinations). In Turkish, el has 58 rather than 84 distinct inflected forms; in Latin, manus has 7 distinct forms rather than 12.3 Relatively speaking, manus economises more heavily than el, fielding only 58.33%, as opposed to 69.05%, of the team of inflected forms it could potentially muster up. If manus, as it were, decided to discontinue expressing any single further categorial distinction (say that between Ablative and Dative Singular), this would automatically reduce the number of its exponents by one, regardless of whether one of the previous distinct exponents (-»: or -«/:) is re-used or a new one (say -o;) is replacing both. By contrast, el would not gain anything from such obliterations of minimal categorial distinctions: if, for instance, one wanted to conflate Dative and Ablative Singular Unpossessed, the overall number of exponents would remain the same if -e or -den were used for this purpose, or would even increase by one if a new exponent were used — as long as Dative and Ablative continued to be distinguished by -e vs. -den in the Plural Unpossessed and the Singular and Plural Possessed forms. It matters, thus, whether exponents are cumulative or separatist if distinctions are to be undone.

Of abundance and scantiness in inflection

7

With manus there is one categorial distinction made in Table 2 that is not reflected by a formal contrast in any subparadigm, and hence might seem gratuitous: that of Vocative and Nominative. Nominative and Accusative, and Ablative and Dative, both pairs conflated in the Plural, are distinguished in the Singular; and Genitive Singular, conflated with Vocative/Nominative/ Accusative Plural, is distinguished from Genitive Plural as well as from other Cases in the Singular. With el, on the other hand, all of the categorial distinctions unexpressed in some environments are expressed in others: the six Cases are distinguished everywhere, the two Numbers in all Unpossessed and most Possessed forms, the three possessor Persons as well as Unpossessed in the Absolutive Singular and Plural and partly also elsewhere, the two possessor Numbers in all Singular noun and all 1st and 2nd Person possessor Plural noun forms. The formal identities in the set of inflected forms of manus cannot plausibly be accounted for in terms of synchronic Classical Latin phonology or morphophonology; their pattern is one that needs to be stated in morphological terms. There are apparently similar identities in the inflection of el: in particular those between Accusative -/ and 3rd Singular Possessed -/' (hence eli SgUnPossAcc/Sg3sgPossNom, elleri PlUnPossAcc/PBsgPossNom) and between Genitive -in and 2nd Singular Possessed -in (hence elin SgGenUnPoss/ Sg2sgPossNom, ellerin PlUnPossGen/P12sgPossNom). The majority of the identities in Table 1, however, are accidents of Turkish phonology and morphophonology, rather than genuinely morphological characteristics of the inflection of el. Morphologically, one expects the Plurals of 3rd Plural Possessed forms to be el-ler-leri, el-ler-lerin-i, etc., and it is evidently owing to the haplological elimination of one ler that they come out as elleri etc., identical to Singulars. The 3rd Person Possessor suffix -/' appears as -in when followed by a Case suffix, and this allomorph looks like 2nd Person Possessor -/«, hence the identities of 3rd and 2nd Person Possessed forms in Cases other than (exponentless) Nominative. What does not quite look like a nonmorphological accident is the partial identity of Plural -ler and 3rd Plural Possessed -leri (perhaps to be segmented as -ler-ΐ), which causes Plural nouns for 3rd Singular Possessor to coincide with Singular nouns for 3rd Plural Possessor (el-ler-i hand-Pl-3sgPoss — el-leri hand-3plPoss). This might still be considered an inadvertent consequence, though not of (morpho-) phonology, but of the failure of Person and Number exponents of the 3rd Person Possessor to be sequenced like those of 1st and 2nd Person Possessors, where Number comes after Person. Something which cannot be inferred from Tables 1 and 2 is that el is far more representative of the entire nominal inflection of Turkish than manus is

8

Frans Plank

of that of Classical Latin. As a matter of fact, the exponents for Number, Case, and Person and Number of Possessor are not always the same as those seen in Table 1 for all Turkish words inflecting for these categories. However, the choice among alternative exponents expressing the same meanings is to a large extent conditioned by their phonological environments, and never by morphological classifications of noun stems. All exponents of Table 1 come in two (-lerl-lar, -del-da, -den\-dari) or four forms (-//-«/-//-A, etc.), with their vowels required to harmonise, in frontness/backness and (the fourfold suffixes) rounding, with the last vowel of their base. Of Case suffixes with initial vowel, those of Accusative and Dative add j and that of Genitive adds n when preceded by a vowel, thus avoiding hiatus (cf. gece 'night' SgNom, geceyi Ace, gecenin Gen, geceye Dat). The initial consonants of Locative and Ablative -de and -den are devoiced after a voiceless consonant (thus, kitap 'book', belonging to the class of nouns which voice their final consonant when it precedes a vowel, kitabt Ace, kitapta Loc, kitaptan Abl). The suffixes of 1st and 2nd Person Possessor drop their initial vowel, and that of 3rd Person Singular adds s, when following a vowel (thus, gecem, gecen, gecesi 'my, your, her/his night'); unlike el, vowel-final stems therefore consistently distinguish 2nd and 3rd Singular Possessor in Singular forms: gecen Nom, geceni Ace etc. vs. gecesi, gecesini etc. And, the single instance of morphological conditioning, 3rd Person suffixes -i\-si\-leri add a final « when followed by a Case suffix. 4 All these exponents occur in the same variations also with adjectives when used on their own, and, with minor modifications, with personal, demonstrative, and interrogative pronouns as well. The 1st Person personal pronoun has Genitive -im instead of -in; its Plural as well as that of 2nd Person is in -i% (as in Possessed nouns) rather than in -Jer;s and interrogatives may or may not use -/ for Accusative, which Case may thus coincide with Nominative. The scope of the application of Person and Number exponents is even wider since they are also used, albeit with additional alternations, as Person and Number markers of verbs. Notwithstanding their superficial variety, the paradigms of different words, at least within the nominal sphere, do not differ from one another on any of the parameters on which Turkish el was seen to differ from Latin manus — except accidentally, as in the case of nouns such as hat 'line', which retain their final geminate consonants prevocalically and simplify them elsewhere, and thus, by phonological chance, fail to distinguish Dative and Locative Singular (hatt-a Dat, hatf-ta Loc). Latin inflection is far less uniform, and allomorphs of exponents, often so dissimilar as not to be relatable to one another phonologically, are largely conditioned by their morphological environment, in particular by the stems

Of abundance and scantiness in inflection

9

Table 3. Classical Latin noun declensions

Sg Voc Nom Ace Abl Dat Gen PI Voc Nom Ace Abl Dat Gen

Sg Voc Nom Ace Abl Dat Gen PI Voc Nom Ace Abl Dat Gen

IK

HM

UN

IVM;(.

IV N

capra capra capram capra: caprae caprae caprae caprae capra:s capri:s capri:s capra:rum 'goat'

lupe lupus lupum lupo: lupo: lupi: lupi: lupi: lupo: s lupi:s lupi:s lupo :rurn 'wolf

helium helium helium hello: hello: belli: bella bella bella belli:s belli:s hello :rum 'war'

manus manus manum manu: manui: manu:s manu:s manu: s manu:s manibus manibus manuum 'hand'

cornu: cornu: cornu: cornu: cornu: cornu:s cornua cornua cornua cornibus cornibus cornuum

'horn'

V,,M

IIIaM,i.·

IIIa N

Illab

IIIb M / F

Illb x

die:s die:s diem die: diet diet die:s die: s die:s die: bus die:bus die:rum 'day'

re:g? re:gs re:gem re-.ge re:gi: re:gis re:ge:s re:ge:s re:ge:s re:gibus re:eibus o re:gum 'king'

no:men no: m en no:men no:mine no:mini: no:minis no:mina no:mina no:mina no:minibus no:minibus no:minum 'name'

urbs urbs urbem urbe urbi: urbis urbe: s urbe:s urbi:s urbibus urbihus urbium 'town'

ignis ignis ignem igni: igni: ignis igne:s igne:s igni:s ignibus ignibus ignium 'fire'

mare mare mare mart: mari: marts maria maria maria maribus maribus marium

'sea'

Note: a Orthographically rex. Otherwise Latin orthography is used, with the addition of vowel length.

to be inflected. There is some phonological conditioning as well, although usually only within the confines of particular inflection classes. Thus, Nominative and Vocative Singular, for instance, tend to be zero rather than -us when nouns of Declension II (see Table 3) end in -er (cf. puer 'boys' vs. lupus I lupe 'wolf); Nominative Singular is zero rather than -s when nouns of Declensions Illa/lllab end in l\r\n\s (cf. consul 'consul' vs. re:gs 'king'); Vocative/Nominative/Accusative Singular is zero rather than -e when neuters

10

Frans Plank

of Declension Illb are polysyllabic and end in l\r (cf. animal 'animal' vs. mare 'sea'); Genitive Plural is -urn or -ium with nouns of Declensions III differing in stem-final vowels and consonants, but the relevant phonological distinctions are so intricate that this alternation is perhaps better treated as morphologically conditioned (hence the different classes in Table 3). As to other parts of speech inflecting for Number and Case, adjectives, themselves heterogeneous like nouns, largely follow nominal patterns, though not always slavishly, while the various kinds of pronouns increase the diversity of exponents. Some nominal exponents recur in verbal inflection, but fortuitously so since verbs do not share inflectional categories with nouns, as they do in Turkish (e. g. Latin -/:, expressing Genitive, Dative, Ablative Singular, or Vocative or Nominative Plural with nouns, serves as 1st Person Singular Perfect Indicative Active with verbs). Whereas one-to-many relations between inflectional meanings and forms can essentially be accounted for by general phonological and allomorphic regularities in Turkish, they are traditionally dealt with in Latin by dividing the words inflecting for the same categories into several classes, with arbitrarily (i. e. neither phonologically nor semantically) determined membership, and further subclasses, with membership determined, somewhat less arbitrarily, by Gender, and by spelling out the full set of inflectional forms for each of these classes and subclasses, regardless of partial similarities. Table 3 illustrates, perhaps not quite exhaustively, the Classical Latin noun declensions, omitting the controversial Locative, phonologically conditioned zero alternants (as mentioned above), and a good deal of synchronic variation in the shape of some exponents. The noun inflection of Latin is, thus, clearly more complicated than that of Turkish, involving subparadigms (shown above to be redundant for Turkish) as well as inflection classes and requiring lexical stipulations of class membership for each noun. 6 Moreover, while consistently favouring cumulative rather than separatist exponence and tightly bound morphological combinations, these classes differ among one another on some of the parameters so far examined. Thus, increasing the size of the paradigm by the inclusion of a Locative seems better motivated in some classes than in others, owing to the existence of forms which are actually distinct from those of all other Cases (e. g. domi: 'at home' from Declension IV); and the setting up of a paradigmatic distinction in the first place is more immediately suggested by some classes than by others (e. g. that between Vocative and Nominative, for which only Declension II provides distinct forms in the Singular). Some classes include a form, viz. Nominative Singular, which is without overt exponent, if only as one alternative in certain phonological contexts (i. e. nouns like puer, consul, animal

Of abundance and scantiness in inflection

11

of Declensions II and III). Such zero forms facilitate morphemic segmentation and the identification of a single base. But there are other classes where inflection is evidently double-stemmed (cf. no:men\no:min from Declension IIIaN, or, not susceptible to phonological treatment, iter\itimris 'way' (Nom/ GenSg), iecur\iecinoris or iecoris 'liver' (Nom/GenSg), nix\mvis 'snow' (Nom/ GenSg)) 7 and where the boundary between stem and inflection is more difficult to locate (e. g. in aeta:s 'age', otherwise inflecting like re:gs, the stemfinal consonant, audible in aeta:tis etc., appears to be absorbed by the inflection). While none of the eleven classes in Table 3 can boast twelve distinct forms, none conflates Case and Number distinctions in exactly the same manner as manus does, either. All these paradigms are, at any rate, completely filled with forms, even if these are not always distinct or are unlikely to have been heavily used. This cannot be said for all nouns of Latin, relatively many of which simply appear to lack an appropriate form for one or another Case or Number for no especially convincing semantic or pragmatic reasons (e. g.faex 'dregs' only has Nominative and Dative Singular and Ablative Plural, vi:rm 'slime' only Nominative, Genitive, Ablative Singular, pondo: 'in weight' only an Ablative Singular). Defective words of this kind seem another peculiarity of inflection in the style of Latin nouns.

