Historical Development of Auxiliaries 9783110856910, 9783110109900

Table of contents :
Section (a): General Problems
Introductory Paper
From Auxiliary to Desinence
Copula Auxiliarization: How and Why?
Auxiliaries and Ergative Splits: A Typological Parameter
Section (b): Germanic Languages
The Auxiliarization of the English Modals: A Functional Grammar View
On the Use of the Modal Auxiliaries Indicating ‘Possibility’ in Early American English
Semantic Change in Modal Auxiliaries as a Result of Speech Act Embedding
Syntactic and Semantic Change Within the Modal Systems of English and Afrikaans (Discussion Paper)
Section (c): Romance Languages
The Strategy and Chronology of the Development of Future and Perfect Tense Auxiliaries in Latin
Syntactic Restructuring in the Evolution of Romance Auxiliaries
The Interaction of Periphrasis and Inflection: Some Romance Examples
The Evolution of Romance Auxiliaries: Criteria and Chronology (Discussion Paper)
Section (d): ‘Exotic’ Languages
Auxiliaries in Khotanese
The Development of the AUX-category in Pidgins and Creoles: The Case of the Resultative-perfective and its Relation to Anteriority
AUX in Basque
Auxiliaries in ‘Exotic’ Languages (Discussion Paper)
Subject Index
Index of Names
Index of Languages

Citation preview

Historical Development of Auxiliaries

Trends in Linguistics Studies and Monographs 35

Editor

Werner Winter

Mouton de Gruyter Berlin · New York · Amsterdam

Historical Development of Auxiliaries edited by

Martin Harris and Paolo Ramat

Mouton de Gruyter Berlin · New York · Amsterdam 1987

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

Library of Congress Cataloging in Publication Data

"Library of Congress Catalog Current Number 87-17312"

CIP-Kurztitelaufnahme der Deutschen Bibliothek

Historical development of auxiliaries / ed, by Martin Harris and Paolo Ramat. — Berlin ; New York ; Amsterdam : Mouton de Gruyter, 1987. (Trends in linguistics : studies and monographs ; 35) ISBN 3-11-010990-5 NE: Harris, Martin [Hrsg.]; Trends in linguistics / studies and monographs

Printed on acid free paper.

© Copyright 1987 by Walter de Gruyter & Co., Berlin. All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced in any form — by photoprint, microfilm, or any other means — nor transmitted nor translated into a machine language without written permission from Mouton de Gruyter, a Division of Walter de Gruyter & Co., Berlin. Typesetting and Printing: Arthur Collignon GmbH, Berlin. — Binding: Dieter Mikolai, Berlin. — Printed in Germany.

Foreword

As part of the 7th International Conference on Historical Linguistics held in Pavia, September 1985, a workshop was organized to discuss the much debated questions of the definition and possible membership of the category AUXILIARY, considered especially from a diachronic point of view, in accordance with the general historical approach of the Conference. The present volume was conceived in order to publish the proceedings of that workshop, which ranged from contributions drawing their evidence from well-documented linguistic traditions, such as Germanic and Romance, to papers dealing with languages like Warlpiri, lacking any accessible diachronic depth, or pidgins and Creoles whose diachronic dimension can be but an indirect one. Unfortunately, it has proved impossible to publish exactly the same set of papers as that presented at the full-day workshop; neither have we been able to reproduce the stimulating interventions from many competent specialists who took part in the lively discussion. Many of their observations have however been incorporated in the final versions of the papers and the editors thankfully acknowledge all the interesting suggestions raised in the informal and friendly atmosphere of the discussion that followed every paper. It is useful to reproduce hereafter the original program of the workshop. Some references to missing papers will thus become understandable to the reader of the Proceedings; these references have at times been maintained — not of course in a detailed form — in order to give an idea of the many subjects (and problems!) dealt with by the contributors to the workshop: (a) General Problems: Discussant Paolo Ramat (Pavia) 9.00— 9.30 Paolo Ramat: Introductory paper 9.30— 9.45 Henning Andersen (Kobenhavn): From Auxiliary to Desinence 9.45 — 10.00 Simon Dik (Amsterdam): Copula and auxiliary: synchronic and diachronic aspects 10.00 — 10.15 Eloise Jelinek (Arizona): Auxiliaries and ergative splits: a typological parameter 10.30 — 11.15 General discussion

VI

Foreword

(b) 'Exotic' Languages: Discussant Annarita Puglielli (Roma) 11.30 — 11.45 Ronald Emmerick (Hamburg): Auxiliaries in Khotanese 11.45 — 12.00 Thomas Stolz (Bochum): The development of the AUX-category in pidgins and Creoles: the case of the resultative-perfective aspect and its relation to anteriority in Creoles 12.00 — 12.15 Annarita Puglielli: Discussion paper 12.15 — 12.45 General discussion (c) Germanic Languages: Discussant Martin Harris (Salford) 15.30 — 15.45 Louis Goossens (Antwerpen): The auxiliarization of the English modals: a Functional Grammar view 15.45 — 16.00 Merja Kytö (Helsinki): On the use of modal auxiliaries indicating possibility in Early American English 16.00 — 16.20 Martin Harris: Discussion paper 16.20 — 16.45 General discussion (d) Romance Languages: Discussant John Green (Bradford) 17.15 — 17.30 Harm Pinkster (Amsterdam): The use of motion verbs as auxiliaries in Latin 17.30 — 17.45 Giampaolo Salvi (Budapest): The Romance auxiliaries: a case of syntactic reconstruction 17.45 - 1 8 . 0 0 Ed Tuttle (UCLA): The spread of ESSE as a universal auxiliary in ItaloRomance 18.00 — 18.15 Nigel Vincent (Cambridge): 'Venire' and 'andare' as auxiliaries in Italian 18.30 — 18.45 John Green: Discussion paper 18.45 — 19.15 General discussion

Two new papers have been added to the original program: C. J. Conradie's paper, transferred from the main Conference, and that by Esmeralda Manandise, who was unable to attend the workshop as originally planned. Despite the regrettable absences, we believe that the present collection of papers represents a valuable contribution to the ongoing discussion on a topic that in recent years has proved to be of great interest within many different theoretical approaches. The diachronic approach of this volume may help to elucidate some of the central issues in the debate about the problematic category of AUXILIARY, and what may reasonably count as an exponent thereof. Finally we would like here to thank all those who helped to make the workshop a success, above all of course the contributors and other participants but also our hard-working secretarial colleagues Marco Mazzoleni, Caterina Pagani, Claire Robinson and Stella Walker. , , TT ,^ ^ Μ. H. and P. R.

Table of contents

Section (a): General Problems

1

Introductory Paper Paolo Ramat

3

From Auxiliary to Desinence Henning Andersen

21

Copula Auxiliarization: How and Why? Simon C. Dik

53

Auxiliaries and Ergative Splits: A Typological Parameter . . . . Eloise Jelinek

85

Section (b): Germanic Languages

109

The Auxiliarization of the English Modals: A Functional Grammar View Louis Goossens

Ill

On the Use of the Modal Auxiliaries Indicating 'Possibility' in Early American English 145 Merja Kytö Semantic Change in Modal Auxiliaries as a Result of Speech Act Embedding 171 C. J. Conradie Syntactic and Semantic Change Within the Modal Systems of English and Afrikaans (Discussion Paper) 181 Martin Harris Section (c): Romance Languages

191

The Strategy and Chronology of the Development of Future and Perfect Tense Auxiliaries in Latin 193 Harm Pinkster Syntactic Restructuring in the Evolution of Romance Auxiliaries 225 Giampaolo Salvi

VIII

Table of contents

The Interaction of Periphrasis and Inflection: Some Romance Examples 237 Nigel Vincent The Evolution of Romance Auxiliaries: Criteria and Chronology (Discussion Paper) 257 John N. Green Section (d): 'Exotic' Languages

269

Auxiliaries in Khotanese Ronald Emmerick

271

The Development of the AUX-category in Pidgins and Creoles: The Case of the Resultative-perfective and its Relation to Anteriority 291 Thomas Stolz AUX in Basque Esmeralda Manandise

317

Auxiliaries in 'Exotic' Languages (Discussion Paper) Annarita Puglielli

345

Subject Index Index of Names Index of Languages

355 363 367

Section (a):

General Problems

Introductory Paper Paolo Ramat

In this discussion paper it is my intention to sketch some of the general problems and issues raised by the reading of the contributions offered to the workshop. I will examine first some proposals for a definition of AUX(iliary) as a category, drawing from the evidence adduced in the papers. Then I shall examine whether all the adduced examples can really be considered as being auxiliaries, according to the criteria usually employed in different approaches to the question of the definition of AUX, and conclude with a negative answer. The category AUX seems to be a gradient notion, in effect 'fuzzy' though with prototypical instances. This is of course a situation that, in the historical dimension, applies to many aspects of language. Finally, from the historical point of view, I will tentatively sketch four stages in the auxiliarization process — that is in the grammaticalization process of lexical items — relying on four empirical properties to arrive at an operative definition of AUX. The first impression one gets from reading the whole set of papers presented for this workshop — and not only those appearing under the heading 'General Problems' — is that the definition of what is meant by 'Auxiliary' is by no means uniform and consistent in spite of several recent attempts to give a comprehensive definition (see also the discussion by John Green in this volume: 257 ff.). Take for instance the notional definition proposed by Akmajian et al. (1979:2) according to which 'AUX is a category — i. e. distinct in its syntactic behavior from the behavior of other syntactic categories — labeling a constituent that includes elements expressing the notional categories of Tense and/or Modality'. In addition to this it must be noticed that the category AUX is said to contain a specified, fixed and small number of elements, usually occurring in a fixed order within the AUX constituent. These elements may mark subject, object, subject and object agreement, question, evidential,

4

Paolo Ramat

emphasis, aspect, and negation (see Manandise, this volume: 319 f., drawing on Steele 1981:21 - 2 2 ) . Does such a definition really include all the verbs which are usually considered as instances of AUX and only those? If a parameter for the auxiliarity of a verb is its peculiar syntactic behavior, how have we to consider the serial constructions of the type quoted by Simon Dik (his example (61)) (1)

'he finish he eat' or 'he eat he finish' for 'he has eaten'

where 'verbs sharing a common core argument are merely juxtaposed with no complementizers or intervening constructions' (Foley —Van Valin 1984:186). This is the case of Mandarin Chinese (2)

Tä lä-käi le men 3 sg. pull-open PERF door 'he pulled the door open'

or of Yoruba (3)

ό mit iwi wä 3 sg. took book went 'he brought the book' (Foley-Van Valin 1984:189),

where we find two full verbs, none of which shows a peculiar syntactic behavior. I am by no means a specialist for Australian languages, but it seems to me that Warlpiri, too, may be said to show a similar feature since the verb root ka functioning as auxiliary may present different forms according to different tenses. Thus alongside (4)

( = Jelinek 5 d.) Ngarrka-ngku ka wawirri man-ERG AUX kangaroo panti-rni spear-NON PAST 'The man is spearing the kangaroo'

with ka = PRES 3 sg NOM 3 sg ACC, we find also (5)

Wawirri-Φ kapi-rna-0 kangaroo-ABS FUT-1 sgNOM-3 sgACC panti-rni yalumpu-Φ spear-NON PAST that-ABS Ί will spear that kangaroo'

Introductory

Paper

5

with kapi-. (On the 0-form of the arguments see Jelinek, this volume: 90.) We have here clitic auxiliaries bound to a fixed position and carrying no individual semantic content but only a syntactic load — as usual in Australian prefixing languages with a fairly transparent agglutinative structure (Capell apud Dixon (ed.) 1976:618) and therefore it seems that these examples can be compared to the Yoruba ones only to a certain extent: however we see that AUX-forms give morphological information on a par with the 'main verb'; in other words there are two inflected verbal forms — which is not of course the case in English or French. The same remarks as for Mandarin Chinese and Yoruba seem to hold true also for Khotanese: (6)

kädägane yäde evil deeds 3 sgPERF TRANS MASC Tyä 3 sgOPT ( < Root ah- 'to be') 'he may have done evil deeds' (see Ronald Emmerick, this volume: 272).

or Basque (examples from Dik, Nr. 48, 49; for further discussion of the Basque case see the final part of this paper): (7)

Liburu hori ni-k irakurri-a book that(-ABS) I-ERG read(PERF)-DEF sg d-a 3 sgABS-izan ( = be) 'That book is such that I have read it'

(8)

Ni-k liburu hori irakurri I-ERG book that(-ABS) read(PERF) d-u-t 3 sgABS-ukan ( = have)-l sgERG

and this certainly holds true for many other languages, where the socalled 'AUX' shows no particular syntactic behavior. If we follow the traditions of Indo-European linguistics and insert among the definitional properties of AUX that in a verbal complex the AUX must be the only inflected form, all the cases discussed so far would necessarily be excluded from the set of examples of AUX. From the semantic and functional point of view this definition entails

6

Paolo Ramat

that the combination of the verb considered to be an AUX and the form of the other verbal lexeme must have just the same argument structure as the simple forms of that verbal lexeme (see Pinkster on habere, this volume: 201). This means, in other words, that the finite verb form ( = the AUX) imposes no restrictions on the non-finite forms. In her stimulating contribution to Auxiliaries and Related Puzzles (Frank and Richards (eds.) 1983:21—46) Eloise Jelinek has argued for the existence of an AUX invariant node and AUX as a universal category, rejecting Pullum's assumption that auxiliaries would represent but a special subset of verbs generated in a way similar to that of the other verbs. If AUX is a universal category it must be represented by a Phrase Structure Node in all grammars and the large crosslinguistic variations we may actually observe must be explained by the (semantic) features of the words inserted under the AUXnode. This is an up-to-date reformulation of the fundamental rule of the classic Transformational Grammar: (9)

S

NP (AUX) VP

where AUX -»• Τ Μ A (Tense, Mode, Aspect). Within Chomsky's Government Binding Theory (10)

S -> NP INFL VP

so that in this respect AUX and INFL would practically be identical. There is, in fact, a widespread agreement that AUX should be considered a category dominating Tense, Mode, and Aspect. However, Eric Reuland has argued in his contribution to Frank — Richards (1983) for the superiority of the INFL category and the non-universality of the AUX category, since the first one exhibits all the fundamental properties of the 'head' S which are relevant for GB theory. Consequently AUX would be just a special case of INFL, and moreover it would be incapable of accounting for non-temporalized clauses as Dutch bare infinitival constructions: (11)

dat Annamaria 1983:139)

Patrick

de ratten zag vangen (Reuland

'that A. saw P. catch the rats' or English (12)

I understand John behaving foolishly (Reuland 1983:115).

Introductory Paper

7

I have dwelt upon the discussion within the Generative Theory just in order to show that also among generativists there is by far no agreement as to what AUX actually means. Actually I have argued elsewhere that the Universale of Generative Transformational Grammar, are often not 'Universale of Language' but 'Universale of the Linguist', i.e. methodological devices of the analysis the linguist applies to all languages (cf. Ramat 1984: 56 — 57). But leaving now aside this larger epistemological and methodological problem, let us recall from the foregoing discussion that AUX must have the property of being capable of expressing Tense, Mode, Aspect. The problem is whether AUX must be the only representative of Tense, Mode, Aspect. If the answer is 'yes', then examples (2) — (8) should be excluded from the realm of auxiliaries. This is, I think, the reason why Simon Dik (this volume: 71) doubts whether serial constructions consisting of two fully inflected verbs have to be considered on a par with the other cases of auxiliarization he has examined (namely the 'Localist channel' and the 'Property channel'). According to A. Capell, Tense, Mode, and Aspect are represented, in Warlpiri and other Australian languages, discontinuously in the clause, not by elements in an auxiliary word, but by suffixes in the verb stem, and free aspectual and modal auxiliary roots (Capell apud Dixon (ed.) 1976:623). Now, some 'auxiliary' roots (with quotation marks added by Dixon 1980:426) are still reducible to semantically autonomous full verbs: 'Consideration of verbs that take -wa- "to fall" as AUX is suggestive of metaphorical connections' (Dixon 1980:127 — comparing Engl, fall in love). If this is true, we are not far from the state of affairs exemplified by the outcomes of Port, acabar 'to finish' in Creoles and pidgins, so nicely studied by Thomas Stolz: (13)

(Ilha do Principe) Ε kabä falä ( = Stolz, example (5)) PRO PERF V 'He finished talking'

or, still more evidently: (14)

(Sri Lanka) E:li ja:-fdla e:w ja:-ka:- fdla: PRO PAST-V PRO PAST-PERF-V fdla:-tu QUOT-PERF 'He said he (had) told (you)' (Stolz, example (8))

8

Paolo Ramat

with -ka: (< acabar) as a tense/aspect modifier of the main verb (Stolz, this volume: 296). The large palette of examples from Portuguese-based Creoles enables Stolz to sketch a stratification which may be considered representative of the diachronic evolution of acabar from full verb with its own lexical meaning as perhaps in (15)

(Papia Kristang) Yo ja kabä PRO PAST PERF kumi ( = Stolz, example (7)) V Ί have finished eating'

to a real auxiliary as in (13), and then finally a temporal conjunction: (16)

(Cabo Verde) El cende cander, el PRO V Ν PRO senta pel d'cara, cabä el bä abri V Ν PERF PRO V V 'She lit a candle, caressed her face and went then to open the door' ( = Stolz, example (6)) thus joining other temporal conjunctions derived not from verbs but from nouns, as lo 'then' ( < logo), used also as future marker. As Schuchardt rightly pointed out, 'das Verbum sinkt vielfach zu der Rolle eines Adverbiums, einer Präposition, einer Konjunktion herab' (quoted after Stolz, this volume: 311). In other words, we are faced with the weakening of restrictions, a development which is typical of the diachronic evolution towards the category AUX, so accurately illustrated by Harm Pinkster referring to the emergence of temporal auxiliaries in Latin (see Pinkster, this volume: 210 ff.). From the examples presented in several papers (and especially in those of Andersen and Stolz) we may tentatively sketch the process of auxiliarization — i. e. of grammaticalization of lexical items — as follows:

I. Full verbs acabar, habere, shall, werden (-ννα- ?), etc. have their full semantic meaning as in

Introductory

(17)

Paper

9

multa bona bene parta habemus (Plautus, Trin. 347) QUANT Ν ADV ADJ V "we have many properties (which are) well acquired"

whose constituent analysis is (17) a.

Vp[ N P [ Q U A N T [ m u l t a ] N [bona]

ADj[bene

parta]] v [habemus]]

(see Ramat 1984:145). This construction may still be found as late as the 5th century: cf. Cassianus, Inst. 4,7 [A. D. 426]: qui habet curam peregrinorum deputatam '(a monk) who has received [from others! The Subj. of habere is not the same as the Subj. of deputärel] the task of taking care of foreign visitors"; see Pinkster, this volume, example (14). In these cases the non-finite verbal form, whose subject may also be different from that of the finite verb (see Salvi, this volume: 228), may also be absent and the sentence will not lack its meaning: multa bona habemus and qui habet curam peregrinorum are meaningful sentences. This is no longer the case when the non-finite verbal form becomes the necessary complementation of the finite verb (Phase II).

II. Predicative construction Yo ja kabä kumi (example (15)) without kumi would mean simply Ί have finished'. Compare also (18)

Nam hominem servom / suos domitos habere oportet oculos et manus (Plautus, Mil. 563 — 564) In fact a man servant / his tamed to have needs eyes and hands. Ά servant must indeed keep his eyes and hands submissive'.

Without domitos (18) would have no meaning at all (see Ramat 1984:144). Pinkster is right to note that in most cases as (19)

or

( = Pinkster, example (8)) [Flamines] caput cinctum habebant filo (Varro, L. L. 5,84) 'the flamines had their hair girt with a woollen filum'

10

(20)

Paolo

Ramat

( = Pinkster, example (15a)) nam et capillos nostros ipse utique creavit et numeratos habet (Augustinus, Serm. 62.10,15) 'for he has certainly created our hairs and has them counted'

the predicative past participle cannot really be omitted: ?'the flamines had their head', ?'he has our hairs'. It is perhaps precisely in such cases of inalienable possession that periphrastic verbal forms made the first step towards auxiliarization of finite verbs. Note that the process may be repeated with new lexical material. Salvi quotes the Italian example (21)

Tengo gli occhi aperti Ί keep my eyes open'

with tenere like Port, ter < Lat. tenere, already used as an AUX in cases as persuasum, traditum tenere (Pinkster, this volume: 214 f.): ? tengo gli occhi Ί keep my eyes' would sound very strange! This applies also to the new category of 'semi-auxiliaries' quoted by John Green (this volume) as Ital. venire, andare which show also modal nuances (e. g. la cosa andava fatta 'the thing had to be done'); on modality see below.

III. Periphrastic forms Examples as the well-known Late Latin (22)

episcopum invitatum habes (Gregory of Tours) 'you have invited the bishop',

discussed by Pinkster, belong here, as well as the already quoted Ε kabä falä (example (13)) "he finished talking". We are dealing here with real (new) periphrastic perfect forms. But the same applies also to the forerunners of the Romance futures and conditionals: possidere habet, dare habes > pussideravit, daras where the finite verb (to be considered in terms of a Categorial Grammar as the 'Operand': Ramat 1984:155 — 156) is really the marker for Tense, Mode, and Aspect, with no autonomous semantic meaning. We arrive finally to stage IV.

Introductory

Paper

11

IV. Agglutination This stage may be exemplified by the already quoted Late Latin (23)

Ρ VSSIDERA VIT, dar as

and, of course, by the Romance future and conditional. Or also by ja:-ka:-fdla: '(had) told' of the Sri Lanka Creole (example (14)) — and, I would add, by the Warlpiri examples (4) and (5). See also the 'not-easy-to-analyse' fusional forms of the Basque examples (27), (28) as the final stage of an agglutination process. The evolution sketched here is strongly reminiscent of the scale of grammaticalization suggested by Louis Goossens for the process of desemanticization of the English modals: (24)

full predicates > predicate formation > predicate operators (Goossens, this volume: 118).

Now, it is certainly noteworthy that in English, too, the instances of AUX, when reduced to tense or mood markers, may have an extremely reduced phonetic form: I'll see; I'd like; I've been, etc. The AUX is here reduced to a simple morphological sign, a prefix agglutinated to the main verb ('erosion' of the form). As was rightly observed by Ernst Pulgram in the workshop discussion, ΊΙ, 've and also gonna (in substandard forms as Ilyoujhe gonna do this) are on the way towards becoming prefixed inflectional morphemes, whether or not they are spelled out as separate lexemes. The same evolution can be observed in ModGk. θά πώ 'I'll say' < θέλει ϊνα εΐπω, lit. 'will (3sg!) so that I speak', where the cliticized form no longer distinguishes verbal person. (Pulgram quoted also French dialectal forms of the type jejtujil va chanter.) The same evolution can be observed — as is well-known — in the Slavic languages that cliticize the verb 'to be', in this case postposed to the main verb: Henning Andersen in his paper presents evidence for the enclitic agglutination of the verb 'to be' to the finite verb forms in Old Polish, where, however, the eventual outcome was person and number markers, unlike the extreme trend noted in the above English and French examples. Finally, a further parallel may be underlined in this drift towards cliticization, namely that beside the reduced clitic forms many languages do continue to have full orthotonic forms of the same verb when they possess an autonomous meaning: Old Pol.jesm

12

Paolo

Ramat

Ί am' vs. -(e)sm/-(e)m; Rum. noi avem 'we have' vs. noi am vezut 'we've seen': the final stage of this reduction or phonetic erosion is precisely Ital. -emo in vedr-emo 'we'll see' vs. (dial.) avemo or French -ons in nous verr-ons vs. nous avons (see Ramat 1984:159). On the one hand, Frans Plank, critically reviewing the 'Modals story' as told by David Lightfoot, affirms that the rise of the English Modals 'must be seen in the larger context of an overall diachronic tendency towards the reduction of inflectionally expressed mood oppositions [e. g. indicative ~ subjunctive], or at least of the inflectional apparatus to express such oppositions' (Plank 1984:345, for further critical discussions of Lightfoot's interpretation of the evolution of modals in English see Conradie and Harris, this volume 113 ff. and 182 f.). The analytic, periphrastic forms produced by using Modals or pre-Modals follow the general principle if 'iconicity' or transparency (Gabelentz 1901: 256 called it 'Deutlichkeitstrieb'). On the other hand, cliticization of AUX — the last phase of the desemanticization and erosion process that auxiliaries undergo in their grammaticalization (see Dik, this volume, referring to Meillet) — represents the final result of the second general principle operating in language (and language change), the principle of least effort (Gabelentz called it 'Bequemlichkeitstrieb') moving towards synthetic, symbolic and no longer analyzable forms. Stages I — IV represent focal instances of the process between the two poles — a process which is continuously going on without breaks (and Goossens, too, has placed some modal items like can or shall on the intersection lines between the three domains of his scale). The first steps of this gradual drift occur first in unmarked environments where the original meaning of the verb on the way to becoming AUX may easily become redundant (e.g., habere in the case of inalienable possession: recall IFlamines caput habebant). Things being so, no wonder that there exist different opinions among linguists as to what must actually be considered an AUX: where should we draw the line between full verbs and AUX? Following the line proposed in Talmy Givon's article of 1971 it has recently been suggested that also the serial constructions ( = type (1)) should be considered as the first, most primitive stage of the AUX development (Moreno, forthcoming). From the evidence gathered from the papers of our workshop it is not possible to support this hypothesis — and in fact, as was stated at the beginning, there are substantial differences between serial constructions and what is

Introductory

Paper

13

traditionally considered an auxiliary verb, although such an evolution cannot theoretically be excluded (see below). We thus come back to the problem of a consistent definition of AUX, the problem with which we began. After having quoted the definition proposed by Akmajian, Steele, and Wasow, let us also consider a recent analysis within the framework of Montague Grammar: Greg Carlson ranges the AUX-verbs among the functional morphemes that do not contribute to the meaning of the sentence but are syncategorematic to the lexical morphemes to which they apply (in this case the main verbs), preserving their category (Carlson 1983). Evidently this applies well to the traditional set of auxiliaries in the (Indo-)European languages. Many years ago, discussing the status of the semi-Modals like Engl, need or Germ, brauchen, I tried (Ramat 1971 and 1972) to outline the set of Germanic Modals as a morphosemantic field starting from a morphosyntactic definition of Modal Verb, intuitively adequate to the set of mögen, dürfen, müssen, wollen, sollen and related forms in the other Germanic languages. I tentatively defined a (Germanic) Modal as a verb that a) selects another non-finite verbal form (its own infinitive included: Ich will wollen Ί want to want' vs. *Ich gehe gehen '*I go to go' (cf. Manandise's example (11 a. und b.) where Basque izan and ukan can be used in combination with the same auxiliary forms!); b) is not able to select a NP of its own, and c) shows a particular paradigm, different from the 'normal' one has also a particular syntactic pattern, different from that of other verbs (e. g., he need not know how many books are there not *He needs not to know or *he does not need; need he come? not *does he need to come?).

and the and and

I concluded that formal criteria alone are inadequate to delimit the set of Modals: Er lernte lernen 'He learned to learn', though having a weak paradigm behaves as Ich will wollen, but for semantic reasons evidently cannot be considered to be a Modal. On the 'fuzziness' of the modal verbs see also Conradie (this volume: 171). A kindred characterization of the present-day English Modals is now proposed by L. Goossens who includes also a semantic criterion beside the formal and syntactic ones: 'Semantically we find that the

14

Paolo

Ramat

items that share these syntactic and formal properties cover a broad range of modal meaning (going from facultative to deontic and epistemic modality) plus a number of other meanings' (this volume: 112 f.). This situation is confirmed also by Conradie from a diachronic point of view in his study of the meaning shifts of the Afrikaans Modals, passing from Knowledge to the expression of bare Futurity ( = AUX!). (On the semantic changes Germanic Modals underwent in their evolution see the careful analysis of the uses of can/could by M. Kytö in this volume and compare also, more generally, Ramat 1972, where it is shown that the changes are not random but on the contrary may be encompassed in a relatively limited set of possibilities, since shifts of meaning imply the existence of semantic categories common to different meanings.) Every definition is of course conventional and arbitrary, and can be called neither true nor false, but simply adequate or non-adequate. But I think that Goossens is right in considering in the formulation of his definition the three levels of linguistic structure at the same time. There is otherwise the risk of widening too much the notion of Modals (or AUX): if we do not consider the morphological aspect as relevant for the definition (property (c) above) it is impossible to exclude Germ, brauchen already alluded to: (25)

Er brauchte nicht kommen (popular)

has precisely the same syntactic frame as (25) a. Er wollte / sollte / konnte (etc.) nicht kommen. Or why not include to take, tomar, prendere, vzjat' among the auxiliaries, since we have the well-known type studied by Coseriu (1966) (26) a. b. c. d.

she took and died tomo y muriose prese (su) e mori voz'mi da i pomri (etc.; cf. Coseriu 1966: 39)

We find here a paratactic and analytic construction building a morphological unity where 'to take' has no lexical meaning at all but gives, on the contrary, expression to the unitarian character of the action (Coseriu speaks of 'Entschlossenheit, schneller Vollzug, Überraschung, Erregung, Einheit und Unteilbarkeit der Handlung', 1966:48), i. e., of notions that are typical of the aspectual dimension).

Introductory

Paper

15

Noting a certain similarity with the serial constructions, we could surely imagine that verbs like 'take' or 'up' (in she up(ped) and married him), 'go' (in ερχουμι κι κάσονμι, 'je viens et je m'assieds', Ί come and sit down' could eventually develop into instances of AUX, but this would again concern the historical dimension of the evolution of the category, not its definition. (In the Sinto dialect of the Gypsy language in the USSR, the periphrastic future is built with the verb lav 'to take': lav te gäva Ί shall go' lit. '(I) take that (I) go', les te gäsa 'You shall go', etc. On the other hand, in standard Romanis of the Danubian area (Vlax), we find a prefix kam- (< kamän Ί will'): kam-keräv 'I'll do', kam-keres 'You'll do', etc. (see Soravia 1977:36.) To sum up: — A historical view of the category AUX helps to elucidate the gradual coming-into-being of auxiliaries and their evolution but cannot lead to a categorial definition. It also makes understandable why there are so different views as to what may be considered an example of AUX. The decision, however, depends on the theory the linguist has chosen, not on the historical development of particular examples. Finally, the historical perspective, with its 'pre-Modals' and 'semiModals' is in keeping with the notion of prototype which is now becoming popular in many definitional procedures (cf., e.g., Comrie 1981:100—104; Lehmann 1984:37). Grammatical categories have a range of characteristics, some of which will be more central than others to the definition of the category under scrutiny. Consequently, there will be focal and marginal instances of the category. In a very interesting article, Karin Aijmer (1985) has outlined the evolution of OE willan to ModE will. Willan shows the prototypical properties of a volition verb: a) active desire; b) human subject; c) the main verb denotes an activity. Almost the only function of ModE will is to mark future. This drift points to a progressive moving of the verb out from its prototypical place and progressive weakening of the original meaning. But also the reverse movement is possible and will may in present-day English express modality, too (probability, inference, likelihood, etc.). We have thus a to and fro movement between full verbs (e. g. OE willan), auxiliaries marking Tense, Mode, Aspect (e. g., I will write a letter tomorrow), and Modals (e. g., This will be the book you were looking for, I suppose). Each one of these categories will be represented by prototypical, focal instances.

16

Paolo Ramat

From the literature I have quoted and from the papers presented for our workshop I think we may retain four criteria for an operationally useful definition of what may be labelled as prototypical instance of the AUX-category: a) An AUX must have the same Subject as the uninfected dependent verb. In other words, although AUX in terms of the Categorial Grammar is to be considered as 'Operand' (main verb), it no longer selects its own ARG(ument) but — as rightly stated by Nigel Vincent in his brilliant study of habere and esse (1982) — fills that position in its structure taking over the ARG of the non-finite verbal form. b) More generally, Vfin and Vinf must have the same PRED(icate)frame. c) No semantic restrictions are imposed by the AUX ( = Vfin) on the Vinf (Pinkster, this volume). In his very accurate contribution to the workshop Edward Tuttle proceeded, in fact, to show that habere and esse may be blended and confused in many Italian dialects: e.g. Ntonio e [vs. standard Ital. ha] rotta la brocca "Antonio has broken the pitcher", and, conversely, the rather widespread type ha stato [vs. standard Ital. e stato] '(he/she/it has been'; see Tuttle 1986). (Interestingly enough we find such a blend in other linguistic traditions too: in ModFris. one can say both Ik ha siik west Ί have been sick' and Ik bin siik west, lit. Ί am sick been', without any significant difference (cf. Popkema 1985). d) The function of AUX is to express syncategorematically tense, mood or aspect relations (so that Modals may fulfill the role of AUX) — though, of course, Tense, Mode, and Aspect may also be represented by other means. If we accept these criteria (or others of the same kind), making use of concepts of the general theory of language like PRED or Subj., but not deriving the notion of AUX from the theory itself (as is the case in Transformational Grammar), the question whether AUX must be seen as a language universal will receive an empirical answer by checking whether the language under scrutiny will show, or not show, the features established as definitional criteria. A further general point must finally be emphasized. Relativizing the notion of AUX and restricting it to the languages where the chosen prototypical definition does really apply is by no means equivalent to denying the real existence of a category 'AUX', both in the general theory of language and in those languages. In other words, as was rightly remarked in the workshop discussion, AUX is

Introductory

Paper

17

not an arbitrary construction of the linguist; it may also not be a language universal, but for the languages where the definitional criteria apply it is doubtlessly real in the sense that it belongs to the morphosyntactic structure of these languages. (And the definition proposed by the linguist has to take care of the language reality if it seeks to be operative!) Take for instance the case of Basque, extensively discussed by E. Manandise. We may also agree with Manandise's view that, from a synchronic standpoint, ukan 'to have' and izan 'to be' are not full verbs with auxiliary function but simply markers for tense, person, mood, etc. in (27)

Nik etxe bat dut I: ERG house one: ABS it: non past: 2 arguments (ABS/ERG): indicative Ί have one/a house' ( = Manandise, example (38))

(28)

Ni gizona naiz I: ABS man: the/a: Sg: ABS I: non past: 1 argument (ABS): indicative Ί am a/the man' ( = Manandise, example (37))

In many languages possessive and existential statements do not have in fact any form corresponding to 'to have' and 'to be'. But in this case we will simply have to conclude that the category AUX does not function in Basque sentences such as (27) and (28) and that the diachronic evolution of the two verbs has reached its final stage reducing the verbs to phonetically not recognizable formatives, much in the sense previously alluded to (under IV. Agglutination). The category AUX is no longer found in (27) and (28) and only a diachronic approach helps us in understanding where dut and naiz come from. A theory-oriented categorial definition of AUX will on the other hand not permit considering dut and naiz as auxiliaries any more. The historical perspective will thus give a sensible answer to the much debated question whether the development of AUX must be thought of in terms of Thorn's catastrophic approach or as a result of a steady, gradual evolution (see also Harris, this volume). The linguistic change leading to auxiliaries is gradual but the recategorization of the items according to the definitional criteria chosen by the linguist will follow a yes/no-strategy, i. e. a catastrophic point of view.

