Semantic Universals and Universal Semantics 9783110870527, 9783110133912

Citation preview

Semantic Universals and Universal Semantics

Groningen-Amsterdam Studies in Semantics (GRASS) T h i s s e r i e s of b o o k s o n t h e s e m a n t i c s of n a t u r a l l a n g u a g e c o n t a i n s c o l l e c t i o n s of o r i g i n a l r e s e a r c h o n selected topics as w e l l as m o n o g r a p h s in t h i s area. Contributions f r o m linguists, philosophers, logicians, computer-scientists and cognitive psychologists are brought together to promote interdisciplinary and international research.

Editors

Johan van Benthem

Alice ter M e u l e n

University

of

Martin Stokhof

Henk Verkuyl

Editorial

Co V e t

University Board

Renate Bartsch

University

of

University

of of

Amsterdam Utrecht Groningen

Amsterdam

O t h e r books in t h i s series: 1. A l i c e G.B. ter M e u l e n (ed.) Studies in Mode/theoretic Semantics 2. J e r o e n G r o e n e n d i j k , T h e o M.V. J a n s s e n a n d M a r t i n Stokhof (eds.) Truth, Interpretation and Information 3. Fred L a n d m a n a n d Frank V e l t m a n (eds.) Varieties of Formal Semantics 4. J o h a n van B e n t h e m a n d A l i c e G.B. ter M e u l e n (eds.) Generalized Quantifiers in Natural Languages 5. V i n c e n z o Lo Cascio a n d Co Vet (eds.) Temporal Structure in Sentence and Discourse 6. Fred L a n d m a n Towards a Theory of Information 7. J e r o e n G r o e n e n d i j k , Dick de J o n g h , M a r t i n Stokhof (eds.) Foundations of Pragmatics and Lexical Semantics 8. J e r o e n G r o e n e n d i j k , Dick de J o n g h , M a r t i n Stokhof (eds.) Studies in Discourse Representations Theory 9. M i c h a e l M o o r t g a t Categorial Investigations 10. Irena Bellert Feature System for Quantification Structures in Natural Languages 11. R. Bartsch, J. van B e n t h e m and P. v a n Emde Boas (eds.) Semantics and Contextual Expression

D. Zaefferer (ed.)

Semantic Universals and Universal Semantics

FORIS PUBLICATIONS Berlin • New York 1991

Foris Publications Berlin • New York (formerly Foris Publications, Dordrecht) is a Division of Walter de Gruyter & Co., Berlin.

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

Library of Congress Cataloging in Publication Data Semantic universal and universal semantics / D. Zaefferer (ed.). p. cm. - (Groningen-Amsterdam studies in semantics : 12) Includes bibliographical references and index. ISBN 3-11-013391-1 (acid-free paper) I. Semantics 2. Semantics (Philosophy) 3. Universals (Linguistics) I. Zaefferer, Dietmar. II. Series. P325.S374 1991 91-33618 401' ,43-dc20 CIP Die Deutsche Bibliothek Cataloging in Publication Data Semantic universals and universal semantics / D. Zaefferer (ed.). - Berlin ; New York : Foris Publ., 1991 (Groningen-Amsterdam studies in semantics ; 12) ISBN 3-11-013391-1 NE: Zaefferer, Dietmar [Hrsg.]; GT ® Copyright 1991 by Walter de Gruyter & Co., D-1000 Berlin 30 All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Printing: ICG Printing, Dordrecht Printed in The Netherlands.

Contents

Dietmar ZaefFerer

Preface Introduction: Universals and Semantics

vii 1

Part I: General Questions Johan van Benthem Linguistic Universals in Logical Semantics Manfred Immler

Johannes Bechert

17

Is Semantics Universal, or Isn't It? On the Relation of Language, Thought and Semantic Structure

37

The Problem of Semantic Incomparability

60

Part II: Basic Issues Hans-Jürgen Sasse David Gil

Predication and Sentence Constitution in Universal Perspective Aristotle Goes to Arizona, And Finds a Language without 'And'

75 96

Part III: Special Topics Godehard Link

Quantity and Number

133

Manfred Krifka Ekkehard König Dietmar ZaefFerer

Some Remarks on Polarity Items Concessive Relations as the Dual of Causal Relations Conditionals and Unconditional: Cross-linguistic and Logical Aspects

ISO 190 210

List of Authors and Addresses Index of Names Index of Subjects

237 238 241

Preface When I first came to Stanford as a postdoc and visiting scholar to leam about Situation Semantics from Barwise and his associates and about linguistic universals from Greenberg and his students, I was initially struck to see that there was virtually no contact between the two internationally famous schools of thought on language. But a moment's reflexion reminded me that this situation was quite representative: There is a general tendency for linguists working on linguistic universals and for logicians working on universal grammar to ignore each other. It seemed obvious to me that this situation, due to the historical development of the traditions involved, is rather unfortunate and should be changed. And it is surely no accident that one of the colleagues who I heard express most pronouncedly the same opinion, David Gil, is a student of the man who represents the single most outstanding exception to the rule just mentionend: Edward Keenan, renowned both as a formal semanticist and as a typologist. Although he could not contribute to this volume, I think that it is written in his spirit. But it would not have been realized without the encouragement by Alice ter Meulen. Follwing a hint of Johan van Benthem's, I told her about my idea of bringing together representatives of the two groups involved in a workshop and she spontaneously offered the GRASS series as an adequate forum for the proceedings. This was in July 1987 on the patio of the Stanford Tresidder Union 'Coffee House'. The workshop took place seven months later as part of the annual meeting of the 'Deutsche Gesellschaft fur Sprachwissenschaft' (German Linguistic Society) from March 2 through March 4, 1988 in Wuppertal. The final discussion on 'Plausiblity and problems of hypothesizing semantic universals' showed on the one hand that the gap between logical and typological research on universal semantics is still rather wide (some contributions had a perceivably polemic undertone), but it also showed that more and more scholars are interested in bridging it (some contributions expressed curiosity for what the other side considers interesting). So the meeting presented a quite representative picture of the current situation between the two parties involved. There is only a partial overlap between the eleven talks given at that workshop and the nine contributions to the present volume: Johan van Benthem, Manfred Bierwisch, Uwe Monnich, and Sebastian Lobner had already committed their presentations for publication elsewhere, Volker Heeschen's and Wolfgang Raible's talks were too far away from the logical side of the subject to be included here. So I had to look for contributions to complement the five core papers by Sasse, Gil, Link, and Krifka. They were not hard to come by since Johan van Benthem wrote another paper especially for this volume, Johannes Bechert had prepared a contribution without reading it due to an illness, I had written a paper which had not been read for lack of time, and Manfred Immler, who had

viii

PREFACE

attended the workshop without presenting a paper, volunteered to submit an additional contribution discussing a central question of the topic at hand. It took considerably more time than anticipated first to get the final versions of all the papers in and then to get the straightening of the layout done and I apologize for any inconveniences the delay may have caused to those who were awaiting the volume eagerly, but I am confident that it is not outdated in any respect and that it will contribute in a way that does make a difference to the growth of a new cooperation between logical and typological work on human language. I have mentioned already Johan van Benthem's and especially Alice ter Meulen's role in the origin of this book, but I owe additional thanks to Alice for patiently accompanying all stages of its progress and above all for her editorial suggestions, which even took stylistic details into account. To Gabriele Hollmann I am indebted for helping with my English. Last, but not least, I want to thank for invaluable support my wife Guadalupe and my sons Sirtan and Manuel, who tolerated with empathy my partial absence during so many weekends. Dietmar Zaefferer, Munich, March 1991

Introduction: Universals and Semantics Dietmar Zaefferer Institutfür Deutsche Philologie Universität München Schellingstr. 3 D-8000 München 40 Semantic universals are the properties the semantics of all languages have in common. Universal semantics is that part of semantic theory which is concerned with general semantic properties of language (singular) as opposed to the specific semantic properties of particular languages (plural). In other words, universal semantics is about semantic universals. But universal semantics is a term which covers quite diverse kinds of scientific study. This diversity can at least partly be traced back to different ways the common subject matter, semantic universals, is conceived of. So it will be useful to begin by first clarifying the different readings of this latter notion or, less specifically, the general notion of linguistic universals.

1. Kinds of linguistic universals Linguistic universals are simply properties shared by all languages. So they naturally fall into two classes: those shared by virtue of the notion of language, the so-called analytical universals, and all the rest, the synthetical ones. Sometimes statements saying that some property is shared by all languages are also called linguistic universals, but we will keep them apart by calling them universal statements or universal hypotheses. So far, the notion seems to be rather clear. But there are two sources of complication hidden in its specification: First, what notion of language does it presuppose? And second, what does the quantification 'all languages' range over? To begin with the second issue, there are basically two choices. First, we can choose to consider only observable languages, i.e., languages that either are still alive or at least reconstructable. In that case, the synthetical universals are just the empirical ones. Empirical universals in this sense may be accidental, i.e., hypotheses about them may be falsified by the discovery or emergence of systems that fail to have the properties in question, although they are languages by all general standards. Suppose, for instance, that the Khoisan and the neighboring Bantu languages haven't been discovered yet. Then the absence of clicks from the phoneme inventories would be an empirical universal, which would disappear at the moment where these languages are discovered, since they do have click consonants in their phoneme inventories.

2

DIETMAR ZAEFFERER

The other possibility is to include in the set of languages also those that are not reconstructable anymore and those that still may come into existence. In this case, the non-analytical universals cannot be identified with the empirical ones anymore, since what can be observed may be accidental, and the question comes up whether there is anything left. Stating universals that hold not only for observable but also for possible languages, doesn't this amount to defining what a possible language is, in other words to stating analytical universals? But then the empirical universals are just the accidental ones and we are left with a dichotomy of accidental versus analytical universals which would make the whole research on universals rather uninteresting, since accidental universals may be falsified by the next language we come across and analytical universals can be read off the very notion of language. So there must be something wrong with the simple picture we have drawn so far. The interesting universals are those that are neither analytical nor accidental, but something in between, namely those properties which are due to certain constraints that hold systematically without being part of the very notion in question. For example, suppose counterfactually that all American bachelors have a phone number containing the digit one, and that in America everybody prefers what he lacks over what he has. Under these circumstances, being unmarried would be an analytical universal property of American bachelors, having a phone number with the digit one would be an accidental universal property of them, and preferring to be married would be a third universal property of them, which would be neither analytical nor accidental, but rather induced by the fact that American bachelors are American people and that Amercan people prefer what they lack over what they have. So the most interesting universals are the non-accidental non-analytical ones, which may be called induced universals. They are induced by background regularities that again are neither accidental nor analytical. Induced universals are not accidental in that counterexamples, although logically possible, are empirically utterly improbable, since they would have to escape the relevant constraints. The introduction of the induced universals has been achieved by complementing the original analytical-synthetical dichotomy with the accidental-systematical dichotomy. Induced universals may now be equally well characterized as the non-analytical systematical or as the non-accidental synthetical ones. Induced universals may further be subdivided according to the kind of constraint they are induced by. For instance, biologically induced universals are properties of language which go back to the genetic endowment of humans. One extreme, but influential assumption, held by Noam Chomsky and his followers, is that all induced universals are biologically induced (Chomsky 1968: 76). Others assume that the interaction of different kinds of constraints is the source of induced universals. The plausibility of these

INTRODUCTION: UNIVERSALS AND SEMANTICS

3

positions will not be discussed here, since this book deals with the universals themselves and not with their possible origins.

2. Kinds of universal semantics There are at least two major ways of doing universal semantics in research on language. One is mainly pursued by data-oriented linguists, who browse through the reports of field workers, grammars, and typological surveys and care about sample size before they generalize. The other one is mainly followed by philosophers of language, languageoriented logicians, and philosophically interested linguists who prefer to carry out indepth investigations into particular languages and to generalize from there.1 The former, the breadth-oriented extensive approach, is primarily concerned with synthetical universals in general and tends to postpone discussions of the accidental-systematical distinction. The latter, the depth-oriented intensive approach, is only interested in fy

systematical universals and avoids discussing the analytical-synthetical distinction. The central question in this latter distinction is: How strong a notion of language do we assume? Obviously, if we stick to a rather weak notion, many more general phenomena turn out to be synthetical universals than if we assume a rather strong and therefore narrower notion. Defining a language as any system of signs, i.e., of correlations of observable forms with inferrable contents, identifies linguistics with semiotics and leads to a notion of language that is too weak. In order to adequately strengthen the notion, we need at least three more ingredients. First, a sign system that is to be a language in our sense should be a general-purpose system serving all general encoding and communication needs as opposed to special-purpose systems like the 'language' of traffic lights. Second, it should be used or capable of being used by humans beings as opposed to animals or machines. Third, it should have developed (or possess the potential to develop) and stay alive naturally, i.e., mainly as a self-organizing system, with at most marginal external planning and control. So the notion of language we are interested in can be defined as a natural human general-purpose sign system. But how about the arbitraryness or conventionality of most linguistic signs? Shouldn't we include it as well in our notion of language? This seems to be dispensable. Only limited purpose human languages can exclusively get along with natural, iconic signs, because general human encoding and communication needs are simply too vast The extent of human encoding and communication needs is a non-accidental synthetical fact and therefore the arbitrariness of most linguistic signs is an induced universal.

4

DIETMAR ZAEFFERER

The same holds for what van Benthem (this volume) calls their computability, the fact that they can be processed by humans with normal cognitive capacities, and similarly for what Barwise and Perry (1983) call their efficiency, their potential for serving different purposes in different contexts, since the fact that they can be used by human beings with limited resources involves that they are learnable in a quite restricted amout of time and that their use has a reasonable cost-benefit ratio. Our definition of language would also classify as a synthetical universal the fact that human natural general-purpose languages are primarily systems of acoustic and not of visible signs and it would imply that such languages could have evolved or could evolve as visual systems that only later, if at all ,are correlated with acoustic representations, should there be no constraints precluding this possiblity. A richer notion would include the orality feature and exclude purely or primarily visual sign systems from its extension. So the exceptional nature of, e.g., sign languages, for example those developed for the deaf, would be an analytical rather than a synthetical feature. For our purposes we do not have to decide which of these two notions is preferable because we are concerned with semantic universals, and having a semantics is an analytical universal since (a) languages are systems of signs, (b) signs correlate sign-vehicles, perceivable forms, with inferrable sign-contents, information units or concepts, and (c) to have a semantics means to include such a correlation.

3. Examples of semantic universals If we leave with these remarks the general universals for the specifically semantic ones, the next question that comes up is the following: Is this analytical universal, that all languages have a semantics, the only semantic universal, or are there other universals that are more interesting and that, in the analytical case, are useful building blocks for a fruitful specification of our central notion, or that, in the induced case, have a fair chance of surviving attempts to falsify them? It might seem a plausible proposal for an analytical semantic universal to require that the contents of linguistic signs be humanly graspable concepts, i.e., possible contents of human cognition. The plausibility of this idea vanishes however if one takes into consideration that linguistic sign-vehicles in our sense, i.e., elements of a general-purpose language, can be of arbitrary (within certain limits3) complexity, which means, given the rough, but certainly positive correlation of sign-vehicle size with sign-content complexity, that linguistically encodable concepts can be of arbitrary complexity too. But certainly human cognition is not ready to grasp arbitrarily complex concepts. So this universal

INTRODUCTION: UNIVERSALS AND SEMANTICS

5

would better be conceived of as an induced universal that follows from the analytical ones plus the trivial constraint mentioned above, namely that humans can only use concepts within the range of their cognitive capacities. By contrast, let us now consider an example of a universal semantic property that clearly could only be called synthetical right from the outset. Since semantics is concerned with the correlation between form and content of signs, the most simple kind of synthetical semantic universal would state that certain linguistic signs occur in all languages. This presupposes that there are phonological forms which occur in all languages of the world. Since phoneme inventories differ considerably (cf. Maddieson 1984), and so do phonotactic constraints, this is a problematic assumption. Yet, some simple syllables like ka and pa seem to pertain to all languages; therefore it is conceivable that there are also universally shared meanings associated with them. But there aren't any. What comes closest to this very concrete kind of semantic universal is the so-called global etymologies, families of related forms, paired with related meanings which can be reconstructed in all language families of the world (cf. Bengtson and Ruhlen to appear). But at present, the status of these etymologies must rather be called tentative, and so we are left without a good example for our first, most straightforward kind of semantic universal. So we have to look for more abstract types of universal properties, such as meanings universally encoded not by expressions with a given phonological form, but with given syntactic properties, i.e., belonging to a certain syntactic category. This is the kind of universal Johan van Benthem (this volume) is interested in, who defines semantic universals as general laws about meanings expressible in all human languages. The main question is: How are they expressed? If all languages have, in principle, the same general-purpose expressive power, as seems reasonable to assume, the differences must lie in the kind, especially in the size of the linguistic sign needed to encode a given meaning. One language may possess a single word to express a meaning another language would need a whole story to encode. It has been suggested (e.g., in Zaefferer 1990) that the hierarchy of coding (1) reflects some hierarchy of basicness in the notional structure of a given language: The lower a means of encoding in the hierarchy, the more fundamental the encoded concept. (1)

grammatical (unmarked) < grammatical (marked) < lexical (root) < lexical (stem) < lexical (compound) < phrasal

Singularity, for instance, is in most languages a more basic concept than plurality, since, although both concepts may be encoded by grammatical means like inflection, plurality tends to be more marked.4 Likewise, in English, four (root), fourty (stem), four hundred

6

DIETMAR ZAEFFERER

(compound), almost four (phrase), in this order, express notions which are decreasingly basic. This hierarchy is in line with a general tendency of coding economy correlating the means of expression with the meanings expressed: The more basic a concept in a system of linguistically encoded concepts, the lower the cost of encoding it. One notion may be more basic in one linguistic community than in another one, but there are many areas of tendential conformity. For instance, the cost of encoding the notion of the speaker of an utterance doesn't seem to exceed in any language the size of one rather short word. A more specific example of this type of semantic universal, which is much 'harder' since it is not statistical and which is also discussed by van Benthem (this volume), is Keenan and Stavi's (1986: 260) 'Extensional determiners in all languages are always interpreted by conservative functions.' Extensional determiners denote functions from properties (sets) to sets of properties, and for a function/to be conservative means that the value for any set P in the domain o f / i s a subset of the powerset of P, in other words, values are built from the arguments, without going beyond. It would seem at first glance equally plausible to define extensional determiners as those interpreted by conservative functions and to state as a universal hypothesis that all natural languages have expressions which are determiners in that sense, as Barwise and Cooper (1981: 179) had done. However, by weakening the notion of determiner, Keenan and Stavi have managed to turn a statement that in the Barwise/Cooper framework would express an analytical semantic universal into a synthetical universal which probably is not an accidental one and hence not void of interest. Let us finally consider an even more abstract example of a synthetical universal semantic property of language. It is what Weinreich calls the 'limited sloppyness' of natural languages, i.e., the fact "that languages are universally less 'logical,' symmetrical, and differentiated than they could be if the components and devices contained somewhere in each system were uniformly utilized throughout that system." (Weinreich 1963: 190) In other words, natural languages tend to be systematic only to a certain degree and they tend to contain a considerable amount of 'exceptions,' and this, of course, does not only hold of the semantic subsystems. If we assume this as a synthetical and not as an analytical universal, we are free to accept languages like Esperanto as natural languages, provided we weaken our notion of natural, requiring self-organization only for the maintenance of the system, and not for its origin. Not everybody will be willing to subsume artificial languages like the international auxiliary languages under the natural languages. But even if one does, other artificial languages that are clearly non-natural, like programming languages or all the diverse languages of formal logic, would not yet fall under this rather weak notion, since they fail to

INTRODUCTION: UNIVERSALS AND SEMANTICS

7

meet the generality requirement by being special-purpose languages for human-computer communication and the encoding of certain aspects of logical inferences, respectively. Wherever the notional border is drawn, investigating universals of human languages across the natural-nonnatural distinction is an interesting task that may still bring important discoveries.5 By contrast, the investigation of universals of natural languages across the human-nonhuman distinction, interesting as it may be, is on a different level, since the knowledge we have about animal languages seems to support the hypothesis that they are special-purpose rather than general-purpose sign systems.6 But the generalpurpose/special-purpose distinction is a graded one, and so are other key notions in our discussion: Living beings can be more or less human (think about possible sign systems of Pithecanthropus) and more or less natural (does Esperanto begin to be natural once it is learned as a first language?). With this cautionary remark we will leave the general reflections about how semantic universals, universal semantics, and their relation can be conceived, and turn to an overview of the themes discussed in this volume. They fall into three groups whose order reflects their degree of specificity. The most general topics are placed at the beginning. Next come the issues which, although more specific, have a direct bearing on the foundations of predicate and propositional logic, those almost indispensable tools of semantic analysis. The last group presents a selection of currently discussed problems.

4. General questions about semantic universals The first three papers address some general questions which may be formulated as follows: Are there non-trivial semantic universals at all, and if so, where are they to be located, especially how are they related to logics? The opening paper by Johan van Benthem is concerned with those systematical universals which are due to the fact that languages are constrained by being systems produced and handled by human cognition. The question which possibilities and limitations this implies is still quite open and may be partially answered through research on universals. He gives an expert overview of recent advances in logical semantics that shed some light on this issue. Among the most interesting findings in this area is the discovery of transcategorial phenomena such as monotonicity or, more generally, inferential sensitivity, of which Krifka's contribution on polarity items gives a stimulating illustration. Another fascinating result is the recurring organization of meanings or linguistically organized concepts into groups of four as in the lexical squares, which come from the combination of operators with different kinds of negation (internal vs.

8

DIETMAR ZAEFFERER

external), yielding what Sebastian Lobner (1990) calls 'duality groups'. A plausible candidate for such a pair of dual operators, concessive and causal subordinators, is discussed in Konig's contribution. Manfred Immler starts his paper from a position that seems quite sympathetic with a sceptical attitude towards the question whether there are non-trivial semantic universals at all. After discussing a considerable number of different positions, he concludes that although there may be no interesting semantic universals, the superficial diversity of semantic systems only covers a deeper commonality: not meanings are universal, but the cognitions behind them. The gross disagreement with the preceding article diminishes to a certain degree if one takes into account that Immler, maybe due to the fact that he mostly discusses more traditionally oriented books and articles, emphasizes lexical meanings, where the diversity is obvious, at the expense of grammatical meanings, where more commonalities can be observed, and it is the latter which are the focus of interest of more recent, logically oriented investigations such as the ones referred to by van Benthem. Bechert is even more radical in doubting the existence of interesting semantic universals: For him the observable degree of diversity among natural languages is too high to leave hope for the discovery of semantic invariants behind it. Even where we might be inclined to admit some, he warns us to be extremely careful to avoid the trap of the effects of colonization. He does not use this term in a metaphorical way, as one might be inclined to think, referring to the influence of our ways of speaking on our way of analyzing differing languages, but rather literally, referring to the influence of our ways of speaking on the languages themselves. In other words, many common features of existing languages may be due to language contact, whereby the colonized language acquires properties of the colonizing one. One has often to go back in history to a point where influence from a colonizing language has not yet been possible in order to detect the real variety of semantic systems. This is very difficult for languages without any historical documents, because one depends exclusively on the highly problematic job of reconstruction. But otherwise, the descriptive linguist easily comes into the situation of the ethnologist who, by misfortune, always arrives at a tribe after the missionary and therefore states monotheism as a religious universal. On the one hand, the current linguistic situation in the world is characterized by a high degree of colonization and consequently of convergence of different linguistic systems. Human language, on the other hand, probably goes back to a single or at least a small number of roots so that, looking backwards, one is also confronted with convergence. Therefore, and this could be the bottom line of what can be learned from Bechert's contribution, the best place to look for genuine (as opposed to enforced) universals is that period in history when there was maximal diversity of languages due to maximal

INTRODUCTION: UNIVERSALS AND SEMANTICS

9

dispersion of mankind together with minimal group-external communication, in short: before divergence turned back into convergence.

5. Semantic universals and the foundations of classical logic The second group of papers addresses two basic issues: Are sentences universally constituted by subjects and predicates, and are constituents universally joined by conjunctions such as 'and'? This entails the question whether standard first order logic (FOL) can. still be regarded as the core of a universal formal semantics, since its atomic formulae are constituted by subjects and predicates (in the one-place case), and its complex formulae make crucial use of prepositional logic with its connectives 'and' and 'not' (the well-known fact that both can be reduced to the Sheffer-stroke meaning 'not both' seems to be a rather superficial phenomenon). Sasse answers the first question negatively, because in the languages of the world, sentence constitution by subject and predicate is only one of three observable basic types that may be further specified according to the kind of device used in relating the state of affairs expression (in our languages the finite verb) to the participant expressions (the complements). Besides the subject-predicate sentence constitution type (Sasse's type 2), where the mere concatenation operation on a subject expression and a predicate expression encodes the semantic operation of proposition formation, there are two other types with segmental, i.e., phonologically observable encoding of this operation that can be distinguished according to the well-known analytical-synthetical distinction (syntactical encoding by grammatical words vs. morphological encoding by inflectional properties). The analytical type (Sasse's type 1) requires a third element, such as a tense-aspectmood marking word, besides the state of affairs expression and the participant expressions in order to form a complete sentence, whereas in languages that are synthetical in this respect (Sasse's type 3), the proposition forming operation is encoded in the morphological form of the states of affairs expression, which allows the latter to constitute a complete sentence all by itself. So the minimal type 3 sentence consists of one word, whereas in the other two types at least a dummy subject (type 2) or an additional grammatical word (type 3) is needed. Does this challenge the widely held assumption that universal semantics, be it cast in the framework of possible worlds semantics (Montague, Cresswell), or discourse representation theory (Kamp, Heim), or situation semantics (Barwise/Perry, Peters), can be based on FOL? I think it does not, since FOL cannot be identified with any one of its notations, and it is possible to devise notations that correspond to all three of Sasse's

10

DIETMARZAEFFERER

types. I will illustrate this with the critical examples of a zero place predicate P and a oneplace predicate Q, the extension to predicates with more arguments being obvious. Type 1 requires a syntactical constituent, written as a symbol that is separated by a blank from its coconstituents, to form a sentence, i.e., a well-formed formula. The difference with type 3, where a morphological element replaces the syntactic constituent, is marked by leaving out the blank in the latter case. FOL notations often use pairs of parentheses for the coding of the proposition forming predicate-argument relation (semantically the settheoretical membership relation), which are discontinuous constituents and therefore syntactically quite complex, so I will in addition use a single colon as an alternative which is closer to natural language. Accordingly, a type 1 syntax will state that if a is an individual constant, then P () and Q (a) are formulae, or alternatively, with the colon in final and the argument in initial position, if a is an individual constant, then P : and aQ : are formulae. A type 2 syntax will state that if a is an individual constant, then e P and a Q are formulae, where e is a semantically empty expression of appropriate category. (Recall that the question of a parentheses/colon alternation does not arise, since type 2 codes the proposition forming operation only by concatenation of functor and argument expression.) A type 3 syntax will state that if a is an individual constant, then P() (without any blank) and Q( a) are formulae, or alternatively, with the colon in suffix notation and the argument in initial position, if a is an individual constant, then P: (without any blank) and a Q: are formulae. This playing around with logical notations shows that those who think Sasse's typology is a challenge to the universality claim of FOL are trapped by a naive identification of natural syntactic subjects with logical argument expressions and of natural syntactic predicates with logical predicate expressions. The logical distinction between predicates and arguments is on a rather abstract level and can be surface-syntactically realized by any one of Sasse's three basic types, not just by the subject-predicate pattern of type 2. Nevertheless, the three types should be kept in mind by all semanticists who try to keep their semantic representations close to the object language's surface. The situation seems to be similar with Gil's contribution. Gil answers negatively the question whether constituents are universally joined by conjunctions such as 'and.' He presents ample evidence that there is at least one human language, namely the Yuman language Maricopa 7 (of the large Northern Amerind phylum), that entirely lacks the syntactic category of a coordinate construction, defined as a construction of some given category with at least two constituents of the same category plus one coordinator. If English sentences with coordinated constructions are translated into Maricopa, a host of different constructions can be used, none of which is exacdy parallel to a coordination; hence it seems that the Maricopas are missing a generalization. But Gil does not conclude

INTRODUCTION: UNIVERSALS AND SEMANTICS

11

from this finding that the logical and-connective is absent from the Maricopa semantics since it might be present without having a direct surface-syntactic counterpart. So the investigation does not provide conclusive evidence for or against the universality of prepositional logic. But it does give a good example of a category whose presence would probably be assumed by many as a universal, contrary to facts. And again, although the diversity found in human languages is surprisingly high, it turns out that the structures of formal logic are much too abstract to be affected by these findings.

6. Number, polarity, cause, and condition: Some special topics in universal semantics The last group of contributions consisting of four papers deals with special topics, namely the syntactic and semantic notions of number, polarity items, causal and concessive operators, and conditional constructions and their relatives. Link's paper, which presents an intriguing combination of linguistic findings with philosophical discussion and formal semantic exploration, deals with the non-trivial relation between the universal cognitive semantic category of quantity and the universal morphosyntactic category of number (cf. Greenberg's universal 42). 8 It discusses the ontological prerequisites for counting and then discusses the phenomenon that in general there is no direct way of inferring quantity from number except in the case of the socalled autonomously referring expressions (Keenan). In the last part a formal way of dealing with this problem is proposed. Polarity items, both negative and positive, are a class of expressions whose universal existence is far from having been confirmed. The reason is that too many descriptive grammars are not sufficiently informative in this respect. But the phenomenon is obviously widespread and interesting enough to be discussed in the context of a volume like the present one. These items are more aptly characterized as expressions whose occurrence is restricted to contexts exhibiting a certain inferential behaviour, allowing for - sometimes limited - truth-preserving local strengthening and weakening, respectively. Krifka's paper shows how the surprising generalizations gained from the semantic approach, which goes back to Ladusaw's pioneering work, have to be restricted in order to fit the data. It furthermore demonstrates how the seemingly isolated phenomenon can be integrated into the general picture if one adopts the independently motivated view that in natural languages propositions have to be interpreted against the backdrop of an ordered set of alternatives.

12

DIETMAR ZAEFFERER

Ekkehard Konig's contribution explores the evidence for and the implications of the hypothesis that concessive and causal subordinators can be conceived of as dual operators, i.e., pairs of operators that are interdefinable with the help of internal and external negation. Although the resulting type of duality group is somewhat special in that it seems to lack both left and right mononicity, which are at least partially present in most of its structure-mates (cf. van Benthem's paper), logical behavior and an impressive amount of cross-linguistic data support the hypothesis. Additional corroboration comes from the fact that it allows to explain the semantic change from a causal to a concessive reading through reanalysis of the relative scope of negation and adverbial, as with English for and German darum. All this makes the paper an excellent illustration of the potential for mutual clarification inherent in general comparative and logical semantic analyses which has been the chief motivation for conceiving the present volume. Zaefferer's paper on typological and logical properties of conditionals and related constructions is a further example of the very same methodological approach. It first formulates this approach in its introductory part, and then gives a cross-linguistic survey of the kinds of constructions used in natural languages to encode the conditionalization of a given proposition. It argues that the findings support those logical analyses which treat conditionals as restricted modal operators, such as Kratzer's theory. The paper closes with the outline of an analysis along these lines that accounts not only for standard conditionals, but also for modus ponens conditionals (assertions of a conditional proposition in a context where the antecedent is already accepted) and for unconditional (conditional constructions with a disjunctive or generalizing antecedent which implicate the truth of at least one of the propositions covered by the antecedent).

7. Conclusion Universal semantics tries to account for semantic universals, for general properties found in the meaning systems of all natural human general-purpose sign systems. At present it seems that an exhaustive list of relevant hypotheses is still far from being established, let alone from being confirmed. And the more challenging enterprise of explaining those semantic universals that are amenable to explanation, the systematical ones, seems even further away from completion. But it is to be hoped that the present volume has clarified at least two points: First, the task of an interesting, empirically contentful universal semantics is quite an involved one. And second, it is probably too difficult to be accomplished by either the data-oriented extensive approach or the logically oriented intensive approach alone: The future of universal semantics in particular and of universal

INTRODUCTION: UNIVERSALS AND SEMANTICS

13

grammar in general depends on how they interact. There is reason to believe, and this is the third and main point the present volume gives ample evidence for, that substantial progress will ony be made if they join forces and continue to learn from one another.

Notes 1. This opposition corresponds to the two major approaches to universal grammar outlined by Comrie (1981: If.) and hallmarked with the names of Joseph H. Greenberg and Noam Chomsky (cf. the two readers Greenberg 1963 and Bach/Harms 1968). The more recent developments of Chomskian generative grammar with their emphasis on universal principles and language specific parameters and with an increasing number of languages that are taken into account has lead, however, to a certain convergence of the two schools, cf. Comrie (1988: 458ff.).

2. There is a third strand of work within the philosophy of language and especially within logics that is neither concerned with accidental nor with induced properties of natural languages, but rather with properties of ideal languages, and which is called ideal language philosophy. Obviously, it can only investigate analytical properties of its constructs; therefore it is of no direct interest in our context.

3. Computational linguists assume the upper complexity bound of natural human language expressions to be in the region of the indexed languages, i.e., between the context-free and the context-sensitive languages (Gazdar/Mellish 1989: 137).

4. This follows from Greenberg's universals 35 ("There is no language in which the plural does not have some non-zero allomorphs, whereas there are languages in which the singular is only expressed by zero....") and 42 ("All languages have pronominal categories involving at least ... two numbers") (Greenberg 1963:112f.).

5. Van Benthem 1989 may be considered as a first step in this direction.

6. Cf. Demers (1988: 333): "What all nonhuman communication systems lack, however, is the unboundedness in scope that is the central feature of human language." 7. Johan van Benthem (p.c.) has pointed out to me that the Maricopas should be known to the readers of Karl May: They appear in his novel Im Tal des Todes, unfortunately on the wrong, i.e., "bad' side. 8. "All languages have pronominal categories involving at least three persons and two numbers." (Greenberg 1966: 96).

14

DIETMAR ZAEFFERER

References Bach, Emmon, and Robert T. Harms (eds.): 1968, Universals in Linguistic

Theory,

Holt, Rinehart and Winston, New York. Barwise, Jon, and Robin Cooper: 1981, 'Generalized Quantifiers and Natural Language', Linguistics and Philosophy 4, 159-219. Barwise, John, and John Perry: 1983, Situations and Attitudes, MIT Press, Cambridge, MA. Bengtson, John D., and Merritt Ruhlen: to appear, 'Global Etymologies,' in Vitaly V. Shevoroshkin (ed.), Genetic Classification of Languages, University of Texas Press, Austin. Benthem, Johan van: 1989, 'Semantic parallels in natural language and computation', in H.D. Ebbinghaus et al. (eds.), Logic colloquium, Granada 1987, North Holland, Amsterdam, 331-375. Chomsky, Noam: 1968, Language and Mind, Harcourt, Brace & World, New York. Comrie, Bernard: 1981, Language Universals and Linguistic Typology, Blackwell, Oxford. Comrie, Bernard: 1988, 'Linguistic Typology', in Frederick J. Newmeyer (ed.), Linguistics: The Cambridge Survey. I Linguistic Theory: Foundations, Cambridge University Press, Cambridge, 447-461. Demers, Richard A.: 1988, 'Linguistics and animal communication', in Frederick J. Newmeyer (ed.), Linguistics: The Cambridge Survey. Ill Language: psychological and biological aspects, Cambridge University Press, Cambridge, 314-335. Gazdar, Gerald, and Chris Mellish: 1989, Natural Language Processing in PROLOG, Addison-Wesley, Wokingham. Greenberg, Joseph H.: 1963, 'Some universals of grammar with particular reference to the order of meaningful elements', in Joseph H. Greenberg (ed.), Universals of language, MIT Press, Cambridge, MA, 2nd ed. 1966, 73-113. Löbner, Sebastian: 1990, Wahr neben Falsch. Duale Operatoren als die Quantoren natürlicher Sprache, Niemeyer, Tübingen. Maddieson, Ian: 1984, Patterns of Sounds, Cambridge University Press, Cambridge. Weinreich, Uriel: 1963, 'On the semantic structure of language', in Joseph H. Greenberg (ed.), Universals of Language, MIT Press, Cambridge, MA, 2nd ed. 1966, 142216.

Zaefferer, Dietmar: 1991, 'On the Coding of Sentential Modality', in Johannes Bechert, Giuliano Bernini, Claude Buridant (eds.), Toward a Typology of European Languages. Proceedings of the Workshop held at Consiglio Nazionale delle Richerche, Rome, 7-9 January 1988, Mouton de Gruyter, Berlin, 215-237.

Part I General Questions

Linguistic Universals in Logical Semantics Johan van Benthem Institute for Language, Logic and Information Department of Mathematics and Computer Science University of Amsterdam

1. Logic and Universal Grammar Contemporary research into the semantics of natural language has produced a system of abstract meaning structures, often inspired by logical paradigms. The aim of this paper is to discuss the potential of this system as a source of semantic universals, that is, general laws about meanings expressible in all human languages. Perhaps, such a question seems surprising. After all, ever since Antiquity, there has been an ideal of 'Universal Grammar', beyond the peculiarities of particular languages: of which the discipline of Logic was a guardian. Thus, are not the outcomes of logical semantics universally valid automatically, so to speak, due to their a priori conceptual nature? Not necessarily. There is no reason for premature complacency here, as there are still matters of fact to be respected. Notably, empirical research in comparative linguistics is bringing to light a staggering variety of linguistic forms and associated conceptual schemata, which present the following challenge. What is the universal validity of the conceptual frameworks, and claims based upon these, thought up in the arm-chair of a native speaker of some Indo-European language? Can purported truths about Language (singular) reached in this convenient manner be maintained in the light of evidence about other languages (plural), obtained through hazardous journeys across arid deserts and steaming jungles? Thus again, how can we relate the abstract meaning structures of logical semantics to valid and significant claims about natural language? Before addressing this question more concretely, let us consider some methodological points in advance. To begin with, we are dealing with a subject which is less concrete than, say, linguistic syntax. In semantics, we postulate an 'invisible world' of meanings behind natural language forms, so that comparing the meaning structures of different languages is not a straightforward business. In particular, it would be naive to translate syntactic differences directly into semantic ones, since languages may have many resources to serve given semantic functions.1 But also, this aspect of abstract 'construction' in the very evidence may make certain proposed semantic universals 'self-made', reflecting features of our chosen framework of representation, rather than hard facts. Indeed, this is

18

JOHAN VAN BENTHEM

unavoidable, and not necessarily harmful: one should merely be aware of the phenomenon. Actually, one wants further theoretical considerations to come in at some stage anyway. This is a quite common situation in scientific enquiry. Current proposals for semantic universals reflect an initial 'botanical' stage of research: they express simple regularities in phenomenological terms. Examples are proposed typological patterns (either SVO and related sentence types, or some more sophisticated variant thereof), or observed correlations between syntactic forms of expression for certain semantic phenomena. Then, there comes a time when one would like to systematize and explain such regularities in an explanatory background theory. And, again as in any healthy science, there is no need whatsoever to postpone such theorizing until after all relevants facts have been acquired: because we also need such a theory beforehand, to guide us in first defining and then searching for what are the 'relevant facts', if we are not to be drowned in an ocean of details. Where are such explanatory theories to be found? Natural language is a diverse field, and there may be no uniform answer to this question. Clearly, genetic information about the actual development of languages is relevant, and so are various findings from empirical psychology. On the other hand, there is also a more theoretical approach, viewing language as a cognitive system, both for individuals and for communities, whose design can be studied by methods from mathematics and computer science, as well as 'cognitive science' generally. It is especially in the latter perspective that logical semantics may be able to contribute, both as to suitable structures for representing knowledge and as to algorithms for transforming and communicating these. This theme will be elaborated in what follows.

2. Logical Semantics Modern semantics of natural language was put on a systematic footing by Richard Montague around 1970. He showed how one could assign suitable abstract meanings to expressions of a natural language in a uniform, systematic manner. It will be sufficient here to recall only an outline of what was achieved. Type-Theoretic Semantics On the abstract side, we assume that meanings come in a functional hierarchy built up as follows. At the most basic level, there are individual objects (which are assigned a primitive type e), as well as truth values (of type t). Other primitive types may be added

LINGUISTIC UNIVERSALS IN LOGICAL SEMANTICS

19

as the need arises, such as a type s for 'possible worlds', 'situations' or 'states'. Next, compound functional types (a, b) may be formed out of components a, b: standing for functions from a-type things to b-type things. For instance, (e, t) stands for unary functions from individuals to truth values: i.e., one-place predicates of individuals; and iterating, ((e, t), t) stands for one-place predicates about the former predicates. These abstract denotations can now be correlated with traditional syntactic categories of expression, so as to ensure systematic interpretation. For instance, here are two basic sentential patterns with corresponding types: "Mary sings": proper name

intransitive verb

e

(£Jl t

• "No frog sings": common noun

determiner ((e. t). ((e. Ü.

ieal

intransitive verb (e,t)

((e, t), t) Here is a more extended category-to-type correspondence which, although simplified, conveys the flavour of the approach: proper name

e

sentence

t

intransitive verb/common noun

(e,t)

complex noun phrase

((e, t), t)

adverb/adjective

((e, t), (e, t))

determiner

((e, t), ((e, t), t))

one-place connective

(t,t)

two-place connective

(t, (t, t))

transitive verb

(e, (e, 0)

Thus, grammatical categories of expression in natural language are systematically related to certain types of abstract semantic objects. For instance, a determiner such as "no", "every", "some", "most", will be of type ((e, t), ((e, t), t>); which may be read as a function from two one-place predicates to a truth value, i.e., as a binary relation between unary predicates. This fits, as, e.g., "every PQ" expresses inclusion of P in Q, and "some PQ" overlap between P and Q.