2. Visions of system Individually the dozen or so differences we have seen between Turkish and Latin noun inflection may be trivial, curious, noteworthy, or only footnoteworthy, depending on one's inclinations and interests. But are they significant? Consider the imaginary paradigm in Table 4, already providing morphological analysis. The inflection of elman 'hand' resembles that of Turkish el in several respects: the citation form is without overt exponent; the exponent for Dative Plural is morphologically not strongly bound to the noun stem; there are separatist exponents for Accusative, Genitive, and Ablative, and, with these three Cases, also for Plural, and these inflected forms are therefore longer than others, especially the corresponding Singulars. But it is also reminiscent of the inflection of Latin manus: the paradigm of elman has about the same number of members; it is based on two stems, elman and elmaner; morphological cohesion is strong, except in the Dative Plural; the exponents of Case and Number appear to be cumulative for Nominative (unless -/ is analysed as an allomorph of Plural -/er, with Nominative being -0 as in the Singular), Dative, and Locative; two exponents are non-distinct for no

12

Frans Plank

Table 4. Inflection of elman (hypothetical)

Nom Ace Dat Gen Loc Abl

Sg

PI

elman-0 elmaner-um elman-u elman-u:s

elman-i elman-ler-um elman-ibuf dmamr-ler-u-.s elman-ui: elman-ler-u

elman-u

Note: * elman et ped-ibus hand(DatPl) and foot-DatPl elman parv-i:s hand(DatPl) small-DatPl

obvious extra-morphological reasons, viz. Dative and Ablative Singular; and unlike manus but like other nouns in Latin, elman is defective, lacking a Locative Singular. The inflection of elman, thus, represents a mixture of what might be called el properties and manus properties; and hypothetical though our example is, it demonstrates the logical possibility of such mongrel paradigms. In fact it is possible for individual inflected forms to partake of both property sets: elman-tbus in Table 4 has an exponent which cumulates Dative and Plural (a manus property) but is not morphologically bound to the stem (an el property); elmaner-ler-u:s has an extended stem as base (a manus property) but Plural and Genitive are expressed separately (an el property). Further, if the inflection of a particular word in a language has exclusively either el or manus properties, there is no logical necessity for other words of the same word-class to be of the same type. And if an entire word-class happens to be uniform in its preference for either el or manus properties, there is no a priori reason why other word-classes in the same language, inflecting for different categories, should have to follow suit. The el and manus properties would acquire significance if they turned out to set an empirical limit on the theoretically possible diversity of inflectional systems. The strongest conceivable constraint that could thus be formulated would demand that either el properties or manus properties, but no mixture of both, be chosen (a) for each individual inflected word form, (b) for all inflectional forms of individual words, (c) for the inflection of all words within a particular word-class, and (d) for all classes of inflectable words of a language, no matter how different the categories for which they inflect. This amounts to the claim that all el properties mutually imply one another and are mutually exclusive with manus properties, and vice versa. It would

Of abundance and scantiness in inflection

13

follow from this claim, if valid, that the properties characterising the inflection of Turkish el on the one hand and of Latin manus on the other are principled rather than accidental collections, notwithstanding their logical independence from one another. This is a fairly strong universal hypothesis, and it would presumably not lose its appeal entirely if its validity turned out to be probabilistic rather than absolute. It is not a novel hypothesis, of course. It is sometimes believed that this essentially is what Morphological Typology, as flourishing in the nineteenth and early twentieth century, was all about. This is perhaps not entirely accurate. In spite of the preoccupation with language classification elaborating on bipartitions such as analytic-synthetic or tri- or quadripartitions such as isolating-agglutinative-flective(-polysynthetic), 8 the typology of inflectional systems was, so far as I can see, no matter of prime importance in the golden age of 'classical' Morphological Typology. Nonetheless, in the short textbook of Kuznecov (1956) that begins with a historical survey (covering the brothers Schlegel, Humboldt, Schleicher, Steinthal/Misteli, Fortunatov, Finck, and Sapir), several of what we have referred to as el and manus properties are mentioned as almost self-evidently interrelated differences between agglutination and flexion. But credit seems due in particular to Vladimir Skalicka (see Skalicka 1979 for a sample of his writings between 1935 and 1966), who did much to highlight such property bundles as constituents of his 'typological constructs', which also subsumed various non-morphological features. Skalicka's favourite correlation was that between cumulative or separatist exponence and the presence or absence of inflectional synonymy (i. e. inflection classes) and inflectional homonymy; but others of the properties discussed above played a part too in his version of the isolating-agglutinative-flectiveintroflective-polysynthetic typology, or were added by more recent writers inspired by Skalicka (such as Dressier 1985). All such correlations, however, are supposed, by Skalicka and his followers, to be mere ideals which real languages, agglutinative or flective etc. only to certain degrees, will not always attain. The prevailing view in the Skalicka tradition appears to be that the allegedly interdependent typological properties mutually imply each other, with none of them being recognisably more basic than any other. This was not quite how Adam Smith and an anonymous F^dinburgh contemporary had seen it, in momentous, but often unacknowledged or not fully appreciated, early contributions to language typology (or 'geniology').9 In his Considerations Concerning the First Formation of Languages, and the Different Genius of Original and Compounded Languages of 1761 (more accessibly republished in 1767), Adam Smith, the moral philosopher and political

14

Frans Plank

economist, championed an excrescence, rather than coalescence, theory of the origin of inflections, suggesting that originally meaningless terminal parts of primary words (nouns, verbs, later also adjectives) are increasingly utilised, by the early formers of languages, to distinguish accessory categories such as Gender, Number, and Case, or Tense, Mood, Voice, and Person and Number. Growing out of originally invariable words, these terminations that become subject to variation will be "thoroughly mixed and blended" with stems. (Recall our manus properties of non-straightforward segmentability and strong cohesion.) Owing to randomly different shapes of the original simple words, the terminations newly semanticised will initially differ from one word to the other, and even the "love of analogy", responsible for the subsequent levelling out of some of these differences by way of transferring sets of inflections from one word to others, will not accomplish uniformity (i. e. there will be inflection classes). When more than one accessory category needs to be expressed with a single word, only its internally unsegmented termination will be available to carry the multiple burden; therefore, if there are as few as three Genders (the maximum Smith had encountered), ten Cases (as supposedly in Old Armenian), and three Numbers (as in Greek, Gothic, and Hebrew), a noun or adjective would need as many as ninety variants to distinguish them all cumulatively by contrasts in its terminations. The unpredictable diversity of the shapes of these inflections of different words further multiples the amount of forms that need to be memorised at this stage of language formation. The anonymous author of the article Language in the first edition of the Encyclopedia Britannica (1771), familiar with Smith's scenario, saw a way of limiting this profusion: speakers of such languages, in order not to overburden their memories, might opt for sacrificing accuracy by making individual inflected forms "serve a double, treble, or even quadruple office", as they did in Latin, where for example puellae takes the office (in fact the quintuple one) of Genitive, Dative, and Ablative Singular and Nominative and Vocative Plural. When Original' languages of this kind, evidently modelled on Ancient Greek and Latin, are transformed into 'compounded' ones in the course of mixtures of peoples, inflections, too difficult to be learned by adults not used to them, will be abolished and be replaced by function words such as prepositions or auxiliaries. And these new expressions for accessory categories, familiar from the modern idioms of Europe, are not intimately joined to noun or verb stems, are uniform for all nouns and verbs, do not cumulate categories, and — as Anonymous added — are not homonymous in the manner of inflections. Had Smith and Anonymous been acquainted with Turkish, they would no doubt have noticed that inflectional suffixes there share the last three properties of function words

Of abundance and scantiness in inflection

15

while still cohering with stems, if less tightly than those of Ancient Greek and Latin. The four properties of systems of expression for accessory categories that are recognised as interdependent in this theory are not all directly correlated with all others. Morphologically conditioned allomorphy of inflections, resulting in inflection classes, is claimed to be contingent on a tight cohesion of stems and inflections — which follows from Smith's evolutionary assumptions. Stated as a one-way implication — if inflection classes then tight cohesion of stems and exponents —, the only co-occurrence ruled out would be that of inflection classes and non-cohesive exponents, while that of uniform inflection and cohesive exponents would be allowed.10 Secondly, cumulative exponence is likewise assumed to be conditional on "thoroughly mixed and blended" combinations of stems and inflections, and perhaps vice versa. This point is not explicitly argued for by Smith; but what he may have had in mind as the reason for the mutual implication between these two properties was that they both evidence an aversion to clear-cut segmentation. Lastly, the practical utility of inflectional homonymy is conditional on cumulative exponence and inflection classes, which jointly create the danger of the fund of inflectional forms growing unmanageably large in the first place, without which there would be no need to economise by using single forms in more than one function. It is fitting that the man poised to elucidate the principles determining the abundance or scantiness of the necessaries and conveniences of life clearly recognised, or pointed the way to the recognition of, abundance and scantiness of formal resources as complementary properties of inflectional systems — united by "moral certainty" rather than "physical necessity", as the author of Language in the Encyclopedia Britannica aptly put it. The way these matters are seen today, excepting perhaps the growth of inflections, is still essentially Smithian. The 'Insensitivity Claim' of Carstairs (1981: 4 — 16), for instance, distinguishes two classes of phonologically bound forms, depending on their association with single parts of speech or with syntactic constituents (where the bound form may be hosted by various parts of speech), and prohibits members of the second class from varying in shape in accordance with morphological features of the words to which they happen to be bound. 11 The exponents of Number, Case, and possessor Person and Number in Turkish, not uniquely bound to nouns, would thus be prevented from displaying inflection-class variety, whereas Case and Number exponents in Latin are entitled to allomorphy conditioned by Gender or any arbitrary morphological features of nouns, which alone may host them. As pointed

16

Frans Plank

out by Pike (1965: 205 f.), cumulative exponents are efficient in terms of compactness of signal (since one, often short formative carries two or more meanings) but are hard to learn and remember, simply because there are so many of them if paradigms are to be large and their members are to be kept distinct; while doing worse in compactness, separatist exponents are more efficient in that not so many of them are needed to staff even large paradigms. There should, thus, be a preference for cumulative exponents to be used in small paradigms (like those of pronouns) and for separatist exponents to be used in large ones. (Recall the different paradigm sizes in Tables 1 and 2.) Another consequence, as I have argued elsewhere (Plank 1986), is that the use of cumulative exponents limits the size of such paradigms as are crosslinguistically variable (e. g. those involving Case) more severely than that of separatist exponents. Surveying the incidence of inflectional homonymies (of a particular kind), Carstairs (1984a; 1987: 87 — 146) has noted that it is higher in cumulative than in separatist paradigms, and his explanation is that homonymy is more useful with cumulative exponents, where it conveniently reduces their number. Formal economy is a built-in advantage of separatist exponence, rendering such secondary measures of economy as the nondistinction of the members of a paradigm superfluous. And there in fact is another, more convenient method of economising available to systems which do not cumulate categories (and which would be problematic for systems which do), viz. to employ distinctive markers for individual categories not obligatorily but only where they are not contextually redundant. Examples of such economy of use (as opposed to systemic economy; cf. Plank 1987) are the non-use of Plural when a noun is accompanied by a numeral, or the non-use of Accusative when a direct object, on account of its indefiniteness or for other reasons, is unlikely to be mistaken for a subject 12 — phenomena alluded to above in pointing out differences in the use of identically named members of the paradigms of el and manus, which are, thus, less arbitrary than might have seemed.

3. Some things that it would be good to know more about The division of labour, so dear to Adam Smith, is a tried principle in economics, so why not let it guide us when carrying on the enterprise of Smith & Co in the field of inflection? One branch of it that needs to be boosted is to do with the empirical testing and refinement of generalisations about the clustering of properties like those surveyed above. By now, instances

Of abundance and scantiness in inflection

17

of the intermingling of what we have called el and manus properties in single paradigms, in the various paradigms for single parts of speech, and in the paradigms for different parts of speech within a language are too well known for anyone to insist on their incompatibility being perforce absolute. What one would want to know more about is whether some el and manus properties are more likely to be intermingled than others, and whether some members of individual paradigms (e. g. citation forms), some paradigms of those coexisting for a single part of speech (e. g. those of words with the highest frequency of occurrence), and some parts of speech (e. g. verbs) are more prone than others to such intermixtures — and if so, why. Over the centuries the typological notions of agglutination and flexion have acquired quite a reputation, though not an impeccable one; yet praise as well as damnation may have been premature, based on insufficient knowledge about permissible cross-linguistic variation between the opposite extremes of paradigms like those of Turkish el and Latin manus. Another direction where much headway is yet to be made is the search for regular patterns in the manifestation of individual properties from either of the two typological sets. Especially interesting here are the two deviations from one-to-one mapping between inflectional meanings and forms which are conditional on cumulative exponence (itself a syntagmatic deviation from biuniqueness), viz. synonymy and homonymy. Unlike Turkish noun inflection, that of Latin does, uneconomically, involve synonymous exponents the choice between which depends on morphological classifications of nouns. More frequently than Turkish noun inflection, that of Latin does, economically, involve exponents which cut across paradigmatic distinctions. Turkish and Latin noun inflection can, thus, be assigned to opposite classes defined by these two criteria (more opposite on the first criterion than on the second). But what one might want to know further is whether synonymies and homonymies as observed especially in Latin are in any way predictably patterned. Can inflection classes be arbitrarily few or many? How uniform or diverse may they be? May arbitrarily few or many paradigmatic distinctions fall victim to homonymy? Are all paradigmatic distinctions equally susceptible to homonymy? Can all cases of non-distinctness of forms for distinct meanings be treated on a par, and are they all equally random? The contributors to the present volume are employed in labouring at this second batch of questions. They examine how paradigmatic distinctions, as opposed to mere meaning distinctions, are established, and how and why they may occasionally be masked; how the abundance and complementary scantiness of forms furnishing paradigms is patterned; and what these patterns imply for the structure of paradigms and their grammatical and/or lexical

18

Frans Plank

representation. The remainder of this introduction is devoted to tracing some of the footsteps in which those are likely to follow who are in pursuit of paradigmatic patterns.