18

Paolo Ramat

References 1985. The semantic development of will, in: Historical semantics: Historical word-formation, ed. by Jacek Fisiak, 11—21. Berlin: Mouton.

AIJMER, KARIN.

AKMAJIAN, ADRIAN; SUSAN M . STEELE;

and

THOMAS WASOW. 1 9 7 9 .

The category

in Universal Grammar. Lin 1 0 . 1 — 6 4 . A. apud DIXON 1 9 7 6 . CARSON, G R E G N . 1 9 8 3 . Marking constituents, in: Frank & Richards

'AUX'

CAPELL,

1983.

Vol.

I:

69-98. COMRIE,

BERNARD.

1981.

Language universale and linguistic typology. Oxford:

Blackwell. 1966. Tomo y me voy. Ein Problem vergleichender europäischer Syntax. VR 25/1.13-55. DIXON, R. M. W. (ed.). 1976. Grammatical categories in Australian languages. New Jersey: Humanities Press. COSERIU, EUGENIO.

The languages of Australia. Cambridge: University Press. 1 OLEY, W. Α.; R. D. VAN VALIN. 1983. Functional syntax and Universal Grammar. Cambridge: University Press. FRANK, H . , AND B. R. RICHARDS (eds.). 1 9 8 3 . Linguistic categories: auxiliaries and related puzzles. 2 vols. Dordrecht: Reidel.

DIXON, R . M . W . 1 9 8 0 .

1901. Die Sprachwissenschaft. Ihre Aufgaben, Methoden und bisherigen Ergebnisse. (2. Aufl. 1972. Tübingen: Narr).

GABELENTZ, GEORG VON DER.

1971. Historical syntax and synchronic morphology. Papers from the 7th Regional Meeting. Chicago Linguistic Society: 394—415.

GIVÖN, TALMY.

1983. Person-Subject marking in AUX in Egyptian Arabic, in: Frank & Richards 1983. Vol. I: 21 - 4 6 . LEHMANN, CHRISTIAN. 1 9 8 4 . Der Relativsatz. Tübingen: Narr.

JELINEK, ELOISE.

Forthcoming. Diacronia y tipologia: hacia una superaciön del punto de vista sincronico. PLANK, FRANS. 1 9 8 4 . The modal story retold. Studies in Language 8 . 3 0 5 — 3 6 4 . MORENO, JUAN CARLOS.

Hy hat/is heechlearaar west, in: Miscellanea Frisica (Festschrift Miedema), ed. by Ν . R . ÄRHAMMER et al., 1 9 3 — 2 0 3 . Assen: Van Corcum.

POPKEMA, J . 1 9 8 4 . Η . Τ. J .

RAMAT, PAOLO. 1971.

Die Analyse eines morphosemantischen Feldes: die germanischen

Modalverben. IF 7 6 . 1 7 4 - 2 0 2 . 1972. Analisi etimologica del campo morfosemantico dei verbi modali germanici. Studi Germanici 10.43 — 79. RAMAT, PAOLO. 1 9 8 4 . Linguistica tipologica. Bologna: II Mulino.

RAMAT, PAOLO.

Government and the search for AUX.s: a case study in crosslinguistic category identification, in: Frank & Richards 1983. Vol. I: 99 — 168. SORAVIA, GIULIO. 1977. I dialetti degli Zingari. Pisa: Pacini. REULAND, E . J .

1983.

M. et al. 1981. An encyclopedia of AUX: a study of cross-linguistic equivalence. Cambridge, Mass.: MIT Press. TUTTLE, EDWARD F . 1 9 8 6 . The spread of ESSE as universal auxiliary in Central ItaloRomance. Medioevo Romanzo 1 1 . 2 2 9 — 2 8 7 . STEELE, SUSAN

Introductory Paper

19

1982. The development of the auxiliaries HABERE and ESSE in Romance, in: Studies in the Romance verb, ed. by Nigel Vincent & Martin Harris, 71—96. London: Croom Helm.

VINCENT, NIGEL.

Paolo Ramat Facoltä di Lettere Universita 1-27100 Pavia

From Auxiliary to Desinence Henning Andersen

0. Introduction It is an old insight that today's morphology is yesterday's syntax. The idea that affixes originate as full words which are grammaticized, become phonologically reduced, and are then agglutinated as bound elements to original free forms goes back, in its rudiments at least, to the 1500's (Robins 1967:101, 157). It was a commonplace at the beginning of the 1800's, when it came to play a central role in the reconstruction of Indo-European morphology (cf. Robins 1967:173). The complementary idea that grammatical means are renewed through the creation of novel syntactic constructions is neatly adumbrated by Whitney (1867:283). Over the years a great number of such developments from word to affix have been noted in a wide range of languages, involving the development of derivational affixes from compounding, case markers from adpositions, nominal possession and definite markers from pronouns, subject and object markers on verbs from pronouns, tense and mood markers from auxiliary verbs, and so on (cf. Tauli 1958:82-112, Serebrennikov 1973). But the details of such developments are still relatively unknown. It is characteristic that neither the major works of synthesis in linguistics (Whitney 1867, Gabelentz 1891, Bloomfield 1935, Serebrennikov 1975), nor the more or less specialized textbooks (e.g., Anttila 1972, Bynon 1977, Boretzky 1977) attempt more than the most superficial account of what has become known as the linguistic cycle (Hodge 1970), if they even mention it. It is characteristic, too, that the speculative 'theory of agglutination', current since the 1800's (cf. Gabelentz 1891:250 ff.), according to which the position of affixes directly reflects the word order of an earlier language state has had to wait until recently to be put into proper perspective (cf. Comrie 1980).

22

Henning

Andersen

In this paper I will review one of the best documented developments of an auxiliary verb from word to desinence, the history of the Modern Polish inflection of the preterite indicative, of the subjunctive, and of the present indicative of the verb 'to be'. The data to which I will refer are well-known to Polonists. They are briefly mentioned in all historical grammars of Polish and have been the subject of a fair amount of scholarly literature, including two monographs (Decaux 1955, Rittel 1975). My aim in reviewing this development is twofold. I want to draw attention to this development as a topic for future research. Although it is richly documented, some of the known details of this development have not been satisfactorily interpreted. I will mention a few of these below and offer my own thoughts on them. But apart from this, the data which are available in Polish texts from the 1400's to the present can yield a great amount of new information if they are approached with the right questions. These will arise as our theoretical understanding develops and they should be addressed to the Polish data in the future. The second aim with this presentation is to show how the observed gradualness of this development can be reconciled with a conception of language structure that recognizes categorical distinctions. Faced with data documenting a gradual development of an autonomous word to a fixed desinence, some linguists might be happy to take this as evidence that there is not a clear-cut distinction between free forms and bound forms, but rather an infinitely graded cline with 'free' and 'bound' as extreme values. I view the current preoccupation with gradience as a very natural manifestation of the 'post-binary bliss' the field of linguistics is experiencing in connection with the shift of interest from language structures to discourse. But in the longer run the problem remains of reconciling the categorical distinctions (oppositions) encoded by grammars with their gradient application in discourse and — in diachronic terms — bridging the gap between categorical shifts in the chronological transmission of a grammar and their gradual attestation in texts. To tackle this problem one needs a suitably articulate conception of language structure, as I will suggest in sec. 2 below.

From Auxiliary

to Desinence

23

1.0. The historical development In this chapter I will present a summary of the development of the set of desinences that in Modern Polish characterizes the preterite indicative and the subjunctive of all verbs, and the present tense indicative of the verb 'to be' (hereafter: BE). The presentation will concentrate on the later stages in this development, which are documented in the historical period. But the account must start with some facts about the Common Slavic tense system, in which BE functioned as an auxiliary (sec. 1.0.1), and must describe the variant present tense forms of BE in the pre-Polish version of this system (sec. 1.0.2). We can then sketch the radical restructuring of the pre-Polish tense system on the eve of the historical period (sec. 1.0.3), which is the point of departure for the centurieslong process through which original forms of BE have metamorphosed into desinences of the preterite (sec. 1.1), the conditional (sec. 1.2) and the present tense of BE (sec. 1.3.1). A number of issues of general relevance to this development are most conveniently discussed along the way, such as morphological univerbation (sec. 1.1.3), prosodic univerbation (sec. 1.1.4), and segmental univerbation (sec. 1.3.2). A number of other general questions will be touched on in sec. 2. 1.0.1. The pre-Polish tense system In the Common Slavic period, constructions of BE with the so-called resultative participle in -/- (showing subject agreement in gender and number), formed from any verb, are idiomaticized and then grammaticized as compound retrospective tenses (e. g. 'has spoken', 'had spoken') opposed to the simple, non-retrospective tenses (e. g., 'speaks', 'spoke') as in (1), which lists the respective 3rd person singular, masculine forms of the Old Church Slavonic verb 'speak'. Tense systems of this structure are amply attested in Old Church Slavonic (Vaillant 1974:81 ff.) and Old Russian (van Schooneveld 1959), and a similar system can confidently be posited for prehistoric Polish. Besides the grammaticized compound tenses listed in (1) there are also constructions of BE -I- passive participle (past or present), probably grammaticized as expressions for the passive voice, and more or less occasional constructions of BE + active participle (past or present). These will not be discussed here at all.

24

(1)

Henning

Andersen

retrospective perfect mlüvilü jestü plu-I mlüvilü be perf. II mlüvilü bease

non-retrospective present mlüvitü aorist mlüvi imperfect mlüvljaase

There also appears in Old Russian and in Old Polish, perhaps only in connection with the development sketched in sec. 1.0.3, a 'double perfect' (as van Schooneveld calls it), a sort of'passe surcompose', composed of /-participle plus perfect of BE (e. g. ORuss. mülvilü bylü jesti), which will be relevant below. Since the development of this tense system is entirely prehistorical, I will not discuss its particulars (see Vaillant 1974: 81 ff.).

1.0.2. Pre-Polish BE In prehistoric Polish, the present tense of BE develops, beside the original orthotonic forms, which are maintained for rhematic and emphatic use, a set of phonologically reduced, atonic forms for unmarked use; cf. (2) (the parenthesized e occurs after forms ending in a consonant). While the orthotonic forms can occur anywhere in a clause, including initial position, and are serialized in accordance with the information structure of the utterance, the atonic forms become subject to Wackernagel's rule, that is, they are treated on a par with the other enclitics of the language and placed — in grammatically fixed sequential relation to these — in the position following the first orthotonic word of the clause (cf. Decaux 1955:16 ff.). Old Polish 1st sg. 2nd sg. 3rd sg. 1st pi. 2nd pi. 3rd pi. 1st du. 2nd du. 3rd du.

orthotonic jesm jes jest/ jesc/je jesm(y) jesce sg jeswa jesta jesta

enclitic -(e)smj-(e)m -(e)s -0

Modern Polish ~(e)m ~(e)s -0

-(e)smy -(e) see 4 -(e) swa -(e)sta -(e)sta/-0

-(e) smy -(e)sce -0

From Auxiliary

to Desinence

25

The outcome of the development of enclitic BE-forms is well attested in the earliest Polish texts we have, some of which are written in a clearly archaic form of the language — they are copies of much older texts, which have been lost. It is interesting to note that some of these texts show influence from the Latin texts from which their originals were translated; for instance, the Old Polish 3rd person perfects regularly have a zero auxiliary (cf. (2) and the examples in (4)), but where the perfects render Latin deponent verbs or passives in these texts, the full auxiliary (jest or sg) is consistently used (cf. Decaux 1955:127). But despite their somewhat artificial character, these texts show clearly enough that both the phonological reductions of the present tense forms and their syntactic change to sentence enclitics were completed before the date of the oldest preserved texts. They show, too, that zero 3rd person forms of BE are usual from the 1300's on. As orthotonic forms, jest, jesta, and sg continue to be used — but with diminishing frequency, and only for emphatic predication — through the 1500's (Decaux 1955:116,127-128). Little by little their function passes to other means of expression, as we shall see in sec. 1.1.1.

1.0.3. The old Polish tense system Still in prehistoric Polish, the system of preterite tenses is sharply reduced. The compound tense forms take over the functions of the imperfect and aorist. As these fall into disuse, also the subsystem of compound tenses (in which both imperfect and aorist of BE figured as auxiliary) is reduced. Only two compound tenses survive (cf. (3)), the earlier perfect (henceforth referred to as the preterite), and the earlier double perfect (henceforth, the pluperfect; this tense will not be discussed here apart from a brief mention in sec. 1.0.4). This reduction of the finite preterite system is completed shortly before the period of continuous attestation. The most archaic Old Polish texts — which, as mentioned, are copies of much older ones — have preserved for posterity two dozen imperfect and aorist forms comparable to the ones known from early East and South Slavic (Klemensiewicz et al. 1974: 367 ff.). Note that the future of lexical verbs is composed of the infinitive of the main verb and the inceptive

26

Henning

Andersen

aspect (traditionally called 'future tense') of BE. The other compound tenses are based on the original /-participle. (3)

Old Polish tense system, 3rd sg. masc. forms of the verbs 'speak' and BE present mowi jest/0 preterite mow if (jest) byt (jest) pluperfect mowit byt (jest) byt byt (jest) future bgdzie mowic bgdzie

The recasting of the tense system has important consequences, which have traditionally been overlooked, but which are essential to the entire following development. In the Old Polish tense system, the original present tense forms of BE are no longer opposed to any past tense forms with person marking. Furthermore they occur as parts of compound past tense forms. Herewith the background is given for two reinterpretations. The original present tense forms of BE can be reinterpreted as simple person and number markers. And the original /-participles can be reinterpreted as finite non-present forms, the -/- in particular as the preterite ('distal tense') marker. Both of these reinterpretations are covert and hence difficult to pinpoint in time. It is not certain that they occurred immediately when the conditions for them arose, but it is certain that they did occur. Polish historical grammars have traditionally spoken of the person and number markers as 'auxiliaries' with reference to all periods of the history of the language (thus also Decaux 1955 and Rittel 1975), just as they have called the original /-participles 'participles'. I will refer to the latter as 7-forms' where their status is uncertain, and otherwise simply as preterites.

1.0.4. Excursus on the future and the pluperfect Although the development of the Polish future tense is not relevant to the central topic of this paper — the history of the person and number markers — it may be relevant to the question of when the /-forms were reinterpreted as finite preterite forms. In prehistoric Polish the future is grammaticized as a construction of inceptive aspect of BE plus infinitive, as in (3). The earliest Old Polish texts show this to be the canonical expression for the future. A variant construction of b§dzie + /-form (i. e. b§dzie mowit 'he:will

From Auxiliary

to Desinence

27

speak') occurs, but has the earmarks of a recent innovation. It is attested first in verbs with infinitival complements (e.g. 'be able'; thus bgdg czynic 'they:will make', but b§dg mogli stac 'they:will be:able to:stand' in the St. Florian Psalter, ca. 1400). Later it gradually spreads to other environments. In the language of Mikolaj Rej (1505 — 1569), the /-form future is regular in the singular (b§d§ mied 'I:shall have'), but not yet in the plural (e. g. bgdziemy miec 'wershall have'), except with verbs that take infinitival complements (e.g. dobrze czynic bgdziemy chcieli 'well to:do we:shall want', bgdziecie mieli mowic 'you:will have to.speak') (cf. Kuraszkiewicz 1970:158). The change from infinitival to /-form future was never completed. Distributions similar to that attested in Rej's writings are current also in modern varieties of Polish. A simple interpretation of these facts might be that the /-form future arose as a participial construction parallel to the participial preterite, possibly motivated by a change in the status of the auxiliary from tropic to phrastic function (cf. Lyons 1977:749 ff.). The gradual generalization of this innovation would then very naturally have lost its impetus when the /-form preterite was reinterpreted predominantly as a finite preterite, perhaps in the 1500's. One might wonder whether also the history of the pluperfect can be linked to this change. Note that at the time when Old Polish mowii could be interpreted simply as a finite preterite (accompanied by person and number markers), it was opposed to the pluperfect mowii byt\ this could be interpreted either (a) as a participle plus finite preterite of BE or (b) as a combination of two fmites (a serial verb construction). The (a) alternative would make it an exception to the status of the /-form as finite preterite; the (b) alternative would make this the only serial verb construction in the language. Polish grammarians note the decline of this tense from the 1700's on. It has been defunct in the standard language for a hundred years, but survives in some dialects.

1.1.1. The person and number markers The 3rd person markers. In Old Polish the difference full vs. zero form expresses the distinction emphatic vs. unmarked predication; see (4). This distinction gradually passes to other means of expression (word order, sentence stress), and the full forms go out of use in the

28

Henning Andersen

1400-1600's, first jest 'sg.\ then sg 'pl.', last jesta 'dual' (Decaux 1955:127 ff.), each of them regularly being omitted earlier with the /-forms byi 'was', miai 'had(to)', mögt 'could', chciai 'wanted (to)', kazaf 'said' than with other verbs (Rittel 1975: 55). Henceforth there are no 3rd person markers in any number in the preterite or pluperfect. Number (and gender) continues to be expressed by the desinences of the /-forms. The 1st and 2nd person markers. The full, originally orthotonic forms (cf. (2)) go out of use in the 1500's (Decaux 1955:133). The distinction emphatic vs. unmarked predication is at first taken over by the presence vs. absence of the original 3rd person markers, jest 'emphatic predicator' and sg 'emphatic predicator, pl.'. This is illustrated in (4), which shows a full form in (a); a combination of 'emphatic predicator' jest and enclitic person and number markers in (b), (c); and the more common use of enclitic markers in (d), (e). (4) a.

b.

c.

d.

e.

Old Polish preterite Wiem ze stworzyciela wszego luda porodzita jes I:know that creator of:all mankind bore 2sg Ί know you bore the creator of all mankind' To-ra jest ogl^daia that-lsg emph saw 'That I did see' Jest ja ciebie zepchn^i albo uczynit-em tobie co ztego? emph I thee repulsed or did-lsg thee any harm 'Did I repulse thee or do thee any harm?' Ani-i mi? zepchn^i, ani rzucit, ani-ί niektore ztosci not-2sg me repulsed nor deserted nor-2sg any harm uczynii did 'Thou neither repulsedst, nor desertedst me, nor didst me any harm' Bo-cie-m si? caia darowata for-thee-lsg refl entire gave 'For I gave myself wholly to thee'

The examples incidentally illustrate the common omission of person and number markers in the presence of a nominative personal pronoun (c) and when two or more verbs are conjoined (d) (cf. Decaux 1955: 30 ff.).

From Auxiliary

to Desinence

29

The demise of the full forms is attested earlier in texts that reflect spoken language than in bookish texts, and in these, earlier in secular than in religious texts (Rittel 1975:68 ff.).

1.1.2. The process of agglutination During the period from the 1500's to the present there are gradual changes in the placement of the person and number markers. First of all, deviations from Wackernagel's rule increase in frequency (cf. (5)). The presence of an intonational caesura after a heavy initial thematic element favors the occurrence of the person and number markers after the word or phrase following the caesura. Also, sentence stress on any word in the rhematic part of a clause apparently makes this word attract the person and number marker. It looks as if the enclitics change status from clause enclitics to phrase or word enclitics (see the examples in (5)). Throughout the development, clause-second placement is more frequent in subordinate clauses than in main clauses, more frequent when the clause initial word is a conjunction than any other word, more frequent when it is a pronoun than when it is a noun. Total no. of examples 1500's 1600's 1700's 1800's 1900's exp. prose

580 1303 1439 1988 3325 569

Deviations from Wackernagel's rule 12 2% 64 4% 4% 62 308 15% 15% 503 usual

Agglutination to preterite form 130 23% 649 49% 994 68% 1395 80% 2817 84% 525 92%

Secondly, the person and number markers are with increasing frequency attached to the /-forms, as the figures in (5) show. Unfortunately these statistics from Rittel (1975:91) do not separate the examples of agglutination to clause-initial /-forms (which must account for most of the cases counted in the early centuries, when deviations from Wackernagel's rule are few) from those where the /form occurs further to the right in the clause. This is why the percentages in the 'Agglutination' column present the picture of a

30

Henning

Andersen

much longer gradual development than those in the 'Deviations' column. This is a weakness in these statistics which limits their usefulness; but the figures are at least suggestive. Agglutination becomes common earlier in main clauses than in subordinate ones (Rittel 1975:88), and the singular person and number markers are agglutinated earlier than the plural ones, and in the singular, the 1st person marker earlier than the 2nd person one (Decaux 1955:28, Topolmska 1961:47). Both deviations from Wackernagel's rule and agglutination of the person and number markers to /-forms are century by century reflected less widely in poetry than in prose, and in the modern period less in artistic prose than in expository prose (the last row in (5); Rittel 1975:92). In modern Polish, tmesis (the separation of person and number markers from the /-form) is more common in written than in spoken language, in speech more common in formal than in casual styles, and more usual in the speech of older than in that of younger people. The markers have ceased to be mobile in the northern parts of the language area, but not yet in the south (Topolinska 1961: 55). This difference seems sometimes to be reflected in the attitudes of grammarians to the problems of synchronic description, linguists from the north (e. g., Warsaw) apparently being much more prone to view the person and number markers as bound morphemes, and tmesis as an archaism, than their colleagues from the south (e. g., Cracow).

1.1.3. Morphological univerbation Rittel's use of the term 'agglutination' calls for comment. This term is evidently good enough for the purpose of registering data, but it lumps together a number of different phenomena that must be distinguished even in a summary treatment of this historical development. In the first place her use of this word prejudges the issue of when the person and number markers changed from enclitics to desinences by speaking of agglutination even with reference to the early centuries when the person and number markers are practically never attached to /-forms unless these are in clause-initial position. Secondly, she glosses over the question of the nature of the observed change. Polish grammarians have traditionally been content to remain on the level

From Auxiliary

to Desinence

31

of observation, where it appears that it is the enclitics that change; 'their autonomy is gradually blurred, though they maintain a measure of mobility', as Topolmska puts it (1961:48). But since the outcome of the development is a change in the syntactic properties both of the person and number markers (which become desinences) and of the /-forms (which become stems), we should wonder about the relationship between these different syntactic changes. Note that Rittel's figures in (5) show quite clearly that the change in the syntactic behavior of the person and number markers (deviations from clause-second position) was not actuated in tandem with the agglutination change. As mentioned in sec. 1.1.2, the changes have not been consummated in all varieties of modern Polish; the standard language, for instance, still admits stylistic (pragmatic) variation such as that exemplified in (6). In describing this synchronic situation Polish grammarians have taken two different tacks. One is to describe the person and number markers a having syntactically distinct allomorphs — they are viewed as desinences when they are attached to an /-form, but as enclitics otherwise (thus Topolmska 1961:28). The other is to describe the /-forms as stems and the person and number markers as desinences susceptible to being moved (thus Laskowski in Grzegorczykowa et al. 1984:184; I will argue in favor of this latter approach below). However one decides to describe the synchronic situation, from a diachronic point of view it is clear that the example we have here of a morphological univerbation in progress raises questions about the changes in both of the constituents involved. (6) a. b. c.

Nigdysmy tego nie mysleli Nigdy tegosmy nie mysleli Nigdy tego nie myslelismy

'We never thought that'

1.1.4. Prosodic univerbation Morphological univerbation must be distinguished from phonological univerbation, and here a distinction must be made between segmental univerbation (to which we turn in sec. 1.3.2) and prosodic univerbation, the change by which combinations of /-form and person and number marker become subject to the Polish penultimate stress placement rule.

32

Henning

Andersen

To appreciate the importance of this distinction, consider the effects of the penult rule in the preterite in different varieties of the language; see (7). (7)

Northern dialects mow'item mow'iles m'owit mowil'ismy mowil'iscie mow'ili

Orthoepie standard mow'item mow'ites m'owit mow'ili-smy mow'ili-scie mow'ili

Southern dialects m'owit-em m'owit-es m'owit mow'ili-smy mow'ili-scie mow'ili

It is apparent from these stress paradigms that the penult rule may ignore the person and number markers in the northern dialects, but must pay attention to the plural person and number markers in the standard language, and to both singular and plural person and number markers in the (archaic) southern dialects. Knowing (a) that Polish has developed penultimate stress (from earlier initial stress) some time since the Middle Ages and (b) that the modern preterite desinences have become agglutinated to the /forms during the same period, a simple, mechanistic account of the differences illustrated in (6) would suggest that (a) occurred before (b) in the southern dialects, and (b) before (a) in the northern dialects, whereas in the dialects on which the standard language is based, (a) intersected with (b). This intersection, it will be noted, agrees well with the textual attestation of agglutination earlier and more widely in the singular than in the plural (cf. sec. 1.1.2). This interpretation is perhaps not all wrong, but it needs to be tempered with some additional information. First, the history of Polish rhyme since the 1500's and the testimony of Polish grammarians from the 1600's on show that in the standard language the agglutinated 1st and 2nd singular forms of the preterite have had penultimate stress in this entire period, whereas the corresponding plural forms have had antepenultimate stress (Topolmska 1961:30). But since the 1700's the plural forms have evinced a tendency to fall under the penult rule — characteristically, Polish normative grammars begin insisting on antepenultimate stress around 1800 — a tendency which is now so much in evidence that the orthoepic antepenultimate stress in these forms can be considered outmoded

From Auxiliary

to Desinence

33

(Topolmska 1961:200). Secondly, a completely parallel tendency towards prosodic univerbation can be discerned in dialect data from the southern parts of the language area recorded during the last one hundred years (Topolmska 1961:55). In consequence, there is in fact no telling in which order developments (a) and (b) occurred in the northern dialects. Since the penult rule is perfectly productive, the prosodic univerbation is to be understood as an elimination of exceptions to the penult rule, and as such it could occur any time after the morphological univerbation. It is important in considering the relation between morphological and prosodic univerbation to note that the prosodic univerbation is only relevant to combinations of /-form and person and number marker (Topolinska 1961: 55) cites an insignificant number of deviations from assorted dialects). When the syllabic person and number markers are attached to any other words in a clause, these retain their normal penultimate stress, cf. (8). Thus prosodic univerbation seems to presuppose morphological univerbation. At the same time, the fact that the prosodic univerbation in part took place long before the person and number markers lost their mobility, in fact, prior to the 1500's, when the process of agglutination (even by Rittel's generous standards) was only beginning, shows that there is no way of understanding the interaction of these changes if one confines oneself to the level of observation. As I will suggest below, to understand this entire development one must distinguish between covert innovations and their observable consequences. Specifically, if the morphological univerbation is interpreted as a covert change in the syntactic properties of the morphemes involved, both the observed tendency towards agglutination and the prosodic univerbation turn out to be well motivated secondary effects. (8)

Wcz'oraj-em prz'yszed-t Wcz'oraj przysz'edt-em

Ί arrived yesterday'

1.1.5. The structure of the person and number markers Before we turn to the subjunctive, it should be pointed out that the person and number markers, which so far have been treated as unanalyzed wholes, in fact are synchronically composed of recurrent partials as in (9), which carry grammatical meaning.

34

(9)

Henning

Andersen

1st sg. 2nd sg. 1st pl. 2nd pl.

-m -s -s -s

-m -y -c -e

The final vowels in the plural are identical with regular nominative plural desinences in the nominal system. The -m- of the 1st person markers can be correlated with the 1st person pronouns (singular: mi§ 'me; accusative', mnie 'me; dative', enclitic mi\ plural: my 'we; nominative'. The -c- can be correlated with the oblique stem of the 2nd person singular pronoun (ciebie 'thee, you; accusative', enclitic ci§ 'accusative', ci 'dative'). The content of the initial -s- can be defined as 'subject other than speaker'. The agglutinative structure of the person and number markers is in part historically fortuitous: the final -e and the pre-final -c- did not originally have the content assigned them here; nor did the -s-, which originally was part of the stem of BE. But three innovations in the history of these forms support the analysis given above. The final -y has been attached in pre-Polish times, thus establishing iconic conformity with the 2nd plural and iconic contrast with the singular markers. Secondly, the Old Polish 1st plural marker -smy has changed to -smy (cf. (2)). This innovation, which is attested from the 1500's (Klemensiewicz et al. 1964: 372), makes sense as semantically motivated under the above analysis, and had a real logical alternative, preferred in some dialects, which will be mentioned in sec. 1.2. Thirdly, the morpheme boundary following -s- has in the history of Polish sandhi been treated identically to the boundary that precedes the person and number marker -my '1st pl.' in the imperative (which we will touch on again in sec. 1.3.2). The reason for emphasizing the agglutinative character of the person and number markers will be obvious in the next section.

1.2. The conditional Old Polish inherited from the pre-Polish period a periphrastic conditional composed of /-participle plus the conditional of BE (see (10)), which had a unique paradigm going back to the pre-Polish (perfective) aorist of BE, e. g., mowif bych Ί should speak'. It also developed a

From Auxiliary

to Desinence

35

past conditional, e. g. mowii byi bych Ί would have spoken', which we will not consider here (see Klemensiewicz et al. 1964:379). Already before the attested period the forms of the conditional auxiliary had become clause enclitics, serialized in relation to other enclitics in clause-second position. (In certain functions they did remain orthotonic into Old Polish; cf. Decaux 1955:122.) Originally composed of stem (by-) and desinences, the forms of this paradigm were reanalyzed as enclitic -by followed by equally enclitic person and number markers. In the 1400 — 1500's, these enter into variation (stylistic, regional) with the person and number markers of the preterite, and this variation eventually yields a single paradigm of person and number markers for both preterite and conditional (as in (10)), with regional differences (some dialects unifying the two sets as -(e)ch, (e)s, -(e)chmy, -(e)-scie; cf. Decaux 1955:130ff.; 145ff.).

1st sg. 2nd sg. 3rd sg. 1st pi. 2nd pi. 3rd pi. 1st du. 2nd du. 3rd du.

Old Polish bych by by bychom, bychmy byscie bychg. bychowa, bychwa bysta bysta

Modern Polish by-m by-s by by-smy by-scie by

From the 1700's on, the conditional enclitic complex (by + person and number-marker) begins to gravitate out of clause-second position and towards agglutination to the /-form. The earliest discernible pattern in this development is agglutination to the /-forms byi 'were', zostai 'became', miai 'had', mogi 'could, might', chciai 'would', which continue to account for a large proportion of the examples of agglutination (1041 of the 1741 counted for the 1900's in (11) (Rittel 1975:120; 146), and to modal predicates with sentential subject (e. g., mozna 'may', trzeba 'must') (Rittel 1975:136, 146). Deviation from clause-second position is favored by the occurrence of heavy initial thematic elements followed by a caesura, but it is strongly disfavored by the occurrence of conjunctions in clause-initial position (Rittel 1975:143). Thus agglutination is far and away more common in main clauses than in subordinate clauses.

36

Henning

Andersen

(11)

1500's 1600's 1700's 1800's 1900's

Total no. of examples

Deviations from Wackernagel's rule

Agglutination to preterite form

879 1137 693 1542 4648

33 62 36 410 393

65 134 120 712 1741

4% 6% 5% 27% 9%

7% 12% 17% 46% 38%

There is no prosodic univerbation in the standard language, where the orthoepic norms prescribe antepenultimate stress in singular form (e.g., m'ogia-bym Ί might; fem.') and pre-antepenultimate stress in plural forms, usually with a secondary stress on the penult (e.g., m'ogli- ,bysmy 'we might; masc.'), unless of course the /-form is monosyllabic (cf. sz'edi-bym Ί would walk', szl'i-bysmy 'we would walk'). However, even though the process of agglutination is clearly still in its initial stages, deviations with straight penultimate stress do occur in speech, apparent evidence of a beginning prosodic univerbation (Topolmska 1961:48).

1.3. Present tense of BE The Old Polish copula and verb of existence had two present tense paradigms (orthotonic and enclitic) identical to the ones cited above in (2). The original present tense forms were reinterpreted as person and number-markers for the reasons mentioned in sec. 2. The full, originally orthotonic 1st and 2nd person markers go out of use by around 1500 (Decaux 1955:133). But the original 3rd person forms, which, as mentioned in sec. 1.3.1 were reinterpreted as person-unspecific predicators, continue into modern Polish as verb of existence and copula. In Old Polish these are used with the enclitic person and number-markers for 1st and 2nd person in clause second position. In the examples in (12), (a) illustrates the orthotonic copula, which remains current until the early 1500's; (b) and (c) show the use of the predicator jest as verb of existence and as emphatic copula, accompanied by person and number markers; and (c), (d) illustrate the use of the person and number markers alone. This use of zero

From Auxiliary

to Desinence

37

copula with person and number markers survives into modern Polish as an option with a small number of predicate adjectives (e. g. powinien 'obliged', pewien 'sure', gotow 'ready', ciekaw 'curious'; Decaux 1955:113 f.), cf. (13). But otherwise the explicit predicators jest and sa are generalized, and the person and number markers are soon agglutinated to these to form the paradigms of (14). (12) a. b.

c.

d.

e.

