Possible Worlds in Humanities, Arts and Sciences: Proceedings of Nobel Symposium 65 [Reprint 2010 ed.] 9783110866858, 9783110112207

Citation preview

Possible Worlds in Humanities, Arts and Sciences

Research in Text Theory Untersuchungen zur Texttheorie Editor Jänos S. Petöfi, Bielefeld Advisory Board Irena Bellert, Montreal Antonio Garcia-Berrio, Madrid Maria-Elisabeth Conte, Pavia Teim A. van Dijk, Amsterdam Wolfgang U. Dressler, Wien Nils Erik Enkvist, Abo Peter Hartmann f, Konstanz Robert E. Longacre, Dallas Roland Posner, Berlin Hannes Rieser, Bielefeld Dieter Viehweger, Berlin/DDR Volume 14

W DE

G Walter de Gruyter · Berlin · New York 1989

Possible Worlds in Humanities, Arts and Sciences Proceedings of Nobel Symposium 65 Edited by Sture Allen

W DE

G Walter de Gruyter · Berlin · New York 1989

Printed on acid free paper (ageing-resistant - pH 7, neutral)

Library of Congress Cataloging-in-Publication Data Nobel Symposium 65 (1986: IBM Nordic Education Center) Possible worlds in humanities, arts and sciences. (Research in text theory, ISSN 0179-4167; v. 14 = Untersuchungen zur Texttheorie) Symposium held Aug. 11-15, 1986. I. Allen, Sture, 1928. II. Nobelstiftelsen. EQ. Title. IV. Series: Research in text theory; v. 14 AC5.N6 1986 082 88-37349 ISBN 0-89925-301-6 (U.S.: alk. paper)

Deutsche Bibliothek Cataloguing in Publication Data Possible worlds in humanities, arts and sciences: proceedings of Nobel Symposium 65 / ed. by Sture Allen. - Berlin; New York: de Gruyter, 1989 (Research in text theory; Vol. 14) ISBN 3-11-011220-5 NE: Allen, Sture [Hrsg.]; Nobel Symposium (65, 1986, Lidingo); GT

ISSN 0179-4167 © 1988 by Walter de Gruyter & Co., Berlin 30 and the Nobel Foundation Printed in Germany All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced in any form - by photoprint, microfilm or any other means - nor transmitted nor translated into a machine language without written permission from the publisher. Typesetting: Dörlemann Satz, Lemförde Printing: W. Hildebrand, Berlin Binding: Lüderitz & Bauer, Berlin

Preface

The Nobel Symposium on Possible Worlds was arranged as part of the bicentenary celebration of the Swedish Academy in 1986. It was held on August 11-15, just like its predecessor, the Nobel Symposium on Text Processing in 1980 (the proceedings of which are available from Almqvist & Wiksell International, Stockholm). The Symposium was sponsored by the Nobel Foundation through its Nobel Symposium Fund. We met in Lidingö on the outskirts of Stockholm at the IBM Nordic Education Center, which was generously put at the disposal of the Symposium by IBM. The members of the Organizing Committee were Professors Lars Gyllensten and Ulf Linde, members of the Swedish Academy, Professors Tore Frängsmyr and Bengt Nagel, members of the Royal Swedish Academy of Sciences, and myself, chairman. Several colleagues were helpful in practical matters: Mr. Dieter Huber, University of Göteborg, whose organizational skill was admirably demonstrated; Dr. Lars Arosenius of IBM and the staff of the Education Center, who kindly gave us the benefit of their experience; Mrs. Gerd Bengtsson, University of Göteborg, and Ms. Monica Holmgren, of the Swedish Academy, who provided professional support as secretaries; and Dr. Bo Svensen, of the Swedish Academy, whose proofreading included many valuable observations. Thanks are due to all those mentioned as well as to Professor Jänos S. Petöfi, University of Bielefeld, who kindly suggested the inclusion of this volume in the series where it appears. And thank you very much, dear participants, for making the Symposium a memorable occasion! When you discover your next planet, don't forget to look for its name. March 1988

Sture Allen Permanent Secretary of the Swedish Academy Professor of Computational Linguistics, University of Göteborg

CONTENTS

Preface

V

Sture Allen Opening Address

l

Session l: Philosophy Thomas S. Kühn Possible Worlds in History of Science

9

Arthur I. Miller Discussion of Thomas S. Kuhn's paper "Possible Worlds in History of Science"

33

Tore Frängsmyr Discussion of Thomas S. Kuhn's paper "Possible Worlds in History of Science"

42

Thomas S. Kühn Speaker's Reply

49

Jaakko Hintikka Exploring Possible Worlds

52

Gunnar Olsson On Doughnutting. Discussion of Jaakko Hintikka's paper "Exploring Possible Worlds"

74

William R. Shea Transworld Journeys. Report on Session 1: Philosophy

82

Session 2: Linguistics Barbara H. Partee Possible Worlds in Model-Theoretic Semantics: A Linguistic Perspective

93

!

Contents

John Perry Possible Worlds and Subject Matter. Discussion of Barbara H. Partee's paper "Possible Worlds in Model-Theoretic Semantics: A Linguistic Perspective" 124 Hannes Rieser Some Caveats with Respect to Possible Worlds. Discussion of Barbara H. Partee's paper "Possible Worlds in Model-Theoretic Semantics: A Linguistic Perspective" 138 Barbara H. Partee Speaker's Reply

152

Nils Erik Enkvist Connexity, Interpretability, Universes of Discourse, and Text

Worlds Wolfgang Heydrich Possible Worlds and Enkvist's Worlds. Discussion of Nils Erik Enkvist's paper "Connexity, Interpretability, Universes of Discourse, and Text Worlds"

162

187

Ulf Teleman The World of Words - and Pictures. Discussion of Nils Erik Enkvist's paper "Connexity, Interpretability, Universes of Discourse, and Text Worlds" 199 Jänos S. Petöfi Possible Worlds - Text Worlds. Quo vadis Linguistica? Report on Session 2: Linguistics 209

Session 3: Literature and Arts Lubomir Dolezel Possible Worlds and Literary Fictions

221

Nicholas Wolterstorff Discussion of Lubomir Dolezel's paper "Possible Worlds and Literary Fictions"

243

Thomas Pavel Fictional Worlds and the Economy of the Imaginary

250

Samuel R. Levin Duality and Deviance: Two Semantic Modes. Discussion of Thomas Pavel's paper "Fictional Worlds and the Economy of the Imaginary"

260

Contents

DC

Inge Jonsson Fictional and Ontological Landscapes. Discussion of Thomas Pavel's paper "Fictional Worlds and the Economy of the Imaginary"

267

Lars Gyllensten Possible Worlds - A Chorus of a Multitude of Souls. A Writer's Perspective

272

Frederik Julius Billeskov Jansen Possible Worlds: A Historical Perspective in Literature. Discussion of Lars Gyllensten's paper "Possible Worlds A Chorus of a Multitude of Souls. A Writer's Perspective"

293

Kurt Johannesson Possible Worlds - But for Whom? Discussion of Lars Gyllensten's paper "Possible Worlds - A Chorus of a Multitude of Souls. A Writer's Perspective" 302 Lars Gyllensten Speaker's Reply

308

Ulf Linde Image and Dimension

312

Arthur C. Danto Pictorial Possibility. Discussion of Ulf Linde's paper "Image and Dimension"

329

Gerard Regnier Discussion of Ulf Linde's paper "Image and Dimension"