4. Perceptions of patterns Phenomena such as those dealt with in this volume have traditionally been seen primarily as constituting practical problems in the description of particular languages. Questions which used to make life difficult for the grammarian of Latin, for instance, typically included these: Assuming that Cases are to do with the encoding of syntactic and semantic relations, without, unfortunately, corresponding to these biuniquely, how many Cases should there be recognised? Should the same Cases be recognised in the Singular and in the Plural, in spite of some of them (Nominative and Vocative, and Dative and Ablative) always looking alike in the Plural? Should the same Cases be recognised with all words inflecting for Case, in spite of great differences among many of these words as to the formal distinctions they actually displayed? Into how many inflection classes should nouns (and adjectives) be divided, considering that the different inflectional repertoires rarely differed completely but overlapped more or less extensively? Should Gender be seen as being subordinate to Declension, considering that there were inflectional patterns (such as the non-distinctness of Nominative and Accusative) characterising all Genders regardless of Declension? Sometimes recourse could be had to purportedly general principles (such as that demanding such paradigmatic Case distinctions as would render syntactic rules of Case assignment most simple); sometimes the ordinary working grammarian's descriptive solution would be ad hoc and arbitrary; usually the preferred solutions would be in line with tradition. It was rarely appreciated that what complicated the description of a particular language might be amenable to cross-linguistic generalisations. Formal non-distinctness was early perceived not to be a unitary phenomenon. Given the existence of a paradigmatic opposition, it could fail to be reflected by a contrast of form in what seemed to be a more accidental or a more systematic manner; and the more systematic a non-distinction, the greater seemed its impact on the structure of the language to be described,13 as well as on its historical development. Many contributions to this volume attempt to refine this differentiation, once a favourite subject for terminologymongers. When generalisations were suggested about possible and impossible,

Of abundance and scantiness in inflection

\9

likely and not so likely, non-distinctions in paradigms, these were commonly assumed to apply to the more systematic, genuinely morphological instances only. Perhaps the most familiar of these is that similarity of meaning is conducive to systematic non-distinction. Bazell (1960) adds that semantic similarity must be accompanied by formal similarity in order to be able to provoke syncretism (his perspective is diachronic)14 — which somewhat blurs the distinction between the accidental and the systematic, between non-distinctions which are mere by-products of phonology and such as morphology alone is accountable for. The problem with the thesis that non-distinction is encouraged by semantic affinity, supported as well as questioned in this volume, is its great adaptability. Expressing terms of the same categories, all members of a paradigm are of course similar in meaning, but to determine which are more and which less similar is notoriously controversial. Often there are equally natural alternatives of grouping terms (e. g. Cases may be divided into syntactic and adverbial ones as well as into independent and oblique ones, Persons into those referring to the speaker and to others as well as into those referring to speech-act participants and to non-participants), and sometimes groupings that seem natural to one analyst may strain the credulity of others (of which the Jakobsonian Case 'correlations', often referred to in this volume, are the most striking example). If meaning similarity can, thus, be defined almost at will, this constraint risks becoming vacuous, no matter how numerous the particular instances where its appeal is undeniable. 15 Even more worryingly, instances can also be found where one is lured to a diametrically opposite position, holding that non-distinction is favoured by dissimilarity rather than similarity of meaning. German verb inflection is a case in point (see Table 5), since here 1st and 3rd Person, no natural class on any plausible criterion, coincide in the Singular of all Moods and Tenses except Indicative Present, and with preterite-presents in this subparadigm too, and generally in all Plurals, with -e or -0(Sg) and -en (PI) serving as non-distinct exponents. The odds presumably are that this is not the most typical pattern of reducing inflectional resources. But who can really tell, with reliable cross-linguistic information still lacking for most paradigmatic categories? As will have been noticed, individual subparadigms in our examples (Tables 1 — 5) were set out as lists, with no internal structure other than linear order, and formal non-distinctions usually involved neighbours. Where non-neighbours coincided — e. g. 3sgPoss and 3plPoss of all Cases in the Plural in Table 1, Gen and Voc/Nom in the Singular of ignis in Table 3, or Abl and Dat in the Singular of the hypothetical noun in Table 4 — the subparadigms concerned could easily have been rearranged accordingly. There is, however,

20

Frans Plank

Table 5. German verb inflection Pres

Sg 1 3 2 PI 1 3 2 Sg 1 3 2 PI 1 3 2 Sg 1 3 2 PI 1 3 2

Past

Ind

Sub

Ind

Sub

jei'-i' gib-t gib-st geb-en geb-en geb-t lern-e lern-t lern- st lern-en lern- en lern-t kann-0 kann-0 kann-st kann- en konn-en kann- 1

geb-e geb-e geb-est geb-en geb-en geb-et lern-e lern-e lern-est lern-en lern-en lern-et konn-e konn-e könn-est konn-en konn-en konn-et

gab-0 gab-0 gab-st gab-en gab- en gab-t lern-t-e lern-t-e lern-t-est lern-t-en lern-t-en lern-t-et konn-t-e konn-t-e konn-l-est konn-t-en konn-t-en konn-t-et

gäb-e gäb-e gäb-est gäb-en gäb-en gäb-et lern-t-e lern-t-e lern-t-est lern-t-en lern-t-en lern-t-et könn-t-e könn-t-e könn-t-est könn-t-en konn-t-en konn-t-et

'give' (strong)

'learn' (weak)

'can' (pret-pres)

an empirical question here, viz. whether the different subparadigms for one category can always be in the same linear order, as determined on the evidence of non-distinctions. Recall that Latin Cases were presented in the order VocNom-Acc-Abl-Dat-Gen in both Singular and Plural subparadigms; comparing these two subparadigms for each noun of Table 3, it will be seen that in each case the non-distinct forms turn out to be neighbours in the same lists. For ignis (Declension Illb) the re-ordered list would have to be Voc-Nom-GenAcc-Abl-Dat, but this too would account for non-distinctions in the Singular (Voc = Nom = Gen, Abl = Dat) as well as in the Plural (Voc = Nom, Abl = Dat). If this were feasible in general, these uniform linear orderings across subparadigms would seem to capture a significant aspect of paradigm structure — and, to the extent that non-distinctions are constrained by semantic similarity, of the structure of meanings expressed in paradigms as well. On the other hand, patterns of non-distinctions are conceivable which would require representational structures more complex than lists. If, for instance, a Case paradigm consistently had non-distinctions only within the syntactic and/or within the adverbial subsets of Cases, a tree structure (or an analogue, likewise involving higher-level classificatory features) would be more appropriate for its representation.16 The lists, i. e. parts of columns, in

Of abundance and scantiness in inflection

21

Table 5 suggest that 3rd Person could coincide with 2nd (its downstairs neighbour) as well as with 1st; but 1st = 2nd Person in unattested. This pattern too would be more adequately modelled in a tree where 3rd Person is linked more closely with 1st than with 2nd:

1

gab-0

φ =

gib-t gab-0

^ φ

gib-st gab-st etc.

What this tree also implies is that 2nd Person might coincide with 3rd and 1st if these two coincide with each other. Another not entirely satisfactory feature of this tree is its incompatibility with the pattern of stem allomorphy which in the Singular Present Indicative unites 3rd Person with 2nd (gib-} in contradistinction to 1st (geh-}. Wherever more than one segment of an inflected form participates in the signalling of paradigmatic distinctions (such as, in the present example, an ending, sometimes preceded by a Past Tense suffix, and the verb stem itself), non-distinctions may, in some such manner, be distributed differently with different segments, necessitating more elaborate representations, such as the superimposition of one tree on another. But again, if non-distinctions in general are induced by similarities of meaning, one would still expect to find semantic correlates to any different groupings of different segments of paradigm members. Another general hypothesis, appearing in various guises, is that nondistinction, at least of the more systematic kind, is the less likely the more important an opposition. One way of gauging importance is in terms of redundancy: a distinction is important if it is not contextually recoverable. Thus, it could be argued that keeping the Genitive distinct from Cases marking head nouns is, ceteris paribus, more important than distinguishing subject and object Cases such as Nominative and Accusative, because attributive nouns could otherwise be mistaken for heads whereas the inherent semantics of nouns and verbs renders the confusion of subjects and objects not very likely even if not distinctively encoded (cf. Plank 1980). The Latin nouns of Table 3 would seem to confirm this prediction, showing eleven instances of Nominative coinciding with Accusative (mostly, however, in Neuters) but only three of a non-distinct Genitive. Of all three Persons of German verbs it is the 3rd where Number distinction is most important in a similar sense: finite verbs are obligatorily accompanied by subject noun phrases which themselves usually distinguish Number, except when consisting

22

brans Plank

Table 6. Plural inflection of definite article and nouns in German (partial)

Nom

Sg PI

d-er d-ie

Letter -0 Leitet 0

Ace

d-en d-ie

Leitet 0 Leitet •0

Gen

d-es d-er

Lei ter- s Leiter -0

d-es d-er

Messer-s Messer-0

'the leader' (Masc)

Sg PI

d-as d-ie

Messer Messer

d-as d-ie

Messern 0 Messe A-0 'the knife' (Neut)

Sg PI

Leiter-0 Leiter-n

Leiter-0 Leiter-n

Leiter-0 Leiter-n

'the ladder' (Fern)

of the 3rd Person pronoun sie, which may be Singular Feminine or Plural. But no Person actually does fail to distinguish Numbers.17 (Lern-t may be Singular or Plural, but not of the same Person.) More in line with the present hypothesis, the inflections of words which tend to co-occur syntagmatically sometimes do complement each other in their economising. For instance, Masculines and Neuters of some noun classes in German lack a distinctive Plural suffix -» in the Nominative, Accusative, and Genitive which corresponding Feminines have;18 definite articles, by contrast, distinguish Singular and Plural of these Cases in the Masculine and Neuter subparadigms but not in the Feminine (cf. Table 6). All in all, however, such neatly complementary patterns seem rather sporadic, as are non-distinctions where unimportance is defined in terms of limited communicative utility. To mention only the bestknown example of this kind: Why should inanimate nouns and pronouns, which are not very likely to be used as transitive subjects, have a distinct Case form specialised for this function? Hence the non-distinction of Nominative and Accusative with Indo-European Neuters, the original class of inanimates. All these importance-based patterns could be interpreted as analogues of the optional rather than obligatory expression of paradigmatic contrasts so familiar from languages where el rather than manus properties are prevalent (cf. Plank 1987). Markedness is another notion which has traditionally been brought to bear on the question at issue here:19 non-distinction is likelier to occur in marked than in unmarked contexts. Thus, on the assumption that Plural is

Of abundance and scantiness in inflection

23

marked vis-a-vis Singular, Plural subparadigms should exhibit more nondistinctions than corresponding Singular ones. They indeed do so in our examples. There are, in Turkish, more non-distinctions of Person and Number of possessor in the bottom Plural section of Table 1 than in the top Singular section; and Plural possessor forms once, with 3rd Person, fail to distinguish Number of nouns, which Singular possessor forms do never. In Latin (Table 3), nouns fail to distinguish Cases more frequently in the Plural than in the Singular; none has more distinct forms in Plural subparadigms than in Singular ones. Likewise German verbs (Table 5) may have three distinct Person forms in the Singular, but have only two in the Plural. On the assumption that Nominative is unmarked vis-a-vis all other Cases, its subparadigms should be the most distinctive ones. Accordingly, in Turkish (Table 1) 2nd and 3rd Person Singular possessor forms are only distinguished in the Nominative but in no other Case. Contrary to this prediction, however, it is precisely in the Nominative (and Vocative) where Latin nouns (Table 3) ever fail to distinguish Number: see the paradigm of die:s.20 On the assumption that Subjunctive is marked vis-a-vis Indicative, it does not come unexpected that German verbs (Table 5) in this Mood never distinguish all three Persons, as happens in the Indicative (Singular Present) of strong and weak verbs. The same pattern recurs when Present, unmarked, is compared to Past, the marked Tense. On such evidence, even if not entirely unequivocal, markedness valuations are something not to be neglected in structuring paradigms. One wonders whether traditional paradigm designers had an inkling of the significance of this notion when placing certain terms — the unmarked ones — in the first or other peripheral positions of their lists or matrices. There may be a sense in which distinction can be said to be more important in unmarked than in marked contexts.21 When non-distinctions were claimed to distribute unevenly over the several categories a word may inflect for, importance certainly was a chief consideration. (The uneven distributions discussed so far were over the several terms realising one category.) A commonly encountered hypothesis along these lines is that if nouns and other nominal words inflect for both Case and Number, Case, a surface syntactic category, is less important than Number, a referential semantic category, hence will more readily admit non-distinctions.22 Latin nouns (Table 3) behave as predicted: there are all kinds of non-distinctions of Cases, but only in Declension V (die:s) do Singular and Plural coincidence for two Cases. In the list (i. e. column) arrangements in Table 3, non-distinct forms thus are usually within either the Singular or the Plural subsection (where they are almost always neighbours); and in the matrix arrangement of Table 2 identities are vertical (and once diagonal: manu:s SgGen = PIVoc/Nom/Acc) but never

24

Frans Plank

Table 7. Inflection of Latin manus

Voc

Sg PI

Nom

Sg PI Sg PI Sg PI Sg PI Sg PI

Ace Abl Dat Gen

manus manu:s manus manu:s manum mantes manu: manibus manui: manibus manu:s manuum

horizontal. If the paradigm of manus is rearranged as in Table 7, where forms are first grouped by Case and only subsequently by Number (it was done the other way round in the lists of Table 3), non-distinct forms are scattered among the different Case subparadigms, rather than being corralled in single subparadigms. Graphic representations as in Tables 3 and 7 differ in facilitating the recognition of non-distinctions (of Case) and distinctions (of Number) respectively. Traditionally those like Table 3 have been preferred over those like Table 7, presumably because they were felt to reflect more adequately which category is dominant over the other, and which should therefore provide the first criterion for grouping.23 Two-dimensional representations as in Table 2 are neutral in this respect, lending themselves to vertical as well as horizontal grouping — though apparently more easily to the former (cf. Pike 1963), hence the inclination to interpret columnal categories (in Table 2 Number) as dominant. Determining the relative importance of categories competing for paradigmatic dominance is of course not unproblematic. One promising measure might be Sapir's (1921: ch. 5) conceptual typology, distinguishing, from most to least 'important', basic, derivational, concrete relational, and pure relational concepts, which has recently been reformulated by Bybee (1985) in terms of semantic 'relevance'. Greater basicness or relevance of a category tends to be reflected by the greater proximity of its exponents to stems, if the categories concerned are not expressed cumulatively. Owing to its concrete relational or even derivational nature, Number should thus be expressed more closely to noun stems than Case, a pure relational concept — as it actually is in Turkish and with some regularity also elsewhere (cf. Universal No. 39 of Greenberg 1963). With cumulative exponence one analogue to positional