(13)

Old Polish copula and verb of existence Ty jes uliczka rajska 'thou art road to: paradise' Ty, jen- ze-i jest przez pocz^tku ... thou who -2sg emph since beginning 'Thou, who hast been since the beginning ...' Bo ja jest-em Maria for I emph-lsg Mary 'For I am Mary' Bo-m ja pan Bog wasz for-lsg I Lord God your 'For I am the Lord, your God' Ale my, co-smyl but we, what-1 pi 'But we, what are we?' A ciekawa-m, czy Pan to wie ) 'But I wonder if A jest-em ciekawa, czy ... ) you know that'

In the standard language and most dialects of Polish, morphological univerbation is long since completed — though tmesis still occurs in some archaic dialects. The stress in the paradigms in (14) is consistently penultimate in all varieties of Polish, which indicates that prosodic univerbation in this paradigm is older than in the preterite. (14)

Present tense of BE, copula and verb of existence Old Polish Modern Standard tmesis Polish dialects 1st sg. -(e)m jest jestem j'estem 2nd sg. -(e)s jest jestes j'estes 3rd sg. jest jest j'est 1st pi. -(e)smy jest/sg. s^smy jest'esmy 2nd pi. -(e)scie jest/s^ s^sce jest'escie 3rd pi. s'ij S3

38

Henning Andersen

1.3.1. Segmental univerbation A development of new inflectional affixes such as the one we have considered in the preceding sections cannot be evaluated without reference to the (morpho)phonological rules which may apply when morphemes are combined into syntagmatic units of whatever size. We cannot make a survey of Polish morphophonemics here, but it must be mentioned that in the inherited inflectional system of this language, when an affix beginning with a vowel is attached to a stem, it either occasions no change in the stem-final consonant or it conditions one of several consonant mutations in it. In addition, affixation may be accompanied by changes in stem vocalism. The mutations stem-final consonants are subject to in the inherited inflectional system involve changes in place and/or manner of articulation, but they never affect the voicing properties of stem-final obstruents. The person and number markers have historically had a different status. The Polish language area is traversed by a phonological isogloss that divides it into two major areas with different treatment of final obstruent followed by initial vowel or sonorant. The difference at phrase internal boundaries is illustrated in (15). The rules of voicing sandhi and devoicing sandhi apply as well at enclitic boundaries, regardless of whether these coincide with a morphological word boundary, as in (16) or are internal to a morphological word, as in (17). Northeast

a. brat Adama 'Adam's brother' b. dziad Adama 'Adam's grandfather'

West and South [-da-] [-da-] West and South

a. jak-em wroctt 'when I returned' b. otk^d-em wrocit 'since I returned'

[-ge-] [-de-]

[-ke-] [-te-]

[nuzem] [riuzes] [rius] [vuzem] [vuzes] [vus]

[nusem] [riuses] [rius] [vusem] [vuses] [vus]

(15)

(16)

(17) a. niosl-em Ί carried' niosl-es 'you carried' niost 'he carried' b. wiozl-em Ί drove (trans.)' wiozl-es 'you drove' wiorf 'he drove'

[-ta-] [-ta-] Northeast

Front Auxiliary

to Desinence

39

It is significant that in some dialects, including the ones on which standard Polish is based, the 1st and 2nd person singular masculine forms exemplified in (17) were remade, already by the 1500's, so that they are formed with the allomorph of the /-form used in the feminine and neuter singular and the (now) non-virile plural, cf. (18). (18)

Fem. sg. niosiam niostas niosta

Masc. sg. niostem niostes niost

Pan-Polish [noswam] [noswem] [noswas] jnoswes] [rioswa] [nus]

Through this morphophonemic innovation, the application of the (de)voicing sandhi rules has been obviated, and in effect a segmental univerbation achieved. In complete conformity with the sandhi differences mentioned above, the present tense paradigm of BE has different regional variants as in (19), which additionally presents the present tense paradigm of a verb with stem-final ft/ to illustrate the normal present tense endings, the regular stem-final consonant mutation pattern in verbs of this type, and a regular concomitant vowel alternation in the stem. (19)

West and South lstsg. jezd-em 2nd sg. jezd-es 3rd sg. jest 1st pi. jezd-ezmy 2nd pi. jezd-esce 3rd pi.

Northeast jest-em jest-es jest jest-esmy jest-esce s^

'to braid' plot-? plec-e-s plec-e plec-e-my plec-e-ce plot-3

Although the effects of the sandhi change are clear enough in the paradigm from the West and South, from a synchronic point of view it is doubtful whether the (de)voicing rules are applied now. Since Polish consistently neutralizes the voicing opposition in obstruents in word-final position, the paradigm of the West and South can be derived from a stem /jezd-/, that of the Northeast from a stem /jest-/. It would seem that segmental univerbation is complete, but the paradigms still stand out by their suppletion, their special person and number markers, the absence of stem-final consonant mutation and of any concomitant vowel alternation. Only one more effect of the sandhi changes remains to be mentioned, the application of the voicing rule before the 1st person plural

40

Henning

Andersen

marker in (e)zmy; cf. (19). In the analysis of the person and number markers in sec. 1.1.5, I identified a morpheme -s- occurring in the person and number markers of the 2nd singular and the 1st and 2nd plural. There is no doubt that originally this segment was voiceless. There is no doubt either that it is followed by a morpheme boundary. But there is no way of deciding, in any dialect of Polish, whether synchronically the (de)voicing sandhi rules apply at this morpheme boundary. If they do, it is undecidable whether the segment is voiced or voiceless. If they do not, it must be a -z- in the West and South, and a -s- in the Northeast. These remarks on the segmental phonology of the person and number markers show that they are still treated as enclitics when they are attached to items other than /-forms (cf. (16)), which is becoming less and less common, and that also the received preterites exemplified in (17) still show evidence of their status of enclitics. Outside the standard language, there is no evidence on the chronology of segmental univerbation. But in the Polish imperative, (de)voicing sandhi does apply at a boundary internal to a morphological word which is also a prosodic word, as shown in (20). One may infer that neither morphological univerbation nor prosodic univerbation can hinge on a prior completion of segmental univerbation. (20) a. uniesmy 'let us b. uwiezmy remove' (tr.)

1.3.2. Excursus: the Polish

West and South [unezmy] [uvezmy]

Northeast [unesmy] [uvesmy]

jest

In concluding this survey I will briefly mention its relevance to a widely known account of the origin of the Polish present tense paradigm of BE. This paradigm has been touted by several textbooks of historical linguistics as a prime example of morphological levelling, a rebuilding of an inflectional paradigm "on the third person singular form as if it were the basic stem", whereby the verb has been '[integrated] into the productive conjugational system of the language so that the inflectional endings ... are those found in other verbal paradigms' (thus Bynon 1977:132).

From Auxiliary

to Desinence

41

The account given in the preceding pages of the development of this paradigm (sec. 1.0.2, 1.1.1, 1.1.5, 1.3.1) and of its exceptional status in modern Polish conjugation (sec. 1.3.2) relieves me of the task of showing at this point how little correct information there is in Bynon's statements. The present tense paradigm of BE has indeed been rebuilt. And the history of this paradigm does perhaps make it well suited as a textbook example. But not of paradigm levelling; to illustrate this we should choose other examples. Polish BE should rather be used as an example of what it is. It is a rather special, but interesting, detail in a development of inflection through the univerbation of original free forms and movable clitics, by which clitic allomorphs of one verb, after being reinterpreted, have ended up as desinences on the orthotonic allomorph(s) of the same verb (see (9)). What it might illustrate in a textbook is that in morphology as elsewhere, there is a difference between a diachronic correspondence and the change or changes that have given rise to it, and that a simple account of the differences between a paradigm at one stage of a language and the same paradigm at another stage may not bear any resemblance to the actual historical development that connects them. Or, on a more general level, it might be used to drive home the point that a diachronic linguistics without language history is nought.

2. Interpretation The data I have surveyed in the preceding pages illustrate a gradual drift from one ordered state — person and number markers in clausesecond position — towards another ordered state — person and number markers as desinences. We cannot claim that this development has run its course. Even limiting ourselves to standard Polish, we must acknowledge that the person and number markers remain free to occur in tmesis in the preterite and predominantly do so in the conditional. But the development as a whole is so far along, and presents so many indications of morphological and phonological univerbation, that its end point is, if not clearly in sight, at least plain to see. Like all other cases of drift, this one confronts the language historian with the problem of explaining why one orderly and perfect-

42

Henning

Andersen

ly transparent state of affairs would yield to centuries of fluctuating usage, and this in turn give way to a steadily clearer new order. To solve this problem we must go well beyond the registration of changes in usage which are observable in the data. In the following lines I will try to sketch the relations of motivation which, as far as I can judge, link the major innovations in this historical process. I base this sketch on Coseriu's conception of the articulation of language (see, for instance, 1968), which distinguishes usage from grammar, and views a grammar as a hierarchy of different levels of organization, norms, system, and type. The norms are the unobservable rules of usage which at any given point guide the members of a speech community in their selection of alternative means of expression and confer socially recognized values on their choices. They comprise both the usage that is motivated only by tradition and that received usage which can be derived by the productive rules of the grammatical system. It is in terms of the norms that usage can be evaluated as, for instance, innovative or current or old-fashioned. The grammatical system comprises the productive rules of the language. There is much in the norms of any given language state that cannot be derived by these rules. On the other hand, the very productivity of the rules of a grammar allows for the production of usage that has not yet been codified in the norms. By type is understood the abstract principles of structure that underlie a given concrete grammatical system. Just as usage may conform to or deviate from the established norms, and just as the norms of a language in part conform to the system of productive rules, in part deviate from it, so a concrete grammatical system may be in greater or lesser conformity with its type, as when a basically isolating language has certain productive morphological patterns, or a basically SVO language contains productive rules for the formation of verb-final clauses. Unconformities between different levels of organization, and between usage and grammar, are major sources of motivation for historical innovations. I will mention here the types that seem relevant to the Polish development under discussion. I add in parentheses terms I have used to describe such types of innovations in other writings. Innovations in the system may be motivated by indeterminacies in existing usage (abductive innovations) or by the type (deductive).

From Auxiliary

to Desinence

43

Innovations in the norms may be directly motivated only by existing usage (abductive innovations). Innovations in usage may be motivated by the system (deductive innovations), and when such new usage is adopted by the norms, these innovations in the norms may be said to be indirectly motivated by the system. But innovations in usage may also be motivated by specific communicative needs of the members of a speech community (accommodative innovations); such innovations need not, but may be motivated additionally by the system. In these relations of motivation, only the norms play no active role. Norms safeguard continuity, they do not give rise to innovations. In the following lines I will try to pull together what we can say (however hypothetically) about the history of the Polish person and number markers against this background. I will speak of them mainly with reference to the preterite, for which we have the fullest documentation for the entire development.

2.1. Usage The statistics cited in (5) show a gradual change in the placement of the person and number markers, which over the last five hundred years with increasing frequency are attached to /-forms in preterite sentences. During the same period, the person and number markers occur with increasing freedom outside clause-second position. A usage oriented description of this development might call the person and number markers desinences whenever they follow /-forms and enclitics whenever they do not (cf. sec. 1.1.3) and would thus see here a gradual change whereby the person and number markers have become steadily less enclitic and more and more desinential. This is in essence Rittel's approach. She does not acknowledge that in Old Polish an /-form in clause-initial position might be followed by an enclitic person and number marker, but speaks of agglutination regardless of where in a clause an /-form is followed by a person and number marker, and so by her very terminology is unable to pose the question of how and why these enclitics turned into desinences. One consequence of this is that she is unable to integrate her description of the changes in person and number marker placement with her observations about the apparent differences in norms at different times during this development.

44

Henning

Andersen

2.2 Norms The gradual change in usage contrasts with a relatively clear-cut change in the norms. It is a great merit of Rittel's investigation that she has given considerable attention to the question of the stylistic value of person and number marker placement in different periods. Her analysis of the materials shows that in the early part of the development, deviations from Wackernagel's rule are pragmatically motivated, whereas in the more recent part of it person and number marker placement in clause-second position can be evaluated as a sign of old-fashioned, casual, or popular speech, and is utilized in artistic prose, especially in dialog, with these values. From the point of view of the norms, then, the development falls into two periods. In the earlier one, the unmarked placement of the person and number markers is after the first orthotonic word of the clause. Deviations from this norm are permitted when specifically motivated. One type of deviation is placement after an initial phrase (rather than the first word), which safeguards the adjacency of its constituents. Another type is placement after the first word or phrase that follows an intonational caesura, which helps set off the thematic elements that precede it. Yet another is placement after a word that carries emphatic stress. A final category of enclitic displacement is remedial, aimed at evading disturbing homophony with unintended grammatical forms, or at avoiding cacophony. All of these types of innovations in usage are pragmatically motivated (they are accommodative) and they serve the ends of clarity of expression. In the second period, which by Rittel's evaluation begins at the end of the 1600's, the unmarked placement is after the /-form. Permitted deviations from this norm include the pragmatically motivated types just mentioned and the clause-initial placement hallowed by tradition, but synchronically unmotivated. Interesting evidence of the speakers' awareness of these permitted alternatives and of their different values can be found in discrepancies between a writer's manuscript and the printed edition of his work, in writers' corrections of their own manuscripts, and in their revisions of later editions of their works. Rittel gives a few such examples from the 1800's (1975:71 —72), but one could wish to see more such data, especially from earlier periods, for they offer invaluable information on the otherwise unobservable norms that govern usage.

From Auxiliary

to Desinence

45

2.3. System On the basis of the change in the norms outlined above, one might well want to speak of the person and number markers as basically enclitics in the earlier period and basically desinences in the later period. One could then interpret the dynamics of the development the following way. In the earlier period the enclitics gained more and more freedom of placement, for as each type of specifically motivated deviation from their previous privileges of occurrence lost its marked character (through fading), new well motivated additional deviations would become possible. In the later period, after the person and number markers came to be perceived as desinences, they would then gradually have lost their freedom of movement, because this freedom is incompatible with their status of desinences. This is a possible drift scenario, which posits accomodative innovations in usage and series of minor abductive alterations before and after a major change in norms, and it may well be close to the truth. But it cannot explain the history of the Polish person and number markers for two quite different reasons. In the first place, this interpretation does not take into consideration that the preterite underwent prosodic and segmental univerbation already in the 1500's (cf. sec. 1.1.4 and 1.3.2). These changes in the system of conjugation are evidence that the person and number markers began to be treated as desinences centuries before they ceased to be treated as enclitics. In the second place, this interpretation makes the mistake of viewing the person and number markers as if their occurrence as enclitics or desinences depended only on their own syntactic properties. But just as it takes two to tango, an inflectional univerbation requires not only an element that can serve as desinence, but also, and perhaps first of all, a stem. It may be recalled that in sec. 1.0.3, I argued that after the restructuring of the tense system on the eve of the historical period, the original present tense auxiliaries were reinterpreted as person and number markers, and the /-forms as finite preterites. I would suggest that this reinterpretation of the content of the /-forms entailed a revaluation of their syntactic status: they became stems. (I will return to this hypothetical entailment relation below, in sec. 2.4.)

46

Henning

Andersen

I propose that the entire drift towards agglutination observed over the last five centuries has been powered by the consistent valuation by speakers of this language — throughout this period — of the /forms as finite verb stems, that is, bound forms which according to the morphotactics of this language require the attachment at their right margin of an element marking person and number. Note that the segmental univerbation described in sec. 1.3.2 consisted in the replacement of the free masculine singular /-form by the bound allomorph used for other genders and numbers. An analogical rebuilding of a masculine singular on the model of more marked gender and number forms is unusual, but this one can be understood as motivated morphotactically (rather than by the markedness values of the associated content categories). As mentioned, the segmental univerbation is attested from the 1500's on. Note further that even in dialects where this innovation did not take place, a prosodic univerbation did occur, an indication that in such dialects the originally free masculine singular /-form was revaluated as a bound form. Also the consistent distinction in stress placement between /-forms and other words followed by a person and number marker makes sense on this interpretation: the penult rule counts back from word boundaries (cf. (8)), but not from stem boundaries (unless followed by a terminal zero desinence); (cf. (7). The essence of the morphological univerbation, then, is a new assignment of syntactic status to the /-forms, an unobservable change that can only be known by its observable consequences. It was reflected in usage through the utilization of an existing bound stem instead of the free masculine singular stem or, in other dialects, through a revaluation of this last mentioned stem as bound. It was reflected in the prosodic univerbation, first in the singular, subsequently in the plural, after a delay that can only be understood against the background of a full history of Polish stress (see Topoliriska 1961). And it was reflected in the gradual change in person and number marker placement which was facilitated by the pragmatically motivated enclitic displacement. The morphophonemic and prosodic innovations were early adopted by the norms. The gradual change in person and number marker placement was actuated much more slowly, presumably not just because of the remarkable linguistic conservatism of this speech community, but rather because enclitic placement continued to carry essential pragmatic information.

From Auxiliary

to Desinence

47

I mentioned in sec. 1.1.3 that Laskowski in his description of modern Polish verb morphology describes the preterite as composed of an /-stem and (person and number) desinences susceptible to being moved. This description may very well have been appropriate since the 1400 — 1500's — as a description of the system. The norms — the set of rules each learner of a language must acquire to adapt his speech to that of his fellows and to the social and stylistic values he perceives this usage has — evidently did not come to conform to the system for a long time. Rittel's guess that the shift in the norms occurred towards the end of the 1600's may be right, but this guess is based on the fact of usage that from this time on more than 50% of preterites are followed by person and number markers. In fact, however, we cannot know what overall frequency might be correlated with the shift in the norms, for more likely than not some varieties of usage were more important for the constitution of the norms than others. In the preceding paragraphs I have deliberately used the terms enclitic and desinence with different values depending on the level of language organization under discussion. What I would suggest is that in a description of usage in spoken or written texts, we can speak of the person and number markers as clause-enclitics, phrase-enclitics, word-enclitics or desinences according to their actual placement in utterances. And we must then acknowledge that some occurrences can be assigned more than one label. In a description of the norms we can speak of the person and number markers as enclitics or desinences according to the speakers' (apparent or avowed) perception of their basic value. When we speak of the system of the language we can maintain that they have been desinences throughout the period in question, because right from the beginning of this period there have been stems ready to receive these person and number markers as desinences (and because the entire development has tended to translate this ideal state into reality). Recent studies of clitics have demonstrated the difficulty of distinguishing between clitics and desinences and have highlighted their essential identity (Zwicky 1977 and 1985; Klavans 1985). Both types of entities are affixes, but while desinences are affixes to stems, proper clitics (Zwicky's 'special clitics') are affixes to phrases. In a language that presents synchronic variation of the kind attested for the person and number markers in Polish since the beginning of its attested history, the task of description does not consist in assigning appropri-

48

Henning

Andersen

ate labels to the entities observed in the texts. As I have suggested here, it is rather a matter of determining the relation between the observed usage and the grammatical system that lies behind this usage. The attested drift in this language is a good illustration that where there is a tension between usage and system, tests for cliticness do not afford much insight into the system of the language.

2.4. Summary To summarize this interpretative account, let me briefly mention the major innovations it has involved and the relations of motivation in which each of them, as far as I can judge, should be understood. The reorganization of the Polish tense system in the Middle Ages resulted in usage that made the present tense (21)

Synopsis of developments discussed

TYPE:

SYSTEM:

NORMS: U: M:

USAGE:

1500 1600 1700 1800 1900 2000 Agglutinative Finite verb forms comprise bound stem + person and number marker /-forms are stems Movable person and number markers are suitable desinences person and marker placement: 2nd position Suffixation to /-forms Pragmatically Pragmatically and motivated stylistically displacement motivated tmesis Movable person and number markers attached to /forms: 23% 49% 68% 70% 92% 1500

1600

1700

1800

1900

2000

forms of auxiliary BE reinterpretable as person and numbers markers and the /-participles reinterpretable as finite preterite forms. Both these abductive innovations occurred.

From Auxiliary

to Desinence

49

The reinterpretation of the /-forms entailed that they were revaluated as stems. This revaluation was a deductive innovation in the system, motivated by general structural properties of the language: it was (and is) of a predominantly agglutinative type (with some fusion and symbolism) in which finite verb forms (the present, the imperative) comprise a bound stem followed by markers of person and number. The revaluation of the /-forms motivated the deductive morphophonemic innovations by which, in some dialects, the free masculine singular /-forms were replaced by bound allomorphs, while in other dialects they were revaluated as bound. These morphophonemic innovations facilitated a prosodic univerbation, similarly motivated by the /-forms' morphosyntactic status as stems. The morphophonemic innovations and the prosodic univerbation were innovations in usage, early codified in the norms. Pragmatically motivated innovations in enclitic placement usage gradually modified the norms so that the person and numbers markers gained greater freedom of (relatively unmarked) occurrence. In the process, they came to occur with ever greater frequency attached to /-forms, particularly those of 'grammatical verbs'. This suggests that these innovations in usage were in part motivated by the syntactic status of the /-forms. The blurring of the enclitic placement usage made a shift in the norms possible, whereby the norms were brought into basic conformity with the system: person and number marker placement after /forms became unmarked. Subsequent innovations in usage, motivated by the system, have gradually made the norms correspond ever more closely to the system; pragmatically motivated deviations from attachment to /-forms have decreased in frequency to the point where — in the northern parts of the language area — tmesis is evaluated as an archaism. A similar account could be given for the present tense of BE and for the conditional, but with significant differences in chronology. While the development was completed for the present of BE several centuries ago, at least in the standard language, the development of the conditional apparently has not reached the important shift in the norms yet. Time will tell if and when it does.

50

Henning Andersen

References Abductive and deductive change. Lg. 4 9 . 5 6 7 - 5 9 5 . 1975. Towards a typology of change: bifurcating changes and binary relations, in: Historical linguistics, ed. by John M. Anderson and Charles Jones, 2.18 — 62. Amsterdam: North-Holland. ANDERSEN, HENNING. 1 9 8 0 . Morphological change: towards a typology, in: Historical morphology, ed. by Jacek Fisiak, 1—50. The Hague: Mouton. ANTTILA, RAIMO. 1972. An introduction to historical and comparative linguistics. New York: Macmillan. BLOOMFIELD, LEONARD. 1935. Language. London: Allen & Unwin. BORETZKY, NORBERT. 1 9 7 7 . Einführung in die historische Linguistik. Reinbek bei Hamburg: Rowohlt. BYNON, THEODORA. 1 9 7 7 . Historical linguistics. Cambridge: University Press. COMRIE, BERNARD. 1 9 8 0 . Morphology and word order reconstruction: problems and prospects, in: Historical morphology, ed. by Jacek Fisiak, 7 1 — 8 2 . The Hague: Mouton. COSERIU, EUGENIO. 1 9 6 8 . Sincronia, diacronia i tipologia, in: Actos del XI Congreso Internacional de Lingüistica i Filologia Romänicas, 1.269 — 283. Madrid. Reprinted as: Synchronic, Diachronie und Typologie in hist: Sprache. Strukturen und Funktionen. XII Aufsätze zur allgemeinen und romanischen Sprachwissenschaft ( = Tübinger Beiträge zur Linguistik, 2), 91 —108. Tübingen: Gunter Narr. DECAUX, ETIENNE. 1955. Morphologie des enclitiques polonais. Paris: Institut d'Etudes Slaves. GABELENTZ, GEORG VON DER. 1891. Die Sprachwissenschaft. Ihre Aufgaben, Methoden und bisherigen Ergebnisse. Leipzig: Tauchnitz. GRZEGORCZYKOWA, RENATA; ROMAN LASKOWSKI; and HENRYK WRÖBEL (eds.). 1 9 8 4 . Gramatyka wspolczesnego jfzyka polskiego. Morfologia. Warszawa: Panstwowe Wydawnictwo Naukowe. HODGE, CARLETON T. 1 9 7 0 . The linguistic cycle. LSci. 13.1 — 7. KLAVANS, JUDITH L. 1 9 8 5 . The independence of syntax and phonology in cliticization. ANDERSEN, HENNING. 1 9 7 3 . ANDERSEN, HENNING.

Lg. 6 1 . 9 5 - 1 2 0 . KLEMENSIEWICZ, ZENON. 1 9 7 4 .

Historia j?zyka polskiego. Warszawa: Panstwowe Wyd-

awnictwo Naukowe. KLEMENSIEWICZ, ZENON; TADEUSZ LEHR-SPLAWINSKI; a n d

STANISIAW

URBANCZYK.

1964. Gramatyka historyczna j?zyka polskiego. Warszawa: Panstwowe Wydawnictwo Naukowe. KURASZKIEWICZ, WITOLD. 1970. Podstawowe wiadomosci ζ gramatyki historycznej jezyka polskiego ζ wyborem tekstow staropolskich do cwiczen. Warszawa: Panstwowe Wydawnictwo Naukowe. LYONS, JOHN. 1977. Semantics I —II. Cambridge: University Press. RITTEL, TEODOZJA. 1 9 7 5 . Szyk czionow w obr^bie form czasu przeszlego i trybu przypuszczaj^cego. Wroclaw, Warszawa, Krakow: Ossolineum. ROBINS, ROBERT HENRY. 1 9 6 7 . A short history of linguistics. London: Longman. SCHOONEVELD, CORNELIS Η. VAN. 1959. A semantic analysis of the Old Russian finite preterite system ( = Slavistic Printings and Reprintings, 7). The Hague: Mouton. SEREBRENNIKOV, BORIS ALEKSANDROVIC. 1 9 7 3 . Verojatnostnye obosnovanija Ν komparativistike. Moskva: Nauka.

From Auxiliary to Desinence

51

BORIS ALEKSANDROVIC (ed.)· 1 9 7 5 — 1 9 7 6 . Allgemeine Sprachwissenschaft I —III. Berlin: Akademie-Verlag. TAULI, VALTER. 1 9 5 8 . The structural tendencies of languages ( = Annales Academiae Scientiarum Fennicae, ser. B, torn. 115, 1). Helsinki: Suomalainen Tiedeakatemia. TOPOLINSKA, ZUZANNA. 1961. Ζ historii akcentu polskiego od wieku XVI do dzis (Prace J^zykoznawcze, 27). Wrodaw, Warszawa, Krakow: Ossolineum. VAILLANT, ANDRE. 1974. Grammaire comparee des langues slaves, III. Le verbe, 1 — 2. Paris: Klincksieck. WHITNEY, WILLIAM D W I G H T . 1867. Language and the study of language. Twelve lectures on the principles of linguistic science. London: Trübner. ZWICKY, ARNOLD M. 1977. On clitics. Bloomington: Indiana University Linguistics Club. ZWICKY, ARNOLD M. 1 9 8 5 . Clitics and particles. Lg. 6 1 . 2 8 3 — 3 0 5 . SEREBRENNIKOV,

Henning Andersen Institut for Lingvistik K.0benhavns Universitet Njalsgade 96 DK-2300 K0benhavn S

Copula Auxiliarization: How and Why? Simon C. Dik

0. Introduction As is well known, many genetically unrelated languages use that verb which primarily functions (or originally functioned) as a copula in a variety of auxiliary functions. Thus, English has the copula be as an auxiliary in the passive (1) and the progressive (2): (1) (2)

John was fired John was reading

Dutch has the copula zijn 'to be' in the passive perfect (3), in the active perfect of a subclass of intransitive verbs (4), and in a construction which is more or less equivalent to the progressive (5): (3)

(4)

(5)

Jan is ontslagen John is fired 'John has been fired' Jan is geslaagd John is succeeded 'John has succeeded' Jan is aan het vissen John is at the fish (inf) 'John is fishing'

In some of these languages, we can even get combinations of copulas in different auxiliary functions. Compare: (6) (7)

John was being treated in the hospital Jan is wezen vissen John is been fish (inf)1 'John has been (involved in) fishing'

I will use the term Copula Auxiliarization for the diachronic process whereby the copula comes to be used in such auxiliary functions. In

54

Simon

Dik

this paper I will discuss three questions relevant to Copula Auxiliarization: (i) (ii)

What kind of process is Copula Auxiliarization? How does Copula Auxiliarization come about, i.e. what routes or channels does it follow? Why does Copula Auxiliarization come about, i.e. what factors can explain the occurrence of Copula Auxiliarization?

(iii)

1. What kind of process is copula auxiliarization? Meillet (1912:148) was perhaps the first to introduce the notion (and the term) 'grammaticalization'. 2 By this he meant a process whereby an originally independent word with independent meaning (mot autonome) develops into an auxiliary word (mot accessoire), and finally ends up as a grammatical marker (element grammatical). This process is characterized by a concurrent weakening of both the meaning and the form of the word in question: L'affaiblissement du sens et l'affaiblissement de la forme des mots accessoires vont de pair; quand Tun et l'autre sont assez avances, le mot accessoire peut finir par ne plus etre qu'un element prive de sense propre, joint ä un mot principal pour en marquer le role grammatical. Le changement d'un mot en element grammatical est accompli. (Meillet 1912: 139) With Heine & Reh (1982) and Lehmann (1982) I will use the term 'desemanticization' for the weakening of meaning, and the term 'erosion' for the formal weakening involved in grammaticalization processes. According to Meillet, Copula Auxiliarization is a matter of desemanticization. In his view (1912:131), the copula has less and less specific meaning in the following constructions: (8) a. b.

Je suis celui qui je suis Ί am the one who I am' Je suis chez moi Ί am at home'

suis est mot autonome suis a encore une certaine autonomie

Copula Auxiliarization:

c. d.

Je suis malade Ί am ill' Je suis parti Ί have left'

How and Why?

55

suis n'est presque plus qu'un element grammatical suis est tout ä fait element grammatical

Of course, the idea that the copula is, in certain occurrences, an independent verb which, in other occurrences, has lost something of its original meaning represents a long-standing view on the nature of the copula. I do not believe that this view is correct, nor do I believe that (8 a —d) represent a scale of desemanticization. My reasons are the following: (a) (b)

it is doubtful whether, even in constructions such as (8 a — b), the copula can be said to have independent meaning. the auxiliary in (8d) is less far removed from its original function than Meillet would have it.

I will now discuss these two points in somewhat more detail.

1.1. Does the copula ever have independent meaning? Within the framework of Functional Grammar I have analyzed the copula as a supportive verb which in certain conditions is inserted into predications with non-verbal predicates in order to help express those grammatical distinctions which are otherwise coded in the verbal predicate (Dik 1980: ch. 4; 1983). The rule of Copula Support which serves to insert the copula in the relevant conditions is judged to be part of the expression rules, and thus to have no contribution to make to the semantic content of the predication. The grounds for this view can be explained as follows. Compare the following constructions: (9) a. b. c.

John is in the garden John is the winner John is ill

(10) a. John — in the garden b. John — the winner c. John — ill

Where certain languages have a copula in constructions of type (9 a—c), others do without a copula, as in (10 a—c). There is no reason to assume, however, that there is any meaning difference between (9 a) and (10a), or between (9b) and (10b). Therefore, the copula in (9a—b)

56

Simon

Dik

cannot be said to have independent meaning. If it had, it would be indispensable in (10 a —b) as well. This reasoning does not only apply across languages, it also applies within languages: in most languages, constructions with non-verbal predicates occur without a copula in certain conditions. Compare: (11) a. b. c. d.

I want John (to be) in the garden by ten I consider John (to be) the winner I mean the chairs (which are) in the garden John was a boy (who was) extremely intelligent for this age

Again, the possible absence of the copula is incompatible with the idea that the copula has an independent semantic contribution to make to the content of the expression. Note that I do not claim that full verbs might not develop into copulas. What I claim is that, as soon as they have acquired copula function is in (9 a —c), they function as a purely supportive device without independent meaning of their own. From this it follows that Copula Auxiliarization cannot be a matter of desemanticization.

1.2. Is the auxiliary so different from the copula? In Dik (1983) I have shown that copular and auxiliary occurrences of be in English can all be described by means of one and the same rule of Copula Support. In other words, even in its auxiliary functions be can be treated as a supportive device. The reasoning behind this is as follows. Compare: (12) a. b. c. (13) a. b. c. (14) a. b. c.

the boy is extremely intelligent the boy who is extremely intelligent the extremely intelligent boy some students were invited the students who were invited the invited students some children were crying the children who were crying the crying children

Just as the absence of the copula in (12 c) does not affect the meaning of the expression, so the absence of the auxiliary in (13 c) does not

Copula Auxiliarization:

How and Why?

57

make the construction less passive, and the absence of the auxiliary in (14 c) does not make the construction less progressive. It may be concluded that the passive character of be invited does not reside in be, but in the passive participle invited·, likewise, the progressive character of be crying does not reside in be, but in the present participle crying. If this is correct, then the status of be in (13) and (14) is not that different from its status in (12): it is a supportive verb which is inserted into predications with non-verbal predicates in certain conditions. This means that the underlying predications of (12) —(14) may have similar structures, which can be represented as: (12)' (12)' (14)'

extremely intelligent (the boy) invited (some students) crying (some children)

Similar observations can be made about Latin, in which both the copula and the auxiliary esse may in many cases be absent in the same or similar conditions: (15)

(16)

(17)

Ciceronem eximium consulem (esse) puto Cicero-ACC excellent consul-ACC (be) I-consider Ί consider Cicero (to be) an excellent consul' Ciceronem victum (esse) puto Cicero-ACC defeated-ACC (to be) I-consider Ί consider Cicero (to be) defeated' Ciceronem eras venturum (esse) Cicero-ACC tomorrow coming-FUT-ACC (to be) puto I-consider Ί think that Cicero will come tomorrow'

Again we find that (16) is no less 'passive perfect' without esse; nor is (17) any less 'future active' for the absence of esse. We must conclude that the relevant semantic aspects do not reside in esse, but in the participles with which esse is combined in these constructions.

1.3. Copula auxiliarization is a matter of expansion If what is said in 1.1 and 1.2 is correct, then Copula Auxiliarization can be said to be a matter of Expansion (in the sense of Heine &

58

Simon

Dik

Reh 1982) rather than of Desemanticization. Heine & Reh define Expansion as the process whereby some grammatical element receives an extra, equally grammatical function. Expansion is involved, for example, when an originally locative preposition acquires the extra function of acting as an Recipient marker, or when an original Direction marker comes to be used as a purposive subordinator. In the case of Copula Auxiliarization we can say that the copula comes to be used in constructions which are integrated into the Tense-MoodAspect system of the predicate. Once this integration is completed, the copula can be said to have auxiliary status, while retaining its basic character of a semantically empty supportive verb. So far I have made the following two points: (a) (b)

the copula is a semantically empty supportive verb in all its occurrences; the copular auxiliary is not that different from the copula as such. In fact, it can also be treated as a supportive verb, the privileges of occurrences of which are often similar or even identical to those of the copula.

If these two points are correct, they help us understand the process of Copula Auxiliarization in the sense that the 'distance through grammatical space' covered by Copula Auxiliarization is less extensive than it would seem to be at first sight. Copula Auxiliarization is more a matter of gradual expansion into the domain of the verbal paradigm, than a categorical clear-cut acquisition of entirely new functions by the copula.

2. The aspectual origin of copula auxiliarization Periphrastic constructions containing a copula may synchronically play different roles in the grammatical system. Very often, they are used to express some aspectual distinction. But they may also have a purely temporal value, or be used for the expression of voice distinctions (as in English (1)). I believe that there is room for the view, however, that Copula Auxiliarization always originates in periphrastic constructions with an aspectual meaning, and only later potentially shifts to other uses.3

Copula Auxiliarization:

How and Why?

