For Trudy
Comparison and Universal Grammar LEON STASSEN
U, F. Pr,
CCPGL
Basil Blackwell
)() n(i...!}o - eo"' s. ~e.' ~A"" co ~
c.z
~
.z,qo, 4Xl
t&.or.~i
©Leon Stassen 1985 First pubhshcd 1985 Basd Blackwell Ltd 108 Cowley Road, Oxford OX4 ljF, UK Basil BlackweJIInc 432 Park Avenue South, Suite 1505, New York, NY 10016, USA AI! nghts reserved. Except for the quotation of short passages for the purpNes of cntKISm and rev1ew, no part of th1s pubhcatJon may be reproduced, stored m a retrieval system, or transmitted, tn any form or by any means, electromc,
mechamcal, photocopying, recordmg or otherwise, without the pnor permiSSIOn of the publisher. Except m the USA, thts book IS sold subJect ro the condmon that It shall not, by way of trade or otherwise, be lent, re-sold, h1red out, or otherwise cm:ulated Without the pubhsher's prior consent many form of bmding or cover other than that tn whtch It IS published and Without a s1milar conditiOn mdudmg this condition being Imposed on the subsequent purcl\aser. Brtltsh Library Catalogumg m Pubilcatmn Data Stassen, Leon
Companson and umversal grammar: an essay in umversa! grammar, J. Grammar, Comparative and general I. Tn:le 415 P201 ISBN 0-631-14058-1
Ltbrary of Congress Catalogmg m Publtcahon Data Stassen, Leon. Companson and umversal grammar.
Bibliography: p. Includes index. 1. Companson (Grammar) 2. Grammar, Comparauve and general. I. T1de. P255.S7.1 1985 415 84-28247 ISBN CHi31-14058-1 (hb. bdg.)
Typeset by Freeman Graph1c, Tonbndge, Kent Prmted m Great Bntam by T.J. Press Ltd, Pad&tow, Cornwall
BIBLIO.II)(:A
< ;-,-;_ ',\t
f"~"-'7<<~J I;1"5>&>,, 1.?..?-s:-gf-1
I
Sm'I'OR DE Cl1tNCI;\S HUMANAS, LET~tAS
8F.TOR
L'E
8 A, TES EDUCACA.G----
OntelltS
Acknowledgements List of Abbreviations Part One: A Cross-linguistic Typology of Comparatives 1 1.1
Introduction General background
1.2 Stages in a umversalist research project 1.3 Issues in universahst methodology 1.4 Outline of the following chapters
2
The Typology of Comparative Constructions
2.1 The definition of the concept ·comparative consuuctton' 2.2 Parameters in the typology of comparatives 2.3 Five major types of comparative constructions 2.4 Particle Comparatives
2.5
Mtxed cases
3 In Search of a Determinant Factor 3.1 Word order as a poss1ble determinant 3.2 Spatial relations as a possible derermm3nt 3.3 Temporal chaining as the determmam of comparative types 3.4 Preliminary data
4 Types of Syntactic Chaining 4.1 Introduction 4.2 The definition of the notion ·consec:uttve chatn' 4.3 The formal expresston of consecutive chaining 4.4 Aspects of consecunve deranking 4.5 The formal expression of stmultaneous chaining 4.6 Summary: syntactic types in temporal chaimng 4.7 Language types in temporal chatnmg
Contents
vi 5
Comparative Type and Chaining Type
5.1 Introduction 5.2 The set of chaining-based universals 5.3 Further remarks on the umversals
Part Two: Testing the Set of Chaining-based Universals 6 The Separative Comparative 6.1 6.2 6.3 6.4 6.5 6.6
Introduction Direct corroboration from SOY-languages Indirect corroboratiOn from SOY-languages Corroboration from non-SOY-languages Counterexamples Conclusion
7 The Allative and the Locative Comparative 7.1 Introduction 7.2 The Allauve Comparative 7.3 The Locauve Comparative 7.4 Indeterminate cases 7.5 ConclusiOn
8 The Exceed Comparative 8.1 Introduction 8.2 The Exceed-1 Comparative: serialization 8.3 Other types of the Exceed Comparative 8.4 Factors m the surface variation of the Exceed ComparatiVe 8.5 Conclusion
9 Derived-case Comparatives 9.1 Introduction 9.2 The Conjoined Comparative 9.3 Particle Comparatives 9.4 The comparative construction in Dutch and English 9.5 Conclusion
10 An Examination of Secondary Chotces 10.1 10.2 10.3 10.4 10.5 10.6
Introducnon Secondary denved-case comparat::tves Secondary Exceed Comparatives Secondary Locative and Allative Comparatives Secondary Separative Comparatives Conclusion
Contents Part Three: Towards an Explanation of Comparative-type Choice
11 Theoretical Background Assumptions ILl IL2 11.3 IL4 llS •
12 12.1 12.2 12.3
Introduction
Three levels of linguistic structure Further remarks on cognitive structure Structural levels for comparative constructions
Conclusion
Cogmtive Strategies in Comparative Formation Introduction
Bas1c features of cognittve strategies Three cognttive strategtes
12.4 Conclusion
13 13.1 13.2 13.3 13.4
Grammatical Procedures in Comparative Fonnation Introduction Functional aspects of grammatical procedures Derankmg and identity deletion Opttons in identity delet1on 13.5 Conclusion
14 14.1 14.2 14.3 14.4
254 257 259 261 266 267 267 171 276 277 277 278 182 289
Optimal and Non-optimal Language Types Introduction
290
The Principle of Procedural Depcr-,dency The Pnnaple of Optimal Harmony Opttm.al and non.-opnmal procedure types
291 29S
15 An Explanatory Model of Comparative-type Choice 15.1 Introduction 15.2 The prediction of comparative· type occuaaence 15.3 The predlction of language distributions 15.4 The predict10n of double options 15.5 An evaluation of the new model Notes Appendix A: Alphabetical Listing of the Sample Appendix B: Genetic and Areal Stratification of the Sample Bibliography Index of Topics Index of Languages
298 304 305 319 32! 334
336 347
352 356 367 371
Acknowledgements
Th1s book is a rev1sed version of my doctoral dissertatiOn, which was presented at the University of Nijrnegen in November 1984. Part of the research contained in it was conducted during a stay at the Netherlands Institute of Advanced Study in the Humanities and Social Sciences (NlAS) m Wassenaar, The Netherlands. My dissertation was supervised by Pieter Seuren, who played an active part in the development of my ideas, and who possessed the priceless quality of insisting on those changes which he knew would be for the better. I am also mdebted to Melissa Bowerman, Joan Bybee, Herbert Clark, Bernard Comrie, John Hawkins, Remmert Kraak and Lachlan McKenzie, for their valuable comments on the manuscript at various stages of its complenon. This does not imply, of course, that any of these kind people can be held responsible for the facts or the fiction put forward m this study. I hope that this book, whatever its further merits may be, will be accepted as a tribute to those linguists who, far from the turmoil of theoretical debate, devote their lives to a minute description of the facts in what is called 'exotic' languages. Writing a book like th1s simply would not have been possible if these specialists had not been willing to share their knowledge with me. In the course of my investigation, I have also become deeply impressed by the profundity of the insights of older, often long-forgotten, descriptivist grammanans. Hanoteau (1896), Hoffmann (1903), Taylor (1921), Lafitte (1944) or Lukas (1953) are not exactly household names in current linguistics; but they have become my friends, and their work has an enduring value, to which this essay may bear testimony.
Abbreviations
In the glosses of the example sentences the following abbreviations have been used:
ABL ABSTR ACC ADESS ADHORT ADV AG AOR ART ASS CAliS CLASS
ablative case abstract form ' accusative case adessive case adhortative mood adverbial marker • agenttve case aorist tense amcle assertive marker causative marker
classifier
COMPLET completive aspect CONCESS
COND CON) CONT CONY COP OAT OEM DEP
DET DIS) DUAL OUR ELAT
concessive mood conditional mood conjunctive mood • • contmuanve aspect converb
copula dative case demonstrative dependent mood determiner disjunctive mood
dual durative aspect dative case
EMPH
ERG FEM FliT
GEN GER GOAL HAB !NAN IMP !NO INDEF !NESS INF LOC MASC MED
emphasis marker • ergative case
feminine gender future tense genitive case gerundial marker goal case habitual aspect manimatc marker Imperative mood
indicative mood mdefinite marker • messtve case infinitive
locattve case masculine gender mediative case
MOMENT momenranous aspect NARR narrative mood
NEG NOM NONFUT NOUN PASS PAST PCP PERF
negative marker •
nommatlve case non-future tense nominalization marker passtve votce past tense parttctpial marker perfect marker
Abbreviations
X
!PL., 2PL, 3PL 1st, 2nd, 3rd person
plural POSS PRES PROG PRT Q REFL REL REM 1SG,2SG,3SG
Possessive Present tense Progressive aspect Particle (unspecified function) question marker reflexive relative marker remote past tense
1st, 2nd, 3 rd person singular
SER. MARK SUB) SUB)CT SUBORD SUP TEMP TOP TNS TRANS UNSPEC
serial marker subject marker subjunctive marker subordination marker supplementary element temporal mood topic marker tense marker transitive marker marker of non-
specificity VN
verbal noun
Part One A Cross-linguistic Typology of Comparatives
1 Introduction
1.1
General background
The present essay must be placed Within the framework of Typological Universal Grammar) a trend in linguistic investigation which is relatively young, but wh1ch constantly increased in importance throughout the sc. venties. Taking the pioneering research on word-order by Greenberg (1 963, 1966) as a starting point, universalist authors have begun to tackle a variety of topics, such as coordinate ellipsis (Sanders, 1976; Harries, 1978), relative-clause formation {Keenan and Comrie, 1977, 1979; Downing, 1978; C. Lehmann, 1984), reflexivity (Faltz, 1977), causative formation (Comne, 1975, Shibatani, 1975), the expression of grammancal functions and the phenomenon of ergativity (Keenan, 1976a, !976b; Schachter, 1977; Comrie, 1978a; Plank, 1979; Hopper and Thompson, 1982), verbal a.pect (Comrie, 1976) and word-order variation (W. Lehmann, 1973; Vennemann, 1974; Steele, 1978; Hawkms, 1979, 1980, 1984); in all these cases, new dt!.coveries and dluminanng ms1ghts into the nature of human language have been brought to hght. The history of this new umversalist trend has been documented in Ferguson (1978). Basic principles of the approach, and discuso;;Ions of the result~ m some of the betterRknown areas of universalist research, can be found m the textbooks by Comne (1981) and Mallinson and Blake (1981). As I see it, the goals of TypologiCal Universal Grammar do not d1ffer essenuaHy from those of other forms of lmguisuc mquiry. Universal Grammar, too, tnes to c:ontnbute to a solution of the problem of how to define the nonon 'human language' In terms of a set of restrictive principles; that is, like any other approach m theoretical ltngUistics, Umversal Grammar IS in search of the essennal features of the rule system (or rule systems) known as 'natural human language'. The differences between Universal Grammar and other schools w1thm the field of
Introduction
3
thcorcnc.lllinguiMks arc, in my opmion, mninly a mattrr of method and perspective. I think it is safe to ~ay that most of the recent research in theoretical linguistics {e.g., the research conducted withtn the framework of the Extended Standard Theory; see Chomsky, 1981) has tried to arrive at the underlying basic principles of human language by means of an indepth investigation of a very small set of languages; usually, English is the sole language which is taken into consideration. While universalist authors do not deny the validity of this type of 'narrow' approach, they nevertheless feel that a broadening of the scope of linguistiC investigation is in order; therefore, Universal Grammar bases its inquines on data from an extensive sample of (preferably unrelated) languages. It is expected
that, by a comparison of the structural properties of a large variety of languages, new generalizations as to the nature of human language may come to be formulated. These generalizations may then be used as a supplement (or, as the case may be, as an evaluation measure) of the regularities which have been discovered in the study of single instances of natural language. lr will be obvious that the broad, survey-type perspective adopted by Typological Universal Grammar calls for a specific type of methodology, which differs from established lingwstic practice in a significant number of respects. Therefore, the remaining part of this chapter will be devoted to an exposition of the way in which a universalist linguistic investigation is conducted, and a discussion of a number of methodological problems which may be raised in connection with this type of research. Through· our this chapter, it should be kept in mind that universalist methodology is still in its infancy, and that therefore no hard-and-fast rules of proper conduct can be prescribed. However, notwithstanding this rather early stage of development, there are a number of issues which must be clarified before any universalist research proJeCt can be undertaken; and the least that can be asked of any universalist grammarian is that he state explicitly what solutions he has adopted towards these preliminary methodologicaJ questions.
1.2
Stages in a universalist research project
From a methodological pomt of view, the conduct of a umversalist research project can be split up in a number of successive stages. As a first step, one must establish a language sample, which forms the empincal basts of the research project at Issue. The choice of an adequate language sample for a given descriptive purpose is not without its problems; I will say more about this point in section 1.3.1. For the moment, however, I
4
A Cross-linguistic Typology of Comparatives
will assume that we can succeed in setting up a language sample which meets at least some general requirements of representativity. Once a more or less adequate language sample has been assembled, one arrives at the stage of typology. At th1s stage, the languages in the sample are investigated for one or several structural features, which form the parameter of the typology, and which must have been defined beforehand in a language-independent fashion (see section 1.3.2). When this basic feature has been attested and documented in all of the languages in the sample, a number of different situations may arise. On the one hand, it may turn out that none of the languages under investigation has the features for which the survey was undertaken. In that case (given that the sample which is used has some degree of representativity), one may formulate one's findings in a statement of the following general form: No human language exhibits feature/property X. Statements of this form are known in the literature as absolute negative
universals of language. The oppos1te Situation may also be encountered. That 1s, it is also possible that all languages m the sample exhibit the feature upon which the typological survey was based. In such a case, the results of the investigation may be summarized in a statement of the following general form, a so~called absolute positive universal of language: All human languages exhibit feature/property X.
It will be obvious that absolute universals, whether they be positive or negative, tell us something about the restrictions on the notiOn 'possible human language' m a very straightforward way; they formulate conditions which any rule system must meet if it is to be called a natural language, and as such they can be viewed as the ultimate and optimal research result for any form of linguistic inqmry. Recently advanced instances of absolute universals include, among others, the island constramts established by Ross (1967), the Subjacency Condition proposed by Chomsky (1973), and the Noun Phrase Accessibility Hierarchy put forward by Keenan and Comrie (1977). All of these are abstract princtples whKh are meant to constrain the structural possibilities of natural language systems m a non-trivtal way. Absolute universals, however, are not the only kind of results to be produced by universalist inquiries, nor are they necessarily the most uucresting ones. When searching for absolute umversals, one deliberately abstracts from the highly characteristic and stgnificant phenomenon of variation among languages, and it is, more ohen than not, this variation
Introduction
5
which is particularly revealing as to the restrictions that are imposed on natural language systems. It is for thrs reason that, at least up to now. some of the more excitmg results of universalist research have come from cases m which a 'real' typology could be established, that IS, cases m which the parameter X had been defined m such a way that the following descriptive result could be obtained: Feature/property X is exhibited by natural languages m n different ways.
In cases where such a situation holds, the linguistic manifestations of the parameter X across languages can be classified into a number of different types; these types represent the possible optzons wh1ch languages may select in the formal expression (or encoding) of the parameter X. Related to the classification of (construction) types, the languages m the sample can also be classified into a number of categories, on the basis of difference and similanty m the ways in which various groups of languages select their options for the encoding of the parameter X. ln short, a cross-linguistJc typology consists of tv.:o related categorizations, viz. the typology of a certain construction type and the classification of the sampled languages in relation to the types attested in the typology of the parameter. Typologies, in the sense defined above, are tnterestlng for a number of lingmstic and non~lingurstic reasons. Their main lingutsttc importance lies in the fact that such typologies can be used as data for a further exploration into the non-rmzdomness of linguistic encodrng. That natural languages show variation in their encoding propernes ts an Irrefutable empirical fact; but it is a basic assumption m all universahst work that languages do not vary in unpredictable ways and that, therefore, typological vanarion can be subJect to explanation. Of course, this pomt of vtew IS a matter of faith; it cannot be refuted by single counterexamples, and It will be abandoned only tf the research which is based on tt does not y1dd sufficient results. In other words, the assumpnon that typological variation among languages is non-random belong-; to the core of the research programme 1 of Universal Grammar, and is therefore immune to direct falsificatiOn. If we accept the basic premise that lin gUts tiC encoding across languages is (at least Ln pnnaple) non-random, we may conclude (following Sanders, 1976: 15) that the major functiOn oftypologtes ts 'to serve as the raw materials for explanation, the most refined and manageable raw materials that are available concerning the nature of the objects they typologize'. Thus, typologies are adequate to the extent that they 'generate significant questtons that are dear, explicit and likely to be
6
A Cross~ling1niltc Tyfmlogy of Com{utratirJt•s
p10Ju~..11vdy .llbwcr.lhlc' (~.mJcr.,~
d)JJ.). Now, g1ven thJt our JJta* sampling results m a typology m wh1ch the languages in the sample are classtfied mto a number of different categories, the explanatory questions that can be a~ked are at least of the following two distinct types. The first explanatory questiOn with regard to a typology concerns the question of the occurrence of attested and non-attested categones. When a typology has been established, it w11l generally be the case that It is not immediately dear why that typology contains just these attested types, instead of other, non-a nested but also imaginable, alternanves. Thus, it ts perfectly JUstifiable to ask the question: 'Why ts the typology as a whole the way it is' (Sanders, 1976: 15). Clearly, a pnncipled answer to this question wiH lead to a further understanding of the restrtctwns which delineate the concept of 'possible human language'.
The second, and related explanatory question concerns the attested distribution of languages over the types in the typology; that is, it is a quesuon about the explanation of type-membership. One might phrase this question in the following form: why should it be that certain languages in the sample are members of category X, and not of category Y? In other words, typological analysis assumes that the grouping of languages in the typology reflects a division into natural classes, and attempts to formulate a bas1s for the explanatiOn of this naturalness.
In summary, then, we may say that Typological Universal Grammar wtll try to discover a set of statements which predict attested and nonattested types, and which can account for the attested distribunon of languages over these typologtcal van ants. Among the strategies which are employed m Umversal Grammar to solve the explanatory problems posed by a multi·categorial typology, a natura) and widely used strategy involves the identification of a determining 'outside' factor, that is 'some additional common distinguishing property or set of properties of all members of a given type' (Sanders, 1976: 15), In practice, this strategy leads to the formation of a second typology, which ts based on a new parameter; this additional typology should be set up in such a way that its categorizatiOns provide a match for the distinctions which were attested in the o:.ginal typology. If >uch a new parameter (or set of parameters) can be identified, it should be possible to formulate so-called tmpltcational untversals of language, whu:h have the followmg general form: If a language belongs to category X m typology A, it belongs to category Yin typology B. In statements of this form, one of the properties mentwned refers to the 'outside factor' or 'determmant', which is used a!. the basis for the
lulrmlItt' trun
7
predoctton of the other typological property; this latter property commonly refers to the category in the origmal typology to which the language in question belongs. In this way, various typologically relevant structural properties of languages can be brought together in a cluster of implicattonal relations.
It should be pointed out immediately that implicational unj,crsals cannot, in themselves, count as an explanation of the attested facts in the
first typology. As they stand, universals of this kind merely state a correlatron between two different rypologi
certainly a valuable contribution tn the progress of linguistic theory, and one of the most nrgent tasks of Universal Grammar is to state such clusters of properties as precisely as possible, and for as many properties as is feasible. However, it will also be clear that, if such implicational statements are to transcend the level of pure description, some further requirements should be imposed on them, in order to ensure their status as an explanatory framework. The concept of explanation is far from clear in linguistics in general, and even less so in a relatively young field such as Universal Grammar. As I see it, linguists commonly employ a more or less intuitive notion of explanation; they would say that a ccrtain analysis explams a body of facts if that analysis leads to a deeper and hitherto unformulated insighr, which esrablishes regularity m a seemingly irregular phenomenon. Thus, explanation presupposes the demonstration of a non~randomness, or regularity, in the data, by means of principles which are, m some intuitive
sense, viewed as the causal factor of that regularity. Now, 1f we apply rh1s intuitive notion of explanation to the version of Universal Grammar
which is adopted in th1s study, we can single out at least three conditions which the second typology m (a set of) 1mplicational umversals must meet if it is to be rated as the exp/anans of the first typology, rhe explanandum. The firs< condition on explanarory typologies wh1ch I would like to advance is the following. In order for a typology A ro count as an explanation of a correlared typology B, ir should be rhe case that the categories in typology A exhaust rhe theoretically posSJble variations in the expression of the linguistic parameter upon which
It
is based. That is,
if A is tn count as an explanarion of B, ir should be the case that the categories of which A consists can, in some exphat way, be shown to cover all the possible categories for this typology. (An example of a typology which is exhaustive in th1s sense would be a typology in which
8
A Cross-lingwsttc Typology of Comparatives
languages are classified on the basis of whether or not they have the possibility of Eqm-NP~Deletion. 2 Given this particular parameter, it will be clear that such a rypology will maximally consist of only two categories, one contammg the languages which do have Eqm-NPDeletion, and another which contains the languages without Eqm-NPDelet!On. Thus, a two-category typology 15 exhaustive of the theoretical possibilities of variation for this particular parameter.) I take it to be a defensible conclusion that, tf a certam typology B is correlated with a typology A whtch is exhaustive in this sense, we can say that typology A explains typology B. In such a situation, the attested occurrence of types in B is no longer a matter of chance; it can now be shown to be nonrandom, by virtue of the fact that these type~ in Bare in correlation with a typology whtch contatns all possible variations of its parameter, and 1s therefore by definition non-random. Apan from this notion of exhaustiveness, there is a second condition which, if met, will mcrease the credtbility of a certain typology as an explanans for another typology. This condition has to do not so much with the explanans-typology itself, but rather with the kind of correlation which exists between the two typologies at issue. I think it is justifiable to say that a rypology A will stand a better chance of being accepted as the explanation of a correlated rypology B if the match between the categones m the two typologies is optzmal. That is, it should nor be the case that only some categories in A can be correlated to categories m B, while other categories in A do not have their match tn B; conversely, one may require that all, and not just some, categories in B have thetr counterpart in ~orne category (or categories) in A. If such an optimal match between the two typologies can be demonstrated, we are in a position to say that the facts in the explanandum-rypology Bare fully and exhaustively predtctable from the facts in A. In other words, m such a situation there is a sense in which we can say that typology B is the way It is because typology A is the way it is, and this formulation corresponds largely to the intuitive notion of explanation outlined above. The two conditiOns on explanatory adequacy discussed so far can both be regarded as formal requzrements on explanans-typologzes; they mvolve properties of typologie~ which are independent of the actual parameters on whtch these typologies Jre based. In addition, there ts also a conceptual factor which determine!:. the explanatory value of addttwnal typologtes. In order for a typology A to count as the explanatton of a typology B, one will generally reqmre that the parameter of A represent some 'deeper-lymg', 'more elementary' or 'more fundamental' lingUistic property than che parameter upon which typology B is based. Of course, the notion of 'degree of fundamcntallty' which is mvolved here ts very
Introduction hard to operationabze; moreover, any claim as to the 'fundamentality' of one linguistic feature over another is bound to meet with controversy,
since such a claim will inevitably be tied up with a priori ideas about the aims and methods of linguistics in general. Nevertheless, there are at least some areas of linguistic theory where the fundamemahty of certain
concepts over others has been exphcitly advocated. To be specific, 7
gramn1arians of the so-called 'localist school have da1med that various types of constructions m natural language~ (such as possess1ve constructions, existentials, aspectual expressions and types of case marking} t-an
be shown to be derived from the expression of spatia-temporal relatrons. Accordmgly, this latter type of relations may be advanced a. a candtdate for the status of 'fundamentallinguisttc feature'. 3 As woll become cleao m the followmg chapters, my own approach to the explanatton of the omplocatoonal universals which I will propa.e can be said to be •ympathcnc to the localist viewpoint. It must be understood, however, that I do not necessarily adhere to all the opinions and analyses that have been put forward by authors who work within a localist framework. Needless to say, the above three condition~ are not intended to provtde
a full and explicit account of the concept of explanation in TypologiCal Universal Grammar. They are meant as a first approximation, which
should gove us at least some foothold in deciding upon the explanatory value of implicational universals. If one or more of these conditions are met by an additional typology, I think we have some extent of JUStification for the claim that this second typology is more than just a correlate of the first; it can now be viewed as the determinant of that first typology, that is, as a deeper-lying causal principle by wh1ch the non·randomneS& of variation in the original typology can be predicted and explained. Naturally, implicational universal• of the kind discussed above cannot be the la•t word on the subject of human language, even if some degree of explanatory validity can be attached to them. As IS the case in Jny worthwhile form of scientific investigation, the statement of regularities of the kind that are latd down m 1mplicational umversals gives ri-;e to
further problems of explanation. Clearly, on pursuong these problem• one will inevttably reach a point where the mvesttgatiOn must tran~J~o:end the boundaries of lingms[lcs proper; the ultimate explanation of hngutst!c umversals, 1f It can ever be reached at all, will have to be found by a oombmation of efforts from variou~ scientific branches, such as lmg~tisncs,
psychology, neurology, biology and perhaps even physics. Therefore, to require of Umversal Grammar that 1t have a definitive explanation foJ' all the regularities 1t dtscovers would be too much to ask. What can be asked, though, IS that Umversal Grammar make a thorough and exact inventory of these regularities, and that it play an active role tn the
10
A Cross-linguistic Typology of Comparatives
,.rltH t' tt • li11d pnn1.1 plcJ cxpl.1n.1tlon~ lor them. In the followmg chapters, I w1ll apply the methods of Typological Umversal Grammar to the description of one structural feature of human language, viz. the various possibilities which languages possess to express comparison of inequality. I will present a rypology of comparative conmucnons which IS based on a sample of 110 languages, and followmg that l w1ll propose a set of implicational umversals from which the typological varaation in the encoding of comparison can be predicted. However, before presentmg the actual rypological data I would like to pomt out a number of methodological considerations which determine 111tc' dI\\, llliiiM ry
the framework within which the cross-linguistic invesriganon of com·
par.ltive constructions ts gomg to be conducted.
1.3 Issues in universaliss methodology The above sketch of the conduct of a universalist research proJect is of course very much an ideahzat1on. In reality, every stage in the process
w1ll confront the researcher with specific problems, both of a theoretical and a practical nature. Now, as I pointed out ..,rJier, the methodology of universalist lingu!Sttc research has not yet reached a state of such matunry that generally applicable guide-lines for the proper conduct of an mvesrigation can be derived from it. As a matter of fact, I think it can be said that the methodology of recent universalist grammar, in as far as It exists at all, is a by-product of actual descriptive work, rather than a prehmmary framework in which all research has to be fitted; universalist me1hodology is developed 'aloug the way', so to speak. This is not necessanly a bad tlnng. In my opmion, a new development in a science
should not be stilled beforehand by all too rigorous requirements on me1hod. As loug as some general conditions on representativity and repeatabiliry are met, I think it is best to give universalist grammar, at lea" for the ttme being, a generous amount of freedom to develop its own standards. In the end, of course, it is always the significance of the research results whJCh makes a specific branch of science worthwh1le; nobody needs a rype of hngmstics which, for all its methodological impeccabdtty, comes up wtth conclusions that are ultimately boring. All thts, however, must not be taken to imply that universalist
researchers should be totally free to make any methodological decis10n wh1ch happens to suit them, nor that they should be absolved from the obligation to provide ample JUsnfication for such decisions. The very least one may expect Js that a universali~t grammanan be w1lhng to state
exphc1tly the pros and cons of the solutions he has adopted, so that the
Introduction
11
potl'lltl.ll ~trcngth or Wt'tlk!H.'~.'! ol ]11.'! .1ppro.H.:h c.m bt• .1\:0.t'.'!M.·d from the out~et. In the following sections, I Intend to meet thts minimal requirement: I will deal with a number of methodological problems which I have encountered in the course of my project, and I will dtscuss the solutions whtch appeared to me to be the most appropriate.