20

JOHAN VAN BENTHEM

Incidentally, how far is this account of semantic structure biased towards the syntax of some particular language, or family of languages? It should be noted, in any case, that the given type system can serve many other systems of categories, in addition to the above traditional one. Thus, even if certain languages lack (prominent) determiners in the above spirit, other semantic counterparts may reappear elsewhere in the type structure. Also, even as it stands, the above assignment is not unduly respectful of syntactic surface structure. For instance, as shows in the above view of determiners, there is little difference between common nouns and (intransitive) verbs: semantically, both express predicates of individuals. Accordingly, both categories have been assigned the same type - and so have the corresponding 'modifier categories' of adjectives and adverbs. And this seems right: certain languages do not even have a syntactic distinction between the two classes. And also within, say, English itself, these two categories exhibit many similarities. So, the Montagovian scheme certainly tries to keep a judicious distance toward the syntactic accidents of particular languages. Denotational Constraints Up until now, we have been mainly concerned with semantic structures, not with further claims based on these. And indeed, despite its grand title, Montague's "Universal Grammar" was a compendium of definitions, rather than an enlightening theory, about natural language semantics. (In this respect at least, his contribution must be ranked below Chomsky's achievements in natural language syntax.) Gradually, however, such further claims have been forthcoming in the resulting research program of logical semantics. We shall review a few examples here, which have led to the formulation of semantic univers a l (cf. Barwise & Cooper 1981, Keenan & Stavi 1986, Zwarts 1986, Lobner 1987). A first noticeable theme has been the search for so-called denotational constraints on the interpretation of categories of expression. The general idea is that certain grammatical categories are interpretatively 'free', in that simplex linguistic expressions belonging to them can freely denote any item in the corresponding type of semantic objects. Examples are proper names (type e ) or intransitive verbs (type (e, t)). Other categories, however, obey certain constraints, in that certain mathematical possibilities for denotations seem to be excluded. One famous case is that of the above determiners. These cannot just denote any binary relation between predicates, but only those which are conservative, in the sense of being restricted to their first argument: D P Q if and only if D P ( P n Q ) . For instance, "no minister resigned" is equivalent to "no minister was a minister-whoresigned".3 In fact, Keenan and Stavi (1986: 260) have proposed the following semantic universal, at least for 'extensional' determiners, without intensional aspects of meaning:

LINGUISTIC UNTVERSALS IN LOGICAL SEMANTICS

21

"Determiners in all languages are always interpreted by conservative functions".4,5 Patterns of Inference A following theme is that of inference, and in particular, stability of semantic information under certain changes of context. Again, this can be illustrated by the case of determiners. For instance, the statement that "every cat jumped" remains true if the predicate "jump" isreplacedby the weaker predicate "move": "every cat moved". Moving has a wider extension than jumping, and thus, we can describe this inference structurally as a case of so-called upward, monotonicity: every PQ, Q c Q ' imply every PQ'. Other determiners can be downward monotone. For instance, "no cat jumped" implies "no cat (jumped gracefully)": no PQ, Q'c Q imply no PQ'. Moreover, monotonicity may also occur in the left-hand argument, again upward or downward. This gives us four 'double monotonicity' types, of which the traditional logical quantifiers are the prime examples. In a self-explanatory notation: •I every T i no 1

T some t T not every i Other quantifiers may be monotone to a lesser degree. For instance, "most" is upward monotone on the right-hand side, but neither upward nor down-ward monotone on the left. The sentence "most Germans are thrifty" implies neither "most Europeans are thrifty" nor "most Southern Germans are thrifty". Monotonicity is an important notion, with uses all across logic and linguistics (cf. Zwarts 1986). It also strongly conveys a certain psychological flavour: we probably need basic concepts with some degree of stability under incoming information. In fact, Barwise and Cooper have proposed the following semantic universal:

22

JOHAN VAN BENTHEM

"In human languages, all simplex determiner words denote (rightward) monotone relations; or conjunctions of the latter". Here, the reason for the final clause is that a determiner like "one" may mean that precisely one P is Q, a relation which is not monotone by itself - although it is equivalent to the conjunction of "at least one" (upward monotone) and "at most one" (downward monotone). 6 Also, with complex determiners, we can define non-monotone relations ("some but not all", "between two and five"). Besides monotonicity, there are also other ways of extracting universals from observed inferential behaviour of determiners. For instance, Zwarts 1986 discusses syllogistic patterns of inference exemplified by basic determiners, such as DPO

DOR

D PR

DPO or

(Transitivity)

D QP (Symmetry)

He then observes the occurrence of certain systematic gaps: there are plausible-looking syllogistic patterns which do not seem to be realized by any expression in natural language. Accordingly, he conjectures the following semantic universals concerning analogues of the above two schemata: In human languages, no determiner validates Asymmetry

('from D PQ to not D Q P ' )

Circularity

('from D PQ, D QR to D R P ' ) .

or

Why should this be so? We shall return to the matter of explanations below. Remark. It may be of interest at this stage to point at an analogy with the work of C.S. Peirce. This far-seeing logician and semiotician tried to develop an account of 'natural reasoning' which should be in harmony with the basic mental operations that humans can perform on knowledge representations. Not surprisingly for a linguist, Peirce's operations turn out to be the following three: copying, addition, deletion. (See Sánchez Valencia 1989 for a modern exposition.). And these are precisely what has been uncovered in our notions of Conservativity (which involves copying of restricting predicates) and Monotonicity (which involves adding or deleting features when replacing predicates by stronger or weaker ones). Again, this points at a possible psychological

LINGUISTIC UNIVERSALS IN LOGICAL SEMANTICS

23

connection for basic notions of semantics. A more general system of 'natural logic1 on linguistic forms exploiting these ideas may be found in Sánchez Valencia 1990. 'Lexical Squares' Further examples of universal semantic phenomena arise once we consider the interaction of determiners with other important linguistic operations. A prominent example is that of negation. For instance, in the earlier 'Square of Oppositions' for the standard logical quantifiers, we encountered combinations with both external and internal negation: (all P)Q

(all P) not Q ("no")

("some") not ((all P) not Q)

not ((all P)Q)

Many expressions in natural language tend to congregate in such families, with various mutual dualities.7 Lobner 1987 presents a list of examples suggesting strongly that this is a universal phenomenon. Note also that such a theoretical schema serves one further function indicated earlier on: it suggests a systematic view-point from which to search for comparative facts, for instance, concerning lexicalization of the various comers in the Squares. (See also ter Meulen 1988 on further elaboration of this idea in the field of temporal expressions.) Transcategorial Phenomena It should be emphasized that the above examples of semantic themes with universal implications are not tied up with any particular grammatical category of expression (say, that of determiners). In fact, there is a noticeable tendency in contemporary semantics towards describing general phenomena across categories and types. One instance of this is the phenomenon of Monotonicity, which can be described quite generally in arbitrary 'Boolean types', so as to cover all kinds of 'inferential sensitivity'. (For instance, we also want to say that a prepositional phrase can be upward monotone, in the sense that "with a cudgel" implies "with a blunt instrument"; and so forth.) For formal details, see Keenan & Faltz 1985, van Benthem 1986. This generality is even forced upon us by the well-known polymorphism of certain linguistic expressions, which seem to travel freely across varying syntactic environments. Notably, this occurs with the so-called Boolean operators of negation ( "not" ), conjunction ( "and") and disjunction ("or"), as has been amply demonstrated in Keenan

24

JOHAN VAN BENTHEM

and Faltz 1985.8 Thus, a good deal of current research is devoted to a liberalization of the earlier Montagovian category-to-type schema, allowing families of types for one expression with suitably related denotations in each type. In particular, one may study admissible type changes and corresponding meaning changes for linguistic expressions. For instance, a proper name (denoting an individual object of type e ) may be 'lifted' to the type ((e, t), t) of complex noun phrases: where it will then denote the set of properties satisfied by that former individual object. And similar systematic recipes can be given for Boolean particles, as they change from sentence operators to predicate operators: say, from truth-functional negation to set-theoretic complement One mechanism for implementing such ideas in great generality is the lambda calculus behind current versions of Categorial Grammar (cf. Buszkowski et al., eds., 1988 or Oehrle et al., eds., 1988). This calculus gives us a principled logical procedure for computing denotational constraints and inferential properties in different syntactic contexts. Thus, for instance, even a determiner occurrence in a prepositional phrase like "to every mountain" will acquire the intended semantic conservativity effect, being the restriction of the corresponding argument role in the binary predicate "walk-to". (This is not to deny that other mathematical principles of 'semantic metamorphosis' may be operative in the type structure as well: witness the analysis of 'homomorphic lowering' in van Benthem 1986, chapter 3.) A full logical exposition of this framework is found in van Benthem 1991. Notably, in this light, the proper transcategorial perspective on Conservativity emerges. It becomes one instance of the general phenomenon of what may be called domain restriction in the course of evaluation of a sentence: individual argument roles for predicates encountered will be restricted to varying subdomains indicated in the syntax. And so, we find a more 'dynamic' semantic difference between (intransitive) verbs and common nouns after all: the former supply the predications, the latter the relevant argument restrictions. The mechanism of meaning shifts can itself be exploited again for the purpose of a more universal semantics. Thus, Keenan 1987 uses the modern Categorial Grammar treatment of noun phrases in direct object position (where they must combine with a transitive, rather than an intransitive verb) in order to provide a structural, syntaxindependent definition for the universal notion of case.9 Of course, the above kind of generality still concerns the categorial structure of one particular language (though now in its full length and width). But, the transcategorial perspective should make it easier to arrive at cross-linguistic generalizations as well. Or at least, such is our hope.10

LINGUISTIC UNIVERSALE IN LOGICAL SEMANTICS

25

3. Language as a Cognitive System Let us now return to the issue of general features of natural language, viewed as a cognitive system, and possible logical explanations for these. Here are some general assumptions about natural language which many people find plausible. It is • expressive

, in that every useful meaning can be verbalized,

• efficient

, in that it is geared toward fast performance of its tasks,

• computational

, being based on (simple) algorithms;

and we may add such features as its learnability and adaptability. So, the question becomes if we can explain certain semantic regularities as contributing toward a design having the above features. In more anthropomorphic terms, would we use such semantic notions if we had to design natural language ourselves ... ? n What we shall do here is merely show that various results obtained in contemporary logical semantics may be viewed as reflecting some (modest) aspect of the above general features of natural language. For a start, expressiveness has featured in various publications, under various names. E.g., Keenan and Stavi 1986 use the term 'effability', proving that, on fixed finite domains of individuals, every conservative relation between predicates can be defined using some (complex) determiner expression in natural language. Another set of examples may be found in van Benthem (1986, part I), where it is shown that all doubly monotone relations between individual predicates are expressible in natural language, using the classical calculus of the standard logical quantifiers. Other examples, concerning definability of so-called 'invariant' semantic objects, may be found in van Benthem 1989c. Thus, natural language does provide names for some of the more important special kinds of semantic structures that exist a priori. In this light, it seems all the more remarkable that certain systematic gaps occur, as was noticed before. But as it happens, this is often a matter of logical necessity. Again in van Benthem (1986, part I), one finds mathematical arguments showing that there just cannot be any conservative quantifier relations obeying Asymmetry or Circularity. In other words, the earlier semantic universal proposed by Zwarts is true for a priori reasons, and hence devoid of empirical content (though therefore not of interest!). This again illustrates an earlier methodological observation. One of our tasks is precisely to gauge how much empirical, as opposed to how much a priori content there is to proposed semantic universals.

26

JOHAN VAN BENTHEM

Even without expecting full expressiveness all the time, we can use our semantics systematically to investigate lexicalization of possible meanings.12 Here is one example from a quite different linguistic field. With temporal expressions, we often have strong language-independent 'pictorial' intuitions about the structure of time. For instance, events may be regarded as occupying certain convex intervals in space-time; and then, their possible comparison must fall into one of a pre-computable number of possibilities: the 'magic number' as it occurs in many 1^ publications, is thirteen. Then, we can ask in how far natural languages allow us to explicidy refer to these relations, by means of simplex or complex temporal connectives. For instance, it turns out that we do not have thirteen basic temporal connectives matching these cases: the natural candidates being only "before", "until", "while", "since", "after". There is much more to be said about this particular example, of course; but here, it will have served its purpose as an illustration of a possible systematic way of thinking about the 'design' of temporal expressions.14 The latter are an interesting example anyway, because, on the one hand, they reflect invariant 'basic physics', whereas on the other, even quite closely related languages show striking differences in temporal representation. (Compare English perfect tense versus Dutch or German perfect tense; and likewise their progressives. For a detailed empirical account, including also Aktionsart and temporal adverbs, see Oversteegen 1989.) The possible universal implications of recent work on temporal semantics remain largely to be explored. Next, as for efficiency of natural language, various general phenomena have been connected with this. For instance, Barwise and Perry 1983 use it to explain the presence of indexicality and of ambiguity, both serving to make one expression re-usable over many varying situations. This does not support very specific predictions, of course, but the idea certainly sounds plausible. For more specific claims concerning efficiency, however, we may have to pass on to the third feature in the above list, namely computability. Given the present-day interest in computationally oriented linguistics, there are many general ideas floating around here. Indeed, one of the earlier denotational constraints itself has a computational flavour. As we have seen, Conservativity really expressed a restriction, when understanding a sentence, to certain specifically indicated subdomains of the whole universe of discourse. This showed in patterns like D AB, which reduced to D A(B n A) - and it becomes even more conspicuous in derivative patterns for determiner complexes with transitive verbs: DiA R D2B if and only if DiA R n (A x B) D2B. That is, in understanding a sentence like "every leak needs a plug", we need only consider the relation "need" as it runs from leaks to plugs. This may be viewed as a

LINGUISTIC UNIVERSALS IN LOGICAL SEMANTICS

27

reflection of how our short-term memory actually works with a system of restricted domains.15 (See also Westerstdhl 1985 on this issue.) Next, one can also analyze specific expressions and their corresponding meanings as to their computational complexity. For instance, quantifiers are viewed as 'semantic automata' in van Benthem (1986, Chapter 8): that is, little procedures for computing truth values, given the extensions of their two argument predicates. Then, we find a hierarchy of complexity: some expressions have meanings which are easier to compute than others. In particular, the standard simplex determiners are easy to compute, using simple finite state automata - whereas, e.g., the procedure for the non-standard quantifier "most" requires a push-down store automaton possessing arbitrary memory. 16 Thus, natural language semantics, like natural language syntax, admits of a significant Hierarchy of Complexity, which can be used to make predictions about learnability of meanings. (On the latter topic, see also van Benthem 1987.) Indeed, there is a more general procedural aspect to language, which has not come out too well so far in the usual style of doing model-theoretic semantics. No doubt, many linguistic expressions give us information, not just about their eventual denotations, but also about the particular procedures (to be) followed in computing them. For instance, an ordinal expression like "every third captive was beheaded" provides not only quantified information, but also a particular order of 'surveying': whether in the perceived reality, or only with the mind's eye. Likewise, many expressions invite us to 'travel' along semantic scales: witness, e.g., Fauconnier's well-known work on polarity, or the 'phase quantification' of Lobner 1987. These procedural aspects are slowly coming to the fore even in more recent treatments of logical constants in natural language, as operations of 'control' over the process of updating information states by incoming linguistic material. For instance, then, a conjunction "and" will come to indicate sequential execution of tasks, whereas a disjunction "or" points at a choice. (See van Benthem 1989a for a survey.) In particular, current computational work on syntactic/semantic parsers (cf. Moortgat 1988) is producing a test-bed for various ideas about the 'dynamics' of constructing semantic denotations. This whole new family of well-defined procedural approaches to semantics may have considerable universalist interest After all, the idea now becomes that, in addition to its descriptive functions, natural language also serves as a programming language for modifying information states. Thus, semantic regularities may also be sought at the level of computational design. For instance, how does its system of explicit 'control expressions' (such as the above logical constants, or particles setting up subroutines, like "if' or "although" ) compare to the structures encountered in the usual programming languages?

28

JOHAN VAN BENTHEM

4. Higher Levels of Language The above survey still lies well within traditional boundaries, namely the semantics of words and expressions within single sentences. But, in order to get at significant universals concerning human languages, we must also ascend to higher levels of aggregation, being those of text and discourse. For instance, it is only at the level of texts that we get a proper view on such phenomena as anaphoric connection or coherence and architecture in general. No doubt, many important insights are waiting to be discovered at this level - although few non-trivial universals have been proposed so far. 17 This level also invites the drawing of a distinction between actual semantic structures where 'real' denotations live, and more ephemeral (discourse) representations (cf. Kamp 1984). The interplay between these two levels must eventually be taken into account as well, if only to provide a suitably rich setting for already the computational concerns broached in the preceding Section. As there is still no elaborated framework in this field comparable to Montague's semantic structures, it may be somewhat premature to expect precise testable universals at this level rightaway. Nevertheless, some intriguing proposals have been made by Fauconnier as well as Sperber & Wilson (cf. the survey of discourse semantics in Guenthner 1989), employing general principles of 'informative economy' when constructing representations. With a consideration of discourse, we also arrive at a richer picture of the various functions of natural language. Up till now, our story might have been about isolated agents searching for knowledge and information. But of course, language is also a social system for communication - and many of its broad features might be explained through this source. To get a grip on this, we will have to widen our scope from linguistic forms (including semantic structures) to the rules governing language use. And again it makes sense to inquire into universal versus more idiosyncratic features of languages (or linguistic communities) in this respect. One example may serve to show that this is not mere speculation. For instance, there appears to be a wide-spread stable linguistic practice of 'optimizing available information', which has been noted by linguists, philosophers and computer scientists alike (cf. van Benthem 1989b). That is, in addition to their literal meanings, expressions in given dialogue settings also suggest that all relevant information has been provided. An answer to the question "Who killed the lion of El Bahr?' might be

LINGUISTIC UNTVERSALS IN LOGICAL SEMANTICS

29

"Greifenklau and Von Eschenrode", with a literal meaning that at least these two were involved, but a 'semantic surplus value' to the effect that only these two were. That the latter is not strictly implied may be seen by the possibility of cancelling: the answerer may add a third person upon reflection, without contradicting herself, and the questioner may refuse the additional implication by going on to ask "And who else ?" (Thus, the additional inferences beyond the basic semantic information are cancellable; or non-monotone, as current jargon has it.) Sometimes, in fact, the linguistic form of certain answers itself encodes a literal, rather than a minimal reading. For instance, this is one of the procedural functions, over and above its purely descriptive content, of the expression "at least" in an answer to the above question like "At least two Prussian noblemen". Once observed, such phenomena of 'surplus meaning' turn out to be ubiquitous across a wide range of language use.18 Again, such conventions may be tested for universal occurrence among linguistic communities, as well as theoretical grounding in the necessities of cognitive computation (cf. McCarthy 1980). More generally, the phenomenon encountered here is the existence of what may be called various "modes" of taking incoming linguistic information. We may have to add it to our existing knowledge, or rather retract it, or merely 'entertain' it for the time being. We may take it in its literal form, or in the above strengthened minimal manner (while 'maximality' principles also occur). Again, what needs to be studied is both the abstract structure of these phenomena, and its degree of encoding in the lexicon or the syntax of natural languages.

5. Conclusion What we hope to have illustrated with this modest survey is how insights from current logical analysis of natural language may be useful in suggesting and analyzing semantic universals. Of course, much of the above is speculative, while also lacking a well-developed unifying theory allowing for parametrization toward specific languages.19 Nevertheless, it seems clear that more theoretical and more empirical approaches stand in need of one another here. Paraphrasing an elegant dictum of Immanuel Kant, "Theory without facts is empty, Facts without theory are blind."

30

JOHAN VAN BENTHEM

Notes 1. One striking example concerns the presence or absence of articles in natural languages. For instance, it has been claimed that Russell's theory of 'definite descriptions' would not be comprehensible to speakers of languages lacking expressions of the form "the so-and-so". But in fact, for instance, a Russian author will have no difficulty whatsoever in explaining Russell's problem and solutions: witness E. D. Smirnova, 1986, Logicheskaja Semantika i Filosofskije Osnovanya Logiki, Izdatelstvo Moskovskogo Universiteta. Likewise, quantifier combinations requiring indefinite articles ("every boy loves a girl") will remain perfectly comprehensible in Russian, either in the absence of the form "a", or with another type of expression moving in to fill its rôle (viz. "nekotoroi"). And similar observations apply to the purported absence of predicative "is" in Russian. Thus, one has to be very careful about making direct semantic inferences from initial syntactic observations.

2. Another quite different source of explanation may lie in physical reality itself. For instance, it has often been suggested that our spatio-temporal vocabulary directly reflects physical space-time as it happens to be. In fact, already in the last century Helmholtz put forward the intriguing hypothesis that it is the geometrical invariants of space-time, i.e., those notions unaffected by the usual transformations of translation and rotation (corresponding to our bodily movements), which are lexicalized in human languages. More generally, lexical semantics might require a wide range of explanations, whereas purely constructional semantics might be more strictly cognitively and computationally oriented.

3. Note again the distance from syntax here. There is no straightforward way of expressing the intersection or conjunction of Q and P, and we must use various circumlocutions. And yet, native speakers seem to have no trouble whatsoever in understanding the point of the proposed regularity.

4. This universal already illustrates the earlier-mentioned interaction between semantic observation and semantic theory. For instance, one possible counter-example to the equivalence is the expression "only": "Only men are cruel men" is not equivalent with "Only men are cruel". But nowadays one finds that this failure of conservativity rather becomes one in an already existing list of reasons for not counting "only" as a determiner (but rather, say, an adverb). Thus, succesful pieces of theory exert pressure on previous empirical schemes of classification.

LINGUISTIC UNIVERSALS IN LOGICAL SEMANTICS

31

5. There may be a deeper mathematical background to the distinction between interpretatively 'free' and 'constrained' categories. At least to some extent, the distinction runs parallel with the lack of, or presence of a large number of logical items, when the latter are explained in terms of invariance for permutations of individuals. (See van Benthem 1989c.) Generally speaking, categories having many logical terms correspond to predicates whose arguments posses a lot of internal structure which may be respected by that predicate. (See de Mey 1990 for a general account of linguistic semantics in terms of 'respect of argument structure'.)

6. Actually, the counter-example is debatable, since "one" usually means 'at least one' - with the 'precisely one' reading only forced by stress. In that case, the proposed universal would be even cleaner. Cf. Thijsse 1983.

7. The reason why this is a natural scheme is that further iterations serve no purpose. Mathematically, with -i for external negation and - for internal negation on a noun phrase 'Det P', we get the closed composition table of Klein's "Vierergruppe" (Here, # stands for - , - . ) :

-

- ,

~

#

Actually, this duality has already been exploited frequently in Philosophical Logic, where operators often come in squares (as in modal and tense logic). Significantly, it also figures prominently in Freudenthal's LINCOS, a language for cosmic intercourse' designed in the fifties, which was supposed to teach other intelligences from outer space the essences of human thinking.

8. As conjunction seems such a basic phenomenon, it should be added that there are also non-Boolean forms of conjunction ; as in "John and Mary quarreled" which has a reading where "and" merely forms some collective (John, Mary} about which something is predicated. This is equivalent to "John quarreled with Mary", a construction closer to the 'accompaniment' form encountered in various languages for all conjunctions.

32

JOHAN VAN BENTHEM

Incidentally, again it is precisely our theoretical perspective on Boolean particles which enables us to see such distinctions. (But, see Link 1984 for an 'assimilationist' view, respecting the unity of English or German syntax here. For an up-to-date discussion of the literature, compare also Krifka 1989.) In fact, it may be shown that Boolean and non-Boolean conjunction do not exhibit the same kind of polymorphism (cf. van Benthem 1988).

9. Briefly, 'nominative case' corresponds to NPs in their basic type meaning, and 'accusative case' to their type-shifted meaning when accommodated to transitive verbs using the so-called "Functional Composition Rule" of modern Categorial Grammar.

10. Another way of achieving such maximum generality within one single language is to follow several variants of one type of expression. For instance, in English or Dutch, quantifying expressions with determiners in front can also let their determiners 'float', as in "All fairies were invited": "The fairies were all invited". Thus, they become more like adverbs; something which is even more noticeable in: "Most fairies were invited": "The fairies were invited for the most part". Other such shifts occur with the help of (quasi-) temporal adverbs: "All children are cruel": "Children are always cruel", or "Many children are cruel": Children are often cruel" (in one of its readings). A truly general theory of quantification, even for one single language, should encompass such shifts and show how the general theory of Conservativity, Monotonicity and the like fares under these changing circumstances. Thus, we already allow for other natural languages, where the latter type of construction may be dominant over the first. (And similar points could be made concerning the jumps between determiners and adjectives with numerical expressions "one", "two", "three"

)

11. This formulation is not wholly facetious. For, there is at least one laboratory' where we may observe languages being conciously designed: namely in Computer Science. As has been observed occasionally in the literature, it would certainly pay to study existing programming languages from a linguistic point of view, in order to draw comparisons with computational features of ordinary human languages. Conversely, of course, the control structures of programming languages may also serve as a model for our thinking about natural languages, once we take a more 'procedural' view of the latter, as will be explained in Section 3.

LINGUISTIC UNIVERSALS IN LOGICAL SEMANTICS

33

12. Note that this issue, like the earlier one of 'denotational constraints', does not even make sense until we have decided upon two systems, one of concrete linguistic expressions and one of abstract possible meanings, which are to be compared. This contrasts shows, for instance, in the interplay between the syntactic categorial component and its accompanying type theoretical semantics in the framework of Categorial Grammar. (Cf. van Benthem 1991.)

13. One counts positions as in the following pictures: 1 1

2

("before")

2

("until")

2 1

("overlap")

Technically, this amounts to enumerating those relations between open intervals of numbers which are invariant for order-automorphisms of the underlying number line. Such an approach is well-known from theories of Measurement. Compare also the in variances mentioned in Note 2.

14. The celebrated Jespersen / Reichenbach schema for linguistic tenses would be another case in point.

15. This fits in with the earlier observation about a non-strictly denotational, but rather computational difference between common nouns and verbs: the latter set up frames of predication, whose ranges are then restricted by the former.

16. This may be understood intuitively as follows. To check, e.g., whether the first-order relation "all PQ" holds in the relevant semantic domain, it suffices to have the automaton inspect all P's (Conservativity!) one by one, seeing if they satisfy also Q. At each stage of this process, we remember only one of two states: 'success so far' (accepting) or 'at least one failure' (rejecting). Once the last object has been inspected, this produces the right outcome. With a higher-order quantifier like "most", however, we have to remember an unlimited finite record of previous successes or failures, since, e.g., early successes may be needed to offset later failures, and vice versa.

17. One possible example is Robin Cooper's claim that all natural languages will be found to contain expressions (viz. pronouns or some such category) which can be both bound, when there is a suitable intra-textual antecedent, and deictic, otherwise. Another example are the left-to-right 'progression principles' found in Barwise & Perry 1983, or Hintikka & Kulas 1983. The game-theoretic paradigm advocated in the latter work has also been used to state various 'strategic principles' of language use that may be relevant

34

JOHAN VAN BENTHEM

18. For instance, choices as to readings with "at least" versus "precisely" already occurred in Note 6. They also emerge in discussions of the validity of Boolean laws. Does "This dish tastes well when prepared with pepper and with salt" imply "This dish tastes well when prepared with pepper" ? (This example is due to Jan Koster.) The answer is positive if the conclusion "with pepper" is taken to list at least one necessary ingredient, but negative if it is taken exhaustively as listing all that is needed.

19. Some suggestions as to computational parametrization for syntax and semantics are coming up, however, in the flexible Categorial Grammar mentioned at various places in Section 3.

References Barwise, Jon, and Robin Cooper: 1981, 'Generalized Quantifiers and Natural Language', Linguistics and Philosophy 4, 159-219. Barwise, John, and John Perry: 1983, Situations and Attitudes, MIT Press, Cambridge, MA. Benthem, Johan van: 1984, 'The Logic of Semantics', in Fred Landman and Frank Veltman (eds.), Varieties of Formal Semantics, Foris, Dordrecht, 55-80. Benthem, Johan van: 1986, Essays in Logical Semantics, Reidel, Dordrecht Benthem, Johan van: 1987, 'Towards a Computational Semantics', in P. Gardenfors, ed., Generalized Quantifiers: Linguistic and Logical Approaches, Reidel, Dordrecht, 31-71. Benthem, Johan van: 1988, 'Vragen om Typen', Glot 10, 333-352. Benthem, Johan van: 1989a, 'Language in Action', Report LP-89-04, Institute for Language, Logic and Information, University of Amsterdam. (To appear in Journal of Philosophical Logic). Benthem, Johan van: 1989b, 'Semantic Parallels in Natural Language and Computation', in H-D Ebbinghaus et al. (eds.), 1989, Logic Colloquium. Granada 1987, NorthHolland, Amsterdam, 331-375. Benthem, Johan van: 1989c, 'Logical Constants Across Varying Types', Notre Dame Journal of Formal Logic 30, 315-342. Benthem, Johan van: 1990, 'Categorial Grammar and Type Theory', Journal of Philosophical Logic 19, 115-168. Benthem, Johan van: 1991, Language in Action. Categories, Lambdas and Dynamic Logic, North-Holland/Elsevier, Amsterdam.

LINGUISTIC UNIVERSALS IN LOGICAL SEMANTICS

35

Buszkowski, Wojciech, Witold Marciszewski and Johan van Benthem (eds.): 1988, Categorial Grammar, John Benjamins, Amsterdam and Philadelphia. Guenthner, Franz: 1989, 'Discourse: Understanding in Context', in H. Schnelle and N. O. Bernsen (eds.), 1989, Logic and Linguistics, (Research Directions in Cognitive Science: European Perspectives), Lawrence Erlbaum Associates, Hove and London, 127-142. Hintikka, Jaakko, and Jack Kulas: 1983, The Game of Language, Reidel, Dordrecht. Kamp, Hans: 1984, 'A Theory of Truth and Semantic Representation', in J. Groenendijk, T. Janssen, and M. Stokhof (eds.), Truth, Interpretation and

Information,

Foris, Dordrecht, 1-41. Keenan, Edward L.: 1987, 'Semantic Case Theory', in J. Groenendijk, M. Stokhof, and F. Veltman (eds.), Proceedings of the Sixth Amsterdam Colloquium, Institute for Language, Logic and Information, University of Amsterdam, 109-132. Also in R. Bartsch, P. van Emde Boas, and J. van Benthem (eds.), 1990, Semantics

and

Contextual Expression, Foris, Dordrecht, 33-56. Keenan, Edward L., and Leonard Faltz: 1985, Boolean Semantics for Natural Language, Reidel, Dordrecht. Keenan, Edward L., and Jonathan Stavi: 1986, 'A Semantic Characterization of Natural Language Determiners', Linguistics and Philosophy 9, 253-326. Krifka, Manfred: 1989, 'Boolean and Non-Boolean "And"', Seminar für NatürlichSprachliche Systeme, Universität Tübingen. Link, Godehard: 1984, 'Hydras: On the Logic of Relative Clause Constructions with Multiple Heads', in Fred Landman and Frank Veltman (eds.), Varieties of Formal Semantics, Foris, Dordrecht, 245-257. Löbner, Sebastian: 1987, 'Quantification as a Major Module of Natural Language Semantics', in J. Groenendijk, D. de Jongh and M. Stokhof (eds.), Studies in the Theory of Generalized Quantifiers and Discourse Representation, Foris, Dordrecht, 53-85. McCarthy, John: 1980, 'Circumscription - A Form of Non Monotonie Reasoning', Artificial Intelligence 13,295-323. Meulen, Alice ter: 1988, 'The Semantic Properties of English Aspectual Verbs', NELS 21, U. of Massachusetts, Amherst. Mey, Sjaak de: 1990, Dyadic Quantifiers, Plurals and Pronominalization, forthcoming dissertation, Instituut voor Algemene Taalwetenschap, Rijksuniversiteit Groningen. Montague, Richard: 1974, Formal Philosophy, (R. Thomason, ed.), Yale University Press, New Haven.

36

JOHAN VAN BENTHEM

Moortgat, Michael: 1988, Categorial Investigations. Logical and Linguistic Aspects of the Lambek Calculus, Foris, Dordrecht. Oehrle, Richard, Emmon Bach, and Deirdre Wheeler (eds.): 1988, Categorial Grammars and Natural Language Structures, Reidel, Dordrecht. Oversteegen, Eleonore: 1989, Tracking Time, dissertation, Institute for Language, Logic and Information, University of Amsterdam / Nederlands Instituut, Rijksuniversiteit Utrecht. Sánchez Valencia, Victor: 1989, 'Peirce's Prepositional Logic: From Algebra to Graphs', report LP-89-08, Institute for Language, Logic and Information, University of Amsterdam. Sánchez Valencia, Victor: 1990, Studies on Natural Logic and Categorial Grammar, dissertation, Institute for Language, Logic and Information, University of Amsterdam. Thijsse, Elias: 1983, 'On some Proposed Universals of Natural Language', in A. ter Meulen, ed., Studies in Model-Theoretic Semantics, Foris, Dordrecht, 19-36. Westerstâhl, Dag: 1985, 'Determiners and Context Sets', in J. van Benthem and A. ter Meulen (eds.), Generalized Quantifiers in Natural Language, Foris, Dordrecht, 4571. Zwarts, Frans: 1986, Categoriale Grammatica en Algebraïsche Semantiek, dissertation, Nederlands Instituut, Rijksuniversiteit Groningen. (To appear with Kluwer, Dordrecht).

Is Semantics Universal, Or Isn't It ? On the Relation of Language, Thought and Semantic Structure Manfred Immler Projekt METAL, SNIAP 312 Siemens Nixdorf Informationssysteme AG MchP/LZ Carl-Wery-Str. 22 D-8000 München 83 1. The Apparent Universality of Semantics 1 Research in linguistic universals, i.e., the quest for features which can be found in all human languages, has in the last years become a central concern of linguistics (again); presumably in the hope to get closer, in this way, to an understanding of the fundamental nature of human language. But really universal linguistic universals (in the sense of Greenberg/Osgood/Jenkins's "unrestricted universals" and "universal implications" (Greenberg 1961UL, 258)) seem to be very rare, and hard to find. Even a plausible and seemingly obvious generalisation such as: "in every human language there is at least one nasal consonant" apparently does not hold for Quileute and some other Salishan languages (cf. Greenberg 1961UL, 259 and 44). But other presumed universals also do not seem to stand on very solid ground, such as: [ phonology: ] "If in a given language there is only one PNC [ primary nasal consonant ] it is /n/" (Ferguson 1961, 56) "No language has NVs [ nasal vowels ] unless it has one or more PNCs." (I.e., 58) "Every language has syllables of the form CV (consonant followed by vowel)" (Greenberg 1961UL, xxv) - whereas there seem to be languages which do not have syllables of the form VC. "There are gaps, asymmetries or 'configurational pressures' in every phonological system, no matter when examined." (Hockett 1961,26) [ grammar: ] "All languages have two distinct types of stems which ... can appropriately be labelled nouns and verbs." (Hockett 1961,4; cf. also Hopper/Thompson 1983) "In declarative sentences with nominal subject and object, the only or neutral (unmarked) order is almost always one in which the subject precedes the object." (Jakobson 1961, 269; cf. Greenberg 1961UG, 77 - but cf. Sasse 1981 for the problematic character of the categories 'subject' and 'object' presupposed here.) "When the general rule is that the descriptive adjective follows [ the noun ], there may be a minority of adjectives which usually precede, but when the general rule is that descriptive adjectives precede, there are no exceptions." (Greenberg 1961UG, 77) "If either the subject or object noun agrees with the verb in gender, then the adjective always agrees with the noun in gender." (I.e., 93)

38

MANFRED IMMLER

"If a language has the category of gender, it always has the category of number." (l.c., 95) [ semantics: ] "The phenomenon of depletion is surely a semiotic universal." (Weinreich 1961, 181) "Vico did not hesitate to regard anthropomorphic metaphor [ i.e., the derivation of expressions referring to inanimate objects from expressions denoting the human body and its parts - M.I. ] as a linguistic universal." (Ullmann 1961,241) A language which contains a word (lexical unit, lexeme ) for 'blue' will always contain a word for 'red' but not necessarily vice versa, (cf. Berlin & Kay 1969, 2-4 etc.) For, to all of these assertions the following statement applies: these universals "[have] a certain fragility. We would be quite astonished if someone discovered a language which did not have this [property], but we cannot give any reason why this should not be found" (Greenberg, Osgood and Jenkins [ concerning the universal syllable structure 'CV' ] in Greenberg 1961UL, xxv - emphasis mine): because (or as long as) we do not know the reason for any one of these universals we can not exclude with certainty the possibility that an exception to it could still be found - as Bloomfield expressed it in this passage which was quoted by several participants of the Dobbs Ferry Conference of 1961 (Greenberg 1961UL): "The only useful generalizations about language are inductive generalizations. Features which we think ought to be universal may be absent from the very next language that becomes accessible." (Bloomfield 1933/1934,20) And yet Bloomfield was wrong here: for we could begin and name actually hundreds of valid linguistic universals as soon as we turn to the domain of semantic universals (more precisely: denotive universals - see below, p. 50). In this respect, it is revealing to see how Hockett himself - certainly a strict exponent of the principle of inductivity expressed by Bloomfield - introduces a semantic universal which he wants to propose: "with reasonable confidence", he says, we may establish the following universal: "Among the deictic elements of every human language is one that denotes the speaker and one that denotes the addressee." (Hockett 1961,21) But how can this "reasonable confidence" be motivated, where does it come from ? The same undoubting - and in fact justified (see below) ! - confidence that certain semantic properties will be discovered in the majority of, or even in all the languages of the world is also expressed, e.g., in the fact, that the three contributions in Bach/Harms (1968), "Universals in Linguistic Theory", which deal with syntactico-semantic phenomena - Bach, Fillmore, McCawley -, essentially analyse only one language, English (only in Fillmore's article other languages play a certain role), and that Bierwisch

IS SEMANTICS UNIVERSAL, OR ISNT IT ?

39

(1967) derives his proposal for some elements of a "universal set of semantic markers" (l.c., 34) solely (!) from the analysis of German examples. Or consider the following proposals for semantic universals: (1)

- "In all languages a combination of signs [ more exactly: of their meanings - M.I.] takes the form of either linking or nesting, and all languages use both patterns in kernel sentences." (Weinreich 1961,167 - for more details, please, see there)

(1') - universal components of word meanings are, among others: "'generation', 'sex', 'light' vs. 'dark', 'dry' vs. 'wet', 'young' vs. 'old', 'alive' vs. 'dead', 'incipiency' vs. 'steady state'" (I.e., 188) (1") - in every language of the world there are names, designations for the following objects, facts and events (but see the paragraph about 'cognitive' vs. 'semantic' universals below, p. 50): "rustle, soil, [many animals], [many plants], [parts of the body], sleep, big, small, heavy, light, fast, slow, sick, talk, call, ask, believe, decide, birth, wave, up, down; hunger, life, death, danger, fear, want/will, power/authority, be allowed, be obliged, mother, man, woman, caress, high, deep, warm, cold, air, water, rain/snow, wind, sun, pain, pleasure, 'we', 'they', group, drink, shelter, make love." (cf. Immler 1974,41) (1'") - "all languages have some metaphorically transferred meanings" (Greenberg 1961UL, xxii). The interesting thing here is that we are immediately convinced of the validity of these universals, not only so: we are sure of them - and this without having verified them by empirically looking at all the languages of the world, which would, of course, never be achievable: only compare the number of languages covered in some big comparative or typological studies - 83 in Lehmann (1984), 105 in Comrie (1981), 132 in Greenberg (1961UG) and Bechert (1976) or, in an extremly restricted domain, 474 (Jakobson 1960 and Murdock 1959) - with the 6170 living languages which there are in the world (according to Ethnologue 1988,740). Cf. also Koschmieder 1959,209: "Statements ... such as ...: in every language there must exist nouns are not verifiable empirically, inductively. For an empirical proof completeness of evidence would be indispensable. But this can be attained neither practically nor theoretically." How can this "reasonable confidence" be motivated, then? These generalizations and universals ean noi be - there is no way of avoiding this conclusion - results obtained by induction; rather, our knowledge in these cases stems from another source: as soon as we know the reason for the necessary presence of a feature in human language, it follows automatically that this feature must be present in all languages of the world; we then know that it must be a linguistic universal. Now we know, for instance, that "we can come to communicate via language about anything that we are capable of experiencing" (this, too, is a valid generalization about all human language (Hockett 1961, 13 -

40

MANFRED IMMLER

emphasis mine) which could not possibly have been obtained inductively) - we know, in other words, that every human language is complete (not in a static, but in a dynamic sense) relatively to the contents to be expressed in it; and from this knowledge we can then derive the validity of semantic universals such as those listed in (l1) and (1"). We therefore are forced to give up the principle that induction is the only possible or legitimate source of generalization within the science of language. In the above cases we find ourselves quasi on another "level" of knowledge: universals like these as a matter of fact are explications of our (possibly hitherto unreflected) knowledge about human language: we know that every language needs the means to express the ideas mentioned above, because we know about the central role of language for all human society; we know (by extrapolation from our implicit knowledge about our own language) that no human society could afford not to be able to express these ideas in its language. This knowledge therefore is a (more or less conscious) generalization about our (implicit) knowledge of ourselves and our own language. The reason and background for this generalization is, so we might say, a 'natural' conviction of the "original unity of [human] mind" (Ast 1808, quoted in: Vermeer 1975,10). There is a relatively widespread tendency in linguistics today to think that what is a linguistic universal must also be innate: but this is not necessarily so: if we have found a linguistic feature present in all languages of the world it may still be that this feature is universal because it is needed, and therefore created, by men in all cultures. This is the case, to my conviction, for most or even all of the features named above in (l1) and (1").

2. Semantics Is Not

Universal 3

In the same way, a West European or North American linguist will tend to take for granted that in all the languages of the world there will be conjunctions like 'and1, and that in all languages the semantic structure of the sentence will be segmented into a) things, 'entities' about which something is said, and b) that which is said about them, cf.: "and thus every proposition necessarily embodies two terms: the first is called the subject and is that of which one predicates... and the second is called the predicate and is that which is predicated..." (Arnauld and Lancelot 1676,67); "A Proposition has necessarily two terms, as men [and] mortal, the former of which, or the one spoken of, is called the subject; the latter, or that which is affirmed or denied of the subject, [is called] the predicate." (Boole 1847, 20) "To a large extent, the rules of the base [ which among others include the rule: S -» NP VP ] may be universal." (Chomsky 1965, 141); and others (e.g. Lewis 1971, 203).

IS SEMANTICS UNIVERSAL, OR ISNT IT ?