59

Such a hypothesis is empty, of course, if we do not have a working definition of what distinctions count as aspectual, as opposed to temporal (and modal) distinctions. Without being able to fully motivate this in the context of this paper, I shall here take the following view of this matter. First, I restrict the terms Tense and Aspect to distinctions which are grammatically (rather than lexically) coded. This means that I do not include Aktionsart in the definition of Aspect. By Aktionsart I understand the internal properties of some State of Affairs (SoA) as expressed by some predicate-frame. In Functional Grammar, the relevant distinctions are captured in a typology of States of Affairs which I will not discuss here in detail. The relevant parameters for this typology are:4 (18) a. ± Dynamic:

whether or not the State of Affairs involves any kind of change b. ± Control: whether or not the State of Affairs owes its existence or occurrence to some controlling entity c. ± Telic: whether or not the State of Affairs presupposes a natural end-point d. ± Momentaneous: whether the State of Affairs lasts only a moment of time (a temporal point) or a stretch of time (a temporal interval).

The resulting typology of States of Affairs is relevant to Tense and Aspect in the sense that there are complex interactions between State of Affairs types as defined by lexical (or derived) predicates on the one hand, and grammatical distinctions of Tense and Aspect on the other. Having restricted Tense and Aspect to oppositions which are grammatically rather than lexically coded, I now propose the following definitions for these categories:5 (19)

Tense: Tense distinctions locate some State of Affairs on the temporal axis in relation to the moment of speaking (absolute Tense) or to a reference point defined by some other State of Affairs (relative Tense).

The relevant parameters are whether the State of Affairs precedes, coincides with, or follows the moment of speaking (ms) or the reference point (r). Secondarily, the relative temporal distance between the State of Affairs and ms/r may be relevant to Tense systems.

60

Simon Dik

This may lead to such distinctions as between Recent Past and Remote Past. 6 The relevant Tense relations can be represented as in the following diagram: (20)

SoA

SoA

SoA

ms r The following schema of progressive articulation of Tense distinctions covers the Tense system of many languages: -Past (21)

Tense

-Recent Past -Remote Past

Present N o n - P a s t ^ ^ - I m m e d i a t e Future t uture Remote Future

As for the notion of Aspect, I believe that, even if this notion is restricted to distinctions which are grammatically rather than lexically coded, it covers a non-homogeneous group of at least three distinct subsystems which in different combinations and to different degrees may be relevant to the grammatical organization of a language. These three subsystems are those of (Im)Perfectivity, Phasal Aspect7 and Quantificational Aspect. These three types of Aspect may be defined as follows: (22) a. (Im)Perfectivity: This subsystem hinges on the question of whether the State of Affairs designated by the predication is presented from an external point of view, as a complete whole (Perfective), or from an internal point of view, as being non-complete (Imperfective). b. Phasal Aspect: Phasal Aspect distinctions serve to describe what is the case at some reference point on the temporal axis, in relation to the occurrence of some State of Affairs. c. Quantificational Aspect: Quantificational Aspect distinctions express different quantifications over a set of occurrences of some State of Affairs.

Copula Auxiliarization:

How and Why?

61

The Imperfective value of the (Im)Perfectivity subsystem merely signals that the State of Affairs described is in some sense 'open', not fully achieved; it suggests that either 'there is yet more of the same State of Affairs to come', or 'it is unknown or immaterial whether the State of Affairs in question ever came to an end'. By virtue of this semantic value of 'openness', the Imperfective may receive different contextually determined interpretations, such as 'progressive', 'habitual', 'iterative', 'continuous'. These interpretations correspond to values which may get distinct and explicit coding in the Phasal or the Quantificational Aspect subsystems. The most relevant Phasal Aspect distinctions can be represented as follows: SoA

(23) 2

3

Suppose that the State of Affairs as considered in (23) is 'John's writing a letter'. I take it that Phasal Aspect distinctions now serve to answer the question: 'what can be said at reference point i (i = 1...7) in relation to the State of Affairs of John's writing a letter?'. As the reference point shifts in relation to the State of Affairs, different Phasal Aspect distinctions are involved. These distinctions can be labelled and illustrated as follows: (24)

1 Prospective Aspect:8 2 Immediate Prospective Aspect: 3 Ingressive Aspect: 4 Progressive Aspect: 5 Egressive Aspect: 6 Immediate Perfect Aspect: 7 Perfect Aspect:

John is going to write a letter 9 John John John John John John

is about to write a letter starts writing a letter is writing a letter finishes writing a letter has just written a letter has written a letter

Note that all the terms in (24) are intended to be taken in their aspectual sense. As we shall see in 4 below, there is a tendency for aspectual forms to be reinterpreted as expressing Tense distinctions. For this reason, terms such as Perfect are often used for expressions which have already been (partially) integrated into the Tense system.

62

Simon

Dik

As formulated above, however, there is always a principled distinction between Tense and Phasal Aspect, even when the same 'constellation' of States of Affairs is involved. Consider the following examples: 4

(25)

John be rich man John ms a. Future:

It is stated at ms that — the State of Affairs 'John is a will obtain after the moment of b. Prospective: It is stated at ms that — John is such at ms that — the State of Affairs 'John is a will obtain after the moment of

rich man' speaking

rich man' speaking.

Thus, the Future simply places some State of Affairs in the stretch of time after the moment of speaking. The Prospective projects a state obtaining at the moment of speaking into a future State of Affairs. In the same way, we can distinguish between Past and Perfect: (26)

John be rich man John ms a. Past:

b. Perfect:

It is stated at the moment of speaking that — the State of Affairs 'John is a rich man' obtained before the moment of speaking. It is stated at the moment of speaking that — John is such at the moment of speaking that — the State of Affairs 'John is a rich man' obtained before the moment of speaking.

Again, the Perfect is a projection of a present state into a past State of Affairs. Quantificational Aspect answers the question: 'How often does State of Affairs occur?', or 'How often is entity χ involved in a State of Affairs?'. We can here make the following distinctions:

Copula Auxiliarization:

(27) a. Habitual: b. c. d. e.

Continuous: Semelfactive: Iterative: Frequentative:

How and Why?

63

State of Affairs occurs (repeatedly) due to a propensity of χ for State of Affairs State of Affairs occurs all the time State of Affairs occurs just once State of Affairs occurs several times State of Affairs occurs many times

As presented here, Quantificational Aspect and Phasal Aspect cannot be subsumed under one unified higher notion of Aspect. As noted by Comrie (1976:26), this seems to run counter to the fact that in many languages one and the same expression type (the Imperfective) is used for expressing both Progressive (Phasal) and Habitual/Continuous (Quantificational). According to Comrie this implies that these various distinctions must be thought of as forming one unified concept. I here take the alternative route of supposing that this type of uniform expression is due to a higher level generalization of the form: (28)

If χ has the habit of/is continuously involved in State of Affairs, then χ (in a sense) never finishes being involved in State of Affairs.

I thus assume that the Imperfective (in the sense mentioned above) generalizes over Phasal and Quantificational Aspect in terms of the common property of'Non-Completion'. Note, by the way, that there are many languages in which Progressive and Habitual/Continuous are duly distinguished both formally and semantically. If Tense and Aspect are defined as in (19) and (22 a —c), they are in principle independent of each other. Aspect distinctions such as those of (24) and (27) do not by themselves locate the State of Affairs on the time axis. And Tense distinctions per se are not sensitive to such reference points as 1 through 7 in (23). Complexities arise, of course, when Tense and Aspect distinctions are combined with each other. As indicated in the following diagram, the product of Tense and Phasal Aspect results in a great number of potential TenseAspect combinations: (29)

1234567

1234567

1234567

Past

Present

Future

64

Simon

Dik

For example: (30)

Past4 = PastProgr = John was writing a letter Pres5 = PresEgr = John finishes writing a letter Futl = FutProsp = John will be going to write a letter etc.

In a similar way, Tense may combine with Quantificational Aspect into such combinations as: (31)

PastHab = John used to write letters PresCont = The roof leaks (all the time) FutFreq = John will often come to visit you

Finally, certain Aspects may combine with each other, and the resulting combinations may combine with Tense into yet further possible combinations such as: (32)

PresPerfProgr = John has been writing a letter PastHabProgr = John used to be waiting for me FutProspFreq = This boy is going to frequently cause trouble etc. etc.

The present paper is obviously not the place to give a full elaboration of the sketchy outline of Tense and Aspect given above. What must be retained for our present purposes is that Phasal Aspect distinctions can always be regarded as property-assigning expressions. They always allow for paraphrases of the following form: (33) a. χ has the property at reference point i b. of such and such involvement in specified State of Affairs The fact that Phasal Aspect distinctions can be conceived of as property-assigning makes it understandable that especially copular constructions, being property-assigning constructions par excellence, provide a constant source for the innovation of the aspectual system.

3. Channels of copula auxiliarization There is evidence that the copula can develop into an aspectual auxiliary along different routes, or 'channels' in the terminology of

Copula Auxiliarization:

How and Why?

65

Heine & Reh (1982). The most important of these channels are the following: (i) (ii)

the Localist channel, the Property channel.

These will be dealt with separately in the following sections.

3.1. The localist channel The Localist origin of periphrastic constructions with a copula acquiring an aspectual meaning is insightfully discussed in Claudi & Heine (1985). Following the theory of Lakoff and Johnson concerning metaphor, they point to an underlying metaphor in which a State of Affairs is conceptualized as a spatial object which one can be close to, enter, be in/on/at, and exit from. According to this metaphor, local expressions can come to be used for expressing aspectual distinctions such as those diagrammed in (23), according to the following pattern: 10 (34) a. John is near the building b. John is near the going "John is about to go" ( = Immediate Prospective) (35) a. John is in the building b. John is in the going "John is going" ( = Progressive) (36) a. John is on the mountain b. John is on the going "John is going' ( = Progressive) Indeed, many languages have aspectual forms which can be traced back to such a localist metaphor. A case in point is Welsh, in which we find such forms as the following (Awbery 1976): (37)

(38)

mae ef ar weld y ddrama is he on seeing the play 'He is about to see the play' ( = Immediate Prospective) mae 'r dyn yn gweld y ci is the man in seeing the dog 'The man is seeing the dog' ( = Progressive)

66

(39)

Simon Dik

mae 'r dyn wedi gweld y ci is the man after seeing the dog 'The man has seen the dog' ( = Perfect)

As Awbery demonstrates, the complement of the prepositions ar, yn, and wedi has all the criterial properties of a nominalization, so that the relevant constructions can correctly be paraphrased as 'The man is at/in/after the seeing of the dog'. Just as illustrated in (30), these constructions can combine with different Tenses to form such expressions as: (40)

budd y dyn yn gweld y ci will-be the man in seeing the dog 'The man will be seeing the dog' ( = Future Progressive)

And just as illustrated in (32), different aspectual distinctions can also be combined with each other: (41)

mae Ifor wedi bod yn darllen llyfr is Ifor after being in reading book 'Ifor has been reading a book' ( = Present Perfect Progressive)

In terms of Functional Grammar we can represent this metaphorical process as follows. First consider a simple locative predication with its underlying structure (as suggested in Dik 1980: ch. 4): (42) a. John is in the garden b. Pres{(dlxj: garden^x^LocXdlXi:

JohnN(Xi))0

In the underlying structure the locative term 'in the garden' is applied as a predicate to the term 'John'. In English, the application of such a non-verbal predicate will trigger the rule of Copula Support, so that (42 a) will result from the expression rules. We can now use (42 b) as a model for a construction which literally means: (43)

John is in the reading of the book

In this construction, when analyzed in the same way as (42 a), 'in the reading of the book' is applied as a predicate to 'John': (44)

Pres{(dxj: [ready (xi)Ag(the book) Pat ](x j )) Loc }(dlx i : JohnN(xO)0

Copula Auxiliarization:

How and Why?

67

Literally, then, John is said to have the property of being inside the entity Xj, which itself is defined as 'his (i. e., John's) reading the book'. As Awbery convincingly argues, constructions such as (37) —(41) should be analyzed as complex predications which, however, have been integrated into the Tense/Aspect system of Welsh, thus providing expressions for several aspectual distinctions. Once such constructions as these have been integrated into the Tense/Aspect system through this kind of metaphorical extension, they tend to lose some of their original locative character: they tend to be cut loose from their semantic roots, so to speak, and processes of formal erosion may contribute to the drifting apart of the two constructions. 11 In Welsh this formal erosion is hardly noticeable, except for the fact that the preposition yn, in its aspectual usage, lacks the property of the local preposition yn of effecting a nasal mutation of the following consonant, and may be reduced to 'n in its aspectual, but not in its locative usage (Comrie 1976:100). But in other languages the aspectual construction has drifted further away from its locative model, although the latter can clearly be shown to be the original source. A very interesting example of this is given by Claudi & Heine (1985) from Ewe, a West African language of the Kwa family. Compare the following constructions: (45) a. me-le I-be Ί am b. me-le I-be Ί am

to dzi mountain on on the mountain' yi-yi dzi going on on the going' ( = Progressive)

In this case, the aspectual construction is fully parallel to the locative one, both having the same postposition dzi. But now compare: (46) a. me-le I-be b. me-le I-be Ί am (47) a. me-le I-be b. me-le I-be

xo me house in yi-yi m going going' ( = Progressive) anyf gbe ground at/place yi-yi ge going ( = Immediate Prospective)

68

Simon

Dik

Note that in these cases the original locative preposition has worn down to a reduced form, which now unmistakably signals that the construction has the particular aspectual value indicated. At this point there is no further need to use complex underlying structures parallel to (44) above, because the construction no longer signifies the literal paraphrase Ί am in/at the going'. Rather, a construction such as (46 b) can be simply derived from an underlying representation of the form: (48)

Pres Progr yi v (me) Ag

where Progressive will be mapped onto the combination of -le ... m which now acts as a kind of discontinuous Aspect marker. 12 The Localist metaphor is responsible for major and minor construction types in quite a few languages. In older stages of English we find constructions of the form he waes on fechtinge 'he was on fighting', which later eroded to he was a-fighting, a construction which survives as an archaism in Modern English. 13 Compare also Germ. Er ist am Singen, Dutch Hij is aan het zingen 'he is at (the) singing' (cf. (5) above), both used for expressing Progressive Aspect.

3.2. The property channel We saw in 2 above that Phasal Aspect distinctions can be conceived of as property-assigning expressions. A constant source of aspectual forms is therefore provided by expressions of the following form: (49) a. Prospective: John is (a person who is) going-to-win b. Progressive: John is (a person who is) winning c. Perfect: John is (a person who is) having-won An interesting, rather complex example of this type of construction from Basque is discussed in Shiratsuki (1985). In Basque the original Perfect has already been partially reinterpreted in terms of Tense. For expressing 'Perfect' in the aspectual sense, there is a complex construction type of the following form:

Copula Auxiliarization:

(50)

How and Why?

69

Liburu hori ni-k irakurri-a book that-(ABS) I-ERG read(PERF)-def.sg d-a 3sg.ABS-izan ( = be) 'That book is such that I have read it'

As Shiratsuki demonstrates, this construction is only seemingly transitive. If it were really transitive, then (i) the auxiliary would not be izan 'be' but ukan 'have', and (ii) both Agent and Goal would be coded in the verb through agreement. Compare the transitive parallel to (50): (51)

Ni-k liburu hori irakurri I-ERG book that(ABS) read(PERF) d-u-t 3sg. ABS-ukan-1 .sg.ERG

Construction (50), then, is comparable to property-assigning constructions of the form: (52)

Koldo handi-a d-a Koldo(ABS) big-def.sg 3sg.ABS-be 'Koldo is big (a big one)'

Correspondingly, Shiratsuki analyzes (50) as a construction in which the property of 'having been read by me' is assigned to 'the book', the result being a truly 'Perfect' or 'Resultative' construction. For another example of exploitation of the Property channel, consider Latin periphrastic forms consisting of a participle + esse. In the passive, such combinations provide the standard expression of the Perfect:14 (53)

Caesar victus est Caesar having-been-defeated he-is 'Caesar has been defeated'

With verba deponentia, such combinations can also have active Perfect value: (54)

Caesar haec locutus est Caesar these having-said he-is 'Caesar has spoken thus: ...'

70

Simon

Dik

The difference between Tense and Aspect comes out most clearly, however, in the opposition between the simple Future and the Immediate Prospective periphrastic form with a participle in -urus: (55)

(56)

Proficiscar leave-FUTlsg Ί will leave' Profecturus sum going-to-leave I-am Ί am going to leave' = Ί am on the point of leaving' Ί intend to leave'

The intentional shade which may attach to the Prospective may be understood as resulting from the meaning 'it is a property of χ now that χ will later'. One of the ways in which a property of χ many be projected into the future is through χ having the intention at reference point of doing State of Affairs after reference point. Other such 'bridges' between reference point and State of Affairs after reference point may be that χ has the plan or the purpose of doing State of Affairs (cf. Pinkster 1984:290). In terms of Functional Grammar, constructions which originally underlie the creation of aspectual forms through the property channel can be represented in the following way: (57)

This type of predication expresses that some entity (xj) has (had, will have) the property of being an entity of type Xj such that xj is going to be involved in (is involved in, has been involved in) pred v -ing. For a concrete example, consider the analysis of the Latin example: (58) (59)

Profecturus erit "He will be going to leave" Fut {(ilxj: Prosp proficisciv(Xj))} (dlxj: 3sg(xj))0

The application of the non-verbal predicate will, in appropriate conditions, trigger Copula Support; and the predicate will be expressed in the Prospective participle profecturus. The Property channel will often make use of participles, because these are quite suitable for expressing properties which are defined in terms of verbal predicates, as in (57). If the present analysis

Copula Auxiliarization:

How and Why?

71

is correct, however, participles are only one particular means for expressing term predicates of the form: (60)

(ilxj: predv(Xj))

Other such means would be headless or headed relatives of such forms as: (61) a. a person who predy b. one who predy c. who predy We may thus expect to find other languages which, as in Basque, use expressions such as (61) for the creation of aspectual forms.

3.3. Other possible channels The Localist channel and the Property channel no doubt account for the great majority of developments through which copular constructions become integrated into the Tense/Aspect system of the verbal predicate. According to Heine & Reh (1982:116) there is at least one further channel through which copulas can develop into auxiliaries: the Serial channel. This channel involves constructions of the following type: (62) (63)

(64)

Prospective: he go he eat Perfect:

'he is going to eat'

Progressive: a. he sit he eat 'he is eating' b. he is he eat 'he is eating'

Constructions of the form (64 b), then, may be a source for Copula Auxiliarization along this channel. I have some misgivings, however, about the question whether constructions of type (64 b) can truly be classed as serial constructions. It is characteristic of serial constructions that two or more verbs which originally signify separate aspects or 'phases' of one complex State of Affairs later amalgamate into one closely-knit

72

Simon

Dik

combination within which they may lose something of their concrete meaning. This characterization holds for all the other examples in (62) —(64). But what is the original 'independent meaning' of the copula in (64 b)? I suspect that either the verb glossed 'to be' in constructions of the form of (64 b) should be traced to verbs other than the copula, or that constructions of this form cannot properly be called 'serial'. Consider in this respect the structure of the Swahili verbal complex. Swahili simple verbs can contain just one instance of a rather considerable number of Tense/Aspect elements (cf. Ashton 1947), as in: (65) a. ni-na-som-a I-PROGR-read-IND b. ni-li-som-a I-PAST-read-IND c. ni-ta-som-a I-FUT-read-IND d. ni-me-som-a I-PERF-read-IND

Ί am reading' Ί read (Past)' Ί will read' Ί have read', Ί have finished reading'

Swahili allows for a great number of combinations of these Tense and Aspect operators. But whenever such a combination is expressed, the copula kuw- is needed as an auxiliary, so that the main verb will still contain only one Tense/Aspect marker. This results in such combinations as: (66) a. a-li-kuw-a a-me-lal-a he-PAST-be-IND he-PERF-sleep-IND 'He was in the state of having fallen asleep' = 'He was asleep' b. a-ta-kuw-a a-na-som-a he-FUT-be-IND he-PROGR-read-IND 'He will be in the state of being involved in reading' = 'He will be reading' Again it may be doubted whether constructions such as these can be properly called 'serial'. An alternative view is that the copula is exploited as a supportive device in order to express combinations of Tense/Aspect operators, while still retaining the principle that only one such operator can appear per verb. It remains true, however, that constructions such as (66 a —b) consist of two fully inflected verbs, which can each occur on their own in other constructions. In

Copula Auxiliarizatum:

How and Why?

73

this respect they differ from the combinations of copula + participle which result from developments along the Property channel.

4. Why does copula auxiliarization take place? Now that we have considered the different channels along which copular constructions can be integrated into the aspectual system of the verbal predicate, we can ask the question of why this process should take place in comparatively uniform ways in the history of quite different types of languages. An interesting hypothesis about this matter was suggested by Meillet (1909; 1912; 1920). This hypothesis can be formulated as follows: (67) (i) Aspectual forms tend to get reinterpreted as temporal forms; (ii) To the extent that (i) is the case, there is a need for new aspectual forms; (iii) These new forms are provided by periphrastic combinations with auxiliary verbs such as 'be', 'have', 'go', 'want' etc. Meillet is not entirely clear, however, about the exact cause-effect relations implied by hypothesis (67). In one interpretation, the changes implied by (67) could be represented in terms of a dragchain, as represented in: (68)

Aspect

Tense stage I

—•

Eo

E,

stage II

E,

stage III

According to this interpretation, an original aspectual form, E 1; is reinterpreted as a purely temporal form; this reinterpretation leaves a vacant aspectual position, which is then filled by the innovative form E 2 . This interpretation is suggested by Meillet's discussion (1912:141 ff.) of the reinterpretation of the Indo-European Perfect as either a simple past or a stative present:

74

Simon

Dik

Mais la forme, en disparaissant, laissait un vide. Car on eprouve le besoin de bien marquer Taction achevee dont on envisage le resultat. On y parvient le plus souvent par le groupement d'une forme nominale rattachee a un verbe accessoire. In this interpretation, then, the initial event is the reinterpretation of Ej as a purely temporal form, and the resulting gap in the system is the trigger for the appearance of the innovative form. Hypothesis (67), however, can also be interpreted in terms of a push-chain of events, as illustrated in: (69)

Aspect

Tense stage I

E, > F

F

E2

stage II Ei

stage III

In this interpretation, the first step is the creation of a new periphrastic form E2, which then allows Ej to be reinterpreted in a purely temporal sense. Again, some of Meillet's formulations seem to suggest this push-chain interpretation rather than the drag-chain interpretation represented in (68). See for instance his description of the replacement of the simple past form by the composite, originally aspectual periphrastic construction in a number of different languages (1909:154 ff.). Both the drag-chain and the push-chain interpretations leave certain questions unanswered. These questions can be formulated as follows: (70)

Drag-chain problem: If the expression of aspectual distinctions is so essential, why should an aspectual form be fully reinterpreted in terms of Tense before a new aspectual expression is available?

(71)

Push-chain problem: If the original aspectual form is still fully aspectual, why should a new aspectual form be created?

Copula Auxiliarization:

How and Why?

75

I believe that these problems might be overcome by adopting a third possible interpretation of hypothesis (67), in which drag-chain and push-chain elements combine into the following sequence of events: (72)

Aspect

-

Tense

Ei E,

stage I

Ei · · · · • · · · E E,t

stage II

.• E η 2 , EJ π . . . . . . . E] ρ >

E2

Et Ε,

stage III stage IV

According to this interpretation, an original aspectual expression Ei is partially reinterpreted as a temporal form. By partially I mean that it acquires purely temporal interpretations in certain contexts/ occurrences, while retaining its aspectual value in other contexts/ occurrences (stage II). In such a situation, the need arises for a new, unequivocal expression for the aspectual distinction (stage III). Once this expression E 2 has been introduced, Ei is free to shift fully to the temporal dimension (stage IV). I believe that (72) is the most plausible interpretation of hypothesis (67), since it avoids both the drag-chain problem (70) and the pushchain problem (71).15 Indeed, (72) can be taken to represent a dragpush-chain type of development. It is also compatible with the 'Functionality Principle' formulated by Bossuyt (1983), which says that languages cannot develop in such ways as to result in stages in which essential functions cannot be unambiguously expressed. This would be the case if the course of events would be as pictured in (68), in which the aspectual form is first reinterpreted as a temporal one, thus creating a void which is later filled in by the new aspectual expression.

4.1. Temporal reinterpretation of aspectual forms In any interpretation of hypothesis (67), the temporal reinterpretation of originally aspectual forms plays an important part, either as cause, or as effect, or as both cause and effect, of the introduction of

76

Simon Dik

innovative aspectual forms. There is, indeed, a vast amount of evidence for such a reinterpretation from quite different language families.16 How does the temporal reinterpretation of aspectual forms come about? I believe that two factors can be held responsible for this process: (i) (ii)

inductive generalization; semantic simplification.

Let us look at these two factors separately.

4.1.1. Inductive generalization I have interpreted Aspect as pertaining to the question: 'What can be said at reference points 1, 2, 3, ... in relation to the occurrence of a certain State of Affairs?'. Since the reference points 1, 2, 3, ... can in principle move freely along the time axis, and take any position with respect to the moment of speaking ms, Aspect is basically independent of Tense. However, both within and across languages, the following two principles appear to obtain: (73) (i) If the reference point for an aspectual form is specified, it is much more often specified as coinciding with moment of speaking than as lying in the Past or the Future, (ii) If the reference point for an aspectual form is not specified, it is commonly taken as coinciding with moment of speaking by default. Through these two principles, the reference point for aspectual forms tends to get anchored in the moment of speaking. Through inductive generalization this may lead to a situation in which only the moment of speaking may be taken as the reference point for an aspectual form. We then have a partial temporal reinterpretation of the originally aspectual form. Now consider what this means for the position of the relevant State of Affairs on the time axis: (74) (i) If the reference point for a (Recent) Perfect coincides with the moment of speaking, then the relevant State of Affairs lies in the Past.

Copula Auxiliarization:

How and Why?

77

(ii) If the reference point for a Progressive coincides with the moment of speaking, then the relevant State of Affairs lies in the Present. (iii) If the reference point for an (Immediate) Prospective coincides with the moment of speaking, then the relevant State of Affairs lies in the Future. Thus, the anchoring of aspectual reference points in the moment of speaking by implication places the relevant State of Affairs in the Past, the Present, and the Future, respectively. 4.1.2. Semantic simplification Even when the aspectual reference point coincides with the moment of speaking, there is still a semantic difference between the relevant Aspect and the corresponding simple Tense. Compare the Present Perfect with the simple Past: (75)

(76)

John kissed Mary 'At the moment of speaking it is stated that (a) the State of Affairs 'John kiss Mary' took place before the moment of speaking' John has kissed Mary 'At the moment of speaking it is stated that (a) John is such at the moment of speaking that (b) the State of Affairs 'John kiss Mary' took place before the moment of speaking'

From these paraphrases it is clear that the Perfect is semantically more complex than the simple Past even when both are anchored in the moment of speaking. The more complex meaning of the Perfect can now be simplified in two ways. First, the element (a) can be dropped, so that the original Perfect fully coincides with the simple Past. Second, the element (b) can be dropped, so that only the present state of John is involved, with no reference to the original State of Affairs which brought him into that state. In this latter case, the original Perfect coincides with a simple (stative) Present. In both ways, the original aspectual form is fully reinterpreted as a simple Tense form. In a similar way, semantic simplification would lead to a reinterpretaion of the Progressive as a simple Present tense, and of the (Immediate) Prospective as a simple future tense.

78

Simon

Dik

This brings us to our final question: if the above account of the temporal reinterpretation of originally aspectual forms is at all valid, how and why is it that semantic simplification comes about? Here, again, some suggestions of Meillet's may lead us to an interesting hypothesis.

4.2. Language acquisition a factor in semantic simplification? Meillet (1909:152) cites Stern & Stern (1907) for the statement that the German Imperfekt ( = simple Past) appears very late in the speech of children, and is of rare occurrence afterwards. This means that periphrastic forms such as Der Hund hat gebellt 'the dog has barked' are much more common in the speech of children than Der Hund bellte 'the dog barked'. Meillet connects this fact with the supposition that the periphrastic forms, being morphologically more perspicuous, are easier to handle ('plus commode a manier' — 1909:155) than the more irregular non-periphrastic forms. We could represent this situation in the following way: (77) simple Past periphrastic Perfect

form

meaning

more difficult

less complex

less difficult

>

^

more complex

Let us now suppose that children in acquiring the verbal system start out with the less difficult forms, associating these with the less complex meanings. This would mean that they wrongly connect form and meaning along the dotted line. If this type of connection were to be maintained into the adult language, then the semantic simplification of the original Perfect would, in fact, have been accomplished. The following facts on Dutch language acquisition would seem to support the type of development postulated here. From Schaerlaekens (1977) we can derive the following summary of the development of the verbal complex: until the age of about 2.6, Dutch children use almost all verbs in either stem or infinitive form, without any sign of

Copula Auxiliarization:

How and Why?

79

Tense/Aspect morphology. When tense distinctions become relevant, these are first expressed by periphrastic combinations of a limited number of auxiliaries plus infinitive of the lexical verb. This happens even when such combinations are ungrammatical in adult Dutch, a fact which suggests that children at this stage use the auxiliary + infinitive pattern as a general strategy for the expression of Tense distinctions. The initial Tense system of Dutch children can be represented as follows: (78)

(Recent) Past

hondje heb gebijt doggie has bitten

Present (Immediate) Future hondje bijt/bijten doggie bite hondje doet bijten hondje gaat bijten doggie does bite doggie goes bite

Note that at this stage there is no sign of the Past/Perfect opposition between beet 'bit' and heeft gebeten 'hae bitten'; that ί/o-support is ungrammatical in Dutch in all circumstances; and that the auxiliary gaan is used for the Future where standard Dutch would require zullen 'shall/will'. We can thus say that children at this stage use a system of periphrastic forms, some of their own making, in association with a simplified Tense system, in accordance with (77).17 Lalleman (1981; 1983) found a similar overgeneralization of periphrastic patterns in the speech of foreign workers learning Dutch. She relates this phenomenon to the complexities of Dutch word order patterns in which finite verbs (Vf) must take clause-second position, but non-finite verbs (Vi) take final position, thus 'embracing' the Object, as in: (79)

Subj. a. de hond the dog b. de hond the dog c. de hond the dog

Vf bijt bites heeft has gaat goes

Obj de man the man de man the man de man the man

Vi

gebeten bitten bijten bite

The general trend is that patterns such as (79 b —c) appear before patterns such as (79 a).

80

Simon

Dik

We may thus conclude that periphrastic patterns appear to be exploited both by first and second-language learners of Dutch, and that this is due to a desire for simplification in two respects: — with respect to form, since a limit number of inflected auxiliaries is sufficient for all purposes; — with respect to order, because when periphrastic forms are used all main clauses can be ordered according to the pattern Subj-VfObj-Vi. Periphrastic forms typically originate as aspectual expressions. The overgeneralization of periphrastic forms by language learners, and their association with a simplified Tense system, may thus be responsible for the temporal reinterpretation of aspectual distinctions.

5. Conclusion and summary By way of conclusion, I summarize the main theses of this paper: (i)

(ii)

(iii)

(iv)

Copula Auxiliarization is a less dramatic change than it is often assumed to be, since (a) the copula never has any independent meaning in the first place, so that no desemanticization is involved, (b) both the copula and the copular auxiliary can be treated as supportive verbs, to be inserted into predications under certain grammatical conditions. Copula Auxiliarization is rather a matter of Expansion ( = the acquisition of new grammatical functions by some grammatical element). Copula Auxiliarization originates in innovative aspectual forms, which may later be reinterpreted in terms of Tense or Voice. Copular constructions are especially suited for the expression of (Phasal) Aspect distinctions, because such distinctions ascribe a property to some entity Jt at some reference point r in relation to the occurrence of some State of Affairs involving x.

Copula

(v)

(vi)

(vii) (viii)

(ix)

Auxiliarization:

How and Why?

81

There are various channels along which Copula Auxiliarization may come about: (a) the Localist channel, based on the metaphor whereby a State of Affairs is conceptualized as a spatial object; (b) the Property channel, where the property of being involved in such and such a way in some State of Affairs is assigned to x; (c) possibly the Serial channel, although it is doubtful whether the relevant constructions can properly be called serial constructions. It is plausible that innovation of aspectual forms (and thus Copula Auxiliarization) comes about through the tendency of old aspectual forms to become reinterpreted in purely temporal terms. It is plausible that such innovation comes about in a combined drag-chain and push-chain fashion. The temporal reinterpretation of aspectual forms can be understood as due to (a) inductive generalization: if aspectual forms are mostly used in relation to the moment of speaking, inductive generalization may lead to a situation in which they can be only so used; (b) semantic simplification: aspectual forms, when used with reference to the moment of speaking, have the complex meaning: 'It is stated at the moment of speaking that (1) χ is such at the moment of speaking that (2) χ is involved in State of Affairs before/at/after the moment of speaking'. If (1) is dropped, the result is a purely temporal form; if (2) is dropped, the result is a simple Stative Present form. The semantic simplification indicated under (viii) (b) may be due to a tendency on the part of language learners to initially associate the formally less difficult periphrastic forms with the semantically less complex Tense system.

82

Simon Dik

Notes 1. Wezen, usually a stylistic variant of zijn, is in this combination the only possibility. 2. See Lehmann (1982) for an analysis of the notion of grammaticalization, and Heine & Reh (1982) for both theoretical analysis and exemplification from African languages. 3. For the aspectual origin of periphrastic forms in Romance, see Fleischman (1982), Harris (1982), Ramat (1982). 4. For the typology of States of Affairs, see Dik (1978), Vet (1980), Vester (1983), De Groot (1984). For the distinction + Telic, see Comrie (1976). 5. This view of Tense and Aspect is largely based on Comrie (1976; 1985). Compare also Vet (1980), Johnson (1981), Shiratsuki (1985). 6. Compare Comrie (1985), Johnson (1981). 7. For Phasal Aspect, compare Johnson (1981:152): 'Verb aspect involves reference to one of the temporally distinct phases in the evolution of an event through time'. Note that both Comrie (1976) and Johnson (1981) reject the distinction between Phasal and Quantificational Aspect as made here, on the ground that many languages have one formal category (the Imperfective) which covers both (or at least, Progressive and Habitual). 8. For the term Prospective see Anderson (1973), Comrie (1976), Fleischman (1982). Comrie gives the following appropriate definition: Ά state is related to some subsequent situation, such that the seeds of that subsequent situation are already present in the earlier state.' For terminological parallelism, one might prefer the term 'Retrospective' instead of Perfect (Anderson 1973; Fleischman 1982). Note that, if 'Perfect' is retained, it is to be understood as non-identical to 'Perfective' (cf. Comrie 1976). 9. These examples are not necessarily meant as sentences of English, but as paraphrases of expressions in languages in which the relevant distinction is coded grammatically. , 10. See also Comrie (1976:98 ff.). 11. Thus one may well say that desemanticization is involved in this process, but it is desemanticization of the locative predicate, not of the copula. 12. Claudi & Heine (1985:29) note that in some dialects of Ewe, expressions such as (46 b) can optionally be further simplified to (i) me-yi-yi m, or even (ii) me-yi-m. In these reduced forms, there is little that reminds one of the locative origin of the construction. 13. For the status of this construction and its alleged influence on the development of the progressive form, see Scheffer (1975). 14. Note that according to Pinkster (1983; 1984) the Latin Perfect (including the periphrastic passive and deponent forms) has largely been reinterpreted as a simple Past Tense. See section 4.1 below. 15. Suzanne Fleischman (pers. comm.) informs me that much of the relevant Romance data is interpretable in terms of (72). See also Fleischman (1982). 16. See the various studies of Meillet's cited above and Comrie (1976; 1981) for diverse examples of such a development. Examples from the Romance area are discussed in Fleischman (1982; 1983), Harris (1982), Ramat (1982). Heine & Reh (1982) mention various developments of this kind in African languages. Koefoed &

Copula Auxiliarization: How and Why?