1.3.1
The construction of the language sample
A first problem which is certain to be raased m connection with any kind of universalist linguistic research concerns the construction of the sample upon which the typology IS to be based. Given the fact that there are presumably more than 4000 languages in the world, It will be evident that, for practical reasons alone, any umversalist proJeCt will have to hmit itself to a selected subset of languages. It is also obvious that one wall have to make efforts to make this selected subset as representative as posstble, and that known sources of bias should be ehmmared from the sample. Unfortunately, however, at the present time the requirements which an adequate language sample must fulfil are still far from clear; there is not even general agreement among lingmsts about the size which typological samples should have. For some types of universalist projects, the choice of the sample is dictated beforehand by the theoretical question which the typology is expected to elucidate. To give but one example, a linguist who is interested in the variation in the distribution of labio-dental consonants across languages would be well advised to exclude languages which do not have such consonants from his sample. In survey-type universalist studies like the present one, however, no such a priori guide-line is given. Such studaes are of an exploratory nature: it is their very purpose to trace the limtts on possibilities of linguistic encoding, so that an a priori delineation of the languages to be considered would be detrimental to their utility. Notwithstanding the considerable amount of uncertainty as to what constitutes a good sample, there are some rules-of-thumb that have become generally accepted in recent Universal Grammar, For one thing, it is commonly taken as a prerequisite of language samples that they be as free from genetic and areal btas as possible; that is, the languages in the sample should be distributed evenly among the language families and linguistic areas in the world. If this requirement is met, and the sample IS constructed on the basis of genetic affiliation, this procedure will lead automatically to certain conditions on the size of the sample. If all the major linguistic families m the world are to be represented in the sample by at least one member, one arnves at samples which have a size that is
12
A Cross-linguistic Typology of Comparatives
not below 80 languages. The examples of samples given in Bell (1978) all vary in size between 80 and 140 languages; samples which exceed this size are generally considered to be unwieldy. It should be remarked here that the criterion of genetic affihatwn in the construction of language samples can only provide a guide-line, and that it should not be taken too absolutely. For one thing, it is nor clear whether this cntcrion provides the only, or even the most important, factor to eliminate sources of bias from the sample. In the present state of universalist research, all kmds of different typological studieS are based on samples wh1ch have genetic affiliation as rhe1r defining term. This practice can be explamed from the fact that, of all possible sources of b1as, genetic biJ.s has the advantage of at least bemg known, so that it 1s a wise move to eltminate at least this factor from the sampling procedure. It may very well be the case, however, that by using genetic affiliation as a criterion for the stratificatiOn of samples other significant factors in thts stratificauon are overlooked; one may very well imagine a universalist study in wh1ch the sample has been stratified according to, say, the distnbution of word-order types instead of genetic affiliation. As Universal Grammar proceeds, it is easy to imagine that the criterion of genenc affiliation will come to be supplemented, or even superseded, by other criteria that provide a basis of sample construction. Apart from th1s matter of principle, there are also practical factors which tend to diminish the importance of genetic affiliation as a criterion of sample stratification. In practice, the major sources of informanon on the various languages m the sample are written grammatical descnptions and texts; only in a few cases can these data be supplemented by the use of native speakers as informants. Now, as every universalist grammarian is bound to discover, grammatical descriptions of various languages (if one is lucky enough to find them at all) differ enormously in their degree of explicitness and sophistication. Therefore, the sample in most surveytype universalist studies is, from sheer necessity, heavily influenced by consideratiOns of conveniencej one must s1mply make do with whatever data are a\'ailable, however scanty and unreliable they may be. It should be added that the specific subject to be investigated may have its own effect upon the construction of the sample. Quite often, one finds that grammars which are excellent m some areas of grammatical structure are insufficient in others; thus, for mstance, many grammatical descriptiOns are very informative as far as morphology is concerned, but fail to state explicitly the finer pomts of ~yntax. I assume that every universalist researcher will hJ.ve encountered the frustratmg Situation that a grammar fails to state some deta1led p1eces of mformation which are vital to the problem under investigation; m such a case, there is no other choice than
Introduction to cancel that language from the sample. In this connection, the 'centr.~hty'
of the information one looks for may also be a factor. Thus, for example, one will find that information on basic word order can be denved trom almost any grammatical description) no matter how general or amareuristic it may be. On the other hand, mformation on a rather ·margmal' construction type such as the comparative construction is often not to be found m even the most minute grammars. One may conclude, then, that the construction of typological samples wtll, at least for the time being, a.lways involve a certam degree of compromi~ and arb1trariness. [n VIeW of this general situatton, I have constructed my sample on the follow1ng guide-Lme&. I have tried to meet the requirements of genetic spreadmg and size by selecting a sample of 110 langnages, which are cho•en from all the important language fam1hes liSted in Voegelin and Voegelin (1977). An alphabetical h.r of the languages in my sample •• presented in Appendix A; also, I have md1cated there the source or sources from which information about these languages has been gamed. A grouping of the languages in the •ample on the basiS of rheir genetic affiliation is given m Appendix B. I am quite prepared to admit that the •ample used m th1s stud} has various weak pomrs. There remam~ some degree of generic b1as in my sample in that it cont•ms 12langnages of the Indo-European family. ThiS may be JUdged tu be a case of over-representatiOn; I have tried to minimtze this instdnce of bias by choo~ing Indo-European languages that from all the •ub-brancho of this family. Furthermore, one could the Amerind languages are somewhat under-represented; for the 150 gJ:"oups of Amermd languages, a total of 26 languages IS a meagre cho1ce, even given our restriction to a sample of th1s size, I can only agree to thts objection, and add that tt 1s a famiJiar dtfficulty wh1ch, at least for now, IS unsolvable. Despite the large efforts, made m the thirties and afterwards, to document Amencan Indian languages, one has to observe that the structural information contained m the grammars of these languages I!! very often mcomplete, or else is made irretnevable by the ldJOsyncl'atiC and tedious predilection for ragmem1c nota nons. Especially in the ca&e of Amermd langnages one often despairs of the fact that two days of deciphering a grammatical text has not resulted in findmg one good and clear example of the comparative construction. All m all, though, while the sample used in this essay IS far from perfcc~ I am of the opinion that tt does not compare too unf.wourably with samples employed in other c.a.se studJcs in Universal Grammar. As long as a scandardlZarion of survey samples is sttll out of reach, the practice followed m previous research must be one's pnmary gu1de-hnc; in this respect, I feel that the sample used in thts study is reasonably adequate. 4
••Y
14
A Cross-Jm~-:uzstzc Typ~»logy of ComflaratiVes
/. 1.2
l11c dc{llutzun of the ty{JulugiCal basts
At the very stan of the actual eJ(ecution of a typological linguistic invunganon, the researcher ts bound to be confronted with a major methodological 1ssue, vtz. the problem of cross·lingutsttc identification. An example may help to 1llusttate the nature of this problem. Suppose that a umversalist linguist wants to set up a typology of the ways in which the concept of possession is expressed in natural languages. This hngu1st wtll then be immediately fared with the problem of how to decide which structural configurations m each of the languages in hi& sample mu~t be considered to constttute the primary expression of possesstvity. In other words, the lmguist wtll need some criterion for the crosshnguisnc tdennficanon of possess1ve structures, so that he will not compare incomparable items.
With regard to this problem, it " clear that a purely formal defirution of the notion 'possessive structure' cannot be completely satisfactory here. If one were to define the criterion for the not1on 'possessive structure m a natural language' ill purely formal term&, one would Identtfy only those possessive constructions wh1ch have a specified formal manifestation, while leaving out all those instances of natural languages 10 which a different formal way of encoding possessivity has been chosen. In short, the use of purely formal criteria for the identifi· cation of comparable construction& across languages makes this identification unst)lvable for all those cases in winch it is not tnvial. The reason why purely formal cnteria do not work as a method of cross-hngmsnc identificanon lies m the fact that such criteria are, by thetr very nature, language-dependent. However, tt is not the atm of Typolog••oal Universal Grammar to single out those languages m which a con•,tructlon type has a spec1fic formal express1on; Typolog~cal Umversal Grammar attempts co present a survey of all the different ways in which natural languages may encode some linguisttcally relevant property. Hence, It must be concluded that the definition of this propeny (that 15, the baSic feature wh1ch is to serve as the parameter of the typology) should be stated 10 terms wh1ch are mdependent of the characteristics of smgle languages or groups of languages. In recent umversalist studies which deal with lllOipbolog~cal or syntactic variation among languages, the common strategy has been to employ semantic (or perhaps better: cognitive) definitions for the parameter of the typology. That IS, the feature upon which the typology is to be based IS defined in terms of tts semantic content or function, rather than in terms of tts alleged structural characteristics. This semantic solutiOn to the problem of cros~-lmgu1sttc identification has been used m a number
IntroductiOn
lS
of prt•.,tigJ0\111 UlliVCI'""lll~t puh!lt...HIOII\, Mu.:h d~ Kccn.u1 .1nd Comrie ( 1977), where the typology of relallvization is based upon a semantic defi.1'l.it10n of the notion 'relative clause'. The semantic strategy is even-
though somewhat less explic1tly - adhered to in the early word-order studies by Greenberg (1963, 1966). It can be observed that Greenberg's universals refer to notions like 'verb', 'subject' and 'dtrect object', terms which cannot (or at least not exhaustively) be defined cross-linguistically by invoking formal characteristics alone. In conformity with this strategy currendy favoured among umversahst
linguists, I have chosen to use a cogninve cnterion for idennfying cases of the comparative construction across the languages m my sample. Stating, for the moment, the defimtion somewhat loosely and informally, I will say that a construction counts as a comparative construction (and will therefore be taken into account in the typology), if that construction has the semantic function of assigning a graded (i.e. non-identical) position on a predicative scale to two (possibly complex) objects.
Thu>, typical instances of the comparative m the languages of my sample will include those expressions which have a semantic function of the type exemplified by English expressions like john is taller than Bill or Harry is more cunning than brave. Fortunately, 11 turns out that a semantic definition of this kind can be used in practice without serious difficulties; there is a considerable amount of unanimity among grammarians of various languages with respect to the question of what counts as a comparative in their respective subjectwlanguages. In th1s conr.ecllon, let me add that a semantic or cognitive definition will be employed in this study not only m the definition of the notion ~comparative construction', but also m referring to several other nonons, such as 'separative construcnon' (chapter 2), 'temporal chaining' (chapter 3) and 'consecutive cham' (chapter 4}. The term 'separative construction', for instance, must always be taken to refer to any type of construction in a natural language which has the semantic function of indicating the source of a movement, regardless of whether this functiOn is expressed formally by mfAI"-"" of a preposition, a postposition, a case affix or some other syntactic or morphological device. It would, of course, be perfectly feas•ble to set up a typology of the various ways in which this 'separative'
meaning is formally encoded in natural languages. However, such a typology, interesting though it may be in its own nght, hes beyond the descriptive scope of the present investigation.
16
A Cross-lmguistic Typology of Comparattves 1.3.3
The construe/ton of a typology
Once the relevant data have been estabh~hed across the sample by means of some language·mdependent cntcrion of identification, one 1s m a posttion to start the actual work of typological investigation, vtz. the construction of a typology. At this stage, the languages in the sample are classified Into a number of categories, on the basis of differences and similariu~ in the way in whtch the parameter of the typology is formally
expressed by them. Accordmgly, the typology reflects the attested variation of hnguistic encoding among languages with respect to the basic hngm~tlc propeny under dtscussion.
Now, it might be thought that, once the set of relevant data has been assembled, the construction of d typology can be a fairly straightforward procedure. One mtght argue that the classific.ttion of languages mto categories, and the classification of constructions into types, should be the result of a mechamcal application of a series of decisions, which are motivated by empJCically attested dtfferences and similarities in the encoding properties of the languages under observation. In reality, however, such a 'bhnd', mechanical procedure seldom yields interestmg results. It must be kept m mind that the construction of a typology IS not a goal m itself; the whole business of constructing typologies derives its rheorettcal s1gmficance from the fact that these typologies function as 'the raw material for principled inquiry and explanation' (Sanders, 1976: 15). Thus, rhe adequacy of a typology depends on the extent to which it 'generates significant questions that are clear, explicit and likely to be productively answerable' (Sanders, ibid.). In other words, the categorization in which a typology results should be theoretically fertile. Given this central function of typologies, it will be clear that a typological categorizatiOn of languages or construction types must necessarily mvolve an evaluation of the theoretical significance and relevance of the criteria on whteh the division rests. In particular, It wt\1 be the case that not every observable formal difference between language<; or construction types 1s of equal Importance in the typology: the researcher will always be forced to assign relattve wetght to the various differences which can be obJectively observed m hts data. In some cases, a blind application of a divh.ion procedure on the basis of observable formal differences can result in a typology which is too detailed for the purpose for which It was constructed in the first place. To giVe an example, suppose that our data show a difference among languages with respect to the fact that some languages use prepositions in their constructions, whereas other languages prefer postposttions. Now, whether this observable formal difference is a significant fact that must be
Introductwn
17
reflected 10 a typologtcal distinction depends entirely on the use one wants to make of the typology for explanatory purposes. If, for mstance, the typology IS supposed to elucidate various regularit1e~ which arc hnked up with word-order variation, the difference between prepositional and postpositional languages might be judged to be a central fact, which may give rise to a divislOn of languages mto tv..·o !.>Cparatc categories. In other cases, however, one m1ght dectde that thts observed dtfference in word order 1s only a concomitant fact which has nothi11g to do with the central explanatory questiOn at issue. In that case, the researcher may choose to 1gnore this d1fference, and hence he will (other things being equal) lump prepo~ItiOnal J.nd po~tpo~itwnal IJ.ngu.:tges together mto one single typological category. While m some cases 'blind' apphcation of formal dtvision procedures can lead to typologies that are too detailed for our explanatory purpo~e, the oppo!.>ite situatiOn may also be encountered. That ts, one may also arrive at a situation in which a typology that is based on formal distinctiOns ts not mformative enough. Again, an example may help to illustrate this case. Suppose that our cro&&-linguisttc data on a certain construction type give nse to a division of constructions mto two groups with formally distinct characteristics: m one group, a certain relevant NP in the construction is m the nominative case, whereas in other constructions this same NP is put into an oblique case, marked by some adposition. Now, it is conceivable that a categorization on the bas1s of th1s observable formal difference does not go far enough for our explanatory purposes; it might be the case that it is of theorerical sigmficance for us to know whether the adpo~ition in the ~econd group of constructions is a locative marker, an ablative marker, a dattve marker, an mstrumental marker, and so on. In such a case, purely formal differences are no longer of any use, since notions like 'locanve' or 'ablative' can be defined cross~ltngUistically only m semantic terms. In other words, it is not only the case that observable formal differences are not always relevant to our division of languages and constru..::non typeb; there are also cases in which such dtffercnce3 are not suffictcnr, .md where they must be supplemented by further div1~ional cntena of a • semantiC nature. Thus, in the end, typologies must be looked upon as data of an explanatory linguistic theory. Since it is a well-known fact of general methodology that, m all branches of sctence, data-gathenng 1s a proM cedure which is at least partially mfluenced by theoreucal consideration~, it will come as no surpnse that the constructton of hngUJstic typologtes will necessarily be guided by JUdgements (or premomttonb) of theoretical relevance and sigmficance.
18
A Cross-lmguzstrc Typ()/ogy of Comparatives Apart from tha~ m.lJHT mt:thodologJCal i!lMIC in the con~truct10n of
typologies, there are a few other pomts which need comment here. One que~tion concerns the admzsszbtlity of diachrontc and etymologtcal data in the construction of a typology. For a number of languages, rt can be established that changes in the1r encodmg properties have occurred m the course of their history; m such cases, one may argue about the 'true' typological status of those languages. One way to settle such a dispute is simply to bamsh all references to diachronic data from the typology; usually, such a strategy is motivated on the grounds that diachronic data are available for only a small subset of natural languages, so that no systematic use of this type of data can be made. Moreover) a decision to exclude all diachromc data is in line with the strict separation of synchronic and diachronic linguisncs, wh1ch was advocated by de Saussure and has been a longstanding article of faith in structuralist linguiStiC theory. In my opmwn, there 1s not much to be gained from an a prion rejection of diachromc data in typologtcal hnguistic research. On the contrary, I clunk that one might argue that such a reJection could even have harmful effects on the overall adequacy of a typology. In recent Universal Grammar, the insight has gained ground that natural languages are, with regard to their structural properties, often compromises betvleen various mterfering strategies, the balance between which may vary from language to language, and from one developmental stage of a language to another. s Diachronic change is one of the results of such a change m balance; another phenomenon which may be explained by taking this point of v1ew 1s the fact that many construction types in natural languages have 'hybrid' features, and can therefore be subject to 'multiple analyses, (Hankamer, 1977). If we accept this perspective, diachronic data are a valuable clue m findmg explanations for typological distinctions; they illustrate areas of linguistic structure where categones may blend into one another, and they can, therefore, be seen as indications of the underlying principles which are assumed to determine the distnbution of attested and non-attested language types. Related with this point is a second problem in the construction of typologies, which has to do with the indeterminacy of categortzation which one may encounter for some languages. I thmk we can safely assume that, in any linguistiC typology, one will come across at least a few instances of languages wluch cannot be categorized into one single class in a fully straightforward fashion. One of the causes of this mdeterminacy may be the fact that the language in questiOn has undergone (or is in the process of undergoing) diachronic change; another possibility IS that the language, in Its synchronic encoding
Introduction
19
pnK:cJurc, turns out to be J 11ttix' of two or more cru:oding po~~ibilitics. In such cases, I thmk the best way to act lS to accept this indeterminacy as an empirical fact, rather than to try forcing the language into one single category. Cases of indeterminate category status should, I feel, be treated as data in the same way as cases where the category status is straight~ forward; and the explanation one has for the typology as a whole should be able to handle not only the dear cases, but should also indicate why indeterminacy is possible. In this respect, I think there is a parallel with the view on idiolect vanation which is put forward by Eliott, Legum and Thompson (1969). These authors claim that Idiolect variation among native speakers (and, as a result of this, divergence of judgements of grammaticality) should be looked upon as a primary concern for linguistiC theory, rather than as a mere nuisance factor. ln therr view, cases of indeterminate grammaticality are empirical facts, which deserve explanation as much as those cases in which such indeterminacy is absent.
1.3.4
The evaluatton of universals
As I stated earlier, a first step in explaining the data which a linguistic typology offers us involves the formulation of implicational universals. In implicational statements, the defining characteristics of a particular typological option are linked causally or functionally to some mdependent characteristic of the languages in question; in this way, a correlation is formulated between the typological properties of a set of languages and some additional property, which serves as the basis of prediction for the observed typological features. As is generally the case in scientific inquiry, no sure-fire rules can be given as to the way m which such linguistic correlations are to be discovered. It is true that one may invoke a number of discovery techniques in th1s process (such as the use of a computer in testing a broad range of correlational possibihties); but, at least for the time being, one is forced to conclude that the procedure for discovering implicational universals is, in the words of Popper (1934), still very much a matter of •art, not science. Therefore, let us assume without further elaboration that the research project has resulted in the establishment of a set of implicational universals, which are intended to predict the typological variation that we have encountered in the sample. Obviously) such universals are, at this stage, nothing more than hypotheses about assumed regularities; they should be validated by·testing them against the facts of each separate language. At this point, we need to answer an rmportant question: what are the criteria of falsification for a universal? Or, to phrase this problem
20
A Cross-lingutstic Typology of Comparatives
in a slightly different way: how many counterexamples are we allowed to tolerate, before we must reject a universal as false? Regrettably, I thmk we have to admit that, for the moment, there
is no
linal and generally applicable answer to this question. Practice in recent Universal Grammar has shown that practically no universal of any significance is completely free of exceptions; by their very nature, lingu6ti< universals formulate tendencies rather than laws. From this general fact, it follows that universals, tf they are to have any theoretical fung~cal Grammar has produced in recent years, this would be too high a price to pay for methodological purity. One can agree with Comrie (1981: 20), who writes: 'To say that the universal has no validity because there are counterexamples to it, and to leave the discussion at that, would be to abrogate one's responsibility as a linguist to deal with significant patterns in language.' Nevertheless, even if we grant it that a universal may have at least some exceptions, we are still faced with the problem of the extent to which such a universal may be in coolhct with the facts. It may be useful to point out here that statistical evaluation measures, such as employed in physics and soc1ai science, seem to be of very limited use in current
linguistic typological research. Given the fact that there is no generally accepted evaluation metric for the representativity of the language samples from which the data 10 typological linguistic smdies are drawn, the computation of exact p·values for linguistic uruversals would be a fruitless, and even pompous, th1ng to do. Obviously, if a universal were to be verified in, say, only 50 per cent of all cases, nobody would reasonably mamtain that this universal shonld be COOSldered to be correct; the pomt is, however, that we lack the tools m typological linguistics to determine exactly what the maximum of counterexamples to a parnculor universal may be 10 relation to the total number of cases which it is supposed to cover. All 10 all, then, we are left With the uncomfortable, but nonetheless very real sttuanon that there lS In fact no straightforward procedure to evaluate the validity of proposed linguistic universals. Whether a umversal will be accepted as a correct generalization depends on a number of considerations which are hard to weigh exactly, and which are interwoven in ways that are as yet largely unclear. Among these
considerauons, we may hst at )east the following:
Introduction
21
(a) The number of counterexamples in relation to the total number of cases covered by the universaL In order for a universal to stand a chance, counterexamples should be iutnitively rated as 'inc1denral' devtations, which do not distort a dear over-all picture. (h) The extent to which a universal is significant, that is, the extent to which a principled explanatton for it IS available. A universal stands a better chance of surviving falsifications if It can be fitted into a general explanatory theory. (c) The extent to whrch a universal has proven to be theorettcally fertile. An example of such a universalist research result is the Noun Phrase Acccssibihry Hierarchy in Relative Clause Formation, formulated by Keenan and Comrie (1977). Since this hierarchy has turned out to be applicable in a number of different areas of grammatical structure, its immunity to falstfication by smg!e counterexamples has undoubtedly increased.
In summary, we may state that it is funle to think that lmgui.tiC universals can ever be pr()ven to be concct. Typological lingmst1c analysis should set itself a more modest goal: the best we can hope for IS to arrive at a certain degree of plausibility fur our uruversals, ro the effect that other researchers in the field are wtllmg to accept them a• a workmg hypothesis in thetr own investigations.
13.5 Model-neutralrty As a last methodological runark. I would like to comment briefly on the overall theoretical persjK!Ctive which wrll be adopted in thiS study. In particular, I want to make it clear from the outset that my general approach shonld be conceived of as being model-neutral. I am of the opmion that cross-Imgmstic work should not affiliate itself outright with any of tbe theuretical fr.uneworks cutrently in favour, such as Chomskyan GB-theory or Relational Grammar. For this reason, I have taken care to state my findings in terms which are largely uncontrovcrsial, and which belong to the common stock of tradmonal grammatical theory. Of some terms it should be understood that they are used here in a very general, almost metaphorical sense, and that no specific theoretical relevance should be attached to them. For example, when I use the term 'deletion', it should not be inferred that I claim that a specific procedure of stringdenvation by means of deletion rules should be part of a grammar (let alone universal grammar). The term 'delenon' Will be used as a matter of convenience, and must 'be taken as merely a recognition of the consistent syntactic and semantic relationships between ... ~entence types, not as
22
A Cross-lingutsttc Typology of Comparatives
..1n ,,.,lnowi.,·JgcmcntthJt one type docs J~,.•,·ivc from the othl'r (or, lube n1ore ex ace, shares an underfyrng structure)' (MaUmM>n and BJake, 1981; 36). 6 Of course, 1t may turn out later that my results are more m line WJth certain theorettcal alternatives than with others, but it 1s not my aim here to bring such deos1ons about. What I want to do is to estabhsh some charactensttcs of the comparative construction wh1ch must be incorporated tnto any theory of grammatical structure. This attirude, whach has been practised m the majonty of recent universalist studtes, ts
a healthy one, I think; It IS the best posSible guard agamst theoretical myopia, and it frees the empmcal work from the dogmatism which bas marred quite a large amount of recent theoretical linguistic discussion. It goes without saymg, of course, that I do not want to advocate the
po>ttion that universaltst grammar should be opposed to the formulation of theories, and that it should confine itself to mere data-gathering. Like all other forms of scientific actiVity, typological linguistic studies are a waste of time if they do not lead to theoretical progress, and the only way to achteve this goal is by means of the testing of theories. I do feel. however, that linguistic theories ha~re of late reached the point of nearsterihty, due to the fact that their empmcal baSis IS kept too narrow. Therefore, I agree wholeheartedly w1th Comrie's (1978b) dictum that, first and foremost, 'hnguisua. 1s about languages,' instead of being mainly about the techmcalines of grammatical models. 1.4 Outline of the following chapters In the following chapters, I will present the results of a typologtcal
mvesngation of the ways in which the concept of comparison of mequaltty is encoded in natural languages. Chapter 2 contains a discus· Sion of the cnteria which have been employed in establishing a typology of comparative constructions, and presents the actual typology of comparatrve opttons which results from the application of these criteria. In chapter 3, three candidates for the function of determinant of the typology of comparative constructions are considered; the conclusion ts that, of these three parameters, the parameter of temporal chatntng is the most promising. Consequendy, chapter 4 deals with the construction of a new typology, m which the various ways of codifying temporal chaining are laid down. In chapter 5, a set of impltcational universals is defined, in wluch the categones in the second typology are explicitly matched with the categories that have been attested in the typology of comparative constructions.
The chapters in part two of this study (i.e., chapter 6 to chapter 10)
IntroductiOn
23
lont.untbt• t•mtnrl(a/ data upon wh11.h tlu.· lll\IL'~t1g.lt1on n·-.t~. In ca~.:h of the~e chapters, one of the umver~~tls wh1ch have been formulated in
chapter 5 as examined for tts empmcal validrty, by re,-,ring It against the actual facts of the relevant languages in the sample. As a general conclusion, it can be said that the set of proposed tmphcational universals can be shown to be confirmed to such a degree that they may rightfully be considered to express valid correlatiOns. Fmally, in part three we take a closer look ac the explanatory value of the proposed set of universals. Smce this set of universals is found to be wannng m this respect, a new model for the explanation of comparativetype choice is d~veloped, in whtch the anginal set of universals has been incorporated. In the final chapter of part three, the empirical and explanatory adequacy of thts new model is put to the test> and found to be superior to the earlier model of comparative-type choice developed in chapter 5.