Of abundance and scantiness in inflection

25

closeness to stems would be greater resistance to non-distinction, representable in terms of paradigmatic dominance, which one would then expect to be cross-linguistically quite uniform. As usual, however, reality falls somewhat short of expectations. Many languages indeed follow the Latin noun pattern and economise on Case rather than Number distinctions. But languages are on record which do the reverse, among them Finnish (not as consistently agglutinative as Turkish), where all fourteen or so Cases are distinct from one another in both Singular and Plural, while in two Cases, Comitative and Instructive, the two Numbers are not. While it would be rash to renounce all pretensions to predictability, such cross-linguistic diversity suggests that at least some categories will have to be ranked on a language-particular basis, if dominance is to correspond to less extensive non-distinction, as is acknowledged by Hjelmslev (1935: 107 f.) and Carstairs (1987: 118-124). With some categories dominance relations seem reasonably uniform, even though conceptually they hardly admit of any ranking. Thus, Person and Number in verb inflection are both purely relational and not very 'relevant', hence should be equally susceptible to non-distinction; but in German and elsewhere Person distinctions are sacrificed more readily than those of Number. In independent pronouns, on the other hand, where the 'relevance' of both categories increases since their function here is less ancillary than in agreement or crossreference, non-distinctions of Person are less frequent than such of Number. The requirement that paradigmatic categories must be ranked, no matter how unpredictably, can still be interpreted as constraining non-distinctions. It effectively implies that a choice must be made between categories in cutting down on distinctions: no two categories must be simultaneously affected, at least not to the same extent. With nouns of Declension V in Latin both Number and Case are affected, but Number less heavily so than Case (in only two out of six instances, as compared to six out of ten instances for Case). In German verb inflection (Table 5) Number is affected less than Person, and Tense less than Mood, neither Number nor Tense. being affected at all. Whether Person or Mood is the primary victim, however, is difficult to tell especially with weak verbs, where Indicative and Subjunctive are not distinct in nine instances out of twelve where they could have been distinguished, and 1st and 3rd Person are not distinct in seven instances out of eight; but the consistent distinction of 3rd and 2nd presumably gives Person the edge over Mood, and makes this paradigm conform to the hypothesis of categorially selective non-distinction. At first sight this hypothesis does not bear on a paradigm like that of object prefixes on verbs in Fore (a Papuan language of the Eastern Highlands of New Guinea), since all three Persons and three Numbers are fully distinguished, as shown in Table 8a. It is only when these

26

f-'rans Plank

Table 8. Object prefixes in Fore a. internally unanalysed Person Number

Sg PI

Du

1

2

3

natatasi-

katitisi-

aiisi-

b. internally analysed

1

Sg PI Du

"

n LJ

a

κ

a

0

a

m

\

~T\

0

~T

0

i

a

uJ

1

LJ

si

a

si

prefixes are internally analysed into what Pike (1963) calls 'matrix formatives' that partial identities are recognisable (see Table 8b), and some of these are not categorially selective. The formative -si, a straightforward morpheme for Dual, is the same for all Persons, and 0-, the zero exponent for 3rd Person, is the same for all Numbers; but the formatives -a, -i, and /- fail to distinguish Persons as extensively as Numbers. It remains to be seen whether undiscriminating non-distinction only occurs in such circumstances where syntagmatically combined segments, and not even proper morphemes at that, co-express inflectional categories. The patterns of inflectional abundance have attracted far less generalisation-seeking attention than those of scantiness. Perhaps the most popular subject under this heading has been suppletion: this blatant disregard for economy by providing two or more entirely dissimilar stems per word to be inflected is supposedly something only particularly frequent words (e. g. general verbs meaning 'to be', 'to have', or 'to do', or common evaluational, quantificational, or dimensional adjectives such as 'good/bad', 'many/few', 'large/small') will be able to afford. Further, suppletion as well as less radical allomorphic stem alternations, as seen above in Latin noun inflection (Table 3) and German verb inflection (Table 5), have been supposed to be preferably conditioned by those inflectional categories which are paradigmatically dominant, i. e. are closest to the stem and/or relatively immune to non-distinction. (Note, however, that stem allomorphy of non-weak verbs in German is

Of abundance and scantiness in inflection

27

conditioned by Person (geb-\gtb-~] and Mood (gab-\gab-, kann-\konn-, konn-t-j konn-t-\ two non-dominant categories.) As to inflection classes, brought about by exponential allomorphy conditioned by stems, the constant disagreements about how many of them to distinguish in particular languages seem to have long discouraged comparatists from looking for general principles that might curb this confusing profusion. At least the question of how many inflection classes there can possibly be in a single language could perhaps be answered derivatively: about as many as there may be semantic classes, out of which inflection classes are assumed to develop in a process of grammaticisation (cf. recently Wurzel 1986). But for answers to questions such as this one might be inclined to turn to the Guinness Book of Records rather than to the theory of inflection. Some elementary arithmetic might also be useful here, though, given a little information about the inflectional resources of a language. Imagine a language, Residual Latin, where nouns inflect for three Cases and two Numbers and there are these exponents for Case-Number combinations, with allomorphs conditioned idiosyncratically by noun stems: SgNom:-o, -us Ace: -am, -urn, -em Gen: -/ PI Nom:-iZi Ace: -as, -os Gen: -arum, orum How many separate inflectional repertoires could there possibly be, differing at least on one Case-Number combination? Multiplying the numbers of exponents for each category combination with one another gives the correct answer: twenty-four ( 2 x 3 x 1 x 1 x 2 x 2 ) . All of them are listed in Table 9. Minimally there would only be three noun declensions in Residual Latin: this is the highest number of allomorphs for any single Case-Number combination (SgAcc). Having as many as twenty-four inflection classes for a single part of speech would be somewhat unusual for a real language; having as many as real Latin could have for its nouns would be ludicrous (assuming nine allomorphs for SgVoc, nine for SgNom, six for SgAcc, six for SgAbl, six for SgDat, five for SgGen, seven for PIVoc, seven for PINom, eight for PlAcc, three for PlAbl, three for PIDat, six for PIGen, and ignoring the Locative, the sum total is 1,851,776,640) — hundreds of millions of them would lie idle for the lack of nouns!24 What the 'Paradigm Economy Principle' of Carstairs (1987: ch. 3) essentially predicts is that the actual number of inflection classes, far from ever approximating the theoretical maximum, will

28

Frans Plank

Table 9. Possible inflection classes in Residual Latin

II

III

IV

V

VI

VII

VIII

-a -am -i -ae

-a -am -i -ae

-a

-a

-a

-a

-urn

-urn

-urn

-urn

-i -ae -as

-a -am -i -ae -as

-i -ae

-OS

-i -ae -as

-i -ae

-OS

-i -ae -as

-OS

-OS

-arum

-orum

-arum

-orum

-arum

-orum

-arum

-orum

IX

X

XI

XII

XIII

XIV

XV

XVI

-a -em -i -ae -as

-a -em -i -ae

-a -em -i -ae

-us -am

-us -am -i -ae

-us -am -i -ae

-OS

-OS

-us -am -i -ae -as

-OS

-OS

-orum

-arum

-orum

-arum

-orum

-arum

-orum

XVIII

XIX

XX

XXI

XXII

XXIII

XXIV

-us -um -i -ae -as

-us -i -ae

-us -um -i -ae

-us -em -i -ae

-OS

-us -em -i -ae -as

-us -em -i -ae

-OS

-us -em -i -ae -as

-OS

-OS

-orum

-arum

-orum

-arum

-orum

-arum

-orum

I

Sg Nom -a Ace -am Gen PI Nom

Ace Gen

Sg Nom -a Ace -em Gen -i PI Nom -ae Ace -as Gen -arum XVII

Sg Nom -us Ace -urn Gen -i PI Nom -ae Ace -as Gen

-arum

-urn

-/'

-ae -as

in fact be the minimal one possible, i. e. will equal the number of allomorphs of the category combination most richly endowed with allomorphs. On the assumption that Vocative as well as Nominative Singular have nine allomorphs each in Latin (viz. -a\-e\-um\-us\-u:\-e:s\-s\-0\-is and -a\-us\-um\-u:\-e:s\ -sl-0/-isj-e), there should, thus, be no more than nine declensions; but in Table 3 we distinguished eleven. This is hardly a serious transgression; moreover, it is sanctioned by a refinement of Carstairs's principle which permits two or more paradigms to be counted as one if the differences between them correlate consistently with lexically determined syntactic properties. Gender is such a property in Latin noun inflection, which for Nominative Singular, for example, partners -us and -urn, -s and -0, and -is and -e as Masculine or Feminine and Neuter variants, bringing the total of allomorphs which count down to six, now the maximum reached by any Case-Number combination.25 Nonetheless, even after Declensions IIM and IIN, IVM/F and IVN, IIIaM/F and IIIaN, and IIIb M/F and IIIbN are combined to single 'macro-

Of abundance and scantiness in inflection

29

paradigms', the inflection classes in Table 3 still number seven, one more than permitted (and two more than traditionally recognised in grammars of Latin, given to glossing over the heterogeneity of the Third Declension). We may identify Declension Illab (urbs\ resembling Declension Ilia (re:gs) in the Singular and Declension Illb (ignis) in the Plural, as the offender: such principled mixtures of paradigms are in fact licensed by a 'Slab Codicil' to Carstairs's Paradigm Economy Principle (1987: 81). There are actually two questions here, which are beginning to shade into one another: How many inflection classes may coexist? How may the inflectional resources available be distributed among the several classes? Purely in terms of resources it would make no difference whether Latin had six, seven, eleven, or 1,851,776,640 declensions of nouns, the amount of Case-Number exponents that need to be memorised remaining the same. It is the manner in which the allomorphs for Case-Number combinations are assembled which may cause or save the learner a lot of trouble. Only if the allomorphs available for each Case-Number were assembled in all possible combinations could there be 1,851,776,640 declensions in Latin (or 24 in Residual Latin); and every inflectional set would then have to be memorised allomorph by allomorph. It would facilitate the task of the learner if the choice of an allomorph for one Case-Number would follow from the choice made for another CaseNumber. Note that if there were such interdependencies, this would automatically reduce the diversity of inflection classes. Of course the ideal here would be for all Case-Number exponents mutually to imply all others; but this is only feasible if inflection is uniform or no two inflection classes share a single exponent. Paradigms involving cumulative exponence typically meet neither condition. The allomorphs available for different paradigm slots usually vary somewhat in number (from three to six or nine, depending on how one is counting, in Latin noun inflection, and from one to three in Residual Latin), and this precludes complete distinctness of classes. The empirical question, then, is whether the assemblage of inflectional sets from the available allomorphs can still be constrained in any principled manner. One pertinent idea already alluded to, due to Carstairs (1987), is to look for constraints on the mixing of classes. Consider a version of Residual Latin further impoverished by the loss of Accusative Singular allomorph -urn (which eliminates the possible Declensions V —VIII and XVII —XX). Assume it has three Declensions, of which one is I and the second XXIV. Which one might be the third? Carstairs's thesis is that if an inflection class has no exponents of its own, it can only collect them wholesale, as it were, from subparadigms (called 'slabs' by Carstairs) of classes which do have exponents of their own. Thus, in our case, a third, mixed class could borrow all Singular exponents

30

l-rans Plank

from Declension I and all Plural exponents from Declension XXIV (yielding Declension IV), or vice versa (yielding Declension XXI); or it could borrow all Nominative exponents from Declension I (Plural -ae being in fact common to I and XXIV, like SgGen -i) and all Accusative and Genitive exponents from Declension XXIV (yielding Declension XII), or vice versa (yielding Declension XIII); or all Accusative exponents from Declension 1 and all Nominative and Genitive exponents from Declension XXIV (yielding Declension XIV), or vice versa (yielding Declension XI); or all Genitive exponents from Declension I and all Nominative and Accusative exponents from Declension XXIV (yielding Declension XXIII), or vice versa (yielding Declension II). Illicit unprincipled mixtures would be Declension III, IX, X, XV, XVI, and XXII. With principled mixtures the choices of allomorphs for different slots are still linked by mutual implications within each subparadigm. For Impoverished Residual Latin with Declension I, XXIV, and IV, the following mutual implications would hold: SgNom SgNom PlAcc

-a SgAcc -us SgAcc -as PIGen

-am -em -arum

PlAcc

-os

5?

SS

^S 55

55^ '
'^ '§> '&> -1

"3

5ί

B

w

56

|o ,|c ,| ,| j| g

-§ '«^ '&> '& 'δ1 '&1 -

-S -S -5 ~3 -^ -3 -1 ~S V,

'-,

"a,

B

o o W

u

-73

Q ^ «

rt

s s

u

u

w

Q

g

ΐii111H

| U

_rt

ι 1

'-W

(Λ

^i ,

7^3 ·*».