336

Umberto Eco Report on Session 3: Literature and Arts

343

Session 4: Natural Science J. S. Bell Six Possible Worlds of Quantum Mechanics

359

Claes Äberg On Characterizing Possible Worlds of Physics. Discussion of J. S. Bell's paper "Six Possible Worlds of Quantum Mechanics" . . . 374 Bengt Nagel Discussion of J. S. Bell's paper "Six Possible Worlds of Quantum Mechanics"

388

X

Contents

Martin]. Rees Our Universe and Others: The Limits of Space, Time and Physics

396

Jean-Claude Pecker How to Describe Physical Reality? Discussion of Martin J. Rees's paper 'Our Universe and Others: The Limits of Space, Time and Physics"

417

Bengt Gustafsson The Uniformity of the Universe. Discussion of Martin J. Rees's paper "Our Universe and Others: The Limits of Space, Time and Physics" 428 Walter Thirring Report on Session 4: Natural Science

439

List of Contributors

445

STURE ALLEN Opening Address

Dear Colleagues, It is my pleasure and privilege to welcome you on behalf of the Swedish Academy and its Organizing Committee. I trust that you are familiar with the theme of this Nobel Symposium, "Possible Worlds in Arts and Sciences". We note with great satisfaction that such a distinguished group of scholars have come together to discuss such a challenging subject in the year of the 200th anniversary of the Academy. With your permission, I also bring a greeting from a possible world (Figure 1). As you can see, it is a rather superreflective one. The picture by Guy Billout is part of a series called "Imaginary Gardens". The theme of the Symposium is truly interdisciplinary, encompassing several academic disciplines, even representing what I hate to call "the two cultures". Thus, we are happy to see participants from natural science as well as philosophy, linguistics, literature, and the visual arts. Nobel Symposia are sponsored by the Nobel Foundation through its Nobel Symposium Fund. They are arranged in accordance with the intentions of Alfred Nobel, the founder of the Nobel Prizes. Over the last twenty years, a series of symposia have been organized, attracting a great deal of attention in their respective fields. This is one of several Nobel Symposia supported by the Swedish Academy, whose domain is language and literature. Some of you, a very small minority, were right here at the IBM Nordic Education Center exactly the same days six years ago, when we held the Nobel Symposium on "Text Processing", discussing aspects of text representation, text analysis and generation, and text typology and attribution. As veterans you will recall that, in our deliberations on text-understanding problems, the notion of possible worlds was touched upon. That was where the idea of the present Symposium was born. Even if, accordingly, the impetus of the Symposium was linguistic, it is obvious that a concept of possible worlds is used as a tool in many other fields, particularly those I mentioned. This comes natural if we think of the general setting put forward in the invitation. In the Actual World, there are various Media of Communication, which have developed gradually. Such media of great importance are natural language in

S. Auen

Fig. l

Speech and writing, formal languages including mathematical and logical notation as well as programming languages, and also visual art and music. It is possible to use these media for a wide range of purposes, such as practical, scientific, and artistic. The result in each case is a Manifestation. This can be a text of a matter-of-fact or belletristic type, a piece of art, or a scientific theory or model. The design and the interpretation of such a manifestation can be regarded as the establishment of a Possible World. Both processes involve a number of interesting problems. The main point concerns the relationship between man, the medium, the actual world, and the possible world established. This includes, of course, the problem of representation or mimesis versus self-reflexivity or autonomy, as well as the fundamental questions of whether a possible world created by art can

Opening Address

3

be an instrument for understanding the actual world and, conversely, whether scientific theories and models can also be regarded as fictional in some way. In the contributions presented, this ground is covered admirably, a host of important issues being brought up for discussion. Among these there are those of differences between language communities, truth-valuation and modalities, extension and intension in semantics, universes of discourse, fictionality and performative speech acts, ontological landscapes, polyphony and revelation, syntactic fiction and illusion, quantum mechanics and physical ambiguity as well as the role of a conscious observer and creation "ex nihilo". The normal sense of possible seems to be verbal: 'that can exist, be done, or happen' (The Concise Oxford Dictionary). The underlying Latin word is derived from posse 'be able'. This is also reflected in ordinary, general-purpose language. Consider one of the pictures by the artist Oscar Reuterswärd (Figure 2). It has been presented in the following way: "He has designed something which doesn't exist, which is not even possible." What does possible mean here?

Fig. 2

4

S. AUen

Since the picture is there, the painting of it is evidently feasible or realizable in the actual world. However, the quotation is not concerned with the picture but with the object, and such an object is certainly not feasible in that sense. Thus, in the current negated context, possible can be taken to mean 'feasible'. For obvious reasons, feasibility is a main concept involved in the use of possible in everyday language. Sometimes there is the additional flavour of suitability, a sort of functional modality, as in one reading of "the only possible answer". In a wider perspective, it is equally evident that the shape created, although not feasible, is imaginable. In terms of imaginability, possible includes, consequently, both feasibility and non-feasibility. The latter type isn't as common as the former in everyday language. Now consider the following sentence: "It is possible that the knows." This example is different in an important way. It means 'there is a chance, or maybe a risk, that he knows'. In other words, this is a quantitative utterance, although vague. It seems that the probability involved is usually rather low. In this way we have discerned three aspects: imaginability, feasibility, and probability. These days, when foreign submarines are said to traverse Swedish waters and are in fact sometimes embodied, a new quantitative expression has been coined. Certain observations are interpreted as indicating "a possible submarine". This phrase can probably be taken to be a nominalization of an adverbial construction like "it is, possibly, a submarine". In everyday usage there are, in addition, a number of set phrases: as fast/ much/soon as possible, where possible has an intensifying function, if possible, make possible, etc. There is also a substantival version: possibles meaning 'reasonable candidates' is a weaker forecast than probables meaning 'likely candidates'. For language users in general, it is obvious that the actual world has priority among the worlds. From a theoretical point of view, it need not be so. Yet, in a human perspective, one might think of all possible worlds, in one mode of thought, as ultimately entered from the actual world. Special-purpose language is chiefly general-purpose language. Its main characteristic is a set of special terms and maybe some particular phrasing. This also applies to the papers of this Symposium. The terminology used is inventive and revealing, as shown by the following examples: possibility, possibly true statements, lexically possible worlds, non-possible^worlds theories, possible psychological reality, impossibility, compossibility, physically possible worlds, non-actual possibles, fictionally authentic impossible worlds, possibilist, possibilism. One of the fascinating aspects of our theme, observed in several contributions, is that of dimensions. The superstrings proposed elsewhere as the vehicles of microcosmos are reported to require tens or hundreds of dimensions for their description. The theory of relativity offers four, three spatial and one time-like. If we restrict ourselves to the dimensions normally perceived, we find that, basically, sculpture works in three dimensions, painting in two, and language in one dimension. There are evidently things which you cannot sculpture but

Opening Address

5

which you can paint. There are likewise things which you cannot paint but which you can express in speech and writing, as, e.g., a paper on possible worlds. If we interpret this, in an admittedly straightforward way, as a case of inversely proportional entities - the decrease of dimensions being accompanied by an increase of expressiveness - we obtain zero, the empty space or the empty page, as the all-time-high of expressive force, the black hole, as it were. This, in turn, evokes the second stanza of the famous "Ode on a Grecian Urn" by John Keats: Heard melodies are sweet, but those unheard Are sweeter, therefore, ye soft pipes, play on; Not to the sensual ear, but, more endear'd Pipe to the spirit ditties of no tone. Dear colleagues, fellow-participants! Quoting Keats, I certainly don't invite you to keep silent. After all, our joint venture has begun.