2 The Typology of Comparative Constructions
In th1s chapter, 1 will apply the methods of Typological Universal Grammar in the construction of a cross~linguisttc typology of comparative constructions. First, r will discuss the definitiOn of the concept 'comparative construction', and deal with some practical limitations and decision~ which I have made in my investigation. Next, I will present an outline of the criteria whtch have been used in the categorization of comparative constructions into types. The last sections of this chapter wtll be devoted to a discussion of each of the individual categories that have been established m the typology~ and a presentation of a few cases in which such a categonzation seems to be indeterminate.
2.1
The definition of the concept 'comparative construction'
In the preceding chapter (see section 1.3.2), the conclusion was reached that the basiS of a cross-linguistic syntactic typology should be defined in language-independenr terms. Accordingly, I proposed a semantic definttion of the notion 'comparative construction', which I repeat here for • convemence:
Definitzon: a construction m a natural language counts as a comparative construction (and will therefore be taken into account in the typology) if that construction has the semantic function of assigning a graded (i.e. non-identlc.tl) position on a predicative scale to two (possibly complex) obrects. Thus, a case of comparison of inequality rmmmally involves three things: a gradable predzcative scale, whtch represents the property on which the comparison is made, and two concepts, one of which represents the standard against which the other is measured and found to he unequal. The conceptual content of a comparative construction is elegantly phrased hy Small (1929: 12-11), who writes:
-.
...........
~
The Typology of Comparatrue Constructions
.!5
the speaker who uses compartson as a means of mdtcatmg the Intensity of a gJVen quality man object castl> about in hts mind for a second obje(t well known to the hearer which has that same quahty, perhaps m a greater or lesser degree. If he JS fortunare enough to htt upon a ~cond ObJeCt that, to the best of h1s JUdgement, ha!. the qu~hty m exactly the same degree as the object he IS dt.!.CU!>Mng, he may md1cate the mtensity of the qualtcy b)· equating the first object wtth the secor·d, thus: john ts as tall u the gate·poSt, or, The dog runs a.!o rapidly J~ the b1rd fl1cs. Instead of a second obJect of companson the speaker may also refer to a seco·1d rondttton of the first object ttself, thus: john 1s JUSt as accurate as he was m the
r1fle match a year ago. Th1~ wa.y of pomnng out the mtensny of a gi\'Cn J.ttr&bute may be termed the compamcm of equality. Should the hypothctic.1l speaker be unable to h1t upon J. ~ccond obJect or condltlon that exactly matche!t the firSt tn the qnabty observed, or !:>hould he desire to contrJ.St the fir!:>t ob1ect with the second, he w1ll call up to the: aucnnon of the hearer another ob1cct havmg the same quahty, but e1ther m a h1gher or a lower degree of mtens1ty. Thuo,: John IS taller than Maryi Dog!:> J.rc fncndbcr than cats; The patient JS now weaker than he was; I he poet wrote more Vlv&dly than the arust pamted. Tins sort of thmg may be~t be referred to a~ the compmtson of mcqualtty.
In what follows, we will be concerned exclus1vely with cases of the comparison of mequahty. Hence, I will usc the term 'comparative construction' or 'comparative' for the more accurate, but also more cumbersome term 'coru.truction wh1ch represents the companson of inequality'. Given th1s semantiC defimoon of the comparative, we should be able to identify the croS>·lmguistic codilicatton of thiS concept, thereby cstabh,h· ing the data base of the typology. However, before I can present an exposition of the attested typological vanants of the comparative construction, I must first pomt out a number of complications wh1ch .ue connected with the definitwn of the notton 'comparative consrrucnon', and mention several practical decisions whtch further restnct the M:OJX of the mqulfy. First, I have found it useful to confine my typology co those cases of comparatiVe constructions m whtch two obJects or mdtviduals {typzcolly expressed in the form of NPs) are being graded agam>t each other. The reason for this decision is entirely pracncal; whereas all grammars of the languages in the sample indicate the way m which NP-comparatlve are formed, they do not, in general, prov1de for sutfic1ently reliable clara on constructions in which other elements (say, adJecnves, verbs or clau~es) are involved. Thus, the prototypical sentence for our mve~ngatton will be a sentence along the lines of those in (1), and not of tho•e m (2):
A Cross-itngtllstrc Typology of Comparatwes
26 i/) 11.
/NI,fl\/1,
b.
f'he tree ts taller tha11 the house Jltkc J'aml'la better than J.ucy
(2}
ENGI./Sll:
,,
The general was more cunmng than brave
'
~I he team
b.
plays better than last year The prestdent ts smarter than you think
..
•'
Constructions hke those m (2) will be taken mto account only when we look at ~ome languages in detaiL In particular, they will provide evidence for our analysts of the comparative construcnons m Engltsh and Dutch (st c chapter 9). G1ven thJs hmJtauon to cases of NP-companson, the followmg termm-
ology wtll be adopted throughout the dtscusston. The lmgmsuc coddicaoon of the predicattve scale in a cornparat1ve construction wtll be termed the comparatzve pred:cate or stmply the predicate; in the maJOrity of examples, the comparative predicate wiH have the form of a predicadvely used adJective. Ot the two NPs m the construction, the NP which mdtcates the object that serves as a yardstick for the comparison (that ts, to use Small's terms, 'the second object') will be referred to as the standard NP. The other NP in the constructwn, whtch refers to the objective of the mental operatiOn of companson, will be called the comparee NP, Thus, in a ~entence hke (3), the NP Mary IS the standard !SP, and the NP John the comparee NP: (3)
ENGLJS/1:
John is taller than Mary The predtcat!ve adJeCtive tall, whtch names the scale on wh1ch the comparison takes place, Is the comparative predKate m this parncular comparanve construcnon. As a second prehmmary pomt, it must be realized from the outset that my typology of comparatives may suggest a d1v1sion among languages whtch looks more defimte and neat than tt actually IS, For one thing, I must concede that a number of languages m my sample have comparatives which do not seem to fit neatly mto one of the established categories. These language~ appear to have a 'mtxed comparative', that IS, a comparative m whtch the fundamental characteristics of two typological variants are combmed; I w11l say more about these cases m sectwn 2.5. Another phenomenon wh1ch weakens the strictness of the typology IS the fact that quite a few languages appear to have more than one alternative to express NP-comparbon. An example of a language in whtch thts is the
case
IS
Latm. In this language, the gradmg of two objects
agam~t
each
The 'typoiot,'Y of Comparatwe Constructions othl:r ~,., typtcally expresst:d by mc;1ns of a constructton like the one in (4)~ whtch con tam~ the comp,n..aivc p,trucle (flhlllt:
(4)
LATIN:
est eloquentror quam Cicero C.-NOM is more-eloquent than C.-NOM 'Cato is more eloquent than Cicero' Cato
However, as is noted m Kiihner-Gerth (1955), Latm has also the posstbility of a comparative like the one in (5). In th1s case, the particle quam 1s no longer present, and the standard NP 1s put mto the ablattve case: (5)
LATIN:
Cato
Ctcero-ne
eloquentwr est C.-NOM C. -ABL more-eloquent is 'Cato is more eloquent than Cicero' Faced wtth this double opnon for comparative-type chotec, I have chosen to categorize the languages m my sample in tw"o d1fferent ways. I take 1t that for each language there is a primary comparative construction, whiCh is somehow more 'natural' or 'unmarked' than its posstble alternatives. In the typical case, this primary comparative can be used more widely than any of its alternative options; 1 thus, for example, we will see in section 9.4.1 that the ablative comparative in Latm is restricted m tts use by certain spectfic conditiOns, whereas the quam-comparattve ts not. For this reason, Latin will be classified pnmarily as a language of Type 6 (a so-called Parttcle Comparative), and the ablative comparative wt!l be rated as a secondary comparative option for this language. Later on, I wtll argue that a language i:. not completely free m the choice of tts ~econdary comparanve(s); this choiCe can be shown to be governed by the same principles which determine the selection of its primary comparative form. To conclude these prehmmary remarks, one final pomt needs to be mentiOned. As can be easily observed, 1t would be possible to split up the languages in the sample into tw"o groups, on the basts of the fact that some languages require an overt marking of the predicate in their comparative constructions, whereas other languages do not. Restnctmg ourselves to predicanve adjectives, we find that a ma)onty of the languages under investigation do not use such an overt marking; m these languages, predicanve adjectives m comparatives retain thetr unmarked, 'positive' form. Some languages, however, mark a prediCative adjective m a comparative construction by means of a special affix (e.g., -er m English, German and Dutch, -ior m Latm, -bb m Hungarian, -ago m
BU:MA!i/d LBTRAS E AHTB', c .f}f;
F:Of>C/' f'' l
28
A Cross-linguistic Typology of Comparatives
Basque) or a special adverb (more in English, plus in French). I have not been able to find a principled way to account for this phenomenon of morphologic•! marking; that is, I have not succreded in finding an explanatory prmctple on the basis of which the presence or absence ol th1s marking can be predicted. 2 Hence, I will assume that the phenomenon of comparative~markmg IS Irrelevant to our typology of comparative constructions, and that it must be explained eventually in terms of (as yet unknown) regular~ ties which are independent of those that detennme the choice of a particular type of comparative construction. Therefore, I wtll not mdicate systematically whether or not a given language requires morphological marking of the comparative predicate; I will, however, refer to the comparative morpheme of mdividuallanguages at various
points in the dtscussiOn. 2.2 Parameters in the typology of In the following ;ecttons, I wtll dtscuss the parameters of my typology ol comparatives, that IS, those features of comparative constructions wh1ch constitute the basts of a categorization mro typeS. As I observed m sectton 1.3.3, the choice of such parameters is not completely theory-independent: objecnvely poShlble parameter> will always have to be weighed agamsr one another for their theorencal relevance and ut1iity. In my case, tlus evaluatiOn of possible starting points for a cross-linguistic typology of comparatives has led to a categorization in which the encoding of the standard NP (and the variation which can be observed in that encoding) is taken to constitute a highly significant factor. 2.2.1
Case assignment of the standard NP
Wnh regard to the encodmg of the standard NP in comparatives, a first dichotomy mvolves a split between derived-case comparatives and fixedcase comparattves. The parameter upon which this first distinction IS
based has to with the procedure by whtch grammatical case is assigned to standard NPs. In one group of constructions, this case assignment appears to be non-unique; that is, the standard NP is not put into one single case in all the instances of the comparative construction. Instead, the standard NP appeares to derive its case assignment from the case mto wh1ch the comparee NP m the construction has been put. Thus, if the comparee NP in the construction happens to function as a sub1ect, and IS therefore put mto the nominanve case, the standard NP in that construenon w11l also have nominative case; 1f the comparee NP IS 10 rhc
The Typology of Comparatrve Constructions
29
accusative case, they standard NP wJII also be marked for accusative, and so on. In short, we can observe here a parallehsm in case assignment to the two NPs m the construction, to the effe~t that the case assignment of the comparee NP appears to determine the case a~stgnment of the standard NP. Comparative constructions in which this ~ituation obtains will be termed instances of derived case. In contrast to this, other comparative con~tructions are more singleM minded, in that they employ one particular case form for the standdrd NP. That is, these constructions put the standard NP into one and the same obltque case form in all instances of companson, no matter what the case form of the comparee NP in the construction may be. In other words, the case assignment to the standard NP IS independent of the case assignment to the comparee NP in th1s type of construction. Comparatives
which exhibit this feature will be called instances of fixed case. The opposition between instances of denved case and fixed case can be illustrated by the two comparative constructions in Latin, which we briefly mentioned above. The Latin quamMcomparative can be shown to be an instance of derived case. Thus, tn sentence (6) the standard NP (that is, the NP following quam) IS marked for nommat1ve case, whereas 111 (7) the standard NP has received accusative marking: (6)
LATIN: • Brutum ego amo non mtnus quam tu love-lSG than you-NOM B.-ACC I-NOM not less 'I love Brutus no less than you (Jove Brutus)
(7)
LATIN: • Brutum ego amo quam te non mtnus B.-ACC I-NOM not less love-!SG than you-ACC '!love Brutus no less than (I love) you'
The variation in case marking of the standard NP m the~e two Latm sentences corresponds with a difference in interpretation. The Hrst sentence tn the pair must be translated as 'I love Brutus no less than you love Brutus', while the other sentence means 'I love Brutus no less than I love you'. Thus, m the first sentence it is the subJect NP ego (marked by nominative case) which must be interpreted as the comparee NP, and the scale on which the comparison is based can be phrased as 'the vanous degrees of intensity With which people love Brutus'. In the second sentence, the direct object NP Brutum (marked for a<:cusative c.tse) functions as the comparee NP; here the ~calc of comparbon m1ght be described as 'the various degrees of intensity With which I love various individuals'. We can see, then~ that in thi.., Lnin comparative thC' l a<.,c
A Cross-lmgutSttc Typology of Comparatives
30
the . . 1.1mL1rd NP mil\\ lw 1ur.1lld to thl'l';\\C.' form wlul.'h ha\ bt.'Cn as~1gncd to the- compJ.rce NP. In additiOn to Its quam-comparative, Latm has also a comparative whtch must be looked upon as an instance of fixed case. In this construction, the standard NP •s mvariably put 1nto the ablative case, regardless of the case form of the comparee NP. As a result, a sentence like (8), which IS the fixed-case counterpart of the sentence (6) and (7), IS ambtguous, due to rhe fact that It call be eith.er the subject NP ego or the obJect NP Brutum whtch may be taken as the comparee NP; the fixed ablanve form ofrhe standard NP does not g1ve us any clue as to the right choice of the comparee NP for this construction. In other words~ sentence (8) may have cirher the readmg whtch we assigned to (6) or the reading of fo1m
!)J
(7): (8)
LATIN:
non mmus te amo B. -ACC I-NOM not los you-ABL love-ISG Brutum ego
Accordmg to Kuhner-Gerth (1955: vol. 2, 466), it IS because of this amb1gmty that the ablanve comparative is avoided in Latin constructions ,.here rhere are rwo poss•ble candtdates for the function of comparee NP.
2.2.2 Subcategorizat10n affixed-case comparatives Within the category of fixed-case comparatives, rhere is a further typological criterton which read1ly presents itself. Given rhe fact rhat all these constructiOns share the feature rhat rhey encode rhe standard NP in a fixed-case fotm, one may try out a further division of rhese constructions, whtch IS based on the particular type of case form employed. As it turns out, the quest1on of the nature of this case assignment to rhe standard NP provides us with a useful typological parameter, in that it gives rise to a number of comparative rypes which can be defined m a relatively straightforward fashion. A first dichotomy m the category of fixed-case comparatives concerns a chstmctlon between constructions in wh1ch the standard NP is encoded as a dtrect obJect, and constructions in which the standard NP is a
The Typology of Comparative Constructions
31
NP c,l!l be governed; thi~ rcquu·cmt·nt become~ pilrttcularly prcs~ing if the ~.:omparativc predtcarc lt~cH t~ nut a tr.Hl'litivt• verb, but, sny, .111 mtransttive predicative adJective. The typical solution to tlus problem of government as the inclusion of a speczal transitzve verb m the comparative construction, wh1ch has the general meaning of 'to surpass', 'to excel', 'to exceed' or 'to be more than', and which may be looked upon as the signal of a case of comparison of inequality in that particular language. 3 Thus, in comparatives of this type, the surface structure typically contains two predicates, one of which represents the scale of companson, and another whach provides the syntactic means to represent the standard NP as a direct object; the comparee NP in the construction ts encoded as the subject of this latter transitive verb. There 1s~ wuhin this type, a cenam amount of variatiOn in the way m which these two predicates are formally expressed in surface structure; we w1ll present a global discusSion of these variants m section 2.3.4, and postpone a more detailed exposition of this variation to chapter 8. For the moment, I will confine myself to a single 1llustrauon of this particular type of comparative, by way of a random example: (9)
DUALA:
Nin ndabo e kolo buka nme this house it big exceed that 'This house is btgger than that'
In opposition to fixed-case comparatives in which the standard NP has the syntactic function of a direct obJect, a second group of fixed-case constructions represents the standard NP as a constituent of an adverbial phrase. Obviously, since adverbial phrases do not 1mpose any structural requirements of government, the addttion of a spectal predicate to the construction is unnecessary here. Hence we find that, in the typical case of this category, the comparative pred1cate is the only predtcative form in the construction, and that it is modlfied syntactical1y by the adverb1al phrase containing the standard NP. Comparative constructions of this type will be called adverbwl comparatwes. While direct-object comparatives constitute a termmal category in our typology (i.e., a category that cannot be split up mto further subclasses), adverbial comparatives do allow for further subcategorization. The parameters of thts subcategorization will be discussed in the next section.
2.2.3
Subcategonzation of adverb tal comparatives
As tt turns out, a large maJOnty of the languages wh1ch possess an adverbial comparative choose to represent theu standard NPs in the form
32
A Cross-linguistic Typology of Comparatives
of an adverbial phrase which has a clearly recognizable locational inter-
pretation. In other words, most languages in this category model the codification of thetr standard NPs on one of the options which they have in the formal expression of the semantic system of spatial relations. Given tills fact, it becomes possible to introduce a further typologtcal div1sion withm the category of adverbial comparatives, based upon the
particular Iocattonal option which a language selects for the representation of its standard NP. In order to be able to appreciate in full th1s
sigmficance of locational parameters for the typology of comparative constructions, we wtll first have to make a few general remarks on the semantics of spattal expressions. The semantic system of spatial relations (that is, the semantic network of the ways in which two obJects are conceived of as being related to one another with respect to their locat1on in ~pace) ha~ been a continuous subject of inquiry for several generat10ns of linguists and psychologists,4 and, despite the con~>iderable progress made in thts area, it must be concluded that the stratification of thts system is still not fully understood. Evidently, the present study cannot pretend to be a contribution to thJs speafic field in semantic theory; I will deal with spatial notions only in as far as they are needed for an understandmg of certain typological decisions which I propose for the classlficanon of adverbial comparatives. A first major semantic distinction w1thm the kinds of spatial relations that two objects may mamtain concerns the notion of contact. Thus, in one set of spattal relations, the two objects involved are conceived of as being so closely together in space that rhey are thought of as forming a spatial unity. The ser of spatial relations which share the semantic feature of implymg that the two objects are in contact will be called locative relations, and the linguistic elements which are used to encode such locative relations wtll be referred to a~ locattve markers. Languages may employ different devtces for the formal expression of locative relations, mcluding case affixes, adpositions and verbal infixes; however, as I pointed our earber, the exact morphological procedure for the marktng of locative (and other spatial) relations will not concern us here. Opposed to the set of spatial relations wh1ch imply contact between objects we find a set of spatial relations in which the two objects are conceived of as being in two dtfferent places; that is, these relations tmply a spatial dtstance between the two obJects involved. Now, 1f two objects are stated to be distant from each other, it 1s often the case that a movement of one of the objects is implied. If such is the case, a further classification can be made: on the one hand, relatiOns which imply that the movement IS intended to extend the distance between the two objects and, on the other hand, relations which imply that the movement ~~
The Typology of Comparattve Constructions
33
mtended to dimimsh that d1stance. If the movement of one of the objects has the effect of creating a larger distance between the two obJects, the second object m the pair can be looked upon as the source or orzgm of tht movement of the other objtct. Spatial concepts which imply that one ot the objects IS moving away from lts source w11l be called separattvt relations, and the formallingutsttc means to express them wtll be termcc
separative markers. The mtrror·irnage of a separative spatial notion IS an al/atwe nonon.ln the case of allative relattons, it IS undcr~tood that the movement of one of the objects has the effect of reducing the dlstance between the two objects, and, accordmgly, the ~econd objeCt can be seen as the goal w which the movement of the first object is directed. The formal medn') wh1ch language~ employ to encode these all.\ttvc relations wtll be referred to as allat1Ve markers. Now, if we look at the various ways In which languages With dl1 adverbtal comparative encode their ~tandard NP~, we can ob~erve .1 reflection of the tripartite spatia) distmction between focatlve, separanve and allative relanons. It turn~ out that a large majonty of the languages at issue selects one of these three f..emantJC options as a model m the codification of 1ts comparative constructiOns. If a language chooses to codtfy its standard NP by means of a separatwe marker, the prototypical cho1ee appear:, to be a marker with a meaning wh1ch IS equivalent to that of Enghsh from. Semanncally speaking, one might say that m these languages comparison seems to be 'metaphorized' as a movement m whtch the object represented by the standard NP JS taken as the pomr of origm. Apparently, a marker with the meaning 'from' IS the most unmarked or natural codtficanon of th1s separatJVt"' comparative. We do find, however, occasiOnal in:;,unce~ of languages m wh~eh a dtfferent separative marker has been cho<>en 10 encode standard NPs. Thus, for example, there are languages m whl< h the marker at ISSue has the meaning of 'up from'; m tht~ case, a rno1rker has been chosen which refers exphc1tly to the vertical dimenston of the space in which the separative movement takes place, as opposed to the use of 'from', which is essentially neutral w1th respect to sp.tcc dimensionality. Also, we find a few cases where the separattve marker on the standard NP must be glossed a~ •beyond'. In these cases, 1t seems that an additional parameter is introduced mto the separative relatiOn, namely, the position which the speaker or the observer occupies relative to the two objects involved in the separative relation. Agau1, this is in contrast with the use of 'from', whtch IS essenttally neutral a~ to the spatial position of the ob~erver. Lastly, we come acro~s one or two examples of separative comparative~ where the marker of the standard NP b translated
A ( :ro~s-ltnguJ~!JL 'l'yfJt,/ogy of ( :ompttrti/W('S .l~ 'bl'htnd' or \dt~t ', ~lKh L.I~C:C. llllght bl' 'l't'll ,1, lll,1.11h:t•~ ol '\C..')l.tf.ltiVt' reianons winch mdude an observer~onentauon, but m wh1ch the aspect of the ~eparative predicate IS perleccve rather than imperfective.
Languages which choose an allattve relanon as the metaphor for the cxpress1on of companson typ1cally select a marker With a meamng wh1ch IS eqmvalent to that of the English ptcposmon to. Thus, the prototypical form of the standard NP m thiS type of adverb1al comparative is that of a goal-phrase. As was the case w1th separanve comparanves, a hm1ted amount of dtversdication may be encountered m the choice of al1at1ve comparative markers. A few languages exphc1tly refer to the vertical d1mcns1on of spacey encodmg the1r standard NPs by a marker meanmg 'up to'. We also find markers of standard Nl's wh1ch can be translated as 'on this Mde of', thereby betraymg observer-onentat10n, or markers with the meamng 'm from or or 'before', which might be JUdged to be a case of observer-onented allauve movement in perfective aspect. lastly, there are languages winch model thm codification of compariSon on the expressiOn of /ocatwe spatial relations. The unmarked case here IS the use of a marker which resembles English at or on m meaning. Occastonally, however, locanve comparative markers are encountered whiCh refer to other possibilities of spatial contact, such as 'on top of', 'bes1de; next t0 1 or 'agamst'.
2.2.4 Problems m the c..atcgorrzation of adverb1al comparatives In connectiOn With the above class1ficanon of adverbtal comparatives mto three semantically defined types, I must point out a couple of t.onceptual and practical difficulties. Wlnle thts dassificanon turns out to he fa1rly straightforward m the maJOrity of the relevant languages, there are also several languages m which there may anse some uncertamty as 10 the correct class1ficat1on of the1r comparative consrrucrion. Tins uncertamty may be caused by various factors wh1ch I will briefly touch upon helow. F1rst, 1t can be observed that the formal codtficanon of the locational ·,ystem as a whole showc; considerable vananon across languages. In partlcular, languages vary m the degree of exphomess to which semanncally dtstmct locatiOnal concepts are mapped onto dtsnnctions m formal (t.e., lex1cal and/or structural) representanons. On one stde of the scale of exphcatness, we find languages hke Ubykh or Fmnish, where differences m locanonal shades of meamng are formally encoded m a very elaborate system of case d1stinct1ons. Ftnmsh, for example, ha!, different case forms to represent such locatlonal concepts as lllanve ('movement Within a closed space): mlmstde the house), elanve ('movement out of an enclosed
The Typology of Comparatcvc CcmstructHms
35
'P·11..L'': out of the lmus,·), llll'\.,lvc ('po.,.ltlon ol .111 obj(.'Ct m a do~nl space': msrdelwtthm the house), .u.k~')IVC ('po"tlliOII ol .m obJeCt on tht'
surface of another ob1ect': on/upon the table), ablative ('removal of an obfect from the surface of another object': from/off the table), allanve ('movement towards an object': to the house), and several more. Howc\'er, opposed to languages with this kmd of very fine-gramed system of locat10nal representations, we also find languages m whtch most, tf not all, spatial concepts are represented by one smgle formal mode of
expression. An example of a language in which this situation obtains is Mapuche (see de Augusta, 1903). In this South-Amencan language, the postposition meu functions as a very general locational marker, whtch may be glossed as "m', 'on', 'at',
--~----
36
A Cross-linguistic Typology of Comparatives
m surface structure in the case of the preposition in, which can be used to express both a motion towards an object (lnj viel m de rwier: 'he fell mto the river') and a locat10n of an obJect mside another object {hiJ zwom m de rivier: 'he swam m the nvcr'). A case of a language in which this partiCular type of partial neutralization creates indeterminacy as to the correct classification of the comparative construction is Kanuri. In this language, the standard NP m comparatives is marked by a case suffix which has both an allanvc and a locative Interpretation (see Lukas, 1953). Parallel to cases m which general neutralization of locational concepts has taken place, these cases of parttal neutrahzation will be dtscussed individually, and for each case a spec1fic solutiOn (or, as thf' case may be) the admiSSIOn that there is no ready solution) will be stated. Apart from the more or less pracncal problem created by cases of neutralization, we should mentton one other source of indeterminacy m the classificatton of adverbial comparatives. This particular difficulty IS more of a conceptual nature: it stems from the observation that, in a limited number of languages, the standard NP m the comparative is part of an adverbial phrase which does not have a dearly recogmzable locJ.tional interpretation. Thus, we find standard NPs marked by elements which are to be glos·sed as an mstrumental marker ('by', 'with')) an agentrve marker ('by'), a comitative marker ('with', 'in the company of') or a benefact:ve marker ('to', 'for the sake of). One obvious way to deal with such cases would be to mcrease the number of subcategories into wh1ch adverbial comparatives can be divided; in addition to the three locationally defined subclasses mentioned above, one would then allow for a prohferation of other subclasses, which are modelled upon the formal expressiOn of non-locational relations. In this study, however, I will adopt a different approach, to the effect that the extstence of a few cases of non-locanonal adverbtal comparatives will not be taken to constitute a sigmficant typological factor. The point of view which will be defended here may be characterized as localistic. It re~ts upon the hypothesis that 'spatial expressions are more bas1c, grammatiCally and semantically, than VJ.riou\ kinds of non-spattal expressions ... Spatial expressions are lingUistically more bas1c) accordmg to the locahsts, in that they serve as structural templates) as it were, for other expressions; and the reason why this should be so) it is plausibly suggested by psychologists, is that spatial orgamzation is of central importance in human cogmtion' (Lyons, 1977: 718). Thus, the localist posltlon, whiCh has been defended by lmgu1sts and psychologists alike, ts characterized by the claim that non-~patial expresstons must be seen as ultimately denved from spatta1 exprc~1:!10ns, by means of processes of meanmg-expanston and abMracnon, An obvtous point in favour of thi<>
The Typology of Comparattve ConstructiOns
37
position is •the mcontroveruble fact that temporal expre&slons, in many related languages} are patently derived from local expresstons' (Lyons, ibid.). But apart from temporal expres&ions, there are also le~& obv1ous grammatical categones which may be candidate~ for a locahstlc analysts, such as tense, and various notions connected with the grammaCl\..J.! category of aspect (see Comrie, 1976). What makes the locahstic hypothest~ particularly relevant to the ~ubject of the present C!.~ay ~~ the claim that, on the one hand, agemive and mstrumental phrases can be locahsticaliy conceived of as source phrase&, and that, on the other hand, indtrect objects and benefactive phrases can be seen as the locJh!.tlc meaning-expansion of goal·phrases {see Anderson, 1971 i Anderson and Dubois-Charher, 1975). If we accept th1s locahsnc analysts, we are tn a position to categonze comparatives m which the standard NP l!:t encoded in an agentive or instrumental phrase as spectal mstances of the separative comparattvc. Conversely, comparattve& 111 whtch the ~tandard NP ha., thE fonn of an indirect object or a constitutcnt of a benefactiVe phrase can be classified as special cases of the allative comparative. Wtth respect to adverbial comparatives whiCh have a comttative Interpretation, it ~eem'> semantically plausible to classify them a!:t a (more or 1~::.) non-spaual meaning-expansion of locative comparative~. Ltke locattve phrases, comitattve phrases too tmply a dose contact between obJects or mdiv1duah. Furthermore, we may point out that comnattve phrase!. are common m the formal expression of the concept of posses!.iv1ty across languages. As has been argued in a large number of publicattons,S possessive construe· tions are structurally cognate to locative construction& m many unrelated languages throughout the world.