'SS; 'SS; '§ '§ § fe 5s S; *5

*ί

Syncretism and the paradigmatic patterning of meaning

'S **v

*

l

?r>

.1 l ,1 .,

S5 S5 i^* ^J" ^^* ^^

55 S -*?» ^ ^^* ^»·

^

s3

s: « •Si -as

"E
^ ^ ^

O Q i-l *-,

141

142

Jadranka Gvo^danovic

Genitive syncretism may be claimed to fit in with the notion of animacy in relation to the meaning of Goal (usually expressed by the Accusative), by which an animate Goal is never fully, but only to some extent affected by an event (and this is usually expressed by the Genitive). In other words, it is perfectly compatible with animacy that the meanings of the Accusative and the Genitive are not mutually opposed. This is presumably what enabled the Accusative = Genitive syncretism to spread to animate masculine nouns and the corresponding adjectives in the singular. Once a favourable condition is present, a spread may occur along the expected lines, but neither its occurrence nor its extent appears to be compulsory (as shown by comparison with the other Slavic languages). A syncretism which spread throughout the desinences of all the declensions (with an accentual difference preserved only in a very limited number of nouns) was the Locative = Dative syncretism. The model for it was found on the one hand in the singular of the feminine nouns, and on the other in the singular of the first and second and the third feminine person of the personal pronouns. The additional condition enabling the full syncretism of the desinences was the absence of an opposition between the Locative and the Dative: the Locative developed so as to be used exclusively with one of the prepositions ο 'about', » 'in', na On', po Over', pri 'in the presence of, during the time of, at', prema 'according to, compared with', which never combine with the Dative, whereas the Dative is predominantly used as a free Case form, and as bound by its own set of (mostly relatively novel) prepositions. Given this absence of an opposition, and in view of the fact that all the prepositions which combine with the Locative have "presence of a limit or a definite outline" as a part of their meaning, it is possible to consider the interpretation of "quantifier/limited" connected with the Locative as being due to its prepositional context. Therefore, "quantifier/limited" cannot be a distinctive part of its meaning, but must be a contextually conditioned interpretive variant. The meaning of the Locative is then only specified as "marginal".4 In Standard Serbo-Croatian, we have the following situation concerning the "marginal" cases, i. e. the Locative, Dative, and Instrumental: — the meaning of the Locative is specified as "marginal", with the contextually obligatory interpretation as "a quantifier/limited", and it is unspecified for "directionality"; — the meaning of the Dative is specified as "marginal" and "directional"; — the meaning of the Instrumental is specified as "marginal" and "nondirectional".

Syncretism and the paradigmatic patterning of meaning

143

In Serbo-Croatian dialects, there is nowadays a widespread system (reported by I vie in several studies, including 1961: 43) in which Instrumental forms have ceased to be used as free Case forms, but are either adverbialised or obligatorily bound by a preposition (i. e. s(a) 'with', medu 'between', nad 'above', pod 'under', pred 'in front of, before', or %a 'behind'), none of which can be combined with either the Dative or the Locative. This means that the Instrumental, Dative, and Locative occur in complementary distribution, and that their interpretational characteristics in these contexts are shared by the contexts and may be ascribed to these contexts. In paradigmatic analysis, it is, at the level of meaning, sufficient to specify these Cases as "marginal" and, at the level of form, to specify the desinences of the Instrumental as characteristic of the "marginal" Case in the context of the given prepositions, and also the accent alternations in the Locative of a subgroup of masculine inanimate nouns and of ^'-sterns (indicated in 3 a) as variants in the context of the prepositions o, u, na, po, pri, prema. There are thus no oppositions among the "marginal" Cases Locative, Dative and Instrumental, nor between the Instrumental and the Nominative, as the latter occurs exclusively as a free Case form. This is paralleled by a relatively recent elimination, not only in the dialects, of the only context in which the Locative used to be opposed to the Genitive, which was at the same time opposed to the Accusative, viz. the context of the preposition a, which is now no longer used with the Genitive in the dialects and spoken Standard Serbo-Croatian. These developments can be explained only if we assume that the absence of a syntagmatic possibility of choice corresponds with the absence of a paradigmatic opposition, and that it is able to trigger the syncretism of the corresponding forms. We thus suggest the following revised analysis of the Case system of Standard Serbo-Croatian (4 a) and of its contemporary dialects (4b): (4) a. Nominative

Accusative

Instrumental

Dative (0 etc. Locative) (directional)

b. Nominative

•Accusative

Genitive

(central) (marginal)

(quantifier/limited)

Genitive

(s etc. Instrumental) (Dative) (o etc. Locative)

(central) (marginal)

Additional independent evidence in favour of this analysis can be deduced from data on Case decay, observed in Serbo-Croatian native-speaking children

144

Jadranka Gvo^danovic

of Yugoslav immigrants in Sweden, which at the beginning of their school education acquire Swedish at a fast rate and lose their Serbo-Croatian at the same time. This holds especially for the Serbo-Croatian Case paradigm, for which Swedish has no counterpart, as it has only lexicalised relics of its former Cases, next to the possessive -s, originating from the former Genitive Case. Durovic (1983, 1984) reported on Serbo-Croatian Case usage in narratives of immigrant children of approximately seven years of age. The data in these reports were collected irrespectively of the dialectal background of the children's parents (possibly with an older or different Case system than in the standard language), and there was no systematic longitudinal investigation of the individual children, but in spite of these shortcomings the reported results still reveal important regularities about the Serbo-Croatian Case system. The Case inventories discovered by Durovic (1984: 90) in the idiolects of Serbo-Croatian speaking immigrant children in Sweden are listed in (5). (5)

1. 2. 3. 4.

NA NAG NAGL NAGLI NAGLyl NAG1 5. NAGL y ID NAGID 6. NAGLylDV

(3) (5) (2) (2) (6) (4) (4) (1) (2)

(The numbers in brackets denote the number of idiolects with a given inventory. The letter j indicates that in the given idiolect the Locative, always governed by a preposition, has been replaced once or several times by the Accusative.) These data, based on the nominal and adjectival declension, led Durovic (1984: 91) to the partly inaccurate conclusion that "there is an implicative relation NAGLIDV between the 7 cases". What can be induced from (5) instead is the set of implicative relations given in (6 a). (6)

a. N, A < G (< L) < I < D < V

The data on which Durovic's analysis is based, published in (1983), show that for the personal pronouns the implicative relations given in (6 b) can be established. (6)

b. N, A (= G/< G) < D

Syncretism and the paradigmatic patterning of meaning

145

The 1983 data show, further, that in the process of Case decay, free Case forms are replaced either by the Accusative or by a preposition with the Accusative or with the Nominative form. Bound Case forms are also replaced by either the Accusative or the Nominative form, in the context of the (preserved) preposition. An example of a free Genitive form being replaced by the Accusative is given in (7 a), and by a preposition with the Nominative equalling the Accusative form in (7 b). (7)

a. sad

samo moju mamu

ocka

now only my A mammy A daddy N lives 'now only my mammy's daddy is alive' b. slika

od mesec

picture^ of moonN = Λ 'a/the picture of the moon' Even though this decay of Serbo-Croatian has been triggered by an emphasis on the use of Swedish at the beginning of the school training of the children investigated, the way in which this decay proceeds is based on the internal organisation of the decaying system itself. This is illustrated by (7 a), in which a Serbo-Croatian possessive Genitive has been replaced by the Accusative form, whereas Swedish preserves its original possessive Genitive form (-j·) better than any other Case form. The replacement of the Genitive (as well as of the Locative, Dative, or Instrumental) by the Accusative is possible in Serbo-Croatian because the Accusative here, at least in many of its dialects, is the only Case which is directly related to all the other Cases in the paradigm, as shown in (4b) above. The decaying system follows its own regularities, by replacing its other Cases with the Accusative or the Nominative with a preposition and by preserving the "central" cases longer than the "marginal" ones. These developments confirm that syncretism between two Cases originally distinguished by a single feature of meaning can be analysed as neutralisation ofthat feature, and that the analysis given in (4 a) and (4b) has relevance to the native speakers' competence.

2.2. Person and number syncretism in conjugation In the preceding section we have seen that Case syncretisms are not evenly distributed over the other dimensions of a declension, but differ in their extent relative to the category (cf. the extent of the Accusative — Genitive syncretism in the personal pronouns as compared with the nouns and adjec-

146

Jadranka Gvo^danovic

tives), including the gender, the number, and the type of stem (cf. the extent of syncretism in the ^'-sterns as compared with the other stems) in a given declension. In a similar way, syncretisms of person or number within a conjugation may occur with reference to each other, and possibly also to the other paradigmatic dimensions (e. g. gender and/or tense) which co-occur within a given conjugation. It is interesting to analyse syncretism of person and number in so-called "complex pronominalising" languages, where not only the Agent or Experiencer, but also the Goal of an event, if present, is specified in terms of person and number in the conjugational prefixes and/or suffixes. Such languages are, for example, found among the Tibeto-Burman languages of Nepal. The Kiranti Tibeto-Burman languages which will be discussed here belong to the so-called "sensitive separatist" type in Plank's (1986: 32) terminology, i.e. their exponents originally express only one inflectional category, but undergo alternations conditioned by morphological (in addition to phonological) properties of co-occurring morphemes. I shall present here data on two Kiranti Tibeto-Burman languages from East Nepal, both belonging to its Kiranti subgroup (for a classification of Kiranti languages cf. Benedict 1972, Michailovsky 1975 and Gvozdanovic 1985). The data on the first of these languages, Bahing, were collected by Michailovsky (cf. 1975), and on two dialects of the second one, Bantawa, by the present author during field-work in the winter of 1983 — 1984 (in connection with the Linguistic Survey of Nepal carried out at the University of Kiel, Federal Republic of Germany, under the direction of Werner Winter). In the presentation of the Bahing data, I shall follow Michailovsky's notation, according to which, e. g., -rv- and -j- denote epenthetic glides, and the desinences are not further analysed morphologically. I shall do so because I am not able to solve the analytical uncertainties mentioned by Michailovsky (1975: 191 ff.), and these are not at issue in the survey of syncretisms, where a comparison of the full desinences suffices. In the presentation of the Bantawa data, I shall adopt the same system for reasons of internal comparability, even though a fuller analysis would be feasible. The data will be classified on the basis of the person and number of the Agent and Goal referred to by means of the desinences. In Bahing, only suffixes occur, and the same is true for the Chhinamakhu dialect of Bantawa (from the North of Bhojpur; Bhojpur is the district where Bantawa is originally spoken), whereas the Wana dialect of Bantawa (from the South of Bhojpur) has both prefixes and suffixes, as shown in Tables 2 —4. 5

Syncretism and the paradigmatic patterning of meaning

(Λ

_ο 3 _cυ

CL, CO

^s-S

δ e_ ·~ '-S

§

1

Ο

Os

bO

§ 5 δ ? \5, "^

§, -δ ·~ 4- ^ ^ ? ·?

ϊ

·, 55

!5 S § Ϊ5ϊ;

» Ύ I

?·?'?·?

Ύ

I

I

^ ^_ ·3" '— ·3 χ_ ·3

^ Ύ

ρ·?·^!ίί·?'^'^Φ

^ ^ Γ^.,

·, ^

R? Si ^ ^ δ ϊ ϊ

"**

^ « « I

f\*

^— ^

Q ^ 5 5

Ο-.

Ι

147

Ι

("V.

Γ*·»

Ι

ι

Ο

υ α *~—' υ

'S ^ Q

1£5 ^1

!^ ζΐ

i^ ^J

1 ξ Q

^ ^J

α

L** 2° ο

L 5 ^ t ^ ^ >

^ ^ J ^ ^

^ ^ J ^ ^ ;

d

«a

^>

? "?

? "?

? "?

Λ · Ι ^δ ^ 'ζ -5 α ·*» ~— -9 Si-S-S^s-^·^^«

1 »> "? δ ^

ο

ι *-·

^

α* ω Ζ

13

c ' 3t* » 'S» ^ S, 5 ·.. -ά, at -ά^ ? ·?

ϊί

'»^ Ε; ς ?·?'?·?

-§

'g

δ

^sj · 5 < ω'S7 ^ ' ^ • ^ Φ ' Τ ^ ' ^ Φ ^ ^

u

^ • O

1 "u

^-^ t*·.

^

"bO C

'S

'(Λ

α

04

Q?

5 e s

^ '^^

Ο^ξ ? -S

§'ϊ? '^^

O >

-o

^ ^

.^

%"? -^J-^j

Ι § '"^^

ο ?-^

οVH ^^ τ-,

'B

Ό u rt S

C 10

Ι

Ι«

:'κ 'ε; ; ·~ 'ε;

:

'•^'· ? ^'ί

VI

"Ε- ^

αϊ T -

'

αϊ O

τ\τ ^ τ

.-,

4

: "g 'S :

υ

ί Λ " ΰ Τ 3 Τ 3 " ΰ c O T - ' T - ' O 4 c o

^'·? ^ τ

._ ^

§

u

4-1

ί

(Λ

C L , A . C L , C U - * - ' O 4 c O

T.

C

ΟΙ '—ι

V 'V· -5 •5

%

X

Vi

%ϊ

? ?