41

Whorf (1941) had already tried in his analyses of Shawnee and Nootka to show that this is not cogent (cf. also the re-analysis proposed in Weinreich 1961,133f.), but many may have refused to believe him, simply because of the far-reaching conclusions which he derived from his observations, which seemed (as they in fact are - see below, 4.1.) by far exaggerated and therefore inacceptable. Modern semantic and logical analysis of language heavily rests on the assumptions that a) the Boolean operator "A" (the syntax and semantics of which is formed after the natural language conjunctions "and/und/et/e/y" etc.) and b) the analysis of a proposition as a "predication" over one or several "argument(s)", is an essential part of any formal semantic description of natural language. But H.-J. Sasse and D. Gil (both in this volume) have again questioned these two assumptions - without the ideological background of Whorf - and asked whether certain linguistic structures which we tend to take as universal, can justly be so considered. What they found out immediately confronts us with the question again whether there are "universal (syntacto-)semantic structures" and whether it is justified to describe them in terms of Predicate Calculus and Boolean Algebra. Gil defines conjunctions as expressions that combine with two or more expressions of some syntactic category to yield a superordinate expression of the same syntactic category and concludes from his data that "in Maricopa, there are no expressions satisfying the above condition; there is thus no syntactic category of conjunction." (I.e.) Sasse outlines a typology of sentence constitution which has, alongside with the familiar Aristotelian type (subject and predicate) at least two more basic types, exemplified by Tongan and Cayuga. In Tongan there does not take place (like in English or German) a predication of "laughing" over the argument "I"; rather, the noun "laughing" is the center of the statement, "I" is its attribute (in the form of "my"), and over them existence or 'reality' is predicated. In the semantics of Cayuga there seem to exist no isolable 'entities' or actants at all ("all full words are conjugated" in Cayuga (Sasse 1988)); rather, the action or the fact to be expressed is represented (and conceived of - see below) as a network of interlacing and interacting 'processes'. In my eyes Sasse's and Gil's findings inevitably lead to two important questions or problems: - if semantics is not universal what, then, is universal ? - if Boolean Algebra and Predicate Logic are not the appropriate basis from which to construct the semantic structures of all languages - what is the basis ?

42

MANFRED IMMLER

3. Two Problems Resulting from these Observations 3.1. If Semantics Is Not Universal What, Then, Is Universal ? 4 Looking at the examples quoted from Gil and Sasse, we find that the same fact or event (Sachverhalf), expressed in a language, can and must be expressed by a quite different semantic structure in different languages: Not in all the languages of the world a fact or event to be described is analysed into a) one or several 'entities' (the arguments) and b) something which is said about it or them (the predicate); and not in all the languages of the world is there the possibility of joining two statements or constituents simply by means of a word which has the syntactic properties of a conjunction. Comparing sentences of English and Tagalog, or of English and Cayuga, which were translated one into the other, we find that there is no l:l-correspondence between the meanings of the elements6 from which the English sentence is built up, and the meanings of the elements of the Tagalog or Cayuga sentence - while the sentences as a whole are perfect (or maximal) translational equivalences one of the other. This means that the syntacto-semantic structuring of 'messages', of utterances is not universal, is not the same in all the languages of the world: in all the languages of the world the content to be expressed is analysed into (semantic) 'segments' which can then be expressed by words in that language; and the words are then combined into a linear order such that the hearer/reader can derive from it how to construct the sentence meaning from the word meanings - but the specific segmentation of the fact or event into semantic 'elements' is not the same in all languages. Moreover, this structuring seems to be done, in the examples mentioned above, in a fundamentally different way than that known to us from the indoeuropean languages — from which Whorf (1940, 213) had drawn this conclusion: "We dissect nature along lines laid down by our native languages. The categories and types that we isolate from the world of phenomena we do not find there because they stare every observer in the face; on the contrary, the world is presented in a kaleidoscopic flux of impressions which has to be organized by our minds - and this means largely i>y thg linguistic systems in our minds" (emphasis mine, M.I.). But, as I will try to show below, this conclusion goes much too far. What is wrong here is that Whorf claims the leading role for 'language'. The more neutral description which Weinreich gives of the same fact is therefore more exact: "The semantic mapping of the universe by a language is, in general, arbitrary, and the semantic 'map' of each language is different from those of all other languages." (Weinreich 1961, 142)

IS SEMANTICS UNIVERSAL, OR ISNT IT ?

43

If we now understand by the semantics of a language the specific association of the words of this language with elemental units of 'content' into which every thought must be analysed in order to be expressed in that language (and, as a consequence, also the specific mapping of the phenomena in the outer world unto the words of a given language), then we must say that the semantics of the languages of the world is not universal but that probably every language has its specific semantics. But if semantics, if semantic structure is not universal - what is universal, then ?

3.2. If Boolean Algebra and Predicate Logic are not the Appropriate Basis from Which to Construct the Semantic Structures of All Languages What Is the Basis? Our second problem is closely linked to the first one; Gil puts it very clearly: "Conjunctions are a central and indispensable feature of Predicate Calculus, the language of logical form espoused by generative grammarians, and of Boolean Algebra, the basis of formal Categorial Semantics." (Gil 1988). In other words, logical conjunctions (and 'predicate-arguments'-structure) form the very basis of practically all modem attempts to grasp and to formalize the semantic properties of natural language and to establish a universal semantics for all natural languages: "Hence the absence of a syntactic category of conjunction in Maricopa" - and the non-universality of 'predicate-arguments'-structure - "calls into question the adequacy of Predicate Calculus and of Boolean Algebra as universal systems of representation for natural language semantics." (Gil 1988). This problem which I will call 'Gil's dilemma' leaves us with the choice between the following two alternatives (cf. Gil 1988): a. Maintaining the position of the universality of Predicate Calculus and Boolean Algebra: this would, however, involve enormous problems for the plausibility of this approach: it would require to assume, for languages which do not have the syntactic category of conjunctions, a semantic 'deep structure' or logical representation which contains a logical AND, and for languages which do not have a 'predicate-arguments'-structure, an underlying semantic representation which has one predicate and several participants or actants. More specifically this would mean deriving, for instance, a Maricopa sentence with the semantic structure of (2) from an underlying semantic structure such as (3): (2) (John accompanying Bill-acc) will-pl come (3) COME(x) A COME(y);

44

MANFRED IMMLER

likewise, it would mean assigning to a Cayuga-sentence with the semantic structure (4) an underlying semantic representation somewhat like the following - (5): (4) him who man-s it strikes an event of losing it which wallet-s him (5) LOSE(ix(MAN(x)), iy(WALLET(y))) A PROPERTY(y.x). ("That individual x which is a man loses that individual y which is a wallet, and y belongs to x.") But now everybody would ask, with reason: why should exactly this 'eurocentric' way of expressing logical relations between the elements of a sentence be the universal basis for the semantics of all the languages in the world? And above all, this approach would force us to hypothesize very strange transformational processes which will be very hard to justify (hard to justify within themselves, but also with reference to other, related, syntactic phenomena) and which, e.g., will have to make disappear, on the way from the underlying to the surface representation of such sentences, logical operators like 'A' and put in their place a verb with approximately the syntax of "accompany" (together with the necessary syntactical modifications!). b. The second alternative is giving up the claim to universality of those formal theories of semantics which are based on occidental logics - which according to Gil could take the form of a "Universal Semantics" that allows of a parametrisation such that some languages could select a semantic system with conjunctions and others one without conjunctions: "This alternative would of course entail a farreaching overhaul of natural language semantic theories." (I.e., 143) By an unprejudiced consideration of the examples (2/3) and (4/5) it seems evident to me that the alternative (a) can not be a defendable option, a possible way to proceed. The fundamental objection here is that it seems impossible to specify a reasonable, sufficiently motivated, non-ad-hoc set of rules which would convert a structure like (3) into a structure like (2), etc. Therefore, the second question to ask is: If Predicate Calculus and Boolean Algebra manifestly are not suitable as a universal instrument for the semantic description of all human languages - what, then, can be the common level of representation, the universal semantic basis from which the sentences and utterances of all the languages in the world can be 'generated1, 'derived'? What would the semantic structure required by the alternative (b.) have to look like, or: What is the Form of 'Logical Form' (or of Semantic Representation)? In what follows I want to try to indicate the possible direction (and no more than a direction) in which such a "far-reaching overhaul of natural language semantic theorfy]" might be looked for. Due to limitations of space, there are many things which I can only touch upon here but which will need much more further discussion and elaboration. The present article can therefore only be a programme, and I will have to presuppose the following statements and results here (cf. Immler 1987 and Immler, in preparation): 1. it can easily be proven that meanings are "in the head [of the speaker]" (contrarily to Putnam's assertion (1975, 223-227), and to Sinha's (1989, 5)

IS SEMANTICS UNIVERSAL, OR ISNT IT ?

45

conviction that " 'conceptualism' and 'psychologism' ... [have] been powerfully criticized by writers ... as Frege and Wittgenstein"): the meaning of a word is a particular "mental" or "psychic content" of a certain form, which - through the process of language acquisition - is (relatively) constantly attached to this word and which is evoked in us, 'called into our consciousness1 every time we hear or think this word, cf. also: "The students of this period [ i.e., of the first phase of the history of semantics which Antal calls the "age of ingenuousness or doubtlessness" ] which lasted until the time of L. Bloomfield considered it to be obvious that the meaning of a linguistic expression is something psychic; an idea, a concept, or both." (Antal 1975, vii) "... word meanings must be treated as internalized mental representations" (Jackendoff 1983,109) 2. language serves man in two ways: first, it expresses his thoughts in a physically tangible form which becomes in this way perceptible to the senses; second, it enables him to refer to the objects, persons, facts and events in the outer world and psychic events within him and to make assertions, requests etc. about them; this always happens by first thinking these objects (etc.), by first grasping them mentally: when we refer to the things of real world we do this 'through the medium of our thinking' - "verba significant res mediantibus conceptibus": words denote things by signifying concepts which in turn refer to these things by characterizing and thus identifying them: "Spoken words are the symbols of mental experiences (affections or impressions - orig.: pathematSn)" (Aristotle, Peri Hermeneias/De Interpretatione, Ch. 1, translation adapted from Ross 1928 and Cooke 1938). "Speaking is explaining one's thoughts by signs which men have invented for this purpose." (Arnauld/Lancelot 1676,41) "The view that words are signs of thoughts represents a dominant theory in western philosophical thought not subject to serious criticism until the twentieth century." (Rieux/Rollin 1975, 66, fn. 4) 3. Meanings are 'out of the same material', are things of the same kind and structure, as thinking. They only differ in form but not by their substance from human thought. It is exactly by their psychic nature (see above) that they are capable of being the bridge, the link between thought and language: the meanings of a language are the conventionalized concepts and conceptualizations available in that language, are 'chunks' of conceptual structure encoded in a word, and therefore are the 'form' into which every thought must be 'moulded' in order to be expressed in that language.

4. Answer to the First Question: Meanings Are Not Universal, but the 'Intentum'

Behind

Them, the Thought

Expressed,

Is (May

Be)

Universal Looking at the above examples, we discovered: when we translate a German or English sentence into a very different language such as Maricopa or Cayuga, the resulting sentence will have a semantic structure very different from that of the original sentence; the same contents, the same 'message' is expressed in different languages in very different ways - semantics is not universal: where the one language speaks of several

46

MANFRED IMMLER

'participants' which are linked by some kind of action or relation - 'there is a man and a wallet, and he has lost it' - the other speaks only of several interlacing processes - 'him who man-s it strikes an event of losing it which wallet-s him'; where the one language has logical connectives which can join any two actions or participants - 'John and Bill are coming' - the other does not have this possibility and has to 'resort to' quite heterogeneous syntactic means in order to express the same relation. The examples adduced by Gil and Sasse show, in my eyes irrefutably, that the same fact or event is not only expressed differently, but also structured semanticallv in a different way, in different languages. But does this not mean, then, that it is not the same fact (or event) any more, but 'a different fact' depending on the language in which it is expressed ? And that its structure is only imposed on it by the respective language ? Cf.: "But if every language provides its own set of lexicalized concepts, every language suggests its own categorization and its own interpretation of the world; consequently, every language is indeed a different 'guide to reality' (Sapir 1949:162)." (Wierzbicka 1989,22) Can we at all think something without it being structured by the semantics of our language? Consequently, does this not imply that the semantic representations of 'the same fact' can not only vary considerably from language to language, but can even be incommensurable between the languages? And if meanings are not universal, what is left then that is universal across all languages and cultures - that is shared by all the speakers of all the languages in the world and that might warrant their understanding each other across the boundaries of their languages? And does this not mean, then, that the speakers of very different languages have no chance of really understanding what the other wants to say, or only at the condition of learning the other one's language? Must we not reject, then, the "basic maxim in linguistics ... that anything can be expressed in any language" (Lenneberg 1953, 467, quoted in Wierzbicka 1989, 23) and say instead that "what can be said (...) may be quite different from one language-culture system to another" (Grace 1987, quoted in Wierzbicka, I.e.) ? Like Wierzbicka (I.e., 24), but for reasons quite different from hers, I do not think that we have to draw this conclusion: Wierzbicka argues that the universal "alphabet of human thought" (the 'primitive', basic conceptual elements from which all the meanings in all the languages of the world are formed) postulated by her is the "common measure" which will allow communication and understanding between (the speakers of) all languages in the world, whereas I think that this "common measure" must be looked for elsewhere: in the last instance in the real world outside of us, which surrounds us and which is the object of our thinking, and which is common to all of us.

IS SEMANTICS UNIVERSAL, OR ISNT IT ?

47

The answer to the present question lies in the somewhat paradoxical fact that language can 'convey' more than ii 'expresses': Look at the following example from Borst/Motsch 1986,169: a sentence like (6) Ich werde alle nach Hause bringen [ I will take them all home ] "can be uttered in very different situations and represent different thoughts." (I.e.) In the thoughts expressed by (6) - the person referred to by "I" varies depending on the speaker of the utterance - "will" can express, can in the concrete case 'mean' a promise, a challenge, a threat, and much more - "all" can refer to guests, chairs, potatoes, children etc. - "home" can refer a) to my home, b) to the respective homes of the different guests at a party, c) to Europe (when seen from Africa), etc. - "bring" can refer to: transporting on foot, by car, by train, by plane, etc. The utterance (6) has a clearly seizable semantic meaning, i.e., it 'tells you' something very clearly (it conveys a 'message1, a knowledge) independently of any information about who the speaker is, in what place and in what situation he is, etc. On the other hand, within a concrete context and situation the words we use - such as (6) - will 'mean' (= convey) much more than simply their meaning: the thought conveyed or communicated by an utterance includes much more, it is richer and more differentiated than its (semantic) meaning, and it can vary widely depending on the context and the situation in which it is produced. With Bierwisch (1978) we might call this latter the 'utterance meaning' but I prefer Koschmieder's (1965, 212-224, 101-115 and passim, esp. also 72-77) term 'Internum' (i.e., 'that which is meant', that which the speaker knows or thinks or imagines and which he wants to communicate or 'means') to stress the fact that the 'Internum' is more than just another kind of meaning. Hjelmslev (1943/1953, 50) had called it "the purport, the thought itself or the "content-purport" (I.e., 56). (In English we would have to use the expressions 'meaning' and 'that which is meant', whereas in German this is easier to distinguish terminologically by the terms "Bedeutung" and "Gemeintes"). The meaning of an utterance is what we know when we know nothing about the circumstances in which the sentence is uttered; it is determined by its words and its syntactic structure alone: for a word (as a linguistic unit) can convey only the 'linguistic', generalized knowledge that is learnt with it through the process of language acquisition. Meanings therefore are the same every time we hear the word.

48

MANFRED IMMLER

The 'Internum', i.e., that which the speaker wants to communicate and which the hearer can infer from 'sentence + context + situation' is more and richer than sentence meaning alone: "In general it may be said that a linguistic system is able to convey, in the semantic value of a sign, only an incredibly small number of the cognitive potentialities of the mind and of the unnumerable and varied moments of our affective life." (Pagliaro, Afntonino] 1952, II segno vivente. - Napoli: [ Edizioni Scientifiche Italiane ], 310, quoted from DeMauro 1967, 55) This distinction between meaning and 'Intentum' is exactly reflected by Lyons's (1977, 29ff.) distinction between 'text-sentences' and 'system-sentences' - where systemsentences have semantic meaning only while text-sentences (i.e., really occurring utterances) have semantic meaning and an Intentum. It is therefore important on the one hand to stress the proximity and close relationship between meaning and cognitive structure (as Cognitive Grammar does); but it would be wrong to identify them, as some authors of Cognitive Grammar seem to do, cf.: "semantic structure is conceptual structure (i.e., meaning equals conceptualization)" (Smith 1989, 1). It is, consequently, correct to say that "language conveys meaning" (Wierzbicka 1989, 3), but there is more to it: through and above the meanings it conveys a knowledge about the Intentum, i.e., about what he knows or thinks or imagines and what he wants to communicate; and furthermore, through and via the 'Intentum'. we can refer to the objects and events in the outer world or within us about which we want to speak. To sum up: the semantic structure of language may not be universal, but this does not mean that there can be no 'common third' and no understanding between the speakers of very different languages: there is another 'level' of communication different from that of meanings: and on this level of the 'Intentum' the thoughts and ideas of a person may well be impartable across very different languages and cultures, and therefore may be universal: Just as someone can succeed in communicating, in the concrete situation, what he wants to say (and much more than the pure words 'express') - e.g. that the concrete person x ("I") promises to collect all the persons who have been at the party and have drunk too much, and that he will take them each to his/her own home -: by the same transition from meaning to the 'Intentum' it is possible to convey to speakers of very remote languages what one wants to say in one's own language. Consequendy, all the things which are not expressed in some langage (such as: natural gender, or time, or Aktionsart, or defmiteness - cf. fn. 1 in van Benthem 1991), can nevertheless be conveyed through that language:

IS SEMANTICS UNIVERSAL, OR ISNT IT ?

49

What one can convey through language is not limited to the meanings available in that language. A language is therefore not a limit, a barrier to what one can communicate in it (or better: through it). It is certainly the case that some things are easier to express in one language and only very clumsily in some other - but as meaning, semantics is not the only and ultimate level on which communication and understanding takes place, it is indeed possible to "express (better: to communicate) anything in any language". On the basis of this distinction - which is not a separation! - between meaning and thought/cognitive structure, I want to say that cognitive contents can i>£ universal across all cultures and all languages - they will be universal

Jhs extent that the real world

around us which is the same for all humans all over the world is the object of cognition (see below, 5.2.1); as Aristotle already said it with admirable clarity and simplicity: "Just as all men have not the same writing, so all men have not the same speech sounds, but the mental experiences (affections - orig.: pathemata), which these directly symbolize, are the same for all, as also are those things of which our experiences (affections) are the images (representations or likenesses, images, copies - orig.: homoiSmata)." (Aristotle, Peri Hermeneias/De Interpretatione, Ch. 1, translation adapted from Ross 1928 and Cooke 1938) Of course, this is not all that simple, cf. Antal (1963, 85 - emphasis mine): "Although the world of meanings is different and theoretically incomparable and incommensurable from one nation to another, the denotata amongst which people live are, in fact, the same for all humanity or, at least, for a great number of people." -: for, to the extent that people in other cultures make experiences different from ours, due to a different material environment or different social or personal experiences (living in an archipelago or in an environment of perpetual ice and snow; social structures such as castes, marital structures such as polygyny; initiation rites, voodoo, shamanism) - to that extent the 'things' around them and their cognitive structures and contents are not 'the same1. But on the other hand we should not underestimate how much in common humans in all cultures have - cf. the examples in (1') and (1") above. Semantic structures, meanings, may be - and apparently are - language-specific and language-dependent ~ because they exist only as attached to the words of which they are the meanings. But what at least can be common to all humans and all languages, is a) certain mental contents, affections and experiences, and b) - above all - the world outside which often enough is the object of our thinking and speaking. The 'Intentum' need not i e structured by the language in which it is going to be expressed. So, in Gil's (1991) terms, we do not have to be either "universalists" or "relativists" rather, the correct answer would be "relativist" on the level of semantics, and "universa-

50

MANFRED IMMLER

list" on the level of cognitive structure: cognitive structures are (can be) universal, but this cognitive 'substrate' or 'field' is structured differently by the semantics of different languages. Speaking means moulding a thought which as such can be independent of the language in which it shall be expressed, into semantic 'elements' = the conventionalized meanings available in this language, and then mapping these meanings into the 'words' (lexemes) which express them in this language. Understanding means, first, associating the words perceived with their (semantic) meaning and, second, inferring from linguistic meaning + context + situation that which is 'meant', the 'Intentum'. Only this distinction - which can never be a separation - between meaning and 'Intentum' can explain how one can speak in French with its relatively generic, unspecific terms like 'mettre', 'aller* about the same things and express the same thoughts as in German with its much more specific terms 'setzen, stellen, legen, hangen' and 'gehen, reiten, fahren' (cf. Ullmann 1959, 143). Also this feature of human language formulated by Wittgenstein (1918/1922): "4.027 4.03

It belongs to the essence of a proposition that it is able to communicate to us a sense which is new. A proposition must use old expressions to communicate a new sense."

can only be explained by the distinction between meanings and 'Intentum': the meanings are that which is 'old', already known, and the 'Intentum' is that which may be new. And it is this 'step' between meaning and 'Intentum' which explains the possibility of understanding between structurally very different languages: if we activate all our knowledge about the world in which our interlocutor lives, we have a chance of reconstructing through his words and their meanings 'that which he means'. Going back to the (substantive) 'semantic' universals listed above in (1') and (1") we will then say that they are cognitive and/or denoting rather than semantic universals. The form of a semantic universal would be: "in every language of the world there is a word with the meaning 'rustle'" - of which, of course, we can not be sure; whereas a cognitive and/or denoting universal is: "in every language there must exist an expression (or several) which may denote, refer to the event type 'rustle'" - of which we can be sure (see above, section 1). In other words: many authors who seem to investigate universal semantic features, in reality analyse cognitive distinctions so fundamental that we may safely assume that they will be expressible in all languages of the world. I therefore think that A. Wierzbicka (I.e., 31) unjustly criticizes Western psychologists and anthropologists for applying "concepts such as 'mind', 'anger', 'fear' or 'depression'" to foreign cultures; it may be that these cultures do not have words which have exactly the intension and extension of our words 'mind', 'anger', 'fear' and

IS SEMANTICS UNIVERSAL, OR ISNT IT ?

51

'depression', but they sure will have the corresponding concepts - for the simple reason that they have also experienced feelings like these.

4.1. What Can we Derive from this for the

Humboldt-Sapir-Whorf

Hypothesis ? Schaff (1964) and Gipper (1972) as well as others before them proposed and attempted to test empirically, experimentally the 'Humboldt-Sapir-Whorf hypothesis' (= the assumption that language by its structure determines the way in which we see the world). These are, however, unnecessary pains. For, whoever subscribes to the HumboldtSapir-Whorf hypothesis will want to gain insight into the mechanism by which a language could determine thinking, and already here he will encounter unsurmountable difficulties: Whorf (1940, 213f.) says: "We cut nature up, organize it into concepts, and ascribe significances as we do, largely because we are parties to an agreement to organize it in this way - an agreement that holds throughout our speech community and is codified in the patterns of our language. The agreement is, of course, an implicit and unstated one, BUT ITS TERMS ARE ABSOLUTELY OBLIGATORY; we cannot talk at all except by subscribing to the organization and classification of data which the agreement decrees." But we must analyse this contention in detail: how could language do what Whorf says it does? How should language be able to force us to think the way it prescribes to us and no other? There are two possibilities: - either we think in words - which we don't, see the next section, - stereo(phonic) sound Gr. 'a-tomos' = indivisible --> the 'atom' is now known to be divisible Gr. 'scholé' = leisure —> school Latin 'caudex' = trunk of a tree --> book --> writing --> (language as a) 'code' And in practice this is exactly what happens: when we want to express a new idea or a concept we have newly developped we endow a word which already exists in our language with this new meaning, cf.: 'shuttle' (of a weaver) --> shuttle (train) —> space shuttle. The above result already renders unnecessary and even undefendable the assumption that language by its structure could determine the way in which we see the world. Whoever adheres to the Humboldt-Sapir-Whorf hypothesis attributes to language a power, an ability which it does not have: namely the ability to fix the meaning of a word in such a way that this meaning could never change any more. Yet the word (lexeme) has nothing in it which would allow it to fix its meaning, to prevent its meaning from taking any value whatsoever. This is a natural consequence of the fundamental law of the 'arbitrary of the linguistic sign'. Therefore the Humboldt-Sapir-Whorf hypothesis in its strong form simply can not hold. On the other hand: A thought will always be expressed according to the lexical and syntactical meanings available in the language in question; and therefore certain differentiations will be easy to express in one language and only clumsy in another. Even more: many concepts - i.e., 'forms of thought', forms into which we mould or thought - we learn during language acquisition and through its mediation: this means, we acquire these concepts in the form in which our culture uses them (and also with all the biases with which our culture may use them). This also means that from the start we will tend to see the world around us "with the eyes" of our culture, in terms of the concepts which our culture provides us. As Gil (1991) made clear with regard to honorifics and the expression of distributivity in different languages: semantic categories which are linguistically realized in a language,

IS SEMANTICS UNIVERSAL, OR ISNT IT ?

53

are much more prominent and, so to say, offer themselves to thought more easily, are more al hand when the speaker sets about mentally 'analyzing' - analyzing into elemental ideas - a fact about which he wants to think. Seen from a remote language, we can then discover these concepts in our own language, too, but they are not prominent here. For this reason, I do in fact think that, for instance, "the fact that the Chinese language, unlike English, commonly uses distinct terms to distinguish 'if-then' relationships from 'if-and-only-if-then' relationships" will "imply that Chinese students are, on that score ..., better fitted for first year logic than their American counterparts" (Bloom 1981, quoted in Bechert 1991) So, I think that the meanings available in our language will not only influence the way something will be expressed but to a certain degree also the concepts in which it is thought. However, we must not forget that the object of our thinking very often is the real world: the world around us and the world within us. And as it is very important for us to understand and to see the world (as nearly as possible) as it is - because only then we can act successfully on it

the contents and the structure of our thinking is to a great extent

simply determined by the properties and the structure of the real world (see below). I would therefore like to say that 'at the start' our thinking may be determined or influenced by the language we have learnt; but the more thinking and cognition orients itself by the properties of the object (or fact or event) which is its object, the more it will become independent of the concepts available in the particular language and determined only by the properties of the object I therefore think that we have to assume here a polarity, a straining ('Spannungsverhaltnis') which exists between two tendencies: 1. one is the tendency to mould our thoughts in the forms which are made available by the respective language. 2. the other is the wish to form, as far as possible, a 'correct' idea of something, i.e., a concept the properties and structure of which are as nearly as possible determined by the properties and structure of the real object or part of the world on which our thought is directed. So, my result is: a language may suggest to us a certain view of the world, but this can never be obligatory, we can (to my eyes easily, and we do it all the time) transcend this view suggested to us and acquire a more adequate and more differentiated view of the world.

54

MANFRED IMMLER

5. Conclusion and Outlook For reasons of space only half of the paper prepared for this volume could be published here. In the present paper I have tried to convince my readers of some essential features which, to my conviction, an adequate theory of the relation of language, thought, meaning and (Modern Formal) Logic must present The central points of this conception are these: 1. Meanings can easily be proven to be mental objects: psychic, mental "contents" (cf. Immler 1987). It is by virtue of this property of being "in the mind" that they play the central mediating role between thought and language. 2. Meaning (semantic structure), although closely related to cognitive structure, is not identical with it. Semantic structure is language-specific because it is what is encoded - through the process of language acquisition - in the words (and in the syntactic patterns) of a language; but cognitive structure - the internal cognitive representation of part of the inner world and the outer world - may be universal across all languages and cultures. In the second half the following points are discussed: 3. Thought, as well as meaning, has a structure essentially different from that of language; both are something more of the kind of a mental 'image' than of a language: both are non-linear, and iconic with respect to the things or events they represent. The utilisation of language consists in converting a non-linear and iconic cognitive representation into a linear syntactic, morphemic and phonic structure. 4. Formal logical notation is therefore more something like a language than like thought (or meaning). This severely limits its adequacy for the representation of the 'laws of thought' and of semantic structure.

Notes 1.1 thank Hans-Jiirgen Sasse, Klaus Tekniepe, Christoph Schwarz, Dietmar Zaefferer, Mrs. Eva Kollmar and Prof. Dr. Elmar Seebold for stimulating discussions and helpful comments, Mrs. Eva Kollmar for correcting my English in a preliminary version, and Ulrich Stuck.

IS SEMANTICS UNIVERSAL, OR ISNT IT ?

55

2. For reasons of simplicity I will use throughout this text the term 'word' (in addition to 'lexeme') although it is clear that in reality not words are the smallest meanigful units of a language, but morphs, lexemes and idioms.

3. I will not enter here into the discussion on what semantics is but will simply take "the semantics of a language" to mean "the totality of the meanings expressed by the words and the syntactic devices of that language". And the question "Is Semantics Universal ?" is simply meant to be a way of asking: "are there semantic properties, semantic features, meanings and/or semantic structures shared by all human languages ?"; in its strongest interpretation the position of the universality of semantics would imply that all the meanings found in one of the languages of the world will be present in all other human languages.

4. There is now the (technical) problem that, in order to proceed in my argumentation, I have to anticipate certain distinctions which I will argue for and discuss only later. The point is the following: the above examples suggest, to my eyes convincingly, that {h£ same thought can be expressed by quite different semantic means - can be structured semantically in quite different ways - in different languages. But to formulate this conclusion, I have to suppose the distinction between 'meanings' on the one hand and 'same thought' on the other which I want to introduce only later, on the basis of the above examples. Anyhow, let me suppose a rudimentary conception of meaning here, saying that the meanings of a language are the specific forms ('attached' to the words of that language) into which a thought must be 'moulded' in order to be expressed in that language, and that apparently, as the examples of Gil and Sasse demonstrate, 'the same thought' is mapped, moulded into different semantic forms in different languages.

5. In German, we have the very useful term 'Sachverhalt' to denote anything which is or might be the case in the real world or in fictitious worlds: I will use in this paper the expression 'fact or event' in this very sense because I hope it will convey best the idea I want to express.

6. There is one consideration which I must leave aside here for want of place: while there may be no 1:1correspondence between the meanings of the elements in one language and in another, there might exist a l:l-correspondence of meanings between larger segments, such as syntagms, or sentences as a whole, etc., in different languages. But, as far as I can see, this fact in the last instance will change nothing to the conclusions at which I will arrive in due course.

7. This even holds if we take into consideration the level of sentence meanings: although the number of sentence meanings expressible in a language is much higher than the number of word meanings, also the number of sentence meanings is limited because the number of 'combinatory' meanings available by syntactic devices must of course be finite, too (cf. my discussion of the distinction between 'finiteness'

56

MANFRED IMMLER

vs. 'enumerability' in language, in Immler 1974, llff.). This is exactly the reason why I prefer Koschmieder's term 'Internum' to the term 'utterance meaning' (Bierwisch 1978): the 'Intenta' (the thoughts expressible) are what is really unlimited in language while there will always be only a finite number of possible sentence meanings.

References Antal, Laszlo: 1963, Questions of Meaning, Mouton, The Hague. Antal, Laszlo: 1975, Aspekte der Semantik. Zu ihrer Theorie und Geschichte 1662 1968, Athenäum, Frankfurt. Aristotle: see Cooke 1938 and Ross 1928. Arnauld, Antoine and Claude Lancelot: 1676, Grammaire générale et raisonnée, in Rieux/Rollin 1975. Bach, Emmon, and Robert T. Harms: 1968, Universals in Linguistic Theory, Holt, Rinehart and Winston, London etc.. Bechert, Johannes: 1976, 'Bemerkungen zu Greenbergs 'Basic Order Typology", Papiere zur Linguistik 10 (1976), 49-66. Bechert, Johannes: 1991, 'The Problem of Semantic Incomparability', this volume. Benthem, Johan van: 1991, 'Linguistic Universals in Logical Semantics', this volume. Berlin, Brent, and Paul Kay: 1969, Basic Color Terms. Their Universality and Evolution, University of California Press, Berkeley, Los Angeles. Bierwisch, Manfred: 1967, 'Some Semantic Universals of German Adjectivals', FL 3 (1967), 1-36. Bierwisch, Manfred: 1978, 'Utterance meaning and mental states', in - , Essays in the Psychology of Language. Linguistische Studien, Reihe A, Band 114, Akademie der Wissenschaften der DDR, Zentralinstitut für Sprachwissenschaft 1983, Berlin. Bloomfield, Leonard: 1933/1934, Language, Allen & Unwin, London 1973. Boole, George 1847, The Mathematical Analysis of Logic, Reprint Barnes & Noble, New York 1965. Borst, Dieter, und Wolfgang Mötsch: 1986, 'In welchem Maße sind Sprachen ineinander übersetzbar?', ZS 5 (1986), 167-186. Chomsky, Noam: 1965, Aspects of the Theory of Syntax, M.I.T. Press, Cambridge, Mass. Comrie, Bernard: 1981, Language Universals and Linguistic Typology. Syntax and Morphology, Blackwell, Oxford.

IS SEMANTICS UNIVERSAL, OR ISNT IT ?

57

Cooke, Harold P.: 1938, Aristotle. The Organon I. The Categories. On Interpretation. [Translated] By Harold P. Cooke. Prior Analytics. [Translated] By Hugh Tredennick, Heinemann, London; Harvard University Press, Cambridge, Mass. DeMauro, Tullio: 1967, Ludwig Wittgenstein. His Place in the Development

of

Semantics, Reidel, Dordrecht. Ethnologue: 1988, Languages of the World. Eleventh Edition. Ed. Barbara F. Grimes, Dallas, Texas, Summer Institute of Linguistics. Ferguson, Charles A.: 1961, 'Assumptions about Nasals. A Sample Study in Phonological Universals', in Greenberg 1961UL, 53-60. Gil, David: 1988, Abstract for 'Aristotle Goes to Arizona, And Finds a Language without 'And", Paper read at the Annual Meeting of the Deutsche Gesellschaft für Sprachwissenschaft, 1988, Wuppertal. Gil, David: 1991, 'Aristotle Goes to Arizona, And Finds a Language without 'And", this volume. Gipper, Helmut: 1972, Gibt es ein sprachliches Relativitätsprinzip? - Fischer, Frankfurt. Greenberg, Joseph H.: 1961UG, 'Some Universals of Grammar with Particular Reference to the Order of Meaningful Elements', in Greenberg 1961UL, 73-113. Greenberg, Joseph H. (ed.): 1961UL, Universals of Language, M.I.T. Press, 2nd ed., Cambridge, Mass., London 1966 . Hjelmslev, Louis: 1943/1953, Prolegomena to a Theory of Language, The University of Wisconsin Press, Madison 1963. Hockett, Charles W.: 1961, 'The problem of universals in language', in Greenberg 1961UL, 1-29. Hopper, Paul J., and Sandra A. Thompson: 1983, 'The Iconicity of the Universal Categories 'Noun' and 'Verb", in Haiman 1985, 151-183. Immler, Manfred: 1974, Generative Syntax - Generative Semantik. Darstellung und Kritik, Fink, München. Immler, Manfred: 1987, 'Was ist die Bedeutung ? Zur Grundlegung einer Theorie der Bedeutung', to appear in Proceedings of the XIVth International Congress of Linguists, Berlin, August 10-15,1987, Akademie-Verlag, Berlin. Immler, Manfred: in preparation, Bedeutung, Sprache, Denken. Eine Skizze ihres wechselseitigen Zusammenhanges. Jackendoff, Ray S.: 1983, Semantics and Cognition, M.I.T. Press, Cambridge, Mass. Jakobson, Roman: 1960, 'Why 'Mama' and 'Papa'?', in Jakobson, Roman, Selected Writings I. Phonological Studies, Mouton, The Hague, Paris 1971, 538 - 545. Jakobson, Roman: 1961, 'Implications of language universals for linguistics', in Greenberg 1961UL, 263-278.

58

MANFRED IMMLER

Koschmieder, Erwin: 1959, 'Das Allgemeingültige in der Syntax', in Koschmieder 1965, 209-224. Koschmieder, Erwin: 1965, Beiträge zur allgemeinen Syntax, Winter, Heidelberg. Lehmann, Christian: 1984, Der Relativsatz, Gunter Narr, Tübingen. Lewis, M. M.: 1971, 'The linguistic development of children', in Linguistics at Large. Ed. by Noel Minnis, Victor Gollancz, London. Lyons, John: 1977, Semantics. Volume I, Cambridge University Press, Cambridge etc. Murdock, George Peter: 1959, 'Cross-Language Parallels in Parental Kin Terms', American Anthropologist 1 (1959), 1-5. Putnam, Hilary: 1975, 'The meaning of 'meaning", in —, Mind, Language und Reality. Philosophical Papers. Vol. 2, Cambridge University Press, Cambridge etc.. Rieux, Jacques and Bernhard E. Rollin: 1975, The Port Royal Grammar, Mouton, The Hague, Paris. Ross, W.D.: 1928, The Works of Aristotle. Translated into English under the Editorship of --. Volume I, Clarendon Press, Oxford. Sasse, Hans-Jürgen: 1981, "Basic Word Order' and Functional Sentence Perspective in Boni', FoL XV (1981), 253-290. Sasse, Hans-Jürgen: 1988, 'Prädikation und Satzkonstitution in universeller Sicht', Paper read at the Annual Meeting of the Deutsche Gesellschaft für Sprachwissenschaft, Wuppertal. Sasse, Hans-Jürgen: 1991, 'Predication and sentence constitution in universal perspective', this volume. Schaff, Adam: 1964, Sprache und Erkenntnis, Europa Verlag, Wien etc. Sinha, Chris B.: 1989, 'On Representing and Referring', to appear in Proceedings of the Duisburg Symposium on 'Cognitive Linguistics', March, 1989. Smith, Michael B.: 1989, Handout for his paper read at the Duisburg Symposium on 'Cognitive Linguistics', March, 1989, 'Case as Conceptual Categories: Evidence from German' (cf. also Linguistic Agency University of Duisburg, Series A, Paper No. 260, Duisburg). Ullmann, Stephen: 1959, Précis de Sémantique Française, Francke, Berne. Ullmann, Stephen: 1961, 'Semantic Universals', in Greenberg 1961UL, 217-262. Vermeer, Hans J.: 1975, Allgemeine Sprachwissenschaft. Eine Einführung, Rombach, Freiburg. Weinreich, Uriel M.: 1961, 'On the Semantic Structure of Language', in Greenberg 1961UL, 142-216. Whorf, Benjamin Lee: 1940, 'Science and linguistics', in Whorf (1956), 207-219. Whorf, Benjamin Lee: 1941, 'Language and logic', in Whorf (1956), 233-245.

IS SEMANTICS UNIVERSAL, OR ISNT IT ?

59

Whorf, Benjamin Lee: 1956, Language, Thought and Reality. Selected Writings of --. Edited and with an Introduction by John B. Carroll, The Technology Press of Massachusetts Institute of Technology, [Cambridge, Mass.:]; Wiley, New York; Chapman & Hall, London. Wierzbicka, Anna: 1989, 'The Alphabet of Human Thoughts', Linguistic Agency University of Duisburg, Series A, Paper No. 245, Duisburg. Wittgenstein, Ludwig: 1918/1922, Tractatus logico-philosophicus. Logisch-philosophische Abhandlung, Suhrkamp, Frankfurt 1966.

The Problem of Semantic Incomparability Johannes Bechert Institutfür Allgemeine und Angewandte Sprachwissenschaft Fachbereich 10 Universität Bremen D-2800 Bremen 0. Introduction In this paper I shall argue that languages can be semantically distinct to such a degree that it seems useless to assume semantic invariants underlying their observable diversity, and further, that facts which seem to support our own general semantic conceptions may be due to colonization, i.e., to the influence exerted by our own ways of speaking on the languages that exhibit these facts. In order to show the significance of this contention I think it best to discuss the meanings of grammatical categories and the semantic implications of syntactic structures, function words, etc., because cases of distinctness in the domain of the lexicon, i.e., of content words - though (or because) they are ubiquitous might be dismissed as trivial, anecdotal, or due to language-external factors. In particular, I intend to adduce examples supporting the thesis that wherever cultures are distinct we have to be prepared to face far-reaching semantic differences between the languages associated with the respective cultures, and especially so with regard to grammatical categories, function words, and syntactic structures of those languages.

1. Three Examples Example 1: K. Gr0nbech (1936, pp. 23-24) argues that in Old Turkic no difference existed between substantives and adjectives (translation mine): "The nouns divide into several subclasses since numerals and pronouns are marked off by some formal and syntactic peculiarities. On the other hand, there is no discernible difference whatsoever between substantives and adjectives, neither in formal nor in syntactical respects. The reason for this is that it is not possible to distinguish between them notionally either. The nouns do not denote living beings or things but their concepts to an indeterminate extent, neither delimited substantivally nor adjectivally; this is to say that the concept is mentioned without the decision about whether it is to be embodied as a thing or as a quality. A first corollary from this notion of the noun is that, originally, the differentiation between singular and plural was not expressed linguistically through inflection. The oldest stage of the language lacked plural forms... The fact that all the Turkic languages of today have a plural suffix proves that the notion of the noun has shifted in the course of time, but the absence of the plural suffix in the inscriptions shows that the idea of plurality as an opposite to the individual is an innovation...