Session 1: Philosophy

THOMAS S. KÜHN

Possible Worlds in History of Science"" The invitation to open this symposium's discussion of possible worlds in history of science has been particularly welcome, for several issues raised by the topic are central to my current research. Their centrality is, however, also a source of problems. In the book on which I am at work, these issues arise only after much prior discussion has led to conclusions I must here present as premises. Limited illustration and evidence for those premises will follow, but only in the later portions of this paper, where they are put to work. What I am presupposing will be suggested by the following claim: to understand some body ofpast scientific belief, the historian must acquire a lexicon that here and there differs systematically from the one current in his own day. Only by using that older lexicon can he or she accurately render certain of the statements that are basic to the science under scrutiny. Those statements are not accessible by means of a translation that uses the current lexicon, not even if the list of words it contains is expanded by the addition of selected terms from its predecessor. That claim is elaborated in the first of the three sections of this paper, and its relevance to an ongoing debate in possible-world semantics is briefly suggested in the second. The third section, an extended analysis of some interrelated terms from Newtonian mechanics, illustrates the entanglements of the lexicon with the substantive claims of a scientific theory, entanglements which may make it impossible to change the theory without changing the lexicon as well. Finally, the closing section of the paper examines the way such entanglements restrict the applicability to scientific development of the conception of possible worlds. A historian reading an out-of-date scientific text characteristically encounters passages that make no sense. That is an experience I have had repeatedly whether my subject was an Aristotle, a Newton, a Volta, a Bohr, or a Planck.1 It This paper has been considerably revised since the Symposium. For much relevant criticism and advice during that process, I am grateful to Barbara Partee and to my M.I.T. colleagues Ned Block, Sylvain Bromberger, Dick Cartwright, Jim Higginbotham, Judy Thomson, and Paul Horwich. For Newton, see my "Newton's '31st Query' and the Degradation of Gold," Isis, 42 (1951), 296-298. For Bohr, see John L. Heilbron and Thomas S. Kühn, "The Genesis of the Bohr Atom,"

10

Th. S. Kühn

has been standard to ignore such passages or to dismiss them as the products of error, ignorance, or superstition, and that response is occasionally appropriate. More often, however, sympathetic contemplation of the troublesome passages suggests a different diagnosis. The apparent textual anomalies are artifacts, products of misreading. For lack of an alternative, the historian has been understanding words and phrases in the text as he or she would if they had occurred in contemporary discourse. Through much of the text that way of reading proceeds without difficulty; most terms in the historian's vocabulary are still used as they were by the author of the text. But some sets of interrelated terms are not, and it is failure to isolate those terms and to discover how they were used that has permitted the passages in question to seem anomalous. Apparent anomaly is thus ordinarily evidence of the need for local adjustment of the lexicon, and it often provides clues to the nature of that adjustment as well.2 An important clue to problems in reading Aristotle's physics is provided by the discovery that the term translated 'motion' in his text refers not simply to change of position but to all changes characterized by two end points. Similar difficulties in reading Planck's early papers begin to dissolve with the discovery that, for Planck before 1907, 'the energy element hv' referred, not to a physically indivisible atom of energy (later to be called 'the energy quantum') but to a mental subdivision of the energy continuum, any point on which could be physically occupied. These examples all turn out to involve more than mere changes in the use of terms, thus illustrating what I had in mind years ago when speaking of the "incommensurability" of successive scientific theories.3 In its original mathematical use 'incommensurability' meant "no common measure", for example of the hypotenuse and side of an isosceles right triangle. Applied to a pair of theories in the same historical line, the term meant that there was no common language into which both could be fully translated.4 Some statements constituHistarical Studies in the Physical Sciences, \ (1969), pp. 211-290, where the nonsense passages that gave rise to the project are quoted on p. 271. For an introduction to the other examples mentioned see my "What are Scientific Revolutions?" in L. J. Daston, M. Heidelberger, and L. Krüger (eds.), The Probabilistic Revolution, vol. 1: Ideas in History (Cambridge, Mass., 1987), pp. 7-22. Throughout this paper I shall continue to speak of the lexicon, of terms, and of statements. My concern, however, is actually with conceptual or intensional categories more generally, e. g. with those which may reasonably be attributed to animals or to the perceptual system. For the support this extension receives from possible-world semantics see the paper by Barbara Partee in this volume. For a fuller and more nuanced discussion of this point and those that follow see my "Commensurability, Comparability, Communicability," in P. D. Asquith and T. Nickles (eds.), PSA 1982, vol. (East Lansing, 1983), pp. 669-688. My original discussion described non-linguistic as well as linguistic forms of incommensurability. That I now take to have been an overextension resulting from my failure to recognize how large a part of the apparently non-linguistic component was acquired with language during the learning process. The acquisition during language learning of what I once took to be incommensurability with respect to instrumentation is, for example, illustrated by the discussion of the spring balance in the next section of this paper.

Possible Worlds in History of Science

11

tive of the older theory could not be stated in any language adequate to express its successor and vice versa. Incommensurability thus equals untranslatability, but what incommensurability bars is not quite the activity of professional translators. Rather, it is a quasi-mechanical activity governed in full by a manual which specifies, as a function of context, which string in one language may, salva veritate, be substituted for a given string in the other. Translation ofthat sort is Quinean, and the point at which I aim will be suggested by the remark that most or all of Quine's arguments for the indeterminacy of translation can, with equal force, be directed to an opposite conclusion: instead of there being an infinite number of translations compatible with all normal dispositions to speech behavior, there are often none at all. With that much, Quine might very nearly agree. His arguments require that a choice be made, but they do not dictate its outcome. In his view, one must either entirely abandon traditional notions of meaning, of intension, or else one must give up the assumption that language is, or could be, universal, that anything expressible in one language, or by using one lexicon, can be expressed also in any other. His own conclusion - that meaning must be abandoned - follows only because he takes universality for granted, and this paper will suggest that there is no sufficient basis for doing so. To possess a lexicon, a structured vocabulary, is to have access to the varied set or worlds which that lexicon can be used to describe. Different lexicons - those of different cultures or different historical periods, for example - give access to different sets of possible worlds, largely but never entirely overlapping. Though a lexicon may be enriched to yield access to worlds previously accessible only with another, the result is peculiar, a point to be elaborated below. In order that the "enriched" lexicon continue to serve some essential functions, the terms added during enrichment must be rigidly segregated and reserved for a special purpose. What has made the assumption of universal translatability so nearly inescapable is, I believe, its deceptive similarity to a quite different one, in this case an assumption that I share: anything which can be said in one language can, with imagination and effort, be understoodby a speaker of another. What is prerequisite to such understanding, however, is not translation but language learning. Quine's radical translator is, in fact, a language learner. If he succeeds, which I think no principle bars, he will become bilingual. But that does not ensure that he or anyone else will be able to translate from his newly acquired language to the one with which he was raised. Though learnability could in principle imply translatability, the thesis that it does so needs to be argued. Much philosophical discussion instead takes it for granted. Quine's Word and Object provides a notably explicit case in point.5 I am suggesting, in short, that the problems of translating a scientific text, whether into a foreign tongue or into a later version of the language in which it W. V. O. Quine, Word and Object (New York, 1960), pp. 47, 70 f.