.^

I | ^' ^'1 ' 1

' ' *: -s:

•»ί

|

ii ί i

i

-5

^

'§

" "*

.53

•?l-

(Λ

U

l

1

i

Q U,

ίξ §

5 S J5 > A. fl£ -5·»

·»

M

S· 8- t t ι

ι

Ί*

^

** \

^

t»

\

Ό

Si

l

3t

Χ

S$

Goal combinations in Bantawa of Chhinamakhu: Agent Goal 1s any 2s 3s, Idc, 3d, 1 pe, 3ρ 3s 3 s, 3 d, 1 pi, 1 pe, 3 ρ Ide 2s, 2d, 2p 2d l de, l pe 3d 3d, Ipi, Ipe, 2p, 3p 1 pi any 1 pe any 2p 3s, Ide, 3d, Ipe, 3ρ 3ρ 3 d, 1 pi, l pe, 2 ρ, 3 ρ (10)

a. Syncretisms in number and person of the Agent in the transitive conjugation in Wana Bantawa:6 Agent Goal 3d=3p 2s 3s=3d Idi 2s=2d=2p Ide 3s = 3d, l d e = l p e 2d

3s=3d(NP), 3d(P) = 3p

Ipi

2s = 2d = 2p, 3s = 3d(NP), 3d(P) = 3p

Ipe

l d e = l p e , 3d(NP) = 3p

2p

b. Syncretisms in number and person of the Goal in the transitive conjugation in Wana Bantawa: 6 Agent Goal any, except 3d 3d = 3ρ 2s l de = l pe Ide 2d = 2p 2d 1 de=l pe Ipe 2s = 2d = 2p 2p l d e = lpe c. Tense syncretisms for particular Agent —>· Goal combinations in Wana Bantawa: Agent Goal ls any

2s

3 s, l de, 3 d, 1 pe. 3 p

Syncretism and the paradigmatic patterning of meaning

3s Ide 2d 1 pi 1 pe 2p 3ρ

\ 53

3 s, 3 d, 1 pi, 1 pe, 3 p 2s, 2 d, 2 p Ide, Ipe any any l de, l pe 3d, 1 pi, Ipe, 2p, 3p

These genetically related language systems differ considerably in the distribution of syncretisms observed in their transitive conjugations. Whereas Bahing has no tense syncretisms, both dialects of Bantawa have tense syncretisms sensitive to person and number. On the other hand, whereas Bahing and Bantawa of Chhinamakhu have syncretisms in person and number, Wana Bantawa has only syncretism in number. However, in all three systems, syncretisms in person and/or number are too regular to be attributable to chance. The regularities are the following: — Number syncretisms may involve the dual and the plural, or the singular, the dual, and the plural in all three systems; in Bahing and Wana Bantawa, they may also involve the singular and the dual, and in one attestation in Chhinamakhu Bantawa (i.e. 2s = 3s = 3p —> Is) the singular and the plural, which shows that there are asymmetries involved in the patterning of number in Chhinamakhu Bantawa. — Person syncretism never involves the first person singular; it may involve the first person dual or plural, which then coincides with the second person, or with the third person on the condition that the first person exclusive is involved, which has in its meaning an overlap with the third person; in addition, the first person dual or plural participates in person syncretism only if it also participates in number syncretism; unlike the first person, the second and the third person may syncretise without restrictions. On the basis of these regularities, we are justified in assuming that syncretism as an identity in form presupposes an identity at some level of meaning, unless this formal identity is due to phonological developments or distributional restrictions. In addition, whenever syncretisms are clearly restricted, we are justified in assuming that such restrictions correspond to a meaning patterning. In the case of number syncretisms, Bahing and Wana Bantawa exhibit a restriction barring straightforward singular-plural syncretism, with the only

154

Jadranka Gvo^danovic

possibilities being syncretisms of singular-dual, dual-plural, or singular-dualplural. \\ e may translate this into three possible patternings, a non-hierarchical one in (11 a) and two hierarchical ones in (11 b) and (11 c). (11) a. One' — 'two' — 'many' singular-dual-plural b.

ne

'two' + singular-(dual-plural) c.

many

'two'

plural-(dual-singular) In Bahing and Wana Bantawa, there is no asymmetry which would distinguish either the singular from the other two numbers, or the plural from the other two members. In the absence of asymmetry, no unequivocal markedness valuations can be established, and a hierarchy is justifiable only if an additional argument in favour of level distinctions can be found (as in the patterning of person in Chhinamakhu Bantawa discussed below). Number syncretism in Chhinamakhu Bantawa, on the other hand, reveals two asymmetries for the following reasons: — the dual and the plural typically syncretise with each other, and — the singular either syncretises with the dual if this coincides with the plural, or exclusively (in the single case o f 2 s = 3s = 3p -> Is) with the plural. One asymmetry distinguishes the singular from the dual and the plural, and another the dual from the plural. These two asymmetries are in accordance

Syncretism and the paradigmatic patterning of meaning

155

with markedness valuations at the two levels of the hierarchical patterning presented in (12).

(12)

One' m u 'two' m u singular-(dual-plural)

Asymmetry is the only conclusive formal criterion for markedness and the corresponding hierarchy. Other formal and substantive criteria are inconclusive. Considerations of substance (by which One' and 'two' are conceptually singled out as opposed to 'many') are relevant to language patterning only if they are paralleled by considerations of form (as in Chhinamakhu Bantawa, which has markedness and hierarchy in the patterning of number, whereas Bahing and Wana Bantawa do not have these valuations). In the case of person syncretisms, both Bahing and Chhinamakhu Bantawa have the restriction prohibiting the first person singular from syncretising with any of the other persons. This may be taken as an argument in favour of the markedness of the first person as opposed to the second and the third person. This markedness is, however, bound by the number context. In both Bahing and Chhinamakhu Bantawa, the first person of the dual and the plural does syncretise with the second person. In Chhinamakhu this holds only for the inclusive first person, and in Bahing irrespectively of inclusiveness (i. e., the first person exclusive may syncretise either with the second person of the same number, or with the third person, with which it has an overlap in the meaning, "exclusive" referring to the speaker and a person or persons outside of the speech situation). In both Bahing and Chhinamakhu Bantawa, it does not follow from the observed syncretisms that the first person of the dual or plural is marked as opposed to the other two persons. It also does not follow that the second and the third person are patterned at a level which is hierarchically distinguished from the level at which the first person is patterned. The only argument in favour of a hierarchy as based on level distinctions in the dual and the plural can be deduced from the 2s = 3s = 3p —> l s syncretism, which cannot be explained if there is a level difference between the third person singular and the third person plural. It is in view of this evidence on levels combined with the evidence on the markedness of

156

Jadranka Gvo^danovic

the first person singular that the analysis in (13) of the number and person patterning in Chhinamakhu Bantawa can be proposed. (13)

'speaker' m u

'speaker'

'hearer'

'hearer'

Is

2s

3s

'speaker'

Id

2d

3d

'hearer'

2p

3p

This analysis shows that a hierarchical patterning may exist in the absence of markedness as well, but on the condition that considerations of form point to the presence of levels. The analysis proposed for Chhinamakhu Bantawa would seem to be a plausible analysis of a system which, obligatorily in the dual and the plural, combines the "speaker" either with the "hearer" (i. e. "inclusive") or with somebody who is not the "hearer" (i.e. "exclusive"). Tibeto-Burman systems which are losing the "inclusive" vs. "exclusive" distinction may be assumed to have developed markedness of the first person in the dual and the plural as well, giving preference to the first person signalization as opposed to the other persons in these numbers as well. The analysis presented here fits with the evidence from other TibetoBurman languages as presented by Bauman (1975) and DeLancey (1981), pointing to a general tendency in Tibeto-Burman to express the first and the second person in preference to the third. On the basis of their data, markedness of the first person as opposed to the second and the third one, and markedness of the second person as opposed to the third one can be established. The data on Bahing and Chhinamakhu Bantawa presented here add details that could not be seen in the other languages. The possibilities of hierarchical analysis appear to be compatible with the analysis of person in Standard Average European and other languages offered by Plank (1985). The Bahing and Chhinamakhu Bantawa hierarchy of persons also seems to

Syncretism and the paradigmatic patterning of meaning

157

be compatible with Silverstein's (1976: 117) hypothetically universal person hierarchy in languages with an "inclusive" vs. "exclusive" distinction (14), but there is a fundamental difference in the analysis of number.

(14) a. , b. c.

A B * C D E F G H I J K

[ + /— ego] \ ' , [ + /— tu] [ + /— plural]

, r , · π d. [ + /— restricted]

A. B. C. D. E. F.

+ _ _ _ _ _ _ + + _ + + _ + + - (+) + - (+) + - (+) + - (+)

first person inclusive dual first person inclusive plural first person exclusive dual first person exclusive plural first person singular second person dual

G. H. I. J. K.

person number

second person plural second person singular third person dual third person plural third person singular

The reason for the difference in the analysis of number is not one of data, but one deriving from the theoretical standpoint adopted by Silverstein, by which feature specifications are considered to equal markedness, and are as such required to be consistent with surface distributional properties such as frequency. It is on the basis of this approach (in itself inconclusive unless frequency is weighed against its formal and substantive context) that the dual can be viewed as the most marked number, because there are fewer attestations of it than of the plural or the singular. Silverstein's hierarchy cannot explain the patterns of syncretism in Bahing and Chhinamakhu Bantawa, and the general reason for it can be found in the treatment of markedness and its relation to hierarchy. The language data presented here show that markedness may, but need not, be attached to feature specifications. Whenever there is an asymmetry of the type attested with the first person singular in Bahing and Chhinamakhu Bantawa as opposed to the other persons, this asymmetry can be ascribed to markedness, which is as a rule connected with an unambiguous patterning. In the absence of asymmetry, no markedness can be established and the patterning is ambiguous unless there is a formal indication of the presence of levels (of the type attested with the third person in Chhinamakhu Bantawa). Presence of an unambiguous meaning patterning in a language is reflected in the meaning of each morpheme which is part of the given paradigm. For example, the meaning of -ji (NP) and -//' (P) in Bahing is to be stated in terms of the Agent —> Goal relation: 'someone who is not the speaker (i. e. unmarked for 'speaker') —> the singular speaker' (and the same holds for -na

158

Jadranka Gvo^danovic

(NP) and -an (P) in Chhinamakhu Bantawa), and the distinction of second and third person of the Agent is here a matter of (contextually or situationally determined) interpretation.

3. Conclusion The general conclusion is that patterning of meanings in a language is revealed by the corresponding forms and their combinatory possibilities, and that an evaluation of this patterning is always part of the entire paradigm in which the given meanings participate. A restriction with respect to identity of form corresponds to an asymmetry at the level of meaning, translatable into hierarchy and/or markedness. Whenever it is present, it forms (part of) an unequivocal pattern, which is relevant to further language developments.

Notes 1. Abbreviations: A = Accusative, D = Dative, d = dual, e = exclusive (i.e. excluding the hearer), G = Genitive, I = Instrumental, i = inclusive (i. e. including the hearer), L = Locative, m = marked, N = Nominative, NP = non-preterite, P = preterite, p = plural, s — singular, u — unmarked, V = Vocative (in its meaning, the Vocative instigates the entity to participate in the event), 1 = the first person, 2 — the second person, 3 = the third person. 2. Notation of the Serbo-Croatian prosody: (") = the short falling accent, ( Λ ) = the long falling accent, ( ') = the short rising accent, ( ') = the long rising accent, ( ~) = postaccentual vowel length. 3. There is a general Dative = Instrumental syncretism in the dual in the oldest written documents in Slavic, i. e. in the pre-Serbo-Croatian period. From the thirteenth century onwards there is a merger of the original Dative and Instrumental plural endings, with the Instrumental used for the Dative plural and the Dative for the Instrumental plural. This merger of Dative and Instrumental in the plural preceded the first attestations of the dual Dative = Instrumental desinences in the plural and the corresponding abolishment of the dual as a category in (Stokavian and Cakavian) Serbo-Croatian. See Belie (1950: 101 — 114) for the data. 4. In an article on Case developments in various Slavic languages, Comrie (1978) has shown that the formal classification of Cases tends to correlate highly with their functional classification. Comrie's results are compatible with my own proposals, except in the following respect. Comrie distinguishes between "syntactic" Cases (originally in Slavic the Nominative and Accusative, later also the Genitive), defined as "correlating more readily with syntactic functions" and expressing "a wide range of semantic roles", and "semantic" Cases (the remaining ones), whereas in my view, which is in agreement with that of Jakobson (1936,

Syncretism and the paradigmatic patterning of meaning

159

1958), a morphological Case is "semantic", i.e. paradigmatically opposed to one or several other Cases, if it can occur within the same configuration as the other Case(s) and, by doing so, systematically contributes a differential meaning to this configuration. In this categorical sense the Cases classified by Comrie as "syntactic" are still "semantic", and the shift from semantic to syntactic Case functions discussed by Comrie has (so far) been restricted to shifts in configurations in which the Cases may occur, still in accordance with the original Case meanings; through sense shifts some paradigmatic oppositions among the Cases have ceased to exist, especially when the Locative ceased being usable as a free Case form, opening the way for new syncretisms. While very valuable, Comrie's analysis is incomplete here. 5. Bantawa is originally spoken in Bhojpur and, due to migrations, also in the more eastern districts of East Nepal. The only data on its dialectal distribution have been collected within the framework of the Linguistic Survey of Nepal, in connection with which I have collected the data used for the present study. On the basis of the total data, preserved at the Linguistics Department of the University of Kiel, we can see a dialectal differentiation of Bantawa in Bhojpur into northern und southern dialects on the basis of discriminatory features of a phonological (especially palatalisation) and grammatical nature (especially suffixes vs. prefixes). In view of these features, Chhinamakhu Bantawa is a typical northern dialect, and Wana Bantawa a typical southern dialect, as attested from the two recorded speakers. At the time of the recording, in 1984, the Chhinamakhu speaker was aged 35, born in Chhinamakhu and with his family living there; he himself stayed for longer periods in Kathmandu. The speaker of Wana Bantawa was aged 67 in 1984; he and his wife had migrated to Ham 50 years before, but continued to speak Wana Bantawa among themselves and with their children. These speakers were the only representatives of the given dialects interviewed for the Linguistic Survey of Nepal. 6. As far as person and/or number syncretisms go, there are no differences between non-preterite and preterite, with the exception(s) indicated in brackets.