THE PROBLEM OF SEMANTIC INCOMPARABDJTY

61

Those words that we call adjectives because they frequently denote a quality do not differ in any respect from the rest of the nouns; when standing in isolation they inflect like any other noun: adgug "your advantage" [with possessive suffix of ¿he 2nd person singular], aqyy "the white (horse)" [accusative]; and nominal compounds such as adgu kisi "good people", aq at "white horse" are formed in the same way as those that have a mass noun denoting a substance or a specification of sex as their first component: tamir qapyy "iron door", qyz o/ul "daughter" [lit. "girl son"]. An adjective such as, e.g., uluy could be defined just as appropriately as a substantive: "he who is big"... Thus, the concept of the noun is so complex that it comprises several of our linguistic categories. A Turkic noun, therefore, is neither a substantive nor an adjective but precisely both at the same time... What the Turk conceives of as genus (species) is expressed by us either as a singular or as a plural; what he imagines as a concept must be rendered by us either as a thing or as a quality,.." This view of the Old Turkic noun has gained general currency; see A. von Gabain (1974), p. 64 (§ 72), pp. 147-148 (§ 328), concerning the plural pp. 84-85 (§§ 168174), p. 147 (§ 327), and O. Pritsak (1982), p. 37 (§ 25), and about the plural p. 38 (§27). Example 2: E. Lewy (1911) discusses coordinate constructions in Hungarian, Ostyak (Xanty) and Votyak (Udmurt). He shows that originally there was no room for the coordinating conjunction "and" in these languages; cp. his summary (pp. 99-101, my translation): "138. A basic tendency of the Finno-Ugric language structure seems to be a striving for sameness of arrangement of successive sentence parts which denote similar things. [Ostyak] jianka, man tam notem varat oumot, tejat oumot, veda-jem "sister, I this arrow-my with-blood flows, with-pus flows, I-am^beingkilled" [text and translation quoted after the wording in §91] .., or: untivet ei tapa vejot, urivet ei tapa vejot "from-the-wood a chip he-took, firom-the-pinegrove a chip he-took".. 139. If it says (Wiedemann, 1884, p.] 105): "The grouping of two verbs together in order to denote a concept for which, in German, we have .. a simple special verb is very widespread [in Zyrian (Komi) and Votyak]" - then it is obvious that the person who wrote these words tended to see a single concept where a single verb of the German language appears as an expression. It is obvious that this single concept has no more validity than the double concept that underlies the Finno-Ugric expression. [Ostyak] segai poxtai"he was struck and kicked" (in English: "he was beaten up")..; [Votyak] um suyimaskei^e, um bordeyje "we would not weep, we would not mourn"., [quoted with the translation from § 100]. 140. The Finno-Ugrian has a strong tendency to single out two points from a conglomerate of perceptions, of a group of impressions, and to express them even where we are hardly or not at all able to find duality ..: cp., e.g., [Hungarian] tatva nyitva talala a ket elso szobat "opened, unlocked he-found the two first rooms" [in English: "wide open", cp. § 135].. and the examples just quoted. He even goes so far in his predilection as to supplement a word - that also occurs independently and expresses part of the conception - with a second, newly created one that further depicts the situation but does not occur independently and therefore cannot readily be called a word at all; he especially likes to do this by means of a labial consonant, just as he likes to make the second word begin with

62

JOHANNES BECHERT

a labial in highlighting two points of the phenomenon by ready-made words. [Examples from §§ 106 and 109: Hungarian amul-bamul "to be greatly astonished", cp. amul "to be astonished", bamul (older form bavul) "to be astonished"; ir-ffr "to compile, to scribble", cp. ir "to write" -fir does not exist as a separate word. The Hungarian verbs are quoted in the 3rd person singular of the present indicative of the indefinite conjugation, a form that exhibits the verbal stem without a suffix.].. 143. It was impossible for me to separate a parallelism of form from a parallelism of content; the Finno-Ugric form is at the same time quite a real content 144. If the same relation or condition of two things or events is sufficiently denoted by the same arrangement of their expressions, then Finno-Ugric does not leave any room at all for a coordinating conjunction .. which briefly expresses a closer connection where equal relations hold. .. If we take the line of the meticulous world-view of the Finno-Ugrians who mainly stick to the facts at hand, the connection by "and" is bound to appear as something utterly superfluous, even as a sort of hubris looking for relations everywhere and expressing them although they often do not exist in reality at all..." That the facts reported here are generally acknowledged can be seen by other descriptions, e.g., A. Laanest (1982, p. 308, translation mine): "In the early period of the development of the Finno-Ugric languages, sentences and phrases were joined to each other without conjunctions. Instead of (E[stonian]) isa ja ema tulevad "father and mother are coming", people used to say isa tuleb, ema tuleb ["father is coming, mother is coming"]. Sentence constructions of this kind are not uncommon in dialectal registers even today ..; in the Finnish and Estonian literary languages, on the other hand, such an expression is mainly considered to be a stylistic device." Example 3: A. H. Bloom (1981, pp. 13-14) introduces the main topic of his book on language and thinking in China by the following opening: "In 1972-1973, while I was in Hong Kong working on the development of a questionnaire designed to measure levels of abstraction in political thinking .., I happened to ask Chinese-speaking subjects questions of the form, "If the Hong Kong government were to pass a law requiring that all citizens born outside of Hong Kong make weekly reports of their activities to the police, how would you react?"; or "If the Hong Kong government had passed such a law, how would you have reacted?". Rather unexpectedly and consistently, subjects responded "But the government hasn't;" "It can't;" or "It won't." I attempted to press them a little by explaining, for instance, that "I know the government hasn't and won't, but let us imagine that it does or did..." Yet such attempts to lead the subjects to reason about tilings that they knew could not be the case only served to frustrate them and tended to give rise to such exclamations as "We don't speak/think that way;" "It's unnatural;" "It's unChinese." Some subjects with substantial exposure to Western languages and culture even branded these questions and the logic they imply as prime examples of "Western thinking." By contrast, American and French subjects, responding to similar questions in their native languages, never seemed to find anything unnatural about them and in fact readily indulged in the counterfactual hypothesizing they were designed to elicit. The unexpected reactions of the Chinese subjects were intriguing, not only because of the cross-cultural cognitive differences they suggested, but also because the Chinese language does not have structures equivalent to those through which English and other Indo-European languages mark the counterfactual realm.

THE PROBLEM OF SEMANTIC INCOMPARABIUTY

63

Could having or not having a counterfactual construction in one's language play a significant role in determining how inclined one will be to think in counterfactual terms?" Bloom gives evidence for statistically significant differences between reactions of Chinese and American subjects to questions implying reasoning in the counterfactual mode. Then he proceeds to investigate the context of this phenomenon (pp. 33-44): "Moreover, the link between language and thought in the counterfactual becomes even more interesting when one considers that it appears to constitute only one aspect of a much more general pattern of language and thought interconnections evident in the psycholinguistic habits of speakers of English and Chinese. In one sense, the counterfactual acts like any ordinary linguistic label might act in encouraging speakers of English to add one more structured perspective to the array of structured perspectives through which they ordinarily make sense out of, code and process information about the world. But in another sense, the counterfactual differs from most other linguistic elements in that it seems to constitute a member of a special set of English and, more generally, IndoEuropean linguistic devices that lead speakers to develop cognitive schemas specifically designed to enable and encourage them to shift from describing, questioning, or even commanding within their baseline models of reality, to projecting and operating with theoretical extractions from those baseline models. And not only does the Chinese language not have any structures equivalent to the counterfactual but neither does it have structures equivalent to the additional members of this special set of English and, more generally, Indo-European elicitors of theoretically extracted thoughts."

2. Summary of the Findings I have quoted the examples at some length with a view to showing their common traits as well as their divergences. The authors of the two older descriptions use the philological method to set off the languages they are exploring against our languages; Gr0nbech and Lewy make interpretations of grammatical phenomena in a metalanguage slightly tinged with philosophical traditions and psychological theories, respectively. The author of the third text is interested in psycholinguistics and cross-cultural social psychology rather than in describing and interpreting grammatical subsystems; he applies empirical methods taken from sociology and psychology to show the actual impact of linguistic and cognitive differences between cultures. Empirical tests would not have been feasible in Gr0nbech's case since his data were ancient inscriptions and manuscripts and his enterprise a kind of archaeological reconstruction of a past mentality from the data. On the other hand, tests would have been possible - and still are - in the case of Finno-Ugric coordination without conjunctions. The features common to all three examples are the following:

64

JOHANNES BECHERT

1.

The authors assume that there is some connection between language and thought,

more specifically between grammatical structures and habitual ways of thinking in different cultures. 2.

The three cases each seem to show a lacuna in the foreign culture/language, a lack

of something that we possess (and take for granted): the difference between substantives and adjectives in Old Turkic, the coordinating conjunction "and" in Finno-Ugric, counterfactual sentence constructions and other features in Chinese. I suspect that we would be less ready to ascribe cultural significance to such observations if the facts turned out to be the other way round; we are not prepared to perceive something that we do not have as constituting a significant lacuna in our system. Bloom mentions an example on p. 1: "Does the fact that the Chinese language, unlike English, commonly uses distinct terms to distinguish "if-then" relationships from "if-and-only-if-then" relationships imply that Chinese students are, on that score at least, better fitted for firstyear logic than their American counterparts?" - The author does not give an answer, but I think most of us would consider such a detail to be of little or no consequence. 3.

Although the examples show rather radical semantic differences between

cultures/languages, it is quite clear that the systems cannot be mutually impervious; otherwise cultural/linguistic influences from one system to another would be difficult to envisage and the linguists, psychologists, or cultural anthropologists, i.e., the observers, would neither be in a position to interpret and understand anything "foreign" nor even notice that there is a difference that calls for interpretation.

3. The Problem of the Missing Adjective/Substantive Distinction How, then, should we describe the semantic diversity of languages/cultures? 3.1. We could assume semantic invariants, e.g., the presumable statistical (i.e., nearlygeneral) language universal substantive as a word class which denotes individuals and has certain syntactic properties, and a second statistical language universal adjective as a word class denoting qualities and having syntactic properties different from those of substantives; both of these invariants might or might not exist in a given language. But this would certainly not do, as the case of Old Turkic shows. The state of affairs in this language is not simply that either substantives or adjectives are lacking but that it is impossible to distinguish between the two and that the word class in question - let us call it AS (from adjective!substantive) - neither denotes individuals nor qualities but concepts more general than both. We have to acknowledge that the relatively self-contained character of the system renders the comparison with other systems rather difficult.

THE PROBLEM OF SEMANTIC INCOMPARABUJTY

65

Moreover, we should be aware of the fact that there is no lacuna whatsoever in the system per se; the impression of a lacuna solely arises out of the attempt to compare it to other systems. But then again we are not completely outside the foreign system insofar as we are able to see and describe its strangeness. 3.2. An alternative readily comes to mind: we could set up language universals as prototypes.

In that case, categories existing in particular languages would again be

expected to conform to certain items on the list of universals, but only with a higher or lower degree of accuracy. A sort of overlap would be sufficient for a language-specific category to meet the conditions on identification with a certain prototype from the list We would now be able to declare the category AS in Old Turkic to be a language-specific approximation to the prototypical noun. As we all seem to know, the prototypical noun preferably denotes a human being, an animal, or a concrete perceptible inanimate object. The possibilities of denoting collective units (of animate beings or inanimate things), noncountable masses of concrete substances, or abstract ideas (e.g. qualities) only come as an afterthought: these are clearly less prototypical functions of nouns. The foundations of such a prototypical structure seem to reside in the biological make-up of the human species, i.e., in the general heritage that we all have in common irrespective of cultural traditions and their diversity: it is easier for us to conceive of human beings, animals, and concrete things than of collective units, non-countable masses or abstract ideas. Now this again fails in the case of Old Turkic, because the assertion that prototypical nouns denote animate beings or concrete inanimate things boils down to the statement that they denote individuals. In fact, the Old Turkic AS denotes quite the opposite: it signifies the genus, as Gr0nbech puts it, that is, the general notion of a species, or the collective unit of all real or thinkable instances of that species, in other words, something rather similar to our abstract ideas, non-countable masses or collective units (all three in one, as it were). Our prototypical noun turns out to be somewhat culture-specific, and it might require a rather deep dig to get down to whatever biological foundations it may have. 3.3. This does not mean that the concept of prototype should be given up. The advantage of this notion lies in avoiding the problem of category boundaries, cp. E. Rosch (1978, pp. 35-36): "Most, if not all, categories do not have clear-cut boundaries. .. cognitive economy dictates that categories tend to be viewed as being as separate from each other and as clear-cut as possible. One way to achieve this is by means of formal, necessary and sufficient criteria for category membership. The attempt to impose such criteria on categories marks virtually all definitions in the tradition of Western reason. .. Another way to achieve separateness and clarity of actually continuous categories is by conceiving of each category in terms of its clear cases rather than its boundaries. As Wittgenstein (1953) has pointed out, categorical judgments become a problem only if one is concerned with boundaries - in the normal course of life, two neighbours know on whose property they are standing

66

JOHANNES BECHERT

without exact demarcation of the boundary line... By prototypes of categories we have generally meant the clearest cases of category membership defined operationally by people's judgments of goodness of membership in the category." It is obvious that the clearest cases of membership in the Old Turkic category AS must be different from the clearest cases of membership in the European category noun. Consider the case of proper names: since they denote individuals, it stands to reason that they are by far more marginal as members of the Old Turkic category AS than they are as members of our category noun. However, since there are some syntactic and even semantic similarities between Old Turkic AS and our noun, we can decide to call both of them nouns - as long as we do not forget the differences. 3.4. It is probable that a continuum exists which comprises both categories, Old Turkic AS as well as our noun, allowing transitions between them and even a bifurcation from AS into substantive and adjective. Otherwise it would be difficult to understand how the Turkic languages of today were able to develop categories similar to our nouns!substantives and adjectives out of the Old Turkic AS. Furthermore, it is probable that this continuum need not be established by abstract theorizing but can be gathered from observation: I venture to suggest that it is realized in the geographical dimension like other typological continua. In any case, the development of categories like our nouns/substantives and, to a lesser extent, adjectives in the Turkic languages is a corollary of the migrations of the Turks from Central Asia towards the West and Southwest and of the influence of the urban civilizations/languages of South Asia, the Middle East and Europe upon their culture/language; see Gr0nbech (1936, p. 61) on Uighur. Such an influence may be called colonization. - On the other hand, Chinese nouns have a semantic structure which is similar to that of AS in once-neighbouring Old Turkic: they also denote abstract ideas rather than individuals, cp. U. Kolver (1982, pp. 175-176, 178) and J. Bechert (Ms.). 3.5. These rather sketchy remarks in connection with our first example are intended to bring into focus a kind of research in language universals that would take the existing semantic differences between languages seriously.

4. The Problem of the Finno-Ugric Speaking in Pairs The second example demonstrates that there are semantic categories in some languages that we cannot anticipate from our own repertoire of habitual thinking. At first sight we are not prepared to recognize the Finno-Ugric way of speaking in pairs as a genuine expression of pairs of ideas; instead, we tend to conceive of those idioms in terms of poetry: they sound like poetic devices to us. But, as Lewy emphasized, "it is obvious that

THE PROBLEM OF SEMANTIC INCOMPARABILITY

67

this [i.e., our] single concept has no more validity than the double concept that underlies the Finno-Ugric expression" (1911, p. 100). 4.1. Other inherited Finno-Ugric expressions show the same way of speaking/thinking; the following quotations are taken from G. Dicsy (1965, p. 168, translation mine): "Two special parts of a whole, forming a compound, denote that whole: Hungarian orca "face" (orr "nose" + sza "mouth"), Vogul (Mansi) noltus "face" (noI "nose" + tus "mouth"), Zyrian (Komi) nyrvom "face" (nyr "nose" + vom "mouth"), Ostyak (Xanty) notseem "face" (not "nose" + seem "eye"), Votyak (Udmurt) ymnyr "face" (ym "mouth" + nyr "nose"), Estonian dial, suusilmad "face" (suu "mouth" + silmad "eyes"); Finnish maailma "world" (maa "land" + ilma "air", originally "sky"); Votyak nylpi "child" (nyl "girl" + pi "boy"); Vogul dial, neexum "married couple" (nee "woman" + xum "man"); Hungarian szantdveto "peasant" (szanto "the ploughing one" + veto "the sowing one"), Finnish kyntaja-kylvaja "peasant" (kyntaja "the ploughing one" + kylvaja "the sowing one"), etc." 4.2. A different but probably semantically cognate relation between pairs and wholes is found in the following idioms (Ddcsy, /. c., my translation): "Parts of the body or things forming a pair are conceived of as wholes; as a consequence, a single part of such a whole counts as a half, cp. Hungarian fe I szemmel "with one eye" [lit.] ("with half an eye"), Ostyak seem peelek "oneeyed" [lit. "half-eyed"], Zyrian sin pov "id."; Hungarian fel kezeben "in one (half) of his hands", Ostyak kootl peelegne "id."; Hungarian fel kesztyu "a (single) glove" [lit.] ("half a glove"), Zyrian kepys pov "id."." 4.3. The central place in this somewhat enigmatic mode of thinking seems to be occupied by the idea of duality as unity. This is completely alien to us. Yet expressions of the type "half-eyed" still exist in Eskimo-Aleut and in Irish, i.e., at the northern and westem margins of our European area as well as in the eastern regions where Finno-Ugric languages are spoken. (See K. Bergsland (1956), pp. 165-173 for Eskimo-Aleut and J. Pokorny (1927), pp. 240-241 for Irish). So they may belong to our own past: to the ways of speaking and thinking that were in use on our continent before its colonization by the Indo-Europeans and, later on, by Greco-Roman civilization. 4.4. The modern literary languages Finnish, Estonian, and Hungarian have been colonized by the Indo-European languages surrounding them and by the civilization of our continent; the idioms quoted are archaic residues within their contemporary structures, and coordination without a conjunction has been replaced by expressions whose literal translations sound familiar to us. The words for "and" are loanwords: from Germanic in the Balto-Finnic languages, in most other Finno-Ugric languages from Russian; the origin of Hungarian es, s "and" is unknown, but it does certainly not derive from the inherited Finno-Ugric vocabulary, and Hungarian meg "and" has been

68

JOHANNES BECHERT

developed from an adverb meaning "back, behind" and later on "again", see L. BenktS (1967, p. 793; 1970, p. 874).

5. The Problem of Chinese Thinking, Or: Who Defines the Rules of the Game? Concerning the third example I shall be very brief. It proves that there arc real differences between the ways of habitual thought in China and the West. Nevertheless, I suspect that Bloom's description is skewed insofar as it does not show possible advantages of Chinese thought over ours - for example, advantages that could go with the more abstract, less individualizing semantic nature of the noun in Chinese (see § 3.4) - but only its alleged shortcomings. This is due to the fact that we are the people who define the rules of the game: the European/American way of abstract thinking counts as an achievement that has to be spread over the world. According to C. A. Ferguson (1968, p. 32), "[the] modernization of a language may be thought of as the process of its becoming the equal of other developed languages as a medium of communication; it is in a sense the process of joining the world community of increasingly intertranslatable languages as appropriate vehicles of modern forms of discourse." This is a clear statement to the effect that modernization means European/American colonization. But it does not say anything in favour of our ways of speaking/thinking; the definition of modernization is simply a question of power - so it may change.

6. Linguistic Colonization as a 'Solution1 to the Problem of Semantic Incomparability The literature quoted in this paper consists almost exclusively, and quite intentionally, of handbooks and philological classics, i.e., collections of well-known facts - but wellknown to whom? Only to the specialists of the philology in question, as it seems. The compartmentalized set-up of our discipline sometimes has the effect of burying the acquired knowledge, as it were, instead of making it operational for further research in our field. Now I think is the right moment to take some of the lines that I wrote as an abstract and to append them as a sort of coda to this paper, especially since the content of the paper does not fulfil all the promises implicit in them.1 The following paragraphs may be

THE PROBLEM OF SEMANTIC INCOMPARABIUTY

69

understood to be a suggestion for further work; needless to say that there is enough of it around for many hands. The assumption of universal semantic structures, abstracted with European methods from mostly European language data, is a self-fulfilling prophecy insofar as the expansion of European culture is bound to lead to an ever-increasing conformity of the languages and cultures of the world with these "universals". The exact knowledge of the semantic systems of languages that are as unfamiliar as possible would be a prerequisite for the discovery of semantic universals that are worth their denomination. This would require close cooperation with cultural anthropology, since semantic systems are links between language and culture. A major source of error in this respect is the fact that the semantic systems of colonized languages become hollow once their culture is dead; as a consequence, semantic subsystems of English, Russian or other colonizing languages can be found in supposedly exotic languages and thus conformity with our pseudo-universals comes about. For example, as P. Muhlhausler (Ms., pp. 44-45) writes, "the grammatical adjustment that is encountered in most Pacific languages that have come under the influence of expatriate missions and education systems is hardly less serious than language death itself. - What is meant by the above statement is that a number of apparently viable languages (in terms of numbers of speakers and social institutionalization), such as Fijian or Samoan, have nevertheless disappeared, in the sense that what has remained is primarily their formal properties and what has gone is their semantic and pragmatic aspects. The continuation of mere lexical forms of earlier languages raises the question of identity of linguistic systems over time, external pressure having introduced a degree of discontinuity and restructuring that renders the notion of historical continuity useless." There are also practical conclusions to be drawn. The insight into the semantic diversity of cultures/languages might help us to develop a more reasonable approach to human beings who live in different traditions and, consequendy, have ideas different from ours. Just as we should stop destroying our natural environment, we ought to think of ways to keep alive the cultural and linguistic heritage of peoples surrounded and threatened by our civilization. In both cases, conservation would mean a better chance for our own survival. In the second case, the reason is that some existing but disregarded ways of viewing the world might be more beneficial to mankind than some of ours have proven to be. So it would be worth wile to learn about them.

70

JOHANNES BECHERT

Note 1.1 am grateful to Eberhard Klein and Walter Morris for correcting my English and to Dietmar Zaefferer for his translation of the German original of the abstract; the responsibility for any remaining errors rests, of course, with the author.

References Bechert, Johannes: Ms., 'Universalienforschung und Ethnozentrismus', to appear in Proceedings of the XlVth International Congress of Linguists, Berlin, August 1015,1987, Akademie- Verlag, Berlin. Benkö, Loränd (ed.): 1967. 1970. 1976, A magyar nyelv törtäneti-etimolögiai szötära [Historical and etymological dictionary of the Hungarian language], 3 vols., Akad&niai Kiadö, Budapest. Bergsland, Knut: 1956, The Uralic "Half Eye" in the Light of Eskimo-Aleut', UralAltaische Jahrbücher 28,165-173. Bloom, Alfred H.: 1981, The Linguistic Shaping of Thought: A Study in the Impact of Language on Thinking in China and the West, Lawrence Erlbaum, Hillsdale, NJ. Decsy, Gyula: 1965, Einführung in die finnisch-ugrische Sprachwissenschaft, Harrassowitz, Wiesbaden. Ferguson, Charles A.: 1968, 'Language development', in Joshua A. Fishman, Charles A. Ferguson and Jyotirindra Das Gupta (eds.), Language problems of developing nations, Wiley, New York, pp. 27-36. Gabain, Annemarie von: 1974, Alttürkische Grammatik, 3. Auflage, Harrassowitz, Wiesbaden. Gr0nbech, Kaare: 1936, Der türkische Sprachbau, I, Levin & Munksgaard, Copenhagen. Kölver, Ulrike: 1982, 'Klassifikatorkonstruktionen in Thai, Vietnamesisch und Chinesisch', in Hansjakob Seiler/Christian Lehmann (eds.), Apprehension - Das sprachliche Erfassen von Gegenständen, Teil I: Bereich und Ordnung der Phänomene (Language Universals Series, Vol. 1/1), Gunter Narr Verlag, Tübingen, pp. 160185. Laanest, Arvo: 1982, Einführung in die ostseefinnischen

Sprachen, Autorisierte

Übersetzung aus dem Estnischen von Hans-Hermann Bartens (gegenüber der estnischen Ausgabe von 1975 vom Verfasser verbessert und ergänzt), Helmut Buske Verlag, Hamburg.

THE PROBLEM OF SEMANTIC INCOMPARABHJTY

71

Lewy, Ernst: 1911, Zur finnisch-ugrischen Wort- und Satzverbindung, Vandenhoeck & Ruprecht, Göttingen. Mühlhäusler, Peter: Ms. (1986), 'The Politics of Small Languages in Australia and the Pacific1, to appear in Language and Communication. Pokorny, Julius: 1927. 1928. 1930, 'Das nicht-indogermanische Substrat im Irischen', Zeitschrift für celtische Philologie 16, 95-144, 231-266, 363-394; 17, 373-388; 18, 233-248. Pritsak, Omeljan: 1982, 'Das Alttürkische', in B. Spuler (ed.), Handbuch der Orientalistik, 1.5.1: Turkologie, Nachdruck mit Ergänzungen (Erstausgabe 1963), Brill, Leiden/Cologne, pp. 27-52. Rosch, Eleanor: 1978, 'Principles of Categorization', in Eleanor Rosch/Barbara B. Lloyd (eds.), Cognition and Categorization, Lawrence Erlbaum, Hillsdale, NJ, pp. 27-48. Wiedemann, F. J.: 1884, Grammatik der syrjänischen Sprache mit Berücksichtigung ihrer Dialekte und des Wotjakischen, St. Petersburg. Wittgenstein, Ludwig: 1953, Philosophical Investigations/Philosophische Untersuchungen (Blackwell, Oxford).

Part II Basic Issues

Predication and Sentence Constitution in Universal Perspective Hans-Jürgen Sasse Universität zu Köln Institut für Sprachwissenschaft Meister-Ekkehart-Str. 7 D-5000 Köln 41

0. Introduction In the 1970's there was a trend among linguists to question the universality of the subject. It was thought that the subject-predicate relation, modelled upon the typical IndoEuropean pattern, was not applicable to all languages of the world (cf. the contributions to Li 1976, in particular Keenan, Li and Thompson, Schachter and Schwartz, also Sasse 1978 and 1982, and many other works at that time). Strangely enough, research into relational typology concentrated almost exclusively on the subject (and adjacent areas such as topics, ergativity, etc.), eventually extending to other participant relations such as objects, although it would have seemed warranted to direct the main interest of research to the predicate, the most basic relational entity in sentences. Given the mutual dependency of subjects and predicates, the former cannot be understood without the latter, and vice versa. As a result, if we have doubt about the universal validity of the subject, we are in reality challenging the generally accepted universal concept of a bipartite "Aristotelian" subject-predicate structure. The Aristotelian concept of predication was problematized very early by language philosophers such as Frege, Brentano, and Marty, though with little, if any, impact on linguistics. The most predominant models in grammatical theory basically relied on the subject-predicate dichotomy as the main factor of sentence constitution. The reason is that there was not enough typological evidence available, even some fifteen years ago (not to mention the time when Frege, Brentano and Marty developed their ideas), to conclude what other types of relations could be involved in constituting the self-contained, utterable linguistic expression of a proposition. In recent years the situation has drastically changed. Our knowledge of the languages of the world has increased as a result of numerous comprehensive descriptions of hitherto undescribed or underdescribed languages. Interest in syntax has increased enormously, with the result that grammars of "exotic" languages now provide a wealth of syntactic information. The question of the subject-predicate structure has become the subject of considerable attention due to Chomsky's continual insistence on "configurationality" as the basic principle of sentence constitution, even though a number of linguists working on individual languages have had difficulty dealing with the syntax of their respective lan-

76

HANS-JÜRGEN SASSE

guages in terms of a basic NP-VP split.1 In addition, typologists such as Comrie, Van Valin, Giv6n, and others have increased our awareness of non-configurational languages. A possible way out of this dilemma would be to stop assuming the universal existence of a bipartite "configuration^" structure, postulating instead a more abstract predicate-argument relation, realizable, as far as overt constituent structure is concerned, in varying ways. Formal semanticists will certainly welcome such a theory ("We have been teaching this for decades"). Yet, things do not seem to be that easy. My first personal doubts about the possibility of a universal representation of sentence structure arose when I investigated various American Indian languages and found that the strategy of processing information in some of these languages was so totally different from what we are accustomed to in European languages that I found it difficult to believe that the basic syntactic organization of every language of the world was modelled on the same pattern. If there is a universal underlying structure, say a "configurational" one, why should a language deviate so drastically from this structure in its formal syntactic patterning? One can hardly imagine that human beings are capable of reacting so strongly against such an economical correlation between speech and thought. Thus, given the enormous variation of syntactic patterns, it is likely that at least many, if not all of the variants make sense, i.e., that they have their own functional values, that they exist for a certain reason, and not only because "variatio delectat". On the other hand, however different the basic syntactic organization seemed to be, the principles on which it was based were always very similar. On the basis of these observations I gathered more material and tried to arrange it in some kind of systematic order. By doing so I arrived at a rudimentary typology of the simple sentence, which I would like to present here for the purpose of discussing the numerous consequences these chiefly syntactic observations may have for the question of universal semantics. Before presenting my typology, I would like to make a few remarks about the methodology on which it is based. 1. It is a functional "ideal typology". Ideal, because it establishes "ideal types", 1.e., fixed points in a continuum of areal and historical variation, holistically defining the typological characteristics they represent in the most consistent manner. Functional, because the fixed points, as we shall see, are explained in terms of their functional coherence. 2. This typology was not invented at my desk, but worked out empirically on the basis of many years of experience with the syntax of so-called "exotic" languages. 3. The methodological steps which played the most fundamental role in the exploration of the syntactic structure of these languages were the following:

PREDICATION AND SENTENCE CONSTITUTION IN UNIVERSAL PERSPECTIVE

77

a) a basically semasiological approach to the interpretation of individual language - from expression structure to content structure; b) an accompanying contrastive-onomasiological approach with the aim of localizing, in the languages under consideration, the functions found in betterknown (e.g., European) languages.

1. A Typology of the Simple Sentence We will perhaps all agree that a proposition is normally based on the conception of an event or a state, in which certain individuals may or may not be involved. Unless otherwise proven, I will proceed from the assumption that this type of proposition formation is universal and constitutes the semantic aspect of what could be called a universal sentence concept. As far as its content is concerned, we would thus have two basic constituent parts of the sentence, the event or state, henceforth called the state of affairs, as the obligatory constituent part, and the individuals (individual), which take part in it, henceforth called actants, as optional constituent parts. Between the former and the latter ones, if any, there is a logical relation, which could be called the propositional relation, presumably identical with the concept of predication in predicate logic. This logical relation does not interest us for the moment; what concerns us here is the syntactic device for expressing it. Then, in order to become manifest at the expression level of language, a proposition is made utterable by a syntactic act, which I will simply call the sentence-constituting

act. The question of whether this is the same as Searle's

"propositional" act or Austin's "rhetic" act or something else altogether must be left for future discussion. What is important about my concept of the sentence-constituting act is that it does not represent a philosophical chimera, but something that can be grasped as part of the overt language structure. It manifests itself in the form of certain formal features of utterances, whatever these may be, which serve as the criteria of differentiation between sentences (or clauses) and constitutents, between complete and incomplete sentences, between the norm-type of sentence in a language and deviant, marked types. Note that in itself the expression of the relation between the state of affairs and the actant(s) need not yet constitute a sentence; it becomes a sentence the moment it is realized in such a way that the respective language-specific mechanisms of sentenceconstitution are overtly expressed. For instance, the phrases barking of a dog, and a barking dog, although they express a state of affairs/actant relation, are not sentences in English, but can be made sentences by adding a verb: there is ...,... can be heard, etc.

78

HANS-JÜRGEN SASSE

In the languages we are familiar with, sentence constitution is mainly expressed by the relation of subject and predicate. In the Classical Languages, in English, German, etc., subject and predicate make up the relational fabric defining the sentence as a formal unit of these languages. This fact has been clearly recognized in traditional linguistics and has been accounted for in traditional sentence definitions. Moreover, subject and predicate are semantically defined with respect to the basic constituent parts of the proposition: the subject is one of the actants and the predicate denotes the state of affairs it is involved in. It is for this reason that we will not say there is barking of a dog, but a dog is barking. The subject is characterized by the predicate as being involved in a certain state of affairs. The Indoeuropean predicate thus unites three related functions: characterization, expression of a state of affairs, and sentence constitution. Only a small minority of the languages of the world, however, construct their sentences according to this pattern. Not everywhere do we find an "underlying" (hypokeimenon) subject and a "contradictory" (kategoroumenon) predicate. This certainly does not mean that the others do not have sentences, but only that these are constructed by some other mechanism rather than that of the combination of subject and predicate.3 Before we proceed let us consider some examples. The following three sentences are translations of typical English subject-predicate structures into Cayuga, an Iroquoian language of Canada: (1)

ake-

tshahnf'k

a:-

it-»me fright

yo-

hpn'at-a:t-

optative it—»it potato

kph

reflexive rot

'I was afraid the potatoes would rot' (2)

he-

'kg:'-

eh

ho-

hgn'at-

I-»him younger sibling diminutive it->him potato

a-

k'ate'

linker be many

'My younger BROTHER (focus) has many potatoes' (3)

ne:'

ki'

Sampson

this is particle S.

Delilah

hni'

kagtat-

hawahk-

sg'

D.

and they-»they be parent distribute

'These are Sampson's and Delilah's children' It is worthwhile translating these three sentences into English literally, i.e., exactly imitating the expression structure of Cayuga: 'it frightened me; it would rot itself potatowise'; 'I younger-sibling him; it manies him potato-wise'; 'this is; Sampson and Delilah; they parent them (are to them in a parent-child relation multiply distributed)'. Where is the subject? Where is the predicate? Where is the difference between the actant and the state of affairs expression? Except for the names Sampson and Delilah, all full words occurring in these three sentences are verbs!

PREDICATION AND SENTENCE CONSTITUTION IN UNIVERSAL PERSPECTIVE

79

Now compare the sentence structure of Tongan, a Polynesian language.4 (4)

na'e

ui

'e

Si one 'a

preterite call ergative S.

Pita

absolute P.

'Sione called Peter' (5)

na'e

sio

'a

Sione kia

preterite see absolute S.

Mete

locative M.

'Sione saw Mary' The normal structure of what would appear in the translation as a transitive sentence is (4). The "verb" is preceded by a tense-aspect-mood marker, whereas the agent is preceded by the "ergative" preposition 'e, and the patient by the 'absolutive' preposition 'a. There may be other combinations as in (5), where one actant takes the locative preposition kia, but at least one actant is always in the 'absolutive' case. Interestingly enough, the absolutive is not entirely fixed as far as its semantic role is concerned; it may be understood as the agent or as the patient: (6)

na'e

ui

Sione

'a

preterite call absolutive S. 'Sione called'

'Sione was called'

This is perfectly explainable if one correctly understands Tongan sentence structure. In reality 'a is nothing but a genitive preposition (= English of) with a very unspecific semantic value, cf. (7)

(h)e

tamasi'i

article boy

'a

Sione

of S. (definite)

'Sione's boy' The genitive or possessive relation is also apparent in constructions with pronouns, where the absolutive form is identical with the possessive form. The verb-absolute combination is thus constructionally a verbal noun-genitive combination (like barking of the dog) and can in fact be used in this way as a phrasal constituent: (8) a ) n a '

ku

kata

preterite absolutive: lsg laugh 'I laughed'

80

HANS-JÜRGEN SASSE

b) ki

he-

'e-

ku

kata

locative article possessive marker my laughing (definite) '... about my laughing' An English-imitating translation of (4), (5), and (6), respectively, would be: 'it was calling of Peter by Sione'; 'it was seeing of Sione to Mary'; 'it was calling of Sione' (none of the semantic roles being syntactically defined). The final example is from Basque. Like Tongan, Basque is an ergative language, which means that the patient ("object") of a transitive verb is expressed in the same way as the "subject" of an intransitive sentence. Upon closer inspection, it seems difficult to identify "subjects" and "predicates":5 (9)

haurr-

e-

k

liburu-

child

definite ergative book u-

a

definite: singulanabsolutive

irakurri

d-

0-

te

read

absolutive:3sg auxiliary ergative plural

'The children read the book' (10) haurr-

child

ak

jiin

d-

ira

definite:plural:absolutive

come absolutive:3pl auxiliary

'The children have come' (11) rti- k

I

gizon- a-

ergative man

d-

a-

-ri

ogi-

a

definite:singular dative bread definite:singular:absolutive kar-

absolutive:3sg tense bring

kio-

t

dative:3sg ergative:Is

'I bring the man the bread' Although the sentence structure of Basque totally differs from that of Cayuga and Tongan, all three have in common the fact that basic grammatical relations (in the sense of Indoeuropean subjects, objects, and predicates) cannot be identified. In Basque, the predicate is a complex form, containing pronominal indexes for as many as three actants (ergative, absolutive, dative) located in the auxiliary. The verb-complex therefore constitutes a minimal sentence with a perfect relational structure, which does not require any further complementation for its support. If full nouns occur, they are added to the verb-complex in the form of appositions ('the children, the book, they read it'). The above observations suggest two parameters for a typology of the simple sentence: 1. the localization of the sentence-constituting operation (as one of the aspects of "predicativity", i.e., what are the formal means of sentence constitution?)

PREDICATION AND SENTENCE CONSTITUTION IN UNIVERSAL PERSPECTIVE

81

2. the syntactic device for connecting the actant expressions to the state of affairs expression ("actant-binding"). The localization of the sentence-constituting operation can be related to the two basic constituents of propositions: state of affairs and actant. Three types may be found in the languages of the world: TYPE 1: The element expressing the sentence-constituting operation is located outside the state of affairs!actant complex. I.e., the syntactic combination of the state of affairs and the actant does not yet in itself constitute a sentence, but only a phrase requiring a further predicative element in order to become a sentence. TYPE 2: The sentence-constituting operation is expressed by combining a selected actant, functioning as the predication base, with a characterizing state of affairs expression. This is the Aristotelian type of predication: a bipartite predication expressing the fact that state of affairs S exists with respect to actant A. TYPE 3: The sentence-constituting operation is inherent to the state of affairs expression and is therefore located within the state of affairs expression. I.e., the expression denoting a state of affairs constitutes a sentence in and of itself. Schematically, if we symbolize the element expressing sentence constitution by p, the expression of the state of affairs by S, and the actant by A, we would have the following three constellations (disregarding linear order): 1.

(S

2.

P S

P

A) A

3.

(S

P)

A

There is a fourth logical possibility, namely: 4.

(A

p)

S

i.e., an actant expression with a predicative force, to which an expression denoting a state of affairs is in some way connected. Supposing that nouns in English were impersonal

82

HANS-JÜRGEN SASSE

verbs, an utterance of the form it mans would be a complete sentence with the meaning 'there is a man'. An expression denoting a state of affairs could be attached to it in the form of an adverbial, e.g., it mans in a sleeping way for 'the man is sleeping'. As far as I know, this type of sentence-constitution has never been attested. There are good reasons to assume that it does not exist, since the central information conveyed in an utterance normally relates to a state of affairs rather than an individual; that's why we expect the state of affairs expression to be obligatory, the actant expression optional. But there are a number of languages in which a certain marginal sentence type could be seen as representing an example of type 4. Consider the following French example: (12) Qu'est-ce

qu'il y a?-C'est

'What's going on?-

maman qui me bat.

MA's (Focus) hitting me' (lit. 'it is Ma that's hitting me)

This type of utterance is characteristic for so-called thetic statements ("all-new utterances") of an entity-central nature, expressing the monolithic character of thetic statements by a nominal phrase with an attribute in a presentative or existential construction (for further discussion, see Sasse 1987). But this construction is highly marked in these languages. There are no known examples of languages in which an unmarked sentence is in any way comparable to this construction. We would be left, then, with the three types of predication mentioned above. Now to the second parameter, the syntactic device for relating actant expressions to the state of affairs expression. Thus far, five different devices for actant-binding have been discovered (whereas there may be more, still waiting to be found): 1. Predicative. This device is closely related to sentence-constitution type 2. An actant is bound to the state of affairs expression in that it is selected to constitute the predication base. 2. Attributive/possessive. The actant is bound to the state of affairs expression in the form of an attributive/possessive ("genitive"-like) construction. 3. Adjunctive. The actant is bound to the state of affairs expression and forms an oblique phrase by means of a linking mechanism (e.g., an adposition); the linker provides semantic information with respect to role structure. 4. Appositive. This device is closely related to sentence-constitution type 3. S expressions of type 3 frequently contain pronominal indexes, to which A expressions can be added in an appositional relation ("he, the X, acts upon him, the y"). 5. Rective. The actant is governed by the S expressions via lexical valency.

PREDICATION AND SENTENCE CONSTITUTION IN UNIVERSAL PERSPECTIVE

83

In sentences containing more than one actant devices for actant-binding may be combined in differing ways. If more than one method of binding is used, there is normally a primary one which occurs in all sentences regardless of the number of actants involved, and an additional one employed to differentiate secondary from primary actants in sentences containing more than one actant. We will see below that the choice of the actant-binding device(s) does not occur independently of the sentence-constituting device.

2. Examples 2.1. Cayuga, Basque and Tongan Let us now study a number of examples. First of all, consider the above-mentioned examples of non-Aristotelian sentence-constitution. In Cayuga the method of constituting a complete utterable proposition consists in combining "verbal" roots, i.e., roots denoting states of affairs, with pronominal affixes acting as referential indices. Consequently, every full word constitutes a complete sentence, even those which in European languages would appear as 'nouns', i.e., as words denoting individual entities: (13)

e-

nestan-ya'k-tha'

she/one->it board

cut

with

'one cuts boards with it' = 'saw' (14) ohn-a' it -»it fat "nominal" 'it fat-s it' = 'it is fat' = 'fat, grease' Cayuga belongs to sentence-constituting type 3. Since every full word represents a complete utterance, and since all roots, even those denoting individual entities, are semantically state of affairs expressions, sentence constitution takes place within the state of affairs expression. Under these circumstances it is clear that Cayuga cannot be expected to possess a direct equivalent for the typical combination of S and A expressions characteristic of the European sentence type. How, then, does Cayuga express the relation between actants and states of affairs? We have already seen above that Cayuga sentences are characterized by concatenation of predications. The first predication in a sentence is always the most important (cf. Mithun 1987) and may be considered the equivalent of the European sentence predicate. The subsequent predications obviously serve to more closely elaborate the initial one. Reference is established by means of coreferential/anaphoric relations between pronominal indices:

84

HANS-JÜRGEN SASSE

(15)

a-

ho-

htg:'

past it-»him lose t

ho-

tkwft-

it-»him wallet

f

a'

n§:ky$

"nominal" this

I

h-

gkweh'

he-»it man |

Literally: 'it was lost to him, it is his wallet, this one, he is a man' = 'this man lost his wallet'. Elaboration is achieved through a step-by-step explanation of the pronominal indices ('it' as 'wallet' and 'he' as 'man'), which themselves help to construct an anaphoric network for the identification of props in discourse. The method of actantbinding is appositive, with the peculiarity that actants themselves are predicative.6 A very similar connection between sentence-constitution type 3 and appositive actantbinding exists in Basque, even though the end-result appears different due to the fact that in Basque actant expressions are normal nouns with no predicative force. As for the rest, the following principle applies: sentence-constitution takes place in the verb complex containing pronominal indices, whose referents are defined by the noun phrases connected with them in a syntactically appositive relation: 'the children, the book, they read it'. Tongan, on the other hand, belongs to type 1, with the predicative element outside the S-A complex. Type 1, in its most outstanding form, is found in some languages as a marked sentence, e.g., in Egyptian Arabic where it is sometimes used for thetic statements, e.g. (16)

gara eeh?

fiih bee?

tazaakir

what's the matter?

exist sale of tickets

'What's happening? - tickets are being sold' The above pattern is "existential marker + S expression in the form of a verbal noun + A expression in a genitive-like construction". It is obvious that the structure of the intransitive sentence in Tongan (e.g., example 6) follows exactly the same pattern. Tense/aspect markers such as na'e 'preterite' serve the same function as Arabic fiih 'there is'; they set a situative frame within which the S expression is embedded. S + A form a possessive phrase, of which S is the head and A the modifier. Tongan is an example of a language employing two different methods of actant binding. While one of the actants is always in a possessive relation, all of the others (if more than one actant is involved), are syntactically adjunctive in that they appear as locative, instrumental, etc. adpositional phrases. The status of the "genitive" preposition is somewhat problematic, since it could also be inteipreted as an adjunct marker, but its

PREDICATION AND SENTENCE CONSTITUTION IN UNIVERSAL PERSPECTIVE

85

special role as a possessive marker is evident in view of the fact that it can be replaced by a possessive pronoun, which is not possible with any other "adjunct".