12

Th. S. Kühn

was written, are far more like those of translating literature than has generally been supposed. In both cases the translator repeatedly encounters sentences that can be rendered in several alternative ways, none of which captures them completely. Difficult decisions must then be made about which aspects of the original it is most important to preserve. Different translators may differ, and the same translator may make different choices in different places even though the terms involved are in neither language ambiguous. Such choices are governed by standards of responsibility, but they are not determined by them. In these matters there is no such thing as being merely right or wrong. The preservation of truth values when translating scientific prose is very nearly as delicate a task as the preservation of resonance and emotional tone in the translation of literature. Neither can be fully achieved; even responsible approximation requires the greatest tact and taste. In the scientific case, these generalizations apply, not only to passages that make explicit use of theory, but also and more significantly to those their authors took to be merely descriptive. Unlike many people who share my generally structuralist leanings, I am not attempting to erase or even to reduce the gap generally thought to separate literal from figurative use of language. On the contrary, I cannot imagine a theory of figurative use - a theory, for example, of metaphor and other tropes - that did not presuppose a theory of literal meanings. Nor, to turn from theory to practice, can I imagine how words could be employed effectively in tropes like metaphor except within a community whose members had previously assimilated their literal use.6 My point is simply that the literal and the figurative use of terms are alike in their dependence on preestablished associations between words. That remark provides entree to a theory of meaning, but only two aspects of that theory are centrally relevant to the arguments which follow, and I must here restrict myself to them. First, knowing what a word means is knowing how to use it for communication with other members of the language community within which it is current. But that ability does not imply that one knows something that attaches to the word by itself, its meaning, say, or its semantic markers. Words do not, with occasional exceptions, have meanings individually but only through their associations with other words within a semantic field. If the use of an individual term changes, then the use of the terms associated with it normally changes as well. The second aspect of my developing view of meaning is both less standard and more consequential. Two people may use a set of interrelated terms in the same way but employ different sets (in principle totally disjunct sets) of field coordinates in doing so. Examples will be found in the next section of this paper; meanwhile the following metaphor may prove suggestive. The United States 6

See my "Metaphor in Science," in Andrew Ortony (ed.), Metaphor and Thought (Cambridge University Press, 1979), pp. 409-419.

Possible Worlds in History of Science

13

can be mapped in many different coordinate systems. Individuals with different maps will specifiy the location of, say, Chicago by means of a different pair of coordinates. But all will nevertheless locate the same city provided that the maps are scaled to preserve the relative distances between the items mapped. The metric that accompanies each of the various sets of coordinates must, that is, be chosen to preserve the structural geometrical relations within the mapped area.7 The premises just sketched have implications for a continuing debate within possible-world semantics, a subject I shall briefly epitomize before relating it to what has already been said. A possible world is often spoken of as a way our world might have been, and that informal description will very nearly serve present purposes.8 Thus, in our world the earth has only a single natural satellite (the moon), but there are other possible worlds, almost the same as ours except that the earth has two or more satellites or has none at all. (The "almost" allows the adjustment of phenomena like the tides which, the laws of nature remaining the same, would vary with number of satellites.) There are also possible worlds less like ours: some in which there is no earth, others in which there are no planets, and still others in which not even the laws of nature are the same. What has recently excited a number of philosophers and linguists about the concept of possible worlds is that it offers a route both to a logic of modal statements and to an intensional semantics for logic and for natural languages. Necessarily true statements, for example, are true in all possible worlds; possibly true statements are true in some; and a true counterfactual is a statement true in some worlds but not in that of the person who made it. Given a set of possible worlds over which to quantify, a formal logic of modal statements appears within reach. Quantification over possible worlds can also lead to an intensional semantics, though by a more complex route. Since the meaning or intension of a statement is what picks out the possible worlds in which that statement is true, each statement corresponds to and may be conceived as a function from possible worlds to truth values. Similarly, a property may be conceived as a function from possible worlds to the sets whose members display that property in each world. Other sorts of referring terms may be conceptually reconstructed in related ways. Even so brief a sketch of possible-world semantics suggests the likely significance of the range of possible worlds over which quantification occurs, and on this issue opinions vary. David Lewis, for example, would quantify over the entire range of worlds that have been or might be conceived; Saul Kripke, at the Some preliminary indications of what these cryptic remarks intend are supplied in my "Commensurability, Comparability, Communicability", op. at. Barbara Partee's essay in this volume provides an elegant summary of the objectives and techniques of possible-world semantics as seen by both linguists and philosophers. Readers unfamiliar with the topic are advised to read it first.

14

Th. S. Kühn

other extreme, restricts attention to possible worlds that can be stipulated; intermediate positions are available ana some have been filled.9 Partisans of these positions debate a variety of issues, most of which have no present relevance. But the debate's participants all appear to assume, with Quine, that anything can be said in any language. If, as I have premised, that assumption fails, additional considerations become relevant. Questions about the semantics of modal statements, or about the intension of words and strings constructed from them, are ipso facto questions about statements and words in a specified language. Only the possible worlds stipulatable in that language can be relevant to them. Extending quantification to include worlds accessible only by resort to other languages seems at best functionless, and in some applications it may be a source of error and confusion. One relevant sort of confusion has already been mentioned - that of the historian who tries to present an older science in his or her own language - and the next two sections will explore some others. At least in their application to historical development, the power and utility of possible-world arguments appears to require their restriction to the worlds accessible with a given lexicon, the worlds that can be stipulated by participants in a given language-community or culture.10 I have so far dealt in general assertions, omitting both illustration and defense. Let me now begin to supply them, acknowledging again that I shall not complete the task in this place. My argument will proceed in two stages. This section examines part of the lexicon of Newtonian mechanics, especially the interrelated terms 'force', 'mass', and 'weight'. It asks, first, what one need and need not know to be a member of the community that uses these terms, and, then, how possession of that knowledge constrains the worlds which members 9

10

Partee provides a fuller account of these divisions as well as a useful bibliography. A more analytic account is included in Robert C. Stalnaker, Inquiry (Cambridge, Mass., 1984). The debate focusses on the ontological status of possible worlds, i. e., on their reality: differences about the range of quantification appropriate for possible-world theories follow directly. Partee emphasizes that possible worlds are not conceivable worlds, points out "that we can conceive of there being possibilities we can't conceive of", and suggests that restricting possible worlds to conceivable worlds may make it impossible to deal with such cases. Talk with her since the symposium has made me realize the need for still another distinction. Not all the worlds accessible or stipulatable with a given lexicon are conceivable: a world containing square circles can be stipulated but not conceived; other examples will recur below. It is only worlds access to which requires restructuring of the lexicon which I mean to exclude when quantifying over possible worlds. Note also that to speak of different lexicons as giving access to different sets of possible worlds is not simply to add one more to the standard kind of accessibility relations discussed at the start of Partee's paper. There is no type of necessity corresponding to lexical accessibility. Excepting statements that stipulate an inconceivable world, no statement framable in a given lexicon is necessarily true or false simply because it can be accessed in that lexicon. More generally, the issue of lexical accessibility seems to arise for all applications of possibleworld arguments, thus cutting across the standard set of accessiblity relations.