83

Van Marie (1985) discuss a case of temporal reinterpretation of aspectual forms in Sranan Creole. 17. Schaerlaekens (1977:71) notes, however, that constructions with the auxiliaries doen 'do' and gaan 'go' as exemplified in (76) are also characteristic of the 'motherese' spoken to children, e.g. in picture book reading. This means that there is probably an interaction between learner and teacher, where both use the auxiliary + non-finite verb combination as an easy entry into the verbal system.

References An essay concerning aspect. The Hague: Mouton. Swahili grammar, including intonation. London: Longman. AWBERY, G. M. 1976. The syntax of Welsh; a transformational study of the passive. Cambridge: Cambridge University Press. ANDERSON, JOHN. 1 9 7 3 . ASHTON, E . O . 1 9 4 7 .

1983. Historical Functional Grammar: an outline of an integrated theory of language change, in: Advances in Functional Grammar, ed. by Simon C. Dik, 301-325. Dordrecht: Foris. CLAUDI, ULRIKE, and BERND HEINE. 1 9 8 5 . From metaphor to grammar: some examples from Ewe. Afrikanische Arbeitspapiere (Köln) 1:17 — 54. COMRIE, BERNARD. 1 9 7 6 . Aspect. Cambridge: Cambridge University Press. COMRIE, BERNARD. 1 9 8 1 . Aspect and voice: some reflections on Perfect and Passive, in: Tedeschi & Zaenen (eds.) 1 9 8 1 : 6 5 - 7 8 . COMRIE, BERNARD. 1 9 8 5 . Tense. Cambridge: Cambridge University Press. BOSSUYT, ALAIN.

D I K , SIMON C . 1 9 7 8 .

Functional Grammar. Amsterdam: North-Holland

(19813,

Dor-

drecht: Foris). Studies in Functional Grammar. London: Academic Press. C. 1983. Auxiliary and copula be in a Functional Grammar of English, in: Linguistic categories: auxiliaries and related puzzles, ed. by Frank Heny & Barry Richards, 121-143. Dordrecht: Reidel.

D I K , SIMON C . 1 9 8 0 . D I K , SIMON

The future in thought and language: diachronic evidence from Romance. Cambridge: Cambridge University Press. FLEISCHMAN, SUZANNE. 1 9 8 3 . From pragmatics to grammar: diachronic reflections on complex pasts and futures in Romance. Lingua 60:183 — 214. FLEISCHMAN, SUZANNE. 1 9 8 2 .

GROOT, CASPER DE. 1 9 8 3 .

lands 1 9 8 3 , ed. by Foris.

H.

Typology of states of affairs, in: Linguistics in the NetherBennis & W . U . S . van Lessen Kloeke, 7 3 - 8 1 . Dordrecht:

B. 1981. The 'past simple' and 'present perfect' in Romance, in: Studies in the Romance verb, ed. by Martin B. Harris & Nigel Vincent, 42 — 70. London: Croom Helm. HEINE, BERND, and MECHTILD R E H . 1 9 8 2 . Patterns of grammaticalization in African languages. AKUP: Arbeiten des Kölner Universalienprojekts 47. JOHNSON, MARION R . 1 9 8 1 . A unified temporal theory of Tense and Aspect, in: Tedeschi & Zaenen (eds.) 1 9 8 1 : 1 4 5 - 1 7 5 . KOEFOED, G. A. T., and J. VAN MARLE. 1985. Requisites to reinterpretation. Paper, University of Utrecht/University of Amsterdam. HARRIS, MARTIN

84

Simon Dik

LALLEMAN, JOSINE. 1981. The F G view on word order: data from spoken Dutch of foreign workers, in: Perspectives on Functional Grammar, ed. by T. Hoekstra et al„ 1 5 7 - 1 7 3 . Dordrecht: Foris. LALLEMAN, JOSINE. 1983. The principle of elimination: establishing word order regularities in the Dutch of foreign workers. LBer. 87:40 — 63. LEHMANN, CHRISTIAN. 1982. Thoughts on grammaticalization; a programmatic sketch, vol. I. AKUP: Arbeiten des Kölner Universalienprojekts 48. MEILLET, ANTOINE. 1909. Sur la disparation des formes simples du preterit, in: Meillet 1982:149-158.

MEILLET, ANTOINE.

1912.

Devolution

des

formes grammaticales,

in:

Meillet

1982:130-148.

MEILLET, ANTOINE. 1920. S u r les c a r a c t e r e s d u verbe, in: Meillet 1 9 8 2 : 1 7 5 — 198.

MEILLET, ANTOINE. 1982. Linguistique historique et linguistique generale. Geneve: Slatkine; Paris: Champion. PINKSTER, HARM. 1983. Tempus, Aspect and Aktionsart in Latin (recent trends, 1961 — 1981), in: A N R W 2 9 / 1 : 2 7 0 - 3 1 9 . Berlin: D e G r u y t e r .

PINKSTER, HARM. 1984. Latijnse syntaxis en semantiek. Amsterdam: B. R. Grüner. RAMAT, PAOLO. 1982. Ein Beispiel von 'reanalysis', typologisch betrachtet. FoL 16: 3 6 5 - 3 8 3 .

SCHAERLAEKENS, ANNA M . 1977. D e t a a l o n t w i k k e l i n g van het kind; een orientatie in

het Nederlands-talig onderzoek. Groningen: Wolters-Noordhoff. SCHEFFER, J. 1975. The Progressive in English. Amsterdam: North-Holland. SHIRATSUKI, NORIKO. 1985. Two types of resultative in Basque and their non-aspectual meanings, in: Predicates and terms in Functional Grammar, ed. by A. Machelt Bolkestein et al., 1 8 3 - 2 0 2 . STERN, CLARA, a n d WILHELM STERN. 1907. D i e K i n d e r s p r a c h e : eine p s y c h o l o g i s c h e

und sprachtheoretische Untersuchung. Leipzig: Barth. TTEDESCHI, PHILIP, a n d ANNIE ZAENEN (eds.). 1981. Tense a n d A s p e c t . S y n t a x a n d

Semantics 14. New York: Academic Press. VESTER, ELSELINE. 1983. Instrument and Manner expressions in Latin. Assen: Van Gorcum. VET, J. P. 1980. Temps, aspects et adverbes de temps en frangais contemporain; essai de semantique formelle. Geneve: Droz.

Simon C. Dik Instituut voor Algemene Taalwetenschap Universiteit van Amsterdam Spuistraat 210 NL = 1000 G D Amsterdam

Auxiliaries and Ergative Splits: A Typological Parameter1 Eloise Jelinek

The development of copular auxiliary verbs from pronominal clitics or affixes has been well documented in many language families — Ancient Chinese, Semitic, and several Native American groups (Li & Thompson 1977). A number of languages, particularly in Africa, have developed negative copular verbs from the fusion of pronominal and N E G elements. Ken Hale (1973) first noted the many parallels in structure and function between the second position AUX clitic cluster (which is almost universal in aboriginal Australia, and in many Native American language families) and auxiliary verbs in language families such as Indo-European. This second position AUX clitic cluster contains elements marking the person and number of the verbal arguments, along with other inflectional categories such as tense, aspect, mood, modality, and sentential polarity or negation. Auxiliary verbs or clitic clusters typically occur with participle-like verbal forms inflected for aspect, voice, or mood. In Indo-European, it is ordinarily just the subject that is marked in the auxiliary verb, as in Span, estoy cantando, he cantado. But in many language families, object arguments may also be marked in the verbal inflection or in AUX, as in Basque, or in the aboriginal languages of Australia and North America that I am primarily concerned with here. The marking of all the verbal arguments in AUX or in the verbal inflection has profound consequences for the syntax. Frequent among these is the presence of ergative features. The dependency between the marking of grammatical relations in the inflectional morphology and the distribution of ergative features across languages will be the focus of this paper. Attempts at assimilating the phenomena ergativity into a theory of universal or core grammar have been faced with the wide diversity of languages that have been termed 'ergative'. It has frequently been noted in the literature on ergativity (cf. Dixon 1979) that languages that exhibit ergative features are rarely, if ever, 'wholly' ergative; that

86

Eloise

Jelinek

is, some morphological or syntactic components of the grammar can be seen as operating on the more familiar accusative pattern. Mixed ergative-accusative patterns within a language are commonly referred to as 'ergative splits'. In another sense, splits may be said to occur within a family of closely related languages; that is, the members of a language family may differ as to their type or 'degree' of ergativity. Anderson (1977) points out that this variation across related languages indicates that languages may become or cease being ergative, and that the study of historical change in such families should shed light on the dependencies between ergativity and other aspects of a grammar, and on what the mechanisms of such changes may be. Within a language, ergative splits may be conditioned by clause type (main vs. subordinate, etc.), by verb type, by tense/aspect or by semantic features of the verbal arguments; this latter is generally referred to as an 'NP split'. Dixon (1979) stresses that the evidence on historical change now available does not suggest that any single kind of split may underlie the numerous and varied kinds of ergative patterns found across languages. Evidence appears to be accumulating to the effect that for some types of ergative systems, historical origins may be clearly traced; while for other kinds of ergativity, origins are obscure, and conflicting claims as to the nature of historical change have been advanced. Historical connections between passive and ergative constructions have frequently been suggested. Chung (1976; 1977) has documented the development of ergative constructions from passives in Polynesian. Chung notes that in contemporary Pukapukan, accusative, passive, and ergative constructions currently exist side-by-side, and the speaker's choice among them relates to his evaluation of the speech situation along a scale ranging from polite/formal (ACC) to casual/slightly improper (ERG). This would appear to demonstrate change in progress. Across languages, passives are frequently restricted to certain tense/aspects. Dixon (1979:99) notes a clear association between ergativity of a passive origin and tense/aspect splits, and cites Anderson's (1977) observation that 'passives are semantically close to perfect in that they generally present a state resulting from a completed action.' Ergatives in Indo-Iranian are clearly derived from earlier perfective aspect forms; see discussion in Trask (1977). It seems clear then that some ergative constructions arise from passives or perfectives, and that these ergatives are predominantly

Auxiliaries

and Ergative Splits

87

tense/aspect conditioned. But, as Dixon notes, very little is known about the origins of ergative patterns showing NP splits. In fact, it is just with regard to languages with this type of ergative split that the historical evidence seems contradictory and confusing. In NP splits, the use of an ergative construction is not determined by utterance context or by features such as verb, aspect, or clause type, but rather by the kind of referential elements that appear in the sentence. Silverstein (1976) claimed that NP splits follow a 'hierarchy of [semantic] features' such that elements higher on a scale of agency or animacy will appear in accusative constructions, while lower ranked elements will trigger ergativity. NP splits occur in many areas including Asia and the Americas, but they are almost universal in the Pama-Nyungan family of languages in Australia. Problems arise with explaining these NP splits with reference to a Silverstein hierarchy, as I will show, and problems arise also in connection with the interpretation of the evidence on historical change. Opposing claims have been advanced concerning the direction of diachronic change in Australia, ERG -»• ACC or ACC ERG. Early historical speculations assumed that a change from ERG to ACC in Australian languages was a kind of evolutionary progress towards an Indo-European grammatical type. Hale (1970) argued that the direction of change was the reverse, ACC ERG, and that ERG constructions evolved from passives, by means of a passive rule that became obligatory. Blake (1977) comments that recent change from ERG to ACC has been documented, and adds that it is his impression that ergativity is on the decline in Australia at present. We shall turn to recent evidence on this point in sec. 3. Dixon (1977) reconstructs both ERG and ACC case in Proto-Australian. The reconstructed agentive case marker seems incontestable, and it seems equally certain that ACC marking was also present on some pronouns. This lends support to neither of the claimed directions of historical change, ERG ACC or ACC -»· ERG, and suggests that an NP split was present in Proto-Australian. I will argue that Proto-Australian was accusative, and I will propose mechanisms whereby both the morphological ergativity that is common in the Pama-Nyungan family and the syntactic ergativity that is rarely found (Dyirbal is a primary example of a language with syntactic ergativity) can have developed from Proto-Australian. These claims will rest on the identification of particular distinctive aspects of the syntax of contemporary Pama-Nyungan languages — specifi-

88

Eloise

Jelinek

cally, the constraints upon permitted argument types — and upon evidence from current change in Australian languages as they succumb to pressure from contact with English. In short, I suggest that certain typological features of Proto-Australian can resolve the puzzle of the conflicting historical evidence, can account for the emergence of ergative features, and can also resolve certain other 'puzzles' that have often been identified in connection with ergativity. These include: 1) What is the origin of the NP splits? 2) Why is ergativity usually so 'shallow' (with the exception of Dyirbal and marginally, a few others)? 3) Why is ERG case invariably homophonous with or closely related to some other case in Australian languages? 4) Why do NPs often receive 'double' case marking? 5) And finally, why are passives rarely found in the Pama-Nyungan languages?

1.

A typological parameter: languages with pronominal vs. languages with lexical arguments

There is a profound typological distinction among languages: those that place constraints upon the class of elements that may serve as arguments, and those that do not. A language may permit a verb (or other predicate that determines the argument structure of a clause) to take only elements that mark just the simple indexical features of person and number (and occasionally, gender) as their arguments. Navajo is an example: (1)

yiyiilts4 3 A CC-3NOM-saw 'He saw him'.

Or a language may permit predicates to take full lexical items with any semantic feature whatever to serve as their arguments. English has no constraints upon argument types, and permits both lexical and pronominal arguments. (2) a. b.

He saw him. David saw Goliath.

This typological contrast may be stated as follows:

Auxiliaries

(3) a. b.

and Ergative Splits

89

The Argument Type Parameter Pronominal Argument languages permit only pronouns and anaphors as arguments. Lexical Argument languages do not restrict arguments; lexical arguments are allowed.

Sentences may serve as arguments to higher verbs in Lexical Argument languages. Pronominal Argument languages lack embedded clauses. Clausal adjunction, with coreference of pronominal arguments across clauses, is found instead. Pronominal Argument languages permit, under specified conditions, optional adjoined nominals that amplify the information (person and number) given on the referents of the pronominal arguments. (4) a.

b.

c.

d.

Ashkii yiyiilts4 boy 3 A CC-3NOM-saw 'He/she saw the boy'. At'eed biilts^ girl bi-3A CC-3-NOM-saw 'The girl saw him/her'. At'eed ashkii yiyiilts4 girl boy 3ACC-3NOM-saw 'The girl saw the boy'. At'eed ashkii biilts4 girl boy bi-3ACC-3NOM-saw 'The boy saw the girl'.

The change in the yi-/bi- pronominal prefix in these examples indicates whether the preceding nominal is an adjunct to the agent or patient pronominal argument (Sandoval & Jelinek in press). Examples from Warlpiri, a Pama-Nyungan language of Australia, are as follows (Hale 1983:6-7). (5) a.

b.

c.

Panti-rni ka spear-NONPAST PRES-3sg.NOM-3sg.ACC 'He/she is spearing it'. Wawirri ka panti-rni kangaroo PRES-3sg.NOM-3sg.ACC spear- NO N3 A ST 'He/she is spearing the kangaroo'. Ngarrka-ngku ka panti-rni man-ERG PRES-3sg.NOM-3sg.ACC spear-NONPAST 'The man is spearing him/her/it'.

90

Eloise

d.

Jelinek

Ngarrka-ngku ka man-ERG PRES-3sg.NOM-3sg.ACC panti-rni spear-NONPAST 'The man is spearing the kangaroo'.

wawirri kangaroo

Word order in Warlpiri is free except for AUX, in second position. Pronominal Argument languages permit nominals to be adjoined to the pronominal arguments via rules that refer to the subcategorization of the verb. (Adjunction rules for Warlpiri are given in Jelinek 1984; for Apachean (Navajo) in Sandoval & Jelinek in press). Since nominal adjuncts are optional, no PRO or other empty category is required in the analysis of (1), (4 a, b) or (5 a, b,c). There is nothing 'missing' in these sentences since the argument positions of the verb are filled by the verbal prefixes or AUX elements, in accordance with the Projection Principle (Chomsky 1982); the speaker has simply failed to exercise his prerogative to adjoin all the permitted nominals, since he is confident that in context his hearer will not mistake his intended reference. A frequently encountered, but non-definitional attribute of Pronominal Argument languages, one which does not follow from the Argument Type Parameter itself, is that in these languages, third person (singular) pronominal is often phonologically null. There are languages that have only pronominal arguments that do have an overt third person pronoun; as in the bi prefix in the Navajo example sentence (4 b) above. However, if a feature is necessarily marked in a construction, and if that feature is marked by the mutually exclusive members of a paradigmatic set, then one member of that set is often ZERO or phonologically null, and the interpretation, as Hale observes, is by default. If no other member of the pronominal paradigm is present, there is no ambiguity; a fixed third person singular interpretation is present. Since the set of permitted argument types is so constrained in these languages, ZERO arguments are permitted. Note that third person dual and plural are overt: (6)

Warlpiri AUX clitics (Hale 1973:315-316, 328) NOMINATIVE -rna -n(pa) -rlijarra

ACCUSATIVE -ju -ngku -jarangku

1st sg. 2nd sg. 1st du.

Auxiliaries

-rli -n(pa)-pala -rna-lu -rlipa -nku-lu ZERO -pala -lu

and Ergative Splits

-ngalingku ( ~ -ngali) -ngku-pala -nganpa -ngalpa -nyarra ( ~ -nyurra) ZERO -palangu -jana

91

1st & 2nd du. 2nd du. 1st pi. 1st & 2nd pi. 2nd pi. 3rd sg. 3rd du. 3rd pi.

Phonologically null third person marking is also found in Lummi, a Pronominal Argument language that belongs to the Coast Salish group of the American Northwest. (7)

(8)

xci-t-0-s know- TRANS-3ABS-3ERG 'He knows him'. ye ? -0 go-3ABS 'He left'. (Jelinek & Demers 1983:316)

Example (8) shows a predicate with no added phonological material, that is interpreted only as a finite declarative sentence with a third person argument. 2 Evidence given in Jelinek (1984) for classifying nominals in Pronominal Argument languages like Warlpiri, Navajo and Lummi as adjoined nominals rather than arguments bearing grammatical relations may be recapitulated briefly here.

1.1. Sentential constituency Nominals in these languages are optional, while predicates never appear in finite clauses without the affixes or AUX elements that mark person, number, and grammatical case. An excellent report on Walmatjari (Hudson 1978) contains the following examples (renumbered). (9)

Although many examples in this paper contain several phrases, this is not typical of the language. Because of the person-number information contained in the verbal auxiliary

92

Eloise

a.

b.

c.

Jelinek

the clause often has only two constituents, the verb and the verbal auxiliary. nganpayi-wu ma-rna-rla jularni wangki-0 man-DAT MRl-lsgS-3sgDAT told word-ABS ngaju-ngu I-ERG Ί told the man something'. Sentence a) is most likely to be heard as: jularni ma-rna-rla Ί told him'. For emphasis one phrase may be included: jularni ma-rna-rla ngaju-ngu Ί told him'. (Hudson 1978:17)

The independent pronoun is included in (9 c) for contrastive referential emphasis.

1.2. Absence of grammatical case on nominals In Navajo, nominals, including free pronouns, carry no case marking, as opposed to the verbal pronominal prefixes, which mark grammatical case. Nominals may be adjuncts to a verb or to an inflected postposition (possessive, locative, directional, etc.). In Lummi nominals carry only oblique case, and the AUX clitics and predicate suffixes are the pronominal elements that carry the grammatical relations. In Warlpiri, the AUX clitic paradigms mark the grammatical cases (NOM/ACC/DAT) and nominals carry lexical case. DAT is both a grammatical and lexical case. ERG/ABS/DAT case on nominals permit them to be co-indexed (interpreted as co-referential) with AUX clitics; other lexical cases (Allative, Elative, Perlative, etc.) cannot be so co-indexed.

1.3. Constraints on adjunction In Pronominal Argument languages there are certain predicates (weather, modals) that exclude nominal adjuncts to their pronominal

Auxiliaries

and Ergative Splits

93

subjects; that is, there are no pleonastic subjects. Reflexive and reciprocal anaphors also commonly exclude adjuncts. Other constraints are language particular. In Navajo there is a "fourth" person verbal inflection, the obviative, used to refer to someone the speaker does not wish to name, that is so specific in reference that it excludes adjuncts (Willie, in press). (10) a. Tom jiztal 3-4-kicked Ά certain person kicked Tom', b. Tom hwiztal 4-3-kicked 'Tom kicked a certain person'. Fourth person subject is marked by ji, and fourth person object is marked by ho. A second nominal may be added to utterances such as these only if a pause and falling contour indicate that they are outside the main clause: (11) a. dine, Tom 'The man, b. dine, Tom 'The man,

1.4.

jiztal he kicked Tom', hwiztal Tom kicked him'.

Specialized function of free pronouns as adjuncts

Independent pronouns in Pronominal Argument languages are adjuncts that generally appear only in contrastive referential emphasis, as in example (9 c) above. A Navajo example: (12)

(13)

ni niish'i you 2nd sg.-lst sg.-see Ί see you', shi yishaäl I 1st sg.-walk 'I'm walking'.

These sentences are marked constructions, as compared to those without the emphatic free pronouns. 3

94

Eloise

Jelinek

1.5. Absence of 'agreement' In Navajo, nouns (except for a half-dozen referring to human beings) do not mark number, just as they do not mark case. In Lummi also nominals often do not mark number; nominals are derived expressions based on predicates that may be reduplicated to show plurality of action or actors. In Warlpiri, nominals need not agree in person or number (and of course, case) with the AUX clitic to which they are co-indexed. The following example is from Hale (1983): (14)

Puyukuyuku-puru, kula-lpa-rlipa-nyanu fog- WHILE NEG-IMPERF- 1st pi. (INC) NOM-REFL yapa-0 nya-ngkarla person-A BSsee-IRREA LIS 'We (plural inclusive) cannot see one another (as) person(s) (i. e., our shapes or figures) when it is foggy'.

In this example, the nominal yapa 'person', is third person singular ABS, and is coindexed with the anaphor -rlipa, which is first person (INC) plural ACC. There is no agreement in case, person or number. Janet Sharp provides the following example from Manyjilyjarra: (15)

paa-rninpa-rna-0 cook-PRES CONT-lst sg.NOM-3rd sg.ACC ngurra-ngka-rlu camp-LOC-ERG Ί am cooking the damper in camp'.

tampa-0 damper-ABS

This example shows 'double' case marking, to be described in section 2.5. below. Ν gurr a 'camp' has LOCATIVE case followed by ERGATIVE, to indicate co-indexing with the first person singular subject clitic; 'agreement' is absent. In non-Pronominal Argument languages, nominals are arguments and none of the characteristics outlined here are present: nouns are necessary constituents of sentences; they carry grammatical case; pleonastic subjects occur; agreement may be present (as in English third person singular -s) and free pronouns have other uses than simply that of referential contrast.

Auxiliaries

and Ergative Splits

95

2. Problems and puzzles I turn now to a consideration of what this understanding of the Pama-Nyungan languages as placing constraints on argument types can contribute to the solution of the 'puzzles' cited above in connection with ergativity in these languages.

2.1. The origin of NP splits Considerable attention has been given to the problem of accounting for the NP splits that are universal in Pama-Nyungan and found frequently also in Asia and the Americas. The explanations for these splits that have been offered heretofore rest on semantic or pragmatic factors. Silverstein (1976) made the significant observation that in languages with NP splits, referential items may be ranked along the following scale: (16)

1 > 2 > 3 > proper > human > animate > inanimate

(First and second person may fall together, or 2 may outrank 1.) In a particular language, ergativity extends leftward to some point along the scale. (Finer gradations reflecting number, etc. may be present.) Silverstein's explanation for this hierarchy was based on the speaker's and hearer's expectations as to agency; elements higher on the scale were progressively more likely to be agents than lower-ranked ones. Mallinson & Blake (1981) presented text counts (including aboriginal Australian texts) that showed that no association between person and agency was present, and argued that the feature underlying the NP hierarchy was 'topic-worthiness'. DeLancey (1980) claims that the hierarchy can be explained in terms of 'attention flow.' Comrie (1981) emphasizes that attempts to account for NP splits in terms of agency/animacy/topicality, etc. are irrelevant to the kind of split found in Warlpiri, for example, where the AUX clitics are NOM/ACC and the independent pronouns are ERG/ABS. This split, Comrie notes, is 'orthogonal' to the Silverstein hierarchy, since the clitics and free pronouns mark the same semantic features of person and number. It is highly significant that there are no counterexamples to this distribution; no languages where clitics or pronominal affixes are ERG/ABS, while free pronouns are NOM/ACC, just as there are

96

Eloise

Jelinek

no cases where free or bound pronominal paradigms are ERG/ABS while other nominals are NOM/ACC.

2.1.1. A referentiality scale It should be emphasized here that there is a clear association between such semantic and pragmatic factors as agency, animacy, topicality, attention flow, 'old' vs. 'new' information, etc., and pronouns, as opposed to other NPs. But a more central aspect of this ranking of pronouns over other NPs is that of referentiality. In the utterance context, first and second person, speaker and hearer, are uniquely referential in the sentence; third person pronominal is assigned reference in the discourse. Proper nouns and kin terms are also given fixed reference in context. As we go down the NP scale, the specificity of reference diminishes, and ambiguity is increasingly likely; when a new referent is introduced, it must be picked out by the use of some descriptive predicate (or a non-verbal deictic gesture). We make reference more specific by saying 'Jane Brown, not Jane Smith', 'my son the doctor', 'the distant kangaroo', etc. In Pama-Nyungan there are 'discontinuous expressions' where nominals that form a semantic unit of some kind are separated in the sentence, since their case-marking shows their common link to an AUX argument. The following example is from Walmatjari (adapted from Hudson 1978:26). (17)

kunyarr-warnti-rlu pa-lu-nya dog-PL-ERG MR1-3PLS-3PL0 karnanganyja-warnti-0 pajani malji-warnti-0 emu-PL-ABS bit male-PL-ABS purlka-warnti-rlu ngaju-kura-warnti-rlu big-PL-ERG I-POSS-PL-ERG 'My big dogs bit (caught) the male emus'.

In this example, each nominal adjunct to the third person plural AUX clitic makes the reference increasingly explicit: 'dogs', then 'big' then 'my' so that the order of the dispersed adjuncts is precisely the reverse of the order in the English NP 'my big dogs'. Similarly, 'emus' precedes 'male'.

Auxiliaries

and Ergative Splits

97

2.1.2. The syntactic basis of the NP split The referentiality scale reflects a complex of semantic/pragmatic features that are language universal, but not all languages have an NP split. The point is that these semantic/pragmatic factors find varying syntactic expression across languages. In Pronominal Argument languages, the referentiality scale finds formal expression in the constraint placed on argument types: only pronominal elements with simple deictic reference may serve as arguments, and nominals serve as adjoined predicates. This is the syntactic basis of the NP split: arguments have NOM/ACC case, and nominals have lexical ERG/ ABS case. We noted earlier that a feature often encountered in Pronominal Argument languages is phonologically null third person. Among the consequences of this feature often documented in Pama-Nyungan is the development of overt third person pronouns from demonstratives and determiners. These 'borrowed' third person pronouns may retain the ERG/ABS case that reflects their previous non-argumental function. In such instances, the NP split divides first and second person from third person and other nominals. So far as I have been able to determine, 'NP ergative splits' are found only in Pronominal Argument languages. The crucial feature in Pronominal Argument languages is the absence of an asymmetry in argument structure of the kind found in Lexical Argument languages, where objects are lexical items under a VP node. Typically all the verbal arguments in a Pronominal Argument language are marked in the same word or constituent, the verb or AUX; and even where they are distributed across the verb and AUX, all arguments are present within the inflectional morphology, not in separate constituents. Therefore, Pronominal Argument languages lack the basic subject/predicate division in syntactic structure common to both transitive and intransitive sentences that creates pressure for the grouping of transitive and intransitive subjects into a single Nominative case, as opposed to Accusative and Oblique case for objects. Pronominal Argument languages are non-configurational, and have syntactic structures that are equally compatible with either NOM/ACC case or ERG/ABS case marking on the pronominal arguments. In these languages, transitive and intransitive subjects do not-fall together in a syntactic constituent [NP, S] that is distinct from the verb-AUX.

98

2.2.

Eloise

Jelinek

Case homophony and 'double' case marking in Australia

A striking feature of the Pama-Nyungan family is the fact that ergative case is always homophonous with or clearly related to some other lexical case in the language. Ergative case is often the same as Instrumental, or some locative 'by' case. Nash (1980) cites the following example from Hale (1966 fieldnotes): (18)

pulyku-ngku ka-lu wawirri-jangka-rlu sinew-ERG PRES-3rdpl.-3rd sg. 'roo-ELAT-ERG wari-ni tie-NO ΝΡ AST 'They tie it [hook] with sinew from a kangaroo'.

Here pulyku is marked ERG-INST. Wawirri provides an excellent example of what is meant by 'double' case; it carries Elative case (sinew from the animal) followed by ERG-INST case. Neither of these nominale is identical in reference to the third person AUX clitic. Another example from Nash: (19)

kurdu-ngku 0-ZERO-ZERO maliki-0 child-ERG PAST-3rdsg.NOM-3rdsg.ACC dog-A BS ngurra-kurra-[rlu] wajirli-pu-ngu camp-ALLATIVE-fERG] chase-PAST 'The child chased the dog [all the way] to camp'.

In this example, when ERG case follows ALLATIVE case on ngurra, the interpretation is that the child as well as the dog is moving towards the camp. When ERG case is not marked on the directional phrase, the interpretation is that the child caused the dog to move but remained behind. In Walmatjari, nominals interpreted as instruments are marked differently according to whether they are alienably or inalienably possessed. If the instrument is an inalienable possession, ERG case is marked on the noun. (20)

kunyarr-0 pa-0-0 pinya nganpayi-rlu dog-ABS MRl-3rd sg.S-3rd sg.O hit man-ERG kurrapa-rlu hand-ERG 'The man hit the dog with his hand'.

Auxiliaries

and Ergative Splits

99

However, when the instrument is an alienable possession, the Comitative case suffix -jarti appears before the ERG case marker. (21)

kunyarr-0 pa-0-0 piny a nganpayi-rlu dog-ABS MRl-3rd sg.S-3rd sg.O hit man-ERG mana-jawu-rlu stick-COM-ERG 'The man hit the dog with a stick'.

Examples such as these, adapted from Hudson (1978), demonstrate that an ergative marked nominal need not be coreferential with an AUX clitic, but simply associated with it. This association need not be one of instrumentality: (22)

ngarrka-ngku ka-ZERO-ZERO man-ERG PRES-3rd sg.NOM-3rd sg.A CC jarnti-rni karli-0 ngurra-ngka-rlu trim-NONPAST boomerang-ABS camp-LOC-ERG 'The man is trimming the boomerang in camp'. Simpson & Bresnan (1983: 51)

Here is an exuberant example of ergative marking in Walmatjari: (23)

yinya pa-ja-lu purrku-warnti-rlu gave MRl-lst sg.O-3rdpl.S old:man-PL-ERG kariampal-jawu-rlu west-COM-ERG mangarla-jawu-rlu karajarri-jawu-rlu NAME-COM-ERG NAME-COM-ERG ngaju-kura-rlu jarntu-warnti-rlu I-POSS-ERG countryman-PL-ERG 'The old men from the west, who speak Mangarla and Karadjeri and who are my countrymen gave it to me'. Hudson (1978:18)

Examples such as these show clearly that efgative is a lexical, not an argumental grammatical case in Pama-Nyungan.

2.3. The syntactic 'shallowness' of ergativity Comrie (1978) identifies the syntactic 'shallowness' of ergativity as the 'real puzzle' and Blake (1976) emphasizes that (aside from Dyirbal, to

100

Eloise

Jelinek

which we will return) ergativity is generally 'superficial'. Where ERG/ ABS case serves simply to coindex nominals with AUX arguments, we would not expect syntactic principles and processes to show ergative patterning. For example, the phenomena that have been labeled 'control' in Warlpiri (Simpson & Bresnan 1983) operate on an accusative basis, not an ergative one. Infinitival clauses in Warlpiri are complex nominals that carry elements generally labeled COMP that signal whether the infinitival is to be co-indexed with the subject or the object of the matrix clause; thus 'control', or as it has been otherwise described in Australia, switch reference (Austin 1981). Compare the following, adapted from Simpson & Bresnan (1983): (24)

Ngarrka-ngku ka-0-0 man-ERG PRES-3rd sg.NOMSrd sg.A CC purlapa-0 yunpa-rni [karli-0 corroboree-ABS sing-NPAST boomerang-ABS jarnti-rninja-^arra-rlu] trim-INF-LOC-ERG 'The man is singing a corroboree, while trimming a boomerang'.

(25)

kurdu-ngku ka-0-0 karnta-0 child-ERG PRES-3rd sg.NOMSrd sg.ACC woman-ABS nya-nyi [ngurlu-0 yurpa-rninja-A:wrra-0] see-NPAST seed-ABS grind-INF-LOC-ABS 'The child sees the woman grind mulga seed'.

Warlpiri infinitival clauses are case-marked just as any other nominal expression is, to provide for coindexing with an AUX pronominal. The case suffixes karra and kurra are locative lexical cases that mark the preceding non-finite verb as oblique adjuncts; these locative phrases are then 'double' case marked ERG or ABS to provide for co-indexing with an AUX clitic. There is an interesting parallel between karra/kurra and the archaic English a', said to derive from locative at? (26)

The man a' trimming a boomerang is singing a corroboree. The child sees the woman a' grinding mulga seed.