2.2.5 Subcategorzzation of dertved~case constructions With the distinction of direct-object comparatives and the three subtypes of adverbial comparatives, we have established the termmal categones of the class of fixed-case comparative conMructtons. To conclude our discussion of the parameters whtch we employ in our typology of comparatives, we must now pay attention to the other maJor cl.l!.!., vtJ.:. those comparattves m whtch the case of the Mandard NP can be Jenvcd or determmed from the case of the comparee NP. One mtght say that, Jn such constructions, the case of the standard NP 1s 'parasitiC' on the ca·,e ass1gnment which the comparee NP has received. Withm the class of denved~case comparatives, we can smgle out one highly typtcal subcategory, whtch can be tdcnnfied very east!y acro;s languages. For this category, a definmg characteristic I& that the comparat1 ,re constructwn consists of two structurally mdependent clauses; one of
A Cross-ling111stic Typology of Comparatrves
3h
1hc..,r ~ Llll'>l''! l o!!t.llll"
tln·
l'Olllp.m.T
NP, wlu.:n·,,.., tlw othrr
dauM·
<.:ont.tmo., the '>t.lntl.lrd NP. Furthnmorc. the two ci.HJ\(':-, llllJUl'\tlon ~how a .tmctural parallcltsm, to the effect that the grammaw.:.tl functton of the comparee NP Js one of the clau~:oe& IS reduplicated by the grammancal funcnon of the standard NP tn the other clause. If, for example, the comparee NP funcnom as the grammatical subject m 1ts clause, the st.lndard NP will also have :-,ubjcct c;;tatus m Its clause. Smce rhe comparanve con<>trucnon at issue com.ists of two mdependent cl.~uo;es, tt foHows that the construction wJ!l also have to contam two mdependent prediCates. In other word~, a further feature of this type of comparative 1s that the comparatwe predtcate is expressed twice. There arc vanotls different ways to formalize thts double reference to the scale ol compan~on; we will say more ahoutthts pomt in sectiOn 2.3.5. As a last pomt, we should note that the semantiC relation between the two mdependent clauses m this comparative construction is usually to be dt·scnhcd as adversatwe coordmatron; tts hteral mterpretation ts somethmg along the hnes of 'A is p, bur B is q.' Thus, one might say that, in tl- I!> type of comparative, there is no dtrect measurmg of the two objects agamst each other, but the non-equal gradanon of the obJects involved can be mferred from the fact that they are contrasted m an adversative coordmat10n. To quote JUSt one example of this type of conJotned comparattve, constder the following sentence from Samoan: (10)
SAMOAN:
Va loa /ener va'a , ua puupuu lena ts long thts boat IS short that
The Typology of ComparatiVe ConstructiOns
\l.lndard NP in th<.''>l' romp.tr.lltH'\
39
.Kromp.lnH·d by .m dement whtch Wl' HhiY ...:.tiii/Je <.OIIIJ•!Ir.llfi'C f'•ll/1(1('. In dw typtl.d l.l'.. t, thi~ p.~rtidr t.:Jnnot be idcnnficJ a~ J C.I~C m.1rker, ~nKc It\ prc'icncc ~~ mdependent of I'>
the particular case form whtch the standard NP happens to have. A der1ved-case comparative construction m which such a comparative partacle is present will be called an mstance of the parttcle comparative. Examples of particle comparative!3 whtch have already been introduced m the foregoing text are the Latm quam-comparattve and the Enghsh than-comparative. Some prehmmary remarks on thts type of comparative will be made in section 2.4. Chapter 9 will be devoted to a detailed dtscussion of the constructiOns m whtch a comparanve parncle IS present. 2.3
Five major types of comparative constructions
If we apply the above criteria to the cross-lingmsnc data base of our typology of comparatives, we arnve at a classtficanon mto five clearly identifiable vanants. In this section, I w1ll bnefly introduce each of these variants in turn, and list the languages whtch have a comparanve construction of the type at issue as their primary or secondary option. Furthermore, I have included some comments on the b~ic word order of the languages involved; as will become clear, we can observe some degree of correlation between basic word-order type and chotce of comparative type.
2.3.1
Type 1, The Separattve Comparatwe
The Separanve Comparative is an mstance of fixed-case adverbial comparanve constructions. NP-companson is, in this type, expressed m one smgle surface clause. In this clause the comparee NP can, in pnnciple, have any grammattcal functiOn. In contrast, the standard NP IS mvariably encoded as a consntutent part of an adverbial phrase with a (spatial or non-spatial) separative mterpretation. Examples of this type of comparative construction are the following:
(11)
MUNDARI'
Sadom-ete batt
mananga-i
horse-from elephant big -PRES. 3SG 'The elephant i!> bigger than the horse'
(12)
JAPANESE'
Nthon-go wa dozts-go yori muzukashz Japanese TOP German from difficult 'Japanese ts more dtfficult than German'
-·-~
-
-
--- -··-·-------
--
-
40 (13)
A Cross-ltngutstic Typology of Comparatives CARIB:
Kuliali a -kultalt kopo apoto-me mang canoe your-canoe from btg -one IS 'My canoe is bigger than your canoe' A hst of languages which exhibtt this comparative construction as thet primary option reads as follows: Amhanc
Eskimo
Manchu
Andoke Arabic (Class.) Aranda Aymara
Guarani Hebrew (Biblical) Hind• Japanese Jurak Kashmtri Khalka
Mundari Nama Piro Quechua
Bedauye
Btlm Burmese Burushaskt
Canb Cceur d'Alene
Korean Lamutic
Tajik Tibetan Tupi Turkish Vayu
Laz
A secondary separative comparative can be found in:
Albaman Basque Old English
.Finnish Old French Greek (Class.)
Latin Russian Tamil
From dus listing, we may conclude that the Separative Comparanve ~~ very wtdely spread indeed; almost 30 per cent of the languages in rht: sample choose thts type as their pnmary option. A striking charactensti<. of this category is that the languages wh1ch it contains appear to have:: strong preference for SOV word-order. Of the 32 languages wtth ; primary separative comparativej only four (viz. Arabic (VSO), Cceur d'Alene (VOS), Guarani (SVO) and Bibltcal Hebrew (VSO) are contra dictions to th1s tendency. 2.3.2
Type 2: The Allattve Comparative
Like the Separative Comparative, the Allative Comparative is an instanc of a fixed-case adverbial comparative type. It can rightly be regarded a the m1rrorMimage of the Separative Comparative. Again, we find NP comparison expressed m one single clause, with the comparee N appeanng in any grammatical functiOn. The standard NP, however 1doc not form a part of a source~phrase, but is invanably encoded as t'nnstin1cnt part of ;' (sp;:ltlnl or non~spaual) goal~phrase. Examples c th~; type indud<'.
The Typology of Comparative Constructions (14)
41
MAASAI:
Sapuk ol -kondi to I -kibulekeny JS-big the-deer to the-waterbuck 'The deer is bigger than the waterbuck' (15)
BRETON:
]azo bras-ox wid-on he btg -PRT for -me ·He is btgger than me'
Listing the pnmary languages m this class, we get: jacalrec
Maa.al Nuer
Kanun
Smslawan
Breton
Tarascan
An Allarrve Comparative as a secondary option can be encountered 1n Mandmka, Mangarayi and Tamil. We can note here that, at least as far at, the primary opuons a1·e concerned, the A1lat1ve Comparath.·e appears to be hm1ted almo,[ exclusively to languages which h•ve basic verb-rmttal word-order. Kanutl, an SOY-language, is the only exa:pnon here. Indeed, the relatively small size of the class may well be connected wtth the fact that a considerable number of verb-inittal languages prefer a ConJoined Comparative or a Particle Comparative as their primary option m comparative~type choice; for this, see sect10n 2.3.5 and secnon 2.4.
2.3.3
Type 3: The Locatzve Comparatwe
The third instance of fixed-case adverbial comparatives is the Locative
Comparative. In this construction, the standard NP is mvariably encod~d as a constituent of an adverbial phrase whrch is marked by an element that mdicates spatial or non-spatial contact. Apart from the marking of the standard NP, the Locative Comparative IS forntally >imil•r to the Separative Comparative and the Allative Comparative in all sign1ficanr
respects. That JS, the construction again cons1sts of one single clause, and the comparee NP ts free to fulfil any grammatical function. Examples of this comparative type include the following: (16)
CHUCKCHEE:
Gamga-qla'ul-rk qetvu -ci -um all ·men ·on strong·more·1SG 'I :un ~uongcr rhan all men'
A Cross-lmguist~e Typology of Comparatives
42 (I I)
1.11 INAl><:
Ragas-mo m luwa t1 -hek surely-you more man on-me 'You are more of a man than me'
The followmg 12languages are prtmary members of this class: (SOY) (YSO) (SOY) (SOY)
Chuckchce Cebuano Mandmka Mapuche ~condary
M1wok Naga Navaho Salman
(YW) (SOY) (SOY) (VSO)
Tamaztghr Tamil Tubu Ubykh
(YSO) (SOV) (SOV) (SOV)
Locanve Comparanves are documented for:
(SOV) B"'que (SOY) Dakota Gumbamggor (SOY)
Hungarian latvian Maon
(SOY?) (SYO) (VSO)
Nama
Samoan
(SOV) (VSO)
As for word order, the absence of SVO·languages in this class is a stnkmg fact, even 1f we make allowance for the relatively small size of the category. Thus, ot seems that a predtlection lor etther SO V or VSO order among the languages of thos category may be stated with some confidence. This tendency IS mengthened even further when the secondary members
of this type are taken mto account. 2.3.4
Type 4: The Exceed CompllTattve
The Exceed Comparatove IS the fourth and final varoant of fixed-case comparattve constructions. Its mam charactensnc ts that the standard NP os mvanably constructed as the direct object of a special transitive verb, the meanmg of wh1ch can be glossed as 'ro exceed, or 'to surpass'. Furthermore, the comparee NP always funcnons as the subJect of th1s ~exceed'-verb. hom a semantiC pomt of v1ew, the Exceed Comparative may be con~1dered as cognate to the Separate Comparative, in that both con~rrucnons
imply a movement of an obJect beyond or farther away from another obJect. formally, however, the two types of comparatJves are definitely dtstmct. As to the representatiOn of the comparatiVe pred1cate m th1s type, ~trateg1es may dtHer from language to language. Some language prefer a so-cal)ed 'sen a I verb '·constructiOn, m whtch the comparee NP is con-
.rructed as the sub)ect of a verbal complex whtch contams both the comparative pred1cate and the exceed-verb. We wdl dtscuss tlus variant 111 detail in sectiOn 8.1. For the moment, the fo11owmg examples of this Exceed-1 Comparatwe may suffice:
The Typology of Comparative Construe trans (IS)
YOHIIIIA: 0 tobr JU
43
u
he big exceed htm 'He IS btgger than htm' (19)
VIETNAMESE:
Vang qur
bon
bac
gold valuable exceed stlver ·Gold is worth more than silver' In other languages, such as Hausa, we find that the exceed·verb is the only main predicate m the constructwn. In thts vanant, whtch we wtll call the Exceed-2 Comparat1ve, the comparative predicate is expressed in a subordinate form which funcuons syntacttcally as an adverbial phrase: (20)
HAUSA:
Doki ya-fi rago grrma horse It -exceed goat btgness •A horse is bigger than a goat'
Sull other languages construct the comparative predtcate as the sole mam verb m the construction, while the exceed-verb receives some subordinate form (e.g., the form of a parriaple, or of an mfinltlve). An example of thts Exceed-3 Comparatwe can be found tn Swah1h: (21)
SWAHILI:
Mtr huu nr mrefu ku -shrnda ule tree this IS b1g INF-exceed that 'This tree is taller than that tree' We should emphasize here that, whatever the strategy in representing the comparative predicate and the exceed-verb may be, the main characterl~ttc of all the variants of the Exceed Comparanve ts that the constructiOn com1sts of one single sentence, and that It contains a transltlve exceedverb which takes the standard NP as Its duect object. My sample contains 20 languages which have some variant of the Exceed Comparative as their pnmary opnon. These languages are the followmg:
Banda Bari Cambodian Dagomba Dual a Fulani Gbeya
Hausa Igbo ]a bern Kuund1 Mandarm Marg1 Nguna
Swahili Thai Vietnamese Wolof
Yagan Yoruba
..
44
_
A Cross-lmguisttc Typology of Comparatives
A secondary Exceed Comparative can be found in:
Ayma.-a
Quechua
Maasat
Stka
Sranan Tamazight
As for word order, we can observe that for this set of languages SVO order appears to be mandatory, at least as far as the prtmary members of tlns type are concerned.
23.5 Type 5: The Con,omed Comparatwe The Conjoined Comparative is the most conspicuous manifestation of
derived-case comparative constructions, ln this type, NP-comparison is typically effected by means of the adversative coordmation of two clauses; one of these clauses contains the comparee NP, and the other clause contains the standard NP. Furthermore, there is a structural
parallelism between the two clauses, to the effect that the grammaucal function which the comparee NP fulfils in its clause is matched by the grammatical function of the standard NP in liS clause. As a result, thiS type of comparative mvolves two g.-ammancally mdepeudent clauses, which are connected in such a way that a gradation between the two objects can be inferred. W1thin this category, we can distinguish two subtypes, on the baSis of the particular manner in which the double reference to the scale of comparison 1S effected. These two subtypes are:
-Type SA: conjoined comparatives in which the two clauses contain antonymous predicates; and - Type SB: conjoined comparatives m which the comparative predicates in the two clauses ex:hib1t a positive-negative polarity. An example of the first subtype IS the following sentence from Sika:
(22)
SJKA:
Dzarang ttca gahar , dzarang ret kesik horse that b•g horse thiS small 'That horse IS b1gger than this horse' Subtype 5B can be illustrated by the following example: (23)
HJXKARYANA:
Kaw-ohra naha Waraka, kaw naha Kaywerye tall -not he-is W. tall he-is K. 'Kaywerye is taller than Wnmka'
The Typology of ComparatiVe Constructions
45
A list of the 20 primary languages in this class reads as follows: (SYO) (OSY) (SOY) (SOY) Gumbainggir (SOY) Hixkaryana (OYS) Kobon (SOY)
Ab1pon Cayapo Dakota Ekagi
Mangarayt
Maon Menomini
Miskito Mtxtec Motu Monumbo
(SOY) (YSO) (SYO) (SOY) (YSO) (SOY) (SOY)
Nahuatl Pal a
(SYo: (SVO' (YSO· (SOY, (SYO• (SOY1
Samoan
Shipibo Sika Yavapai
A secondary option for a ConJoined Comparative can be found in:
Banda Kirundi
(SYO) (SYO)
llocano
Mapuche
(YSO) (SOY)
Nuer
Swahd1
A first thing to note about this class is that the
language~
(YSOJ (SYO) which
It
contains appear to exhibit a certam degree of geographtcal groupmg. Th~ primary languages in this class are predommandy Austrahan) Papuan or Polynesian, or else can be situated on the American continent. It thus looks as tf the choice for a Conjoined Comparative is (at least to a certain extent) influenced by the characteristics of certain linguistic areas. As regards basic word order, no definite statement can be made for this class. Basic word order does not seem to be a determining factor in the choice for or against a Conjomed Comparattve; afl maJor word-order patterns are represented in this category and, what is more, they occur in
proportions which do not dtffer greatly from those which one would expect to find in a random 7 selection of languages.
2.4 Particle Comparatives In the case of the five typological categone~ which we discussed 1n the preceding section, classification IS fa1rly straightforward; the languagt·~ which are h~ted under each of the respective he.:tdings are ea~ily recogni;able as instances of the type in question. However, I have already pointed out that, for a number of languages in my sample, das~ification IS a bit more problematic. In this section, I will deal With one set of such problematic cases, the so-called Partzcle Comparatwes. Followmg that. I will note a number of cases where classificatiOn seems to be indetermmate as a result of the ~mixing' of two comparative types.
Under the heading of 'Particle Comparatives' I have lumped together a number of constructions which have in common that the standard NP has derived case. However, these construction\ al ..o lack a numhcr 1}f
41.
A Cross-lmgu1sttc Typology of Comparatwes
fc tlllr<:~ whtch ch,tr.h.:t•.:ntc the m.un class of dcnvcd·ca~c comparatives, VII.. the CtHlJ(Illlt'd (,Oil1J1.ll'.l11Vf..', h1r one thtng, PJrt1de ('(lmp.tr.tllvc~ do n
structurally a& a constituent part of a phrase m the clause which also contams the comparee NP. In this respect, Particle Comparanves restmble (again, as far as thetr 'ynchronic form is concerned) adverbial comparatives rather than conjotned comparanves, where standard NP and comparee NP are constituents of separate and mdependent clauses. A typical characteristiC of all Particle Comparative is the presence of a
spec1fic comparatwe particle, wh1ch accompanies the standard NP. In the typ1cal case, thiS particle cannot be idennfied as some kind of case marker, since the case form of the standard NP in this type of comparative IS derived from the case assigmnent of the comparee NP. 8 Another feature of Particle Comparatives, which we may note in passing, IS that morphological markmg of comparative predicates seems to be exceptionally popular with them. Of the 18 language; with a Particle Compara-
nve, there are 13 languages in which morphological romparativemarkmg of the predicate IS obligatory. While Pamde Comparatives share at least one common feature
(namely, the fact that they are, or have been, instances of derived case), there are also mdicanons that the Particle Comparative IS l10t a homogeneous category. The internal diversificauon of th1s set of comparatives
is brought out by the fact that the comparative particles which are mvolved in the vanous constructions cannot be shown to have the same ongm or categorial status for all of the languages in this class. One gets
the impressmn that the class of Particle Comparatives (m as far as they can rightly be •aid to form a class at all) has assembled its members from a large vanety of sources and that, m languages with a Particle Comparative,. the formanon of the comparatiVe construction has, so to speak,
'gone wrong along the way'. A full d1scussion of the various forms which Particle Comparatives may have Will be presented in chapter 9. Examples of languages with a Particle Comparative are Latin and English, and also the followmg languages: (24)
HUNGARIAN:
Istvan
magasa-bb mmt Peter
L-NOM tall
-PRT
than P.·NOM
'Istvan is taller than Peter'
The Typology of Comparative Constructtons (2\)
47
JA VAm:IF.:
lit111k
dagtttg karo ttetlk
ts-good meat th;ul fi~h 'Meat is better than fish' (26)
MALAGASY:
Lehibe noho ny zana-ny Rabe tall than rhe son -his R. 'Rabe is taller than his son' A pr1mary Particle Comparative of some type can be attested for the following IS languages m the sample: (SYO) Albanian (SOY) Basque (SYO/SOY) Dutch (SYO) English (SYO) French (SYO) Finnish
Gaehc (YSO) GoaJiro (YSO) Greek (CL) (SOY) Hungar1an (SOY?) Ilocano (YSO) Javanese (SYONSO)
Latin Latvian Malagasy Russian Sranan Toba Barak
(SOY) (SYO) (VOS) (SYO) (SYO) (YOS)
Furthermore, a Particle Comparative occurs as a second opuon in Bari (SYO) and Classical Nahuatl (SYO). Looking at the hst of languages with a primary Particle Comparative, I think it IS safe to say that thts comparative type is, to a considerable extent, to be rated as an areal phenomenon. No less than 12 of the 18 languages at issue are members of the European 1 Sprachbund', while four others belong to the Austronesian family. As for word order, this list makes it clear that basic word~order patterns do not seem to constitute a determining factor in the choice for a Particle Comparative. As with conjomed comparatives, all major word orders are represented here, in roughly the proportions which one would expect to encounter m a random selection of languages. All in all, then, we may conclude that there is no correlation between derived-case companson and preference for certam word-order types, while for the vanous subtypes of fixed~case comparison such a preference can be established wtth some certainty.
2.5 Mixed cases As I pointed out earlier, there are certain languages in my sample for whtch the type of ~omparative construction is hard to determine uniquely. These languages appear to have a mrxed comparative, in which the essential features of two different types seem to have been combined.
48
A Cross-linguistic Typology of Comparatives
Now, in section 1.3.3 I adopted the view thatsuch cases of indeterminacy must be conceived of as empirical data, wh1ch require explanation as much as cases m which this indeterminacy is absent. A framework m which the phenomenon of type-mixing in comparatives might find a principled explananon will be presented m chapter 15, For the present, I w1ll restrict myself to a brief presentation of the relevant cases. In the sample we find at least three languages with a comparative
which combines features of rhe Conjoined Comparative and the Exceed Comparative. In all three of these languages, the construction consists of an adversative coordmation of two structurally independent sentences; if we take this feature to be the determmmg factor in the classification of these constructions, we would have to categorize them as instances of the Conjoined Comparanve. Unlike the regular case of the Conjomed Comparative; however, m these constructions the second sentence does
not contain a negation or an antonymous predicate, nor is there a
structural parallelism between the two sentences with respect to the grammatical function of the standard NP and the comparee NP. Instead, the second sentence has an exceed-verb as its pred1cate, which takes the standard NP as its duect object and the comparee NP as its subject. We find this 'Conjoined Exceed Comparative' in Fulani and Achoh: (27)
FULANI:
Samba mawt , o buri Amadu S. is-big , he exceed A. 'Samba IS b1gger than Amadu' (28)
ACHOU:
Gwok mera dit ki kato men dog my big and exceed your 'My dog is bigger than yours'
In the above Fulani example, it is the comparee NP which is mentioned explicitly m the first clause of the comparative. We can, however, also find instances of the oppoSite possibihry, that is, cases of a Conjomed Exceed Comparative in which the first sentence has the standard NP as its subject. Examples of this state of affairs are provided by the (secondary) comparatives in Motu, Tamazight andTemne. Cp.: (29)
MOTU:
Una na namo , ina herea-ia
that is good this exceeds 'Th1s i~ better than that'
The Typology of Comparative Constructions (30)
49
TAMAZIGHT:
Aiis
ennek mularen , oua bin iou(i horse your IS-good that my exceeds
'My horse is betrer than your horse' (31)
TEMNE:
A -seth ane a fino , kere anan a thas the-house this it good but that It exceed 'That house IS better than this house' Qmte a different case of indeterminacy 1s presented by the classification of the comparative construction in GaoJiro. Part of the difficulty here IS that the two sources on this Arawakan language (Celedon, 187S;
Holmer, 1949) seem to contradict one another on an essential pomt. If we follow Holmer (1949: 146), we observe that the GoaJiro comparative obligatonly includes the presence of the Item aulaka, whtch m other contexts must be translated as 'and) or 'too'. Furthermore, the standard NP tn the construction is not only expressed by that NP Itself, but ts al'lo referred to by an adverbial phrase whtch consists of the postposition -Ita 'from' and a proclitic pronominal element whtch agrees wtth the standard NP m person, number and gender. The following sentences may tllustrate this constructiOn: (32)
GOA}IRO:
a.
Au/aka Rupeta mulia'u , hu -Ita Rafaela and R. big-is her-from R.
Faced with this situation, we may venture the hypothesiS that the GoaJlro comparatrve is a Conjoined Comparative in which, untypicaHy, the second clause has the form of a separative expression. Accordingly, we might rate thts construction as a case of mixing between the Conjoined Comparative and the Separative Comparative. In contrast to this representation of the GoaJiro comparattve, Celedon (1878: 20) notes an essentially different state of affairs. For one thing, Celedon does not mention the presence of the item au/aka in this construction. More importantly, this author state~ that the standard NP in Goajiro comparative~ is signalled by the presence of the element nOna; it is plausible to view this item as a phonological or orthographtcal vanant of the element nu-lra 'from himllt' mentioned by Holmer. 9 Now,
10
the C\'ll.:rltral polllt J,') th.ll, ,J(.I.orJurg to CclcJun\ JJt.l. the p.~rtrdc mma ~~ an mv.lnable clement; tt does not show any morphological agreement wtth the standard NP. Thus, Celedon quotes the followmg examples of the GoaJrro comparative: (33)
GOA]/ROc
''·
Kauzu
-sht
Pedro noria juan beatmful-MASC P. than J. 'Pedro rs more handsome than Juan' Kauza -se Marta nona Juana beautJful-FEM M. than ]. 'Mana rs prettier than Juana' Ooyor-shi taya nOna pw fat -MAIC I than you 'I am fatter than you'
If Holmer's analysis were followed here, the particle ndna in these semence~
would have to have the forms nu-na 'from h1m', hu-na 'from her', and pu-rta "from you', respectively. Since this vanation is not attested by Celedon, we must as!!tume that this author takes the Item norza to be a comparattve partiCle, which would make Goa]iro an
instance of a language With a Particle ComparatiVe. In considenng the case of Goajiro, I have dectded to follow Celedon's faces, and hence l have classified thts language under the heading of the Particle ComparatiVe. I thmk that the most plausible way to account for the vanous data on the GoaJ!rO comparative is to as~ume that, in this construction, the cr~tw'hile separative phrase '&om (which has been retamed as such Ill some dialects or by some native speakers) has gradually come to be reanalysed as an invariable adverbial element, in tts most neutral form, vtz. the form 'from 1t'. In other words, my suggestion IS that GoaJiro presents a case m which a Separative Comparative has elltered a penod of transmon mro another type, viz. the Particle Comparative; an mtermcdtate ~rage in this process may be seen in the bet that th1s construction has at least some features wh1ch pomt to a conJomed analysts. In this connection, it may also be useful to point out that a basic separative mterprctauon of the particle 1n a Particle Comparative is by no means uncommon; the comparative parncle asa in T oba Batak IS an adverb with the meaning 'after that', and the same IS true for the Dutch comparative particle dan. But it must be adm1tted, of course, that this solut1on to the problem with which GoaJlrO confronts us contams a high degree of speculation, and that this language remains a 'ase whtch seems to res1st straightforward classificatiOn.
x·
The Typology of Comparattl'e Constntctwns
51
hn.tlly, we mu~t 'Oil\Jdcr om.• r.ulwr pccuh.~r l'.l'>t' oJ 1111xcti NPcontparlson. Telugu, a Orav1d1an IJnguagc, markr, the o,t •.md:1rd NP m rt~ comparatiVes by means of the affixes -kanna or -kante: (34)
TELUGU:
I
-pandu a
this-fruit
-pa11du-kanna trpr -ga undr
that-fruit -llRT
sweet-one 1s 1 11us fruit is sweeter than that fru1t'
The suffixes -kanna and -kante arc not spatial postpasittons m Telugu; they mean ne1ther 'to' nor 'at' nor 'from', nor can they be 1dennfied as p
hke the followmg: (35)
TELUGU:
a.