2.3. Tagalog and Jacaltec Let us give two more examples in order to demonstrate sentence-constitution type 2. The prototypical example of type 2 is the ascriptive nominal sentence, with or without a copula, where the subject forms the predication base (the entity about which something is said), and the non-referential predicate noun the S expression characterizing the subject. There are several languages in which this is the basic pattern of the simple sentence, i.e., in these languages there is no difference between nominal and verbal sentences. Tagalog is a very clear example of this type.7 Its lexicon does not provide any clear distinction between S expressions ("verbs") and actant expressions ("nouns"); every full word may occur in either the one or the other function. This is possible due to the fact that S expressions in Tagalog are normally A-oriented, very much like participles of European languages, i.e., they basically denote individuals involved in a state of affairs rather than the states of affairs themselves (the latter may also be the case but not in the type of sentence we are concerned with here). The Tagalog sentence is composed of two basic slots, which must be obligatorily filled. The first, syntactically marked only by its sentence-initial position, is provided for the "Aristotelian predicate"; the second, syntactically marked by the preposed particle ang, is provided for the "Aristotelian subject" (the "predication base"). In terms of our typology of sentence-constitution we would say that the juxtaposition of these two elements (there is no copula) can be considered a paradigm case of type 2. In terms of the actant-binding typology, the actant occurring in the ang phrase represents a paradigm case of the "predicative" type of actant-binding. There are two optional slots, representing two further methods of actant-binding, a possessive/attributive one, and an adjunctive one. Due to the participial (actant-oriented) character of the predicate, there is always a correspondence between the semantic role of the predicate (indicated by an affix) and the (inherent) semantic role of the a«g-phrase. Since the system is comparatively flexible, there are often several possibilities of semantic role relations for one and the same state of affairs, yielding what has been termed a system of "voices". There are only a few simple rules for the distribution of possessive and adjunctive actants: if one of the basic roles, actor or undergoer, appears in the awg-phrase, the other is in the possessive relation. Oblique roles, such as locative, benefactive, and the like, are bound adjunctively, unless they constitute the arcg-phrase. If the latter is the case, both actor and undergoer appear in

86

HANS-JÜRGEN SASSE

the possessive relation. The following sentences demonstrate the above description: ang is glossed by REF (= reference marker), the possessive particle ng is glossed by LK (= linking element); for each sentence an English-imitating literal translation is given in order to demonstrate the system in a more transparent way. (17)

b-um-ili

ang

babae

buyer

REF

woman LK rice

ng

bigas

'buyer of the rice (is) the woman' = 'the woman bought the rice' (Actor = predication base) (18) b-in-ili

ng

bought

babae

ang

bigas

LK woman REF rice

'bought of the woman (is) the rice' = 'the rice was bought by the woman' (Undergoer = predication base) (19)

i-b-in-ili

ng

babae

bought-for LK woman

ng

bigas ang

LK rice

bata

REF child

'beneficiary-of-buying of the woman of the rice (is) the child' = 'the child was bought rice for by the woman' (Benefactive = predication base) The final example is Jacaltec, a Mayan language; the data are taken from Craig (1977). Mayanists distinguish two sets of pronominal elements, which they traditionally call Set A and Set B. In modern publications, the terms "ergative" for Set A and "absolutive" for Set B have become common, since Mayan languages have an ergative system and play a certain role in the modern typological discussion on ergativity. We will see below, however, that the most interesting phenomenon in Jacaltec is not the ergativity itself, but how it is established syntactically. The two sets of personal pronouns in Jacaltec are as follows: SET A

Sg.

PI.

SET B

C-initial

V-initial

1

hin-mam 'my father'

w-atut 'my house'

meba hin 'I am poor'

2

ha-mam 'your father'

haw-atut 'your house'

meba hach 'you are poor'

3

s-mam 'his father'

y-atut 'his house'

meba 0 'he is poor'

1

cu-mam 'our father'

j-atut 'our house'

meba hori 'we arc poor'

2

he-mam 'your father'

hey-atut 'your house'

meba hex 'you are poor'

3

s-mam 'their father'

v-atut 'their house'

meba 0 'thev are poor'

PREDICATION AND SENTENCE CONSTITUTION IN UNIVERSAL PERSPECTIVE

87

For the two sets of personal pronouns the following functions have been described in the literature: (20)

Set A: A. Subject of a transitive verb: (a) ch-

in

ha-

ASP B i s (b) ch-

'you hit me'

maka

A2s hit

ach

y-

ASP B2s A3

oche

naj

like

KL

'he likes you'

B. Possessor: (a) hin-

'my sandals'

xanab

Als sandal (b) s - bak s- sat A3 pit (c) ay

'the pit of his face'

naj

A3 face KL haw-

EXIST A2s

= 'his eye' 'you have a brother'

ux'taj

brother

C. Possessor of a reflexive pronoun: (a) x 0- w- il hin-ba 'I saw myself ASP B3 (b) cu-

cajyat

Als see Als REEL cu- ba 'We are enemies (of each other)'

Alp enemy Alp REFL D. Object of a preposition: (a) w- et

'to me'

Als to (b) J-

'on top of us'

iban

Alp on top (c) y- ul te' A3 in KL

'in the house'

nah

house

E. Subject of an aspect-less embedded sentence: (a) sab ichi cu- muni ay i 'we started working early' early begin Alp work (b) x0- (y)il naj ASP B3 A3

hin

ha-

see KL B1 A2

mak-

ni

hit

SUFF

'he saw you hit me'

(21) Set B A. Subject of a nominal sentence (with a nominal or a stative predicate): (a) winaj

man

hach

B2s

'you are a m a n '

88

HANS-JÜRGEN SASSE

(b) sicinaj hin

tired

'I am tired'

Bis

(c) nimejal

big

0

te'

riah

'the house is big'

B3 KL house

(d) hin

y-

ahaw

'I am its owner'

Bis A3 owner B. Subject of an intransitive verb: (a) ch-

on

wayi

'we sleep'

ASP Blp sleep (b) xc-

ach

toyi

'you go'

ASP B2s go (c) x -

0

cam no'

cheh

'the horse died'

ASP B3 die KL horse C. Object of a transitive verb: (a) ch-

in

haw-

'you see me'

ila

ASP B i s A2s see (b) ch- on s- col naj ASP Blp A3 help KL (c) x - 0- s- watx'e naj te' ASP B3 A3 make

'he helps us' tiah

'he made the house'

KL KL house

D. Subject of a so-called passive (= stative verbal adjective): (a) x 0'11- lax naj 'he was seen' ASP B3 see PASS KL (b) xc- ach mak- ot 'you were hit' ASP B2s hit

PASS

At first glance, these lists appear relatively heterogenous. The reason is that there has been no attempt so far in the literature on Mayan to postulate a uniform basic function for each of the two sets.8 This is all the more surprising, since the deeper functional coherence is evident. If one compares functions B through D of Set A, it immediately becomes clear that one is dealing with a possessive relation. B is a canonical possessive expression; the cases under C can be explained if terms of the noun 'self, to which possessive prefixes are attached (as in numerous other languages, e.g., Arabic, Turkish, and even English: my-self, your-self, our-selves). As for function D, the term "prepositional object" is a misnomer: the class of the so-called positional expressions in Mayan constitutes a subclass of nouns with very concrete meanings, e.g., j-ibari 'on top of us' is literally 'on our top'.

PREDICATION AND SENTENCE CONSTITUTION IN UNIVERSAL PERSPECTIVE

89

Once a possessive function for Set A has been established, the cases under E ('aspectless embedded clauses') easily fit in: in reality they are not clauses at all, but nouns of action, which explains the fact that they bear deverbal derivational suffixes and are neutral with respect to aspect Their nominal character has been explicitly pointed out for Mam, another Mayan language, by Nora England. This also explains their possessive marking, which was often erroneously understood as an example of split ergativity. (22)

cu-munlayi

'our working'

(23)

hin-ha-makni

'your me-hitting'

In Mam these verbal nouns bind both actants possessively ('my your hitting'). It is not going too far to conclude from these considerations that an extra function 'subject of transitive verbs' does not exist for Set A, and that this alleged function is only a result of the English translation, while in reality it is part of the basic possessive meaning of this set. In other words, the actant corresponding to our subject of transitive verbs in the English translation is marked by a possessive affix on the verb. As for the basic function of Set B, it is likewise clear that it establishes a predicative relation. Pronouns of Set B correspond to subjects of ascriptive nominal sentences (much in the same way as the ang-phrase of Tagalog). There is no copula in Maya, so that the predicative relation is expressed by the form of the pronoun and the juxtaposition of the predicate and predication base (the actant for which the predication holds). A certain copula function is fulfilled by the aspect markers, in so far as they contribute to the overall marking mechanism of the predictive relation. If we accept the hypothesis that the basic function of Set B is to mark an actant in a predicative relation, the entire verbal system of Maya suddenly appears as a functional unit: since the actor is bound to the "verb" in the form of a possessive expression, the verb + actor complex constitutes a phrase, but not a sentence. In order to obtain a sentence, a predicative relation is necessary, which is produced by the ascriptivesentence-like connection with the undergoer actant. The entire verbal phrase with the possessively bound actor serves as the predicate of the ascriptive sentence. The predication base is the undergoer, hence Mayan ergativity. We thus have an expression displaying a type of syntactic relations which we could render in English with the help of a participial predicate: 'I am seeing you' = 'you are my seen (one)'. That is, we have a structure very similar to the one described for Tagalog above. So far we have been dealing with pronominal actants only. The interpretion given above is confirmed when we consider non-pronominal actants. Before doing this, a word must be said about word order. The examples show that the Set A pronouns appear as

90

HANS-JÜRGEN SASSE

prefixes, while the set B pronouns stand in an isolated position in front of the set A pronouns, if occurring together with them, otherwise they are placed after the predicate. The subject of an ascriptive sentence is placed at the end of the sentence, as mentioned above, i.e., it follows its predicate. Now there is a peculiarity about Mayan syntax, which has been mentioned in the literature: nominal actants appear in the reverse order. A sentence with two actants is constructed in the following way: it begins with the "absolutive" pronoun, next comes the "ergative" prefix, followed by the verb form, then the "ergative" noun, and finally the "absolutive" noun. With pronouns, then, the order is A E, with nouns it is E A. This peculiarity of word order follows quite naturally from the character of the actant relations already described. We discovered that the "ergative" relation (i.e., what appears as the subject of a transitive sentence in the translation) is a possessive relation. It is to be expected then that the nominal phrase which expresses the "ergative" actant is marked as a possessor. This is in fact the case. A genitive relation is expressed in Jacaltec by placing the nominal possessor immediately after the possessum, the former bearing a possessive prefix corresponding to the latter: (24) s -

txam riah

POSS:3s edge

house

'its-edge house' =

'the edge of the house'

The possessor may not be separated from the possessum, since juxtaposition is the formal sign of the relation. The so-called ergative shares this formal property with the possessor, i.e., there is a 100% formal coincidence between the possessive phrase and the actor-verb complex. The position of the absolutive, on the other hand, can be explained by the fact that an actant in a predicative relation (i.e., constituting a predication base) always follows its predicate. The transitive sentence is in fact ordered in exacdy the same way as a stative sentence with a possessive phrase as the predicate: (25) s -

watx'e naj

POSS:3s make

pel

cab te'

xila

classifier Pedro two classifier chair

'his made (ones) of Pedro (are) two chairs' = 'Pedro made two chairs' (26) y -

ahaw

te'

riah

POSS:3s owner classifier house 'the owner of the house is Pedro'

naj

pel

classifier Pedro

PREDICATION AND SENTENCE CONSTITUTION IN UNIVERSAL PERSPECTIVE

91

Schematically:

s-

watx'e

naj pel

cab te' xi la

y-

ahaw

te' riah

naj pel

POSSESSUM

POSSESSOR

PREDICATE

SUBJECT

To sum up, Jacaltec possesses two "slots" of actant relations in the sentence, which may be labelled "actor" and "undergoer". It belongs to sentence-constitution type 2 ("Aristotelian predication"), which means that in intransitive sentence the sole actant is always bound to the predicate in a predicative relation. With transitive sentences, the actor is in a possessive relation and the undergoer in a predicative relation. This results in ergativetype syntax, since the undergoer of a transitive sentence is marked in the same way as the sole actant of an intransitive sentence. The system resembles that of Tagalog, with the difference being that it lacks the flexibility of the predication base with respect to semantic role; in Jacaltec it is always the undergoer which constitutes the predication base.

3. Results Let us summarize the results of our considerations. We began with the attempt to redefine the concept of predication as a sentence-constituting device serving to unite elements of a proposition in order to make it utterable. We saw that the bipartite Aristotelian type of predication is only one of at least three possible types of sentence-constitution. Moreover, there is a second parameter involved, that of actant-binding, which cross-cuts the parameter of sentence-constitution and results in a rich array of different combinations languages may employ in constructing their respective networks of grammatical relations. The consequence is that the idea of the subject-predicate structure as a linguistic universal must be abandoned. Instead, each individual language should be more closely examined with regard to the combinations of sentence-constituting and actant-binding devices it employs, since the functional coherence of the simple sentence seems to be based precisely on these two parameters. In the course of our discussion we have seen that these two principles do not function independently of each other, but that there are certain correlations between the localization of the sentence-constituting mechanism and the

92

HANS-JÜRGEN SASSE

method of actant-binding. Moreover, we have seen that there is a third correlate involved, namely the basic lexical semantic characteristics of the main constituent parts of sentences, and in particular those of the state of affairs expressions: whether or not they are actant-oriented ("participial"), whether or not there is a clear distinction between nouns and verbs, and so on. The following table illustrates some of the findings: LOCALIZATION OF THE

TYPE OF STATE

CHARACTERISTIC

SENTENCE-CONSTITU-

OF AFFAIRS

ACTANT RELATION

TING OPERATION

EXPRESSION

outside the S-A complex

'action-noun-like'

rrvpe n 'verb-like'

Examples

combination of possessive,

Polynesian

adiunctive. and others

fTonganl

rective

English

between the state of affairs

German

expression and actant

'participle-

predicative in combination

Tagalog

(Type 2)

like'

with others (possessive,

Maya (Jacaltec)

adjunctive ...t within the state of

'verb-like'

at least one relation appositive Cayuga

affairs expression

Basque

(Type 3)

Latin

So far we have been used to treating the typology of the simple sentence in terms of atomistic parameters such as word order, ergativity/accusativity, etc. The typology presented here makes it possible to proceed directly from syntactic manifestations of basic linguistic operations such as reference and predication, helping to make the functional coherence, immanent structure, and typological 'layout' of a language transparent. For, if the basic principle involved is understood, the functional place value of the details in the overall system becomes self-apparent. Questions of subject and topic, word order, noun/verb distinction, word classes and their grammatical categories in general, even problems that go beyond the simple sentence such as subordination, embedding etc. turn out to form a coherent whole with a common denominator. This leads us finally to the main subject of our joint activity, the question of semantic universality. There can be no doubt that the drastic syntactic differences described above are related to drastic semantic differences. When we speak of 'possessive', 'adjunctive', 'predicative', etc. syntactic relations, we are in fact dealing with semantic relations, or at least with syntactic relations having semantic roles (possession, location, etc.). This means that roughly the same information is conveyed in radically different ways in different languages, and, despite our collective fear of Whorfianism, we are forced to

PREDICATION AND SENTENCE CONSTITUTION IN UNIVERSAL PERSPECTIVE

93

admit that this must have something to do with differences in conceptualization and thought. Of course, it is not my intention to impose upon the Cayuga speaker a totally different conceptualization of a saw or of the loss of a wallet. What I want to show is that the mere fact that the same thing can be said in so totally different ways should make us careful about assuming uniform universal semantics, at least if semantics is understood in the traditional sense as the meaning of morphemes, words, phrases, and sentences (rather than as a cognitive entity). Remember our observation made above that the different ways of sentence-constitution are closely interrelated with the lexical semantics of the constituted parts of sentences: The syntactic system of Cayuga is based on the primarily event-denoting character of its roots, while the structure of Tagalog and Jacaltec is made possible by the principally actant-oriented character of lexical entries. This close interaction of lexical semantics and syntax is a strong argument against the assumption of universal semantics in the sense defined above. What we may expect to find are universal semantic principles (such as reference and predication), which may be definable as semantic universals, but I doubt if we can go much further than that. The reason is that semantics (always understood in the sense defined above) does not represent an autonomous level of linguistic analysis: meaning cannot be separated from the expression, of which it forms an integral part. Consequently, if there are differences in expressions, there are also differences in meaning. This fundamental unity of expression and meaning defines meaning as language-specific and rules out universal semantics as a legitimate field of study.

Notes 1. For example, Brettschneider (1979), and later Bossong (1984), sated (independently of each other) that a description of Basque in terms of subject and predicate relations is inadequate. The configurationality problem has been extensively discussed for Warlpiri by Hale (1983). In my paper on the thetic/categorical distinction (Sasse 1987) I described a monomial type of predication used in many languages as an expression of thetic statements. Himmelmann (1986) even goes so far as to posit a typological continuum between monomial and bipartite predication.

2. A paradoxical discrepancy exists, on the one hand, between the apparent fact that almost every language possesses sentence-constituting mechanisms and, on the other, the empirical observation usually made in discourse analysis that in actual spoken discourse (conversation) you can make out turns and clauses, but rarely sentences. The reason for this is obvious: In conversation, the construction of a "sentence" is a joint activity, in which all conversation partners participate. It may well be that the sen-

94

HANS-JÜRGEN SASSE

tence, traditionally considered the basic unit of syntax, is only one of several possible types of information-processing available, even in our own languages. Since the question does not have an adverse affect on my typology of sentence-constitution it may be disregarded for the moment. A complete typology of the simple sentence, however, would have to take all types of information-processing into account

3. Or by several mechanisms. It is perfectly normal that different sentence patterns are the result of different sentence-constituting mechanisms. - What constitutes a sentence is determined for each individual language on the basis of its specific sentence-constituting devices. This sounds circular, but is not, the primary criterion being that it corresponds to the native speaker's own idea of a complete utterance. 4.1 am indebted to Jürgen Broschart (Cologne) for providing me with the Tongan data.

5. The following examples are discussed at length in Himmelmann (1986). 6. For a detailed discussion of this view of Cayuga syntax, cf. Sasse 1988. 7. The interpretation of Tagalog syntax presented here is based on Himmelmann (1983) and (1987).

8. A methodological question is being touched on here, which I cannot discuss in detail due to lack of space. One of the fundamental heuristic strategies in linguistic analysis should be the attempt to find a uniform function or meaning for each formal linguistic phenomenon, i.e., not to assume a priori that there are irregular mapping relations between form and function (homonymy, etc.), but to proceed from the assumption that there is, in principle, a 1 : 1 relation, each form having one and only one basic function, the subfunction being determined by and explainable in terms of the environment. Assuming of homonymy is acceptable only as the result of the inability to discover a uniform meaning, i.e., as a last resort, so to speak, but never as an acceptable working principle.

References Bossong, Georg : 1984, 'Ergativity in Basque', Linguistics Brentano, Franz von: 1924-28, Psychologie

2 2 / 2 3 , 341-92.

vom empirischen

Standpunkt.

Ed. by O.

Kraus. 3 vols. Leipzig (Vol. 2: 'Von den psychischen Phänomenen', Chapter 7(3): 'Vorstellung und Urteil zwei verschiedene Grundklassen'). Brettschneider, Gunther: 1979, 'Typological Characteristics of Basque', in Frans Plank (ed.), Ergativity,

Academic Press, N e w York.

Craig, Colette: 1977, The Structure London.

of Jacaltec,

University of Texas Press, Austin and

PREDICATION AND SENTENCE CONSTITUTION IN UNIVERSAL PERSPECTIVE

95

Frege, Gottlob: 1918, 'Der Gedanke. Eine logische Untersuchung', in Beiträge zur Philosophie des deutschen Idealismus 1 (repr. in Frege, Gottlob: 1966, Logische Untersuchungen, hrsg. und eingeleitet von Günther Patzig, Vandenhoek & Rupprecht, Göttingen). Hale, Kenneth: 1983, 'Warlpiri and the Grammar of Non-Configurational Languages', Natural Language and Linguistic Theory 1,5-47. Himmelmann Jr., Nikolaus: 1983, Linking im Tagalog (unpublished M.A. Thesis). Himmelmann Jr., Nikolaus: 1986, Morphosyntactic Predication. A Functional-Operational Approach (= akup 62, Institut für Sprachwissenschaft, Köln). Himmelmann Jr., Nikolaus: 1987, Morphosyntax und Morphologie - Die Ausrichtungsaffixe im Tagalog, Wilhelm Fink, München. Keenan, Edward L.: 1976, Towards a Universal Definition of 'Subject", in Charles N. Li (ed.), Subject and Topic, Academic Press, New York. Li, Charles N. (ed.): 1976, Subject and Topic, Academic Press, New York. Li, Charles N., and Sandra A. Thompson: 1976, 'Subject and Topic: A New Typology of Language', in Charles N. Li (ed.), Subject and Topic, Academic Press, New York. Marty, Anton: 1897, 'Über die Scheidung von grammatischem, logischem und psychologischem Subjekt resp. Prädikat', Archiv für systematische Philosophie 3, 174-90 and 294-333 (reprinted in Marty, A.: 1918, Gesammelte Schriften, 2. Band, 1. Abteilung: Schriften zur deskriptiven Psychologie und Sprachphilosophie, Niemeyer, Halle). Mithun, Marianne: 1987, 'Is Basic Word Order Universal?', in Russell Tomlin (ed.), Coherence and Grounding in Discourse, Benjamins, Amsterdam. Sasse, Hans-Jürgen: 1978, 'Subjekt und Ergativ: Zur pragmatischen Grundlage primärer grammatischer Relationen', Folia Linguistica 12,219-252. Sasse, Hans-Jürgen: 1982, 'Subjektprominenz', in Sieglinde Heinz and Ulrich Wandruszka (ed.), Fakten und Theorien, Gunter Narr, Tübingen. Sasse, Hans-Jürgen: 1987, 'The Thetic/Categorical Distinction Revisited', Linguistics 25, 511-80. Sasse, Hans-Jürgen: 1988, 'Der irokesische Sprachtyp', Zeitschrift für Sprachwissenschsft 7, 173-213. Schachter, Paul: 1976, 'The Subject in Philippine Languages: Topic, Actor, Actor-Topic, or None of the Above?', in Charles N. Li (ed.). Subject and Topic, Academic Press, New York. Schwartz, A.: 1976, 'On the Universality of Subject: The Ilocano Case', in Charles N. Li (ed.), Subject and Topic, Academic Press, New York.

Aristotle Goes to Arizona, And Finds a Language without "And" David Gil Department of Linguistics University of Haifa Haifa Israel 1. Is Aristotelian Logic Just Greek Grammar? In his Beiträge zu einer Kritik der Sprache (translated by and cited in Weiler 1970:143), Fritz Mauthner writes: "The whole logic of Aristotle is nothing but a consideration of Greek grammar from an interesting point of view. Had Aristotle been speaking Chinese or Dacota, he would have had to arrive at a completely different logic ..." Alas, we are not in a position to know. The history of ideas does not easily permit the formulation of testable scientific hypotheses; claims such as the above tend to assume the role of articles of faith, believed by some, ridiculed by others, ignored by yet others. Recently, however, developments in linguistics and other cognitive sciences have begun to suggest a more practical approach towards a better understanding of the relationship between language and other domains of mental activity. Linguistic theory posits the existence of distinct representations, among which arc Syntactic Structure, Phonological Form and Logical Form, while a variety of other representations are proposed for fields such as music, vision, and more general cognition. These different representations are interrelated by a variety of Mapping Rules, which enable us to perform variegated tasks such as drawing inferences based on verbal or visual inputs, talking about what we see or hear, singing a song, and so forth. Within the cognitive sciences, it has thus become possible to formulate more explicit and empirically testable versions of Mauthner's conjecture. Specifically, we shall entertain the following three hypotheses: (1)

(a)

Different languages possess different syntactic structures.

(b)

Different languages possess different Logical Forms.

(c)

Different languages are associated with different non-linguistic cognitive representations.

ARISTOTLE GOES TO ARIZONA, AND FINDS A LANGUAGE WITHOUT "AND"

97

Hypothesis (la) is trivially true at the level of S-Structure; however, opinions differ as to whether it is true at the level of D-Structure. Hypothesis (lb) has, to the best of my knowledge, received little or no discussion in the linguistic literature, probably because most scholars tacitly assume it to be false. Hypothesis (lc) is obviously true to the extent that languages mirror their speakers' culture and world view; whether or not languages also affect their speakers' non-linguistic cognitive representations is a question that has occupied the attention of a minority of scholars while being ignored by a much larger majority. Mauthner's conjecture may perhaps be recast in terms of each of the above three hypotheses; however, it is probably most appropriately interpreted as being concerned with hypothesis (lc), specifically with the influence of language upon non-linguistic cognitive representations, or, more specifically, a mental faculty of logical or mathematical reasoning. The above three hypotheses accordingly set the stage for a more seriousminded exploration of Mauthner's conjecture. For this purpose, we shall conduct a thought-experiment. With a time-machine, we shall bring Aristotle forward to the twentieth century. After administering a course in linguistic field methods, we shall escort him to the Salt River and Gila River Reservations in Arizona, where some 500 American Indians speak a Yuman language called Maricopa.1 After a while we shall return, to see how Aristotelian logic has fared among the Maricopas, with reference to the above three hypotheses. Our findings cast doubt on the universality of Aristotelian logic. For it emerges that Maricopa has no and; in fact, it has no uniform device for expressing any of the basic logical connectives of the prepositional calculus. More generally, Maricopa does not possess coordinate constructions, nor does it have a syntactic category of coordinator. Thus, Aristotle's voyage to Arizona provides conclusive evidence for hypothesis (la). However, with regard to the latter two hypotheses, our conclusions are somewhat less unambiguous. The absence of coordination in Maricopa syntax increases the prima facie plausibility that the basic logical connectives are lacking from Maricopa Logical Forms and various other cognitive representations of Maricopa speakers; however, it falls short of tolling a death knell on classical logic. We shall accordingly limit ourselves to a discussion of alternative approaches to the semantics of Maricopa, pointing to various empirical consequences that may be tested through further work in the cognitive sciences. Parts 2 and 3 of this paper deal respectively with the syntax and semantics of Maricopa: in part 2 it is shown that Maricopa syntax has no coordinate constructions and no coordinators, while in part 3 it is asked whether Maricopa Logical Forms and the nonlinguistic cognitive representations of Maricopa speakers make use of conjunctions.

98

DAVID GIL

2. A Language without "And" Coordination is a syntactic construction of the form presented in (2) below, where X denotes an arbitrary syntactic category, and curly brackets indicate constituent structure without linear order: (2)

[xXjX2...Xn

e]

A coordinate construction of syntactic category X thus consists of n expressions also of category X, and, in addition, an expression e, belonging to the category of coordinator. The above formula further specifies that the n coordinated expressions and coordinator e are all directly dominated by the root node of the construction; that is to say, no proper subset of these expressions forms a syntactic constituent. Some examples of coordinations in English are provided below (for each construction, the values of X, n and e as per formula (2) are indicated to its right):3 X

n

e

2

and

(a)

John and Bill

NP

(b)

tall, dark and handsome

AP

3

and

(c) (d)

in town and on the reservation

PP

or

John believes in GB but Bill practices GPSG

S

2 2

(e)

Harry neither drinks nor smokes

VP

2

but neither... nor

In Maricopa, there are no constructions of the above form: there is no coordination, nor is there a syntactic category of coordinator. In order to demonstrate this, we shall examine translations offered by speakers of Maricopa for English coordinate constructions. When asked to translate English coordinations, speakers of Maricopa characteristically offer a wide variety of different construction types. However, none of these constructions are themselves coordinations; moreover, none of them correspond specifically to English coordinations—each occurs also in a wide variety of other contexts, not conresponding to coordinate constructions in English. Thus, Maricopa simply has no equivalent for and, or for any of the other coordinators occurring in English and other familiar languages; it has no coordinate constructions. In order to show that Maricopa has no coordination, we shall examine translations offered by speakers of Maricopa for the following two English sentences:

ARISTOTLE GOES TO ARIZONA, AND FINDS A LANGUAGE WITHOUT "AND"

(4)

(a)

John and Bill will come

(b)

I saw John and Bill

99

The reason for doing so is simple: NPs coordinated with and provide what is perhaps the most natural instance of coordination: if a language has coordinate constructions, it is most likely to use them for renditions of English NPs coordinated with and. Let us now examine some of the strategies used by speakers of Maricopa for translating the above two sentences. The first and simplest strategy involves the bare concatenation of the two NPs:4 (5)

(a)

v9aawuum

Johns

Bills

John-nom

Bill-nom 3-come-pl-fut

"John and Bill will come" (b)

John

Bill

John-acc

Bill-acc

ni^yuuk

pl:obj-l-see-sg-real

"I saw John and Bill' From a formal point of view, the above sentences are the closest Maricopa can get to the English source sentences in (4). However, these constructions fail on a number of grounds to meet the definition of coordination provided in (2) above. First, the constructions in (5) do not contain a coordinator. Nevertheless, one may be tempted to characterize concatenated NPs in Maricopa such as the above as "bare coordinations", containing a phonologically-null coordinator.5 However, the Maricopa facts offer scanty support for such a proposal. In general, when covert, or "zero" elements are posited in linguistic descriptions, they usually form part of a larger paradigm containing other overt, non-zero elements. Thus, for example, positing a phonologicallynull plural indefinite article in English NPs is commonly justified, inter alia, by paradigmatic pressure: the book (singular definite), a book (singular indefinite), the books (plural definite), 0 books.6 However, in Maricopa, there are no overt coordinators coexisting alongside the putative covert coordinators in (5); therefore, no paradigmatic pressure can be invoked in support of phonologically-null coordinators in Maricopa concatenated NPs. A second argument against the analysis of concatenated NPs as coordinate stems from the observation that in constructions such as these, it is unclear whether the two NPs form a syntactic constituent, as is required by the definition of coordination presented in

100

DAVID GIL

(2). Potentially, the two sentences in (5) may be assigned any of the three syntactic structures presented in Figures 1 and 2 respectively:

(a)

(b)

NP

NP

VP

Johns

Bills

John-nom

Bill-nom

3-come-pl-fut

NP

VP

NP Johns

Bills

John-nom

Bill-nom

v?aawuum

v^aawuum

3-come-pl-fut

VP

(c) 9

v aawuum

3-come-pl-fut Figure 1:

Syntactic Structures of (5a) S

(a)

VP ni^yuuk

pl:obj-1 -see-sg-real

ARISTOTLE GOES TO ARIZONA, AND FINDS A LANGUAGE WITHOUT "AND"

101

VP

(b) 7

ni yuuk pl:obj-1 -see-sg-real

YP ni^yuuk

(c)

pl:obj-1 -see-sg-real Figure 2:

Syntactic Structures of (5b)

The structures in Figures la and 2a are "flat" structures, in which each of the two concatenated NPs is an immediate sister of the verb; those in Figures lb and 2b group the two NPs into a single constituent; and those in Figures lc and 2c associate the second NP with the verb. Now for the Maricopa sentences in (5) to contain coordinations, their syntactic structures must be as per Figures lb and 2b. However, I am aware of no evidence to the effect that the two NPs in these constructions form a syntactic constituent. In the absence of such evidence, Occam's razor suggests that such sentences be assigned the simplest available structures, namely those in Figures la and 2a.7 A third argument against characterizing concatenated NPs in Maricopa as involving coordination is semantic. In bona fide cases of bare coordination, such as those illustrated in footnote 5, the construction in question is associated with a specific semantic interpretation. However, the semantics of concatenated NPs in Maricopa is quite vague; in particular, it does not correspond to any particular English coordinator, such as and. Thus, concatenated NPs are also offered by speakers of Maricopa as translations of English sentences containing or:

102

DAVID GIL

(6)

Johns

Bills

John-nom

Bill-nom

v7aawuumsaa

3-come-pl-fut-infer

"John or Bill will come" Comparing (6) with (5a), we see that the difference between the English coordinators and and or finds its expression in Maricopa in the form of the verb: sentence (6) is obtained from (5a) by attaching an inferential suffix -saa to the verb. Thus, "John and Bill" is rendered roughly as "John Bill will...", while "John or Bill" assumes a form similar to "John Bill may ... ". Together, sentences (5a) and (6) show that concatenated NPs in Maricopa do not correspond semantically to and, or to any other specific coordinator in English. The semantic vagueness of concatenated NPs thus provides additional evidence against the existence of a covert, phonologically-null coordinator in Maricopa, and the characterization of concatenated NP constructions as coordinations.8 In fact, it is questionable whether the concatenation of NPs in Maricopa is in any sense the outcome of a syntactic rule. Concatenation per se is not a specifically syntactic notion, or even a specifically grammatical one; concatenated entities are present wherever linearly-ordered behaviour occurs. In dance, motor events are concatenated; in music, pitch and other events are juxtaposed~in neither case, however, does it make any sense to speak specifically of coordination, involving a (covert or overt) coordinator. In language, too, concatenation occurs at a variety of different levels of structure. In morphology, formatives are juxtaposed by compounding, affixation, or other processes; however such concatenation can hardly be considered as coordination.9 Similarly, in discourse, chunks of text are placed together to create larger and larger chunks; for example, a shopping list contains of a sequence of NPs, while this article consists of a series of paragraphs.10 However, the principles governing their construction and interpretation are not syntactic principles; accordingly, they fail to satisfy the syntactic definition for coordination proposed in (2) above. In fact, the contrast between (5a) and (6) suggests that Maricopa may possess no specific rules for the interpretation of concatenated NPs; rather, constructions containing juxtaposed NPs are assigned their meanings via inferences of a general, extragrammatical nature. Just as the meaning of a shopping list, or a sequence of paragraphs, is normally construed as the conjunction of the meanings of its constituent parts, so the meaning of concatenated NPs in Maricopa is most naturally taken as the conjunction of the meanings of the individual NPs. However, particular contexts, such as the inferential verbal inflection -saa in (6), may override this tendency and make other interpretations more readily available. Thus, no grammatical rule of semantic interpretation is necessary in order to provide concatenated NPs with their meanings. This in turn suggests that the

ARISTOTLE GOES TO ARIZONA, AND FINDS A LANGUAGE WITHOUT "AND"

103

concatenation of NPs in Maricopa itself is not the product of a syntactic rule, but rather the outcome of a general discourse process of concatenation.11 Concatenation of NPs is but one of the strategies available to speakers of Maricopa for the translation of English coordinate constructions. More common than NP concatenation, however, are a variety of strategies in which an English coordination corresponds to an embedded clause, whose verb may be loosely characterized as endowed with a coordinating meaning. We shall examine three such strategies, in which the verb of the embedded clause is, respectively, transitive, intransitive reflexive, and copular. Subordination in Maricopa makes use of a system of switch reference marking. Consider the following sentences: (7) (a) Johns

John-nom

Bill

asxamk

Bill-acc

3-hit-sg-real

"John is hitting Bill" (b) Johns

John-nom

Bill

asxamk

Bill-acc

3-hit-sg-ss

v7aawuum

3-come-pl-fut

"John hitting Bill they will come" (c)

Johns

Bill

asxamm

ni^yuuk

John-nom

Bill-acc

3-hit-sg-ds

pl:obj-1 -see-sg-real

"I saw John hitting Bill" Sentence (7a) is a simple transitive sentence exemplifying the basic SOV word order of Maricopa. The subject is suffixed with the nominative case marker -s, the direct object exhibits the zero case marking characteristic of accusatives, and the verb is suffixed with the realis marker -k. Sentences (7b) and (7c) show how the clause comprising (7a) may be embedded in a higher clause, the verb of the subordinated clause being suffixed with a switch reference marker. In (7b), the subject of the embedded verb asxam, namely Johns, is identical to the subject of the higher verb v?aawuum; hence, asxam is suffixed with the same-subject marker -k (which happens to be homophonous to the realis marker in (7a)). In (7c), however, the subject of the subordinated verb asxam, Johns, differs from the subject of the higher verb nPyuuk, which is an understood first person; hence, asxam is suffixed with the different-subject marker -m. Let us now replace the embedded verb asxam "hit" with another transitive verb udaav "accompany". The results of such a move are presented in (8) below:

104

DAVID GIL

(8) (a)

Johns

Bill

uffaavk

John-nom

Bill-acc

3-accompany-sg-real

"John is accompanying Bill" (b)

Johns

Bill

John-nom

Bill-acc

uSaavk

v^aawuum

3-accompany-sg-ss

3-come-pl-fut

"John and Bill will come" H

(c)

Johns

Bill

John-nom

Bill-acc

uifaavm

3-accompany-sg-ds

ni7yuuk

pl:obj-l-see-sg-real

"I saw John and Bill" Morphosyntactically, the three sentences in (8) completely parallel those in (7). Like asxam "hit", uSaav "accompany" is a regular transitive verb in Maricopa: it can be inflected like any other verb, and it occurs in the same syntactic environments-in simple sentences, as in (8a), and in complex ones, as in (8b). However, sentences (8b) and (8c) are offered by speakers of Maricopa as translations of the English sentences in (4). In fact, an embedded transitive clause headed by the verb uSaav "accompany" constitutes what is probably the most common Maricopa equivalent of coordinated NPs in English. Prima facie, then, it would seem as though in Maricopa, and is a verb, uSaav. However, closer inspection reveals that sentences (8b) and (8c) also fail to meet the definition of coordination given in (2). For the two NPs Johns and 5/7/ to be coordinated, they must form a syntactic constituent; however it is clear that they do not. In both sentences, the first NP is marked as the subject of udaav while the second NP is marked as its direct object-regardless of the role of the putative coordinate NP as subject or direct object of the main verbs v7aawuum and ni7yuuk. Taking Maricopa clause structure to be nonconfigurational, the three components of the embedded clause, Johns, Bill and uSaav are all immediately dominated by a (subordinate) S node. Alternatively, if Maricopa has a VP constituent, then the structure of (8b) and (8c) is as as indicated in Figures 3 and 4 below, the direct object forming a constituent with the verb. 12 In neither case do the two NPs Johns and Bill form a syntactic constituent; hence, they can under no circumstances be characterized as coordinated.