The elements in an infinitival expression may be discontinuous, as with any other complex nominal (adapted from Nash 1980):

Auxiliaries

(27)

Ngarrka-ngku man-ERG marlu-0 kangaroo-ABS 'The man shot

and Ergative Splits

101

ZERO-0-0 PASTSrd sg.NOM-3rd sg.ACC marna-kurra luwa-rnu nga-rninja-fcMrra grass-LOC shoot-PAST eat-INF-LOC the kangaroo eating the grass'.

An example of a discontinuous purposive expression from Walmatjari (Hudson 1978): (28)

kuyi-wu pa-0 paryani mana-nga animal-PURP MRl-3rd sg.NOM climbed Iree-LOC wanyjirri-wu lan-u-purru kangaroo-PURP spear-NOMINALIZER-PURP 'He climbed a tree to spear a kangaroo'.

(Purposive case in Walmatjari has the variants wu and purru.) In this discontinuous expression, the general common noun kuyi 'animal' precedes the more restrictive wanyjirri 'kangaroo'. The fact that the elements of a non-finite expression can be interleaved with other sentential constituents in 'discontinuous' expressions is evidence that the syntactic status of infinitival clauses is the same as that of other nominals, and that with both nominals and infinitivals, coindexing with AUX arguments as signaled by case particles is the basis of coreference.

2.4. Passives in pronominal argument languages In a Pronominal Argument language, the position of a nominal in the clause does not reflect its grammatical relation, since nominals do not carry grammatical relations. Therefore, NP movement rules such as PASSIVE do not apply.5 A lexical passive (an intransitive construction marked passive in the verbal morphology) is found in some Pronominal Argument languages such as Navajo. But there would seem to be little pragmatic motivation for a passive in a language like Warlpiri, where a speaker may place focus or emphasis on either a patient or an agent by placing a nominal adjunct to the verbal argument in sentence initial position, the focus position before AUX. Or an AUX subject or object argument may be focused by adjoining the independent pronouns that are used for emphasis.

102

Eloise

Jelinek

3. Dyirbal and historical change in Pama-Nyungan We turn now to the interesting case of Dyirbal, a language apparently unique in the Pama-Nyungan family in its 'degree' or syntactic depth of ergativity. Dyirbal has an anti-passive construction that may be informally described as converting a transitive clause into a derived intransitive, with an underlying agent appearing in ABSOLUTIVE case and the underlying object DATIVE. (29)

(30)

(31)

quma banaga-n y u father-ABS returned 'Father returned'. yabu numa-qgu bura-n mother-ABS father-ERG saw 'Father saw mother'. quma banaga-n y u bural-qa-n^u yabu-gu father-ABS returned saw-ANTIPASSIV Ε mother-DAT 'Father returned and saw mother'. (Dixon 1979)

This change in grammatical relations makes deletion under identity possible in conjoined clauses such as those in (31). Here we see major syntactic processes crucially referring to ERG/ABS case, a situation quite different from that in Warlpiri and the majority of PamaNyungan languages. The significant typological feature that distinguishes Dyirbal is that is it not a Pronominal Argument language; it has no AUX constituent. Dyirbal has lexical arguments; nominals are not adjuncts but verbal arguments, and ERG/ABS are grammatical, syntactically relevant cases. Dyirbal has an NP split. First and second person pronouns are NOM/ACC and as Dixon (1972) notes, there are properly speaking no third person pronouns. There are 'noun markers' that mark the four noun classes, and agree with the noun in case. These noun markers have complex paradigms that express semantic features such as animacy, gender, proximity and visibility and usually occur with a noun, although either the noun or the marker may occur alone; in the latter instance the marker is comparable in function to a pronoun. The majority of Pama-Nyungan languages have "cross-referencing" pronouns in AUX or post-cliticized to the verb, and an NP split where first and second or all the clitic pronouns are NOM/ACC. Hale (1973) claims that in Warlpiri the clitic pronouns are derived from former independent pronouns. Blake (1979) notes that the

Auxiliaries

and Ergative Splits

103

direction of change seems to have been from no bound pronouns to having bound pronouns. A historical process whereby pronouns lose word stress and become second position (AUX) clitics has been widely documented across languages.6 My proposal here is that Proto-Pama-Nyungan was a wholly accusative Pronominal Argument language with independent first and second person NOM/ACC pronouns, and ZERO third person pronouns, as is overwhelmingly the case in the daughter languages today. Nominals in Proto-Pama-Nyungan were adjuncts, adjoined predicates, with lexical cases, including LOCATIVE and INSTRUMENTAL. Most of the daughter languages developed clitic pronouns; a few, including Dyirbal, did not. INST (or LOC) case came to mark nominals coreferential with agent pronouns, as evidenced in the distribution of these cases at present. Dyirbal developed the 'noun markers' which express deictic or other very general semantic notions; these 'markers' are followed by another nominal which makes reference more explicit, as we have seen in the order of adjuncts in other Pama-Nyungan languages. Originally Dyirbal nominals were adjuncts, usually to a third person ZERO pronoun. Evidence for ZERO third person would be the fact that an utterance consisting of a verb alone (in certain tense/aspects) is unambiguously interpreted as having third person arguments; this is the case in many PamaNyungan languages today. The complex Dyirbal nominals (marker + noun), in the absence of an AUX constituent encoding the verbal arguments, acquired argumental status, and thus ERG/ABS case became grammatical cases. Hence the syntactic 'depth' of ergativity in Dyirbal, where we see syntactic processes such as subordination and control operating on an ergative system. Blake (1976:303) observes: 'It seems difficult to imagine a language developing a case [ERG] that has no syntactic basis, much more likely that the ergative marker is a vestige of a once functioning ergative system.' But, as we have seen, ERG case is clearly related to INST and LOC cases that have no syntactic functions (i. e., do not mark grammatical relations) and aside from Dyirbal and a few marginal examples we see no instances of ergative syntax. The crucial feature proposed here for Proto-Pama-Nyungan was a constraint upon argument types, such that only (independent) first

104

Eloise

Jelinek

and second person pronouns could serve as arguments, along with phonologically null third person singular. An interesting transitional case with respect to the development of AUX clitics is Warramunga (Hale 1973), where there are two sentence types, those with independent pronouns and those with unstressed clitic arguments; but the 'free' and 'bound' arguments do not cooccur, as they do elsewhere. Janet Sharp (personal communication) reports similar restrictions on the distribution of free and clitic pronouns in Ngaanyatjarra. The NP split universal in Pama-Nyungan today reflects the earlier status of the proto-language as an accusative Pronominal Argument language; in some of the daughter languages, Instrumental/Ergative case spread to nominals that acquired the syntactic function of arguments, moving into the 'vacuum' of a phonologically null third person marking. It is generally agreed that third person forms in Pama-Nyungan are so variable that no proto-forms can be reconstructed, unlike first and second person forms.

4. Contemporary evidence Recent findings on language change in Australia as the aboriginal languages suffer influence from English, and in many cases move towards extinction, lend support to the interpretation of the function of ergativity in these languages that has been suggested here. Annette Schmidt (1985) in a paper titled 'The fate of ergativity in dying Dyirbal' finds that in the Jambun community under study, the speakers of traditional Dyirbal were 35 or over, while a few younger people spoke an English-influenced Dyirbal. Each of these speakers had his own idiosyncratic 'grammar'. ERG case marking is first made morphologically simpler, then is lost as SVO word order emerges. Syntactic ergativity then declines. The loss of ERG case on nominals means that the NP split is no longer present, and syntactic processes can no longer be organized on an ergative basis. In Dyirbal we see free word order and syntactic ergativity decline along with ERG case marking, since in traditional Dyirbal ERG was a grammatical case that appeared on lexical (third person) arguments. In a paper presented at the 1984 LSA meetings, titled 'Change in Progress in Warlpiri: cross reference clitics in the auxiliary', Bavin &

Auxiliaries

and Ergative Splits

105

Shopen report that the language of young Warlpiri speakers shows the simultaneous loss of ergative case, free word order, and the AUX clitics. The simultaneity of these changes reflects the interdependence of these features in Warlpiri as a Pronominal Argument language. It appears that Warlpiri is moving towards the status of a Lexical Argument language. As nominals become arguments, AUX is redundant ('cross-reference' or 'agreement') and ERG case on nominals and free word order decay as nominals assume governed positions in the clause. The evidence suggests that as Warlpiri continues to decline, it will become a non-Pronominal Argument, accusative, configurational language before it finally becomes extinct. A Lexical Argument language does not need AUX pronominal clitics nor an NP ergative split.

5. Concluding remarks The typological parameter proposed here contrasts languages like English that permit any referring expression to serve as a verbal argument with languages such as Warlpiri that permit only pronominal arguments. I suggest that the term 'cross-referencing' is a misnomer for these languages, where nominal expressions are optional adjuncts. The contrast between the argumental function of bound pronouns vs. the non-argumental functions of nominals is often reflected in an ergative NP split frequently found in Pronominal Argument languages distributed around the world and almost universal in Pama-Nyungan.

Notes 1. It is a pleasure to thank Janet Sharp for sharing her knowledge of Australian languages with me, and for many hours of fruitful conversation on these topics. I also want to thank MaryAnn Willie, especially for her patient explanation of some aspects of Navajo to me. I am greatly indebted to Ken Hale and I also benefited from discussions with Peter Culicover, Richard Demers, Ann Fanner, Chisato Kitagawa, Adrienne Lehrer, and Susan Steele. 2. See Kinkade (1983) for an analysis of arguments and adjuncts in Salish. 3. See discussion in Willie (in press).

106

Eloise Jelinek

4. I do not intend to obscure the contrast between karra that identifies the nominal as an adjunct to the main clause subject, and kurra that identifies the nominal as an adjunct to the main clause object; I intend just to emphasize that both are locative cases. ERG (and ABS) case following these locative cases is optional, perhaps since it is redundant. 5. It is of interest that a rule altering grammatical relations in AUX is present. In Warlpiri, there is obligatory 'advancement' of first and second person goal arguments in ditransitive sentences to ACC case; first and second person are never marked DAT, only third person can be DATIVE. See discussion in Jelinek (1984). 6. See discussion in Steele et al. (1981). 7. In Jelinek (1984) I argued against the designation 'pro-drop' for languages like Spanish. Spanish restricts subjects to pronominal inflection in the verb, and thus there are no pleonastic subjects; there are nominal adjuncts to the subject pronominal inflection, not 'pro-drop'. Arabic also has constraints upon subjects, not upon objects. (See Jelinek 1981; 1983.) When a language is 'mixed' with respect to the argument type parameter, it is always the subject argument that is restricted, while the objects are not. For other kinds of restrictions upon subjects, see Jelinek & Demers (1982).

References R. 1977. On mechanisms by which languages become ergative, in: Mechanisms of syntactic change, ed. by Charles N. Li, 317 — 363. Austin and London: University of Texas Press. AUSTIN, PETER. 1 9 8 1 . Switch reference in Australia. Lg. 5 7 . 3 0 9 — 3 3 4 . BAVIN, EDITH L . , and TIMOTHY SHOPEN. 1985. Change in progress in Warlpiri: crossreference clitics in the auxiliary. Paper presented at the 1984 LS A meeting. BLAKE, BARRY J. 1976. Introduction to Sect. D., in: Grammatical Categories in Australian Languages, ed. by R. M. W. Dixon, 485—494. New Jersey: Humanities Press. BLAKE, BARRY J. 1977. Degrees of Ergativity in Australia, in: Ergativity, ed. by Frans Plank, 291 — 305. London/New York: Academic Press. CHOMSKY, NOAM. 1 9 8 2 . Some concepts and consequences of the theory of Government and Binding. Cambridge, Mass.: MIT Press. C H U N G , SANDRA. 1976. Case marking and grammatical relations in Polynesian. Ph. D. dissertation, Harvard University. C H U N G , SANDRA. 1977. On the gradual nature of syntactic change, in: Mechanisms of syntactic change, ed. by Charles N. Li, 3 - 5 5 . Austin and London: University of Texas Press. COMRIE, BERNARD. 1981. Language universals and syntactic typology. Chicago, 111.: University of Chicago Press. DELANCEY, SCOTT. 1980. An interpretation of split ergativity and related patterns. Bloomington: Indiana University Linguistics Club. DIXON, R . M . W . 1 9 7 3 . The Dyirbal language of North Queensland. Cambridge: Cambridge University Press.

ANDERSON, STEPHEN

Auxiliaries and Ergative Splits

107

R. M. W. (ed.)· 1976. Grammatical categories in Australian languages. New Jersey: Humanities Press. DIXON, R . M . W . 1 9 7 7 . The syntactic development of Australian languages, in: Mechanisms of syntactic change, ed. by Charles N. Li, 366 — 415. Austin and London: University of Texas Press. DIXON, R . M . W . 1 9 7 9 . Ergativity. Lg. 5 5 . 5 9 - 1 3 8 . HALE, KEN. 1 9 7 0 . The passive and ergative in language change: the Australian case, in: Pacific Institute Studies in Honor of Arthur Capell, ed. by Stephen A. Wurm & Donald C. Laycock, 7 5 7 - 7 8 1 . Canberra: ANU. HALE, KEN. 1973. Person marking in Warlpiri, in: A festschrift for Morris Halle, ed. by Stephen R. Anderson & Paul Kiparsky, 308 - 344. New York: Holt, Rinehart and Winston. HALE, K E N . 1 9 8 3 . Warlpiri and the grammar of non-configurational languages. Natural language and linguistic theory 1.5 — 47. HUDSON, JOYCE. 1978. The case of Walmatjari grammar. Australian Institute of Aboriginal Studies, Canberra. New Jersey: Humanities Press. JELINEK, ELOISE. 1981. On defining categories: AUX and PREDICATE in colloquial Egyptian Arabic. University of Arizona dissertation, University Microfilms, Ann Arbor. JELINEK, ELOISE. 1 9 8 3 . Person-Subject marking in AUX in Egyptian Arabic, in: Linguistic categories: auxiliaries and related puzzles, ed. by Frank Heny & Barry Richards, 2 9 - 4 6 . Dordrecht: Reidel. JELINEK, ELOISE. 1 9 8 4 . Empty categories, case, and configurationality. Natural language and linguistic theory 2.39 — 76. JELINEK, ELOISE, and RICHARD DEMERS. 1 9 8 2 . An agent hierarchy and voice in some Coast Salish languages. IJAL 4 2 . 1 6 7 - 1 8 5 . JELINEK, ELOISE, and RICHARD DEMERS. 1983. On the absence of empty categories in Lummi. Working Papers for the 18th International Conference on Salish and Related Languages, Vancouver B. C. KINKADE, M. DALE. 1983. Salish evidence against the universality o f ' n o u n ' and 'verb'. Lingua 60.25-40. Li, CHARLES N . , and SANDRA A. THOMPSON. 1 9 7 7 . A mechanism for the development of copula morphemes, in: Mechanisms of syntactic change, ed. by Charles N. Li, 419 — 444. Austin and London: University of Texas Press. MALLINSON, GRAHAM, and BARRY J. BLAKE. 1981. Language typology. Amsterdam: North-Holland. NASH, DAVID G. 1980. Topics in Warlpiri grammar, unpublished MIT dissertation. SANDOVAL, M., and ELOISE JELINEK. In press. The ^/'-construction in Apachean. To appear in: Working papers in Athapaskan, ed. by Keren Rice & Eung Do Cook. Berlin: Mouton. SCHMIDT, ANNETTE. 1 9 8 5 . The fate of ergativity in dying Dyirbal. Lg. 6 1 . 3 7 8 — 3 9 4 . SILVERSTEIN, MICHAEL. 1 9 7 6 . Hierarchy of features and ergativity, in: Grammatical categories in Australian languages, ed. by Robert Malcolm Ward Dixon, 112 — 171. New Jersey: Humanities Press. SIMPSON, JANE, and JOAN BRESNAN. 1983. Control and obviation in Warlpiri. Natural language and linguistic theory 1.49 — 64. STEELE, SUSAN M. et al. 1981. An encyclopedia of AUX: A study in cross-linguistic equivalence. LI Monograph 5. MIT Press.

DIXON,

108

Eloise Jelinek

TKASK, R. L. 1979. On the origins of ergativity, in: Ergativity, ed. by Frans Plank, 385 — 404. London and New York: Academic Press. WILLIE, MAR Υ ANN. In press. Why there is nothing missing in Navajo relative clauses in: Working papers in Athapaskan, ed. by Keren Rice & Eung Do Cook. Berlin: Mouton.

Eloise Jelinek University of Arizona Department of Linguistics Tucson, Arizona 85721 U.S.A.

Section (b):

Germanic Languages

The Auxiliarization of the English Modals: A Functional Grammar View Louis Goossens

1. Introduction 1.1. Aims The purpose of this paper is double. It wants to deepen our insight into the auxiliarization process that the English modals underwent in their development from Old to Present-day English. Since it is especially Old English that is in the center of interest in this respect I will present two sample studies based on material from the Toronto Microfiche Concordance to Old English to clarify some of the points I want to make. Secondly I will present a Functional Grammar view of the kind of development that we find ourselves concerned with. The indefinite article is particularly significant since modality (and the modals) are largely unexplored territority in Functional Grammar. The investigation presented here will help, I hope, to clarify Functional Grammar insights into modality and the modals. At the same time it does not claim to be more than an interim contribution to a vast problematic area in linguistics where Functional Grammar will have to proceed to extensive further work.

1.2. Structure of the paper The structure of this paper is as follows. In the second section we first emphasize the differences between Old English and Modern English modals with a view to delimiting the set with which we want to be concerned. Having decided to define our set from a modern

112

Louis

Goossens

perspective, we consider in section 3 the transformational viewpoint as formulated in Lightfoot (1979) that the Old English 'premodals' 1 were recategorized as auxiliaries in the sixteenth century; summarizing (i. a.) some of the arguments from an earlier paper, we argue that Lightfoot's hypothesis is much too drastic to be empirically adequate. Section 4 presents my tentative proposals to deal with modality and the modals as formulated in Goossens (1985 a; b) with a slightly modified proposal for the Present-day modals in subsection 4.3. In section 5 then we will look at the Old English counterparts of the present-day English modals considering in some detail two items that present us with two extremes in the Old English spectrum of pre-modals, cunnan and sculan. In this analysis we base ourselves on two samples of 200 instances each from the Toronto Microfiche Concordance. It leads to a proposal as to how the insights gathered can be formulated within Functional Grammar. The general conclusion (section 6) reviews the foregoing first at the level of the Functional Grammar model, then with respect to the general question concerning the auxiliarization of the modals from Old to Present-day English.

2. The modals: Old and Present-day English By way of a preliminary we have to make clear what we mean by 'the modals'. The set is comparatively easy to identify in Present-day English. It contains the verbal items which are characterized by the following features. (i) (ii)

(iii)

Formally: the absence of non-finite forms; and the absence of -s in the third person singular of their so-called present. Syntactically: by the fact that they belong to the so-called 'anomalous finites' (otherwise formulated, by the fact that they exhibit the 'NICE-properties'), i. e. that they do not need do-support in a number of syntactic environments (Negation with not, Interrogative, repetition of the 'main verb' (Code), Emphatic affirmative); also by the fact that they combine with the 'plain' infinitive of some other verb. Semantically we find that the items that share these syntactic and formal properties cover a broad range of modal meaning

The Auxiliarization

of the English Modals

113

(going from facultative to deontic and epistemic modality) plus a number of other meanings, including what are usually referred to as grammatical meanings. Turning to Old English we find that the syntactic criteria do not apply. Z)o-support is a later development in English: and, though the combination with infinitives is not infrequent, it is not a generalized feature (as will amply appear from our analysis of cunnan). Formally, the situation is rather different as well. What seems to offer itself instead of the modern English formal criteria, is that as a rule the pre-modals belong to the so-called preterite-presents (see e.g. Campbell 1959:§§726, 767). But then the classes of preteritepresents contain a number of other items besides the pre-modals. And there is one pre-modal that is not a preterite-present (even if the similarity with the pre-modals is considerable), viz. willan, which Campbell classifies separately among the 'anomalous' verbs (see 1959: §768 (a)). Semantically there are considerable differences as well. The details are beyond the scope of this paper, but will partially appear from the two case studies in section 4. Otherwise we can refer the reader to the discussion in Visser (1969). The gist of all this is that a diachronic study of the modals will have to specify what are the exact items that one wants to concentrate on. In our case, we take the central Modern English modals (can, could; may, might; shall, should; will, would; must) and their predecessors in Old English (cunnan; magan; sculan; willan; motan) as our focus of attention. As already pointed out, I will concentrate on two of the Old English items only {cunnan and sculan). The selection is such, however, that it brings out some essential points about the premodals in Old English.

3. Lightfoot (1978): premodals recategorized as AUX Within the framework of Transformational Grammar, Lightfoot (1979) defends a 'transparency principle', which 'requires derivations to be minimally complex and initial, underlying structures to be "close" to their respective surface structures, and [which] must be conceived as part of the theory of grammar, and not as a component

114

Louis Goossens

of a theory of (syntactic) change' (1979:121). His paradigm example for the radical reanalysis that can be occasioned by this principle is the change 'whereby the grammar of sixteenth-century English developed a new syntactic category of "modal verb"' (1979: viii). In outline his analysis of this change boils down to the following. The pre-modals belonged to the category verb, but became increasingly exceptional until in the sixteenth century a point was reached where this could no longer be tolerated (according to the transparency principle), so that within a short time span a radical restructuring took place which recategorized them from (anomalous) full verbs to become 'modals', members of the deep structure category AUX. In an earlier contribution (Goossens 1984), a belated publication of a paper presented at the English Historical Linguistics Conference held in the University of Durham 1979,1 have argued that the change in question is not purely syntactic as Lightfoot would have it.2 In outline, I pointed out that all the factors which for Lightfoot played a part in the change have a semantic dimension: (i)

the breaking up of the present-past relationship went hand in hand with an increase in the semantic divergence between the pairs;

(ii)

the fact that the pre-modals lost their capacity to take direct objects, directly correlates with the loss of what I called the non-operatorlike meanings which the pre-modals could have;

(iii)

the loss of the non-finite forms, itself a morpho-syntactic accident, is up to a point understandable if we consider the operatorlike status of the modals. Especially in their epistemic and discourse-oriented meanings (where they have scope over the rest of the predication) they do not stand in need of non-finite forms; the quasi-modals (have to, be able to, etc.), which primarily arose out of morpho-syntactic needs, also fulfill a semantic function, in that they are as a rule unambiguous markers of the non-discourse oriented/non-epistemic modal meanings and hence do not exhibit the ambiguity of the real modals;

(iv)

(v)

the fact that can, may and will (which at the time when doperiphrasis was in the making still had their infinitives) were excluded from do-periphrasis in its initial stages can be accounted for on semantic grounds; indeed, periphrastic do

The Auxiliarization

of the English Modals

115

derives from a pattern where it combined with action verbs only and appears to have been resistant for a long time to a combination with the class of verbs to which the modals belong. A second point which I want to add here, is that Lightfoot's view of a change to the category AUX is made dependent upon factors which do not in my view justify the thesis of a radical short-term shift in the course of the sixteenth century as he would have it. (i)

Among the factors that prepare the shift to AUX there are two which continue to be operative beyond the sixteenth century. One of those is mentioned and illustrated by Lightfoot himself (Lightfoot 1979:101; see also the footnote there). It is the continued combination of can with NP objects as in (1). (1)

(ii)

Yet can I Musick too; but such as is beyond all Voice and Touch (1649 Lovelace, Poems (1659) 1 20). A second factor is the supposed break in tense opposition between the modals. Here present-day can and could offer counterexamples, as e. g. in (2) (2) I could jump a lot higher when I was a boy. The operation of a category AUX in Negative Placement and Inversion, which is crucial for the adoption of the category AUX as conceived by Transformational Grammar, was not yet a fact in the sixteenth century. The establishment of do as an obligatory dummy auxiliary in questions dates from about 1700, and it takes until the late nineteenth century before dummy do is generalized as we know it nowadays in negative sentences.3

In sum, interesting though Lightfoot's discussion of the development of the English (pre)modals may be, we can neither accept that it is a primarily syntactic matter, nor that it had a culminating point leading to recategorization within a short time span in the course of the sixteenth century. This conclusion will find further confirmation from an Old English viewpoint in our sample studies of cunnan and sculan. It will, of course, have to be taken into account in a Functional Grammar view of the change, for one thing because, if possible, a semantic account should take precedence over a syntactic one in Functional Grammar.

116

Louis Goossens

4. A Functional Grammar view of the modals in Present-day English Before proceeding to a more detailed investigation of two Old English items, I briefly summarize the (tentative) proposals in Goossens (1985 a) (4.1), and their reconsideration in Goossens (1985 b) (4.2). I round off this presentation with some additional points that I think have to be borne in mind in an Functional Grammar account of the modals, whether synchronic or diachronic (4.3). 4.1. If we try to account for the way in which modals can be integrated into the construction of predications three possibilities present themselves: (i)

Predicate operators. Predicate operators are introduced to capture the uses of the modals in the formation of the future tense (shall, will; predicate operator FUT), in the conditional tenses (should, would; predicate operator COND) and in specific types of subclauses corresponding to subjunctives in older stages of the language (such as should in clauses after main verbs reporting directives or may/might in purpose clauses; predicate operator SUBJU).

(ii)

Predicate formation This is an option which is available to deal with the remaining uses of the modals (i. e. especially when they express epistemic, deontic or facultative modality). It must be weighed off against (iii). If we opt for predicate formation, we get rules like (3): (3)

Input: φ(χ!)...(χ„) Output:

(x!)...(x n )

— φ stands for some predicate, x} and xn for its associated arguments — mv for any modal verb, to be differentiated into mv 1 (epistemic modality), mv 2 (deontic modality) and mv 3 (facultative modality)

The Auxiliarization

of the English Modals

117

— the following restrictions on the input predications differentiate (up to a point) the three modality types: (a) mv 1 requires as a rule a 'situation' (i. e. states or some other type of State of Affairs if it is prefixed by a PROG(ressive), PERF(ective), or HAB(itual) predicate operator (b) mv 2 and mv 3 typically combine with 'events' (i.e. non-progressive, non-perfective, controlled states of affairs) — Subject assignment for the derived predicate works in the same way as for the input predication Predicates in their own right The alternative to (ii) (but again excepting the uses under (i)) is a treatment of the modals as predicates in their own right. As such, they must be states with zero-role arguments. Again we distinguish between three modality types (as mv 1, mv 2 and mv 3). A further distinction which suggests itself is that epistemic modals are one-place predicates and that facultative modals are two-place. Deontic modals can be taken to be two- or one-place predicates, according to whether some (personal) entity is affected by the deontic modality (i.e. is 'under obligation' or is 'granted permission') or not. The single argument in a one-place modal, as well as the second argument of the two-place modals is an embedded predication, which is (as a rule) a 'situation' (for epistemic modals) or an 'event' (in the case of deontic and facultative modals). Hence we get predicate frames like the following: (4)

must m v l (x,: r situation -j) (x,)) < (position) > I (process) J

(5) (6) (7)

may m v 2 (xi: event (x t ))0 may m v 2 (xi) 0 (x2: event (x 2 ))0 will mv3 (x,) 0 (or Exp. ?) (x 2 ))0

Subject assignment is to an argument in the embedded predication (one-place predicates) or to the first argument in the case of two-place modals (with the proviso that X! must also be an argument in the embedded predication which remains unexpressed).

118

Louis Goossens

4.2. In Goossens (1985 b) I reconsidered these proposals from two angles. First I asked the question whether the differentiation into predicates in their own right, predicate formation position and predicate operator origin can be backed up by phonological, morpho-syntactic or semantic arguments. The conclusion was that the differentiation can partly be justified semantically. On the other hand, no such differentiation can be argued for on phonological or morpho-syntactic grounds (for one thing pronominalization tests do not bring out the difference between one- and two-place predicates). Secondly, I looked at the proposal from the point of view of grammaticalization, an angle which has particular relevance for the auxiliarization question in the diachronic perspective. Making use of six parameters to test grammaticalization offered by Lehmann (1982) I had to conclude that on all of these parameters we get indications that the English modals are grammaticalized to some extent (but never fully). Moreover, that the degree of grammaticalization does not appear to be systematically differentiated except for those criteria that relate to the degree of desemanticization of the modals and which I claimed to be reflected in scale (8). (8)

Facultative > Deontic > Epistemic > Futurity Conditional etc.

In other words, the basis for the differentiation of the modals is the same as that for grading their grammaticalization. Looking for a grammaticalization scale within Functional Grammar, we find it in a scale like (9) (9)

full predicates > predicate formation > predicate operators

This naturally leads us to consider the possibility that scale (9) can be used to capture the differences in grammaticalization of scale (8). Accordingly, I tentatively proposed that facultative and deontic modals should be dealt with as full predicates, that epistemic modals would involve predicate formation, and that modals in future and conditional tenses and in certain types of subclauses could be taken care of by means of predicate operators. In the following subsection I would like to reevaluate that proposal, repeating one argument

The Auxiliarization

of the English Modals

119

against it which I already raised in the 1985 b paper, but also adding a few other points which I had not taken into account before. 4.3. As I have pointed out in the 1985 b paper, the proposal presented in 4.2. is at least partially objectionable in its application to the Present-day English modals in that it assigns full predicate status to the facultative and the deontic modals, which on the non-semantic grammaticalization criteria exhibit the same (partial) grammaticalization as the other uses. A particularly important counterindication is the impossibility of pronominalizing the arguments we assign to the different modality types. This should not lead us to discard schema (9), however. Rather, we feel inclined to place the Present-day modals at least in the predicate formation position, whereas independent predicate status will appear to be indispensable for (at least some of) the Old English premodals. Secondly, desemanticization scale (8) is in its present form no more than a first approximation. What is eventually needed, I think, is something that reflects the diachronic shift in individual modals much more precisely and that therefore will be a lot more complex. At present I only want to remark that scale (8) does not imply that the transition from one point (e. g. Facultative) to another non-adjacent one (e. g. Futurity) would necessarily involve the intermediary steps (in our example Deontic and Epistemic). For the purposes of this paper, however, I will not go beyond this (over)simplified version. A third observation concerns the status of predicate formation in schema (9). It is not fully clear to me yet that the kind of intermediate position on a grammaticalization scale between full predicate and predicate operator should be captured by predicate formation. As I have pointed out in Goossens (1985 a), the suggestion to deal with certain modality types as instances of predicate formation comes from Vet (1981), but as yet the exact status of predicate formation within Functional Grammar has by no means been completely explored. Since nothing else appears to offer itself within Functional Grammar, I will stick to labelling this intermediary position as predicate formation. At the same time I will try to give a more precise account of the kind of shift that takes place when the modals move away from their status of full predicates in our discussion of the Old English data. In doing so, we will contribute to the clarification of predicate formation within Functional Grammar, if at least shifts of this kind must come under predicate formation in the model.

120

Louis

Goossens

5. A Functional Grammar view of two Old English modals 5.1. Introduction The next step is to see how the Old English modals would have to figure on scale (9). More specifically we want to find out what arguments can be found to locate them on this grammaticalization scale one way or another. To that end we shall have a closer look at two items which clearly illustrate that different Old English modals (e. g., cunnan and sculan) (may) require different locations on the scale. The choice of cunnan and sculan has been determined by the fact that they appear to me to represent two extremes within the Old English items with respect to grammaticalization and auxiliarization. The two samples which are studied in the following two subsections were taken from the Toronto Concordance of Old English. Each contains 200 instances from iElfric and Wulfstan ( ^ l f r i c largely predominates, a small minority comes from Wulfstan), the samples are therefore late Old English (the Old English from around the turn of the first millennium) and come from a homiletic corpus.

5.2. Cunnan 5.2.1. The sample includes practically all the vElfric and Wulfstan instances in the Toronto Concordance (only a couple which I found it impossible to interpret from the context included were left out) for the following forms of the verb: canst (which yielded 8 instances), cunnan (11 instances), cunne (45 instances), cunnon (43 instances), cupe (68 instances), cupon (25 instances). 5.2.2. To answer the question whether cunnan needs to be considered as an independent predicate or not we have to decide whether we can find proof that it has its own argument structure (with associated subjectivalization and (possibly) objectivalization), or not. Working from the sample sentences, this means devoting special attention to the constituents that appear as subjects and especially whatever else the verb combines with besides the subject.

The Auxiliarization

of the English Modals

121

(a) Subjects The subject of cunnan is always an NP denoting an animate entity (animate entities). The only exceptions are the 14 instances where cupe is not a past tense but an (adjectival) past participle. Obviously in those instances the active 'object' comes in subject position, cf. (10). (10)

his gebyrd and goodnys sind gehwaer α φ ε {AZCHom I (Pref) 2.1.)

As is to be expected, there are as a rule no agent phrases and if the 'active subject' is represented, it comes in an adjunct form which is not typical of agents (be, fram or purh followed by an NP representing an animate entity), but in some other form. 4 (b) Non-subject complements Let's first give a survey of the non-subject complements. We include the subjects in sentences with the past participle cupe between brackets, because as was pointed out under (a), these sentences are no real passives.5 1. The complement is an object NP (a) denoting a person 26 (b) denoting a language 12 (c) denoting some other 'knowable' entity 90 Sentences with cupe as an adjectival past participle (2. 3. The complement is an infinitive (possibly with its own complements) 4. The complement is zero 5. Other complementation types Total

128

14) 38 13 7 200

We first provide examples to this schema, adding a few words of discussion where necessary. Further interpretation will be offered globally in 5.2.3. 1. (a)

(11) (12)

Canst ]du öone preost ^e is gehaten Eadzige (...) (y£LS (Swithun) 21) God gesceop öa haejrenan Jjeah hi hine ne cunnon, ... (aELS (Forty Soldiers) 336) 'though they do not know him'

122

Louis Goossens

(b)

(13)

(c)

(14)

(15) (16)

(17)

Bide nu aet Gode J>aet ic grecisc cunne (ALLS (Basil) 514) ^Elc Cristen man sceal cunnan his paternoster and his credan (AELS (Ash Wed) 261) [note that cunnan is infinitive] Twa lif sind soölice. J)aet an we cunnon. Jsaet oöer ... (ALCHom I, 15 224, 14) O n e we know, the other...' ... and eow laewedum mannum is öis genoh. öeah öe ge öa deopan digelnysse öaeron ne cunnon (ALCHom II, 35 267.228) 'though you do not know the deep mystery thereof ... and swa hraöe swa heo gehyrö Jjaere burge naman J)e heo aer cuj?e ... (ALLS (Christmas) 222) 'and as soon as she hears the name of the town which she knew before'

2.

For examples, see (10) and fn. 4(a), (b) and (c).

3.