Ramarav podugu-vadu ka -du R. tall -one not·IS 'Ramarav ts not tall'
•
I -pandu trpi -dt ka -du this-frutt sweet-one not-is ·Th1s fruit is not sweet'
To be exact, the element kanna is the Infinitive form of this negative copula, a fortn which " used to represent the predicate of subordmate temporal clauses. The element kante must be v1ewed as the Adverbial Conditional of the negative copula, a form which codifies the predicate of subordinate condinonal clauses. Hence, we might say that the compara-
tive construction in (34) has in fact the literal meamng: 'Wh1le/if that fru1t lS not (sweet', th1s frutt IS sweet.' In other words, the Telugu comparative migbt be taken to be a Conjoined Comparative of the polar subtype; as such, Telugu would be deviant only m that 1t has reduced one of che conjoined clauses in the constructiOn to a subordinate form, whereas the regular cases of the ConJoined Comparanve prefer to keep the structural coordmauon of the two clauses mtact. On the other hand, one m1gbt also maintain that Telugu is a language With a Particle
Comparative. The fact that the particle(s) 10 quesnon are verbal fortns w1th the meaning 1not being' cannot be regarded as an argument against this classification, Ill v1ew of the fact that, apparently, particles in such
•
52
A Cross-linguistic Typology of Comparatives
comparatives are of a heterogeneous origin anyway. Given these alternatives, and the lack of argumentation to make a princ1pled choice between them, I have decided to treat Telugu as an indeterminate case which defies straightforward classification. The comparative construction in T elugu will therefore not figure in the discusston of the various comparative types that will be presented in the following chapters; its occurrence will not be brought up until, m chapter 15, a general framework for the explanation of the occurrence and distribution of comparative-types has been developed.
3 In Search of a Determinant Factor
Now that the description of the data on the comparative constructions in our sample has resulted in the establishment of a cross-lmgmstic typology, our next task will be to find a prmc1pled way to account for the distinctions which th1s typology offers. As I noted in section 1.2, such a principled account presupposes that we are able to find an explanation for the attested occurrence and non-occurrence of the categories which figure in the typology, and for the att.,ted d1stnbution of languages over the options wh1ch the typology offers. A natural way to provide answers to these questions is to search for a second typological parameter which can be identified as the determinmg outside factor of our original typology of comparatives. That is, we will try to categorize the languages in our sample in an additional typology, m such a wa} that categunzahOIIS m this second typology match the distinctions which we have attested m the typology of comparative&: this matchmg can be laid down in a set of implicational umversals. In as far as the parameter of the second typology can be argued to be more basic or fundamental to language systems than the first (i.e. than comparison), we can call this second typology the determinant of the typology of comparative constructions; it serves as the basis upon which the non-randomness of variation m the encoding of comparison can be accounted for, and hence It can, in some sense, be said to constitute an explanation for that varia non.
3.1 Word order as a possible determinant At first glance, It would seem to be evident from our data that ba.Mc word-order type is a likely candidate for the function of determinant m our typology of comparatives. As we observed, for at least four of the maJor categories in the t~·pology a specific cho1ce of word order b a very stringent, if not necessary, condition. We may express the relations whkh
A Cross-lmgurstu: Typology of ComparatiVeS
54
we h.wr founll lo l"CI\1 hrtwt•t·n word·ordt·r p.liH'IIl ,liH\ ~.-hml.'(' of cmnp.u.wvc type 111 the followmj!, tour gcncr.l\..,t.ttcnwnt~:
(1) a. If a language has a Separattve ComparatiVe, then its basiC word orderrs SOV. b. If a language has an Al/at1ve Comparative, then 1ts baste word order is verb~mrtral.
c. If a language has a Locative Comparattve, then 1ts b= word order ts etther SO V or verb-inittal. d. If a language has an Exceed ComparatiVe, then tts basic word order ts SVO. Naturally, these four statements should not be taken as absolute unlvcrsa\s of language; they formulate tendenctes, m the sense of Comne (1~81: 27). We may say, however, that, as tendenctes go, our statements seem to be fatrly sound. Restncnng ourselves to the pnmary choices of comparative type, we can observe that statement ( 1a) i& contradicted by only four out of 32 languages m the sample (vil. ClasSical ArabiC, Ca::ur d'Alene, Guaram and B1blocal Hebrew); statement (1 b) is contradicted by one language out of seven (v1z. Kanuri); and statement!; (1 c) and (1 d),
wluch concern 12 and 20 languages, respective-ly, are not contradicted at all m the sample. There thus seems to be some ground for posit1onmg tentatively a correlatiOn between baste word-order type and cho1ce of comparative type; this co1·relanon IS of the kmd wh1ch is expressed m the above four general statement!:!. In what follows, I will refer to the
statements in (1) as word-order-based unwersals of comparative type chotec.
Now, although the four statements m (1) dearly represent vahd umversal tendencte!) 1 one
13
nevertheless left wtth the unea!,y feeling that,
somehow, thts set of word-order· based umversals d<>elo not teD the whole Mory about the attec;ted occurrence of comparat1ve types. Th1::. susp1cton IS confirmed once we examme to what extent the parameter of word· order vanatton can be said to prov1de an explanatory framework for the choice of a pamcular comparative type. If we look at the three require· ments for umversal explanatiOns (see sectiOn 1.2), we can note thJt the parameter of baSic word order w11l, at its best, fulfil only one of these reqUirements, namely, the requirement of bemg the 'more basiC' hngmstic feature of the two parameters mvolved. Since the pubhcat1on of Green· berg (1963)) it has become a common assumption m Umversal Grammar that word order IS a fundamental charactenstic of natural languages, from which vanou~ other cases of cross-hngutsnc vanation can be deduced (for Instance, the order of elements m phrases, the direction of deletion and adjunctiOn procedures, and so on}. Hence, It IS Cl!rtamly
ln Search of a Deternmumt }<'actor
55
JUsttfiable to bring up basic word order as a 'deeper-lymg' causal factor 111 the l111guistk typology uf uunp.H.tdvc tOII..,II'lii..IIOil"· On the other two count~, however, the ~t of word-un.lcr-b.u,cd UI\IVCI'..,,l[~ c.m be !:>hown to fail as an explanatory framework. For one thing, It can be observed that the set of umversals stated in (1) does not exhaust all the theoretically possible cases of combmations of word-order types. It IS a generally accepted fact m Umversal Grammar that natural languages are distributed over three maJor word-order types, vtz. SOY, YSONOS and SYO; these three word-order types are correlated to comparative types m the statements (Ia), (!b) and (ld), respectively. In addition, however, statemem (lc) clauns that a speafic comparative type, viz. the Locative Comparative, 1s correlated with either SOY or VSO word order. Now, given that at least some comparatiVe types can be correlated to two word-order types to the exclusion of the third, we may expect there to be other comparative types for which a correlation with a dtfferent combmation of word-order types can be ancsted. As It is, however, our data show that It is only the SOV-VSO comb1nauon whtch correlates wtth an empirically documented comparative type; there are no comparatt\'e types for which a cho1ce for either SOY or SYO, or a choice for either SYO or YSO, IS prescribed. As It Mands, the set of word-order-based umversals formulated above offers no pnncipled account for this empmcally attested exclusiOn of theoretically poss1ble combinanons of word-order types, and Its explanatory value is considerably weakened by this fact. As a second pomt, It can be !!een that the set of word-order-based umversals covers only four classes of comparative type~ out of six; for the ConJomed Comparative and the vanous mamfestattons of Parnclc Comparatives no correlation whatsoever wtth a parncular type of basic word order can be established. One might argue here that the Particle Comparative is only a mmor class, 1f a das~ at all. But even if we grant this, we are still faced With the uncomfortable situation that the correlation of comparative type to word-order type breaks down m more than 35 per cent of our pnmary cases of comparative constructions; this situation gets even worse once secondary options are taken mto account. l\1oreover, by accepting word order as the primary determmant of the typology of comparatives, one highly typical and Important variant m that typology, VIZ. the ConJOmed Comparative, Js left without any explanation of Its occurrence at all. Constderauons of this kind suggest that the relation between basic word-order types and certain comparative types, while undoubtedly real, should nevertheless be thought of as bcmg of a more mdtrect nature. They suggeM that we should be able to track down a new typological
56
A Cross-lingutstic Typology of Comparatives
parameter whrch intermediates between these two typologies. Of this intermedtate parameter, it should be required that It lead to a typology m which each category IS correlated both to a certain comparative type and, at the same tlme, to a certain word-order type (c.q. a combination of word-order types, or a lack of defimte word-order type). In this way, a word-order-based umversal of comparative-type choice such as (1 a) If a language has a Separattve Comparattve, then it has baste SO V word order
should be replaced by the syllogi;m (2), wh1ch clearly illustrates the concomitant status of word-order type in comparative-type choice: (2) a. Languages with a Separative Comparative belong to type A in the typology of X. b. Languages of type A in the typology of X typically have SOV word order. c. Therefor"' languages wtth a Separative Comparative typically have SOV word order. Thus, the establishment of a new determmant parameter for the typology of comparatives enables us (at least in principle) to maintain the empirical1y attested correlations between word-order types and certam comparative types, while at the same time the drawbacks of a d1rect correlation between these two typologies stand a chance of being eliminated. In the new model, the problems with which the set of wordorder-based universals confront us are transformed into issues regardtng the correlations between word-order types and the variants of the new intermediate typology. Hence, the explanatory value of this new parameter can be examined from two different sides. On the one hand, the new typology should be able to make correct predictions about the attested characteristiCS of the typology of comparative constructions. In addition, the new parameter should abo be able to incorporate the set of word-order-based umversals in a prinCipled, unproblematic fashion.
3.2 Spatial relations as a possible detem1inant
If we accept the lme of reasoning outlmed in the preceding section, our next problem IS of course to find out what this assumed new 'intermediate' parameter m1ght be. At this point, the system of the ways m which spatial relations are encoded in natural languages readily presents itself. As we observed in the foregomg chapter, the relation between spatial expression.!! and comparattves is most conspicuous in the case of
In Search of a Determinant Factor '
57
the various types of adverbial comparatives, but there is also a natural way m which the Exceed Comparative can be viewed as spatially derived. In short, one might venture the hypothesis that the typology of compa ratives ts the way 1t is because comparative constructions in natural language; tend to borrow their linguiStic mantle,tlon from the codtficatton of certain types of spatial notions. If we mvesttgatc this hypothesis for ItS value as an e>.piJ.nat1>ry framework, we can conclude that it IS certatnly satisfactory frorr1 a conceptual point of view. The idea that spatldl relations consntutc a more fundamental concept than comparbon IS not ltkely to meet wtth senou~ dh.agreement among lingmsts or psychologists. There are, how¥ ever, a number of ob1ecttons again~t this hypothesis whtch have to do with the formal way m which the parameter~ of spattal rclatton' and compan~on appear to be connected. First, it should be observed that the hypothesis, as it stands, does not offer a full matchzng of the types in the two rypologi"', While a relatton between spabal expressions and comparatl'\oes is ev1dent in the case of fixed-case comparatives, no such matching can be made m the case of derived-case constructions. Now, it mtght be argued that this tmperfect matchmg is Just a fact of life; one might mJ.IIltain that, apparently, some languages choose a spatial model in their encodmg of comparison, while other languages Simply do not. I think one has to admit that there IS, at present, no way to prove that such a p~ition is mconecr. I do feel, however, that, as long as the force~ behmd the phenomenon of linguistiC variation are still largely a mystery, one should aim at the formulation of universal theories which cover all the different hnguisnc mambtahOns of a given parameter, mstead of only a subset. As a second, and perhaps more damaging, objection, we can note that the hypothesiS, in its above formulation at least, falls short of the Criterion of exhaustiveness whtch I thscu..ed in sectron 1.2. Even for those languages in whieh a direct modelhng of comparison on spatial relations is evtdent, the hypotheSI• offers no restricted framework: it docs not answer the que!:ltlon of why it IS that some of these languages select a separative rel.ttiOn as their structural template of compari~on~ whtlc others prefer an allative or a locative relation. GIVen the plausible assumption that, in prmciple, all of these language~ have the means to encode all of these different spatial notions, we mru.t conclude that the '1ocalist' hypothesis advanced above 1~ not con~tramed enough to enctble us to make precise pred1ctlons vf the choice of comparative type m a given language. Notwi:h~tanding these objections, however, it remam~ an undemable and h1ghly !.ignificant fact that there IS a clo~e connection between the
A Cmss-lmguJSttc Tyf>ology of ComparatiVes
58
t:xpn·,..,wn of . . p.111.1l Jrl.111on.., .mtl romp.m . . on m .1 l.1r~c nwnlll'r of unrela1ed !Jngu.ag;t'-!1. Faced With thl:, ~mmgly par~1dox~e.1l !1-JtnatJon, I have adopted the followmg lmc of argumentation. I a"ume that the mental act of companson 1~, In ib cogmtive representation, indeed based upon ~patlal concepts; a further elaboranon of thi!. assumpnon w1ll be gwen 10 section 11.3. However~ a~t far as the linguistic codification of the concept of compartc;on 1!. concerned, I wtll defend the position that It is not ~panal location wh1ch IS the dctermmant factor m this encodmg, but rather the notion of temporal rhatnmg. Agam, I may pomt out here that locahst authors (e.g., Lyons (1977: 719)) have claJmed the ex~stence of a
progresstve degree of abstracuon m the order of 'spatial location', •temporal locaoon' and •absttact locaoon' (such as grammatical funcnons, <xJstent!ality, posse~~1v1ty and causanvity). GIVen this framework, the <entral cla1m of this study can be pbra,.d as tbe statement that comparative constructiOns m natural language~ are one more case of 'ab&tract locanon'. 3.3
Temporal chaining as the determinant of comparative types
\~a cogmt1ve
nouon, temporal chainmg can be defined as the process by whtch the mmd establishes 'the relation between two events, A and B, as overlappmg, preceding or following each other' (Traugott, 1975: 208). The rt.-sult of the mappmg of tht!. process onto a language system is a temporal cham, 1.e., a semannc configuration m wh1ch t"WO tensed
proposmons (representmg states or events} are presented succcsstvely. We will see m chapter 4 that, semantically, temporal chains can be d1v1ded mto two classes. In one class of cases, temporal chams must be mterpreted as staung that the events or states m questiOn occur at the same pomt m time; these temporal chains w1ll be referred to as stmultaneous (temporal) chams. The other class, m which the events or states refcncd tom the ch.ain must be taken to occur or ohtam one after tbe otherwdl be called consecutwe (temporal) chams. In addmon, I ac;sume that thts c;emantac or cognitive temporal chaining is mirrored formally by a range of syntactic consttucnons m which sentences or pred1ca~ are hnearly ordered w1th respect to one another. Thus, syntactic temporal chammg will be assumed to be the formal hngutstlc correlate of the semantic chammg of propositions; syntactiC chams represent the formal way~ m wh1ch the cogmnve process of temporal chaming IS encoded m natural language. Natural languages may vary m the ways m which they choose to encode temporal chaming, anJ hence It 1s possrble co con~truct a typology of syntactic chams. One
In Search of a Determinant Factor
59
of the h!.'M-known ,md m.110r w.tyl't m which tcmpor.tl ..·h.unmg can be torm.lhLcd 111 n.nur.tl l.mguagc& it:~ the .1pplic.1t1on ul the &trm:tur.tl procedure of coordmation of sentences or prediCates. However, ab we &ball see, coordination IS by no means the only option m syntactiC chaming constructions across languages. The central aim of th1s study IS to provtde evtdcnce for the positiOn that temporal channng must be 1dencified as the determinant of com para· nve-rype chmce m natural languages. Thus, I w1ll argue: that the chotce of .a particular comparative type for a language L can be predicted from the type (or types) of temporal chainmg which that language permits, and
that the attested occurrence of comparanve types across languages
1s
determined by the theoretically possible types of temporal chams m natural langiiages. As will become clear in chapter 4, the possibility of having certain types of temporal chammg m a given language is in Its turn pamally constrained by the basic word-order pattern of that language. As a result, the correlations between comparative types and word-order types can be accounted for. Thus, I will argue for umversal statements of the followmg general kind:
(3)
Languages wrth a comparative construction of type A must have a temporal charning construCtiOn of type X. Smce languages With a temporal chatnrng construction of type X typrcally have the word-order optton W, languages With a comparatrve of type A typrcally have the word-order option W.
In chapter 4 I shall g1ve a survey of the ways m wh1ch temporal chaming is structurally reahzed 111 natural languages, and followmg that I shall present a definitive statement of the universal.;; of comparative-type choice which I defend. Before domg so, however, It may be inreresting to advance a number of prehmmary reason• for thinkmg that temporal chaming might be the proper candidate for the mtervening parameter 111 our model. In particular, I want to quote here a number of facts whiCh seem to pomt to a possible formal relat1onshap between comparatives
and coordinations.
3.4
Preliminary data
Preliminary md1cat10ns that temporal chammg may be the appropnate candidate for the functwn of determinant factor m comparative-type choice stem from a vanety of sources. Fust, we can note that by postulatmg temporal chaining as a typological determmant, we can explam the occurrence of at least two of the maJor classes of compara-
60
A Cross-linguistic Typology of Comparatives
nvcs in a natural and direct fashion. These categories of comparative constructions appear as syntactic temporal chains even in surface structure: the ConJoined Comparative is structurally a case of sentential coordination, while at least some cases of the Exceed Comparative have the surface form of a :.enalizatlon, which ts a maJor syntactic chaining type 1n these languages. 1 Funher preliminary evtdence for the crucial role of temporal chammg in the choiCe of comparative type can be derived from a number of ISolated facts about the morphology of the comparatiVe constructions m various languages. At first sight, these facts may appear to be nothmg more than oddities, but they fall into a pattern once we assume a relatton between comparative formation and the formation of coordinated (1.e., a spectal type of chaining) constructions. On the one hand, we can note that, for a number of languages with a Particle Comparative, the comparative particle seems to be a lexical item that is also in use as a marker of (some type of) coordination between sentences. In Javanese, the element karo not only marks the standard NP in comparatives, but also occurs freely as the coordinating particle 'and'. Cp.: (4)
JAVANESE:
a.
Enak daging karo iwak is-good meat than fish 'Meat is better than fish'
7.
Bapaq menJang mg-desa karo simboq menJang ing-desa uga father go to -field and mother go to -field too 'Father went to the field, and mother went to the field too'
In Toba Barak, the comparative particle asa is also in use as an adverb tal consecutive marker, with the meaning 'and after that' or 'too': (5)
TOBA BATAK:
a.
Dumejak utang~na asa torop di obuk more-many debt -his than crowd of hair 'He has more debts than hairs on his head'
b.
Ningon dapot ho do i , asa gabe ho first have-got you EMPH this then rich you 'First you must have this, then you will be rich'
A similar siruatton as in Toba Batak may be claimed for the English comparauve particle than (see Small, 1923: 8; see, however, Joly, 1967 for a different opmton)) and possibly also for Standard Dutch, where the comparative parttcle dan is homonymous - and perhaps diachronically related- to the sentential consecutive marker dan •then'. Cp.:
In Search of a Determinant Factor (6)
STANDARD DUTCH:
a.
Jan is grater dan Piet J. is taller than P. 'Jan is taller than Piet'
b.
Eerst ga ik , dan gaat Jan first go I then goes J. 'Ftrst I will go, then Jan will go'
Furthermore, we observe that in Bari (Spagnolo, 1933: 266) the comparative particle na gwon functions as the adversative sentential coordinator 'but': (7)
BARf:
a.
Nan ktta bya na gwon do yeye7u I work more than you think 'I work harder than you think'
b.
SOromundi kata , na gwon kala 'bayin groundnuts exist but teeth exist~not 'We have groundnuts, but no appetite'
The situation m Bari is identical to that in Basque and Ilocano, where the comparative particles (baino and ngem, respectively) also function as the adversative coordinator 'but': 2 (8)
BASQUE:
a.
Jakes
b.
Ethorri da , bainan ez gogotik come-PCP. PERF he~is but not voluntarily 'He has come, but not out of free will'
(9)
ILOCANO:
a.
Nainimbag dattoy ngem datta good this than that 'This IS better than that'
b.
Nasayaat ti porion -ko, ngem daan bassiten good~one is carriage~my but already oldish 'My carriage is fine-looking, but It is already somewhat old'
baino lodi-ago da J.-NOM than fat -PRT he-is 'He is fatter than Jakes'
Lastly, the use of the coordinator but as a comparative particle can be attested both in older variants of English and in at least some dialects of modern American English. Sentence (10) is a !me from a play by Wilham
A Cross-lml{ursttc Tyfwlogy of Comparatwes
62
~h,tkc'lp<.:.!rc, JJJd '>l'fHt:lll-C (1 I) ,., .1 4Uotc hom .1 ":.ollg
by
~outhcrn
country-and-western ~mger Hank W1lhams: (1 0)
RENAISSANCE ENGLISH'
Thou knowst no less but all (Twelfth N1ght, 1,4) (11)
AMERICAN ENGWJ-1'
Ain't no more left but the blues to cry
In th1s connection
It
is also worth notmg that, m a number of
languages, the comparative particle which marks the standard NP seems to be identical to the d1s]unct1ve coordmatton marker 'or'. Such a language IS Class1cal Greek. Alexandre (1880: 644) cites the following phrases from the works of Plato: (12)
CLASSICAL GREEK'
a.
Chresthos e poneros good or bad 'Good or bad'
•
Sofoteros e su Wiser than you 'Wiser than you'
The same SituatiOn seems to have heJd m Gothic, where the item thau represented both companson and diSJunctiOn (Benveniste, 1948: 140). In that same paper, Benvemstc mentions the dtsjunct1ve and comparative parllcle It m Old Slavomc. Small ( 1923: 36) states that Swiss and Middle H1gh German dialects employ the disJunctive element weder as a marker of standard NPs m NP-comparatrves. Bcrgmans (1982: 78) report; that 111 several Eastflemtsh and WestO.emish dialects there ts a comparative rartJde of, whJch COIOC!des wtth the diSJUDCUVe element of in Standard Dutch and Standard Flemish: (13)
EASTFLEMISJ-1/Wt.STFLEMISJ-1'
a.
Komt h11 vandaag of morgen comes he today or tomorrow 'Will he come today or tomorrow?' Tk ben grater of mqn broer I am taller or my brother 'I am taller than my brother'
There are also cases m whtch the comparative particle seems to be Identical to the element wh1ch represents the negattvc coordinattve
In Search of a Determinant Factor
63
p.madc 'nor'. In lu~ dl\l.ll~''on of the Engh,h comp.li.IIIVl', Joly (1967: 17) states that nor used as a comparanve pamde '">toll •love m modem d1alects, from the Shetland Islands to Cornwall, from Ireland to the United States'. Example (14a) was observed in a number of West-English dialects, and example (14b) in an American dialect: (14)
NON-STANDARD ENGWH:
a. b.
I know better nor that lt'srichernoryou'lleverbe
Joly further mentions cases In which the po>Itive diSJUnebve irem or IS used as a comparative particle m Engbsh dialects. The use of or as a comparative marker appears to have been Widespread m Northern British dialects from about 1250 onwards. andean snll be found today m ScottiSh dialects:
(IS)
SCOTTISH ENGl.ISH:
tattle Der no a thing at I leek better ir a rosn there not a thmg that I hke better or a roast(?) potato 'There is nothmg !like better than a roast potato' Speaking of Scotland, we should mention that ScottiSh Gaehc use> the negative coordination na 'nor' as the parncle of companson> a feature which is normal for Celnc languages. Cp.: (16)
SCOTTISH GAEUG:
h baine Calum na Seumas IS fa1rer C than S. ~Cal urn is fauer than Sewnas' (17)
MIDDLE WELSH: (Strachan,
!909: 133)
Na werthu na ellwng nor sell-FliT. !SG nor release-FUT. !SG 'I will neither sell it nor Jet it go' Fmally, the situation in Latvtan's primary comparative 1s completely parallel to that m Gaehc: (18)
LATVIAN: (Endzelin, 1922: 233)
a.
Meitas
vecakas ne
mate
daughters older than mothers 'The daughters are older than the mothers'
Man nau ne tes ne mat I have nor father nor mother '1 have neither father nor mother'
A Cross-linguistic Typology of Comparatives
64
Not Only do the comparative particles of various languages appear to exhtbit relations to sentential coordinative (i.e., chaining) markers, but we also find cases where the obligatory morphological marking on the comparative predicate seems to be related to coordinatlllg items. For mstance, in Hmdi (whtch has a Separative Comparative if the standard NP is overtly expres.!led) the predicate in a comparative construction m which there is no overt standard NP must be preceded by the element aur. This item is also commonly employed as the connective element 'and' between dauses and phrases: (19)
HINDI:
a.
Ap aur bare haim you PRT big are 'You are bigger'
•
Usne bang/a sikhi
thi
aur hindi
he Bengali learning was and Hindi 'He learned Bengali and Hindi' A very similar situation is found in Goajiro, a language whJCh we discussed in section 2.5. In Goajiro, the item au/aka 'and, too' is reportedly (see Holmer, 1949) very often found accompanying the predicate in comparative constructions: (20)
GOAJIRO:
a.
Au/aka Rupeta mulia'u hulia Rafae/a PRT R. big than R. 'Roberto is taller than Rafaela'
b.
Au/aka etka ci
ta -melei
and dog this me-with 'And this dog is mine' In Tajik (Rastorgueva, 1963), the conjunction kham 'and, too' is at least phonetically identical to the marker of comparative predicates: (21)
TA}lK:
a.
Ruy-ash az
barf kham safed ast
face-her from snow PRT wh1te IS 'Her face IS whiter than snow'
Shumo na -rafted
man kham
na -raftam you not-go-PAST. l.SG I and/too not-go-PAST. !SG 'You d1d not go and neither did I'
In Search of a Determmant Factor
65
An identical situation obtams in OssetiC {Abaev, 1964): (22)
OSSETJC:
a.
Deu-e7 chuyz-der you-from good -PRT 'Better than you' Stalite serttwtoJ , meJ der skastt stars began-shine moon and/too rose 'The stars began to ~hme, and the moon came up'
Lastly, we note that tn Tarml (Beythan, 1943; Asher, 1982) the standard NP in comparatives is commonly followed by the suffix -um. Thi'> suffix is also the connective particle in conJunctiOns. Cp.: (23)
TAMIL:
a.
Agarar-il -um periyar antanar kings -from-PRT Brahmans are-great 'Brahmans are greater than kmgs' Agarar-um penyar -um kings -PRT Brahmans-PRT 'Kings and Brahmans'
All these facts, then, suggest that 1t mtght be worthwhile to take a look at coordinated (and other types of chaining) constructions as a possible determinant of comparative-type chotec. But, of course, the actual proof of whether such a determmmg relation exJ~ts must be given by examining the predictions which follow from such an a~:'.umption for the languages in the sample. I will turn to this matter in part two of this study, after an exposition of the syntactte types of temporal dummg (chapter 4) and the final statement of the universals which I propose (chapter 5).