ARISTOTLE GOES TO ARIZONA, AND FINDS A LANGUAGE WITHOUT "AND"

Figure 3:

Figure 4:

Johns

Bill

u&aavk

John-nom

Bill-acc

3-accompany-sg-ss 3-come-pl-fut

105

v^aawuum

Syntactic Structure of (8b) S

ni9yuuk

Johns

Bill

udaavm

John-nom

Bill-acc

3-accompany-sg-ds pl:obj-l-see-sg-real

Syntactic Structure of (8c)

Thus, although corresponding to coordinated NPs in English, the Maricopa constructions in (8b) and (8c) do not exhibit coordination themselves. Rather, they present typical instances of clausal subordination, paralleling other instances of subordination, as in (7b) and (7c), which correspond naturally to subordinate constructions in English. A second strategy for the translation of English coordinated NPs into Maricopa embedded clauses involves an intransitive verb with a concatenated NP subject. In (9) below, a clause headed by the intransitive verb xsmee "be tall" is embedded in a higher clause:

106

DAVID GIL

(9) (a)

Johns

Bills

xsmeek

John-nom

Bill-nom

3-tall-pl-real

"John and Bill are tall" (b) Johns

John-nom

v7aawuum

Bills

xsmeek

Bill-nom

3-tall-pl-ss

3-come-pl-fut

"Tall John and Bill will come" (c)

Johns

Bills

xsmeem

ni^yuuk

John-nom

Bill-nom

3-tall-pl-ds

pl:obj- 1-see-sg-real

"I saw tall John and Bill" Sentence (9a) is a simple intransitive sentence with concatenated subject NPs, resembling (5a). Examples (9b) and (9c) show, once again, how the clause comprising (9a) may be embedded in a higher clause, the subordinate verb xsmee "be tall" being suffixed with a switch reference marker-same-subject -k in (9b), different-subject -m in (9c). Let us now, like before, replace the embedded verb xsmee "be tall" with another transitive verb, the reflexive mat teevk "be together": (10) (a)

Johns

Bills

mat; teevk

John-nom

Bill-nom

refl

3-be:together-sg-real

"John and Bill are together" (b)

Johns

Bills

mat; teevk

John-nom

Bill-nom

refl

3-be:together-sg-ss

v7aawuum

3-come-pl-fut

"John and Bill will come" (c)

Johns

Bills

mat

teevm

John-nom

Bill-nom

refi

3-be:together-sg-ds

ni^yuuk

pl:obj-1-see-sg-real

"I saw John and Bill" Morphosyntactically, the three sentences in (10) are completely parallel to those in (9). Like xsmee "be tall", mat teevm "be together" is a regular intransitive verb in Maricopa, taking the same array of inflections, and occurring in the same syntactic environments. However, sentences (10b) and (10c) are also offered by speakers of Maricopa as alternative translations of the English sentences in (4). Indeed, an embedded clause

ARISTOTLE GOES TO ARIZONA, AND FINDS A LANGUAGE WITHOUT "AND"

107

headed by an intransitive verb of a coordinating meaning constitutes yet another common strategy for translating English coordinations. Nevertheless, sentences (10b) and (10c) themselves are not coordinations. The syntactic structures of (10b) and (10c) are represented in Figures 5 and 6 below, following the "flat" analysis of concatenated NPs proposed in Figures la and 2a above:

John-nom Figure 5:

Bill-nom

refl 3-be:to- 3-come-pI-fut gether-sg-ss Syntactic Structure of (10b)

z\ Figure 6:

Johns

Bills

John-nom

Bill-nom

mat, teevm

ni^yuuk

refl 3-be:to- pl:obj-l-see-sg-real gether-sg-ds Syntactic Structure of (10c)

Admittedly, in terms of their geometry, the structures proposed in Figures 5 and 6 above resemble coordinations, with a putative verbal coordinator ma t teev linking the two NPs Johns and Bills. However, for the definition of coordination in (2) to be satisfied, the superordinate constituent must belong to the same category as the coordinated elements, and this condition is not met: combining Johns and Bills with mat teev produces a

108

DAVE) GIL

sentence, not an NP. 13 Most importantly, though, the verb mat, teev "be together" in (10b) and (10c) stands in perfect syntactic parallel to the verb xsmee "be tall" in (9b) and (9c). Since it does not make any sense to characterize xsmee "be tall" as a coordinator, there is also no justification whatsoever for characterizing mat teev "be together" as a coordinator. A third strategy for the rendition of English conjoined NPs into Maricopa embedded clauses involves the copular verb Suu "be" in construction with concatenated predicative NPs. In Maricopa, nominal predicate clauses exhibit somewhat peculiar case marking, with the subject NP occurring with the otherwise accusative zero case marking, and the predicative NP taking the otherwise nominative case suffix -s. 1 4 This case marking is illustrated in (11a), containing subject and predicative NPs, and in (lib) containing an understood subject NP alongside an overt predicative NP:15 (11) (a)

John

kweseSes

John

doctor-nom

Suum

3-be-sg-real

"John is a doctor" (b)

KweseSes

doctor-nom

Suum

3-be-sg-real

"He is a doctor" In construction with concatenated predicative NPs, Suu "be" forms the basis for yet another common translation equivalent of conjoined NPs in English: (12) (a)

Johns

Bills

Suum

John-nom

Bill-nom 3-be-sg-real

"They are John and Bill" (b)

Johns

Bills

nisuum

John-nom

Bill-nom pl:obj-3-be-sg

v7aawuum

3-come-pl-fut

"John and Bill will come" (c)

Johns

Bills

John-nom

Bill-nom pl:obj-3-be-sg

"I saw John and Bill"

niSuum

ni?yuuk

pl:obj-l-see-sg-real

ARISTOTLE GOES TO ARIZONA, AND FINDS A LANGUAGE WITHOUT "AND"

109

Sentence (12a) parallels (1 lb), with kwese&es "doctor" being replaced by concatenated NPs Johns and Bills. Sentences (12b) and (12c) show, once again, how the clause comprising (12a) may be embedded in a higher clause. Sentences (12b) and (12c) are also offered by speakers of Maricopa as additional alternative translations of the English sentences in (4). In general, the copular verb &uu "be" forms the basis for one of the most common strategies for rendering English coordinations into Maricopa. The syntactic strucures of sentences (12b) and (12c) may be represented as in Figures 7 and 8 below:

John-nom Bill-nom pl:obj-3-be-sg 3-come-pl-fut Figure 7: Syntactic Structure of (12b)

John-nom

Bill-nom

Figure 8: Syntactic Structure of (12c)

pl:obj-3-be-sg pl:obj-l-see-sg-real

110

DAVID GIL

Once again, it is evident that the translations of English coordinations into Maricopa constructions involving the copular verb 5uu are not themselves coordinations. While the geometry of the [NP NP V] construction resembles that of coordination, the superordinate constituent is not an NP, as required by the definition of coordination in (2), but rather a VP or a sentence. Moreover, the subordinate clauses of sentences (12b) and (12c), corresponding to the English conjoined NPs, bear the unmistakable form of a Maricopa predicative construction, as exemplified in (ll). 1 6 Accordingly, it makes little sense to characterize these constructions as coordinations. In examples (8), (10) and (12), we encountered three strategies for translating English coordinated NPs into Maricopa, involving embedded clauses headed by transitive, intransitive reflexive and copular verbs. Alongside the bare concatenation of NPs, illustrated in (5), subordinate clauses provide one of the most common means for rendering coordinated NPs into Maricopa. It is eminently clear, however, that such emdedded clauses are not coordinate constructions themselves. Altogether, then, we have examined four lexically and structurally distinct translations of the English coordinated NPs in (4). These four translations do not come close to exhausting the possible strategies for translating English coordinated NPs into Maricopa; a variety of other lexical items and syntactic structures are available in Maricopa for this purpose. Furthermore, for the translation of English coordinate constructions of other syntactic categories, adjectives, verbs and the like, yet additional strategies are available to speakers of Maricopa. Thus, when asked to provide equivalents of English coordinations, speakers of Maricopa offer a plethora of different constructions. As the preceding observations suggest, none of these constructions contain coordinations themselves; nor are any of these constructions limited in their distribution to the equivalents of English coordinate constructions. Thus, Maricopa has no equivalent for English and, or for any other coordinator in English or in other familiar languages. In short, Maricopa has no syntactic category of coordinator, nor does it have any coordinate constructions. The absence of coordination in Maricopa is not an accidental feature of the language, but rather a particular manifestation of a more general typological property, namely a propensity for subordination. As we saw in (8), (10) and (12) above, English coordinate constructions often correspond to embedded clauses in Maricopa. In fact, a variety of other simple English constructions also assume the form of subordinate clauses in Maricopa; among these are quantifiers and adjectival phrases:

ARISTOTLE GOES TO ARIZONA, AND FINDS A LANGUAGE WITHOUT "AND" (13) (a)

tlx ays

boy-nom

111

cumpapk

3-four-sg-real

"The boys are four" (b)

v7aawuum

Mxays

cumpapk

boy-nom

3-four-sg-ss

3-come-pl-fut

"(The) four boys will come" (c)

Mxays

cumpapm

ni?yuuk

boy-nom

3-four-sg-ds

pl:obj-l-see-sg-real

"I saw (the) four boys" (14) (a)

Mxays

xsmeek

boy-nom

3-tall-pl-real

(cf. e x a m p l e (9))

"The boys are tall" (b)

Mxays

xsmeek

boy-nom

3-tall-pl-ss

v7aawuum

3-come-pl-fut

"(The) tall boys will come" (c)

Mxays

xsmeem

boy-nom

3-tall-pl-ds

ni7yuuk

pl:obj-l-see-sg-real

"I saw (the) tall boys" In Maricopa, words like cumpap "four" and xsmee "tall" are verbs, taking the same array of inflections as other verbs, and occurring in the same syntactic environments. For example, in (13a) and (14a), they occur as main verbs of simple sentences. Similarly, in (13b) and (13c), and in (14b) and (14c), they occur as subordinate verbs taking the appropriate switch reference markers. In (13b) and (14b), their subject, mxays" boys", is identical to the subject of the higher verb v7aawuum, and therefore the two verbs are marked with the same subject suffix -k. Conversely, in (13c) and (14c), their subject, mxays, is different from the understood first person subject of the higher verb ni9yuuk, and hence the two verbs are suffixed with the different subject marker -m. Thus, in (13) and (14), English NPs containing quantifiers and adjectives correspond to subordinate clauses in Maricopa. In similar fashion, stacked nominal modifiers in English NPs produce sentences containing multiple embeddings in Maricopa: (15) (a)

Mxays

cumpapk

xsmeek

boy-nom

3-four-sg-ss

3-tall-pl-real

"(The) four boys are tall"

112

DAVID GIL (b)

v9aawuum

Mxays

cumpapk

xsmeek

boy-nom

3-four-sg-ss

3-tall-pl-ss 3-come-pl-fut

"(The) four tall boys will come" (c)

ni7yuuk

Mxays

cumpapk

xsmeem

boy-nom

3-four-sg-ss

3-tall-pl-ds pl:obj-l-see-sg-real

"I saw (the) four tall boys" In each of the above sentences, cumpap "four" is marked with the same-subject suffix -k, since its subject, mxays is identical with that of the next higher verb xsmee "tall". However, while in (15a) xsmee is a main verb, in (15b) and (15c) it too is subordinate, and accordingly takes the appropriate switch reference marker. In (15b) its subject, mxays, is identical to the subject of the main verb v7aawuum, and hence it is marked with the same subject affix -k; however, in (15c) its subject is different from that of the main verb ni ?yuuk, and therefore it is suffixed with the different subject marker -m. Thus, in (13) - (15) above, simple sentences in English correspond to complex sentences in Maricopa. Maricopa accordingly exhibits a strong predeliction for subordination, in contexts where English makes use of flatter structures. Such phenomena may be accounted for by a generalization of the notion of configurationality currendy in vogue in generative grammar. Rather than pertaining exclusively to the presence of X-bar structure, we may view configurationality as a general measure of the complexity, or, more specifically, hierarchical depth, of the syntactic structures of a given 17 language. The above observations suggest that Maricopa syntax may be characterized by a substantially greater degree of configurationality than that of English, the absence of coordinate constructions in Maricopa providing a specific realization of such greater configurationality. Conversely, other languages are clearly characterized by a lesser degree of configurationality than English. Accordingly, while English coordinations often assume the form of embedded clauses in Maricopa, as in (8), (10), and (12), an opposite state of affairs may also obtain, in which coordination or concatenation in other languages corresponds to a variety of English constructions involving a greater degree of hierarchic syntactic depth. Some examples following this pattern are provided in (16) below: (16) (a)

Yoruba

(Stahlke 1970:61)

mo

fi

ada

I

took machete

igi

na

cut tree the

"I cut the tree with a machete"

ARISTOTLE GOES TO ARIZONA, AND FINDS A LANGUAGE WITHOUT "AND"

(b)

113

Vietnamese (Thompson 1965:230-231) Toi di

I

cho

go market

mua

do

buy

things

"I'm going to market to buy some things" (c)

(Thompson and Longacre 1985:175)

Mandarin ta

mei

he neg

nian

shu

ta

da

qiu

le

study

book

he

hit ball asp

"Instead of studying, he played ball" (d)

Hebrew hayeled

yaca

mehaheder

vehiyux

?al

the-boy go:out-past-3:sg:m from-the-room and-smile on

panav

face-3:sg:m

"The boy left the room with a smile on his face" In each of the above examples, a coordination or concatenation in some other language corresponds to an English construction exhibiting a greater degree of configurationality. The first two examples illustrate serial verb constructions characteristic of West Africa and East Asia. In (16a), a sequence of VPs in Yoruba corresponds to an English VP modified by an instrumental PP, while in (16b), consecutive VPs in Vietnamese are matched by an English VP modified by a purpose PP. 18 In (16c), concatenated sentences in Mandarin are rendered into English as a main sentence modified by an adverbial subordinate clause. And in (16d), two Hebrew sentences conjoined with the proclitic ve"and" assume the English form of a single sentence modified by an adverbial prepositional phrase: just as English and disappears in Maricopa, so Hebrew ve- "and" gets lost in translation to English.19 Thus, the above examples show how coordinations and concatenations in other languages may correspond to constructions of greater hierarchic depth in English. They accordingly suggest that the presence of coordination and the productivity of its use in a given language may be correlated with the degree of syntactic configurationality characteristic of that language. As we have seen in this section, Maricopa is endowed with highly configurational syntax, and it has no coordination whatsoever. How extraordinary is the absence of coordination in Maricopa? If we look around at other languages, we will find that it is not that uncommon for English coordinations to be rendered into a variety of constructions not involving coordination. Following are some typical examples:

114

DAVID GIL

(17) (a)

Akan

(Christaller 1875:143, cited in Schachter 1985:49)

ye_-ne

won

abom'

bio

we:sj-be

with-them

have-united

again

"We and they are united again" (b)

Fijian

(Payne 1985:29)

au a

raica

na

I

see

ait chief

past

turaga

kei

na

marama

with

art lady

"I saw the chief and the lady" (c)

Tagalog Naririto

kami

ni

act:top-pres-here

top-l:pl:excl

non:top-pers:sg

Boboy

Boboy

"Boboy and I are here" (d)

Japanese susumu-tati

ga

Susumu-pl

nom

kita

come-pfv

"Susumu and his associates have come" In Akan (17a), English coordinated pronouns assume the form of a transitive clause headed by the enclitic verb ne "be"; while the first pronoun is in subject form, the second pronoun is in object form. The Akan construction is thus reminiscent of Maricopa (8b) and (8c), in which an English coordination is rendered as a subordinate transitive clause.20 In Fijian (17b), coordinated NPs in English are translated into an NP modified by a comitative PP. In Tagalog (17c), coordination of a pronoun and a proper noun in English occurs as a possessive-like construction, consisting of a pronominal head assuming the case, person and number of the entire NP, modified by a genitive proper noun. And in Japanese (17d), an English coordination of the form "[proper noun] and his associates" corresponds to a proper noun suffixed with the "plural" marker -tati. Thus, in the above examples, English nominal coordinations are translated into other languages with a variety of construction types not involving coordinations. Nevertheless, in at least some of the above cases, these constructions exist alongside alternative constructions exhibiting bona fide coordination. What is truly unusual about Maricopa, distinguishing it from most or all of the above cases, is the total absence of coordination in the repertoire of construction types available within the language. The absence of coordinate constructions and a syntactic category of coordinator thus constitutes Aristotle's major finding on his trip to Arizona. This striking property of

ARISTOTLE GOES TO ARIZONA, AND FINDS A LANGUAGE WITHOUT "AND"

115

Maricopa syntax provides conclusive support for version (la) of Mauthner's conjecture, namely that different languages may possess different syntactic structures.

3. A Language without Conjunction? So Maricopa syntax isn't Greek syntax; such a conclusion is hardly surprising. But what about Maricopa semantics: how closely does it resemble Greek semantics? Does the absence of coordination and coordinators in Maricopa syntax entail the absence of logical connectives in Maricopa semantics? Is a language without and also a language without conjunction? How does Aristotelian logic fare amongst the Maricopas? Two different approaches suggest themselves to this problem: universalist and relativist. Each of these approaches enjoys its share of uncritical followers within the linguistic and philosophical communities. Unfortunately, however, there would seem to be precious little critical discussion of the relative merits of these approaches, and possible ways of adjudicating between them. In what follows, I shall accordingly attempt a critical evaluation of these two approaches, pointing to some of the advantages and difficulties inherent therein. The universalist and relativist approaches may be characterized with respect to their attitudes towards versions (lb) and (lc) of Mauthner's conjecture. The universalist approach lays greater emphasis upon commonalities shared by the Logical Forms of different languages and their associated non-linguistic cognitive representations, whereas the relativist approach attaches greater weight to differences between the Logical Forms of different languages and their associated non-linguistic cognitive representations. Of course, few if any universalists would deny the existence of variation in the semantics of different languages, though they might perhaps consider most such differences to be "peripheral" and "uninteresting". On the other hand, only the most stalwart of relativists would adhere to the view that cross-linguistic differences in semantics are so great as to preclude the formulation of semantic universals, though they too might consider most such universals to be "trivial" or "uninteresting". Clearly, then, some semantic categories are universal whereas others are language-specific; the issue is which of these two classes is more important. However, the relative importance of universal and language-specific semantic categories can only be meaningfully assessed within the context of a particular theoretical framework. The universalist and relativist approaches may be illustrated through some specific examples. In Tagalog, like in many East Asian languages, there is a system of honorifics, or politeness markers. One such marker is the clitic po, which, like other clitics in

116

DAVID GIL

Tagalog, occurs in Wackernagel's sentence-second position. Following are some examples of Tagalog constructions containing po: (18) Tagalog (a)

Ibigay

mo

po

pat:top-imp-give non:top-2:sg "Please give me the book" (b)

Ano

po

ang

ang

libro

hon obi obl-l:sg top

book

iinumin

Hindi

akin

ninyo

what hon top pat:top-fut-drink "What would you like to drink?" (c)

sa

non:top-2:pl

po

no hon "No sir" (d)

Mahal

po

ni

Juan

si

love hon non:top-pers-sg "John loves Mary" (0)

John

top-pers-sg

Maria

Maiy

As indicated by the glosses beneath each sentence, there is no unique translation of Tagalog po into English; in each sentence po corresponds to a completely different English form or construction. In (18a) po corresponds to please-, in (18b) it is rendered by the combination of conditional mood and the "polite" lexical item like (as opposed to neutral want); in (18c) it is expressed by sir, and in (18d) it receives no morphosyntactic realization whatsoever in the English gloss. The difficulties faced by a would-be translator of po into English are of course reminiscent of the similar problems faced by someone trying to find an equivalent for and in Maricopa. Analogously, and unsurprisingly, we may conclude that English possesses no morphosyntactic category of honorific. In the present case, however, most linguists would not hesitate to go one step further and claim that English possesses no semantic category of honorific. After all, honorifics are one of the most culture-bound aspects of language, and cultures obviously differ from one another in myriad ways. Why, then, should a particular semantic category be attributed to English just because it is evident in Tagalog? Examples such as these suggest a distinction between between what we might loosely call "cultural" categories, obviously language-specific, and "logical" categories, which are more likely to be universal. However, paradigms similar to (18) above can be constructed for "logical" categories as well. In Georgian, reduplication of verbs, adjectives, quantifiers and adverbs functions as a marker of distributivity:

ARISTOTLE GOES TO ARIZONA, AND FINDS A LANGUAGE WITHOUT "AND"

117

(19) Georgian (a)

vardebi

rose-pl-abs

gamayld-gamayldnen

dist-grow-aor-3:pl

"The roses grew and grew" (b)

imgenma

dzv1r-dzv1r1

saSukrebi

they-erg dist-expensive-abs gift-pl-abs "They bought individually expensive gifts" (c)

(d)

sam-saml

bavsvebi

buy-aor-3:pl

mirbodnen

dist-three-abs child-pl-abs "The children ran in threes"

run-aor-3:pl

mananam

gamoicno

amocanebl

iq'ides

Manana-erg problem-pl-abs solve-aor-3:sg "Manana solved each problem carefully"

prtxll-prtxllat

dist-careful-adv

Again, the English glosses show that there exists no unique translation of Georgian reduplication into English; in each sentence, distributivity is expressed by a totally different English form or construction. In (19a) Georgian reduplication corresponds to a somewhat similarly iterated form in English; in (19b) it is realized in the adverbial individually; in (19c) it assumes the form of a prepositional phrase containing a plural numeral; and in (19d) it is rendered by the prenominal quantifier each. Once again, a single construction in some other language has no unique equivalent in English; this time, however, the construction in question is of a manifestly "logical" rather than "cultural" flavour. We may accordingly conclude that English, unlike Georgian, has no common cross-categorial marking of distributivity. However, the question arises whether to attribute to English a corresponding cross-categorial semantic category of distributivity. Adopting a relativistic approach, we might draw upon the analogy between (18) and (19), concluding that just as English has no semantic category of honorific, so it has no crosscategorial semantic category of distributivity. Alternatively, adopting a universalist approach, we might attempt to use Georgian morphosyntax as a window into English semantics, arguing that a variety of formal strategies, including iteration, adverbials such as individually, the plural numeral construction, and the each operator, all function as 00

markers of a single cross-categorial semantic category of distributivity in English: Similar alternatives face us in our analysis of Maricopa. Adopting a relativistic approach, we might posit different Logical Forms for English and Maricopa, corresponding to their different inventories of syntactic constructions and syntactic categories. Specifically, while English Logical Forms would possess logical connectives, Maricopa Logical Forms would not. In principle, there is no reason not to suppose that

118

DAVID GIL

Universal Grammar may provide for parameters governing variation not just in syntax and phonology but also in semantics. In fact, in the case at hand, a single parameter would be posited to underly both the presence or absence of coordination in Maricopa syntax, and the concomitant presence or absence of conjunctions in Maricopa semantics. However, such a proposal raises certain difficulties. If English syntax were deprived of coordinate constructions, the resulting system, although less powerful, would retain a substantial amount of its original structure. However, within most of the semantic representations proposed for English, logical connectives play a central, indispensible role: remove them, and the rest collapses like a pack of cards. This is equally true for the Boolean systems of formal categorial semantics (Montague 1973, Keenan and Faltz 1986), as it is for the generative grammarians' Logical Forms based on the Propositional and Predicate Calculus (Chomsky 1981, May 1985). So if English semantics is classical while Maricopa semantics is otherwise, the semantics of these two languages will have to be based on fundamentally different systems; their semantics will thus have to differ from one another far more radically than their syntax. Pushing the relativist approach even further, one might posit not only different Logical Forms but also different non-linguistic cognitive representations for speakers of English and Maricopa. In such a case, a single parameter of Universal Grammar would underly not just a syntactic property and a linguistic semantic one, but also a non-linguistic general cognitive property. Specifically, while the cognitive representations of English speakers would contain the usual logical connectives, those of Maricopa speakers would not. From an acquisitional point of view, the child would infer from primary linguistic data whether his language contained coordination, and would then construct his linguistic semantic and non-linguistic cognitive representations in accordance. Such a proposal would thus be consonant with the Sapir-Whorf hypothesis of linguistic relativity. Alternatively, one might propose-this time contra Sapir-Whorf-that the non-linguistic cognitive representations were primary and the linguistic facts derivative. In this case, though, it is hard to envisage what kind of non-linguistic data would be available to the child, enabling him to set the appropriate value of the parameter in question. The attribution of different non-linguistic cognitive representations to speakers of English and Maricopa raises a problem concerning the competence of bilinguals. If the two languages are associated with different non-linguistic cognitive representations, then one might predict that bilinguals would encounter difficulties when faced with the task of fyy

translating from one language to the other. To some extent, this may be the case. However, as shown in the previous section, bilingual speakers are nevertheless capable of translating relatively freely between the two languages. Consider, now, the task performed by a bilingual speaker translating, say, Maricopa sentence (8b), exhibiting

ARISTOTLE GOES TO ARIZONA, AND FINDS A LANGUAGE WITHOUT "AND"

119

subordination, into English sentence (4a), exemplifying coordination. This task may be represented schematically in terms of a derivation beginning with (8b) and passing through n linguistic or general cognitive representations until reaching (4a): (20) (8b)

Rx

R2

R n - > (4a)

Presumably, at some point in the derivation there will exist two consecutive representations R; and R i+1 , such that R ; does not contain coordinators or logical connectives but R i+1 does. Thus, in passing from Rj to R i+1 , Maricopa bilinguals prove themselves capable of creating coordinators or logical connectives ex nihilo, as it were, in order to produce English translations of Maricopa sentences. Given this capacity, the question arises why not attribute it also to monolingual speakers of Maricopa? In effect, this would entail positing a (linguistic or general cognitive) level of representation, that of Rj + i, which would be universal, and would contain coordinators or logical connectives. We thus arrive at the second approach, the universalist one. In its weaker version, the universalist approach posits the same logical connectives for the non-linguistic cognitive representations of speakers of all languages. In its stronger version, the universalist approach assumes also that the same logical connectives are present in the Logical Forms of all languages. Generative grammarians may keep their Propositional and Predicate Calculi, formal semanticists their Boolean Algebras; the difficulties are now shouldered entirely by Maricopa linguists, who must devise a way to get from the exotic syntactic structures of Maricopa to the familiar, comforting forms of classical logic. Quite clearly, some complicated footwork is required, lowering verbs such as uSaav, mat ieev and &uu, and collapsing subordinate structures into coordinate ones. But after all, such rules are not entirely unfamiliar; in fact, a rule of Conjunction Lowering in Maricopa would in many ways present a near mirror image of the currently fashionable rule of Quantifier Raising in English, which takes a determiner within an NP and moves it out into a higher clause. However, the problem remains: how does the Maricopa child acquire such rules? With no primary data to suggest the existence of logical connectives in his semantic representations, the Maricopa child would seem to be faced with a formidable task of acquisition. Of course, orthodox generative grammar has an ingeniously simple solution to this problem: the Maricopa child never has to acquire logical connectives, for he has them all along—they are innate. Indeed, perhaps they are, but one wishes one had some further and independent reason to believe this. Unfortunately, there seem to be no knock-down arguments for either relativist or universalist approaches to the semantic issues raised by the absence of a syntactic category of conjunction in Maricopa. Are these, then, really unanswerable questions? I

120

DAVID GIL

believe not. However, in order to adjudicate between the various alternatives, it may be necessary to step outside the confines of linguistic theory, in order to examine the issues at hand within a broader perspective of cognitive science, and a theory of the human mind. Specifically, in order to determine whether Maricopa speakers have at their disposal the classical logical connectives of Boolean Algebra and the Prepositional Calculus, we may search for evidence that these connectives are required within other cognitive domains, such as music or vision. If, within a theory pertaining to some other cognitive domain, unambiguous evidence arises for the existence of representations containing the classical logical connectives, then it may be concluded that these connectives are available elsewhere, including natural language semantics. If, on the other hand, no such evidence from other cognitive domains is forthcoming, then the likelihood decreases that the logical connectives are a necessary part of the semantics of Maricopa, and, ipso facto, of natural languages in general. Consider, for example, man's presumably universal visual capacity to recognize various two-dimensional images as belonging to a single rigid three-dimensional object in different spatial orientations. This capacity is what enables us to carry out such tasks as identifying a triangle, a sector and a circle as projections of the same three-dimensional cone. In order to be able to perform such feats, we must be able, among other things, to compute the possible rotations of a rigid body in space. Now the set R of rotations of a rigid body in space exhibits a particular mathematical structure, that of a monoid. For any two rotations x and y in R, we can define their composition, xy, resulting from the application of x and then y. Such composition is associative; that is to say, for any x,y,z in R, x(yz) = (xy)z. Moreover, R contains a unit element e, such that for any x in R, ex = x; e is the "zero rotation", in. which the body in question is not moved. The set R of rotations of a rigid body in space thus satisfies the axioms of a monoid. Our capability to compute the rotations of a rigid body in space accordingly indicates that our visual competence presupposes, inter alia, mastery of the algebraic properties of monoids. Now it so happens that a Boolean Algebra also satisfy the axioms of a monoid, the binary operation in question being none other than the conjunction A. Thus, conjunction is associative, A A (B A C) = (A A B) A C; and the unit element X is the maximal element of the Boolean Algebra, the "universe". Given that mastery of monoids is required by our visual competence, the monoid-theoretic properties of Boolean Algebra are thus available to semantic theory free of charge, as it were, without need of positing any additional mental apparatus. Quite simply, if we are capable of constructing compositions of rotations, we must also be capable of constructing conjunctions of propositions, and of other elements in a Boolean Algebra. We may accordingly posit an autonomous algebraic competence, containing, inter alia, mastery of the algebraic properties of monoids; this

ARISTOTLE GOES TO ARIZONA, AND FINDS A LANGUAGE WITHOUT "AND"

121

mental faculty is subsequently drawn upon by other faculties, including vision and semantics. On closer inspection, however, it is evident that the algebraic parallel between a Boolean Algebra and the set R of rotations of a rigid object in space is not allencompassing. The set R, with composition as defined above, also contains inverses; that is to say, for any x in R there is an x 1 such that x x 1 = e. It thus satisfies the more stringent axioms of a group. However, composition is not commutative: for any x and y in R, it is not necessarily the case that xy = yx. Conversely, Boolean Algebras are not endowed with group structure; members of a Boolean Algebra do not generally contain inverses. However, the conjunction A is commutative: for any A and B in the Boolean Algebra, A A B = B A A. Thus, Boolean Algebras exhibit a somewhat different structure, that of a commutative monoid. These observations accordingly weaken somewhat the parallel between composition in R and conjunction in a Boolean Algebra, and, ipso facto, the force of the argument inferring the availability of conjunctions to semantic theory from the existence of compositions in the set R of rotations of a rigid body in space. In order to adduce stronger support for logical connectives in semantic theory, we must therefore seek more specific cases in which they form part of various other mental representations. One possible example of this is provided by Marr and Hildreth (1980), cited in Marr (1982:66). In order to account for our capacity to detect oriented zerocrossing segments, part of our ability to see edges of objects, they posit a logical AND gate, a device that produces a positive output if and only if all of its inputs are positive. Here, within a relatively specialized domain in the theory of vision, there surfaces a clearcut example of a cognitive representation involving a classical logical connective, in fact, the conjunction A. If Marr and Hildreth's analysis is correct, then our visual competence must have recourse to conjunctions. Examples such as this accordingly provide strong prima facie plausibility to the claim that conjunctions are available also to semantic representations, even for languages such as Maricopa, whose syntactic structures provide no linguistic evidence for their existence. Thus, whether Maricopa semantics contains the connectives of classical logic may ultimately be answerable outside of linguistics, within the framework of a broader theory of the human mind. If we encounter solid evidence for the existence of logical connectives within other cognitive domains, then a parsimonious theory of mind, one in which the totality of our cognitive capacities is accounted for with a minimum of theoretical apparatus, leads towards the conclusion that such logical connectives may also be available for the representations of meanings, in Logical Form and in non-linguistic cognitive representations. Conversely, if such evidence is not forthcoming, then the plausibility increases that the representations of meanings in natural languages may not

122

DAVID GIL

make use of the classical logical connectives either. The observations of the preceding few paragraphs suggest that such evidence may indeed be available. However, in order to adjudicate conclusively between universalist and relativist approaches, and decide whether Maricopa semantics contains the connectives of classical logic, we must await the development of a more fully articulated cognitive science, consisting of separate theories of individual autonomous modules and faculties, brought together by an overall theory of the way these modules and faculties interact.

4. An Even Longer Journey As we have seen, Aristode's journey to Arizona does not provide conclusive evidence for or against the universality of Aristotelian logic. Mauthner's conjecture remains just that—a conjecture; hypotheses (lb) and (lc) formulated at the beginning of this paper still await confirmation or refutation. Nevertheless, Aristode's sojourn among the Maricopas has proven beneficial in a number of respects. To begin, it has established that languages may vary with respect to the presence or absence of coordination: while English and many other familiar languages possess coordinate constructions and a category of coordinator, Maricopa is entirely lacking in these syntactic objects. Beyond this substantive result, two methodological morals emerge from Aristotle's voyage. First, natural language semanticists must begin emulating their colleagues in syntax and phonology, and start thinking seriously about looking at natural languages (with emphasis on the plural). In his contribution to this volume, van Benthem asks: "What is the universal validity of the conceptual frameworks, and claims based upon these, thought up in the arm-chair of a native speaker of some Indo-European language? Can purported truths about Language (singular) reached in this convenient manner be maintained in the light of evidence about other languages (plural), obtained through hazardous journeys across arid deserts and steaming jungles?" Elswhere, (Gil 1982a,b,1987,1988), I have argued that we must indeed leave our armchairs to embark on these "hazardous journeys". The results of this paper provide further support for this view. Regardless of the ultimate fate of Mauthner's conjecture, and of the classical logical connectives in Maricopa, the absence of coordination in the syntax of Maricopa raises questions that no natural language semanticist can afford to ignore. However, there is a second, more far-reaching moral. As suggested in the previous section, in order to determine whether the classical logical connectives are available to speakers of Maricopa, we must transcend the confines of linguistic theory and search for

ARISTOTLE GOES TO ARIZONA, AND FINDS A LANGUAGE WITHOUT "AND"

123

logical connectives in other, non-linguistic, cognitive domains. Only within a broader cognitive science containing theories of language, vision and other mental faculties can we hope to resolve Mauthner's conjecture one way or another; only within a wider theory of the mind can we ultimately determine the nature of man's grammatical competence. What this means is that semanticists, together with other linguists, must embark on a journey even more adventurous than that across arid deserts and steaming jungles, namely, a journey through the mind, across the variegated mental faculties with which man is endowed.

Acknowledgements This paper would not have been possible without the help and infinite patience of Polly Heath, who introduced me to the intricacies of Maricopa syntax and semantics. The Maricopa data cited in this paper were elicited within the framework of a Field Methods course at UCLA during 1980/1; I am grateful to Pam Munro, who taught the course, and to my fellow students Vance Hamm, Stephen Siemens and Tracy Thomas-Flinders, for numerous insightful discussions. For the Turkish and Georgian data cited in this paper I am indebted to Deborah Azaryad and Manana Bat-Hana respectively. This paper was presented at the Working Group on Universal Semantic Structures and Their Social Motivation, at the Annual Meeting of the Deutsche Gesellschaft fur Sprachwissenschaft, at Wuppertal, West Germany, on 4 March 1988; I am grateful to several participants of that conference for their helpful comments.

Notes 1. For descriptions of various aspects of Maricopa grammar, see Gordon (1979,1983,1986), Harwell (1976), and Thomas-Flinders (1981). Since the Yuman family is part of the Hokan phylum, Maricopa may be distantly related to Dacota, referred to above by Fritz Mauthner - within the "Northern Amerind" group posited by Greenberg (1987). 2. Occasionally, constructions such as (2), and also expressions such as e, are referred to as "conjunctions". In this paper, we shall reserve the use of the term "conjunction" for the logical connective commonly denoted with the symbol "A".

3. In English, and in other languages, coordinators typically cliticize to one of the coordinated elements; for example, in (3a) above, and cliticizes to Bill. (In some languages, e.g. Hebrew, such cliticization is

124

DAVID GIL

reflected in the orthography, the coordinator and the expression it attaches to being written as a single word~cf. (16d) below.) However, such grouping would appear to be of a phonological rather than a syntactic nature, and hence the definition of coordination in (2) above is not violated.

4. The transcriptions of the Maricopa data cited in this paper are close to the standard IPA conventions; the only point worthy of mention is the distinction between dental t and alveo-palatal f. (However, the present transcription differs from that of most Yumanists, who prefer a more English-like orthography.) In the morpheme-by-morpheme glosses of the Maricopa data cited in this paper, we make use of the following abbreviations: acc "accusative"; ds "different subject"; fut "future"; infer "inferential"; nom "nominative"; obj "object"; pi "plural"; real "realis"; refl "reflexive"; sg "singular"; ss "same subject"; 1 "first person"; 3 "third person". In the morpheme-by-morpheme glosses of data from other languages, we shall also use the following additional abbreviations: abs "absolutive"; act "actor"; adv "adverbial"; aor "aorist"; art "article"; asp "aspect"; dist "distributive"; erg "ergative"; excl "exclusive"; hon "honorific"; imp "imperative"; m "masculine"; neg "negation"; obi "oblique"; pat "patient"; pers "personal"; pfv "perfective"; pres "present"; sj "subject"; top "topic"; 2 "second person". 5. A similar analysis is suggested, for a variety of constructions in other languages, by Payne (1985:2527). Some constructions which might plausibly be characterized as containing zero coordinators include the following: (i)

Turkish Debora

Megi

Debora

Megi

girdiler

• come:in-pfv-pl

"Debora and Megi came in" (ii)

Hebrew axalti

arbafa

hamiSa

afarsimonim

eat-past-l:sg

four-m

five-m

persimmon-pl:m

"I ate four or five persimmons"

In Turkish, bare concatenation of NPs is interpreted as conjunction, with the further proviso that the conjoined constituents be conceptually close; for example, (i) is appropriate in the case of Debora and Megi being sisters. In Hebrew, bare concatenation of numerals is interpreted as disjunction.

6. Similarly, the analysis of the constructions cited in the previous footnote as containing phonologically-null coordinators may be supported by the existence of alternative constructions

ARISTOTLE GOES TO ARIZONA, AND FINDS A LANGUAGE WITHOUT "AND"

125

containing an overt coordinator between the coordinated constituents, ve "and" in Turkish, o "or" in Hebrew.

7. The structures in Figures la and 2a are similar to those occasionally posited for basic sentences in nonconfigurational verb-final languages; those in Figures lb and 2b are appropriate for coordinated NPs as in English (4); and those in Figures lc and 2c are isomorphic to those sometimes posited for basic sentences in verb-final languages with a VP constituent, and for the so-called "double subject" constructions characteristic of Chinese and Japanese-see, for example, Teng (1974). Analogous parsing problems are presented by a variety of constructions in English:

(i)

Ugly little man

(ii)

Harry flew from Athens to Phoenix

(iii)

Whistling Zorba the Greek, carrying his book on Aristotelian Logic, Harry walked into the room

In the above three examples, consecutive expressions belonging to the same syntactic category-AP, PP, and participial phrase, respectively-occur in construction with another expression; in these cases, it is not immediately obvious whether indeed the consecutive expressions form a syntactic constituent, to be characterized as instances of bare coordination, or whether some other syntactic structure is more appropriate.

8. According to Payne (1985:39), citing DOhmann (1974:41-42), similar examples of concatenated NPs being interpretable as either conjunctions or disjunctions occur in Sumerian, Ancient Egyptian and Chinese. 9. For example, in the Sanskrit compound devagandharvam3husoragar3ksas3h

"gods, heavenly

singers, men, serpents, and demons" (Payne 1985:26), nominal stems are concatenated in a morphological analogue to the Turkish bare coordination cited in footnote 5. Similarly, in the Kwakiutl form bdgwo'lbidu' "ugly and little man" (Boas 1911:493), the stem bagw- "man" takes the concatenated suffixes o7 "ugly" and -bidu' "little", in a morphological parallel to stacked adjective constructions such as example (i) in footnote 7. At a somewhat more abstract level, we find concatenation not only of formatives but also of inflectional categories. For example, in the Hebrew form susot "mares", the stem sus- "horse" is suffixed with the single formative -ot, synthesizing the inflectional categories of feminine and plural; here, too, such concatenation can be paraphrased, albeit awkwardly, with an English coordination, "female and numerous horses". Nevertheless, constructions such as the above involving morphological concatenation fail to satisfy the definition of coordination in (2) for the simple reason that they are not syntactic constructions.

126

DAVID GIL

10. In literature, the concatenation of phrases by non-syntactic strategies constitutes an important device characteristic of a number of writers, including Walt Whitman and Dylan Thomas (see Goodblatt, 1990, to appear, for discussion). The following passage, from Dylan Thomas' play Under Milk Wood (Aldine Paperbacks, London, p.2), contains juxtaposed sentences, beginning "It is night...", each sentence in turn containing concatenated NPs, PPs, or participial phrases, with only sporadic occurrences of the coordinator and: "Listen. It is night in the chill, squat chapel, hymning in bonnet and brooch and bombazine black, butterfly choker and bootlace bow, coughing like nannygoats, sucking mintoes, fortywinking hallelujah; night in the four-ale, quiet as a domino; in Ocky Milkman's lofts like a mouse with gloves; in Dai Bread's bakery flying like black flour. It is to-night in Donkey Street, trotting silent, with seaweed on its hooves, along the cockled cobbles, past curtained fernpot, text and trinket, harmonium, holy dresser, watercolours done by hand, china dog and rosy tin teacaddy. It is night neddying among the snuggeries of babies." 11. Similar examples can be constructed in English of concatenated NPs whose interpretation, as conjunction or disjunction, is determined by context: (i)

(ii)

Speaker A:

Which three electives would you like to register for next semester?

Speaker B:

Aristotelian Logic, Government and Binding, Maricopa Grammar.

Speaker A:

Which elective would you like to register for next semester?

Speaker B:

Aristotelian Logic, Government and Binding, Maricopa Grammar (... I really don't care which).

While in (i), the concatenated NPs are interpreted as a conjunction, in (ii) the same NPs are interpreted as a disjunction. Speaker B's response may therefore be most appropriately characterized as the product of an extra-grammatical discourse rule of concatenation, whose interpretation is at least partially determined by context.

12. The tree diagrams presented in Figures 3 and 4, and in subsequent figures, are intended only as broad and (relatively) theory-independent indications of the syntactic structure of the appropriate Maricopa sentences. Many questions are left moot, especially regarding the precise nature of the subordinate switchreference clauses: are they adverbial, as is the Received View amongst Yumanists, or do they constitute nominal arguments of the matrix verbs?

13. Even if we adopt the alternative analysis, mentioned previously, whereby embedded switch-reference clauses are nominal in nature, the NP node constituting an argument of the main verb would still have to dominate an S node; this is evidenced by the fact that (10a) is a grammatical simple sentence. Accordingly, the definition of coordination in (2) would still fail to be satisfied.

ARISTOTLE GOES TO ARIZONA, AND FINDS A LANGUAGE WITHOUT "AND"

127

14. This case marking property is a characteristic feature of Yuman languages; see Munro (1977) for discussion. 15. A certain peculiarity of Maricopa verbal morphology must be noted at this point Maricopa verbs belong to two classes, conveniently referred to as k- and m-verbs respectively. With Jfc-verbs, the realis and same-subject suffixes are the homophonous -k, while the different-subject suffix is -m. All the verbs we have encountered so far have been it-verbs. With m-verbs, however, the -k/-m distinction collapses, and the realis, same-subject and different-subject suffixes are all homophonous, -m. The copular verb &uu is an m-verb, and hence, in the examples below it always occurs with the suffix -m. It can be shown, however, that this difference is a purely morphological one, with no syntactic or semantic consequences. Specifically, when an m-verb is suffixed with one or more of a set of so-called non-final suffixes, such as focus -t, negative -ma, and distributive -xper, it behaves as a ¿-verb, taking either -k or -m suffixes, whichever is appropriate. 16. A minor exception to this generalization is the occurrence of the plural object prefix hi-on the verb ¿fuu in (12), for which I have no explanation.

17. For a more detailed defense of this proposal, see Gil (1983). 18. In serial verb constructions such as in (16a) and (16b) above, it is questionable whether the two verbs form a coordinate construction, with a phonologically-null coordinator, or whether they are the product of a more general rule of concatenation; for discussion of some of the issues involved in the analysis of such constructions, see Giv6n (1975) and Hyman (1975).

19. Example (16d) suggests a sequel to the present paper, titled "Abraham goes to Arizona, and finds a language without 'and'", the language in question being, in the present case, English (restricted to cases like the present one).

20. Constructions such as (17a) form the basis of the following anecdote, reported by Margaret A. Bryan to William E. Welmers, and cited in Welmers (1973:372): "The late Ida C. Ward, like anyone else in the field of African languages, was often asked, 'Aren't those African languages you work on terribly primitive?' Her gentle but effective reply was, "Well, in one West African language I could give you the pluperfect subjunctive of the verb "and"!'"

21. For a detailed description and analysis of reduplication in Georgian, see Gil (1982a,1988).

128

DAVID GIL

22. This latter approach is in fact the one adopted in Gil (1982a,1988).

23. My experiences with one speaker of Maricopa do not permit me to draw any definitive conclusions on this score. However, in her study of a related Yuman language, Diegueflo, Langdon (1970:172) reports that "informants typically are uncomfortable when asked to translate English sentences with conjoined nouns."