References Bauman, James J. 1975 Pronouns and pronominal morphology in Tibeto-Burman. [Ph.D. dissertation, University of California at Berkeley] Ann Arbor: University Microfilms. Belie, Aleksandar 1950 Istorija srpskobrvatskogje-^ika II. 1: reci sa deklinacijom. Beograd: Naucna knjiga. Benedict, Paul K. 1972 Sino-Tibetan, a conspectus. Cambridge: Cambridge University Press. Coleman Robert 1991 "The assessment of paradigm stability: Some Indo-European case studies", this volume. Comrie, Bernard 1978 "Morphological classification of cases in the Slavonic languages", The Slavonic and East European Review 56: 177 — 191. DeLancey, Scott 1981 "The category of direction in Tibeto-Burman", linguistics of the Tibeto-Burman Area 6,1: 83-101.

160

Jadranka Gvo^danovic

Durovic, L'ubomir 1983 "The case system in the language of diaspora children", in Lingua in diaspora (Slavlca Lundensia 9), 21-94. 1984 "The inner consistency of a linguistic continuum", in Signs of friendship: To honour A, G. I~. van Hoik, slavist, linguist, semiotician, ed. by Joost J. van Baak, 89 — 95. Amsterdam: Rodopi. Gvozdanovic, Jadranka 1985 Language system and its change. On theory and testability. Berlin: Mouton de Gruyter. Ivic, Milka 1961

"On the structural characteristics of the Serbocroatian case system", International Journal of Slavic Linguistics and Poetics 4: 38 — 47.

Jakobson, Roman 1936 "Beitrag zur allgemeinen Kasuslehre. Gesamtbedeutungen der russischen Kasus", Travaux du Cercle Linguistique de Prague 6: 240-288. [Reprinted 1971, 23-71.] 1958 "Morfologiceskie nabljudenija nad slovjanskim skloneniem", in Materialy i diskusii H, Trudy IV me^dunarodnogo s'je^da slavistov. Moskva. [Reprinted 1971, 154—183.] 1971 Selected writings II: Word and language. The Hague: Mouton. Jakobson, Roman — R. Colin Cherry — Morris Halle 1953 "Toward the logical description of languages in their phonemic aspect", language 29: 34—46. Reprinted in Roman Jakobson, Selected writings I: Phonological studies, 449-463. The Hague: Mouton, 1962. Michailovsky, Boyd 1975 "Notes on the Kiranti verb (East Nepal)", Linguistics of the Tibeto-Burman Area 2,2: 183-218. Plank, Frans 1985 1986 1991

"Die Ordnung der Personen", Folia Linguistica 19: 111 — 176. "Paradigm size, morphological typology, and universal economy", Folia Linguistica 20: 29-48. "Rasmus Rask's dilemma", this volume.

Silverstein, Michael 1976 "Hierarchy of features and ergativity", in Grammatical categories in Australian languages, ed. by Robert M. W. Dixon, 112—171. Canberra: Australian Institute of Aboriginal Studies.

Rasmus Rask's dilemma Frans Plank

1. Arranging paradigms: The hows and the whys The earliest extant grammatical texts are paradigms. Listing the inflectional forms of Sumerian on clay tablets, they date from around 1600 B.C., the period when Sumerian, the chief literary language of Ancient Mesopotamia, was being replaced by Akkadian as the medium of everyday communication, and those who continued to learn it needed grammatical instruction. By later standards (say, those of Pänini, Dionysius Thrax, Priscian, or his Old English adaptor, /Elfric), these Sumerian paradigms appear somewhat disorderly. Setting less store by uniformity than their latter-day colleagues, the Old Babylonian grammarians, otherwise skilled morphological analysts, would alter the arrangement of inflectional forms from one word to the next and from one part of a paradigm to the other. Admittedly, when listing Indicative forms of verbs (and there are quite a few of these among the 227 forms of gar 'to place', the most extensive paradigm to have survived, and among the less numerous, otherwise independently arranged forms of other verbs), they would always begin with 3rd Person, followed by 1st, with 2nd coming last. Also, in both nominal and verbal paradigms they would usually give Singular before Plural forms, although minimal pairs are not always adjacent. To exemplify the more pervasive pattern of variability, in the fragmentary lists of nominals carrying Case suffixes Locatives sometimes precede and sometimes follow Datives (with the Terminative, however, invariably following both), and Genitive and Dative likewise appear in alternative orders. Another trace of uniformity here is that basic Absolutive forms generally head (sub-) sections, which continue with Comitative and Ablative-Instrumental, which in turn are followed by the Dative, Locative, Terminative group. 1 It does not behove us to disparage the efforts of our Old Babylonian predecessors at setting out paradigms; they after all did not have the benefit of three millenia and a half of experience in work of this sort. It is more appropriate to wonder why indeed they should have felt it desirable to settle on a particular invariant order for the terms of all inflectional categories. After all, what their paradigms were designed for was the communication of knowledge about Sumerian inflectional morphology to their Akkadian-speak-

162

Frans Plank

ing pupils. And did it really matter, then, in what manner, however haphazard, the inflectional forms that needed to be learned were arranged, as long as the listings were complete and each Sumerian entry was properly glossed in Akkadian? Steeped in one or another post-Babylonian grammatical tradition, we have come to expect grammarians to stick to a single order of terms realising inflectional categories throughout all relevant paradigms or parts of them. Arbitrary inversions of the order of Cases in the various declensions or in the Singular and Plural parts of individual paradigms would no doubt jar on the aesthetic sense of the contemporary reader of grammars of Sanskrit, Latin, or Old English. In this respect (and perhaps others), grammars have become a genre where variafio no longer delectat. Uniformity, however, has not yet been enforced on the professional community as a whole. Notwithstanding the preferences that have evolved within particular traditions or schools subscribing to the principle of uniform term arrangement, from the ancient Indians to us postmoderns, individual grammarians have continued to disagree, to some extent, on the most appropriate invariant order of the paradigms of virtually all major inflectional categories. Of course, reference grammars rarely justify explicitly the paradigm arrangements they have opted for; such choices, nevertheless, are not entirely arbitrary. Their motives are practical or theoretical (rather than aesthetic), and it is their possible heterogeneity which is commonly held responsible for the continual disagreement about the best orders. A potent theoretical, or perhaps rather metatheoretical, motive is that if everything in the realm of language — as well as in the animal, vegetable, and mineral kingdoms — is reducible to some natural order (except perhaps the exceptions), it would be odd if in inflectional paradigms, the backbone of grammar, order had to be admitted to be arbitrary. But decisions about the most natural order of paradigms have rarely been unanimous, owing to some extent to the multiplicity of criteria in principle available to the imaginative systematist. It is possible, for instance, to motivate paradigmatic order by syntagmatic order. Thus, Case paradigms have been arranged so as to reflect the normal linear order in which Case-marked constituents occur in a clause (with the Nominative, the subject Case, accordingly coming first, followed by the Accusative, the direct-object Case, etc.), or also some cognitive order underlying the 'logical' surface sequence of constituents (with Cases marking the point of departure, the resting-place, and the goal appearing in that order). Another possibility is to conceive of paradigms as a kind of grammatical thesaurus and to group inflectional forms in accordance with the principle that the closer to one another they are in meaning, the smaller

Rasmus Rask's dilemma

163

the distance between them in the paradigm. In paradigms of words inflecting for Case and Number, forms sharing a Number but differing in Case, accordingly tend to be closer to one another than forms sharing a Case but differing in Number. As in lexical thesauri, judgments of relative semantic similarity, however, invite controversy. Moreover, they may lead to arrangements which are at odds with other semantic criteria, like the syntagmatic ones alluded to above.2 Unlike semantically inspired decisions on uniform paradigm orders, those drawing on formal properties of the items to be ordered are essentially practical, and partly maybe also aesthetic. It seems more natural, for instance, to present less complex forms prior to more complex ones, especially if the latter can be regularly derived from the former, or forms which have implications for other paradigm members prior to these implied or to unpredictive forms. Probably the single most important practical consideration behind the conventions of paradigm designers since those of ancient India is a formal analogue to the thesaurus principle: entries should be the closer to one another in arrangement the closer they are in form. Homonymous entries, representing the extreme of similarity, accordingly ought to be adjacent to one another. It is this motive which will concern us primarily, trifling though it may appear at first sight. The Old Babylonian grammarians were not the only ones to pay little heed to what could seem to be mere vagaries of form. In a Sumerian grammar of post-Babylonian provenance, Thomsen (1984: 88), this guideline is also disregarded when the Case suffixes are presented in this order: (1)

Genitive Absolutive Ergative Dative Locative Comitative Terminative Ablative-Instrumental Locative-Terminative Equative

-ak -0 -e -ra -a -da -se -ία -e

Of the suffixes which share the vocalic segment /a/, for instance, only three (Dat, Loc, Com) are adjacent, and the two which in addition have an initial alveolar consonant in common (Com, Abl-Ins) are separated by a segmentally quite dissimilar suffix (Ter). Most blatant is the large distance between the two Cases whose exponents in fact are identical (-e\ viz. Ergative and

164

Irans Plank

Locative-Terminative.3 This raises the question why it should be more practical in the first place to have the similarity between forms reflected by the distance between them in paradigms. The basic purpose of paradigms in reference grammars is to provide information about the inflectional categories of inflecting words (such as Case), about the terms realising these categories (such as Genitive, Absolutive, Ergative), and about the exponents (affixes, segmental or suprasegmental modifications, root exchanges, zero) expressing these terms and, if exponents cumulatively express terms of more than one inflectional category, term combinations with the relevant inflecting words. Disregarding possible drawbacks of formal similarities between exponents of different terms or term combinations (such as the confusion they might cause on the part of the hearer), identities of exponents tend to have one advantage: they reduce the number of grammatical forms that need to be memorised by the speaker, and in particular the learner, of the language. Thus, due to the homonymy 4 of Ergative and Locative-Terminative, the learner of Sumerian only needs to acquire nine, rather than ten, distinct exponents of Case. The problem that remains, however, is to know precisely which distinctions are neutralised in which paradigms. Unless there is some general strategy of delimiting the range of possible victims of homonymy, little is gained by formal parsimony as such; those who are above all faced with this problem, language learners, might just as well acquire distinct exponents. It is here that the judicious arrangement of paradigms proves pedagogically useful. The convention that those paradigm members are placed next to one another which may be expressed by the same exponents authorises an appropriate generalisation. It rules out non-neighbouring members as possible homonyms, unless they are linked by members also sharing the same exponents. With Case terms ordered as in (1), the Genitive could on this condition be expected to be possibly homonymous with the Absolutive, the Absolutive with the Ergative, the Purgative with the Dative, etc. — but not the Genitive with the Ergative (unless the Absolutive, too, were to share the same exponent), nor the Ergative with the Locative-Terminative (unless the same exponent were to be shared by the Dative, Locative, Comitative, Terminative, and AblativeInstrumental as well), etc. The arrangement of (1) is, thus, impractical insofar as it does not alert the learner of Sumerian to the particular homonymy, that of Ergative and Locative-Terminative, which the language, economising on its formal resources, resorts to. It has so far been taken for granted that paradigms may only be presented in the form of a list, as favoured by the Old Babylonian grammarians and others in their wake. But there obviously are alternatives, and, most impor-

Rasmus Rask's dilemma

165

tantly for us, it depends on the kinds of order one is prepared to endorse which and how many members of a paradigm may be considered neighbours. If a paradigm is presented in the form of a list all « elements of which are linearly ordered, there will be n — \ pairs, n — 2 triples (etc.) of neighbours, with two elements, the first and the last on the list, having only one neighbour, and n — 2 elements having two neighbours each, as is shown schematically in (2): (2)

linear order w I

pairs of neighbours: triples of neighbours:

w —x, x —y, y —z w —x —y, x — y — z

i If the ordering is permitted not to be asymmetric, so that one clement may precede as well as follow another, neighbourhood relations multiply. If the number of elements exceeds two, there are then at least as many pairs and (with more than three elements) triples of neighbours as there are elements directly linked to one another in a circle (all of which have two neighbours), and in the extreme case any element ends up a neighbour of any other, as shown in (3):5 (3)

circular order a. w·«— z l t x —^y

pairs of neighbours: w-x, x-y, y-z, z-w triples of neighbours: w-x-y, x-y-z, y-z-w, z-w-x

b. w-«— z I \^ T

pairs of neighbours: w-x, x-y, y-z, z-w, w-y triples of neighbours: w-x-y, x-y-z, y-z-w, z-w-x

x —^y

c. w^—z I \( t

pairs of neighbours: triples of neighbours:

w-x, x-y, y-z, z-w, w-y, z-w w-x-y, x-y-z, y-z-w, z-w-x

x—^y

It the ordering is permitted not to be connex, so that not all n elements are linearly ordered relative to all others, there will be no more than n — 1 pairs of neighbours (as with linear and unlike circular orders), but particular

166

Frans Plank

elements may have more than two neighbours (as in circular but not in linear orders); and, as also shown in (4), there will also be more triples (etc.) of neighbours than in a linear order. (4)

partial linear order pairs of neighbours: w-x, w-y, w-z triples of neighbours: w-x-y, w-y-z, w-x-z Υ