Possible Worlds in History of Science

15

of that community can describe without doing violence to the language. Some of the worlds they cannot describe are, of course, described at a later time, but only after a change of lexicon which bars coherent description of some worlds describable before. That sort of change is the subject of this paper's last section. It focusses upon the so-called causal theory of reference, an application of possibleworld concepts said to eliminate the significance of such changes. The vocabulary in which the phenomena of a field like mechanics are described and explained is an historical product, developed over time, and repeatedly transmitted, in its then current state, from one generation to its successor. In the case of Newtonian mechanics, the required cluster of terms has been stable for some time, and transmission techniques are relatively standard. Examining them will suggest characteristics of what the student acquires in the course of becoming a licensed practitioner of the field.11 Before the exposure to Newtonian terminology can usefully begin, other significant portions of the lexicon must be in place. Students must, for example, already have a vocabulary adequate to refer to physical objects and to their locations in space and time. Onto this they must have grafted a mathematical vocabulary rich enough to permit the quantitative description of trajectories and the analyses of velocities and accelerations of bodies moving along them.12 Also, at least implicitly, they must command a notion of extensive magnitude, a quantity whose value for the whole of a body is the sum of its values for the body's parts. Quantity of matter provides a standard example. These terms can all be acquired without resort to Newtonian theory, and the student must control them before that theory can be learned. The other lexical items required by that theory - most notably 'force', 'mass', and 'weight' in their Newtonian senses - can only be acquired together with the theory itself. Five aspects of the way in which these Newtonian terms are learned require particular illustration and emphasis. First, as already indicated, learning cannot begin until a considerable antecedent vocabulary is in place. Second, in the process through which the new terms are acquired, definition plays a negligible role. Rather than being defined, these terms are introduced by exposure to examples of their use, examples provided by someone who already belongs to the speech community in which they are current. That exposure often includes actual exhibits, for example in the student laboratory, of one or more exemplary I discuss the acquisition of a lexicon because it is a source of clues to what the individual's possession of a lexicon entails. Nothing about the end product depends, however, upon the lexicon's being acquired by generation-to-generation transmission. The consequences would be the same if, for example, the lexicon were a genetic endowment or had been implanted by a skilled neurosurgeon. I shall, for example, shortly emphasize that transmitting a lexicon requires repeated recourse to concrete examples. Implanting the same lexicon surgically would, I am suggesting, have involved implanting the memory traces left by exposure to such examples. In practice, the techniques for describing velocities and accelerations along trajectories are usually learned in the same courses that introduce the terms to which I turn next. But the first set can be acquired without the second, whereas the second cannot be acquired without the first.

16

Th. S. Kühn

situations to which the terms in question are applied by someone who already knows how to use them. The exhibits need not be actual, however. The exemplary situations may instead be introduced by a description conducted primarily in terms drawn from the antecedently available vocabulary but in which the terms to be learned also appear here and there. The two processes are for the most part interchangeable, and most students encounter them both, in some mix or other. Both include an indispensable ostensive or stipulative element: terms are taught through the exhibit, direct or by description, of situations to which they apply.13 The learning that results from such a process is not, however, about words alone but equally about the world in which they function. When I use the phrase 'stipulative descriptions' in what follows, the stipulations I have in mind will be simultaneously and inseparably about both the substance and the vocabulary of science, about both the world and the language. A third significant aspect of the learning process is that exposure to a single exemplary situation seldom or never supplies enough information to permit trie student to use a new term. Several examples of varied sorts are required, often accompanied by examples of apparently similar situations to which the term in question does not apply. The terms to be learned, furthermore, are seldom applied to these situations in isolation but are instead embedded in whole sentences or statements, among which are some usually referred to as laws of nature. Fourth, among the statements involved in learning one previously unknown term are some that include other new terms as well, terms mat must be acquired together with the first. The learning process thus interrelates a set of new terms, giving structure to the lexicon that contains them. Finally, though there is usually considerable overlap between the situations to which individual language learners are exposed (and even more between the accompanying statements), individuals can in principle communicate fully even though they acquired the terms with which they do so along very different routes. To the extent that the process I am describing supplies individuals with anything resembling a definition, it is not a definition that need be shared by other members of the speech community. For illustrations, consider first the term 'force'. The situations which exemplify a force's presence are of varied sorts. They include, for example, muscular The terms Ostension' and Ostensive1 seem to have two different uses, which for present purposes need to be distinguished. In one, these terms imply that nothing but the exhibit of a word's referent is needed to learn or to define it. In the other, they imply only that some exhibit is required during the acquisition process. I shall, of course, be using the second sense of die terms. The propriety of extending them to cases in which description in an antecedent vocabulary replaces an actual exhibit depends on recognizing that description does not supply a string of words equivalent to the statements containing the words to be learned. Rather it enables students to visualize the situation and apply to the visualization die same mental processes (whatever they may be) that would otherwise have been applied to the situation as perceived.

Possible Worlds in History of Science

17

exertion, a stretched string or spring, a body possessed of weight (note the occurrence of another of the terms to be learned), or, finally, certain sorts of motion. The last is particularly important and presents particular difficulties to the student. As Newtonians use 'force', not all motions signify the presence of its referent, and examples which display the distinction between forced and forcefree motions are therefore required. Their assimilation, furthermore, demands the suppression of a highly developed pre-Newtonian intuition. For children and Aristotelians the standard example of a forced motion is the hurled projectile. Force-free motion is for them exemplified by the falling stone, the spinning top, or the rotating flywheel. For the Newtonian all of these are cases of forced motion. The only example of a Newtonian force-free motion is motion in a straight line at constant speed, and that can be exhibited directly only in interplanetary space. Teachers nevertheless try. (I still remember the contrived lecture demonstration - a block of ice sliding on a sheet of glass - that helped me undo prior intuitions and acquire the Newtonian concept of 'force'.) But for most students the main path to this key aspect of the use of the term is provided by the string of words known as Newton's First Law of motion: 'in the absence of an external force applied to it, a body moves continuously at constant speed in a straight line'. It exhibits, by description, the motions which require no force.14 More will need to be said about 'force', but let me first look briefly at its two Newtonian companions, 'weight' and 'mass'. The first refers to a particular sort of force, the one which causes a physical body to press on its supports while at rest or to fall when unsupported. In this still-qualitative form the term 'weight' is available prior to Newtonian 'force' and is used during the latter's acquisition. 'Mass' is usually introduced as equivalent to 'quantity of matter', where matter is the substrate underlying physical bodies, the stuff of which quantity is conserved as the qualities of material bodies change. Any feature which, like weight, picks out a physical body, is an index also of the presence of matter and of mass. As in the case of 'weight' and unlike the case of 'force', the qualitative features by which one picks out the referents of 'mass' are identical with those of preNewtonian usage. But the Newtonian use of all three terms is quantitative, and the Newtonian form of quantification alters both their individual uses and the interrelationships between them.15 Only the unit measures may be established by convention; the Newton's First Law is a logical consequence of his Second, and Newton's reason for stating them separately has long been a puzzle. The answer may well lie in pedagogic strategy. If Newton had permitted the Second Law to subsume the First, his readers would have had to sort out his use of 'force' and of 'mass' together, an intrinsically difficult task further complicated by the fact that the terms had previously been different not only in their individual use but in their interrelation. Separating them to the extent possible displayed the nature of the required changes more clearly. Though my analysis diverges from theirs, many of the considerations that follow (as well as a few of those introduced above) were suggested by contemplation of the techniques developed by J. D. Sneed and Wolfgang Stegmüller for formalizing physical theories, especially by their