4 Types of Syntactic Chaining
4.1
Introduction
A~ h•.h been noted by theoretical hngut!)t!>, comparatlVt' lmgut~t!>, Jnd gucmmarran~ of smgle exonc languagec;, syntactic chammg of a scnes of <·vem·cxpre~smg sentences or pre
ddferenc ..rates of affaus. Such a senes can be used to express either srmultaneous acllon (m whtch the events de~cnbed by the predJCates m rhc sene~ are taken to happen at the same tune), or consecuttve actzon,
wh1ch 1mphe!> rhat the event& descnbed by the predicates occur one after the other. Thus, from a scmanuc pomt of v1ew, we may conclude that the
'<mantle nonon of temporal clmmng, as defined 1n the precedmg chapter, allow& for two semanttc subtypes, one m which the propo~mons 11 the semantlc reprec;enranon are ordered w1th respect to temporal !.uccesston, and another 10 wh1ch no ~uch ordermg 1~ Imposed. For a number of languages, the semantic dtstmctJon between s1mul· taneous and consecutlVf chammg 1~ matched by a ~yntacnc dtfferen· tlJlion at the level of ~urface ~jfructure. Such IS the ca5.e m fgbo. A-::. JS pOinted out 111 Weimer,\ ( 1973), lgbo has a simultaneous awon construe! 1011, 111 which the fir!!t verb 111 the sen~ can be 1n any tt.'ll~e. All following
verbs m the cham, however, tnU'it appear m the lncomplettve Form, that IS, a type of verbal phra'::>e consJ~tmg of the particle nil followed by a verbal noun. Thu'::> we have: (1)
IGBO:
Ha no n'oce nil Crt nn they ~at down PR. f eatmg food 'They :io.lt .md ate/They s.tt eaung' In the consecutive acuon con'::>tructiOn m Igbo, the fir::.t verb m the sencs may agam be m any tense, but all to1lowmg verbs have to be put mto the :>o~c..aHed N.arrar1ve Form. Jn the regular case, th•~ form conMSts of a verb stem With obligatory low tone and a verbal suffix -if-d. Thm. we get:
Types of Sy11t,u tsc ( :lwmmg (2)
J(;Jj();
() 1111
c,l!,lJt' g/m-e
ilgu
he took gun k1li~NAI{R kop.mJ 'He took a gun and killed the leopard' We can conclude, then, that m Igbo Simultaneous action and consecutive actton are formally kept apart m surface structure. In this particular language, syntactiC dtfferenttauon between the two chammg types takes place by repre.!>entmg non~first pred1cates m a cham by means of fonnally different verbal formatwns. As we shall see m the followmg sect10ns, other languages may employ d1ffercnt procedures to mark thts differen~ tlatwn; but smce the actual mechamcs of symacnc simultaneous and consecunve chammg m spec1fic languages are largely beyond the scope of th1s study, we wtll not pur~ue rlus matter further here. Oppo.<.cd to cases like lgbo, we find other languages m wh1ch the semannc distmctton between !:>Jmulraneous and consecut.JVe chammg 1:. not matched by a syntactiC oppo<.mon m <Jurface structure, at !east not m the typiCal case. For example, the Engh~h coordmanon (3) may be mterpreted both as a case of simultaneous acuon dnd as a case of consecutive acuon, dependmg on whether John and Mary performed a~ a duo or whether they were two separate ac~ m some cultural happemng: (3)
ENGLISH:
john played the flute and Mary rectted a poem by Yeats Sumlarly, Amhanc (Cohen, 1936: 142) encodes tt~ temporal chams by purtmg one of the predicates m the cham mto the so~called Gerund1al Form, while the other predtcate is constructed as a mam predicate. Gerund1ai Forms are subordmate but finJte forms, whtch have no tense of rhc1r own; thetr exact temporal mterpretatwn must be mferred from the ten~e
markmg on the mam verb, and from contextual facrors. Hence, m Amhanc, the same !lubordmate form serve~ a~ the codtficanon of both stmultaneous and consecutive chammg: {4)
AMHARIC:
Ta/ast sastwo tamallasna enemy-NOM flee-GER. 3~G return-PERF IND 1PL 'Whtle the enemy fled, we returned' or 'After the enemy had fled, we returned' A<:, a last example of languages of thts sort, let u& constdcr the syntacnc cha1ning construction in Mandarin (Lt and Thompson, 1978). In this language, a promment way to encode cases of temporal chammg ts the u<:,e of so-called senal verb constructions, m whtch predicates are strung
-
..
68
A Cross-linguistic Typology of Comparatives
together m a series wtthout overt c-oordinative marking. An example of this type of chaming construction ts the following: (5)
MANDARIN:
Tii kiin dtJnymg chi pingguo he see movte ear apple 'He :c-aw a movte and ate an apple'
Now, a~ L1 and Thomp~on (1978: 241) remark, senal verb constructions m Mandarm are essentiaJiy mdeterrnmate as to thetr temporal mterpretation (and, for that matter, as to other semantic nuances). Thus, a senal verb construcuon may, on one case, receive a simultaneous mterprctat1on (see (6a)), whereas in other cases this same syntacttc construction 1s naturally interpreted as a consecutive chain: (6)
MANDARIN:
a.
Tii xze
b.
Tii mtit pzdo Jinqu he buy ttcket go-in 'He bought a ttcket and went m'
Xt:ioshuo melt gUdong he wnte novel sell antique 'He writes novels J.nd sells ant1ques'
Wuh regard to thts ~ttuatlon, Lt and Thompson (ibid.) state: 'What ts evtdent ... JS that the ~enal verb construction IS used to encode a number of dtfferent relatwnshtps between predicates m Mandarin. These predicate relatwn~lups are structurally distinct in most other languages because of the presence of grammatical markers.' Somewhat further on, they observe: A good part of the knowledge of pO!>!>ib!e relattonsh1ps between clauses m a sentence 1s acqu1red by normal people a!l a result of thetr cxpenence m the world. Th1s knowledge !S language-mdependent and results from our perception of and expenence wnh the world. Gtven any two events, there can only be a small, fimte number of rel.ttJonshtps between them. Language merely reflects and codifies these relatiOnships m vJtlOUb multl·predtcate constructions, although d1fferem languages employ different strategws 111 thetr cod1ficanon. (p.244-5)
We may ~ayt rhen, rhar m ca~es like English, Amharic and Mandarm the semanttc distmctiOn between !:.!multaneous action and consecutive action has been formalLy neutraltzed In the codification of temporal chaining. As such, these languages are opposed to cases hke Igbo, in which thts semantiC d1stmction is mtrrored by two different procedures of syntactic encodmg. Ir should be remarked in thts context that the boundary between
69
Types ofSyntactrc Chammg
neutralizing and mirroring languages IS a tlutd one, and that there are also cases of syntactic chainmg which may be called instances of parttal neutraltzatton. As an example of such a case, let us com.tder a type of
cham in latin, the
so~called
Ablattvc Ab1:1olute construction. The details
of this construction wtll be discussed m chapter 10; for the present purpo~e, a few general remark!, may ~uffi~e. First, the Ablattvc Absolute can be used only for those case~ ot temporal chammg 111 whJCh the subJeCt<; of the predtcates m the cham are not Identical. Secondly, m a cham of th1s type one of the predtcares IS constructed as a mam predicate, wh1le the other prediCate receives the form of a non~finite participle; partlctples are nominal-adjectival forms wh1ch, like other nouns and adjectives, are marked for case, number and gender. A third feature of the AblatiVe Ab~olute 1s that the '>llh)ect of the predtcate w1th the participial form is put into an obhque case form, VIZ. the ablative case, and that the partictple of which It 1s the ~ubJect ts made to agree with tt m cas.e, gender and number; hence, the parttnple too ts marked for abldttvc case. Examples of thts type of Lattn chammg comtructwn are: (7)
LATIN:
a.
Dommo bzbente master-ABL. MN!C SG dnnk-PCP. servae cantant
PRE~
ABL MASC.
)C
slave gtrl~~NOM. PL smg-PRI:S.lND. 3Pi 'Wht!e the master drmks, the slave gtrls smg'
b.
Galltii vtctii G.-ABL. FEM SG conquer- PCP PFRl-. PAS~ ABL. I-EM SG Caesar reventt C.-NOM return· PERF. IND. 3SG "After Gallia had been conquered, Caesar returned'
The examples m (7) make 1t clear that the Ablat11·e Ab1olute m Latin can be employed both for cases of Simultaneous actiOn and for case~ of consecutive action: both types of temporal chammg mvolve the '5:lme syntactic procedure. However, it cannot be sard that the Latm Ablauve Absolute con~tltutes a complete neutr ahzatton of the semantic contra-.t at tssue. Although the two sentences m (7) are denved by the same syntactic procedure, the partiCipial form!J which figure m these !>entence~ are formally distinct: for the codification of ~imultancous act1on by the Ablative Absolute the Present Partlaple has to be used, whereas the codification of com.er.:utive action in thi~ construction type requ1res the Perfect Participle. In other words, Latin has a procedure for the cod1ficanon of temporal chaimng which involves the apphcatton of one s1ngle
70 '>tl tlltn.tl pro~o.nhll\' to two lorm,t!ly <.ll~tHKt non-timtt· prcJ!L.lttVl'
form&. We mc1y conclude from thts that the L3nn Ablanve Absolute c,
netther (l cc1.se of complete neutralllanon hkc the senal verb construcnon U\ o\1(1ndartn, nor a case of complete dtverslficatton a~ Illustrated by the dt-,rmct structural
procedure~~o
m lgbo. Hence, cases hke the Lattn A~l:nwe Ahsolute wtll be referred to as. mMance&o of parttal neutrahzatton. In the followmg secnonc;, I wtll present a typology of the wa)-S m whtch th( two ~cmO'\nttc ~ubtype~~t of temporal chammg are syntactically encoded tWI)
m the languages of my sample. Now, the phenomenon of (complete or pal'ttal) neurralmHlOD l'i a firc;t mdtcanon of the fact that languages often US(' parallel or even tdentJcal procedures to cod1fy Simultaneous and consecutive chammg; this tendency wdl be formulated as a general
the final secnon of th1s chapter. llowevery m order to fac!lltate the expmltlon I have found 1t convenient first to deal With each of ~he scman~IC subtype; ;eporately. In domg
In
acuon; .as It turn!! outy It IS this subtype whtch IS the more complex of the two, and whach IS therefore more m need of clarificatiOn.
4.2 The definition of the notion 1Consecutive chain'
Betore we can set up a typology of the ways m whtch consecutive action
-5yntacttcal1y encoded aero&& languages, we will have to formulate a en tenon for the cross-hngUI&tlc Identification of thts conMruction type; m orher words, we need a language-mdependent definmon of the not1on 'consecutive cham' (or C-charn for short). As we &aw in &ecuon 1.3.2, m rypolog~eal syntax such language-mdependent definition~ are necessanly of a ~cmannc (or perhaps better: cogmtiVe) nature. Hence, our first task no"" IS to prov1de d further delmeanon of the notion 'con&ecutlve cham' m sernannc terms. The resulrmg definition of th1s notiOn &hould at least be nght enough to en.thle u~ to ·~alate the relevant con~truct10n 111 each of tht·languages of the sample. ~I mdtcated above, at the centre of the not1on 'consecutive cham' he!! the concept of temporal success1on of events. That 1s, a mmunal reqmremenr for a construction to count as a C-cham IS that It descnbes a &~tuanon m wh1ch events occur one after the other, 111 a fixed temporal orJcr. By thts rcqmrement, cono;ecutlve chams are oppoo;ed to &lmultaneous chams (S-charns), m whtch no &uch order IS 1mphed. However, although the cxptess1on of temporal o;uccco;s•on 1!1 a neces!>ary condltton on consecutive cham&, there <&rc reason!> to as~ume that Jt ts not a sufficient conditiOn. As the ltterarurc on consecuttve chatnmg IS
71
..,how!-., v.mou~ .H1tho1.., :-.t.Hc th.lt .111 ,1dd1tton.11 ..,l'lll.llll!L l.:ondttton ~hould be tmpo~cd: the event~ tn tht: Cwcham should not only be temporally ordered, but m addltlon they should be concetved of as 'parncularly closely related' (Welmers, 1973: 367). The exact nature of th1~ 'clo~e relationship' IS hard to pm down, but 1t seems that, somehow, consecutive ordenng 1mphe:. more than JUSt a mere temporal successiOn; It must also be the case that the ordered event~ in the cham should be
concerved of as successive stages m the progress of one complex 'total event'. Condus1on~ to th1:. effect can be denved from statements such as the one m Crazzolara (1933: 136), who states of the so-called Narrative Mood m Nuer: 'a IS used for connecnng successwe particulars of an event or transaction' (my ttahcs). Another example of an author who stresses the requirement that the events in a consecutive cham should be taken as the constituent particulars of one complex acnon ts Labourer (1934). In descnbmg the so-called In1uncttve Form m Mandmka, th1s author remarks that the verbal construction at l<>~ue 'marks ... that the performance of the actiOn or the acqmsJtJon of the state ts subordmated to the performance of another acnon or the acquiSition of another state' (p.202; my translation}. In connection wtth this additional condition of coherence between the successr"e events m a C-cham, It 1s also a rellmg fact that, m many languages, the pnmary examples of consecutive chammg tend to be constructions m whtch the first predicate m the cham descnbes a motion. Apparently, the k.tnd of temporal successton whtch has the property of being con~ecuttve i\ typically mmared by a movement, whiCh, so to speak, 'tnggers' the sequence of events of whtch the consecutJve series consists. Thus, an mmal movement appears to be preemmently appropnate to set up a context m whtch followmg events can be naturally interpreted as bemg contmgent upon one another; many of the examples of C-chams wh1ch we will encounter are of the kind Illustrated by the English sentences 'He went out and closed the door' or 'He ~tood up and left.' It must be admitted that there n;mam\ a cert.lm vaguenes'l m the above dcfi111t10n of consecutive chammg~ even tf we accept the additional reqlllrement of relatwnshlp or coherence between the evenrs m such a cham. Therefore, m add1t10n to pmtulatmg po~mve condmons, we may also employ a negative strategy to get a clearer picture of what must be counted as a consecutlve cham m a gtven language. We may contrast com.ecut1ve chains wtrh other type~ of con~tructiom which al~:oo express a temporal ordering between events, and which can therefore be assumed to be semantically cognate to C-chams. A first construction which comes read!ly to mmd m thJs connectiOn IS the final or purpostve comtruction. From a semantiC pomr of view,
72
A Cross-lmgutstic Typology of Comparatives
construcuons of thib type are meant to express that, of two successive events, the first event is/was performed With the purpose of brmging about the second event. Typical English examples of this final construenon include sentences m whtch the second event in the chain is encoded in the form of an infimtive, preceded by the conjunction to or in order to. Cp.: (8)
ENGLISJI:
a. b.
John stood up to close the wmdow john told that story m order to embarrass me
Semantically spedbng, final conbttuctlons are cognate to C-cham~ m that both construction types imply a successive ordering between events, and also an mt1mate relationship between the events in that success10n. There is, however, also a clear semantic difference, which can be stated m terms of the truth values of comecutivc and final assernons. Informally speaking, we can say that a final cham like (8a) is true even if the second event did not take place m reality; all that is needed for this assertion to be true is that the first event rook place, and that there was some intentiOn on John's part to perform the second action. In contrast, a Cchain like (9}
ENGLISH'
john stood up and closed the wmdow can only be said to be true if the closing of the wmdow by John actually happened. Notwtthstandmg this semanttc difference, the dtviding line between consecutive and final constmcttons is a diffuse one. This relative m~ determinacy ts at least partially due to the fact that, in consecutive constructions too, a purpose-reading is a natural implicatiOn; a C-chain like (9 ), which, stnctly speakmg, must be read as stating only a mere succession of events, is readily mterpreted as a sequence in whrch a spcc1fic mtenn.on on John's part IS present. In other words, C-chams are likely to be interpreted as a series of events m which the first actwn IS goal-oriented, and hence the demarcation between this type of construction and the neighbourmg final construction tends to be blurred. The semantic overlap between the two constructions has ItS parallel m the syntactic encoding of the two constructions which can be observed m a number of languages. For mstance, m Mangarayi (Eades, 1979) we find that there IS one smgle deviCe for the syntactic encoding of temporally successive events; the syntactic construction type 10 question may have a con~ecut1ve or a final mterpretation, depending on the context. Also, m Mandarin {L1 and Thompson, 1978) the serial verb construction which
Types of Syntactic Chainmg we discussed above sentence like (10)
IS
73
subject to this rype of mdetermmacy. Hence, a
MANDARIN' Ta hut '11a
kiin qmqt he return home see parents cHe returned home and saw/to see h1s parents' may have both a consecutive and final read mg. Even m English, we Cflme across a construction m whtch features of final and con!ie<:Utlve chammg are combined. We can note that a sentence hke (11) has the superficial form of a final cham, but the semantk content doe~ not so much 1rnply a purpose as a close successton of events. A parallel situatiOn can be auested for Dutch. In tlus language, final constructiOns are typic tlly encoded in the form of an mfinitive preceded by the conJunction om tt 'tn order ro· (see (12a}), but there are ca&e& 1n whtch thl!!. construction must receive an interpretation which is more temporally successiVe than final (see (12b)). Cp.:
(11)
ENGLISH'
fohn came home to find his apartment looted (12)
DUfCH'
a.
Ik ga naar Maastrzcht om m'7n ouders te bezoeken I go to M. for my parem> to viSit 'I'm going to Maastricht to viSit my parents'
•
De Boerenparti!
groe;de stonnachtig, om even
The Farmers' Party grew
stormily
snel
for equally rap,dly
weer te verdwtJnen again to disappear
'The Fanners' Party underwent a stormy growth, but had an oqually rapid downfall' In v1ew of thiS fluctuating distinction between final and con>ecurivc chaining, I have committed myself to the followmg guide-hoe. for tloo>e languages tn wh1ch a clear syntactic separation can be dernon!:>tmted between chains that have an excluMve consecut1ve mrerpretatton Jnd chams with an exclu&ive final mterpretatton, I have excluded all mstance& of the latter construction from my data base. Final chains have been admitted to the data baoe only for those languages in which no defimte d1~tinct10ns between both types of consuuctions could be made. Thus, for example, the Mandarin construction dlu~trated m (10) ts counted a<; an instance of a consecutive construction~ even though one mtght equally be justtfied 10 rating it as an mstance of final chain mg.
A Cross-lmRutstu Typology of Comparatwes
74
ltn.d \.li,liiHIIg I'> 1101 tlH· ou\~ r\,ll!lpll' u! -l l\lll'>!t\ldiUII IVjll'
tend::. to mtrudc upon the
'>Clll,llHH..
JreJ covered by
wlud1
r...ollbCC\.ltlVC r...ham~:~.
Most language~ have al~o a type of construction wh1ch we may call consecuttonal. Here, the occurrence of a former event tS <;een as the fulfilment of a condmon by wh1ch the occurrence of a later event t!> made posstble. A typtcal Eng!J~;h example of a consecutwnal construction ts the followmg sentence: (13)
ENGU.\Hc
john worked da)' and nzght so as to pa)' off hts dl'hts
Ab was the case wtth final cham!>. consccunonal cham'l have a number of semantiC features m common wtth consecutive chams: both types of constructton tmply a succe<><;tve ordenng between event<>, and both 1mply a certam close relanon<;htp between these events. Parallel to what we found m the ca\c of finJl chams, we can also note a semantic difference here, whiCh wn be stated m term~ of truth vJlue; to put it bnefly, a consrcutJonal cham doe~ not assert that the second event actually h:1pprncd, wherea~ a comecutive cham does. Despite tht~ ~emanttc dt fference, however, we find that, at least for ~orne languages, the dJ'>tmcnon betwt>en consccutlonal and consecut:we chammg IS not formally cxprc~~cd; Kanun, a language which I w1ll discuss m some detail m se<:non 7.2.3, may be a ca~e 111 pomt. GIVen thl5 s1tuanon, I have adopted the 'tame strategy for con.!.ecutJonal cham'> a!o I employed m the case of fi11al cham mg. Chams whKh may rece1ve a consecuttonal mterpretanon are admmed to the data ba ..e only m those cases where the language has sy<;temaucally neutralized the semantiC dtfferentJatJon between com.ecuttonal and consecutive chammg. Fmally, we find ca~es where con<;tructiOm are neutral as to thetr mrerpretatJon between con~ecunve and causal or concesswe cham mg. In Lltlll, the Ablative Ab:.olutc comtrucnon, of which ~cntence (14) IS another example: (14)
LA/IN:
Urbe
destruGtii
City·ABL. SG. fEM. destroy- PCP PI:RF PA<;~ ABL ~C. H:M
gaudebrmus be happy·fUl IND. lPL 'Whcn/hecau~e 1although
the City ha!, been destroyed, we wdl
reJOict~'
1s essentially neutral between consecutive, causal and concessiVe readtng<;. Agam, we will take note of causal and conccs!:.!Vc cham~ only m as
Types of SyntactiC Chammg
75
l.u ...... 1 1.111gu.1gr Lui . . to dllk1cntL1tr them fo1m.dly lro111 lt't11por.1llv Interpreted t.h.tllb. Concludmg tht~ dJscusston on the notion 'consecutive cham', we may state the followmg. We wtll rake thts notiOn to be defined by at least the followmg semantiC charactenstJcs:
{a) the construction must expre~s a successive temporal ordenng between events; (b) the constructwn must Imply an mnmate coherence between the
events mvolved; and (c) the construction must ao;sert the actual performance of all the events 111 the cham. As we have seen, there IS some degree of semannc and formal overlap between the notton of 'consecutive chammg' and various other construction types whtch exprCS!, sucCC!,SIVC temporal ordenng. For this reason,
the definition of the notion of consecunve chammg 1~ not wholly airtight. It may be satd, however, that, as far as our typological purposes are concerned, the above defimtwn turns out to be reasonably apphcable. I shall note explicitly those cases where there IS some doubt a~ to the correct Identificatwn of the formal expression of consccunve chammg, and arguments for the actual deCISions made m such a case wdl be d1~cusscd . •
4.3
The formal expression of consecutive chaining
In thts section I will mtroduce a first typological di!>tmctlOn m the ways In whtch consecutive chams are crov~-hnguJ<;tJcally encoded. To s1mphfy the dJ.,cu'isJon, I will confine my examples to C-chams m wh1ch the consecutive relation hold~ between JUSt two events. In d1~cussmg such con~ecunve pans, I Will use the term anterror predicate for the predicate whiCh describes the cJ.rhcr event m the chJm; the other predicate 111 the cham wtll be called the posterwr predtcate. In the typical case, the semannc ordenng between antenor and postenor events is matched formally by a left-to-nght ordenng in the syntactic expre~'>!On of consecutive chains; as expenment~ by Eve Clark (1973, 1975) have shown, complex 1>entences in whiCh the leftmm.t clause m surface structure expresses the earher action m a successiOn are cons1derably easier to process than sentences in whtch this matchmg ha!!- been thwarted. Hence, the term antenor predtcate covers a JOmt semantic ,md <;ynt.lCtiC content: tt refers to the prediCate which descnbes the earlier event, and wh1ch IS
A Cross-ltnguistic Typology of Comparatives
76
therefore rhc leftmost predicate m the syntactic C-chain. In the same way, the term posterror predzcate refers to the predicate which describes the later event m a temporal succo~ion, and which IS therefore typically encoded a~ the nghtmost predtcate tn the syntacuc express10n of that •
succe~~10n.