^2

Apart from being favoured by extraneous considerations like that of saving space on a page or clay tablet, the ancient method of arranging paradigms in the form of a list, with all entries linearly ordered, is, thus, also the most restrictive one as regards the neighbourly intercourse among those entries. We have so far dwelt on the practical advantages of arranging paradigms in accordance with patterns of formal differences and identities. We have argued that the ordinary working grammarian, if he wishes to oblige his readers, ought to place those paradigm members next to one another the distinction between which is neutralisable; we have shown that the simple method of arranging paradigms in the form of lists is the one that most severely limits the placement of members next to one another; and we have mentioned the requirement that the arrangement of terms should not vary from one paradigm, or part of it, to the other. These matters, however, are not exclusively practical or aesthetic. On the contrary, it is of considerable theoretical interest to determine whether it will in fact always be possible to comply with these three requirements of good paradigm design — viz. that inflectional homonyms be adjacent, that terms be arranged linearly, and that term orders be uniform. If it is possible, this would imply that inflectional homonymy in the languages of the world is subject to quantitative constraints, because the well-meaning grammarian would be unable to succeed if a language confronted him with such a diverse pattern of homonymies that it cannot be defined in terms of neighbourhood in uniformly ordered lists. Our practical advice, thus, amounts to an empirical hypothesis about permissible patterns of inflectional homonymy — one that cries out to be tested, considering that other attempts to constrain homonymy in semantic terms have only met with limited success and, in particular, have left its unequal distribution among languages of different morphological type unexplained. 6 Obviously, the exclusive force of the neighbourhood condition always depends on a particular sequencing of terms. If, in Sumerian, Ergative and Genitive, or any other pair of terms non-adjacent in (1), were actually found to be expressed by the same exponent, all we would have to do to authorise

Rasmus Rusk's dilemma

167

their homonymy (and to facilitate the learner's task) would be to alter the order of terms accordingly, without having to resort to ordering relations less restrictive than linear ones. Thus, any homonymy pattern can in fact be accommodated in this manner, since it is always possible to place the terms which happen to be affected next to one another in a list. Of what kind, then, are the restrictions to be derived from modes of paradigm arrangement? The Old Babylonian Grammatical Texts VI, VII, VIII, IX, and X contain five more or less detailed verbal paradigms, all differing to some extent in the linear order in which the respective verb forms are given. This could seem to suggest that it is really necessary, in order to master the Sumerian inflectional system, to internalise lists of forms of more than one lexical item of a particular word class. Modern grammars of Sumerian, such as Thomsen (1984), indeed divide verbs into four classes, depending on the way they form their maru aspectual stem; but otherwise the inflectional resources are, on the morphological level, uniform for all verbs and are not distributed among several conjugation classes.7 For nouns, a single paradigm is nowadays deemed sufficient, since there are no different declension classes calling for different sets of markers of Possession, Number, and Case. There in fact is a classification of nouns into those denoting persons and the rest; but this is a general semantic distinction, and its morphological manifestations (only personal nouns take the Plural suffix -ene and may appear in the Dative; the Locative, Ablative-Instrumental, and Locative-Terminative are limited to impersonal nouns, which also lack a Plural, using the basic form also as a Collective), thus, are predictable for the class of nouns as a whole. Some Case exponents may also vary depending on their phonological environment (after vowels, the Dative, for instance, may take the form -r instead of -ar); but this variation too is predictable by general (morpho-)phonological rules, hence does not necessitate the drawing up of separate paradigms — say for vowel- und consonant-final personal as well as impersonal nouns such as, respectively, lu 'man', lugal 'king', e 'place', ig 'door'. Furthermore, there is no need to repeat the list of Case markers, and analogously the markers of Number (in fact only the Plural marker, Singular being unmarked) and Possession (varying as to Person, Number, and, in the 3rd Person, Personness), with the terms of the other inflectional categories: they are the same, no matter which terms of the other categories they are combined with. If they had made a point of limiting their lists to what is unpredictable about Sumerian inflection, the Old Babylonian grammarians, therefore, could well have made do with a relatively small supply of not excessively large tablets. Agglutinative languages, such as Sumerian, where the terms of different inflectional categories (such as Possession, Number, and Case) are typically

168

I-'rans Plank

expressed by separate, segmentable exponents, tend to employ exponents which do not vary with the terms and lexical items they are combined with — or if they do vary, the shape they take is predictable on general semantic or phonological grounds. The overall number of exponents of all words inflecting for particular categories ideally does not exceed the sum total of the terms of these categories. Thus, instead of giving separate subparadigms for different term combinations (such as, for instance, one subparadigrn of Case for the Singular and another for the Plural) and separate paradigms for separate lexical items or classes of them, it ideally suffices to list the terms plus their exponents of each category only once, supplemented by an instruction of how to join them syntagmatically (specifying, for instance, that nominal stems must be followed, in this order, by Possession, Number, and Case suffixes). As long as the number of terms per category, and of categories themselves, is not exorbitant, the fund of distinct exponent necessary to make all paradigmatic distinctions should be of manageable proportions. Economising on exponents by means of utilising them for more than one purpose, i. e. by neutralising paradigmatic distinctions, should hardly be a top priority of the users of such systems. And as a matter of fact, agglutinative languages demonstrably are not the most fertile soil for inflectional homonymies.8 In flective languages, on the other hand, exponents often are not morphologically invariant. Cumulative exponence implies that terms are expressed differently depending on the terms of the categories which are cumulated. What is expressed by individual exponents, thus, are term combinations rather than separate terms. To list all exponents and properly identify them, the terms of each cumulated category have to appear repeatedly in full paradigms. These can then be divided into subparadigms for each category, all but one partner of term combinations remaining constant in each subparadigrn. For example, with six-term Case and two-term Number categories expressed cumulatively, there would be six subparadigms of Number, one for each Case, and two subparadigms of Case, one for each Number. Although this is not a logical implication of cumulative exponence, flective languages, furthermore, typically boast synonymous exponents not distributed along phonological or transparently semantic lines. Different classes of lexical items inflecting for the same categories, thus, differ, partly or wholly, in the set of exponents they utilise. Since the number of exponents of one inflected word here results from multiplying, rather than adding, the number of terms of the categories cumulated, and the co-existence of different inflection classes further increases this number, homonymies come in handy, serving to pare down the fund of distinct exponents when it threatens to become undesirably large. Suppose there are five declensions in our hypothetical flective language

Rasmus Rusk's dilemma

169

with six Cases and two Numbers, the number of exponents could then be as large as sixty if no paradigmatic distinctions were neutralised.'·1 The extent to which flective languages characteristically cut down on formal resources is illustrated by the adjective inflection of Old English, where five Cases, two Numbers, three Genders, and two 'Declensions' (i. e. Definiteness and Indefiniteness) are cumulated, but there are only about ten, rather than sixty, distinct exponents (of which virtually none is confined to adjectives). It is under flective-type circumstances that the neighbourhood condition, in conjunction with ordering regulations, becomes effective. Here paradigms multiply. If the terms of inflectional categories are expressed by alternative sets of exponents, whose distribution needs to be lexically stipulated, we have to set up parallel paradigms for these categories; and in each of them all terms recur depending on the number of terms of the categories they are cumulated with. Term distinctions are susceptible to homonymy in all these separate paradigms and subparadigms. In principle it should be possible for the patterns of homonymy to differ without limit across the various paradigms and subparadigms, even though the problems of the learner, welcoming formal parsimony as such, would thus be aggravated. Of course, the different homonymy patterns could all be accounted for individually in accordance with the neighbourhood condition by appropriately arranging the relevant paradigms and subparadigms. But to meet this requirement, the order of terms could well have to be different on different occasions. There is, thus, the potential of a conflict between the neighbourhood condition and that achievement of post-Babylonian grammarians, the invariance of term order,10 and it remains to be seen whether such conflicts ever materialise in empirical reality. In modern times it was Rasmus Kristian Rask who devoted most attention to the question of term order. No less persistent than the systematisers of nature of the Enlightenment, where Rask had his intellectual roots, he was almost haunted by a desire to restore the inflectional paradigms, especially of the Old Germanic tongues and of Latin and Greek, to their natural order. His efforts did not receive the acclaim they would have merited, largely because they were seen (among others by Jacob Grimm) as being wasted on peripheral matters of at best practical interest. But Rask's quest really was for insight into the structure of language, and the criteria he invoked to order paradigms struck him as no less essential than those of the naturalists pertaining to reproduction. Least convincing perhaps is his appeal to the natural order of ideas as a guideline for the arrangement of paradigms. More tangible are his two main formal criteria: derivability, with 'derived' forms slotted into place after their bases (which often suggests the placement of

170

Frans Plank

Accusatives after Nominatives rather than after Datives or Genitives; cf. the Latin 3rd Person Feminine Singular demonstrative earn, best 'derived' from Nominative ea, rather than from eius Gen or ei Dat), and homonymy, with terms sharing exponents earmarked for neighbourhood. The Case paradigms of Indo-European languages in particular were a constant source of worry for Rask and at times drove him to despair. Originally accepting the old Accusative-final order (Rask 1811), he was soon persuaded that it was more natural for the Accusative to follow immediately after the Nominative (and Vocative). He was less certain about the relative order of Dative and Genitive, although Dative-before-Genitive eventually was to carry the day. A matter he apparently never resolved for good was whether Dative or Ablative should be given precedence, opting for Ablative-first (as such he categorised the Instrumental) in the English version of his Anglo-Saxon grammar (1830), but favouring the reverse order, likewise with good reasons, in some published works and in letters on Latin. Rask has been criticised for his dogmatic insistence on the necessary parallelism of semantic and formal criteria of paradigm order.11 It is of course an empirical issue to determine how close the fit is between independently inspired arrangements. But another, more elementary question is whether Rask's programme could have succeeded if it had been based on the formal criterion of homonymy alone. The position Rask took on this point was the most restrictive one conceivable, as outlined above: he subscribed to the neighbourhood condition on inflectional homonymy, assumed that paradigms were to be ordered linearly, and expected term orders to be invariable throughout, and indeed across, languages. Was he thus heading for a resounding defeat, bound to be inflicted upon him by an utter intrinsic disorderliness of inflectional homonymy? An examination of the Case system of Old English and, in less detail, of various other languages should help us to identify precisely where a position like Rask's is unrealistically restrictive.12

2. Old English Case With five Cases,13 of which one, the Instrumental, is presumably best recognised only with adjectives and non-personal pronouns, the diversity of patterns of homonymy is potentially great. To be precise, the sum total of possible two-, three-, four-, and five-term homonymies is twenty-six:

Rasmus Rusk's dilemma

(5) 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

Nom = Acc Nom = Gen Nom = Dat Nom = Ins Ace = Gen Acc = Dat Ace = Ins Gen = Dat Gen = Ins Dat = Ins

11. 12. 13. 14. 15. 16. 17. 18. 19. 20.

Nom = Ace = Gen Nom = Ace = Dat Nom = Ace = Ins Nom = Gen = Dat Nom = Gen = Ins Nom = Dat = Ins Ace = Gen = Dat Ace = Gen = Ins Ace = Dat = Ins Gen = Dat = Ins

21. 22. 23. 24. 25.

171

Nom = Ace = Gen = Dat Nom = Ace = Gen = Ins Nom = Gen = Dat = Ins Nom = Ace = Dat = Ins Ace = Gen = Dat = Ins

26. Nom = Ace = Gen = Dat = Ins

Any linear arrangement of five terms licenses no more than ten such patterns, as shown abstractly in (6). (6)

list of terms: a b c d e

permitted homonymies: a = b, b = c, c = d, d = e; a = b = c, b = c = d, c = d = e; a = b = c = d, b = c = d = e;

With five terms, the alternative linear arrangements number no less than 120, which for our purposes may be reduced by half since neighbourhood relations are the same in corresponding inverted sequences. There are, thus, sixty different sets of homonymy patterns which could in principle be accommodated by choosing an appropriate linear term order. Since there are numerous classes of nominal, pronominal, and adjectival lexical items differing in the set of Case exponents they require, and also numerous subparadigms of Case, owing to the various categories Case is cumulated with in different classes of words (two- or three-term Number in nouns, adjectives, and pronouns; three-term Gender in adjectives and pronouns; two-term Definiteness in adjectives), the potential of a very diverse composite picture of Case homonymies is enormous, too. Here is a summary of the patterns of Case homonymies and their instances as actually attested in Old English, keyed to the enumeration in (5). 1. Nom = Ace: Singular of all masculine and neuter ajata-nouns and phonologically identifiable subclasses ofy), Indefinite adjectives (Masc -e, Fem -ej-a, Neut -0/-a}, Definite adjectives (-0»), 3rd Person personal pronoun (kte\ki\heo}, demonstrative pronouns (pa, pas}. 2. Nom — Gen: In the Plural, ötfö/wö-nouns have normally -a in Nominative, Accusative, and Genitive in West-Saxon, but in early West-Saxon Accusative has occasionally -e, distinguishing this Case from Nominative and Genitive (late West-Saxon also has Gen -ena). 6. Acc = Dat: Singular, Dual, and Plural of 1st and 2nd Person personal pronouns (me, urn, üs; pe, ine, }; outside West-Saxon, however, there tends to occur a distinctive Accusative (mec, unket, üsic; pec, incit, eowic}, but these forms may later be generalised also to Dative function. 8. Gen = Dat: Singular of feminine /-nouns (-e, which may, however, also extend to Ace), w-nouns (-a}, neuter weak nouns (-an}, feminine athematic nouns (-ej-0, but there may be a difference in root vowels), dental stems (-0 and final -p}, Definite Neuter adjectives (-an; Definite adjectives also lack a distinct Instrumental), Feminine 3rd Person personal pronoun (hire}. 10. Dat = Ins: Plural of all adjectival and pronominal forms having a separate Instrumental in the Singular.

11. Norn = Ace = Gen: Plural of o//o/a'