18

Th. S. Kühn

scales must be chosen so that weight and mass are extensive quantities and so that forces can be added vectorially. (Contrast the case of temperature in which both unit and scale can be chosen by convention.) Once again, the learning process requires the juxtaposition of statements involving the terms to be learned with situations drawn directly or indirectly from nature. Begin with the quantification of 'force'. Students acquire the full quantitative concept by learning to measure forces with a spring balance or some other elastic device. Such instruments had appeared nowhere in scientific theory or practice before Newton's time, when they took over the conceptual role previously played by the pan balance. But they have since been central, for reasons that are conceptual rather than pragmatic. The use of a spring balance to exhibit the proper measure offeree requires, however, recourse to two statements ordinarily described as laws of nature. One of these is Newton's Third Law, which states, for example, that the force exerted by a weight on a spring is equal and opposite to the force exerted by the spring on the weight. The other is Hooke's Law, which states that the force exerted by a stretched spring is proportional to the spring's displacement. Like Newton's First Law, these are first encountered during language learning where they are juxtaposed with examples of situations to which they apply. Such juxtapositions play a double role, simultaneously stipulating how the word 'force' is to be used and how the world populated by forces behaves. Turn now to the quantification of the terms 'mass' and 'weight'. It illustrates with special clarity a key aspect of the lexical acquisition process, one that has not yet been considered. To this point, my discussion of Newtonian terminology has probably suggested that, once the required antecedent vocabulary is in place, students learn the terms that remain by exposure to some single specifiable set of examples of their use. Those particular examples may well have seemed to provide necessary conditions for the acquisition of those terms. In practice, however, cases of that sort are very rare. Usually there are alternate sets of examples that will serve for the acquisition of the same term or terms. And, though it usually makes no difference to which set of these examples an individual has, in fact, been exposed, there are special circumstances in which the differences between sets prove very important. In the case of 'mass' and 'weight', one of these alternate sets is standard. It is able to supply the missing elements of both vocabulary and theory together, and it probably therefore enters the lexical acquisition process for all students. But logically other examples would have done as well, and for most students some of them also play a role. Begin with the standard route, which first quantifies 'mass' in the guise of what today is called 'inertial mass'. Students are presented manner of introducing theoretical terms. Note also that these remarks suggest a route to the solution of a central problem of their approach, how to distinguish the core of a theory from its expansions. For this problem see my paper, "Theory Change as Structure Change: Comments on the Sneed Formalism," Erkenntnis, 10 (1976), 179-199.

Possible Worlds in History of Science

19

with Newton's Second Law - force equals mass times acceleration - as a description of the way moving bodies actually behave, but the description makes essential use of the still incompletely established term 'mass'. That term and the Second Law are thus acquired together, and the Law can thereafter be used to supply the missing measure: the mass of a body is proportional to its acceleration under the influence of a known force. For purposes of concept acquisition, centripetal-force apparatus provides a particularly effective way to make the measurement. Once mass and the Second Law have been added to the Newtonian lexicon in this way, the law of gravity can be introduced as an empirical regularity. Newtonian theory is applied to observation of the heavens and the attractions manifest there are compared to those between the earth and bodies resting on it. The mutual attraction between bodies is thus shown to be proportional to the product of their masses, an empirical regularity that can be used to introduce the still missing aspects of the Newtonian term 'weight'. 'Weight' is now seen to denote a relational property, one that depends on the presence of two or more bodies. It can therefore, unlike mass, differ from one location to another, at the surface of the earth and of the moon, for example. That difference is captured only by the spring balance, not by the previously standard pan balance which yields the same reading at all locations. What the pan balance measures is mass, a quantity that depends only on the body and on the choice of a unit measure. Because it establishes both the Second Law and the use of 'mass', the sequence just sketched provides the most direct route to many applications of Newtonian theory.16 That is why it plays so central a role in introducing the theory's vocabulary. But it is not, as previously indicated, required for that purpose, and, in any case, it rarely functions alone. Let me now consider a second route along which the use of 'mass' and 'weight' can be established. It starts from the same point as the first, by quantifying the notion of force with the aid of a spring balance. Next, 'mass' is introduced in the guise of what is today labelled 'gravitational mass'. A stipulative description of the way the world is provides students with the notion of gravity as a universal force of attraction between pairs of material bodies, its magnitude proportional to the mass of each. With the missing aspects of 'mass' thus supplied, weight can be explained as a relational property, the force resulting from gravitational attraction. That is a second way to establish the use of the Newtonian terms 'mass' and 'weight'. With them in hand Newton's Second Law, the still missing component of Newtonian theory, can be introduced as empirical, a consequence simply of observation. For that purpose, centripetal force apparatus is again appropriate, but no longer to measure mass, as it did on the first route, but now rather to determine the relation between applied force and the acceleration of a mass

All applications of Newtonian theory depend on understanding 'mass', but for many of them 'weight' is dispensable.

20

Th. S. Kühn

previously measured by gravitational means. The two routes thus differ in what must be stipulated about nature in order to learn Newtonian terms, what can be left insteacf for empirical discovery. On the first route the Second Law enters stipulatively, the law of gravitation empirically. On the second, their epistemic status is reversed. In each case one, but only one, of the laws is, so to speak, built into the lexicon. I do not quite want to call such laws analytic, for experience with nature was essential to their initial formulation. Yet they do have something of the necessity that the label 'analytic' implies. Perhaps 'synthetic a priori' comes closer. There are, of course, still other ways in which the quantitative elements of 'mass' and 'weight' can be acquired. For example, Hooke's Law having been introduced together with 'force', the spring balance can be stipulated as the measure of weight, and mass can be measured, again by stipulation, in terms of the vibration period of a weight at the end of a spring. In practice, several of these applications of Newtonian theory usually enter into the process of acquiring Newtonian language, information about the lexicon and information about the world being distributed in an indivisible mix among them. Under those circumstances, one or another of the examples introduced during lexical acquisition can, when occasion requires, be adjusted or replaced in the light of new observations. Other examples will maintain the lexicon stable, keeping in place a set of quasi-necessities equivalent to those initially induced by language learning. Clearly, however, only a certain number of examples may be altered piecemeal in this way. If too many require adjustment, then it is no longer individual laws or generalizations that are at stake but the very vocabulary in which they are stated. A threat to that vocabulary is, however, a threat also to the theory or laws essential to its acquisition and use. Could Newtonian mechanics withstand revision of the Second Law, of the Third Law, of Hooke's Law, or the law of gravity? Could it withstand the revision of any two of these, of three, or of all four? These are not questions that individually have yes or no answers. Rather, like Wittgenstein's "Could one play chess without the queen?", they suggest the strains placed on a lexicon by questions that its designer, whether God or cognitive evolution, did not anticipate its being required to answer.17 What should one have said when confronted by an egg-laying creature that suckles its young? Is it a mammal or is it not? These are the circumstances in which, as Austin put it, "we don't know what to say. Words literally fail us."18 Such circumstances, if they endure for long, call forth a locally different lexicon, one Twenty-five years ago the quotation was a standard part of what I now discover was a merely oral tradition. Though clearly "Wittgensteinian", it is not to be found in any of Wittgenstein's published writings. I preserve it here, because of its recurrent role in my own philosophical development and because I've found no published substitute that so clearly bars responding that additional information might permit the question to be answered. J. L. Austin, Other Minds", in Collected Philosophical Papers (Oxford, 1961), pp. 44-84. The quoted passage occurs on p. 56, and the italics are Austin's. For examples from literature of situations in which words fail us, see James Boyd White, When Words Lose their Meaning

Possible Worlds in History of Science

21

that permits an answer but to a slightly altered question: "Yes, the creature is a mammal" (but to be a mammal is not what it was before). The new lexicon opens new possibilities, ones that could not have been stipulated by the use of the old. To clarify what I have in mind, let me suppose that there are only two ways in which use of the terms 'mass' and 'weight' can be acquired: one which stipulates the Second Law and finds the law of gravity empirically; another which stipulates the law of gravity and discovers the Second Law empirically. Suppose further that the two routes are exclusive; students traverse one or the other so that on each the necessities of the lexicon and the contingencies of experiment are kept separate. Clearly, these two routes are very different, but the differences will not ordinarily interfere with full communication among those who use the terms. All will pick out the same objects and situations as the referents of the terms they share, and all will agree about the laws and other generalizations governing these objects and situations. All are thus fully participants in a single speech community. What individual speakers may differ about is the epistemic status of generalizations that community members share, and such differences are not usually important. Indeed, in ordinary scientific discourse, they do not emerge at all. While the world behaves in anticipated ways - the ones for which the lexicon evolved - these differences between individual speakers make little or no difference. But change of circumstance may make them consequential. Imagine that a discrepancy is discovered between Newtonian theory and observation, for example celestial observations of the motion of the lunar perigee. Scientists who had learned Newtonian 'mass' and 'weight' along the first of my two lexical-acquisition routes would be free to consider altering the law of gravity as a way to remove the anomaly. On the other hand, they would be bound by language to preserve the Second Law. But scientists who had acquired 'mass' and 'weight' along my second route would be free to suggest altering the Second Law but would be bound by language to preserve the law of gravity. A difference in the language learning route, one which had had no effect while the world behaved as anticipated, would lead to differences of opinion when anomalies were found. Now suppose that neither the revisions that preserved the Second Law nor those that preserved the law of gravity proved effective in eliminating anomaly. The next step would be an attempt at revisions which altered both laws together, and those revisions the lexicon will not, in its present form, permit.19 Such