References Benthem, Johan van: this volume, 'Linguistic Universals in Logical Semantics'. Boas, Franz: 1911,'Kwakiutl', in F. Boas ed., Handbook

of American

Indian

Languages, Part I, Bureau of American Ethnology Bulletin 40.1, Government Printing Office, Washington, 423-557. Chomsky, Noam: 1981, Lectures on Government and Binding, Foris, Dordrecht. Christaller, J.G.: 1875, A Grammar of the Asante and Fante Language Called Tshi [Chwee, Twi], Basel Evangelical Missionary Society, Basel, reprinted by Gregg Press, Ridgewood, 1964. Döhmann, K.: 1974, 'Die sprachliche Darstellung logischer Funktoren', in A. Menne and G. Frey (eds.), Logik und Sprache, Francke, Bern, 28-56. Gil, David: 1982a, Distributive Numerals, PhD Dissertation, University of California, Los Angeles. Gil, David: 1982b, 'Quantifier Scope, Linguistic Variation, and Natural Language Semantics', Linguistics and Philosophy 5, 421-472. Gil, David: 1983, 'Stacked Adjectives and Configurationality', Linguistic Analysis 12, 141-158. Gil, David: 1987, 'Definiteness, Noun-Phrase Configurationality, and the Count-Mass Distinction', in E. J. Reuland and A.G.B, ter Meulen eds., The Representation of (In)definiteness, MIT Press, Cambridge, 254-269. Gil, David: 1988, 'Georgian Reduplication and the Domain of Distributivity', Linguistics 26, 1039-1065. Givön, Talmy: 1975, 'Serial Verbs and Syntactic Change: Niger-Congo', in C.N. Li (ed.), Word Order and Word Order Change, University of Texas Press, Austin, 47112.

Goodblatt, Chanita: 1990, 'Whitman's Catalogues as Literary Gestalts: Illustrative and Meditative Functions', Style 24, 45-58.

ARISTOTLE GOES TO ARIZONA, AND FINDS A LANGUAGE WITHOUT "AND"

129

Goodblatt, Chanita: to appear, 'Semantic Fields and Metaphor: A Case Study', Journal of Literary Semantics. Gordon, Lynn M.: 1979, '-k and -m in Maricopa1, in P. Munro ed., Studies of Switch Reference, UCLA Papers in Syntax 8, University of California, Los Angeles, 119143. Gordon, Lynn M.: 1983, 'Switch-Reference, Clause Order, and Interclausal Relationships in Maricopa', in J. Haiman and P. Munro eds., Switch Reference and Universal Grammar, John Benjamins, Amsterdam, 83-104. Greenberg, Joseph H.: 1987, Language in the Americas, Stanford University Press, Stanford. Harwell, Henry O.: 1976, 'The 'Say' Auxiliary in Maricopa, Some Notes and Speculations', Southern Illinois University Museum Studies 7, 63-70. Hyman, Larry M.: 1975, 'On the Change from SOV to SVO: Evidence from NigerCongo', in C.N. Li (ed.), Word Order and Word Order Change, University of Texas Press, Austin, 113-147. Keenan, Edward L., and L. M. Faltz: 1986, Boolean Semantics for Natural Language, Reidel, Dordrecht. Langdon, Margaret: 1970, A Grammar of Diegueno, The Mesa Grande

Dialect,

University of California Press, Berkeley. Marr, David: 1982, Vision, W.H. Freeman and Company, New York. Marr, David, and Ellen Hildreth: 1980, 'Theory of Edge Detection', Proceedings of the Royal Society of London B 207, 187-217. May, Robert: 1985, Logical Form, Its Structure and Derivation, Linguistic Inquiry Monograph Twelve, MIT Press, Cambridge. Montague, Richard: 1973, 'The Proper Treatment of Quantification in Ordinary English', in J. Hintikka, J. Moravcsik, and P. Suppes eds., Approaches to Natural Language: Proceedings of the 1970 Stanford Workshop on Grammar and Semantics, Reidel, Dordrecht, 221-242. Munro, Pamela: 1977, 'From Existential to Copula: The History of Yuman BE', in C.N. Li (ed.), Mechanisms of Syntactic Change, University of Texas Press, Austin, 445490. Payne, John R.: 1985, 'Complex Phrases and Complex Sentences', in T. Shopen ed., Language Typology and Syntactic Description, Volume II, Complex Constructions, Cambridge University Press, Cambridge, 3-41. Schachter, Paul: 1985, 'Parts-of-Speech Systems', in T. Shopen ed., Language Typology and Syntactic Description, Volume I, Clause Structure, Cambridge University Press, Cambridge, 3-61.

130

DAVID GIL

Stahlke, Herbert F.: 1970, 'Serial Verbs', Studies in African Linguistics 1, 60-99. Teng, Shou-hsin: 1974, 'Double Nominatives in Chinese', Language 50, 455-473. Thomas-Flinders, Tracy: 1981, 'Aspects of Maricopa Verb Morphology', in T. ThomasFlinders ed., Inflectional Morphology: Introduction to the Extended Word-andParadigm Theory, UCLA Occasional Papers 4: Working Papers in Morphology, University of California, Los Angeles, 158-186. Thompson, Sandra A., and Robert E. Longacre: 1985, 'Adverbial Clauses', in T. Shopen ed., Language Typology and Syntactic Description, Volume II, Complex Constructions, Cambridge University Press, Cambridge, 171-234. Thompson, L.C.: 1965, A Vietnamese Grammar, University of Washington Press, Seattle. Weiler, Gershon: 1970, Mauthner's Critique of Language, Cambridge University Press, Cambridge. Welmers, William E.: 1973, African Language Structures, University of California Press, Berkeley.

Part III Special Topics

Quantity and Number Godehard Link Seminar für Philosophie, Logik und Wissenschaftstheorie Universität München Ludwigstr. 31 D-8000 München 22

1.

In this paper I am going to focus on the relation between the syntactic notion of

number and the semantic notion of quantity from the perspective of logical semantics. The paper consists of two parts. In the first I discuss the conceptual issue of individuation as a prerequisite to counting. In this context classifier constructions will serve as illustration for some remarks on the philosophical problem of

ontological

relativity. The second part of the paper is concerned with a certain kind of structural relativity in quantifying phrases. Following Keenan (1987) a distinction is made between autonomously referring expressions and those with dependent reference. It is only the former that indicate the true amount of individuals or material in the world that they refer to; the amounts given by the dependent expressions have to be derived from the denotations of noun phrases taking scope over them, and/or often from an iterative interpretation of the given sentence. Each part ends with a semantic model that is meant to capture the basic logic of the linguistic phenomena discussed.

1.

Ontological

Relativity.

2.

It is of course no accident that the English word 'number' (like its Latin origin

'numerus') stands both for the mathematical concept of number and the grammatical distinction between singular and plural. The paradigmatic function of the singular of a given count noun N like apple is to pick out a single object of the kind N\ the corresponding plural stands for some collection of objects of the same sort which can be counted, that is, numerically specified. For the process of counting to make sense at all there has to be some appropriate discrete unit available. In the case of count nouns this unit is naturally given: it is the objects themselves that such a noun is true of. Mass nouns like gold or water, on the other hand, need some special lexical item expressing the unit in terms of which the given quantity is measured (e.g. three ounces of gold, two cups of water). So the process of counting is but a special case of the process of measuring quantities: in general, we have to fix a particular unit first, and only then we can proceed to take multiples or proportions to determine the measure of the quantity at hand. This is the procedure in science (for an account of basic measurement theory, see Krantz et al.

134

GODEHARDLINK

1971), but (experimental) science is no more than theoretically guided and sophisticated cultural practice. This is why we find the practice of measurement extensively encoded in language. Relative to a given unit the absolute amount is represented by the class of numerals (NL; let us ignore proportion expressions like three quarters of); with its strict mathematical meaning their structure shows little cross-linguistic variation. What counts as unit, however, is far less predetermined conceptually, and outside the class of count nouns the measure words for mass terms or transnumeral nouns (for the latter term see Biermann 1981) abound accordingly. An object may be measured along different dimensions (e.g. length, width, volume, weight); within the same dimension there may be several lexicalized units that are multiples of each other {inch, foot, ounce, pound-, discrete objects, too, can be taken in pairs, dozens, etc.). In less standardized contexts we have cups, mugs and bottles. Those measure words or numeratives (NM; Krifka 1989) are obligatory with mass terms. The reason is, of course, that it is the measure word that divides up the continuous stuff denoted by the mass noun into discrete, countable units. The situation is slightly different with collective terms like cattle send furniture. If somebody speaks of "the catde over there" we are able to count the animals referred to, and yet, cattle has to be preceded by the numerative head of before we get a well-formed measure phrase like six head of cattle. One might wonder now what the reference (the "extension") of a term like cattle is; it cannot be the class of bovine animals, or the measure word head of would have no semantic function. While this is possible in principle it is not very likely; rather, the grammaticality of the definite NP the cattle over there gives a clear indication as to what the proper extension is: the term cattle refers not to single animals but to collections of those. It is only when we want to single out individual animals that we have to use the measure word head of The difference, then, between collective terms and mass terms is the feature idiscrete: the former stand for discrete, the latter for continuous totalities. Usually, three head of cattle and the like is considered as one of the rare examples of a classifier construction in English, a pervasive phenomenon in the so-called classifier languages (for a survey see Allan 1977). The nouns in those languages, it is said, refer to unstructures wholes, and it is the classifier (CL) that fabricates the units for counting. Thus, even nouns standing for objects that are perceptually clear-cut enough to be referred to by regular count nouns in Western languages like English have to be preceded by such a classifier. 3.

It is interesting to note that W. v. O. Quine used the phenomenon of classifier

languages as an example to illustrate his famous philosophical thesis of the inscrutability of reference. In his Ontological Relativity (Quine 1969: 35 - 38) he writes:

QUANTITY AND NUMBER

135

In Japanese there are certain particles, called "classifiers," which may be explained in either of two ways. Commonly they are explained as attaching to numerals, to form compound numerals of distinctive styles. Thus take the numeral for 5. If you attach one classifier to it you get a style of "5" suitable for counting animals; if you attach a different classifier, you get a style of "5" suitable for counting slim things like pencils and chopsticks; and so on. But another way of classifiers is to view them not as constituting part of the numeral, but as constituting part of the term - the term for "chopsticks" or "oxen" or whatever. On this view, the classifier does the individuating job that is done in English by "sticks o f as applied to the mass term "wood," or "head of' as applied to the mass term "cattle." Quine says here that there are two possibilities of parsing a classifier phrase NL+CL+N, either (i) [NL+CL]+N or (ii) NL+[CL+N]. Version (i) would mean conceptually to "parametrize" the notion of number; thus we would get, for instance, "animal-five" or "slim-thing-five" as different counterparts of the number Five depending on what is there to be counted. Version (ii), on the other hand, is modeled after the English 'head-of paradigm: here the classifier in an operator on nouns which does the "individuating job" of forming countable units. Now Quine thinks that there is no fact of the matter as to whether we say that the Japanese word for 'ox' refers to the class of individual bovines, and the parametrized numerals are just an extra, albeit obligatory, linguistic device to specify them numerically - this is version (i); or else that that word refers to, as he puts it, "the unindividuated totality of beef on the hoof'. We could only say that the English NP five oxen and its Japanese translation correspond to each other as wholes; on the level of single constituents reference may fail to be invariant. Could there possibly be linguistic evidence to decide Quine's philosophical claim one way or the other? The answer is obviously negative if we talk about the inscrutability thesis in its whole breadth; seen in this context the Japanese case is only an example. But the universal linguist might indeed try to give arguments for or against a particular way of parsing the classifier construction and then see if Quine is justified in deriving support from it for his philosophical point. Now we could try some familiar tests for constituency in order to settle the issue of the proper parse for NL+CL+N. In the case of German, for instance, if we are prepared to argue from analogy and take the German NM for CL, the tests all seem rather inconclusive, as Krifka (1989) notes. Both parses (i) and (ii) are possible as the examples (1) and (2) show; in (2), the a) sentence corresponding to the parse [NL+NM]+N is better than the b) sentence, but they are both grammatical. (1)

a.

[[dreiNL BarrenNM] und [zweif^ Sackchenj^J] GoldN [[threeNL ingotsNM] and [two^ bags,™] (of) goldN

136

GODEHARD LINK

b. (2)

a.

drei^

[[BarreiiNM

thrceNL

[[ingotSNM (of) goldN]

Gold N ] und [Säckchen^ SilberN]] and [ b a g s ^ (of) silverN]]

An AtommüllN wurden [250^l F ä s s e r ^ ] gezählt. of nuclear wasteN were [ 2 5 0 ^ barrelsj^] counted

b.

An [Fässern^

AtommüllN]

wurden 250nl gezählt,

of [barrelsj^j (of) nuclear wasteN] were

2 5 0 ^ counted

In Chinese, like Japanese one of the typical classifier languages, the regular order of constituents in the classifier construction is NL+CL+N. Although there are cases where the classifier appears together with the noun while the numeral is missing (see sentence (3)) the numeral and the classifier form a standard unit (see sentences under (4)). (3)

a.

qianbian

shi tiao CL

he N .

over there

is

riverN

qing

he

beicL

please you

drink

[CL]

kafeitf ba. coffee N [particle]

a.

Ni

you

haiziN

ma?

You,

wo you

you Ni

have mai

children [QPt] guo

shu N

Yes, le

I

b.

You Dui,

buy wo

[past] mai

book(s) [Perf] san le NL

[QPt] ben CL .

Yes,

I

buy

[past]

[CL]

b. (4)

nin

[CL]

three

have

[CL]

ma?

Now curiously, Japanese happens to be not as good a candidate as Chinese, for instance, to illustrate Quine's point. In this language the usual order of constituents in classifier constructions is not NL+CL+N, but N+NL+CL, so that Quine's example five oxen translates into (5)

o-ushi-

go- to

oxen N

5jml

CL

Thus the noun and the classifier are separated by the numeral. That makes the notion of the classifier operating on the noun not a very natural one. Moreover, it is apparently a universal fact that out of the six possible combinations of ordering the constituents N, CL, NL, only those four occur in the world's languages in which the classifier and the numeral are contiguous elements. The most common arrangements are the ones exemplified by Chinese (NL+CL+N) and Japanese (N+NL+CL); some Pacific languages realize

QUANTITY AND NUMBER

137

CL+NL+N and N+CL+NL, while NL+N+CL and CL+N+NL do not occur anywhere (Allan 1977: 288). The conclusion to be drawn from this evidence seems to be that at least in Japanese version (ii) above is not realized, viz. the classifier does not have the individuating function after all. This is somewhat at variance with the picture that a semanticist would like to draw: he would prefer to be able to identify an item in the lexicon that has precisely that individuating function. But there is an assumption operative in the conclusion, namely that the parse of the classifier phrase is in fact a reliable indication of the classifier's conceptual role. This assumption can hardly be proved, however, and thus need not actually be embraced; there may simply not be a close enough fit between syntax and semantics that the evidence from parsing could decide the conceptual issue. (But then, Quine would have to give up his Japanese example altogether, which would in fact not be a loss after all since the linguistic facts concerning Japanese are, as we saw, not the way Quine represented them in the first place.) A more cautious formulation of the above conclusion would be to say only that linguistic evidence makes it more likely that classifiers have a parametrizing than a individuating function. But it might also be that they combine those two functions, so that the reference of the pure noun is still "holistic" or "mass-like". Now there is a distinction to be made here within the notion of mass term; the fact that a noun does not have the feature +count can have at least three reasons: it refers either to (i) some kind of homogeneous stuff like water or mud; or to (ii) some collection of discrete objects, but in a wholesale manner (paradigm example: cattle); or finally to (iii) some abstract concept like solidarity. There is no reason to suppose that in classifier languages these conceptual distinctions cannot be drawn; the only thing that can be legitimately claimed is that apparently there is no reference to single countable entities by the noun itself in those languages. Together with (i) - (iii), let us call this the fourth mode of reference. 4.

The semanticist who wants to model those four modes of reference in a close-to-

language fashion has to provide a set of formal characteristics distinguishing four classes of objects, one for each mode. Now the traditional Tarskian style of formal semantics recognizes only one kind of entities in its domain of discourse, the only postulate being that this domain be non-empty. There are, however, systems of multi-sorted logic which do distinguish different sets of basic individuals (e.g. usual individuals and spatiotemporal locations); moreover, there are sometimes various cross-connections between the different basic domains in those systems which can serve as a paradigm for the modeling task under discussion here (e.g. the postulate that every concrete object occupies some location in time and space). However, apart from such

functional

138

GODEHARDLINK

connections linking two otherwise unrelated subdomains there is the need to establish certain algebraic connections structuring one and the same subdomain. Consider the second and the fourth mode of reference; one picks out collections, the other single objects. When we consider collections and single objects as belonging to two altogether different subdomains we obviously miss a fundamental intuitive relationship between the two: collections are made out of single objects. Thus, if we consider a certain group of catde it is the individual animals that make up this collection. I have proposed elsewhere (Link 1983) to introduce one single domain for collections and individual objects and endow this domain with the algebraic structure of a lattice to represent the basic part-of relationship between those kinds of objects. The individual objects are then called atoms in this domain and are but special collections of "size" one. A general collection is also called individual sum or i-sum, for short. These theoretical devices basically take care of modes (ii) and (iv). The first mode of reference concerns homogeneous stuff. Its manifestations are all the particular quantities of matter of the given kind; they, too, are structured by a fundamental part-whole relation, with the important difference that this ordering does not recognize smallest parts (Link 1983, Krifka 1989). Finally, there are the abstract concepts or properties for the fourth mode of reference. Logical semantics used to model these entities at first as sets of concrete individuals (extensional semantics), then as intension functions, i.e., functions from possible worlds into sets of individuals (intensional semantics). But it is in the spirit of the algebraic approach to go even more "intensional" and to introduce an additional subdomain of basic entities standing for concepts; these are structured with the familiar concept hierachy. Our current ontology, then, consists of the following four subdomains (the operation 'X' forms all possible i-sums over a given set): (6)

A set of atomic individuals

[referential mode (iv)]

E set of individual sums over A,E-

[referential mode (ii)]

Q set of homogeneous quantities, Q^A

[referential mode (i)]

C set of concepts

[referential mode (iii)]

Next, we introduce some definitions. Let a be a noun, llall its denotation, n a natural number, and c a classifier; furthermore, let the down arrow

stand for the operation

that forms from a given element x in E or Q the set of all elements "below" x (in the sense of the relevant ordering relation); finally, '©' is the summing operation on E and Q. Then we define:

QUANTITY AND NUMBER

(7)

Ac

:=

{* e AI x classifies as c-object}

En.c

•=

U e E I 3 yi,...,y» e Ae : x = y&...®yH

CLPnc(llall)

:=

J

llall -n£„ c

)

[a in referential mode (ii)]

Qc

:=

[xb Q\x classifies as c-object}

Z(GC )

:=

the set of i-sums of elements of Qc

Qn,c

•=

U e I ( Q c ) I 3 yu...,yn J

139

e Qc : x = yi®...9yH

}

CLP'nc{\\a\\)

:=

Z(Hall ') n Qn c

[a in referential mode (i)]

SG(llall)

:=

£xr(llall)

[a in referential mode (iii)]

Thus, when c is a classifier for slim things, Ac is the set of all atomic individuals that are classified as a slim object in a given language, and En c is the set of i-sums that are built up from n atoms classified as c-objects. Let a be a noun in a classifier language in referential mode (ii); then it is reasonable to assume that its denotation Hall is the maximal collection x in E such that x qualifies as a collection of kind a. This captures the intuition of "wholesale" reference that is connected with the bare nouns in classifier languages: the bare noun by itself does not single out any particular collection. So if a = o-ushi (ox) then Hall is the collection or i-sum made up of all oxen. Let CLP be the classifier phrase [n+c] with the numeral n standing for the natural number n; then the semantic effect of this CLP is as described by the operator CLPn c: it forms all the subcollections of llall and intersects this set with all collections that consist of exactly n atomic individuals classifying as c; thus for the Japanese CLP o-ushi- go-to (five oxen) we get the set of all i-sums consisting of five oxen (here we assume that the numeral is not quantifying but behaves semantically like an adjective; for a discussion of this issue see Link 1986). If a is a noun in referential mode (i), its denotation llall is the maximal quantity of matter of kind a ; again, the reference of a does not single out any particular quantity. Then the analoguous CLP operator CLP'n c forms the set of i-sums of elements below llall that are at the same time collections of n units of portions of matter classifying as c. Thus if a is wine and c is glass-of then three glasses of wine denotes the set of collections of quantities of wine that make up three glasses. Finally, if a has referential mode (iii) its denotation is a concept in C. This mode is perhaps not that much realized in classifier languages but fits languages better that have a singulative form for their otherwise transnumeral nouns. In Arabic, for instance, the transnumeral form for the concept fly is dabban; from this the singulative dabbane (a fly) can be formed, which is the countable unit (Cowell 1964: 369; cited after Biermann 1981: 234). So the singulative marker seems to play a genuinely individuating role here. If a = dabban its denotation in our model is an element of C, viz. the concept fly, then the semantic effect of the singulative

140

GODEHARD LINK

is very straightforward: it is the familiar extension function Ext forming the set of all objects that the given concept applies to. Now let us come back the the familiar count nouns in Western languages; their mode of reference is individuating right away, without the detour over unstructured totalities or concepts. It is interesting to note, however, that the bare plural is those languages tends to fill the role of the transnumeral in other languages. Thus, bare plurals can refer to collections and also to kinds, which are not to be identified with concrete particular individuals and are best treated in semantics as still another class of entities (for arguments see Carlson/Pelletier, to appear). Some examples are given in (8). The bare plural tigers refers to the natural kind TIGER, and Otto motors to the kind OTTO MOTOR. (8)

a.

Tigers are a subspecies of wild cats.

b.

Otto motors were invented by the German engineer Nikolaus August Otto.

Since it is obvious that no single tiger can be a (sub)species, and Otto motors cannot have been invented over and over again, we see that the number distinction singular vs. plural is neutralized in these contexts. Bare plurals have the form of a plural, but their reference is transnumeral. The above discussion shows that universal semantics has to apply richer models to account for the varieties of referential modes that can reasonably be assumed across languages. To repeat, we are not putting forward a philosophical thesis here as to what the "real" reference of a term is. Our model is based on the assumption that universal grammar gives reasonable clues for the choice of the proper ontology. But this is an assumption that is not uncontested in philosophical quarters favoring metaphysical regimentation in ontological matters. This is not a place to argue a case which could help to resolve that tension.

II. Structural Relativity in the Notion of Number 5.

In the first part of the paper we addressed the basic problem of individuation of

nouns in language and tried to model their various modes of reference in terms of a multisorted, algebraically structured ontology. We now extend our discussion from the nominal to the verbal domain, and from the phrasal to the sentential level. This step opens up the possibility to discuss matters of quantity and number relationally, i.e., to consider

QUANTITY AND NUMBER

141

possible scope relations between measure phrases. Moreover, a fifth mode of reference will come into play, that to events. To start with the latter, a very interesting linguistic phenomenon was recently discussed in the literature (Krifka 1989a), which in view of examples like (9a) I like to call the "transportation facts". (9)

a.

Lufthansa carried five million passengers last year.

b.

Four thousand ships passed through the lock last year.

c.

The library lent out 23,000 books in 1987.

The sentence (9a) is in fact not about five million passengers, nor does (9b) speak about four thousand ships, nor (9c) about 23,000 books. Rather, some collection of related events is counted in each case: in (9a), five million counts the number of tickets sold by Lufthansa last year; (9b) claims that four thousand passages by ships took place at the lock in question; and (9c) speaks of 23,000 events of lending out a book. In each of those situations, there might well have been much less individuals involved than the figures indicate, and that is indeed the typical case. Now counting presupposes individuation, as we said above. Thus linguistic evidence leads us the conclusion that the semantics has to provide for still another kind of discrete, countable entities - this time not individuals of some sort or other, but whole events. There are more classes of examples that show the need for such a move. One case in point in Georgian reduplication-, the following data are taken from an intriguing paper by David Gil (Gil 1988). (10) a.

b.

Orma

k'acma

sam-sami

canta

c'aiyo.

two-erg man-erg \hr&&-dist-abs suitcase-afos 'Two men carried three suitcases [each / each time].' Or-orma k'acma sami canta

c'aiyo.

two-dist-erg man-erg

camed-iig

three-abs

suitcase-aZw

carried-Jig

'Three suitcases were carried by two men [each / each time].' In Georgian, numerals can be reduplicated, which leads semantically to the distribution of the NP containing the reduplication over another NP in the sentence. That means for (10a) that each of the two men carried three suitcases (for more on the phenomenon of distributivity see below). But this is only one reading; reduplication can also have the semantic effect of introducing some reference to events. That is, (10a) can also mean that the same group of men carried the three suitcases several times or on several occasions.

142

GODEHARD LINK

Note that this effect seems to be symmetric in Georgian: if the numeral of the ergative NP is reduplicated as in (10b), the interpretation of the sentence is exactly the opposite. Now this NP (or-orma k'acma) is distributed over the suitcase NP (rendered in English by passivization), with the same ambiguity between the internal distributive reading and the event reading. Call the NP that is distributed over the domain of distribution. Then we can formulate a generalization subsuming the two readings: if the reduplicated NP 2 can find another NPi in the sentence denoting a pluralic entity, then NPj serves as domain of distribution for NP 2 (call it the distributional antecedent for NP2). In this case NPj is an explicit internal antecedent for NP2. But NP 2 can also refer back to an implicit, external antecedent, a pluralic event-like entity, e.g. some contextually given set of occasions at which the action expressed in the sentence takes place. So the reduplicated NP 2 looks for a suitable antecedent as its domain of distribution, which may be either internal (usual distributive interpretation over plural individuals) or else external (event interpretation). The same ambiguity can be observed in Korean, where distribution is asymmetric, however. Consider the sentences under (11) - (13); the data are due to Choe (1987a). (11) a.

b.

[ai-tul]-i

[phwungsen-hana]-rul

sa-ess-ta.

ch\\d-pl-nom

balloon-one-acc

bought

'The children bought a balloon.'

(collective)

[ku-tul]-i

[phwungsen-hana]-rul

sa-ess-ta.

h c-pl-nom

balloon-one-acc

bought

'They bought a balloon.' (12) a.

[ai-tul]-i

[phwungsen-hana-ssik]-ul sa-ess-ta.

c\\i\A-pl-nom balloon-one-AQ-acc 'The children bought a balloon each.' b.

(13) a.

(collective) bought (distributive)

[ku-tul]-i

[phwungsen-hana-ssik]-ul sa-ess-ta.

child-p/-nom

balloon-one-AQ-acc

bought

'They bought a balloon each.'

(distributive)

na-nun

[phwungsen-hana]-rul

sa-ess-ta.

I-top

balloon-one-acc

bought

'I bought a balloon.' b.

na-nun

[phwungsen-hana-ssik]-ul sa-ess-ta.

I-top

balloon-on e-AQ-acc

bought

'I bought a balloon [at several occasions].' Sentences (1 la,b) have a collective reading: the content is that there is one balloon which was bought by the children (by them). Now if the particle -ssik is added to the object

QUANTITY AND NUMBER

143

phrase the sentences get a distributive reading according to which each of the children (each of them) bought a balloon. As in the Georgian example, the NP carrying the distribution marker -ssik (called anti-quantifier by Choe) looks for another plural NP that it can take as domain of distribution. Now interestingly, if no such NP is available, as in (13b), the sentence can still be interpreted: some contextually understood plurality of events steps in as external antecedent, so that the sentence means that I bought a balloon each day or at each store. But note that the distributive particle does not work in the same symmetric way as Georgian reduplication: in the above sentences the subject NP could not carry the particle -ssik with the distributional effect reversed. It is a curious fact that a particle very much like the Korean -ssik exists in German (see Link 1987b). Like -ssik, the German particle je not only triggers distribution in the same way as in Korean, but can also induce an event interpretation if no NP is present in the sentence that qualifies as an internal distributional antecedent. Thus, while sentence (14a) is ambiguous between the collective and the distributive reading, ye in (14b) forces the distributive reading much like the English each; but unlike each, je is much more flexible: thus in (15a) it is not necessary that the subject NP be plural for the sentence to make sense; (15b), which is (15a) except for the singular subject NP, still has the event reading, which is already present in (15a) and makes that sentence ambiguous between the internal individual and the external event interpretation. (14) a.

Die Touristen kauften eine Eintrittskarte.

(collective/distributive)

The tourists bought an admission ticket. b.

Die Touristen kauften je eine Eintrittskarte, (distributive) The tourists bought an admission ticket each.

(15) a.

Die Zollbeamten prüften je einen Koffer,

(distributive)

The customs officers checked a suitcase (each/each time). b.

Der Zollbeamte prüfte je einen Koffer.

(event reading)

The customs officer checked a suitcase (each time). c.

Die Zollbeamten (/der Zollbeamte) priifte(n) jeweils einen Koffer. The customs officer(s) checked one suitcase each time.

The event reading in (15a) becomes stronger when we replace the particle je by the adverbial jeweils; see (15c). Accordingly, jeweils interchanges with je in sentences where an internal distributional antecedent is altogether missing, like in (16a). As for the question of symmetry, it seems to be the case that in certain contexts je b e h a v e s kataphorically; thus, in (16b) the NP den Körben serves as antecedent, and the sentence means that we have one situation with several baskets in each of which there is an apple.

144

GODEHARD LINK

But then, world knowledge excludes the external event reading according to which at each occasion, one apple was "collectively" in the baskets. (16c) remedies this flaw, and the sentence is again ambiguous between the external and the internal reading (preference is for the internal reading on pragmatic grounds). (16) a.

Je ein Apfel war faul.

(event reading)

An apple each was rotten. b.

Je ein Apfel lag in den Körben.

(internal distribution)

An apple each was in the baskets. c.

Je drei Äpfel lagen in den Körben.

(internal > external)

Three apples each were in the baskets. In (16b,c) we have a definite plural NP as domain of distribution. Internal kataphoric distribution becomes worse if not outright impossible, however, when the NP is indefinite plural with a numeral in it. Thus I think that (17a) has only the event interpretation meaning that for each "case" of two diplomats in need of an interpreter one such person was actually dispatched. In (17b) the internal reading is there again; but here the regular order of constituents is reversed, the subject NP and the object NP are switched. (17) a.

Je ein Dolmetscher begleitete zwei Diplomaten,

(external)

An interpreter each accompanied two diplomats, b.

Je einen Dolmetscher erhielten zwei Diplomaten, (external, internal) [An interpreter each] OBJ got [two diplomats]

SUBJ

For more discussion on je the reader is referred to Link (1987b). Here I would like focus now on the question of autonomous vs. dependent reference of numerative phrases in relational plural sentences. Consider (18) and (19). (18) a. b. (19) a. b.

Two interpreters accompanied four diplomats. Four diplomats were accompanied by two interpreters. Three ensembles played two symphonies five times. In five performances, three ensembles played two symphonies.

Sentence (18a) speaks about two interpreters whose job it was to deal with a certain number of diplomats. Without further information the sentence does not say precisely how many diplomats were involved; if the intended reading is collective then the

QUANTITY AND NUMBER

145

interpreters accompanied a group of four diplomats, but otherwise, in its distributive reading there may have been up to 2x4 = 8 diplomats around. The NP four diplomats has dependent reference: its numeral four does not give the true amount of the individuals involved. This number can be anything between 4 and 8. Thus in general, let NP 2 be the dependent NP in a relational plural sentence and NPj a noun phrase taking scope over it; let k be the numeral in NPj, n the numeral in NP 2 , and k, n the respective natural numbers expressed by those numerals. Then the exact number i of individuals in the denotation of NP 2 cannot be completely determined since it depends on pragmatic circumstances; but on semantic grounds alone the interval n CONDITIONAL > COMPARATIVE,

with NPIs most likely occurring towards the left side. Furthermore, I suspect that the list of NPI contexts, though quite impressive, is still not complete; we will have to watch out for other environments where NPIs can pop up. But even so, it is difficult to see a common principle which explains why the known NPI contexts license NPIs. Before outlining the current theories, I will discuss a particularly interesting phenomenon which can be treated similarly to NPIs, although its relation to NPIs has escaped notice until now, as far as I can tell.

1.3. Partitives as Negative Polarity Items There are languages which show a case alternation in some grammatical contexts, especially in the scope of negation. For example, in Russian and some other Slavic and Baltic languages there is an alternation between accusative / nominative and genitive (see e.g. Brooks 1967, Lisauskas 1976, Gundel 1977), and in Finnish there is an alternation between accusative / nominative and partitive (see e.g. Dahl and Karlson 1975, Raible 1976, Heinamaki 1984). We find a similar phenomenon in French with the use of de + common noun under negation. Here, I will concentrate on Finnish. There are several triggers for the switch to the partitive case in Finnish. The most prominent ones are the scope of negation and imperfective aspect. An attempt to explain the use of the partitive as an marker of imperfective aspect can be found in Krifka (1989, 1989a). Here I will concentrate on the non-aspectual use of the partitive. Summarizing the observations of Heinamaki (1984), we can note the following contexts for the non-aspectual use of the partitive. I start with two examples in which the partitive is used in the scope of negation. (21) a.

Mina nain Annelin. I saw Anneli.ACC 'I saw Anneli'

b.

Mina en nahnyt Annelia. INEG saw Anneli.PART 'I didn't see Anneli1

SOME REMARKS ON POLARITY ITEMS

(22) a.

157

Raili hiihti paivan. R. skied day.SG.ACC 'Raili skied for a day'

b.

Raili ei hiihtanyt paivaa. R. NEG skies day.SG.PART 'Raili didn't ski for a day.'

An exception to this use of the partitive is that the accusative occurs in negated questions with a suggested affirmative answer: (23)

Eikòhan juoda kuppi kahvia? NEG-Q drink cup.ACC coffee 'How about drinking a cup of coffee?'

With certain modal operators, like tuskin 'hardly', turha 'needless', vaikea 'difficult', mahdotonta

'impossible', tarpeetonta 'needless', epàviisasta

'unwise', and certain

quantifiers like harva(t) 'few', an accusative can, but need not, change to a partitive: (24) a.

Pirkko tunnisti minut / *minua. P. recognized I.ACC / I.PART 'Pirkko recognized me'

b.

Tuskin Pirkko minut / minua tunnisti. hardly P. I.ACC / I.PART recognized 'It is unlikely that Pirkko recognized me'

Furthermore, the partitive can occur in Yes/No-questions. In this case, the speaker seems to assume that the answer is negative. Also, questions with partitives are considered to be more polite. (25) a.

Mina nain Anjan / *Anjaa katsomossa. I saw A.ACC / A.PART audience-in 'I saw Anja in the audience.'

b.

Naitkò Anjan / Anjaa katsomossa? saw.you.Q A.ACC / A.PART audience-in Did you see Anja in the audience?'

158

MANFRED KRIFKA

Interestingly, a question with the NPI koskaan 'ever' forces an indefinite accusative to switch to a partitive: (26)

Oletko koskaan rekentanut saunaa / *saunan? have.you.Q ever built sauna.PART / sauna.ACC 'Have you ever built a sauna?'

These examples make it at least plausible to explain some of the distribution of partitives by the assumption that the partitive is an NPI. Consequently, we can use the hypothesis that partitives can be NPIs as a heuristic tool to detect other contexts which favor partitives, for example the protasis of conditionals. This is an example which shows that the notion of polarity items might prove useful in the study of languages.

2. Theoretical Approaches to Negative Polarity Items In this chapter, I will discuss the two main lines of explanations of NPI contexts, the syntactic tradition and the semantic tradition. 2.1. Syntactic Explanations The basic assumption of the syntactic approach is that NPIs are triggered by a negation element which stands in a certain syntactic configuration to the NPI. This theory was brought forward in Klima's treatment of negation in English (Klima 1964). He assumed that an NPI has to stand 'in construction with' (roughly: be c-commanded by) an 'affective element', that is, a negation. See also Jackendoff (1968) for this type of treatment. The problem is that not every NPI occurs in the scope of a negation. The remedy proposed by Baker (1970) was to distinguish between two types of NPI contexts: Either an NPI is licensed directly by a negation, or the proposition which immediately contains the NPI ENTAILS a proposition which then directly licenses the NPI. For example, Baker explains why be surprised creates an NPI context by saying that a is suprised that p entails the sentence a does not expect that p, which contains a negation and, therefore, directly licenses NPIs. (Arguments like this one have been influential in the development in Generative Semantics, as they suggested that there is no strict borderline between syntactic rules and semantic rules, like entailment.)

SOME REMARKS ON POLARITY ITEMS

159

The latest proponent of the syntactic-based explanation is Linebarger (1980, 1987).2 Similarly to Baker, she distinguishes between direct licensing and derivative licensing. However, she refines both types of licensing mechanisms: - First, the direct licensing does not apply to some deep or surface structure representation, but to the logical form (LF) of the sentence, where LF is the representation level in Government and Binding theory. NPIs are said to be licensed in a sentence S if, in the Logical Form of S, the NPI is in the immediate scope of a negation operator ('immediate' meaning that no other prepositional operator intervenes between the negation operator and the NPI). Linebarger calls this the 'Immediate Scope Constraint'. - Second, Linebarger has to invoke another principle to explain the remaining NPI contexts. She assumes that an NPI in a sentence S contributes the following conventional implicature: (i) We can infer from S a proposition NI (the "Negative Implicatum"). In the Logical Form of some sentence expressing NI, the NPI is directly licensed by a negation element (ii) The truth of NI guarantees the truth of S. Without going into the interesting arguments for the syntactic approach here, I want to point out some serious problems with it. My general impression is that the basic rule of direct licensing is very strict, but is in need of additional rules which are completely unsyntactic to cover the many remaining cases. So one could suspect that the theory is immunized by the additional rules. This suspicion is confirmed, as the additional rules often are quite problematic. A general problem from a logician's point of view is that from every proposition p follows -i-ip, so NPIs should occur everywhere. Linebarger has to stipulate that spurious entailments like this one cannot induce derivative licensing — which is surely not a convincing treatment. Another problem is that Linebarger wants to explain why NPIs occur in the protasis of conditionals by analyzing a sentence i f p then q as a material implication p —*q and the fact that this is equivalent to —>p v q. However, it is highly questionable whether natural language conditionals can be represented by material implication (see e.g. Kratzer 1987). Finally, Linebarger tries to explain why questions are NPI contexts by saying that they only occur in questions in which the speaker clearly presupposes that the answer is negative, especially rhetorical questions. But we have seen that NPIs can occur in normal questions as well.

2.2. Semantically-Based Explanations The general view in the semantic tradition is that negation is only one trigger for NPIs among others, and that the class of triggers has to be characterized semantically.

160

MANFRED KRIFKA

The semantic tradition can be traced back to the work of Horn (1972) on semantic scales and Fauconnier (1975a, 1975b, 1978) on scale reversals. Fauconnier identified the phenomenon of semantic scales with so-called 'quantificational superlatives'. For example, sentence (27.a) normally is used to express (27.b): (27) a. b.

Mary can solve the most difficult problem, Mary can solve every problem.

The reason for that is that speakers assume a rule: If Mary can solve problem x, and y is a problem which ranks lower on the scale of difficulty than x, then Mary can solve problem y as well. It follows that if Mary can solve the problem which ranks highest in difficulty, she can solve all other problems as well. Now, negation is a scale reverser. For example, we can derive from the rule cited: If Mary CANNOT solve problem x, and y is a problem which ranks HIGHER on the scale of difficulty, then Mary cannot solve problem y either. Conversely, if Mary cannot solve the problem which ranks LOWEST in difficulty, she CANNOT solve any other problem. Therefore sentence (28.a) can be used to express (28.b): (28) a. b.

Mary cannot solve the easiest problem, Mary cannot solve any problem.

NPIs, then, can be analyzed as 'quantificational superlatives', denoting the lower end of scales (cf. also Schmerling 1971, who claimed that NPIs denote 'smallest units'). To get this interpretation, then, they need an operator, such as negation, which reverses the scale. This type of explanation was developed and applied to a range of phenomena by Ladusaw (1979, 1983); see also Zwarts (1981) and Hoeksema (1983, 1986). According to Ladusaw, NPIs only occur in DOWNWARD-ENTAILING (DE) contexts. A context is DE if an expression occuring in it can be replaced by a semantically stronger (that is, more restricted) expression salva veritate (without change of truth of the whole sentence). Correspondingly, a context is called UPWARD-ENTAILING (UE) if an expression in it can be replaced by a semantically weaker expression salva veritate. The following sentences exemplify UE and DE contexts; we assume that Italian ice cream is semantically stronger than ice cream: (29) a.

X in Mary ate X is UE, as it follows from Mary ate Italian ice cream that Mary ate ice cream.

SOME REMARKS ON POLARITY ITEMS

b.

161

X in Mary didn't eat X is DE, as it follows from Mary didn't eat ice cream that Mary didn't eat Italian ice cream

Obviously, DEness is a semantic notion. Ladusaw defines it as a property of operators a. His definition runs as follows: (30)

An expression a is downward-entailing (or polarity reversing) iff VX,Y[X < Y —» q then -iq => -ip as a general rule (contraposition). QUANTIFIED NPS which allow for NPIs in their scope, like no persons, few persons or less than three persons, create DE contexts. For example, less than three persons ate ice cream makes a stronger claim than less than three persons ate Italian ice cream, although ate ice cream is weaker than ate Italian ice cream. Quantified NPs of this type are called MONOTONE DECREASING (Barwise and Cooper 1981). When they are analyzed as Generalized Quantifiers, it can be formally derived that they are DE. For example, if we analyze less than three persons as a second order predicate Q, Q = A.X[#(X n person*) Maiy has visited more Asian countries than four of her colleagues,

b.

Mary has visited more Asian countries than John could ever dream of.