(18) (19)

(20)

(21)

Ne canst fcu huntian buton mid nettum (ALColl 61) Forgif me wisdom. {)aet ic maege J)in miccle folc gewissian. and ic cunne tocnawan betwux god and yfel (ACHom II, 45 336.28) 'Give me wisdom so that ... I can/may discern between good and evil' ... ac hi naefdon öone lareow. öe him cuöe J?a digelan lare geopenian... {JECHom II 29 233.107) '... that could expound to them the secret lore' And he leop sona cunnigende his feöes hwaeöer he cuj)e gan {ALLS (Peter's Chair) 32) 'and he at once leapt up trying his gait whether he could walk' And gecnawe se t>e cunne, nu is se tima ... ( W H o m 5 21) [examples of this type are all interpretable as exhibiting ellipsis of an infinitive; gecnawe se £>e cunne = gecnawe se J)e hit gecnawan cunne]

4.

(22)

5.

I illustrate the different subtypes all of which occur with very low frequencies, as the number of occurrences, which I give between brackets, proves. (23) SeJ>e wylle wacian and wuröian Godes halgan, wacie mid stilnysse and ne wyrcan nan gehlyd, ac singe his gebedu swa he selost cunne ... (ALLS (Pr Moses) 81) 'as skilfully as he can' [cunne supplemented by swa and selost, but one might also argue that this is

The Auxiliarization

of the English Modals

123

another instance of an ellipted infinitive — one instance] (24)

...fa t»e on stane cunnon, and gecwemlice on treowe, J)£et hi on Romanisce wisan araere his cynebotl (/ELS (Thomas) 33) 'those that know about stone and satisfactorily about timber, so that they may build his palace after the Roman fashion' [on-phrase to indicate the entity the subject knows about; 3 instances]

(25)

Cristene men ne motan healdan nu öa ealdan. ae. lichamlice. ac him gedafenaö t>aet hi cunnon hwaet heo gaslice getacnige (sECHom II, 15 151.36) '... but it befits them that they know what it means spiritually' [the complement is a clause introduced by the 'conjunctive' hwaet', there is another example like it with hu as a conjunctive] f a öe bet cunnon and magon. sceolon gyman oöra manna ... (/ECHom II, 15 159.311) 'those that know and can [do] better must direct other men' [with an adverbial comparative; one instance]

(26)

5.2.3. The crucial point in these data is the predominance of instances with NP objects (64%, not including the cases with the adjectival past participle cupe in 2. of the table, which for that matter, though no real passives, have subjects that are of the same type as the NP objects in 1.). Note also that those NP objects can be pronominalized, as is illustrated in (12) and (17). In all these sentences cunnan is clearly an independent two-place predicate. The State of Affairs-type is state. According to Dik (1981) it should take two arguments, but it seems preferable (also to Dik, personal communication) to differentiate the two roles. My (tentative) proposal is Exp(eriencer) for X! and 0 for x2. The selection restriction for the first argument is that it must be [+ human]; in a more refined formulation, employing categorizations that as such have not yet been introduced into Functional Grammar, we can describe the first argument as [+ cognizant], the second as some 'knowable entity',6 in line with the fact that cunnan in all these instances is equivalent to ModE know. Notice also that this cunnan 'know' is frequently recorded in the infinitive (a non-finite form) in the Toronto Corpus as a whole (and in the jElfric and Wulfstan entries in particular).

124

Louis

Goossens

What to think of the remaining categories in the table? Those under 5., if anything, confirm the independent predicate status of the verb, possibly with reduction of two arguments to one,7 but, of course, their number is too restricted to have any but marginal value in the characterization of the verb. More important are those under 3. and 4. They can be grouped together, because the instances under 4., as we have already pointed out, can all be interpreted as showing ellipsis of an infinitive. This would give us a portion of nearly 25% of our sample. A first remark concerns the types of predicate figuring as infinitives. There is a clear predominance of verbs like tocnawan, asmeagan ('consider, investigate, think'), understandan (which are also the verbs that are typically ellipted). Besides there are also action predicates like huntian, temian ('tame'), secgan etc., all of which have an ingredient of (intellectual) insight. In all these instances cunnan is translatable as 'know how to' as well as 'can' or 'be able to'. In fact there is only one case where the 'ability' sense practically completely ousts the 'know how to'-meaning, viz. (21). With respect to the independent predicate status, we can continue to consider cunnan as a two-place predicate with the infinitive, a nominal verb form, as the second argument. 8 Alternatively we can look upon the combination of cunnan with infinitives as a first step away from the independent predicate status: indeed when cunnan combines with another predicate which brings along its own arguments, it shares its first argument with this combining predicate, which, moreover, imposes its additional argument structure (if it is two- or more-place) on the combination. 5.2.4. To capture the foregoing in Functional Grammar-terms we propose the following basic (independent) predicate structure for cunnan: (27) Cunnanv (xj: human (xi))exp (x2)0 — Further refinements for xj would be to restrict it to 'cognizant' entities, for x2 that it must be some 'knowable' ('cognizable') entity — unmarked subjectivalization and objectivalization turns the first argument into the subject and the second into the object. If we decide to bring the combinations with infinitives in line with structure (27) we could adapt it as (28).

The Auxiliarization

(28) (28)

of the English Modals

125

cunnanv (x^ human (XI))E X P (x2: [φ ... ( Xl ) ...](x2))0 receives the following specifications: — xi is 'cognizant' and/or 'potent' — x2 is some predication whose first argument is identical with the Xj of cunnan. — x2 is still objectivalized but takes the infinitival form of the predicate in question.

Alternatively, we can set apart the combinations of infinitives with cunnan as resulting from predicate formation in the following fashion (29)

Input: cp(xt) ... (xn) Output: cunnan cp(xi) ... (xn) Where φ primarily stands for the class of predicates that involve some degree of 'cognizance' (understandan, asmeagan, cnawan and the like; but also action predicates involving 'cognizance'); accordingly x1? which is to be subjectivalized, is, to some extent at least, 'cognizant'.

Examples like (21) would require us to relax the restriction of φ to 'cognizance' and of Xj to 'cognizant': xj becomes merely 'potent'. This second solution would account for a first step of cunnan towards grammaticalization. As the foregoing discussion should have made clear, I would hesitate whether anything more than the beginnings of this stage should be claimed for cunnan.

5.3. Sculan 5.3.1. The sculan sample can be said to be made up of two layers of 100, both in the great majority instances from ^Elfric, again supplemented by a few from Wulfstan. The distribution is as follows: sceal (70 instances: 50 iElfric, 20 Wulfstan), scealt (30 instances, all iöfric); sceolde (50 instances: 40 ^ f r i c , 10 Wulfstan), sceoldon (40 instances, all jElfric), sceolden (10 instances, all ^lfric). Note that this represents only a selection (and obviously an arbitrary one) from the relevant material for the items in question in the Toronto Concordance. The two layers then are sceal/scealt on the one hand

126

Louis Goossens

and sceold- on the other. As will appear there are considerable differences between the two. 5.3.2. A first observation which holds for the whole 200-instance sample is that sculan is combined with an infinitive (phrase). There are only seven exceptions. The first of these is (30 a), where we get a contracted relative as a direct object, which, however, anticipates the following infinitives. (30)a. ... t»aet is J?aet he sceal. öa flaeslican lustas gewyldan: and his lichaman to godes öeowdome symle gebigean {/ECHom I, 12 188.24) 'this is what he has to (do), control the lusts of the flesh ...' There are three that are in combination with a directional constituent, with an 'ellipted infinitive' indicating movement (e. g. gan); an instance of this type is (38) (see section 5.3.5.). Finally, there are another three cases with an ellipted infinitive in a clause of comparison introduced by swa (swa). For an instance, see (61). Notice also that we often find non-animate subjects for sceal/ scealt/sceold- (22 in the whole sample), which we illustrate in (30 b), (30 c) and (30 d) (30) b. He saede Jjaet aefter J)isum faece gewuröan sceall swa egeslic tima swa aefre naes syööan J)eos woruld gewearö (WHom 6 197) 'He said that after this period such a dreadful time will come about ...' (30) c. Ne cwaeö na se symeon J?aet cristes swurd sceolde Jmrhgan marian lichoman: ac hyre saule {JECHom I, 9 146.15) 'Simeon did not say that Christ's sword would pierce through Mary's body, but ...' (30)d. ... And sceal beon gebodod on minum naman daedbot. and synna forgifenyss: on eallum feodum (JEHom I, 15 220.20) 'and there will be announced repentance and forgiving of sins in my name to all nations' 5.3.3. A second point, obviously not derivable from the sample, is that sculan (practically) does not occur in its non-finite forms, as far as the whole Toronto Corpus goes. In the yElfric and Wulfstan samples there is not a single instance of sculan as infinitive, of a past

The Auxiliarization

of the English Modals

127

participle of any form, of a present participle. A glance at the rest of the material revealed only one doubtful instance of an infinitive.9 The conclusion then is that in this respect sculan is not a full verb (any more) in Old English. 5.3.4. Scealjscealt For sceal/scealt we have looked at the distribution of the (semantic) opposition between a necessity or obligation meaning and the expression of futurity. We illustrate this in the following instances; (31) and (32) are instances of N(ecessity), (33) and (34) of F(uturity). (31)

Swa sceal don se gastlica sacerd. he sceal gerihtlaecan godes folc and Jjone ascyrian. and amansumian fram cristenum mannum f)e swa hreoflig biö on manfullum J)eawum £>aet he oJ>re mid his yfelnysse besmit (JECHom I, 8 124.27) [Necessity: obligation]

(32)

Hit is awriten on öaere ealdan ae. t>aet nan mann ne sceal hine gebiddan to nanum deofelgylde. ne to nanum t»inge buton gode anum ... {JECHom I, 11 174.4) [Necessity: negative obligation or prohibition] ... and he ahof öa eadmodan: swa swa crist sylf cwaeö on his godspelle: aelc J)aera f>e hine onhaefö. he sceal beon geeaömet (j£CHom I, 1 202.29) [Futurity: 'each of them who extols himself, will be humiliated']

(33)

(34)

f u s waes gewriten be me. J?aet ic !)rowian scolde: and arisan of deaöe on öam f>riddan daege: And sceal beon gebodod on minum naman daedbot. and synna forgifenyss: on eallum feodum (AZCHom 1,15 220.20) [Futurity: 'and in my name penitence and forgiving of sins will be proclaimed to all people']

Obviously, if we want to assess the degree of grammaticalization of scealjscealt it is important to see to what extent it loses its Necessity-meaning and becomes a signal to express (mere) Futurity. In my sample it appears that the great majority of instances present us with a mixture of Necessity and Futurity. The distribution, which, of course, can only established somewhat tentatively, is as follows: Necessity 16

Futurity 16

Necessity/Futurity 68

128

Louis Goossens

We conclude this subsection with a few illustrations of this mixed category. (35)

(36)

(37)

Crist ableow J?one halgan gast ofer J?am apostolan J?a gyt wuniende on eoröan. for Jjasre getacnunge. faet aelc cristen mann sceal lufian his nehstan swa swa hyne sylfne (/EC Horn I, 16 232.10) [Necessity/Futurity, Necessity predominates '... as a sign thereof that each Christian man shall love his neighbor as himself] J)onne faerlice gewit he of J)yssere woruld nacod and forscyldgod: synna ana mid him ferigende. for öam öe he sceal ece wite J)rowian (A?C Horn I, 4 66.12) [Necessity/Futurity: 'then he suddenly departs from this world, naked and guilty: taking only sins with him; therefore he will/shall suffer eternal punishment'] heo is ece. and naefre ne geendaö. Jieah se lichoma geendige. J)e sceal eft J)urh godes mihte arisan to ecere wununge (JEHom I, 1 20.4) [Necessity/Futurity, Futurity predominates:'... though the body may end, which afterwards through God's might will (shall) arise to eternal dwelling (life)']

5.3.5. Turning to the sceoldejsceoldonjsceolden sample, we concentrate on two parameters that can be assumed to give indications about grammaticalization. The first, (a), is to what extent sceold- is still used as an indicative (to mark past necessity/obligation); the second, (b), concerns the clause-types in which they occur, with the additional question whether they act as a grammatical signal in any of the clause-types in which they are used. (a) Indicative or non-indicative 10 A first important fact about sceoldejsceoldonjsceolden is that in our sample they almost exclusively occur in contexts where they have to be interpreted as non-indicatives. Notice that formally sceolde is both indicative and subjunctive; that sceoldon (the predominating plural form in our sample) is (at least in 'standard Old English') an indicative; and that sceolden is the standard subjunctive form. As is wellknown, however, both -on and -en can be assumed to stand for [an] in late Old English; moreover, spellings in unstressed syllables tend to get confused. 11

The Auxiliarization

of the English Modals

129

If we take factual contexts to require the indicative and nonfactual contexts of various sorts the subjunctive or at least a nonindicative we find the following distribution (the total of this subsample is 100): Indicative 3

Non-indicative 96

? 1

For exemplification of the non-indicatives we refer to the instances quoted under (b), which are all of this type. The three indicatives are (38), (39) and (40); the undecidable case is (41). (38)

(39)

(40)

(41)

god t»a gegearcode aenne hwael. and he forswealh [?one witegan and abaer hyne to |?am lande. he to sceolde and hine J)aer ut aspaw (/ECHorn I, 18 246.12) '... and [it] carried him to the land to which he had to [go] and spewed him out (up) there' On öam getelde hi sceoldon f)a godcundan lac symle geoffrian. for öan öe hi ne mihton on öaere fare cyrce araeran (/EC Horn II, 12.1 144.160) 'In that tent they always had to sacrifice the divine offerings, because Seo ealde ae bebead J>aet gehwilc reoflig man gecome to t?am sacerde: and se sacerd sceolde hine fram mannum ascyrian gif he soölice hreoflig waere (/EC Horn I, 8 124.5) '... and the priest had to separate him from people if he were really leprous' Hit waes gewunelic Jsaet öa magas sceoldon J>am cilde naman gescyppan on J>am eahto}>an daege mid J>aere ymbsnidenysse: ac hi ne dorston naenne of)erne naman criste gescyppan {ionne ... (/ECHom I, 6 94.22) 'It was the custom/the rule that the parents had to/ should assign/assigned the child a name on the ...'

Notice that in the indicative instances sceold- retains its basic necessity (obligational) sense. In (38) we get a combination with a directional constituent and an 'ellipted infinitive';12 (41) can be interpreted in three ways: as a factual obligation (paraphrasable as 'it was the custom that the parents had to assign the child a name'), as a non-factual obligation ( = 'it was the rule that the parents should

130

^ouis Goossens

...'), as a grammatical marker of the subject clause after an expression of the type Hit was gewunelic (where Present-day English would use zero or would). The ultimate decision (if there is one) would have to rely on more elaborate investigation of constructions of this type. The exceptionality with which sceold- functions as the past tense equivalent of obligational scealjscealtjsculon indicates that, more than the latter, sceold- was used in functions which can (with varying degrees) be described as grammaticalized. To find out more precisely what this grammaticalization amounts to, we now turn to an analysis of the different clause types in which sceold- occurs. (b) Clause types Let us first survey the different clause types with sceoldejsceoldonj sceolden. 1. Main clauses 2. Relative clauses 3. Object clauses 20 ^ a. after verbs of volition 25 ( b. after (other) verbs of communication c. after verbs of opinion 1 ( 2 J d. after verbs of 'fear' 4. Content clauses after nominal heads a. after a noun expressing volition or command 6 b. after other nouns 4 | 5. Subject clauses 6. Clauses of consequence/purpose 7. Clauses of comparison 8. Other

10 5 48

10

1

Total sample sceoldejsceoldonj sceolden

2 20 4 1 100

Next we consider each of these categories in some detail, giving at least one example and adding a brief discussion. We restrict ourselves to what we regard as relevant to the question of grammaticalization. 1. Main clauses (10 occurrences) As a rule sceold- is paraphrasable as should/ought to. It occurs in combination with a hypothetical conditional subclause in 4 instances, e. g. (42); this is not the case in the other 4 instances, e. g. (43).

The Auxiliarization

(42)

(43)

of the English Modals

131

Gyf hit J?onne maedencild waere: J)onne sceolde heo hi forhaebban fram ingange godes huses hundeahtatig daga ... (JECHom I, 9 134.18) 'If it were a female child, then she should abstain (herself) from ...' aefter godes gesetnysse ealle cristene men sceoldon beon swa gejswaere. swilce hit an man waere. for wa {jam men J)e öa annysse tobrecö 'according to God's decree all Christian men should be so united as if it were one man, ...'

The other two cases are two indicatives, viz. (39) and (40). 2. Relative clauses (5 occurrences) Again most instances are paraphrasable as (obligational) should (e. g. 44) bordering on was to in (45). (44)

(45)

Hit getimaö forwel oft {jaet öa Jswyran becumaö to micclum hade on godes gelaöunge. and hi f)onne gastlice ofsleaö mid heora yfelnysse heora underjjeoddan: t>a öe hi sceoldon mid heora benum geliffaestan ... (AECHom I, 28 412.20) 'those that they should endow with their services' ... J)a gelamp hit J>a öa hi on J)aere byrig betleem wicodon. t>aet hire tima waes gefylled \>xt hio cynnan sceolde: ... {/ECHom I, 2 30.5) 'that her time was fulfilled that she should/was to give birth'

There is one instance with an indicative past tense, see (38) above. 3. Object Clauses (48 occurrences) As appears from the table this is a diversified category with two high frequency (sub)categories and two which are marginal. 3 a. After verbs of volition (20 occurrences) In this category sceoldejsceoldonjsceolden occur as a rule in peetclauses (there are four instances of Aw-clauses, exemplified by (48)). Sceold- as it were echoes the obligational meaning in main verbs like hatan, (ge)neodan, bebeodan, gewissian and at the same time underscores the non-factual character of the predication in the subclause. As such it gets dissociated from the basic meaning of sculan and becomes a marker of a specific type of subclause. Note that the distinction with the following category, though not problematic on the whole, is not always as clearcut as one might wish, witness (49).

132

Louis Goossens

The following instances illustrate the category. (46)

(47) (48)

(49)

t>a öa Ioseph J>is smeade: J>a com him to godes engel. and bebead him f)aet he sceolde habban gimene seigöer ge J)aere meder. ge öaes cildes ... (AZCHom 1,13 196.17) and ordered him that he should take care both of the mother and the child...' Crist wolde J)aet manega witegan. and eac öa haeöenan sceoldon bodian his tocyme. ... {/ECHom II, 1 10.267) &of J?am cynne aefter |)aere wisan syööan wurdan manege, oö J?set Crist eft astealde on oöre wisan hu bisceophad sceolde of manegan cynrynan syööan aspringan, swa swa ... (WHorn 17 21)'... until Christ ordained in another way how bishophood should originate from a lot of families thereafter ...' ... swa swa Crist laerde t>aet man don sceolde ... (IV Horn 17 65) 'as Christ taught that man should do'; [if we take Iceran as 'imperative teaching' we have to classify it here]

3 b. After (other) verbs of communication (25 occurrences) Whereas in 3 a. we considered subclauses after verbs reporting directives (and verbs of volition), we now turn to those after verbs rendering other speech act types, e.g., cwepan, secgan, 'say', acsian, biddan 'ask', behatan 'promise'. The clause is introduced by poet (or by some interrogative word after verbs reporting questions). Notice that sceold- is paraphrasable as either would/was, were to, as in (50) and (51) (there are 14 instances of this type), or as should (10 instances in all, exemplified by (52) and (53)). (50)

(51)

(52)

God us forbead J>aes treowes wasstm and cwaeö. Ipxt we sceoldon deaöe sweltan gif we his onbyrigdon {&C Horn I, 1 16.35)'... and said that we would perish by death if we ate from it' And öa agunnan hi hine eft acsian dihlice hwaenne Jsaet geweoröan sceolde, & eac ...) (WHorn 2 30) 'and then they began to ask him secretly when that would/ was to happen, ...' t>a men öe beforan ^aem haelende eodon ciddon ongean J>one blindan J?aet he suwian sceolde {ALCHom I, 10 152.17) 'the men that were walking in front of

The Auxiliarization

(53)

of the English Modals

133

the Lord chided against the blind man that he should keep silent' [notice that ciddon acquires a directive meaning component through this occurrence of sceolde in the subclause] öaertogeanes gehaelde petrus blinde and healte. and deofelseoce. and t>a deadan araerde and cwaeö to J?am folce J^aet hi sceoldon forfleon J)aes deofles drycraeft ... ECHom I, 26 376.6)'... and said to (the) people that they should fly away from the witchcraft of the devil' [again the whole acquires a directive overtone]

Notice that the first group reflects the Futurity and Necessity/ Futurity uses of sceal/scealt discussed in 5.3.4.; those in the second group reflect the Necessity uses, but at the same time the combination reporting verb + subclause with sceold- exhibits a shift to directive. In both groups there is an element of desemanticization and grammaticalization of sceold-, either as a signal of (or approaching) past future or as a marker of a clause which acquires a directive overtone. 3 c. After a verb of opinion (1 occurrence) The instance is (54). (54)

Hwast öa fyrmestan on aernemerigen comon wendon Jm J?aet hi maran mede onfon sceoldon. öa underfengon hi aenlipige penegas. swa swa öa oöre C/ECHom II, 5 42.21) 'And see, the first who (had come) at dawn thought that they would get more pay [sceoldon expresses a past future]

3d. After verbs of fear (2 occurrences) We give one instance, (55). (55)

öa com crist on {jam timan t>e seo cynelice maegö ateorode. and se aelfremeda herodes J?aes rices geweold: wearö he micclum afyrht. and anJ)racode ]3aet his rice feallen sceolde. Jsurh tocyme J)aes soöan cyninges {JECHom I, 5 82.3)'... then he became much afraid and feared that his kingdom would/might fall by the coming of the true king'

Sceold- is a marker of past futurity (and at the same time a grammatical marker of a subclause after a main verb expressing fear).

134

Louis Goossens

4. Content clauses after nominal heads (10 occurrences) 4 a. After a noun expressing volition or command (6 occurrences) These parallel category 3 a; the head nouns are ce 'law', regol 'rule', has 'command' and the like. We give one instance. (56)

Feawa manna crist sylf gefullode, ac he forgeaf öone anweald his apostolon. and eallum gehadedum mannum £>aet hi sceoldon fullian mid godes fulluhte. on naman öaere halgan örynnysse ... {/ECHom II, 3 25.214) [onweald = 'power/command']

4 b. After other noun heads (40 occurrences) In three out of the four instances, sceold- is paraphrasable by should (e. g. (57)); the other one, (58), is another instance of a 'past future'. (57)

(58)

... and andwyrde him ]surh öa anlicnysse. J>aet hi him heora lac offrian sceoldon ... (AECHom I, 31 454.21) 'and [he] answered him through the parable that they should offer him their sacrifices' ... J)a com him andswaru fram J>am halgan gaste: J>aet he ne sceolde deaöes onbyrian aer J>an he crist gesawe (/ECHom I, 9 136.8) 'then came to him an answer from the Holy Ghost that he would not taste death before ...'

5. Subject clauses (2 occurrences) We have already had an instance of this low frequency category, viz. (40), where perhaps sceolde is a grammatical marker of the subclause in this context. The other instance has sceoldon paraphrasable as should. 6. Clauses of consequence I purpose (20 occurrences) This is the second most important category. Clearly sceold- is used to mark this type of adverbial clause. We give two instances; notice that consequence and purpose may be difficult to keep apart. (59)

Symon Ipa öa he J)am folce aetwunden waes. getigde aenne ormaetne ryööan innon t>am getefcaerpetrus in haefde J>aet he faerlice hine abitan sceolde (/ECHom I, 26 372.33) '... tied [up] an immense dog inside the gate in which [he] kept Peter so that it (he) might bite him fiercely'

The Auxiliarization

(60)

of the English Modals

135

Ealle öa öing öe crist dyde for us, ealle hi waeron s r gefyrn gewitegode. Jjaet men sceoldon gelyfan f>aet he is soöfaest ... {JECHom II, 1 6.121)'... all these were before prophesied so that men should/might believe that he is trustworthy ...'

7. Clauses of comparison (4 occurrences) These are introduced by swa (swa) (3 times) or ponne (after a comparative). The paraphrase is always should, as in (61). (61)

And se öe nele Godes bodan hyran mid rihte ne godcundre lare ne gyman swa he sceolde, he ... (WHom 17 47) 'And he that will not duly listen to God's messengers nor observe divine lore as he should ...' [sceolde is obligational and non-factual]

8. Other The remaining instance is (62). (62)

Crist wolde J?aet ... \>xt mancynn waere J>aes öe geleaffulre. and öaes J>e gewisre on hwaene hi sceoldon gelyfan ... {AECHom II, 1 10.267) '... so that man would be the more believing and the more certain as to whom they should believe [in]' [this is a clause introduced by a conjunctive dependent on a comparative adjective; sceoldon is obligational non-factual]

Since our interest is in the degree of grammaticalization that sceold- exhibits in this sample, we summarize the foregoing analysis from that point of view. We distinguish three 'degrees' of grammaticalization. (a)

Not grammaticalized: these are the instances where sceoldis the factual/indicative past tense equivalent of obligational sceal(t)jsculon. There are only three instances of this sort.

(b)

Partially grammaticalized Here we group the instances where sceold- retains its obligational sense, but in a non-factual context (it is equivalent to Present-day Eng. should/ought). There are 31 cases like this (8 in main clauses, 4 in relative clauses, 10 in object clauses after verbs of communication, 3 after head nouns (category 4 b), 4 in clauses of comparison, one instance in a subject clause, and the single instance in category 8).

136 (c)

Louis

Goossens

Grammaticalized Here we get the remaining 66 cases, for which we can distinguish the following types: — in object clauses after verbs of will or in content clauses after noun heads expressing an idea of will/command (20 + 6 = 26 instances) — in clauses of purpose or consequence (20 instances) — to express a 'past future' (once in a relative clause, 14 instances in object clauses after 'other' verbs of communication, once after a verb of opinion, once after a noun head of the category 4 b = 17 instances) — the two instances after verbs of fear, and the subject clause after hit is gewunelic (the latter instance might also be assigned to category (b) or even (a)).

5.3.6. What can we conclude about sculan as a whole? (i) There are two general indications that it has lost its status as an independent predicate already in (late) Old English. One is the loss of its non-finite forms. The other is the fact that it combines as a rule with an infinitive (phrase) which imposes its argument structure on the resulting combination sculan + infinitive (in the few instances where there is no infinitive, the infinitive is anticipated by a pronoun or ellipted: see 5.3.2.). (ii) There are arguments to claim that sceal/scealt/sculan are to be considered separately from sceold-. As appeared from our analysis sceold- only exceptionally functions as the past tense of scealjscealtj sculon. Moreover, the degree of grammaticalization is markedly higher for sceold- than for scealjscealt/sculon. We therefore conclude separately for the two. (a)

(b)

sceal/scealt (and there is little doubt that the plural exhibits a similar pattern) predominantly shows in its semantics a mixture of futurity meaning and the original necessity-meaning (68% in our sample), which we could regard as weak grammaticalization. As it happened, the sample showed an equal destribution of the (fully grammaticalized) expression of futurity (16%) and the fully obligational (necessity) sense (also 16%). For sceold- the fully grammaticalized uses predominate (66%), the rest of the sample shows partial grammaticalization, if only because the instances are in a non-factual context

The Auxiliarization

of the English Modals

137

(17%), with the exception of no more than 3 cases where sceold- retains its original meaning. 5.3.7. How can we account for the intricacies of sculan in Functional Grammar? It seems no longer justifiable (at least for my late Old English sample) to assign to sculan the status of an independent predicate. It necessarily combines with an infinitive (which occasionally can be ellipted or anticipated by a pronoun); the argument structure of the combination is completely determined by that of the combining infinitive (notice that, unlike with cunnan, this also results in nonhuman subjects for the combination, cf. 5.3.2.). On the other hand, the opposition between non-grammaticalized uses argues for a differentiated treatment. Since we have discarded the full predicate possibility, this amounts to a treatment in terms of predicate formation as well as in terms of predicate operators. Given the present state of the art in Functional Grammar our proposals are tentative; we give them in main outline only. (i)

As long as sculan retains (some of) its obligational meaning we would derive combinations with it by means of predicate formation rule (63) (63)

(ii)

Input: φ (xi) ... (Xn) Output: sculan φ (xj) ... (xn) Given that sculan has (some of) its obligational meaning we would expect restriction to what we have called 'events' in section 4, but this is only a strong tendency. Subjectivalization is to one of the arguments of φ (it is not restricted to x ls compare 30(d), where φ is passivized). Obviously, it will have to be noted in the lexicon that sculan can only occur in a finite form.

For those instances where the obligational meaning is no longer present and which we have called fully grammaticalized, we propose an origin as predicate operators. The plural implies that we have to consider at least two different predicate operators, viz. FUT and SUBJU. FUT would trigger off the 'pure future' uses, with the proviso that FUT has other realizations in Old English besides sculan. SUBJU is even primarily realised in a different way, viz. by subjunctive marking on the verb; sceold- is an alternative realization for

138

(iii)

Louis Goossens

the subjunctive in a number of specific subclauses, especially clauses of consequence/purpose and clauses dependent on a predicate expressing an idea of will/command. Instances of the so-called past future would have to be taken care of by two combined predicate operators (PAST and FUT). Some of the possibilities uncovered in the sample analysis could be dealt with as blends between predicate formation and a construction of the predication by means of predicate operators: 13 — The Necessity/Futurity cases, which predominate for sceal/scealt present us with the blend between predicate formation and the FUT operator — the non-factual obligational instances with sceold- can be treated as combinations of predicate formation and the SUBJU operator.

5.4. Conclusion: cunnan and sculan As we have pointed out in 5.1. the purpose of this section was to give an idea of how the Old English modals would figure on scale (9) in section 4. The approach chosen was to look at the case of cunnan and sculan on the basis of two late Old English samples. It appeared that there are arguments to place cunnan and sculan in rather different positions. Cunnan is still predominantly an independent predicate with the beginnings of a shift to a usage which would come under predicate formation. Sculan is no longer a full predicate, but partially belongs under predicate formation, partially under predicate operator, with certain uses presenting us with blends between the two. Differentiating between sceal/scealt/sculon (represented by sceal) and sceold-, the following table summarizes the proposals introduced in this section: Full predicate cunnan

Predicate formation

Predicate operator

(cunnan) (sceal) ((sceold-))

sceal (sceold-)

(sceal) sceolde

The Auxiliarization

of the English Modals

139

(the unbracketed positions indicate the predominant use; positioning on an intersecting line points to a blend).

6. General conclusion 6.1. Conclusions with respect to Functional Grammar (i)

Our analysis of cunnan and sculan has proved, I think, the value of grammaticalization scale (9) within Functional Grammar. It also appeared that a three-point scale was not refined enough. Opting for blend positions between full predicate and predicate formation on the one hand, and between predicate formation and predicate operator on the other, gives us two additional points. Further research may reveal other and/or additional refinements.

(ii)

In deciding about the choice between independent predicate status or predicate formation for a modal verb a crucial argument appeared to be whether the modal verb brings along its own argument structure (in which case we take it to be an independent predicate) or whether a combining predicate, which in the case of the (pre)modals results in an infinitive (phrase), imposes its argument structure on the whole combination (this gives rise to a treatment under predicate formation). Cunnan and sculan showed up clear differences in this respect.

(iii)

To assign predicate operator status to a given modal it should, in our view, in addition to the fact that it does not have an argument structure of its own, be used in specific grammatical functions, such as the expression of tense, the marking of certain types of subclauses and the like. What those grammatical functions may be is subject to further investigation within Functional Grammar.

(iv)

Although I have not yet made this explicit, the reader may have inferred that I would view the shift of cunnan to can and of scealjsceolde to shall and should on grammaticalization scale (9) as follows:

140

Louis

Goossens

Full predicate

Predicate operator

Predicate formation

cunnan (cunnan) (can) can (sceal) ((sceold-))

sceal (shall)

(sceal) shall

(sceold-) (should)

sceoldshould

This gives us only part of the picture, of course. E. g., it does not indicate that shall is very much regressive in Present-day English, especially in American English. Or, more importantly, it does not reveal to what extent a given predicate operator is expressible by a given modal (in Old English morphological marking on the verb is a comparatively much more important correlate of SUBJU than in Present-day English, whereas should is comparatively more important than sceold-). (v)

A final point with respect to grammaticalization scale (9) is that it appears to correlate with a decreasing specificity in the (semantic) combinatorial possibilities for the item that can be shown to develop from independent predicate to the other end of the scale. We noted for cunnan that it had a specific argument structure in Old English in which we could characterize the first argument as cognizant and the second as a knowable entity. The transition to predicate formation involves replacement of the second argument by some State of Affairs, originally one that involves a cognizant dimension, the next step being (as for ModE can) that this cognizant element no longer restricts the combining State of Affairs. Similarly for sceal the initial position in the schema involves a State of Affairs which can be characterized as an 'event'. As we move along to a function as predicate operator that restriction is removed. A similar point can, of course, be made about the shift from {{sceold-)) to Present-day English should.

The Auxiliarization

of the English Modals

141

6.2. Grammaticalization and auxiliarization of the English modals In addition to the conclusions in 6.1., I would like to point out the following: (i)

The grammaticalization of the English modals (and hence their auxiliarization) is already a fact in Old English: to some extent for sceal and predominantly for sceold-.

(ii)

In a discussion of auxiliarization of the English modals we must consider the different items individually, not globally. Even items that could be thought of as belonging to the same paradigm set may need to be investigated separately, as appeared from the instance of sceal/sceold- j shallI should.

(iii)

Cunnan and sculan present us with two extremes as regards grammaticalization in Old English. Can and shalljshould do so as well in Present-day English, with the proviso of a further shift on the grammaticalization scale away from the full predicate position for both items.

(iv)

The directionality of the changes on the grammaticalization scale seems to go in one sense only, viz. towards increased grammaticalization. A fully detailed investigation of all the modals will still have to confirm this, but I am rather confident that this will be the case.

Notes 1. This is Lightfoot's term for the ancestors of the Present-day English modals (see Lightfoot 1979:101). 2. Both in his 1974 article, which was the basis for my Durham paper, and in his 1979 book (see Lightfoot 1979:153). 3. See e.g. Traugott (1972:199) and Visser (1969), where ample exemplification is to be found. 4. We list the relevant examples with a few words of comment: (a) Ac uton don swa swa god tiehte: f>aet ure goodan weorc beon. on öa [sic] wison mannum cuöe ( / E C H o m I, 11 180.23) [on öa [R ö®m] wison mannum is a prepositional qualification of the adj. cuöe rather than an agent-phrase]. (b) and ge magon Jje cuölicor to him clipian gif heora lifes drohtnunga eow t>urh lareowa bodunge cuöe beoö (/ECHom I, 37 556.25) 'if the conditions of their life are known to you through the message of teachers'. (c) jjaer beoö cuöe ure ealra daeda eallum öam werodum ... {/ELS (Ash Wed) 172) 'There/then are known the deeds of all of us to all multitudes'.