19

(Chicago, 1984). I have compared an example from the sciences with one from developmental psychology in "A Function for Thought Experiments", reprinted in The Essential Tension (Chicago, 1977), pp. 240-265. At this point I will seem to be reintroducing the previously banished notion of analyticity, and perhaps I am. Using the Newtonian lexicon, the statement "Newton's Second Law and the law of gravity are both false" is itself false. Furthermore, it is false by virtue of the meaning of the Newtonian terms 'force' and 'mass'. But it is not - unlike the statement "Some bachelors are

22

Th. S. Kühn

attempts are often successful nonetheless, but they require recourse to such devices as metaphorical extension, devices that alter die meanings of lexical items themselves. After such revision, say the transition to an Einsteinian vocabulary, one can write down strings of symbols that look like revised versions of the Second Law and the law of gravity. But the resemblance is deceptive because some symbols in the new strings attach to nature differendy from the corresponding symbols in the old, thus distinguishing between situations which, in the antecedently available vocabulary, were the same.20 They are the symbols for terms whose acquisition involved laws that have changed form with the change of theory: the differences between the old laws and the new are reflected by the terms acquired with them. Each of the resulting lexicons then gives access to its own set of possible worlds, and the two sets are disjoint. Translations involving terms introduced with the altered laws are impossible. The impossibility of translation does not, of course, bar users of one lexicon from learning the other. And having done so, they can join the two together, enriching their initial lexicon by adding to it sets of terms from the one they just have acquired. For some purposes such enrichment is essential. At the beginning of this paper, for example, I suggested that historians often required an enriched lexicon to understand the past, and I have argued elsewhere that they must transmit that lexicon to their readers.21 But the sense of enrichment involved is peculiar. Each of the lexicons combined for the historian's purposes embodies knowledge of nature, and the two sorts of knowledge are incompatible, cannot coherently describe the same world. Except under very special circumstances, like those of the historian at work, the price of combining them is incoherence in the description of phenomena to which either one might alone have been applied.22 Even the historian avoids incoherence only by being sure at all times which lexicon he is using and why. Under these circumstances, one may reasonably ask whether the term 'enriched' quite applies to the enlarged lexicon formed by combinations of this sort. A closely related problem - that of the grue emeralds - has recently been much discussed in philosophy. An object is grue if it has been observed to be green before time t or if, alternatively, it is blue. The puzzle is that the same set of observations, if made prior to f, support two incompatible generalizations: "All emeralds are green" and "All emeralds are grue". (Note that a grue emerald, if

20

21 22

married" - false by virtue of the definitions of those terms. The meaning of 'force' and 'mass' is not embodiable in definitions but rather in their relation to the world. The necessity to which I here appeal is not so much analytic as synthetic a priori. In fact, for the Newton to Einstein transition, the most significant lexical change is in the antecedent kinematic vocabulary for space and time, and it moves from there upward into the vocabulary of mechanics. See the reference cited in note 2. In describing the expanded lexicon it is essential to use terms like 'incompatible' and 'incoherent' radier that 'contradictory' and 'false'. The two latter terms would apply only if translation were possible.

Possible Worlds in History of Science

23

not examined before time t, can only be blue.) Here, too, the solution depends upon segregating the lexicon containing the normal descriptive color vocabulary, 'blue', 'green', and the like from the lexicon that contains 'grue', 'bleen', and the names of other occupants of the corresponding spectrum. One set of terms is projectible, supports induction, the other not. One set of terms is available for descriptions of the world, the other is reserved for the special purposes of the philosopher. Difficulties emerge only when the two, embodying incompatible bodies of knowledge of nature, are used in combination, for there is no world to which the enlarged lexicon could apply.23 Students of literature have long taken for granted that metaphor and its companion devices (those which alter the interrelations among words) provide entree to new worlds and make translation impossible by doing so. Similar characteristics have been widely attributed to the language of political life and, by some, to the entire range of the human sciences. But the natural sciences, dealing objectively with the real world (as they do), are generally held to be immune. Their truths (and falsities) are thought to transcend the ravages of temporal, cultural, and linguistic change. I am suggesting, of course, that they cannot do so. Neither the descriptive nor the theoretical language of a natural science provides the bedrock such transcendence would require. I shall not in this place even attempt to deal with the philosophical problems consequent upon that point of view. Let me, instead, attempt to increase their urgency. The threat to realism is the foremost of the problems I have in mind, and it can here stand for the entire set.24 A lexicon acquired by techniques like those discussed in the preceding section gives members of the community that employs it conceptual access to an infinite set of lexically-stipulatable worlds, worlds describable with the community's lexicon. Of these worlds, only a small fraction are compatible with what they know of their own, the actual world: the others are barred by requirements of internal consistency or of conformity with experiment and observation. As time passes, continuing research excludes more and more possible worlds from the subset that could be actual. If all scientific 23

24

For the original paradox see Nelson Goodman, Fact, Fiction, ana Forecast (2nd ed., New York, 1977), Chapt. 3 & 4. Note that the similarity just emphasized is in one very important respect incomplete. Both the Newtonian terms discussed above and the terms in any color vocabulary form an interrelated set. But in the latter case the difference between vocabularies does not affect vocabulary-structure, and it is therefore possible to translate between the projectible 'blue'/'green' vocabulary and die unprojectible vocabulary containing 'bleen' and 'grue'. Contrary to a widespread impression, the sort of position here sketched does not raise problems of relativism, at least not if 'relativism' is used in any standard sense. There are shared and justifiable, diough not necessarily permanent, standards that scientific communities use when choosing between theories. On this subject see my papers, "Objectivity, Value Judgement, and Theory Choice" in The Essential Tension, op. at., pp. 320-339, and "Rationality and Theory Choice," The Journal of Philosophy, 80, (1983), pp. 563-570.