4.3 .1
Two baste strategies: balancing and deranking
The first typological result of a cro~-lmguistic investigation of consecutive constructions can be stated as follow:;. In order to express the situation m which two events occur m a fixed consecutive order, a language may resort to one of two baste strategtes. On the one hand, it may choose to express those two events by means of two mdcpendent clauses (and typically, though not nece%anly, It will connect these clauses by means of a connecttve particle). In this case, the Important thing is that the two predicates wh1ch express the two relevant actions remain structurally of the same rank; that IS, they are embedded at the same level of structure (see Dtk, 1968: 30). Hence, tf the total construction is not embedded, both prediCates m the cham w1ll have the finite form of a main predicate. If the predicates happen to have the same subject, coordinative reduction of one of the subjects may take place, but this will not result m a 'debalanced' structural configuration: a coordinated verb remams a fimte verb, even If it has lost its subjeCt, and it remams equally ranked to the other verb m the cham. Chaming constructions in which predicates remam of the same rank I will call balanced constructions, and languages which choose thts encodmg option I will call balancing languages. Examples of balanced C~ chains are the following: (15)
ENGLISH:
john Jumped out of hts chatr and grabbed a gun (16)
GUARANI: Oi~ke kaagwi
pe ha o -henu petet avu m and he-heard one nOISC
he -went forest 'He went into the forest and heard a notse'
Thus, the definmg characteristic of balancmg languages IS that their syntactic chams are structurally coordinattons. As a consequence, con~ ~ecutivc chams m balancmg languages are subject to the Coordmate Structure Constraint formulated m Ross (1967), as can be seen from the tll-formcdness of (17) and (18):
Types of Syntacttc Chaming (17)
I:NGL/S/1:
'What dtd John Jump out of hts chatr and grab(bed)? (18)
GUARANI: • Ma?e pa ot -ke kaagwt pe ha o -henu ? what Q he-went forest m and he-heard
Rather than keeping the balance, however, a language may also choose to represent the two predicates in a C~cham by reducmg one of these predicates in rank. In such a case, only one of the predicates in the cham retains its finite verb form, whereas the other predicate is represented., a subordinate, usually non-fimte, verbal construct. Language< of thts type I Will call deranking languages, and It• •rntacttc chams I wdl call derankd
constructtons. From a strictly syntactic point of view~ one may say that deranking languages do not have any consecutive coordmatton at all, smce their surface structures do not permtt consecutive predtcates m configurations which embed them on an equal level of structure. That
there is indeed subordmatlon mvolved m deranked chams can be shown from examples
10
Tamil, a language m whtch consecutive chams are
obligatorily deranked. Thus, In Tamtl we find the followmg constructloJO (taken from Annamala1, 1970): (19)
TAMIL:
Avaru
kavide
erudmtu
naaval
he-NOM poetty-ACC write-PERf. GER novel-ACe morzpeyarttaaru
translate-PAST.IND. 3SG 'He wrote poetry and then tran!!,lated a novel'
In (19), one of the predicates (viz. monpeyarttaaru 'he translated') IS represented as a finite main verb in the Indicative Past, while the other predicate (viz.. erudttttu 'having written') has the non-finite subordmnte form of the Perfect Gerund. Now, it can be shown that such Tamil sentences cannot be regarded as structural coordinations. In particul u, the Coordmate Structure Constramt no longer apphes; as the we 11formed example (20) Illustrates, 1t IS poss1ble to relatiV!Ze the complement of the finite verb in (19): (20)
TAMIL: kavtde eruditttu monpeyartta he-NOM poetry-ACC write-PERF. GER translate-PeP. PAIT
Avaru
naaval
novel 'lit: the novel wh1ch he wrote poetr; and translated'
4. 3.2
J)erankmg versus clause~embeddmg
Regardmg the conc<.-pt of derankmg whtch we mtroduced 1n the prev10u~ sectron, It may be useful ro add a few dar1fymg remarks. To ehmm3te a pos\tble ~ource of confusion, tt should be Mre~~ed at thts pomt that the nonon of derankmg wdl be defined here as applymg exclusJvcly to the prediCates of sentences, and not to whoJe sentences or clauses. In other words, I w1H cla~!tlfy a language as a dcrankmg language only tf, m the <.odlficanon of It<> temporal cham<>, n IS the form of the predicate m one of the sentences ttsdf whtch stgnals the subordmat1on of that sentence. Languages may employ vanous formal means to achieve this subordmate markmg of pred1cates. But whatever the exact procedure for derankmg may be, the essential pomt ~~ that, m order for a constructiOn to be called deranked, It mu.r be the predicate of one of the sentences Itself whiCh is marked as a form of non-equal rank to the mam predicate m the chain. The above defimtton of the concept of derankmg IS meant to create a sharp delineation between the derankmg of predicates and another po•Sihduy of subordmanon, >Jz. the embeddtng of whole clauses in the syntacttc encodmg of a temporal cham. As l see It, m encoding a temporal cham a language may choose one of three stmctural option& (and sometime:; more than one of the&e options). Ftrst, the language may choose to encode temporal chamc; m the form of a coordinanon; m that case, the language will be called a balancmg language. A second po~slh1l1ty IS w downgrade the whole of one of the sentence') m the chain, and to ~oubordmate It to the other sentence, wh1ch will then become the main clause 10 the cha1nmg construcnon. Usually, thts second possJhdlty of complete elauoe-embeddmg" effectuated by mean' of a subordinating conJunction. An Enghsh example of a C-cham m whtch this second opnon ha; been cho•cn i>: (21)
f"NGUSH:
After john hmlleft, l poured mysrlfa we/1-desen•rd drmk
In tillS construcnon, the an tenor sentence in the C-cham (viz. john left) ha~o been embedded as a whole m the form of a subordmatc clause, mJJ ked by the subordmanng conJunction after. A typtcal charactenc;nc of a case of dause-embeddmg 1s that 1t IS not the pred1catc of the c;ubord1nate clause~ such whtch has been marked for ~ubordmatlon; It 1\ rather the whole anterior (or, as the case may be, postenor) sentence wh1ch has been marked as such. As a re~ul~ the prcdJcate 1n a su\-oordmatc clause hke the after-claus.e m (21) IS a normal fimte verb form, wh1ch ~~ charactenzed by the same morphologtcal dcvtces as the roam verb m rhe consrrucnon.l
Types o(Syntactrc Chammg Now, I w1ll
.t:-.Mmlc
79
1h.11 h.,l.u~~..m~ .1nd d,IU.,l'-clnhcddtng .Ire two
vanants of one and the ~Jme ~trlKtur.ll ~11-.Jtcgy. Both opuon., ~h.1fc the fe-ature that the clauses as such are kept mtact m surface structure. One might say, eqUivalently, that under both optwns both of the clauses in the cham retam their full sentennal charactenstlcs, a fact wh1ch must be formally represented by the presence of two separate S-nodes in the surface P-markers of these constructions. In my opm10n, the chmce between a balanced coordmatton or the embeddmg of one of the clauses 111 a cham IS not so much a marcer of structural neceSSities as a result of the apphcatton of functional cons1derattons. In certam contexts, there may be reasons for a functional (or, 1f one prefers that term, a pragmanc) backgroundmg of one of the events m the cham. 2 If ~uch a background~ mg occurs, the structural match of this procedure wlil be the embedding of one of the clauses m the cham, whereas m the absence of such a need for backgrounding a coordinate structure appears to be the most appro~ pruue way to codify the cham. In this connection, we may pomt our that, apparently, the boundaries between coordmation and clause-embedding are not very sharp anyway~ for an elaboration of thts pomt see Talmy (1978). At any rate, we will adopt the v1ew here that consecutive (and, m general, temporal) chams which are encoded as cases of clause-embedding do not form a sigmficant variant in our typology of cham formation. l-ienee, such cases will be lumped together wnh balanced constructions m the typology. Opposed to balanced and clause-embedded constructions we find cases of what might be called 'real' derankmg, that Is, cases m which the predicate of one of the sentences in the chain has Itself been marked for subordmaoon. Thus, we wtll count as mstances of dcranking languages only those languages in which the verbal form of the subordinate clause m a temporal cham has undergone a specific formal change1 so that this pred1cate is formally distinguished from fimte mam predicates. The construction in Tamil whtch we tllustrated in sentence (19) is a case m pomt: m this constructiOn, the predicate of the antenor clause m a Ccham receives a speClfically marked morphological form, VIZ. the form of the non~fimte Perfect Gerund. In some languages, derankmg of a predicate leads to the rcducnon of one of the sentences in the chain to the status of a phrase instead of a clause; that Is, sentences m whtch a predJCate has been deranked are no longer dommated by a separate Snode. It should not be thought) however, that thts complete loss of sentential status IS a necessary defimng condJtJon on our notwn of 'deranked strucrore'; other languages may choose less rad1cal solutions m the representatLon of deranked clauses, so that at least some of the sentential properties are retamed in the surface representation of such
80
A Cross-linguistic Typology of Comparatives
clauses. Once more, It should be emphasized that the defining characterIstic of the notion of deranking which is used here consists of the explicit
subordinate markmg on the predtcate of a clause. At the risk of labouring the obvious, let me pomt out that, gtven our definition of deranking, English sentences like the ones in (22) will be rated as instances of deranked chains, smce the predicate in one of their clauses IS, by vrrtue of rts parttctpial form, specifically marked for subordination. On the other hand, the semantiCally equivalent sentence (23) is a case of clause-embedding and hence, in our v1ew, essentially a balanced structure: (22)
ENGUSHc
a. b.
Having locked the door, john undressed After lockmg the door, john undressed
{23)
ENGI.ISHc
After John had locked the door, he undressed Thus, one might say that, of two possible interpretations of the notion of der anking, I wtll choose the defimttOn which defines the narrower set of structures. Clearly, tf a language deranks its predicates (in the sense which we adopted above), that language will also necessarily downgrade its cemporal clauses. The opposite, however, need not be the case. For mstance, we find that m Guarani both coordinated structures and cases of embedded temporal clauses are possible, but that this language appears to lack the means to derank its predicate!. in temporal chains. For this reason, Guarani will be classified not as a deranking, but as a balancing language. To be sure, it is not always easy to decide whether a particular case of a non-coordinated chaining construction IS a case of 'real' predicate deranking, or rather a case of clause-embeddmg. To give bm one example of a language in which this kind of indeterminacy occurs, let us consider the consecutive construction in Tibetan. From the description of this language g1ven in Lalou (1950), we learn that the antenor clause in a Tibetan C~chain contains the element ~nas, a particle which is also in use as the marker of the ablative case m nominals: {24)
TIBETAN'
Yul
~Ia
village~to
bleb -nas rt ~Ia son arnve-frorn hill-to went
'He arrivcJ at the village and went to the hills'
Now, the problem With which this Tibetan construction confronts us is this. From the surface form of th1s construction, it is very hard to JUdge
Types of Syntactic Chaining
81
whether the marker ~nas functiOns as a marker on the predtcate m the antenor clause, or as a subordinating conjunction wh1ch IS used to subordmate the anterior clau&e as a whole. One might argue that the subordinative marker -nas IS morphologically bound to the anterior predicate, and that therefore this construction can be rated as an instance of (consecutive) derankmg. UnfortunJtely, hoY. ever, such an argumentation lacks dec1sive force. In fact, one mtght JUSt as well argue that -nas is a subordinating conJunctiOn wh1ch tS placed m clause-final po&ItiOn. and which IS then attached to the last Item m the clause, by tht' applicanon of some morpho-phonologiCal rule of attraction. Smce Tibetan 15 a very stnct verb~final language, it will normally be tht· predicate which t.') the last item in a clause, so that the attachment of ·na~ to the predH::J:te is only to be expected; but thb does not ncce.<.sJrily imply that the scope of the Item ·nas cannot mclude the antenor clause as a whole. We mu.<.t conclude, then, th.H the attachment of ~nas to the predicate m the anterior clause in JThelf doe.<. not .11low us to rate this pred1cate as a deranked form. Other arguments which, mother languages, can be adduced to force a dectsion between deranked or embedded status of a clause also fail to apply in Ttbetan. For instance, one finds m many languages that deranked predicate, lack a number of verbal categories which 'norma!' finite main verbs have; tn parttcular, it ts common for deranked prt'dtcates to have lost the abihty to agree wtth their subJects in per~on and/or number. This critenon, however, is of no help in the case of Tibetan, smce m this language no predtcate (not even a mam predicate) ever shows any person/number agreement With tts subJect. In other words, the absence of agreement fa1ls to distmguish between dera.nkcd and non-deranked predicates m Tibetan, and, accordmgly, the fact thJt the pred1cate in the anterior clause in (24) does not ~how agreement wi1h ItS ~ubJect is of no consequence here. Luckily, however, we can trace a number of other characten~tl .. ~ which differentiate the predicate in anterior consecutive clauses from main predicate~ m TLbetan temporal chams. We can observe that, wh;le predicates in T1betan are never marked for ~ubJC<..'t agreement, they can be morphologically marked for tense; the verb for 'to drink', for example, has a Present form a-thun, a Future form d·thun, and a Perfect form b-tun. Now, one of the properties of the predicate In a Ttbetan antenor clause like the one in (24) is that It can never be marked fur tense; in this type of constructiOn, pred1cates mvariably appear m thr~Jr root form, denvmg their temporal interpretation from the tense markmg on the followmg mam predicate. From this we can conclude that, m Tibetan, predicate.<. m antcnor consecutive clause~ are at lea~t 111 some
rc '>JWd d1fft'H:nc hom m.un d.m'>t' prnhu!!:..,: they .m: ddcLtJvc 111 -.omc of rhe categoneo of verbal flexwn whiCh th1s IJ.nguagc allows. On these ground,, one may be )UStlfied m ratmg the antenor dauc,;e m (24) as an mstance of 'real' derankmg, rather than as an mstance of 'mere' dau!.eembeddmg. A!! a result, one may dasstfy Ttbetan as a derankmg language.
4.3.3 Dtscu~s1ons
The morphology of deranked predzcates
of cases ~uch as the o1bove example from T1bctan lead us to a !:lecond pom.t wh1ch ts connected wnh the nonon of derankmg, vtz. the vanety of ways 10 wh1ch deranked predicates are formally mamfe!Jted m ~mface structure, As ( stated enriler, the actual morphologiCal outcome 01 the derankmg procedure may dJVerge from one language to another. In T1betan, a der~nked pred1catc takes the form of a verbal root, wh1ch is marked further by the attachment of an advecb1al case ~uffix. In Tarrul, ~peufic gerundial forme; are u~cd to derank con\ecutJve predicates; while Engl1sh can, ~mong other thmgs, represcm a deranked consecutive p1 edrcate m the form of a Perfect Paruaple. These three fonnat1onoc; by no mea no exhaust the options for the morpholog~cal encodmg of deranktng prcd1catec;; m parr two we w1il come across a Iarge variety of deranked fonn•, many of winch are complerely d1fferellt from the ones we have merw far. No"', m our typology of chammg consrruct1ons, the morpholog~cal form wh1ch a dcranked pred1cate takes m a particular language will not bt taken to comtlture a carcgon1mg factor. That IS, I w11l not propose a dn I.S.Jon among derankcd con\trucnons on the bas1s of the d1fferences m the morphological mamfestanon of deranked predicates. In my estJm,ltJOn, the factors whKh mfluence the chmce of, say, a gerund or a parnc1ple a..; the repre'ia); they wtll only be gone mto f<.)r the purpose o£ tllustranon. Nevertheless, It may be of ~orne uoe to the reader 1f he ha\ at least ~orne general1dea of the vanou\ way.. m wh1eh derankmg of predicates can be eflectuated acrO"i<> languagt....;. To gtve ~o~ome global mdtcat10n of thts vanatwn, we can note first that, 111 a constderable number of languages, derankmg of a prediCate '" brought about by a reduction of the Hrbal morphology whtch 1~ normally allowed to predtcat~. Above, we hmted
Tvf'n
o{SI'!If(f< II< C/){lt!IIIIR
Bl
!he bet th,lt dcr,mkcd pn.·tilcatc.., .nc ohcn ch,tr.Ktnucd by the los~ ot per~onal flexwn. In other cases, such as Tibetan, we find that deranked predicates cannot take the normal tense affixe,<;., wh1le m snll other languages deranked predicates appear to be severely restncted m the choiCe of mood and aspect markmg. Moreover, It should not be thought that this reduction of verbal morphology can affect only one verbal category at a time; we find ca~es where der anked predicates have suffered a loss of both personal flexiOn and tense markmg, of both tense marking and mood markmg, and so on. A ,<;.econd general strategy m the derankmg of predicates (whiCh may m fact be a more radical applicatiOn of the reductiOmst strategy mentioned above) mvolves a change of category status of the deranked f>redtcate. We often find cases m wh1ch the deranked predJCate has lost some or all of Its verbal charactensncs, and has been turned mto a nominal form (e.g., an mfinmve or an action nom mal), an advednal form (for whiCh the tradmonal term 'gerund' will be used here), or an adjectival form (e.g., a partiCiple). Thirdly, derankmg may be effectuated by a strategy whJCh IS more or le~~ the opposite of rnorphologtcal reductiOn. In such a case, the deranked predicate is marked by the addttton of some specr(ic marker, or by the application of a special subordmate con]ugattonal form, a socalled 'dependent mood'. Lastly, we should remark that It IS qmte normal for a language to combme different strategic,<;. In the morphological codificatiOn of Its deranked predJc,nes. As the above example from Tibetan shows, the derankmg of a predicate can be effectuated by a reductiOn m tense, while at the same time mvolvmg the addition of a speofic (m this case, ablative) marker.
,It
4.4
Aspects of consecutive deranking
In the precedmg secnon, I stated that the cross-lmglllstiC vanatlon m the morphologiCal outcome of the derankmg procedure will not be regarded a:, a ::.Igmficant typological factor 111 this essay. From this, It should not be mferred that the category of derankcd constructions wtll from now on be treated as an undivided whole. While consecutive constructiOns m balancmg languages show a great deal of structural umform1ty (m that, basiCally, they are all coordmate <;tructures), comecut1ve comtructlons m derankmg languages may differ from one another m at least two respects, which are connected With structural conditiOns that can be Imposed on the derankmg procedure. As I will argue later on, the typological
A CrossMlinguistic Typology of Comparatives
84
variation whiCh lS brought about by these two structural factors IS instrumental in the explanation of (at least a part of) the typological variation of comparattves. Hence, these two factors must be rated as typologically Significant, given the explanatory purpose for wh1ch the typology of chaining constructions IS set up. In the followmg sect10ns, I w1ll comment on these two factors) and discuss the interrelation which appear~ to exist between them.
4.4.1
Condtt!O,altty
If we take a look at the various languages in which derankmg of one of the predicates 10 a consecutive cham ts permttted, we note that, for a number of languages, the applicatton of the deranking procedure appears to be restricted by spcctfic structural condittons. In particular, we find that there are languages which can de rank consecutive predicates only m cases where the two predteates m the C-chain have identical subJects; tf different subJects are involved) the construction must remain balanced. Such a situanon we encounter tn Wolof (Rambaud, 1903). In thiS language, C-chams with identtcal subjects derank their posterior predicates into a subjuncnve form, whtch ts characterized by the presence of a specific particle a and special rules for the placement of pronouns. If the subjects in the C-chain are not tdenttcal, the only option is the use of a coordinate constructwn, which may be asyndetic, but can also contain the coordinative parttcle te 'and'. Cp.: (25)
WOLOF:
a.
Demal a
0
SUB)CT. PRT call-SUB]CT htm 'Go and call him'
go-IMP
•
Nyeu on
na
te
wakh on
come PAST IND. PRT and tell
na
ma ko
PAST IND. PRT I
him
'He came and I told (It to) htm' As far as I know, there ts no tradinonaJ label for the set of construction~ (and languages) m whtch this specific identity-condition on consecutive deranking holds. In what follows, I will refer to this group of construenons as conditionally deranked consecutive constructions, and the group of languages whtch exhibit this possibiltty will be termed languages with conditional (consecutive) derankmg. In opposition to those languages m whtch the deranking procedure is subject to a structural conditiOn on subject-tdentity, we also find languages in which no such condition seems to obtain. That IS, these
Types of Syntacttc Ch.zinmg
85
languages may derank one of the predicates m a C-cham regardless of whether the predicates m the cham have tdenttcal or non-tdenttcal subJects. As a consequence, such languages can have consecutive constructwns in which both the deranked predtcate and the fimte mam predtcate have their own overt subJect. In keepmg With traditional terminology, I wtll use the term absolute constructzon for a construction type m which a deranked predtcate has Its own overt subject; hence, languages m whtch such a structural possibliity exists will be called mstances of absolute derankmg languages. An example of a language in whtch absolute consecutive deranking t'i dearly posstble ts Tamtl. Judging from sentence (26), tt appears that, 1n this language, the deranked consecutive predtcate (I.e., the genmd1al form kuduttu 'havmg gtven') and the fimite main predicJte (poon ..m 'went') may have their own separate subJect (viz. naan 'I' and avan 'he', respecttvely). Cp.: (26)
TAMIL:
Naan 1-I"'IOM
panan kuduttu avan szmmaa-vukku money-ACC gtve-GER. PI: Rl hc·!'-..OM 1110VIe -to
poonan go-PASl. IND. 3SG
'I gave (him) money and he went to a movte' Another example of this ab.!lolute derankmg of C-chams can be found m Maasa1. In this language, deranking take~ place by putting the postenor predicate in a C~cham in the so-called Dependent Mood, a subordmate verbal form whtch IS characterized by the prefix n- and the absence of tense marl.mg. As sentence (27b) shows, identity of subjects is not required m this constructiOn. Cp.:
(27) a.
MAA,Al:
E -iput-a emotz n -e -pik en-kzma she-fill ·PAST pot DEP -she-put on-fire 'She filled the pot and put It on the fire' E -zput~a emoti n -a-pik en-kzma she-fill ·PAST pot DEP-1-put on-fire 'She filled the pot and I put it on the fire'
For the sake of clanty, it should be noted explicitly that our notion of absolute deranking is defined here so as to mclude the poss1b1hty of conditional deranking. That is, languag~ whtch permtt the derankmg of predicates under non-tdentlty of subjects wlll always allow derankmg to take place m cases of subject-identity, but the converse does not hold.
A ( m'>~-lm;.:uNtc h·tmloxy of ComparatiVe'S
Altt'Jn,n•vdy, we m.ty '>t.uc .1~:. .1 umvcr~:.,ll t.lu ol n.ltur,tl l.lllgu.tgc th.lt there Jrc no Lmgu.1gc~:. 111 wh1ch dcrankmg under noJHdentlty of ~uhJcCt~ ~~ poss1blc whde dt the same tune deranking under 1denmy of subjects 1~ forb1dden. App~uently, condJtJonal dt:rankmg ts, somehow, eas1er co e.xecute than derankmg under non-Hlenmy of subjects. Hence, gtven the plauMble pnnctple that one can perform a dJfficult task only tf one has ma>;tered all the ea~1er ones, Jt Js only to be expected that the poss1bliity of absolute (t.e., non-ldcnttty) derankmg Will nece'>'>Jnly 1mply the abtltty to achieve derankmg under identity of subjects. Wnh re~pect to the concepts of condltlonal and absolute derankmg whtch are developed here, two addu10nal remarb need to be made. First, we 'ihould comment on cases of umformtty and dJfferennatton in the procedun'.!. for the~e two types of derankmg. As we stated above, laflguagcs whKh allow for absolute derankmg may derank thcu con· secunvc predtcatc!::> both m cases of tdenmy and m cases of non-1denmy of ~ub)cct<>. Now. m the above examples from Tamil (.!.ee sentences (19) and (26)) and Maasa1 (see •entences (27alb)) we can observe a parallelism m the derankmg procedure, m that the derankmg of a prediCate m a Ccham IS always effectuated by the same morpho-symacnc mechamsm, regardless of whether there IS 1dent1ty or non-ident1ty of subjects 111 the cham. For example, the Dependent Mood in Maasa1 can be used both in deranked cham; w1th 1denucal subject. (see (27a}) and for deranked predtcates whtch helVe a >;ubject that IS different from the subject of the mam predicate ('>ee (27b)). It 'ihould not be mferred, however, rh.u such a uruformJty of the der~mkmg procedure ts m:mdatory for languages with absolute derankmg. We can also find cases of absolute derankmg la:1guages m whJCh derankcd predJCates receive morpho·syntacncally dt He rent forms, dependmg on whether there 1~ Identity or non-tdennty of .!.Ubjccts m the cham. A case m pomt IS Lann. As can be seen from the sentence~ (4alb) and (14), ab!>olute derankmg 111 Latm C·chams reqmres the use of the Ablative Ab-.olute construcnon. However, m Latm C· chams With tdenucal subject!. the Ablative Absolute 1s not employed; m th1s ca~e. the anrenor prcd1c.lte receives the form of a parttctple which agree~ m case, gender and number w1th the subject of the main predicate. Cp.:
(28)
LAT/N:
Pompetus vtctus necatus est P.-NOM defeat·PCP Pf.Rl-. PASS -NOM. ~G. MA.)C k!IJec\ IS 'H:wmg been defeated, Pompems wa~ killed'
Artother case of thts dJfferentJatiOil between the procedures for absolute and condltJOnal dcrankmg can be documented m Burushask1 (Lonmcr,
'I 'yfH!S of Syntm tu (;bamm.~:
87
l!.J l ~ ). 111 tlw, l.mgu.tgt.•, we find th.u .1 {rch.un wHit u.knth..tl ,uhJn."h rcqu1rc1t the derankmg ot the anteriOr predrcate mto ,1 P.l.,t 1>.1rtrapfe (..,ce (29a)). However, when the subjects m the cham are not 1denncal, the antenor predicate must be turned Into a verbal noun, wh1ch IS then put mto the ablanvecase (see (29b)). Cp.: (29)
8URmHASKJ'
a.
Dmm dogaru·su -mt Come-PCP. PAST a;ked ·her-3SG MASC 'He came and made mqume~ about her
7
Xurts
lang
mamntt·tsrnn
11t1lmW
dust-NOM away move·VN-ABL mother-his-NOM yet -su -man saw-h•m-3SG. FEM
'The dust moved away and his mother saw h1m' As J second remark, It should be pomted out that matters are actually somewhat more comphcared than a simple bipartition mto conditional and absolute deranking ;uggem. For some languages (e.g., Lann and Mangaray1), we find that conditional deranking is not only possible if the subjects of the two predicates in the cham are idenncal; m addition, the procedure for condinonal derankmg can also be apphed If the subject of the derankcd prediCate i& tdenncal to some other, non-subJect, NP zn the other clause of the cham. Thus m addition to a case of 'real' condittonal derankmg hke (30a), m wluch there IS subject·ldenmy for borh predicates m the C-chain, Lacm also allm.vs constructton~ ltke (JOb); m thts latter con.!.truction, a predicate has been deranked condmonally on the basts of the fact that its subJect 1~ tdenncal ro rhe tndtrect obJect of the ocher clau~e. Cp.: 7
(30)
LA11N:
a.
Pompetus vtctus P.-NOM
Caesart naves
defeat-PCP. PERf. PAS;. NOM C.-DAT sh1ps·ACC
reddtdtt g1ve back·PER~.IND. 3Mi 'After Pompoms had been defeated, he gave back the ships to Caesar'
Caesar
Pornpew vzcto
C.-NOM P.·IlAT
defeat-P(.I>. PERF
naves PA~). DAT
shlps-ACC
reddtdtt g1ve back-PERF. IND. 3;G 'After Pompetus had been defeated, Caesar gave the shtp; back to h1m'
·-·
11 ..
88
A Cross-linguistic Typology of Comparatives
It must be s.tiJ, however, that this possibility of conditional deranking under Identity wtth a non-subJect IS rather marginal in natural languages. As far as I know, no language permits this type of non-subJect Identity on conditional derankmg while at the same time forbidding condittonal derankmg under subJeCt·tdentlty. On the other hand, there are numerous languages whiCh permit cond1t10nal deranking only if the Identity relatmn holds between two subJects. It must be noted that we will cons1der cases of derankmg like the Latm sentence (30b) as a (marginal) instance of conditional deranking~ anJ nor as an mstance of absolute dcranking. The reason for th1s •s that, m sentences like (30b), there IS at ledst some conditiOn of identity involved, even if this identity does not hold between two subJects. On the other hand, in cases of absolute derankmg no condttion of tdentlty holds between any con~mucnts of the two clauses m the cham.