142

Louis Goossens

5. The brackets indicate that these instances do not really figure on a par with the others; the inclusion is based on the idea that what we get as the subject in real passive sentences corresponds to the (object) complements in their active counterpart. 6. Indeed, also the object NPs denoting persons and languages have to be interpreted with emphasis on their 'knowable' properties. 7. This could be captured in Functional Grammar as an instance of predicate formation. In this case, however, predicate formation would not signal any increased grammaticalization. 8. Note that Old English has no gerund and that the (nominal) inflected infinitive only functions after prepositions taking a dative. 9. The instance is Ben RG1 26: [Haec complens expectat nec cotidie his suis sanctis monitis factis nos respondere debere] Jjis gefyllende anbidiaö mid daedum we sculan. I suspect that sculan glosses debere, but I have not been able to check the edition referred to. 10. Non-indicative means as a rule subjunctive, though there are a considerable number of non-indicatives that are semantically different from the kind of grammatical meaning normally associated with 'subjunctive'. Hence the non-committal term 'non-indicative'. 11. See e.g. Campbell (1959:§379, §735 (e) and (g)). 12. See also 5.3.2. 13. Note that the acceptance of blends would be/is an innovation within Functional Grammar.

References A. 1959. Old English grammar. Oxford: Clarendon Press. C. 1981. Functional grammar. Dordrecht: Foris. GOOSSENS, LOUIS. 1984. The interplay of syntax and semantics in the development of the English modals, in: English historical linguistics: studies in development, ed. by N. F. Blake & C. Jones, 149-159. CECTAL Conference Papers, No. 3, University of Sheffield. GOOSSENS, LOUIS. 1985 a. Modality and the modals: a problem for Functional Grammar, in: Predicates and terms in Functional Grammar, ed. by A. Machelt Bolkestein, Casper de Groot and J. Lachlan Mackenzie. 203—217. Dordrecht: Foris. UOOSSENS, LOUIS. 1 9 8 5 b. Differentiating the English modals in Functional Grammar, in: Antwerp studies in Functional Grammar, ed. by Jan Nuyts, 49 — 71. Antwerp Papers in Linguistics 39. LEHMANN, CHRISTIAN. 1 9 8 2 . Thoughts on grammaticalization: a programmatic sketch. Vol. I (Arbeiten des Kölner Universalien-Projekts 48). LIGHTFOOT, DAVID. 1 9 7 4 . The diachronic analysis of the English modals, in: Historical linguistics I. Syntax, morphology, internal and comparative reconstruction, ed. by J. M. Anderson & C. Jones, 219 — 250. Amsterdam: North Holland. LIGHTFOOT, DAVID. 1 9 7 9 . Principles of diachronic syntax. Cambridge: Cambridge University Press. CAMPBELL,

D I K , SIMON

The Auxiliarization

of the English Modals

143

PAOLO HEALY, ANTONETTE D I , a n d RICHARD L . VENEZKY. 1 9 8 0 . A m i c r o f i c h e c o n c o r d -

ance to Old English. University of Toronto. TkAUGorr, ELIZABETH CLOSS. 1972. A history of English syntax. New York: Holt, Rinehart and Winston. VET, C. 1981. Subject assignment in the impersonal constructions of French, in: Predication and expression in Functional Grammar, ed. by A. Machelt Bolkestein et al., 143 — 163. London/New York: Academic Press. VISSER, F. TH. 1969. An historical syntax of the English language. Part Three. First half. Syntactical units with two verbs. Leiden: E. J. Brill.

Acknowledgements Thanks are due to the participants of the workshop on auxiliarization at VIIICHL, especially Henning Andersen, Joan Bybee, Simon Dik, Martin Harris, Harm Pinkster and Paolo Ramat. Needless to say, the shortcomings of this paper are my own.

Louis Goossens Universitaire Instelling Antwerpen Departement Germaanse Filologie Universiteitsplein 1 B-2610 Wilzijk

On the Use of the Modal Auxiliaries Indicating 'Possibility' in Early American English Merja Kytö

1. Introduction The rise of the British Empire, coinciding with the arrival of the Modern Age, gave birth to a number of overseas varieties of the English language. The British explorers and fishermen found their way to the coasts of the New World as early as the 16th century, but the settlement of North America could be said to have begun seriously with the settlement of Jamestown in Virginia in 1607 and with the arrival of William Bradford and the separatists in Cape Cod in 1620. The exodus of those willing to put their fate at stake on the unknown continent gained more impetus after John Winthrop and his 500 fellow passengers had successfully founded the Puritan-ruled Massachusetts Colony in 1630. Out of the some five and a half million inhabitants of the British Isles, 50 000 to 80 000 left their old mother country for a new one. From the 1680s onwards they were joined by Germans, Irishmen, Scots and French, who soon left the decreasing number of English immigrants in the minority. It has been estimated that by the early 1700s the number of settlers and their descendants had reached the figure of a quarter of a million. Except perhaps for the highest and the lowest, all the ranks of society were represented. From the very beginning the communities in the North and the South developed in different ways: while the tobacco plantations of the South drew settlers from dubious and rootless backgrounds — mainly adventurous-minded young men — those heading for the North came with their families and aimed to stay. Furthermore, the Puritan New England area attracted many learned clergymen and administrators whose university background guaranteed from the beginning that the world of education and literary activities would become highly valued. 1 These two factors, one would think, must have been of significance to the forming of the new linguistic communities.

146

Μ er ja

Kytö

Before describing my material and reporting the findings I have come upon in my study of modal auxiliaries, I would like to briefly discuss the conditions set on the development of the English language in the early Colonies. Migration and isolation of the language and the new socio-cultural settings and contacts of the language users worked in opposite directions, both accelerating and retarding the development. Owing to the levelling effect of the contacts between the speakers of different dialects or socially marked varieties, there is not much point in looking for the origins of American English in a single regional dialect.2 The wish to adapt oneself to the new conditions, in general, must have required, as such, a positive attitude towards changes in one's use of language as well. Further factors favoring the change in the language habits of the early Colonists included the special status that spoken language — the very ground for linguistic development — was enjoying in the Colonies. In New England, in particular, the congregationalist tradition challenged the common man in an altogether new way. Communal institutions such as the town meeting, which required the presence of all members of the congregation, was a new forum for common folk. Another more or less sanctioned form of oral delivery was what became known as the public relation: in order to gain full membership in the congregation one had to declare one's faith in front of the congregation in a convincing way.3 In courts one was supposed to be able to speak for oneself or for one's neighbor — professional lawyers, as advised in the Bible, were regarded as dubious characters in general.4 The core of all Puritan institutions was the sermon, delivered as custom required, ex tempore or with the support of sketchy notes. The members of the congregation, twice or three times a week, were expected to memorize the main points of the sermon either by ear or by short notes taken during the delivery. The preaching style — known as the plain style — aimed at simple and practical presentation and avoided all ornamentation characteristic of traditional rhetoric. 5 Among the factors retarding the development of language was the literary tradition of the old home country which made itself particularly felt in the big coastal cities which provided an opportunity for frequent contacts with England. At the same time, on the other hand, British English developed independently, leaving the migrant language in an archaic state, relatively speaking. The emigrant, on

Modal Auxiliaries

Indicating

'Possibility'

147

the other hand, if he was aware of the possible degeneration of his language, might well have clung to the norm he thought he knew. One would assume, further, that Puritanism as such functioned as a stagnating force, resistant to new influence of any kind. The pressures that the congregation directed towards its members should have, similarly, created a resistance to any non-conformist phenomena. Control and invigilation was made easier by the pattern of settlement which favoured the founding of small and compact villages. Nor could the prestige enjoyed by educational institutions and literary culture encourage any ideology that would favor change.

2. Material and method The socio-demographic and cultural factors referred to above point to New England as a well-grounded starting point for a study of early American English idiom. The documents available — which reflect the basic motives of migrant literature — are representative of utilitarian penmanship in many ways. The texts include such types as private and official letters, diaries, travel journals, history writing, scientific writings, sermons, witness depositions, appeals and answers to the courts etc. The corpus I have collected covers the period from the 1620s to the 1720s and consists of some 926000 words. The texts supposedly recorded verbatim are relatively rare: they stand for a fifth of my corpus providing some 180000 words. This kind of material offers a good basis for a variation-based study on diachronic principles. The language user selects consciously or unconsciously amongst the variants available to express a specific meaning. The study of the distribution patterns of the variants — when collected from texts representing different periods, media and styles — should help us observe the development. 6 In my own study I aim to describe the use of the modals CAN, MAY, COULD, and MIGHT 7 and to account for the factors influencing the choice of the variant in a given context. A diachronic perspective will be provided by what we already know about the pre- and post-seventeenth century development of the modals and by the changes to be observed between the language of the first-generation settlers and their descendants. The study of the modals in this period is particularly interest-

148

Merja

Kytö

ing, as the system of modals met its last major restructuring process in the very early Modern English period. 8 Classification of the texts on the basis of a number of extralinguistic criteria will help us observe the factors which supposedly influence the variation. In my study such factors include the nature of the communicative situation (which can be private or public), the speaker's or writer's relationship to the audience (close or distant), the educational background of the language user and the purpose for which the text has been produced (subject-matter and stylistic requirements). On the basis of these factors the texts have been defined as representative of formal or informal uses of language of either non-speech-based or speech-based origin.9

3. The criteria of classification The models presented for the purpose of analyzing modals have so far been mostly monosemantic — the modal is assumed to have one basic meaning — or polysemantic — the modal is assumed to have a number of independent or related meanings — or both. The lastmentioned possibility seems most rewarding from the point of view of a variational study on historical principles: independent categories of meaning can be distinguished without denying the existence of semantic indeterminacy so characteristic of the modals. Besides, shifts of meaning are known to imply the existence of categories common to different meanings, as these categories allow contextual shadings and neutralizations gradually to become permanent. 10 In traditional studies each modal has been attributed a number of 'uses' such as 'ability', 'permission', 'possibility', 'potentiality' and so on. However, too refined a classification of these intuition-based uses would not seem to be the best way of approaching material that is remote in time.11 Difficulties arise, moreover, as more or less the same uses have been attributed different interpretations. Thus, for instance, CAN has been seen to both express and lack the 'ability' meaning in passive constructions: (1)

'ability' porveythe therfor that thei mow be squarid there, and sentte hedre, for here can non soche be hadde in this conttre (Paston Letters, 11.73.27, 1445)

Modal Auxiliaries

Indicating

'Possibility'

149

'see-to-it therefore that they (the joists) may be squared there and sent hither, for here can none such be had in this partof-the-country' 12 (2)

'ability' He can ( = is able to) beat the world champion. φ 'possibility' The world champion can be beaten by him. 13

The problem is particularly difficult as it has been traditionally held that the original meaning of OE M A G A N — i.e. 'physical capacity' — gradually weakened and finally disappeared in the seventeenth century giving way to CAN used in this sense. Obviously, such notion-based distinctions cannot be totally avoided in a study of modals. In a study of variation the classification of the uses should, however, be anchored in the context as it is only in a firmly defined context that we can reliably make conclusions about the factors that play a role in the choice of a variant. The two poles of modality, 'possibility' and 'necessity', have been regarded as basic notions in the study of modals. 14 Of the many kinds of modality linguists have ascribed most significance to the difference made between epistemic and non-epistemic modality. In an epistemically colored use of a modal the speaker expresses his subjective opinion about the truth-value of the proposition; in a nonepistemic use no such statement is made. 15 Compare examples (3 a) and (3 b), in which the epistemic meaning of the modal MAY can be paraphrased with the phrase 'it is possible that' and the non-epistemic meaning with the the phrases 'it is possible for ... to' or 'nothing prevents': (3) a.

'epistemic possibility' I may be a few minutes late but don't know (Survey of English Usage, S.7.3E.6) 'It is possible that I am' 16

(3) b.

'non-epistemic (root) possibility' I am afraid this is the bank's final word. I tell you this so that you may make arrangements elsewhere if you are able to. (Survey, W.7.9.37) 'It is possible for you to make'

150

Μ er ja

Kytö

In indeterminate examples both interpretations seem possible: (4)

It is important to note that where high concentrations are theoretically possible in the plant evaporators the time required to build them may be considerable. (Lancaster Corpus, 6-1699) 'it is possible that it is' / 'it is possible for it to be'

In my study the first rough classification divides the examples with CAN, MAY, COULD or MIGHT into cases of 'possibility' or 'necessity'. Cases of'possibility' are defined, further, as of'epistemic', 'non-epistemic' or 'indeterminate' meaning (table 1): Table 1. Non-Past:

CAN, MAY 3542 COULD, MIGHT

55.4

Past:

COULD, MIGHT

1896

29.6

Non-past:

MAY, MIGHT

222

3.4

Past:

MIGHT

112

1.8

Non-Past:

MAY, MIGHT

457

7.1

Past:

MAY, MIGHT

144

2.3

Non-Past:

CAN

4

0.1

Past:

COULD

19

0.3

6396

100.0

Non-epistemic

Possibility

Indeterminate

Epistemic

Necessity

Epistemic

Total

Cases of 'necessity' are all epistemic, represented by the negative forms CANNOT and COULD NOT used to indicate logical conclusion as in: (5)

'epistemic necessity' Paul can't be in Liverpool. 'it is necessarily the case that Paul is not in Liverpool' φ Paul must be in Liverpool.

Besides modus, tempus is taken into consideration: the examples are classified into 'non-past' or 'past' categories. The third main criterion of classification is the sentence function, which divides the examples

Modal Auxiliaries

Indicating

'Possibility'

151

into declarative sentences, questions or exclamations. This criterion is necessitated by the unclear status which epistemic uses may have in questions.

4. Results The examples representing the category 'non-epistemic non-past' covers a good half of the material, some 3500 examples of the nearly 6400 attested. All the four forms appear in this category. The category 'non-epistemic past' represented by MIGHT and COULD includes 30% of the examples. This leaves the 'epistemic' category represented by MAY and MIGHT with only some 10% of the material. The role of 'indeterminate' examples is even less significant: the category includes six per cent of the examples. In the following I will discuss the area of 'non-epistemic non-past' meaning in more detail (I restrict my observations to declarative sentences only). In other words I will discuss the use of the four variant forms in examples such as (6 a) and (6 b): (6) a.

You may trust to my Wife, that She will imploy her utmost to accommodate John, or any other of yours what we can. (1702)

(6) b.

if my daughter nan could but starch a little better, I should be very glad she might supply her place — (1649)

As hypotheticality is the feature distinguishing MAY from MIGHT and CAN from COULD in this category, the variation of the forms will be looked at in the combinations CAN vs. MAY and COULD vs. MIGHT. MAY is the form which occurs in 60% of the cases as against the 40% with CAN (for the illustrations see figure 1 given in the Appendix). MIGHT, similarly, prevails in some 60% of the instances. MAY prevails most clearly (in nearly 70% of the cases) in non-speechbased texts defined as formal. CAN, on the contrary, gains ground on the informal side (around 50%). Letters in particular, show this tendency: while MAY dominates in the official letters with 64%, CAN is the form preferred in private correspondence (55%). The figures for COULD, similarly, rise on the informal side and particularly in private letters.

152

Merja Kytö

CAN competes on equal terms with MAY in the texts which come closest to the spoken language (see figure 2 given in the Appendix; only the examples from the category of 'direct speech' have been included in figures 1 and 2; the figures available at this point of analysis include both 'non-epistemic' and 'indeterminate' examples). A difference can be seen, again, in the figures obtained for the formal and informal categories: MAY appears in 60% of the cases recorded in utterances defined as formal and CAN in 60% of those defined as informal. The uneducated witnesses recorded in the Depositions and Salem Witchcraft Trials, in particular, favor CAN. It is of interest to notice, too, that in the only non-speech-based text showing a tendency to favor CAN — that is, in Cotton Mather's report of famous cases of witch identification — most instances attested appear in utterances reported to have been made by the characters involved. Owing to the small number of examples recorded with MIGHT and COULD, no conclusions can be made on the use of the forms in the speech-based texts. On the basis of the figures obtained it seems thus that the forms prevailing in formal and literary language are MAY and MIGHT. CAN and COULD, on the other hand, are characteristic of the more colloquial registers, represented by informal writing and speech. The study of the semantic and syntactic environment shows that negation clearly favours the use of both CAN and COULD (table 2): Table 2. CAN

COULD

MAY

Declarative Sentence

MIGHT

Total

Total Total

Affirmative %

614 (25.6)

1782 (74.4)

2396 (100.0)

109 (38.7)

173 (61.3)

282 2678 (100.0)

Negative %

726 (88.0)

99 (12.0)

825 (100.0)

29 (74.4)

10 (25.6)

39 (100.0)

Total

864

3542

Out of the 825 examples with CAN or MAY, only 99, or 12%, have MAY; the figures for MIGHT are relatively higher — and we will see later why — but cover no more than a quarter of the examples attested with the non-past forms (from this point on the 'indeterminate' examples are removed from the figures and the data for the

Modal Auxiliaries Indicating 'Possibility'

153

speech-based texts include the categories of 'direct speech', 'indirect speech' and 'scribal comment'). I will take up the use of the variant forms first in affirmative and then in negative sentences. To define, further, the syntactic environment, examples will be classified according to the clause type in which they occur. The distributions of the forms are given in table 3 a and table 3 b: Table 3 a. Clause type

CAN

MAY

Total (100.0)

Final clause

337 (100.0)

337

1 (0.4)

274 (99.6)

275

Other

422 (28.6)

1051 (71.4)

1473

Conditional clause

108 (61.0)

69 (39.0)

177

Comparative clause

83 (61.9)

51 (38.1)

134

Total

614 (25.6)

1782 (74.4)

2396

COULD

MIGHT

Total

-

Noun clause indicating volition

Table 3 b. Clause type

(100.0) Final clause

-

51 (100.0)

51

Other

60 (35.5)

109 (64.5)

169

Conditional clause

49 (79.0)

13 (21.0)

62

Total

109 (38.7)

173 (61.3)

282

154

Merja

Kytö

Final clauses exclude the use of CAN and COULD: (7) a.

'final clause 4- may, might What shall I do, that what is yet Lacking in the Image of God upon me, may be Perfected? What shall I do, that I may Live more Perfectly, more Watchfully, more Fruitfully before the Glorious Lord? (1710)

(7) b.

Mr Cotten: Hear we haue bine the mor patient with yow and haue the Longer delayd this bwssines, that we might see whether the Cowrt Could finde owt any Iniqwetie in thear proceedinges. (1641)

This is understandable in light of the development of the periphrastic subjunctive. As Mustanoja in his Middle English syntax (1960) points out, the use of the subjunctive and its equivalent, which consists of a modal and an infinitive, is natural in a final clause, which inherently expresses a prospective event or state. In Middle English the modals used to build the periphrastic form were MAY, SHALL and MOTE, while other modals were rarely found. That neither CAN nor COULD appears in final clauses in my material is thus only to be expected. The use of the periphrastic subjunctive is, similarly, typical of noun clauses which follow volitional expressions such as wishes, exhortations and commands and thus clearly indicate the presence of some kind of modal color:17 (8)

'hope, desire etc. 4- may I send home the Bacheler, and desire your helpe for her disposing. I must of necessity have her returne heere for I may shortly have much vse of here: but I desire they may goe for shares and victuall them selves, which John Wood, and his company are willing to doe. (1636)

CAN appears only once in the 275 examples of the present tense forms in noun clauses expressing volition. No more than twenty examples of these clauses with MIGHT or COULD were attested. Of these MIGHT appears in fifteen examples, and in twelve cases the introductory verb is hope or desire: (9)

'hope, desire etc. + mighf I haue desired longe to speake to you about my poor cosen

Modal Auxiliaries

Indicating

'Possibility'

155

Feakes [Elizabeth Feke, a lady of scandalous fame at that time]. — and now I hear Patricke [Captain Daniel Patrick, died in 1643] is cut of wich makes me hope that by the vse of some good means theer might be more hopes to reduce her. (1644) The five cases with COULD all have the introductory verb wish appearing in the more or less idiomatic phrase I wish I could. The use of COULD underlines the lack of 'ability' of the subject or the impossibility as to the fulfilment of action. The idea is more or less Ί wish I could, but I can't': (10) a. 'wish + could I wish I could tell you some certainty of any good mines in this North America. (1668) (10) b. Dear brother, — I am sorry to hear you are ill, and wish I could be with you. (1706) The clause types favouring the use of CAN and not of MAY, on the contrary, are comparative and conditional clauses; COULD, similarly, is the form preferred in the conditional clauses (the number of comparative clauses attested was too small to base conclusions on). Before taking up these types I will discuss in more detail the types in which MAY and MIGHT prevail. Clause type analysis as such is not sufficient to account for the variation in these cases and a set of criteria is needed to specify, further, the environment of the variant form. As examples with the passive voice are sometimes difficult to place into notional categories, the first rough classification will divide the examples into the categories of 'active' and 'passive'. The examples with active verbs will be classified, further, into the notional sub-categories of 'ability', 'permission' and 'neutral possibility'. With the 'ability' meaning the possibility of action is determined by inherent properties of the animate subject; with the 'permission' meaning the utterance can be paraphrased with the words 'be permitted' or 'be allowed to', and the subject is animate or an authority. In both cases the verb denoting the action is agentive. With the 'neutral possibility' (the term used by Coates is 'Root "Possibility"') there is no clear indication of the restriction or of inherent properties of the subject, but external circumstances make the state or action referred to in the verb possible; the paraphrase is 'it is possible for ... to'. 18 Clear 'core' examples

156

Merja

Kytö

can be classified into the 'ability' and 'permission' groups relatively reliably. But cases in which it is not clear whether inherent ability or permission is implied are more problematic: (11) a. 'permission' / 'external circumstances' and I was angry at at the School of Applied Linguistics when I found that as a student I couldn't get a key (Survey, S.2.4A.45)19 (11) b. 'ability'/'external circumstances' I wrote back as am worded it as diplomatically as I could (.Survey, S. 1.3.14) In these examples the fulfilment of action seems to be the result of a mixture of these factors. The paraphrases are not of much help as there is a very close connection of meaning between the more specific meanings of 'to be able to' or 'to be permitted', which imply the more general concept expressed by 'it is possible for ... to'. I have included these examples in the category of 'neutral possibility'. 'Ability' is the only sub-group to rule out the use of MAY totally. No clear examples with COULD were attested, either: (12)

'ability' + can They say he came from a trading howse which Plymmouth men haue at Qunnihticut, and can speake much Indian — (1637)

'Permission', on the contrary, is the ground for MAY and MIGHT: (13) a. 'permission' -I- may/might My service to your self and Mrs. Richardson: To Mr. Woodbrigg, with whome if you see caus you may Comunicatte the busines of this Letter. (1687) (13) b. I read a Clause out of Dalton shewing when an officer might break open a House. (1705) CAN is recorded only in two clear context-related cases out of the 106 instances with CAN or MAY in this category:

Modal Auxiliaries Indicating 'Possibility'

157

(14) a. 'permission' + can Mary was told that her fault was greatly agrauated, that she haveing beene tempted once or twice, yt she would neither tell her Master, nor locke ye doore, but he can come a first & a second time, & theres no resistance, but the doore stands open, as if she were a common Harlot. (1662) (14) b. — & when he cometh, theres no opposition, when she might have cryed out & cald her master, but he can come a first, a second, a third time, & she is willing to it — (1662) It is in the category of 'neutral possibility' that the forms meet (table 4): Table 4. 'Neutral' Texts

CAN

MAY

Non-speech-based

Total

COULD

MIGHT

(100.0)

Total (100.0)

Informal

88 (47.8)

96 (52.2)

184

22 (88.0)

3 (12.0)

25

Formal

136 (21.3)

504 (78.7)

640

34 (72.3)

13 (27.7)

47

Speech-based

26 (27.1)

70 (72.9)

96

3 (60.0)

2 (40.0)

5

Total

250 (27.2)

670 (72.8)

920

59 (76.6)

18 (23.4)

77

The overall figures for the non-speech-based texts as a group and for the non-speech-based formal and informal texts point, again, to the use of CAN and COULD in colloquial registers. The low number in the frequency of CAN in the speech-based texts is due to the high number of examples representing the categories of 'indirect speech' and 'scribal comment'. Another factor linked with the use of CAN and COULD is the animate subject. As table 5 shows, CAN appears with the inanimate subject in only seven per cent of the 'neutral' cases with the active voice, while the corresponding figure for MAY is thirty per cent:

158

Merja Kytö

Table 5. 'Active'

Subject

'Neutral'

+ Animate

— Animate

CAN

232 (92.8)

18 (7.2)

250

MAY

462 (69.0)

208 (31.0)

670

Total

694 (75.4)

226 (24.6)

920

Total

MAY is linked more frequently than CAN with the existential subjects there and it. CAN, on the other hand, has been associated with animate subject from the beginning — its original use as an equivalent of Lat. scire 'to know how' presupposed the presence of an animate (or human) subject. Different collocational phrases are among the factors that one would presume might favor the use of one form or the other. The main interest all the time is in the category of 'neutral possibility'. For lack of space I restrict myself to the use of CAN and MAY only, omitting COULD and MIGHT. Three environments clearly favor the use of CAN. A CAN environment is seen in the clauses following a superlative or suchlike expression (in 75 examples out of the 80 attested). Though the fulfilment of action depends not only on the 'ability' of the subject but also on the enabling circumstances, the superlative phrase underlines the 'ability' meaning and favors the use of CAN: (15)

'superlative expression + can' Your poor sister will be left alone. Bring a pair of ordinary white specticles, in a wooden case, for old Samll Bebe, which I have promised him; they cost but about a shilling, — let them be of the oldest sight you can get, be shure. (1711)

The same holds with a number of other CAN-prone environments. Thus, for example, relative clauses introduced by what or the combination what + noun are linked with the 'ability'-related meaning: (16)

'what + cart I had allmost made a bargane withe Goodman goold but we

Modal Auxiliaries Indicating 'Possibility'

159

shall now stay, yet I thincke you will wante attenat to youer mind for he will depend no longer, yet I will doe what I can to perswad hime to stay till you come. (1649) (17)

'what + noun + can The beddin sent you must be responsible for when I call for them. Therefor procure flocks with what speed you can. (1689)

To return to the comparative and conditional clauses — in which, as pointed above, CAN and not MAY prevails — table 6 reveals a number of uses characteristic of each verb. Table 6. Comparative clause

'Neutral'

Sub-Type

CAN

comparative + than

17

2

19

so/as

21

8

29

— so/as far as

12

2

14

— so/as/so as ... be

—

15

15

9

9

1

5

6

51

41

92

...as

so/as/so as ... that such as ... that Total

MAY

Total

A comparative expression followed by the conjunction than introduces, almost invariably, the verb CAN (17 cases out of 19). CAN, similarly, follows the preceding adjective or adverb surrounded by the conjunction combination solas ... as, and occurs, in particular, after as far as: (18)

'comparative expression + can' — yet good sir remember the peril and disaduantage of deiection of spirit — but what doe I troubleing you with thees wheerin you are so much better furnisht then I can supply: I much bewayell my insufficiencie to be assistant in anny kynd to you. (1640)

160

(19)

(20)

Μ er ja

Kytö

'as ... as 4- can' I received yours dated July 8, 1687, since which I have one from Mr. Hull giving an account of £20. received which is very well because I am in his debt, and therefore entreat you to send more to him as fast as you can. (1688) 'as far as + can" Mr Cotten: We haue drouen this as far as we cane. Let us now proseed to examine the untruth that are Layd to her charge. (1641)

In all these examples the expression of comparison marks the environment by highlighting the 'ability'-colored meaning characteristic of the use of CAN. With the main verb to be, on the contrary, the form chosen is MAY (in all the fifteen cases attested). As examples (21 a —c) show, the meaning conveyed by the zero-subject phrase as ... as may be does not come close to that of 'ability' but rather to that of 'permission'. The paraphrase could be something like 'as circumstances allow': (21) a. 'as ... as may be' Your wife and I came hither with the Govr in his coach but just now, and the babe came in our coach with Madam Dudly and your mother. It is intended to be layd into the tomb tomorrow about this time, as decently as may be. (1711) (21) b. I am yet not redy to get away, but shall endever as fast as may be. (1716) (21) c. I now resolve to leave the Matter to you, to procure my Satisfaction with as much convenient Speed as may be: for I want the money. (1705) A similar kind of 'permission'-related non-personified use of MAY in the phrase if it may be accounts for the majority of cases attested in conditional clauses with an inanimate subject (table 7) (20 instances out of the 31 attested; notice, again, the closer link of inanimate subject with MAY): (22)

'if it may be' You may shew so much of it as concerns your affairs, and it must be returned by the next, if it may be. (1706)

Modal Auxiliaries Indicating 'Possibility'

161

Table 7. 'Neutral'

Conditional clause Subject

CAN

MAY

-I- Animate

70

12

82

— Animate

1

31

32

71

43

114

Total

Total

With animate subjects, however, we again encounter a stronghold for CAN (70 examples, or 85% of the instances, as against the 12 with MAY): most examples seem to be quite close to the core meaning of 'ability' normally expressed by CAN, as the conditional clause itself puts in doubt the possibility of the action taking place: (23)

'if + (animate subject) + can' I think I can procure the originall deed to the thirteen persons, and, if I can, shall get your coppy strengthened by sending an attested one from hence. (1706)

Conditional clauses are significant, too, in that both the variant forms appear in examples with passive constructions. In the following examples CAN and MAY seem to be used interchangeably (notice the use of the forms collocating with the same main verb in the texts of one and the same author): (24) may

can

nayles of 6 10 and 20. peircer bitts. Sithes for grasse, and two brush sithes. — Trading cloth good store, if mony may be had. (J. Winthrop, 1634)

some more Cowes would be brought, especially] 2: new milche, which must be well mealed and milked by the waye and some goates [and] especially] sheepe (if they can be had). (idem, 1630)

& I know not why wee should not imitate them herein except some other way may be thought of wherein such a stocke may be imployed — (R. Keayne, Will, 1653)

— that so my estate may receive no losse that way or if any other or jus way can be thought of betweene my overseers & she I shall leave it to them, (idem)

162

Merja Kytö

— but alsoe since yow haue gone abowt to Lay blame vpon the church as if thay had not delt Justly with yow, if it may soe be proued, than I hope any Brother and the whole church wilbe content — (R. Keayne, Meeting Records, 1640)

Therfor Let us proseed to that which is playne and admonish her first for that; and if any other thinge cane be proued agaynst her, or this Lye cane be made mor manifest, it will hinder the churches farther proseedinge with her — (idem)

The form originally used in the affirmative passive constructions was, according to Visser, MAY, and CAN began to appear in writing only after the 15th century. Grammarians agree that it is difficult to see any difference between the uses, unless one is ready to say that CAN often stresses the capability as such.20 I will take up, finally, the use of the four forms in the negative sentences. The overwhelming use of CAN in negative contexts has already been pointed out. The use of MAY in final clauses and clauses expressing volition is only to be expected (table 8): Table 8. Clause type

MAY NOT

Final clause

34

Noun clause indicating volition

21

Other

44

Total

99

The negative form CANNOT does not occur in these environments more than twice out of the 726 examples attested. Of the ten cases with MIGHT NOT eight occur, similarly, in final clauses. Neither clause type analysis, nor textual distributions seem to carry us far with the remaining 44 cases with MAY NOT. The examples must thus be classified into notional sub-categories (table 9): Table 9. Sub-group

MAY NOT

'negated permission'

38

'neutral possibility'

6

Total

44

Modal Auxiliaries Indicating 'Possibility'

163

As no clear instances with CANNOT could be found, MAY NOT seems to be the form used to express the 'core' meaning of 'negated permission' (33 of the cases attested are in the active and five in the passive voice; for factors possibly explaining the prevalence of MAY NOT, see Martin Harris's comments, p. 185 — 6: (25)

'negated permission' + may not We may not judge of Men meerly by outward accidents which befal them in this World, since all things happen alike unto all, and no man knoweth either love or hatred by all that is before them. (1684)

CANNOT appears, however, in a pragmatic use, which it shares with MAY NOT, when the writer makes a comment on his way of organizing the text: (26) may not

cannot

They told mee a pleasant story about a pair of Justices in those parts, w^ I may not omit the relation of (1705-06)

heer being now an opertunity I cannot omitt the presenting my loue and servis, with thes lines (1649)

I may not omit this opportunity of saluting your selfe and Mrs. Winthrope (1658) I may not here omite how — (c. 1630)

— yet I shall add a few lines, as myne owne letter to your selfe, to whom I am so particularly obliged, that I cannot omit to present my respectfull salutacions to yourselfe and Mrs Winthrop (1659)

— to stay and take a supper with him whose wonderful civility I may not omitt. (1705-06)

Nor can I omit to intreate your effectuall prosecution of what your purpose toward me — (1654)

The deontic source, though not directly identifiable, is nevertheless implied as the paraphrases Ί must/should not do so' or 'it would be wrong of me to do so' seem to fit the context. In the six cases placed under the heading 'neutral possibility' MAY appears with the collocation not only ... but also, which — though included among the negative sentences because of the element not — is, in fact, affirmative by its semantic deep structure:21

164

Merja

Kytö

>


LO LT) • tO ·>» «N

0)

Lf) —~ • Ο in co lf) CO

iJo.

3 CO

ο

S

IT)

Lf)

ω .ω

OD Γ-Ο CO Ο CD •-ί to

(Ξ •3 Β äa.
> to ΖΖ.

ο co σ> σ> to C.

>
5

co — • oo co OS ro

in ία»

ir> o ^ Ξ (Ο •—

g5 LT) — • LT) Lf) in CO

ο Ε ι ο c

—. CD «— -4· in _ CT•) U3~ 00 vjm CD CO

in ο CM —

} ο Ε ίΟ

u.

. oo 03 CS CO

00 CO • in

s

m ω ζ:

d S

=:
• CO (M in —

t/5 m ω — ζ


ο οο

ο C--

ο CD

ο ιη

ο -
m Ηι-Ο— c

CM oo CN| _ . 00

f- Cvl r- _ Ν Β 9 Μ Ε χ* 1 8. ο.

co ~ • ΙΟ CO 00 —

ο Ε > ο C

. oo r(Ο —

>0Q CO . Ο ο to H.

ο Ε ιΟ IL

CO σ) σ> co —

>