Again, the notion of limited DEness should give us the right analysis in these cases. Let us conclude this exposition of the semantic approach. I think that the semanticallybased explanation of NPI contexts is impressive and even a paradigm case for the explanation of linguistic facts by the means of formal semantics. However, there still remain some problems. A basic problem is that the NPI contexts are characterized by the general semantic principle of DEness, but it is not clear WHY DE operators allow for NPIs. So Ladusaw's generalization must be backed up by an explanation as to how DEness has the property of licensing NPIs. Another problem is that the notion of DEness, as Ladusaw presents it, is restricted to set inclusion for non-propositional items. But set inclusion is not always a good model for Fauconnier's hierarchies. As an example, consider the sentence John doesn't have x, which should be DE in x. It surely is; from John doesn't have ice cream it follows: John doesn't have Italian ice cream. However, the hierarchy which seems to matter in cases like John doesn't have a red cent are amounts of money. So, for example, from John doesn't have five cents it follows that John doesn't have ten cents. But the extensions of the predicates five cents and ten cents are not related to each other by set inclusion; they are simply disjoint: No object to which five cents applies is such that ten cents can be applied, and vice versa. So the i s a P P LATTICE iff (a), (b), (c) a s in ( 3 5 ) , a n d

(d) A' is the unique Y such that for every XGLA, X We have for all properties X, if X e L a .drop.of.wine then X is the property of being a quantity of wine of a certain size. Furthermore, we can assume that if x is a quantity of wine, then at least one of the properties in La.drop.of.wine applies to x. In a formula, with i as a variable over possible worlds and wine' as the property of being wine: Vi,x[wine'(i)(x) 3X[Xe L a . d r o p .of.wine A X(i)(x)]]. Therefore La.drop.of.wine can be called exhaustive with respect to wine'. As for the ordering relation, we assume that X —a.drop.of.wine Y iff it necessarily holds that for any x, y such that x has the property X and y has the property Y, x is smaller than or equal to a quantity of wine y. That is, VX,Y[X Y Vi,x,y[[X(i)(x) A Y(i)(y)] -» [wine'(i)(x) A wine'(i)(y) A x is a smaller or equal quantity than y]]]. Finally, a.drop.of.wine' applies to quantities of wine which are smaller than a certain (small) limit; roughly: Vi,x[a.drop.of.wine'(i)(x) —» wine'(i)(x) A x is smaller than some quantity e]. Obviously, then, a.drop.of.wine' is the least element of the lattice. < a . d r o

P

. o f . w m e

We do not specify how many elements the lattice La.drop.of.wine should have, except that there must be more than one; however, the idea is that there be sufficiently many.

168

MANFRED KRIFKA

Perhaps the most natural assumption here is that La.drop.of.wine is a partition of quantities of wine into equivalence classes, and that la.drop.of.wine is antisymmetric and connected. Also, we do not specify how small the limit e has to be which defines the least element in the lattice. As the idiomatic expression a drop of wine cannot be used in the positive sense to specify some amount of wine, we should assume that e can be deliberately small. This idea could be worked out in some form of game-theoretical semantics (cf. Saarinen 1979, Hintikka 1983), where the speaker (proponent) gives the hearer (opponent or Nature) the opportunity to choose as small a value as she likes. This would involve a more dynamic view of lattice sorts. In order to avoid additional complications, we will stick here to the static view and assume that negative polarity lattices come with a fixed (small) NPI representation. Our second example is quite similar; it is the NPI a red cent. Its NP lattice is - For every Xe

La.red.cent,

it holds that X is a

property of amounts of money; and if we claim exhaustiveness, for every amount of money x and world i there should be a property X in La.red.cent such that X(i)(x). We furthermore have X i s a P P INCLUSION LATTICE i f f ( a ) , ( b ) a s i n ( 3 7 ) , a n d ( c ) f o r

every X e L A , X < A' (that is, A' c X), The ordering of inclusion lattices can be associated with a very general relation, namely set inclusion. In contrast, the ordering relations of non-inclusion lattices seem to be rather idiosyncratic. They are related to orderings such as quantities of matter, monetary value, or amounts of labor. However, we can associate most, if not all, of these orderings with a general ordering relation as well. There are reasons to assume that the domain of individuals is structured by a PART RELATION (see e.g. Link 1983, Krifka 1989,1989a). For example, we can assume that quantities of wine are subject to a part relation, and that this part relation is associated with the relations on which S a .drop.of.wine is based. Proof: If x and y are quantities of wine, and x is a proper part of y, then x is an amount of wine which is equal to or smaller than y. Hence, if X, Ye L a .d r0 p.of.wine> xe X(i), and ye Y(i), then it may be the case that X < a .drop.of.wine Y, but it cannot be the case that Y '[t]](i) and a lattice sort which consists of propositions p for which there is some time t such that p = 0 [ t ] . Thus, ever and anytime 0 have the same semantic representation but different NP lattices. Let us now turn to PARTITIVES. Partitives in NPI contexts, which we observed in Finnish (cf. 1.3), should also be treated as cases of non-idiomatic NPIs. We can analyze the semantic effect of partitive marking as follows: If a nominal predicate a has the semantic representation a ' (a property, type set), then the partitive form of a, has as its semantic representation Xi,x3y[a'(i)(y)

a-PART,

a x Spy], where

S[Y] < S[X],

c.

NEUTRAL WITH RESPECT TO A ' if neither (a) nor (b).

Note that we do not define DEness and UEness generally, as in Ladusaw (1979), but always with respect to a polarity item representation A'. With the definition of UE / DE semantic compositions, we can develop the notion of a polarity lattice which is DERIVED FROM ANOTHER POLARITY LATTICE by semantic composition. As we take into consideration only those derived lattices which are inclusion lattices, it is sufficient to know the lattice sort and the polarity item

176

MANFRED KRIFKA

representation of the derived lattice; the ordering relation is always < restricted to elements of the lattice sort. (44) Let LA = be a polarity lattice and S[A'] a semantic composition which is UE or DE with respect to A'. Then the inclusion lattice generated by , where a. b.

B' = S[A']; Xe L b iff there is a Y, Ye L A , and X = S[Y].

The polarity of the derived lattice depends on the polarity of LA and whether S[A'] is UE or DE with respect to A'. Proof: Let LA = . As L A is an NP lattice, we have for any Xe L A with -iX ) or C(0,A') which is DE or UE with respect to A', and if LA is the inclusion lattice generated by LA and S[A'], then we call [x]}.

186

MANFRED KRIFKA

4. Heim's rule is formulated as follows: Suppose you have a conditional "if X then Y", where X contains the NPI-occurrence A. Let X [ A / g ] be just like X, except with A replaced by B. Let c be the set of presupposed background assumptions. Then A is licensed in "if X then Y" if for any B of the appropriate type: c A [X[ A /B] ->X] A [if X then Y] -> if X[ A /B] then Y.

5. That is, L a is a set of entities of the type of A'. 6. Note that is not necessarily a lattice in the usual sense, as we do not claim that - NOW we can assume a semantic composition C* applying to complex semantic representations. C* can be defined on the basis of the semantic composition C as follows: C*(«x', ALT(a')>, . The rule for the generation of derived polarity lattices is a consequence of this general rule, which allows for the composition of two polarity items in a natural way. The two polarity lattices must have the same polarity in order that the composition is a polarity lattice as well.

References Baker, Carl Lee: 1970, 'Double Negatives'. Linguistic

Inquiry 1, 169-189.

Barwise, Jon, and Robin Cooper: 1981, 'Generalized Quantifiers and Natural Language', Linguistics

and Philosophy

4, 159-219.

Brooks, Maria Zagorska: 1967, 'The Accusative-Genitive Contrast in s o m e Polish Constructions', in To Honour Seventieth

Birthday,

Roman Jakobson.

Essays

on the Occasion

Vol. I, Mouton, The Hague - Paris, 395-401.

Carlson, Greg: 1981, 'Distribution of Free-Choice any', CLS 17, 8-23.

of his

SOME REMARKS ON POLARITY ITEMS

187

Comrie, Bernard, and Norval Smith: 1977, 'Lingua Descriptive Series: Questionnaire', Lingua 42, 1-72. Cresswell, Max, and Arnim von Stechow: 1982, 'De re Belief Generalized', Linguistics and Philosophy 5, 503-535. Dahl, Osten and Fred Karlson: 1975, 'Verbal Aspects and Object Marking: A Comparison between Finnish and Russian', Logical

Grammar Reports 17,

Gothenburg. Edmondson, J.A.: 1981, 'Affectivity and Gradient Scope', CLS 17, 38-44. Fauconnier, Gilles: 1975a, 'Polarity and the Scale Principle', CLS 11, 188-199. Fauconnier, Gilles: 1975b, 'Pragmatic Scales and Logical Structure', Linguistic Inquiry 4, 353-375. Fauconnier, Gilles: 1978, 'Implication Reversal in a Natural Language', in F. Guenthner and S J . Schmidt (eds.), Formal Semantics and Pragmatics for Natural Languages, Reidel, Dordrecht, 289-301. Gallin, Daniel: 1975, Intensional and Higher-Order Modal Logic, North-Holland, Amsterdam. Gundel, Jeannette K.: 1977, 'Towards a more Adequate Description of Genitive Marking in Russian: Part I', Folia Slavica 1, 188-211. Heim, Irene: 1982, The Semantics of Definite and Indefinite Noun Phrases. Ph.D. Dissertation, University of Massachusetts at Amherst, and University of Konstanz, SFB-Papier 73. Heim, Irene: 1987, 'A Note on Negative Polarity and Downward Entailingness', NELS 14, 98-107. Heinamaki, Orvokki: 1984, 'Aspect in Finnish', in Casper de Groot and Hannu Tommola (eds.), Aspect Bound. A voyage into the realm of Germanic, Slavonic and Finno-Ugrian aspectology, Dordrecht, Foris, 153-178. Hintikka, Jaako: 1983, The Game of Language: Studies in Game-Theoretical Semantics and its Applications, in collab. with Jack Kulas, Dordrecht, Reidel. Hoeksema, Jack: 1983, 'Negative Polarity and the Comparative', Natural Language and Linguistic Theory 1, 403-434. Hoeksema, Jack: 1986, 'Monotonicity Phenomena in Natural Language', Linguistic Analysis 16, 25-40. Hoppenbrouwers, G.: 1983, 'Polariteit. Een literaturonderzoek.' NIS Working Papers 2, Nijmegen. Horn, Larry: 1972: On the semantic properties of logical operators in English. Ph.D. Dissertation, University of California at Los Angeles.

188

MANFRED KRIFKA

Jackendoff, Ray: 1968, 'An Interpretive Theory of Negation', Foundations of Language 5, 218-241. Jacobs, Joachim: 1983, Fokus und Skalen, Zur Syntax und Semantik von Gradpartikeln im Deutschen, Niemeyer, Tübingen. Jacobs, Joachim: 1984, 'Funktionale Satzperspektive und Illokutionssemantik', Linguistische Berichte 91, 25-58. Jacobs, Joachim: 1985, 'Negation'. To be published in A. von Stechow and D. Wunderlich (eds.), Handbuch Semantik, Athenäum, Kronberg. Klima, E.: 1964, 'Negation in English', in J. Fodor and J. Katz (eds.), The Structure of Language, Prentice Hall, Englewood Cliffs. Kratzer, Angelika: 1987, 'Conditionals', CLS 23, 1-15. Krifka, Manfred: 1987, 'An Outline of Genericity'. Partly in collaboration with Claudia Gerstner. Forschungsbericht des Seminars für natürlich-sprachliche Systeme der Universität Tübingen 25. Krifka, Manfred: 1989, Nominalreferenz

und Zeitkonstitution.

Zur Semantik von

Massentermen, Pluraltermen und Aspektklassen, Wilhelm Fink, Munich. Krifka, Manfred: 1989a, 'Nominal Reference, Temporal Constitution and Quantification in Event Semantics', R. Bartsch, J. van Benthem, and P. van Emde Boas (eds.), Semantics and Contextual Expressions, Dordrecht, Foris. Krifka, Manfred: 1990 'Polarity Phenomena and Alternative Semantics', in Martin Stokhof and Leen Torenvliet (eds.), Proceedings of the 7th Amsterdam Colloquium, Amsterdam, ITLI, 277-302. Kürschner, Winfried: 1983: Studien zur Negation im Deutschen, Tübingen, Narr. Ladusaw, William: 1979: Polarity Sensitivity as Inherent Scope Relations, Ph.D. dissertation, University of Texas at Austin. Ladusaw, William: 1983, 'Logical Form and Conditions on Grammaticality', Linguistics and Philosophy 6, 373-392. Lewis, David: 1973, Counterfactuals, Harvard University Press, Cambridge, Mass. Linebarger, Marcia C.: 1980: The Grammar of Negative Polarity, Ph.D. Dissertation, MIT, Cambridge, MA. Linebarger, Marcia C.: 1987, 'Negative Polarity and Grammatical Representation', Linguistics and Philosophy 10, 325-387. Link, Godehard: 1983, 'The Logical Analysis of Plurals and Mass Terms: A LatticeTheoretical Approach', in R. Bäuerle e.a. (ed.), Meaning, Use and the Interpretation of Language, de Gruyter, Berlin, New York, 303-323. Lisauskas, Sarunas: 1976, 'Objects of Negated Verbs in Lithuanian', CLS 12, 459-467.

SOME REMARKS ON POLARITY ITEMS

189

McGloin, Naomi Hanaoka: 1976, 'Negation', in M. Shibatani (ed.), Syntax and Semantics Vol. 5: Japanese Generative Grammar, Academic Press, New York, 371419. Montague, Richard: 1970, 'Universal Grammar', in Theoria 36, 373-98. Also in R. Thomason (ed.): 1974, Formal Philosophy: Selected Papers of Richard Montague, 222-246, Yale University Press, New Haven. Partee, Barbara: 1984, 'Nominal and Temporal Anaphora', Linguistics and Philosophy 7, 243-286. Partee, Barbara, and Mats Rooth: 1983, 'Generalized Conjuction and Type Ambiguity', in R. Bäuerle, Chr. Schwarze and A. v. Stechow (eds.), Meaning, Use and the Interpretation of Language, de Gruyter, Berlin, 361-383. Progovac, Ljiljana: 1988, A Binding Approach to Polarity Sensitivity, Ph.D. Dissertation, University of Southern California, Los Angeles. Raible, Wolfgang: 1976, Zum Objekt im Finnischen. Eine

sprachwissenschaftliche

Fallstudie, Christoph von der Ropp, Hamburg. Rooth, Mats: 1985, Association

with Focus. Ph.D. Dissertation, University of

Massachusetts, Amherst. Saarinen, Esa (ed.): 1979, Game-Theoretical Semantics, Reidel, Dordrecht. Schmerling, Susan F.: 1971, 'A Note on Negative Polarity', Papers in Linguistics 4, Linguistic Research Institute, Champaign, 200-206. Stechow, Arnim von: 1988, 'Focusing and backgrounding operators', Arbeitspapier 6, Fachgruppe Sprachwissenschaft, Universität Konstanz. Vendler, Zeno: 1962, 'Each and Every, Any and All', Mind LXXI, 145-160, also in Z. Vendler: 1967: Linguistics in Philosophy, Cornell University Press, Ithaca, 70-96. Welte, Werner: 1975, Negations-Linguistik, Wilhelm Fink, München. Zaefferer, Dietmar: 1984, Fragen und Frageausdrücke im Deutschen. Zu ihrer Syntax, Semantik und Pragmatik, Wilhelm Fink, München. Zimmermann, Thomas Ede: 1989, 'Intensional Logic and two-sorted type theory', The Journal of Symbolic Logic 54, 65-77. Zwarts, Frans: 1981, 'Negatief polaire uitdrukkingen I', Glot 4, 35-132.

Concessive Relations as the Dual of Causal Relations* Ekkehard König Freie Universität Berlin Institut für Englische Philologie Gosslerstr. 2-4 D-l000 Berlin 33 0. Introduction Together with terms like "temporal", "local", "manner", "conditional", "instrumental", "purposive" and "causal", the term "concessive" belongs to the terminological inventory that traditional grammar makes available for the characterization and classification of adverbial relations.1 In contrast to the picture presented by many grammar handbooks and traditional descriptions, this list of adverbial categories should not be regarded as an unstructured collection of well-established and convenient distinctions. Among these adverbial relations two groups can be distinguished and it has even been suggested that these relations can be ranked on a semantic scale of informativeness (cf. Foley and Van Valin, 1984:293). Furthermore, adverbial relations may be connected to each other through a network of semantic relations such as converseness, hyponymy and duality. The distinction between two groups of adverbial relations is based on the following criteria (cf. Thompson and Longacre, 1985: 177ff.): (i) "elementary" or "primary" adverbial relations (place, time, manner) can typically be expressed by monomorphemic, non-anaphoric adverbs and be questioned by simple interrogative pronouns. The corresponding adverbial clauses can generally be paraphrased by relative clauses: (1)

(2)

a.

He came yesterday.

-When?

b.

He sat there.

-Where?

c.

He ran fast.

- How?

a.

He arrived when (= at the time at which) Peter left.

b.

He lost the money where (= at the place at which) he had stopped to buy something.

(ii) "Logical" relations (causal, concessive, instrumental, purposive, etc.), by contrast, cannot be expressed by simple adverbs and if interrogative pronouns are available at all, they are composite and often based in their form on one of the elementary relations. The corresponding adverbial clauses cannot be paraphrased by relative clauses. Moreover, expressions for elementary adverbial relations may give rise to those denoting logical

CONCESSIVE RELATIONS AS THE DUAL OF CAUSAL RELATIONS

191

relations, but not the other way round and this direction of semantic change and extension is also reflected in the order of the acquisition of these expressions. Adverbial relations are not only connected, however, through general processes of semantic extension and semantic change, but also through synchronic semantic relations. Purposive relations, for example, are generally considered as a variety of causal relations, i.e., as a combination of causality and volition or intention (cf. Saeb0, to appear), consecutive (resultative) relations can be analyzed as the converse of causal relations and there is also a close connection between conditionals and concessives, on the one hand, and conditional and causal relations, on the other. It is the goal of this paper to study in some detail the semantic connection between causal and concessive relations. There is an intuition of long standing that concessive expressions are somehow the opposite of causal expressions and this intuition manifests itself in such labels as incausal, inopérant cause, anticause, cause contraire, or condition contrecarré, which have often been used in descriptions of English, German and French. The explication offered below for this traditional intuition is the analysis of concessive adverbials are the dual counterpart of causal adverbials. In the light of this analysis and against the background of the general discussion of duality presented in Löbner (1987, 1990) the cross-linguistic

observations made in König (1988) on concessives

connectives will be reexamined and it will be shown that this analysis sheds some interesting light on a variety of synchronic and diachronic facts relating to the form and use of concessive connectives. The outline of the paper is as follows: I will first summarize the main observations made in König (1988) on the specific properties of concessive relations and the different types of connectives used across languages to encode them. In section 2 evidence will be presented for the view that there is a semantic relation of duality between causal and concessive adverbials. The possibilities that this perspective offers us for the explanation of a variety of synchronic and diachronic facts will be explored in section 3 and 4.

1. Properties of concessive sentences and types of concessive connectives Concessive sentences have a number of special syntactic and semantic properties which distinguish them from most other types of complex sentences with adverbial clauses. These properties are discussed in great detail in König (1988) and will only be summarized here very briefly: The most striking syntactic property of concessive adverbials is that they cannot function as the focus of an operator or as utterance focus.This restriction seems to underlie the following more specific properties and restrictions:

192

EKKEHARD KÖNIG

(i) There are no interrogative pronouns for concessive relations analogous to whv (causal), what for (purposive) or how (manner).2 Nor can concessive adverbials be used as a complete answer to a question. (ii) In contrast to all other adverbials, concessives cannot combine with focus particles such as only, just, also. not...but. etc. (3)

a.

He does it only because/ if / in order to p.

b.

*He does it only although p.

Note, however, that concessive connectives in many languages derive from a combination of a conditional or temporal connective and a focus particle like even or also. (iii) There are no cleft constructions for concessive adverbials. A further striking formal property of concessive adverbials is that the connectives expressing these relations (i.e., conjunctions, prepositions, conjunctional adverbs or affixes) are typically complex, but also transparent in their formal make-up, i.e., it is easy to relate the concessive meaning of these expressions to an earlier meaning or different use. As far as their interpretation is concerned, concessive sentences are a dead-end street for interpretative augmentation. Whereas sentences with temporal, conditional or local clauses can have a concessive interpretation in certain contexts, clauses explicitly marked as concessive can never be reinterpreted in any other way (cf. König, 1986). It could well be a consequence of this fact that not all languages have special concessive connectives. Moreover, concessive connectives seem to develop relatively late in the history of a language and are also acquired relatively late in the linguistic development of a child. These differences between concessive and other types of adverbial clauses seem to be connected with a difference in function: In contrast to other adverbials, concessive adverbials do not identify an important coordinate of the main clause: a time, a place, a condition, a reason, etc. What they identify is an unfavourable circumstance for an event or state, a fact which is reflected in the very term "concessive", which denotes a possible use rather than a two-term relation. I have shown in another paper that concessive connectives can be divided into five types on the basis of their etymology, the semantic domains they come from and the other uses their components may have. The following is a brief summary of this typology. Most of the time, however, new data are used for illustration. The first type of connectives derives from nouns originally denoting human reactions to actions, events or states, i.e., from notions originally only applicable to human arguments:

CONCESSIVE RELATIONS AS THE DUAL OF CAUSAL RELATIONS

(4)

193

TYPE A: E. in spite of; Gaelic a dh'aindeoin (aindeoin 'reluctance, defiance'); It. mal gr ado (chel (malgrado arch, 'discontent'); Rom. in. ciuda (ciuda 'anger'); Basque nahiz (volition + negation); Hung, dacâra (dac 'spite'); Dutch in weerwil van ('against someone's will').

In a wide variety of languages, concessive connectives exhibit a clear affinity to the notion of free choice or universal quantification, i.e., they contain a component that is also used as an operator licensing free choice of values for a variable. Type B contains all the connectives with such components: (5)

TYPE B: E. albeit, however, anyway: G. wiewohl. allerdings: Somali in kastâ oo 'although' (< quantity-whatever-and); Maori ahakoa 'although' ( concessive' can also be observed in the synchrony of a wide variety of languages. Plausible as such explanations may be, they do not suffice to account for many of the formal and etymological properties of concessive connectives described above. The present paper will therefore re-examine these data in the light of a new perspective on the meaning of concessive relations. And such a perspective is provided by analyzing concessive relations as the dual of causal relations.

2. Concessive relations as the dual of causal relations Complex sentences with clauses of cause or reason and sentences with concessive clauses are intuitively felt to be related. In the recent past Hermodsson (1978) has most strongly expressed this view in arguing for a replacement of the traditional term 'concessive' by what he thought was a more appropriate term, namely 'incausal'. The most obvious property that causal and concessive constructions such as (11) and (12) share is their 'factual' character, i.e., they entail both of their component clauses: (11) Since/because John has lost a lot of money , he is unhappy. (12) Although John has lost a lot of money, he is not unhappy. Another common feature of these two types of complex sentences is sometimes seen in a presupposition, which is assumed to be associated with both causal and concessive connectives. This background asumption can be formulated as a correlative statement or 'default conditional' and concerns the general connection between losing money and being unhappy in the case of our examples. The meaning of sentence forms such as (13)a. and (14)a. can be analyzed according to this view as follows (König, 1986; Zaefferer, 1990):

196

EKKEHARD KÖNIG

(13) a.

Since/becausep, q.

b. c. (14) a.

p&q if p, normally q (presupposition) Although/even though p, not-q.

b. c.

p&—q if p, normally q (presupposition)

This description is very problematic, however, since the specific contributions made to the meaning of a sentence by a causal and a concessive connective over and above the factuality of the two clauses exhibit different projection behaviour, as the following examples show: (15) If John was fired because he took part in the strike, we have to take immediate action. (16) If John was fired although he did not take part in the strike we have to take immediate action. The conditions that have to be checked in the case of (16) can be expressed by a coordination: p & q . The presupposition associated with the concessive connective (cf. (14)c.) is inherited by the whole sentence (cf. Metrich, 1983: 98f.). In (15), by contrast, the question whether John's participation in the strike was the reason for his being fired is part of the meaning of the antecedent. We, therefore, have to look for a different way of explicating the traditional intuitions about the antithetical relationship between causals and concessives. Consider pairs of sentences like the following: (17) a. b.

This house is no less comfortable because it dispenses with air-conditioning, This house is no less comfortable, although it dispenses with airconditioning.

These two sentences, which differ only in that (17)a. contains a causal connective where (17)b. contains a concessive conjunction, are clearly equivalent in meaning. Since it would be counterintuitive to conclude that although and because must therefore be synonymous in the context given, this equivalence can only be due to different 'interaction' between the adverbial clauses and the negation in these sentences. In other words, it must be due to a difference in relative scope of the relevant operators and thus to a different composition of the two sentences: not(P) ^

Expressions in diagonally opposite corners of this square manifest the relation of duality. The reason why this relation is of great interest for the semantic analysis of natural languages is that in most cases two or more corners of this square are lexicalized differently. Thus groups of two, three or four expressions can be formed on the basis of this relation and the semantic analysis of one of these expressions automatically gives us the analysis for the rest. The relevant test for such grouping is the equivalence between external and internal negation of sentences with two such expressions. Lobner (1990) has shown that duality structures many domains in the vocabulary of a language and that the following lists of expressions are instances of such duality groups: a. some,

all,

no,

b. sometimes,

always,

c. also,

only,

never, »

d. or,

and,

neither...nor,

e. possible,

necessary,

f. can,

must,

impossible, ~>

g. let,

force,

prevent,

~

unnecessary "

198

EKKEHARD KÖNIG

h. already,

still,

not yet,

no longer

On the basis of tests like (17) and (20), because and although, on the one hand, and by and until, on the other, can be added to this list: (20) a. The pub will not be closed by ten. s b. The pub will be open (= not-closed) until ten. (19) i. because, j. by,

although, until,

The best-known example of a dual pair of expressions are, of course, the existential and universal quantifier of predicate logic, as is shown by the following equivalences:4 (21) a. - . 0 * ) f(x) s (Vx) - f ( x ) b. -.(Vx) f(x) s (3x) -,f(x) Another important aspect of Lobner's analysis of duality that needs to be mentioned here is the possibility of ordering the expressions that enter into a duality group on a scale and of distinguishing thus between four types of expressions. The basis of this ordering is first an intuitive one: the expressions that roughly correspond to the existential quantifier of predicate logic are taken as starting point (type 1), their dual counterpart is taken as type 2 expression and these are then followed by their negations: 3, V, —13, -iV. Later more solid criteria are provided for this ordering, based on general properties of quantifiers: monotonicity, persistence and consistency: Type 1 and type 2 expressions are monotone increasing in the sense of Barwise & Cooper (1981), whereas type 3 and type 4 expressions are 'negative' and thus monotone decreasing. Type 1 and type 4 expressions, on the other hand, are persistent, type 2 and type 3 expressions are not. The four types of expressions distinguished and ordered on the basis of these and other general properties of quantification are then compared in terms of the complexity of their lexicalization and Lobner observes that in going from type 1 to type 4 expressions we find an interesting asymmetry across the world's languages: type 1 and 2 are generally lexicalized, type 3 rarely and type 4 only in exceptional cases. Furthermore, there tends to be an increase in the complexity of the lexicalization along the scale. The asymmetry is very clear between type 1 and type 2 expressions, on the one hand, and expressions of type 3 and 4, on the other. As far as expressions of type 1 and 2 are concerned, the evidence is not so clear, but there is at least some evidence that type 1 expressions are simpler and unmarked in several respects. The examples given in (19) illustrate the ordering and the asymmetries just described.

CONCESSIVE RELATIONS AS THE DUAL OF CAUSAL RELATIONS

199

This is only a very rough sketch of Lobner's analysis of duality and the expressions exhibiting this relation, which is meant to provide just enough information for a discussion of the connection between concessive and causal adverbials from this perspective. I have already used the standard test for duality in (17) to show that causal and concessive adverbials do in fact manifest this relationship. Since the example used, however, was a very special one, it seems appropriate to spend some more effort on demonstrating that the external negations of causal constructions and the internal negations of the corresponding concessive constructions are indeed equivalent. The main problem is to clearly identify the external negation of a causal construction among several possible negative versions of such sentences. Consider our example (11), repeated here (in a slightly modified version) as (22)a., and the two negated versions given below: (22) a.

Because he has lost a lot of money, John is unhappy.

b.

Because he has lost a lot of money, John is not unhappy.

c.

John is not unhappy because he has lost a lot of money, but because...

The first of these two negated versions clearly does not qualify as the external negation of (22)a., since the negation has only narrow scope within the subordinate clause. In (22)c. the negation has the required wide scope, but this sentence cannot be the external negation of (22)a. either, since the negation is a focusing one, i.e., a negation which does not only indicate that a sentence is false, but also indicates which part of a sentence has to be replaced by an alternative expression to turn it into a true one. That this focusing negation, expressed in English by not...but, is special in many respects has been established by a variety of recent studies (cf. Horn, 1985). Unfortunately, however, this focusing interpretation is the most plausible one for negated sentences with adverbials, for reasons we cannot go into here. The negation that makes it possible to paraphrase a causal construction by a concessive one has wide scope, does not interact with focus and relates to the causal connection between main clause and subordinate clause. This means that the causal connnection between two clauses and, as a result, also the content of the main clause is rejected. To express this meaning simply by inserting a negation into (22)a. is possible if the nuclear tone is on the negation (cf. (22)d.). Even here, however, a focusing interpretation may interfere, so that the periphrastic versions (22)e.-f. are perhaps clearest: (22) d.

John is not unhappy because he has lost a lot of money.

e.

Losing a lot of money has not made John unhappy.

f.

John has lost a lot of money without therefore being unhappy

200

EKKEHARD KONIG

The distinction between a wide-scope focusing negation (cf. (22)c.) and a non-focusing negation with the same scope (cf. (22)d.-f.) has a parallel in the following focusing and non-focusing use of an interrogative operator: (23) a. b.

Are we racists because we do business with South Africa (or because...) Are we racists because we do business with South Africa? (Does business with South Africa make us racists?)

These problems with additional, interfering inteipretations can be avoided with examples where two main clauses are linked by a causal conjunctional adverb or 'conjunct'. The following examples are again clear cases of the equivalence relation relevant for the semantic relation of duality:5 (24) a. b.

He lost a lot of money. He has not become a poor man because of that, He lost a lot of money. He has not become a poor man despite that

Again, we must assume that the negation has wide scope over the whole sentence in (24)a., but narrow scope relative to the adverbial in (24)b. In German the equivalence of sentences such as (24)a. and (24)b. has given rise to blending, i.e., both a causal and a concessive conjunctional adverb may appear in the German version of (24), and the same phenomenon is illustrated by the English example in (26): (25) Er hat viel Geld verloren. Deswegen ist er trotzdem kein aimer Mann geworden. (26) He is a grammar-school man, though none the worse for that. A further piece of evidence for the view that there is a dual relationship between causal and concessive adverbials is the fact that concessive sentences can be used to reject an argument with a conclusion introduced by therefore. Using Toulmin's famous example (23)a. J. Klein has pointed out that in rejecting (23)a. with the help of a sentence like (23)b., a speaker accepts the premise as well as the background assumption relating to the connection between the two states (Toulmin's warrant). The only part of the argument that is rejected is the conclusion. (27) a. b.

Harry was born on the Bermudas. Therefore he is a British citizen. Although Harry was born on the Bermudas, he is not a British citizen.

CONCESSIVE RELATIONS AS THE DUAL OF CAUSAL RELATIONS

201

Now that it has been established that there is a relationship of duality between causal and concessive adverbials,6 what remains to be done is to assign these two operators to one of Lobner's types. Unfortunately, Lobner's criteria are not clearly applicable to these cases, so that it is more on the basis of their similarity to other expressions of these two types that I will regard causal adverbials as type 1 and concessive adverbials as type 2 expressions. Thus we get the following duality square: (28)

(because p)q

internal negation

(because p) A

external negation

-i((because p) q) internal negation (although p) —q

external negation

-i((because p) -u?) although p, q

3. The typology of concessive connectives revisited Identifying concessive adverbial as the dual counterpart of causal adverbials does not give us a semantic analysis for the former, but only a criterion of adequacy for such an analysis. A semantic analysis must be formulated in such a way that it captures this relationship and given the fact that the operations of external and internal negation take us from one expression to its dual counterpart, an analysis of causal adverbials automatically gives us the analysis of concessive operators. What the preceding discussion does offer us, however, is a new perspective for an explanation of the properties exhibited by concessive connectives across the world's languages. We will therefore now re-examine some of these properties in the light of the discussion presented in the preceding paragraph. First, we may note that concessive connectives are typically more complex in their formal make-up than causal connectives. On the basis of cross-linguistic observations made about quantifiers, functors and operators in Dohmann (1974), and Payne (1985), there is no reason to assume that type 2 expressions involve a more complex type of lexicalization than type 1 expressions (cf. Lobner, 1990). There is, however, a certain amount of evidence that type 2 expressions are more complex as far as processing is concerned and as far as the context is concerned that they create. If we were right in

202

EKKEHARD KÖNIG

assigning concessive adverbials to Lobner's type 2 expressions, they provide a clear example of the greater complexity of this type. The fact that they cannot be the focus of an operator or an utterance is a manifestation of this complexity. An even more interesting phenomenon in the context of our discussion is the formal similarity between concessive connectives and other type 2 expressions. Free choice expressions and universal quantifiers are a frequent component of concessive connectives (cf. Type B and E). Moreover, E. all and its counterparts in other languages may combine with all types of connectives and may turn an unspecific connective like the German preposition hsi or the French gérondif (en V-anO into a concessive one: (29) a.

G. Trotzdem/trotz allem ist er nett. 'He is nice all the same'

b.

Bei all seinen Fehlern ist er doch sehr nett.

c.

F. Il était d'accord tout en gardant ses objections fondamentales.

'For all his faults, he is still very nice.'

In addition to all, another type 2 expression, namely still, is a frequent component of concessive expressions (cf. G. dennoch. Sp. aunque , Fr. encore que. E. sîill. y£l, etc.) and the counterpart of E. only is used in some languages in the sense of 'but'. Finally, the cross-linguistic regularities discussed in section 1 and more fully in Kônig (1988) show that there is a certain degree of formal similarity between causal connectives and concessive connectives. In four of the 80 languages represented in my sample (Cambodian, Japanese, Mundari, Indonesian) concessive connectives can be derived from causal connectives through the addition of certain particles. A more frequent phenomenon is the use of negated causal expressions as markers of concessivity. Connectives like G. ungeachtet. E. unimpressed by. regardless. Fr. n'empêche que ('does not prevent'), Finn, huolimatta. D. ondanks ('ungrateful, thoughtless'), Russ. nesmotria na lo ¿to ('regardless of), nezavisimo oî togo sto ('independent of), or Mand. bugùo are cases in point.7 Moreover, negative markers of proportional correlation such as E. nevertheless, nonetheless. Lat. nihilo minus, nihilo secius. Fr. néanmoins, etc. can also be regarded as negative forms of basically causal expressions. In the natural sciences statements about coiTelations underlie statements or hypotheses about causal relations and for a child to learn expressions for causal or concessive relations presupposes some knowledge about correlations between states and events. So, given the dual relationship between causal and concessive adverbials, it is not surprising that negative forms of connectives that express correlations or proportional functions should be used as expressions of concessivity.8

CONCESSIVE RELATIONS AS THE DUAL OF CAUSAL RELATIONS

203

4. Semantic changes from causal to concessive The assumption that there is dual relationship between concessive adverbial and adverbials of cause and reason also enables us to explain some otherwise puzzling facts of polysemy and semantic change. In Old English foe was the most common causal connective and Scandinavian languages still use a cognate form in this function (Norw. fordi. Swed. darfor attl. In Modern English, for can still be used in a causal sense, but apart from the conjunctional adverb therefore, this use of for has become somewhat marginal. The conjunction for is very formal and literary in its stylistic quality and as a preposition for is only used in a causal sense in very specific contexts: in combination with nouns denoting services rendered (= in return for), degrees, in combination with comparatives and in a limited number of fixed collocations: (30) a.

He was rewarded for his bravery.

b.

We could hardly see for the thick fog.

c.

...though none the worse for that.

d.

for fear of, for want of, notorious for, for this reason

In addition to this causal and various other uses, for also has a concessive use. The concessive interpretation is in fact the only possible one, whenever ¿21 combines with all: (31) a.

For all his personal responsibility for the manipulation of power, he has not done anything illegal,

b.

He can be very selfish, but he is not unpleasant to work with for all that

That the concessive interpretation of for should require a supportive marker all is not really surprising in view of what was said above about the typical form of concessive connectives, but it is certainly remarkable that the same form should have both a causal and a concessive meaning and that the latter should be derived from the former.9 But if, as we have assumed, concessive adverbials are the dual counterpart of causal adverbials, this development can be regarded as the result of reanalysis. What seems to have happened very early in English, but in none of the other Germanic languages, was a reanalysis of the scope of negation in negative sentences with causal for, such that the adverbial received wide scope relative to the negation. As a result, the interpretation of the adverbial relation was changed from 'causal' to 'concessive'. In other words, sentences like the following retained their overall interpretation, but as a result of different composition:

204

EKKEHARD KÖNIG

(32) a.

He gave us generous financial support. He has not become poorer for (all) that ('causal', wide scope of negation)

b.

He gave us generous financial support. (He has not become poorer) for all that ('concessive', narrow scope of negation)

Such reanalyses of the relative scope of negation and adverbials and the concomitant reinterpretation of the adverbials have often been observed in the history of English and other languages. The development of

(< OE giet. gvt. etc.) from a distributionally

unrestricted adverb with a meaning roughly equivalent to that of Modern English still to a negative polarity item ('already') and the development of dialectal anymore ('nowadays') are two cases in point (cf. König & Traugott, 1982). (33) Life is getting difficult anymore. In French, the development of pour autant (< Latin alterum tantumt from a causal to a concessive connective seems to have been very similar to that of for (all). Up to the 16th century this form was used purely in a causal sense like done. In Modern French, this connective also has a concessive sense and, interestingly enough, it is a negative polarity item in this function, i.e., it only occurs in such contexts where causal and concessive adverbials may lead to the same interpretation and thus to reanalysis: (34) II est riche. II n'est pas pour autant heureux. Reanalysis on the basis of duality also seems to be the reason for the fact that a concessive use is ascribed to the causal conjunctional adverb darum in most dictionaries of German. The examples given to illustrate that use are almost exclusively negative sentences. (35) Er hat gute Arbeit geleistet, aber darum ist er noch kein Meister. 'He has done a good job, but that doesn't mean he is perfect.'

CONCESSIVE RELATIONS AS THE DUAL OF CAUSAL RELATIONS

205

5. Conclusion Many grammarians have expressed the view that causal and concessive adverbials are related in some way, but so far no convincing explication has been given for this intuition. On the basis of the general analysis of duality developed in Löbner (1987, 1990), I have tried to show that concessive relations can be analyzed as the dual counterpart of causal relations. If the arguments presented for this view are conclusive, this means that the relationship between these two types of adverbials is yet another instance of a logical relation that has been shown to structure wide domains of the vocabulary of natural languages. With the help of this new perspective on the meaning of concessive adverbials, I have then re-examined the typology of concessive connectives presented in König (1988) and shown that this perspectives offers new possibilities for the explanation of some properties that concessive connectives exhibit across the world's languages. In fact, some types of connectives, directly reflect the dual relationship between the two types of adverbials. As pointed out above, the identification of concessive adverbials as the dual counterpart of causal adverbials does not give us a semantic analysis for the former, but only a criterion of adequacy for such an analysis. But, according to Löbner, whenever such dual relations are found, they serve as an indication that the lexical elements thus related should be analyzed as quantifiers. It is along these lines that further research on concessivity can fruitfully be pursued. To round off this paper, I would like to mention three problems for which I have not found a satisfactory solution and which thus weaken somewhat the force of the arguments presented above. First, the equivalence between suitable negative versions of causal and concessive constructions only holds if the causal connection is based on an underlying correlation between the events or states in question. In other words, the external negations of causal constructions which contain 'private, specific' reasons rather than general causes do not admit of a concessive paraphrase: (36) a. b.

Not ( I hit him because I did not like his face) Although I did not like his face, I did not hit him.

Does this mean that our analysis has to be restricted to certain contexts or that there are two kinds of causal constructions only one of which is related to concessive sentences? Secondly, the duality square suggests that we can derive a causal construction through a external negation of a concessive sentence. But this is not the case. Concessive sentences cannot be negated and concessive adverbials can hardly ever be found in the scope ot another operator.

206

EKKEHARD KÖNIG

Thirdly, I have assumed throughout the paper that all concessive connectives have more or less the same meaning, so that there is only one type of concessive construction. This, however, is a gross simplification. There at least two types of concessive constructions, often called 'rectifying' and 'argumentative' concessives (cf. Morel, 1983; Metrich, 1983), which are notrelatedto causal constructions in the way described above. How many types of concessive constructions are to be distingushed on the basis of the meaning of the connectives employed and the nature of the terms entering into this relation is another important question for futureresearchon concessives.

Notes * Thanks are due to A. Conte, S. LObner, H.-J. Sasse, P. Seuren and D. Zaefferer for helpful comments on an earlier version of this paper.

1. Cf. Thompson & Longacre (1985) for a recent typology of adverbial clauses. Their list also includes labels like "substitutive" and "circumstantial" for examples like the following:

(i)

He went to the theatre instead of working.

(ii)

a.

He left without saying a word,

b.

He left under protest.

2. Given that interrogative pronouns in Germanic languages are typically parallel in form to conjunctional adverbs (cf. German wann - dann. warum - darum. wozu - dazu) we would expect something like *wennoch as a concessive interrogative pronoun, parallel to dennoch.

3. In Tibeto-Burman languages, concessive connectives frequently have the same form as locative postpositions (cf. Genetti, 1986). But since the latter may also change to temporal and conditional connectives, it is not clear whether there is a direct path from locational cooccurrence to concessivity or whether concessive connectives derive from a locative meaning via a temporal or conditional meaning.

4. The quantifiers also figure in another logical square, the Aristotelian square, which exhibits the implicational relations between the members of the group: implication, compatibility, contrarity (cf. Reichenbach, 1947:91ff.) and should not be confused with the duality square given above.

5. On the basis of this test, conditionals and concessive conditionals could also be considered as a dual pair. Whether (ii) can really be regarded as the negation of (i), however, is a controversial issue:

CONCESSIVE RELATIONS AS THE DUAL OF CAUSAL RELATIONS

(i)

If you ask him, he will help you.

(ii)

Even if you ask him, he won't help you.

207

6. Further support for this assumption comes from constructions of the type 'p. Not that q. But r ': (i)

The authors are to be congratulated on their book. Not that the account isn't flawed, but what account isn't

A sentence introduced by not that establishes a link both to the preceding and to the following sentence. With regard to the preceding sentence, these two words indicate that what follows is a possible, but an unjustified inference on the basis of what was said before. On the other hand, a sentence introduced by not that serves as premise, or emphatic assertion, for a following restriction. In other words, the relationship between such a sentence and the preceding one can be analyzed as the negation of an 'internal causal' relationship in the sense of Halliday & Hasan (1976). It is denied that the utterance p i s a reason for the conclusion q: p. Not that q. This means that texts of the type (i) can roughly be analyzed as follows:

(ii)

Not (p. Therefore q) or not ((because p) q)

On the basis of the central thesis of this paper, we would expect that the relevant relation can also be expressed by a concessive connective, which is indeed the case. The expression in (ii) can be paraphrased by 'Although p, not-