4.4.2
Directronaltty
In addition to the typological vananon brought about by the presence or absence of subJect-Identity condtbon&, there IS a second structural factor which leJds to a typological subcategonlation of deranked consecutive constructiOns. Tins factor concerns the directzon of the derankmg procedure. That "'• we can d1vide deranked C-chatns inro two typologtcally significant clas~es, on the ba~i~ of whether it is the anterior predtcate or the posterior predtcate wh1ch ts affected by the derankmg procedure. In what follows, I wtll refer to these two options as anterior (consecutzve) derankmg and postenor (consecutrve) deranking, respectively. Now, it can be observed that, in the choice between anterior and posterior consecutive deranking, basic word order seems to play a dcterminmg role. In particular, 1t appears to be the rule that deranktng
languages wrth S0 V word order derank all anterior predtcates m a consecutive cham; that is, they derank all predicates 10 a C~chain except the last one. Thus, the unmarked case for a C-cham in a deranking SOYlanguage IS a constructton in whtch only the last predicate in the cham retains its fin1te form, while all precedmg prcdtcates recetve the form of a participle, a gerund or some orher specifically marked morphological complex. The preference of derankmg SOV~languages for anrer10r consecutive dcranking is illustrated in the examples from Latm, Tamtl and Burushaski quoted above, and in the following sentence from Korean:
l
--Types of Syntacuc Charnmg {31)
89
KOREAN:
Pulle il sigiera call-PER<. GER work force-IMP 'Call him and force h1m to work' In derankmg SVO-languages and VSO-language;, the oppo>Jtc dtrectionahcy seems to be the rule. In deranktng languages With the~e baMl word orders, the unmarked case IS that only the first predicate m a com,ecutlvc chain retains its fimte status, whtle all follo\.\.mg pred1catc~ are dcrankcd mto root form~, infimtives, verbal nouns or speafi< conjugatlo•ul forms. Clear examples of SVO-Ianguages wh1ch hav<· posterior consecutive deranking are Igbo, where- dcrJn~cd con\ecunv,· predicates ret.:ctvc a bpcoal NdCrarive Form (~ee (2)), and Wolof, where pobtcnor con~nve prcdrcates Are rn.1rkcd by the particle a and the SubjunctiVe Mood (&ee (2.5J)). A th1rd ex.unplc of the preference of deranking SVO-languages for po~tenor con~ecuttvc dcranklng IS the following sentence from B•n (Spagnolo, 1933). In thts language, de· ranked po~ter10r predtcates in C-ch.nns appear m the form of the socalled Narrative Imperative, a verbal formanon whu:h ·happens to resemble the lmpcrarive, begtnmng each phra~e u~ually warh the copula a' (Spagnolo, 1933: 13). Cp.: (32)
BAR/:
Nan a -tu mede ti nan rene ... I IND. j)A~T -go home PRT I ~weep~Il\1P 'I went home and swept (the house) and •.. ' In VSO~Ianguages, the process of con.~ecutJve dcr.Jnkmg ~~ .a b1t mo1 e elusive. Part of the dtfficulry is that, quite often, VSO-Ianguage; lack many or an CJ:tegoric~ of verbal flex1on~ so ch;:at the dt.~otmction bcmeen finite and deranked verbal form> tends to he blurred from the ouMt. Nevertheless, tn those VSO~Ianguages m whtch. dcrankcd con~t:cuw c form~ can be L~cabh~hcd, we note a dc,u preference for the derJnktng of postenor predu.:ates. Above, we rnennoned the case of Maa!IJL (.~oce (27alb)), a VSO·IJnguagc m whiCh poo;tenor pred1cate~ get mJrkcJ for 'Dependent Mood'. Another example of th~> preference for po-rert<>r derankmg IS jacaltec (Cratg, 1977), where we find that the posterwr predicate m a C~cham must be marked by the ~ubordmanng suffix ~w. Cp.: {33)
/ACALTEC:
Speba tX te pulta sah ~m tx te wentantJ closed she the door opened-PRT she the wmdow 'She closed the door and opened the w1ndow'
90
A Cross-itnKUISftc Typology
of ComJ;aratwes
In '>unun.uy, then, we t.m fornwl.lt'l' the following two o.,tatcnlcnt~ on the dJrct..tlonJ]Jty of con~ccutivc dcrankmg:
(l4) a.
if
an SOY-language deranks
1ts
C-chams,
1t
tllllVl'r'>.ll
wt/1 derank
antertor predrcates.
b. If SVO-languages or VSO-/anguages derank thetr C-chams, they wzll derank posterzor pre.dzcates. It 1s only fair to pomt out here that these two ~tatemenrs must be taken as tendencies rather than law~: they formulate the unmarked cases of directionahry. Thus, for mstance, statement (34a) 1mphes that It IS normal for a dcrankmg SOY-language to prefer anterior derankmg of consecutive prediCates, but thi~ preference r,hould not be conceived of as a law which perm1ts no dcvi<J.tiOn. In our sample, we come across at least one counterexample to statement (Ha); Kanun is a derankmg SOY~ language whKh nonethe\ec:,c; prefers po.,tcrior derankmg m C-chams. 3 Wnh statement (34b), counterexamples are even more numerous. Fm~ msh, Engll~h, Dutch and French are all cases of SVO-languages wh1ch have at least margmal po<,<,ibllltles for derankmg, and these languages all preter antenor derankmg procedures." Similar remarks can be made for Classical Arab1c and B1bhcal Hebrew; these two VSO-Ianguages have a margmal possibility to derank C-chams, and the procedure wh1ch they preter IS the derankmg of the antenor mstead of the postenor predicate. However, dcsptte these counterexamples, I hold that we are JUStified in saymg that the conclusions 1n (34), 1f taken as tendenoes, can be shown to be ia1rly well estabh>hed. I trust that the general empmcal validity of the~e t\vo tendenc1e!:. wdl become more apparent m the chapters of part rwo, where mdl\'ldual mstance~ of consecutive derankmg will be presented. Apart from the quec;non of the empmcal valid1ty of the statements tn (34), 1t ,.., also natural to a<>k whether there 1s a prmc1pled explanation for the ~tate of affatr'i whtch they descnbe. Of course, one mighr very well argue th.n the explananon of these correlattom. between derankmg types and \\'ord order types ts really beyond the scope of our mveo;uganon. It will be recaHed that the motivation for the establishment of a typology of chammg constructions hes 1n the ao;c;umptJOn that such a typology may prove to be useful as a determmant of the typology of comparatives. From rh1<> perspective, the fact that th1s 'second' typology turns out to be con elated with word order preferences may be rated as an interesting, but e!-.~ennally concomitant phenomenon. Although I agree that such a posmon I'i certamly defendable, l would nonetheless hke to offer some speculations on the reasons behmd the word order preferences of certam types of derankmg. To the extent m wh1ch these speculations are plaustble, :-..upport 1!:. g~ven to our clam1 that the word order preferences
Types of SyntactiC Chammg
91
of lVrl,mll,;omp.lr,lf!vl.' typl·:-., wh1d1 we lormul,ltcd in ".cctJon 1.1, .uc not .tccJdcntal, but can be accounted for ,mtomJtH.:,llly once we a:-.:,umc th.n
comparative-type choice ts determmed by the possJblhtles of chamingtype cho1ee. As I see it, the correlations between word order and deranking type wh1ch are latd down m (34) can be accounted for on the basis of the followmg general pnnctple:
The Forward PnnCip/e of Deranktng: If a language deranks rts consecutive ch.uns, dcrankmg w1H affect the postenor predrcate, unless there 1s some special reason dtrectwn.
to
reverse thts
Thus, thts principle cla1m5 that the dtrecttonality of derankmg wh1ch 1s opted for by SVO-languages and VSO-languages ts the unmarked case of chrecnonality. Behmd th1s prmople we may postulate a general funcnonai or perceptual strategy, according to which the decoding of a deranked predtcate ts easter when 1t IS encountered after the rnam predicate m the cham has been decoded. In this ca.!!e, the structural position and function of the deranked predicate has already been defined by the mam predicate, whereas, if this order is reversed, the structural decoding of the deranked prediCate has to be suspended until the rnam predicate has been perceived. That the Forward Princtple of Deranking is not applicable to SOY-languages can be artnbuted to the fact that in these languages it 1S mandatory to have a fimte mam verb in sentence-final position. If SOYlanguages were to follow the Forward Pnnc1ple, fimte predicates would never be able to turn up m rhe1r obligatory position, and hence there Is no other ch01ce for these languages than to reverse the direction of the derankmg procedure. For languages with SVO or VSO order, there Is no obhgation to place finite predicates m sentence~final position, whtch IS why they can fol1ow the Forward Pnnciple and have postenor consecutive derankmg.
4.4.3
The mterrelatzon of cond1t10naltty and dtrecttOnaltty
Jn the two preceding sect10ns, we have noted that the cases of deranked consecutive construction~ can be subcategonzed along either one of the followmg two parameters: (a) ConditiOnality: a deranked C-cham may be either a case of condttJOnal derankmg or a case of absolute derankmg. (b) Dtrecttonality: a deranked C-cham may be etther a case of anterior derankmg or a case of postenor derankmg.
92
A
Cross~linguisttc
Typology of Comparatives
Now, if it were the case that these two parameters were completely mdcpendent, the c.uegorization of deranked C-chains would obviously lead to four ~ubtypes. As It turns out, however, one of these logically pO.!l!!ible subtypes does not occur rn reahty, due to the fa<..'t that there ~eems to exiM an Interdependency of the two parameters in question. To be specific, there ts no counterexample m my sample to the following umversal claim: (35)
If a language has anterior deranktng (and is therefore typically SOV; see sectton 4.4.2), it has absolute deranking.
Thus, the s.1mplc does not provtdc us wtth one single instance of a language in whtch antenor consecutive predicates can be deranked only tf there b tdennty of ~ubjects between the clauses in the chain. In other words, at least one logically possible subtype of consecutive deranking (viz. a chain wnh conditrona/ anterior derankmg) does not appear tO exist. Again, the dtscussions in part two will clearly demonstrate that (35) is a hard empzrical fact. Wtth respect to cases of postenor deranking of consecutive predtcates, we can observe that the two logically possible variants are indeed represented in the data. That is, of the languages which prefer postenor deranking in C-chains some can be shown to have only conditional deranking, while other~ permit the wider option of absolute deranking. It turns out, however, that the choice for absolute or conditional derankmg 1~ not wholly unprediCtable in this case; the options on clus point appear to be firmly ned up wirh the basic word order of the languages in quesnon. For the languages in my sample, the following two statements are nearly Without exceptions: (36) a. If a language has word order is SVO, b. If a language has word order is VSO,
posterzor consecutwe deranking and zts then tt has conditional derankzng. postenor consecutzve deranking and zts then it has absolute deranking.
Wtth regard to these two umversals, I must refer once more to the data in part two, espectally those in chapter 7 and chapter 8. It will become clear there that the statements in (3 6) represent valid generalizations, to which only incidental counterexamples can be found in the sample. The statements m (34), (35) and (36) can be brought together in the following un1versal tendencies, which state the correlations between word-order type and type of consecutive deranking:
93
Types of Syntactic Chaining
(37) a. Languages w1th absolute consecuttve derankmg have etthtT SOV or VSO word order.
b. Languages with absolute antermr consecutive derankzng typtcally have SO V word order.
c. Languages wrth absolute postenOT consecutwe derankmg typtcally have VSO word order. d. Languages wh1ch permit only cond1tronal derankmg of C
it 1s jufJtifiable to
a~k for dn
explanauon of the correldttons la1d down
m
(37). I do not have the pretension that I can offer a conclus1ve answer on rhi&o point, but I would hkc to advJnce the followmg con~IdcratiOil!>. In deranking languages with SOV or YSO word order, the consecutive predicate "h1ch undergoes deranking 1s perce1ved by the hearer (and, presumably, al•o encoded by the speaker) before the second subJect m tbe chain 1s percewed. Thus, in a deranking SOY-language, wh~eh has Cchams of the followmg general form:
s,
0
v,
-
s,
0
v,
the antenor predicate (V1) must be deranked; as can be seen, th1s predicate is ordered in the chain pr10r to the second subject (5 2 ). A similar siruanon obrains m deranking VSO-languages. Agam, the pre· d1cate wh1ch undergoes deranking (1.e., the po~tenor pred1cate V2 ) IS ordered before the second sub jeerS,, Cp.:
v,
s,
o
-
v,
s~
o
In contrast to thiS Situation in SOV·language• and VSO-Ianguag<'s, derankmg SVO-Ianguages have C-chams m wh1ch the deranked pre· dicate (viz. the posterior predicate V2 ) follows the second subJectS,. Cp.:
s,
v,
0
-
s,
v,
0
In v1ew of th1s contrast, the followmg conclus1on suggests '"elf. In deranking SOY-languages and VSO-languagcs, deranking of pred1ca1es is perceived before "!entity or non-Identity of ;ubjects m a cham IS estabbshed, while in derankingSYO-languages 1dent1ty conditions on the subjects in the chain are known at the time that the deranked predicate" perceived. From this, one might venture the hypothesis that derank1ng languages With SOY or VSO order must ha\e the freedom to derank th•'lt consecutive preJICates under any conditions, wherea.!> decankmg SVO-
A Cross-lmgurstrc Typology of Comparatives
94
l.mp,,lgt·'> ~,.In h11111 thnnwh('-. to tlu: 'c,l\ll'f' pH~t.. t•dmc ol uHH.IttiOlhtl dcr.mkmg. It goes Without ~aymg, however, that tb1~ explanatu.m of the fact• on (37) 1s rather a temanve one, and that much p&ychohnguistic
research mto the process of derankmg will be needed before we can tell wh<>thcr there is any plausibility to an analysiS of th1s kind.
4.4.4 Subtypes of consecutr"e derankmg If we accept the vahdny of the umversals presented Ill sect10n 4.4.3, we are m a posltton to reduce the subcategorlzanon of de ranked consecuttve construcnons to a typology whtch contams three baste n~btypes. These three typologiCal opnons form dusten of properties wh1ch are connected wrth condJtionahty, directionality and basic word order, and can be
defined as follows: (a) consecutive chams with absolute deranking of the antertor predrcate. Typically, languages with th1s option have SOV ward
order. •b) consecutive chains with absolute deranking of the posterror predtcate. Typ1cally, languages With this option have VSO word
order. ~c)
consecutive cha1ns w1th conditional deranking of the posterior
(>redicate. Typically, languages with this option have SVO word order. 4.5
The fonnal
exprcs~ion
of simultaneous chaining
In the precedmg section!!- of th1s chapter our mam concern has been with the way in wh1ch the semantiC concept of consecutive chammg 1s caddied m the ~yntact1c surface \trUcturt.'~ of natural languages. I wtll now make a few remarks on rhe codificatiOn of the other vanant of temporal chammg, v1z. stmultaneous chazmng. Our discussion of thts subtype of temporal chnmmg can be relattvely brief, smce a number of notions
v,.h1ch are necessary 1n the descripnon of the various types of simultaneous chams have been dealt With extensively dunng our survey of the subtypes of consecutive cha1ns. As far as the semantic, language-Independent defimtion of the nonon of 1 Simultaneous chaming' 1s concerned, maners are relauvely stratght-
f<Jrward. We can define a simultaneous cham (ot S-cham) crosshngmsncally as that construction type which expresses a situauon in wh1ch two events happen at the same pomt m tzme, or a situation in whtch two states obtam at the same potnt m ttme. Included m thiS
Tyf't'-" uf Syntattu Clwmmg
tldimuon .Ire .tl\<) ~nu.uuuh tn whh.h two cr'cnts or stat('S nverlap m tmw. Thu~~ lnlhc lypu...tl \..•lM.' ~-tho1111" \OIIl"'Jlo!ld 111 lllt'.Hlmp.to tilP'>(' Engh~h
adverbtal dau~c~ which are mtroduced by the conjunctwm. (ilt the ttme) when or uohrlc. A ~omewhat spec1al ca~e of 'l1multaneous chams is formed by those chams m wh1ch the two events or states are set off
agamst each other in a relation of adversatw1ty. Roughly speaking, in this ca~J~e
the two events or states are sa1d to happen c.q. obtam at the same time desp1te expectations to the contrary. A typ1cal cod1ficanon of such an adversative chain in English I& a coordinated structure m whtch cwo
sentences are hnked by the connecnve but. The syntacnc encodmg of Simultaneous chammg parallels the encodmg of consecutive chammg m a number of Important respects. For one thmg, JUSt hke C-chains, S-chams may be expressed euher by a balanced or by a deranked structural configuratiOn. That ts to say, m some languages the simultaneous cham has the shape of a syntactic coordmanon between clauses or sentence& c.q. the shape of a matn dause wtth an embedded
temporal clause, whereas m other languages one of the two predicates m the S-cham is morphologically marked for subordmanon. An example oi a language in wh1ch S-chams have to be balanced IS Pala: (38)
PAl-A:
Da e1tum ra tamat ma da so ra hanszk one·w11l eat the b1g and one-w1ll plant the small 'The b1g (yams) will be eaten and the small ones wtll be planted'
Tibetan, on the other hand, IS a language m wh1ch Simultaneous chams are obligatorily deranked, in the manner shown by example (39): (39)
TIBETAN:
Lam nan-pa -ym-las rta agrul-ma ·thub road bad-one-be -PRT horse walk -not-able 'The road was bad and the horses could not walk' A second parallelism between the cod1ficanon of simultaneous and
consecutive chains hes in the fact that deranked structures of both types of temporal chaining are subject to the same kmd of cond1ttonaltty. In the same manner as we observed for de ranked C-chain>, deranked S-chams can be subdivided into structures m wh1ch absolute derankmg IS pos<~ble (that IS, in which the derankmg procedure operates mdependently of the 1dennty or non-idennty of the subJ~Cts in the chain), and structures m
wb1ch only cond1tional derankmg IS possible (that IS, deranking under identity of subjects). An example of absolute Simultaneous deranlang IS provided by the above sentence from Tibetan (see (39)). Condltlonal stmultaneous derankmg seems to be the rule in lgbo; the deranked
---~-
70
A Cross-linguistic Typology of Comparatives
Simultaneous construction m this language, which we illustrated m example (1), is only allowed in cases m whtch there is subject-identtty between the prcdtc.ues m the chain (see Welmers, 1973}. With re&pect to drrectiorzality, the other condition on the derankmg procedure in consecutive chams, there IS, however, a fundamental d1fference between the two types of temporal chammg. In consecutive chains, the propo&itions which express the events involved are temporally ordered. As we saw, this temporal ordenng has its formal-syntactic counterpart tn the left-to-right ordermg of predicates (or sentences) in the con!:tecutive ,<,tructure, to the effect that the predicate or sentence whtch describes the e.1rlier event m the temporal succession ts ordered structurally as the leftmost predicate or sentence in the chain. We noted in sectton 4.4.2 that thts (temporal and structural) ordering of consecutive predtcates plays an tmportant role in the procedure of consecutive derankmg: It may either be the temporally anterior (and structurally leftmost) predicate or the temporally posterior (and structurally nghtmost) pre· dtc.uc whtch t!-1 affected by the deranking procedure. As we have shown, the choice between these two altemattves ts at least m part mfluenced by the baste word order which the language in question has. Now, tn the case of simultaneous action there is of course no successive temporal ordenng between the two event~ m the chain: both of these events arc supposed to take place at the same point in time. This essential lack or neutrality of temporal ordering m a simultaneous chain is mirrored m semantics by the fact that, m chams wtth a simultaneous readmg, both posstble orderings of the two event-expressing propositions are expression,!, of one and the same state of affairs. In simultaneous chams, the choice for one of the possible orderings in a given situation is not made on semantic grounds (as is the case in consecutive chains), but ts presumably influenced by con,!,iderations of a pragmatic or functional nature. Thus, the English sentences tn (40), when taken in their natural simultaneous interpretation, are semantically equivalent. In contrast to this, the sentence~ in (41), which are naturally interpreted as consecutive cham!-., arc not !.Cmant~eallr equivalent, smce they do not satisfy the same truth conditions: (40)
ENGLISH:
a. b.
john wore a cap and sported sneakers john sported sneakers and wore a cap
(41)
ENGLISH:
a. b.
Gary bought a gun and rode into town Gary rode mto town and bought a gun
Types of Syntactic Charnmg
97
A structural counterpart of this es~nttally unordered nature of Simultaneous chaming IS the fact that, m languag~ wh1ch derank thetr 5-chams,
both of the eveot-expn:ssing predicates m the structure may be candtdates for deranking. For mstance, tn Enghsh, a language whtch optionally deranks its SJmulraneous predtcates, the two sentence> tn (43) are both possible deranked variants of the balanced stmultancous structure (42): (42)
ENGLISH:
John stood on the amter and smoked a cigarette (43)
ENGLISH:
a. b.
Standrng on the corner, john smoked a ctgarette Smoking a ctgarette, John stood 011 the corner
Now, from a structural point of view, simultaneous cha1ns are of course entities which are deployed m nme; therefore, they have a hnear order.
just hke any other lingutstic ennty. Hence, it will alway> be posstble t<> mack pred1cates a!. leftmost or rightmost, even m 5-chams. Som1· languages which derank thetr S-chain~ may have a preference for th{· deranking of a leftmo>t or a rtghnnost predicate in such chams. Ttbetan, for example, obligatorily deranks leftmost prediCates m Mmult ..tneou<; constructtons (see (39)); this preference can be .tttrlbuted to the fact that Tibetan, a strict SOV~language, alway!t rcqutres a finite mam verb m sentence~fi.nal
posation. Despite these structurally mouvated preferences
for direction, however, the essential point remams that,. even m Tibetan, there W()uld be no change m truth value if one were to mvert the clausts in (39) linearly and in terms of der•nking. In other words, the tenrs 'anterior' and 'posteriOr' have no applicatton m the docnpoon uf simultaneou~
chaining constructions. As a consequence of thb feature of simukancou::. chaimng~ we can note that, unlike consecuttve derankmg, stmultaneom.. derankmg does not seem to be subject to conditions whicb are connected wtth baste word order. If a language deranks its $-chains, word order does not come imo play, and therefore no subcategonzanon of srmulrant..-ou~ der.mkmg types can be made on dus pomt. In ocher word~, whdc derankt:d consecutive structure!:. can be subcategonzed mto structures wtth antefliu deranking and posterior deranking, ~tmultattL'"<>US der.mking does nJt
have directional subtypes, and can only be pttted as a whole agamst the alternative of simultaneous balancing. As a result, we can summanze the structUral types of simultaneous chams in the followmg three categones: ca;es of simultaneous balancing (b) cases of condttional simulta11eous derankmg (c) cases of absolute Simultaneous deranking. (a)
4.6
Summary: syntactic types in temporal chaining
In the prccedmg section~ of this chapter, it has been my atm to present a typology of the ways 111 whtch the semantic or cogmttve notion of temporal chammg IS ~yntacncaiiy encoded m natural languages. Our maJor re&ult~ can be summan1ed as follows. FJ"St, the semantic concept of temporal chammg appears to cover two subtvpes, vtz.. stmultancou& actiOn and con!:>ecunve action. For at lea!:>t a number of languages, the&e two &ubtypes are structurally represented by dtfferent surface constructtom; m other languages the semantic chsnncnon between these subtype'> has been formally neutralized m a total or partial manner.
Secondly, for both of these semanttc subtype; a pnmaty spilt in thm possJbdJttes of &yntactJC codificatiOn mvolves the d1stmcnon between the structural opt'IOO'i of halancmg and deranking. Tlurdly, wlthm the category of deranked chammg constructiOns, both consecutive and ~trnultaneous cham~ can be !:>ubcategonzcd ac; instances of either conditional or ab~olute derankmg. Fourthly, m the case of drranked consecunve construcnons, the ~pht ben."' een condtttonal and abc;olute derankmg mtera.ct!:> w1th the d1stmct1on benveen antenor and poc;tcnor derankmg; this latter dJ!:.tmctlOn correlate!:> w1th the basic word order of the languages at issue. The mteractwn of the two di!:>tmctiOn<> produce<; a three-way dtvl!:.ion w1thm the cate':!Q,TV pf de-ranked consecutive chauun~sonsrructlQns. __ _ In ~urn, the typology of the syntactic vanat1on in the cod1ficatwn of temporal chammg consJ'it\ of the foHowmg seven type!:.: A 1 A 2.1 A 2.2 A 2.J B1 B 2.1
B 2.2
consecutive halancmg consecutive conditional (po,tcrior) dcrankmg (typically SVO) absolute (consecutive) antcnor dcranking (typically SOY) abwlute (consecutive) postenor derankmg (typically VSO) ')Jmultaneous balancmg conditional ~Imultaneou!:> derankmg ab~olute Mmu1taneou!:> derankmg
4.7
Language types in temporal chaining
In the prevwus sectwn, we have summanzed the syntactic forms m wh1ch nat mal languages may encode the semantiC concepts of Simultaneous and con,ecutrve actwn. Thts typology of chammg constructiOns can now be
'/)'/It'S
11/ Synla1 //(
u~ed as the baM~ of a typology
( 'lhlmlll,'!
99
of languuges m the expressiOn of temfmrt~l
we w1H now mvesngate the way~ m whJCh languages make their chmce from ch1s }Jst of poss1hle encodings, and, as a result, we wtll e5.tabhsh a dasSificanon of che languages tn the sample on the basts of their options in the formah7atJon of conc;ecunve and Simultaneous • acuon. I assume Jt will be obvwus that any natural language will have to have the possibthty to cod1fy both simultaneous and consecutive acnon, and that, therefore, any language must make a cho1ce from both the posstble types of simultaneous chamt~ and the possible types of consecutJve chams listed m ~ectton 4.6. Now, g1ven that there are four poss1ble ways to codify consecutive actiOn and three poss1ble ways to cod1fy simultaneous action, 1t follows that a typology of language classes m chainmg
chammg. That
lS,
formanon will theoretically conSISt of 12 dJfferent categories of languages. In reality, however, thts number of language types turns out to be ngorously restricted, due to the opera non of a general prmctple whtch regulates the selection of consecutive and smmltaneous chammg types for a given language. Tlu~ prmCJplc can be formulated as follows:
The l'rmciple of Parallel Chammg: In Its codlficanon of temporal chammg, a language will select parallel optiOns for consecutive and simultaneous chams.
It ts, of course, natural to conceive of thts pnnctple as a kmd of economy pnnc1ple operating in the systems of natural languages; basically, the pnnctple is a mamfestanon of the general tendency m rule ~ystems not to comphcate procedure(> beyond what 1s stnctly necc&sary, and thus to employ similar procedures for sim1lar tasks. As we noted m section 4.1, m some languages the Princ1ple of Parallel Chammg ha~ led to the
The Principle of Parallel Chammg can be seen to operate both m the b,J!ancmg-derankmg dtstmctlon m cham formation and 1n the absoluteconditional dl:.tmctJOn m derankcd temporal chams. To start With the fir~t d1stmcnon, we can note that a language whiCh balances Its C-chains will tend to choose the balancmg optwn for Its S-chams, and v1ce versa. Conversely, tf a language deranks Its C~chains, 1t will, as a rule, also de rank lts S-chains, and vice versa. In my sample, there are only very few counterexamples to thts general tendency. One conspJCuous case m wh1ch th1s para1lehsm is absent IS Nuer, a NtlotlC VSO-language, whtch balances its Simultaneous chains, but has the pmsththty of absolute derankmg for its consecutive pred1cates. The same ~uuat1on holds for
A Cross-lznguistic Typology of Comparatives
100
l\l>asai, another Nilotic VSO-Ianguage. Opposed to this, two Australian languages (viz. Gumbamggir and Mangarayt) and the two Polynesian languages in the sample (MJ.on and Samoan) have the possibility to
derank predicate~ m ~unultaneous cham~, but, as far as I know, they lack a dcranking optmn for consecutive chams. However, the general tendenq m my sample IS cle:1rly that language~ wtll employ the same basic strategy m both the formahLJuon of stmultancous action and the formalization of •
consecutive actiOn.
The strw.:tural parallebsm between consecutive and encoJmg can al:)o be observed
10
simultaneou~
rhe kmd of condttionality under wh1ch
the two types of chJtnmg constructions can be deranked in a language, gtven that they can be deranked at aU. In general, it appears to be the case that, if a language ha<; absolute consecut1ve derankmg, it wtll also have absolute simultaneous derankmg, and VICC versa; as the examples above dearly demonstrate, Ttbetan ts a case m point here. Conversely, if a language has only the opnon of conditional consecutive deranking, the same type of restrtctlon tends to be Imposed on the deranking of simultaneous construction~, and vtce ver<;a; lgbo (see sentences (1) and (2)) IS a language in which thiS Situation obtains. A possible counterexample here JS Quechua, a language which seems to have absolute consecutive derankmg, but only conditional stmultaneous deranking (see chapter 10). However, we can state wtth some confidence that the structural parallelism between the types of C~chains and S-chains whtch a derankmg language selects is confirmed by the large majority of the relevant languages m the sample. If we accept the general validity of the Principle of Parallel Chaining, we are in a position to limit the set of language types in chainmg formation to the followmg three categones: (a) Balancing languages. These languages have both balanced Simultaneous construcr1on5 and balanced consecutive constructions; that IS, m these languages the form of S-chams and C-chains is that of a coordination. Since~ typically, coordmation is a umform procedure, it will be expected that neutralization of consecutive and stmultaneous chaining will be very common in the languages of thts type; disambtguization of the chatning construction will typiCally be effectuated by the optional use of temporal adverbs wtth cnher a definite consecutive meaning ('and then\ 'and thereafter') or a defimte simultaneous meaning ("and also', 'and at the same ttme'). Furthermore, tt wtll be expected that the languages in thts class do not have a preference for a particular type of basic word order, smce the choice for a coordmated chaming construction is not many way dependent on the type of word order which a language happens to have. (b) Conditronally deranking languages. The languages of this type
Types of Syntactic Chaming
101
have S-chains and C-chains m which one of the predicates is deranked under Jdentlt}' of subJects; as we have seen m secuon 4.4.3,
It JS
the
postenor predtcate whtch i• deranked m C-chams of rhis type. Smeo' conditional derankmg of C-chains •• typ~<:.dl} • property of derankmg SVO-languages, 1t wtl! be expec-red that l•nguagcs of th~> c