24

Th. S. Kühn

development proceeded in this way, the progress of science would consist in ever closer specification of a single world, the actual or real one. A reiterated theme of this paper has been, however, that a lexicon which gives access to one set of possible worlds also bars access to others. (Remember the Newtonian lexicon's inability to describe a world in which the Second Law and the law of gravity were not simultaneously satisfied.) And scientific development turns out to depend not only on weeding out candidates for reality from the current set of possible worlds but also upon occasional transitions to another set, one made accessible by a lexicon with different structure. Once such a transition has occurred, some statements previously descriptive of possible worlds prove untranslatable in the terminology developed for the subsequent science. These are the statements the historian first encounters as anomalous word strings; one cannot imagine what those who uttered or wrote them were trying to say. Only when a new lexicon has been mastered can they be understood, and that understanding does not provide them with later equivalents. Individually, they are neither compatible nor incompatible with statements embodying the beliefs of a later age, and they are therefore immune to an evaluation conducted with its conceptual categories. The immunity of such statements is, of course, only to being judged one at a time, labeled individually with truth values or some other index of epistemic status. Another sort of judgement is possible and in scientific development something very like it repeatedly occurs. Faced with untranslatable statements, the historian becomes bilingual, first learning the lexicon required to frame the problematic statements and then, if it seems relevant, comparing the whole older system (a lexicon plus the science developed with it) to the system in current use. Most of the terms used within either system are shared by both, and most of these shared terms occupy the same positions in both lexicons. Comparisons made using those terms alone ordinarily provide a sufficient basis for judgement. But what is then being judged is the relative success of two whole systems in pursuing an almost stable set of scientific goals, a very different matter from the evaluation of individual statements within a given system. Evaluation of a statement's truth values is, in short, an activity that can be conducted only with a lexicon already in place, and its outcome depends upon that lexicon. If, as standard forms of realism suppose, a statement's being true or false depends simply on whether or not it corresponds to the real world independent of time, language, and culture - then the world itself must be somehow lexicon-dependent. Whatever form that dependence takes, it poses problems for a realist perspective, problems that I take to be both genuine and urgent. Rather than explore them further here - a task for another paper -1 shall close by examining a standard attempt to dismiss them. What I have been describing as the problem of lexical dependence is often called the problem of meaning variance. To avoid it and related problems from other sources, many philosophers have in recent years emphasized that truth values depend only on reference, and that an adequate theory of reference need

Possible Worlds in History of Science

25

not call upon the way in which the referents of individual terms are in fact picked out.25 The most influential version of such theories is the so-called causal theory of reference developed primarily by Kripke and Putnam. It is rooted firmly in possible-world semantics, and its expositors resort repeatedly to examples drawn from scientific development. A look at it should both reinforce and extend the viewpoint sketched above. For the purpose, I restrict myself primarily to the version developed by Hilary Putnam, for Putnam deals more explicitly than others with problems of scientific development.26 According to causal theory the referents of natural kind terms like 'gold', 'tiger', 'electricity', 'gene', or 'force' are determined by some original act of baptizing or dubbing samples of the kind in question with the name they will thereafter bear. That act, to which later speakers are linked by history, is the "cause" of the term's referring as it does. Thus, some samples of a naturally occurring yellow, malleable metal were once baptized 'gold' (or some equivalent in another language), and the term has since referred to all samples of the same stuff as the original whether they displayed the same superficial qualities or not. What establishes the reference of a term is, thus, the original sample together with the primitive relationship, sameness-of-kind. If the original samples were not all or mostly of the same kind, then the term in question, for example 'phlogiston', fails to refer. Theories about what makes the samples the same are, on this view, irrelevant to reference, as are the techniques used in identifying further samples. Both may vary over time as well as from individual to individual at a given time. But the original samples and the relation sameness-of-kind are stable. If meanings are the sorts of things that individuals can carry around in their heads, then meaning does not determine reference. Excluding proper names, I doubt that there is any set of terms for which this theory works precisely, but it comes very close to doing so for terms like 'gold', 25

26

Views which, like mine, depend on talking about the way words are actually used, the situations in which they apply, are regularly charged with invoking a "verification theory of meaning", not currently a respectable thing to do. But in my case at least that charge does not hold. Verification theories attribute meanings to individual sentences and through them to the individual terms those sentences contain. Each term has a meaning determined by the way in which sentences containing it are verified. I have been suggesting, however, that with occasional exceptions terms do not individually have meanings at all. More important, the view sketched above insists that people may use the same lexicon, refer to the same items with it, and yet pick out those items in different ways. Reference is a function of the shared structure of the lexicon but not of the varied feature spaces within which individuals represent that structure. There is, however, a second charge, closely related to verificationism, of which I am guilty. Those who maintain the independence of reference and meaning also maintain that metaphysics is independent of epistemology. No view like mine (in the respects presently at issue there are a number) is compatible with that separation. The separation of metaphysics from epistemology can come only after a position that involves both has been elaborated. Op. at. Putnam has, I believe, now abandoned significant components of the theory, moving from it to a view ("internal realism") with significant parallels to my own. But few philosophers have followed him. The views discussed below are still very much alive.

26

Th. S. Kühn

and the plausibility of the application of causal theory to natural kind terms depends on the existence of such cases. Terms that behave like 'gold* ordinarily refer to naturally occurring, widely distributed, functionally significant, and easily recognized substances. They occur in the languages of most or all cultures, retain their original use over time, and refer throughout to the same sorts of samples. There is little problem about translating them, for they occupy closely equivalent positions in all lexicons. 'Gold' is among the closest approximations we have to an item in a neutral, mind-independent observation vocabulary. When a term is of this sort, modern science can often be used not only to specify the common essence of its referents but actually to single them out. Modern theory, for example, identifies gold as the substance with atomic number 79 and licenses specialists to identify it by the application of such techniques as x-ray spectroscopy. Neither the theory nor the instrument was available seventy-five years ago, but it is nevertheless reasonable to suggest that: "The referents of 'gold' are and have always been the same as the referents of 'substance with atomic number 79'." Exceptions to that equation are few, and they result primarily from our increased ability to detect impurities and forgeries. For the causal theorist, therefore, 'having atomic number 79', is the essential property of gold - the single property such that, if gold in fact does have it, then it has it necessarily. Other properties - yellowness and ductility, for example - are superficial and correspondingly contingent. Kripke suggests that gold might even be blue, its apparent yellowness resulting from an optical illusion.27 Though individuals may, in fact, use color and other superficial characteristics when picking out samples of gold, that practice tells nothing essential about the referents of the term. 'Gold' presents a relatively special case, however, and what is special about it obscures essential limitations on the conclusions it will support. More representative is Putnam's most developed example, 'water', and the problems which arise with it are still more severe in the case of such other widely discussed terms as 'heat' and 'electricity'.28 For water the discussion divides in two parts. In the first, which is the more familiar, Putnam imagines a possible world containing Twin Earth, a planet just like our own except that the stuff called 'water' by Twin Earthians is not H2O but a different liquid with a very long and complicated chemical formula abbreviated XYZ. " [Indistinguishable from water at normal temperatures and pressures", XYZ is the stuff that on Twin Earth quenches thirst, rains from the skies, and fills oceans and lakes, much as water does here. If a spaceship from Earth ever visits Twin Earth, Putnam writes:

27 28

Of), at., p. 118. The force of Putnam's discussion depends in pan upon an equivocation that needs to be eliminated. As used in everyday life or by the laity, 'water' has through history behaved much like 'gold'. But that is not the case within the community of scientists and philosophers to which Putnam's argument needs to be applied.

Possible Worlds in History of Science

27

then, the supposition at first will be that 'water' has the same meaning on Earth and Twin Earth. This supposition will be corrected when it is discovered that 'water' on Twin Earth is XYZ, and the Earthian spaceship will report somewhat as follows: "On Twin Earth the word 'water' means XYZ."

As in the case of gold, superficial qualities like quenching thirst or raining from the skies have no role in determining to what substance the term Vater' properly refers. Two aspects of Putnam's fable require special notice. First, the fact that Twin Earthians call XYZ by the name 'water' (the same symbol that Earthians use for the stuff that lies in lakes, quenches thirst, etc.) is an irrelevancy. The difficulties presented by this story will emerge more clearly if the visitors from Earth use their own language throughout. Second, and presendy central, whatever the visitors call the stuff that lies in Twin Earthian lakes, the report they send home must take some form like: "Back to the drawing board! Something is badly wrong with chemical theory." The terms 'XYZ' and 'H^O' are drawn from modern chemical theory, and that theory is incompatible with the existence of a substance with properties very nearly the same as water but described by an elaborate chemical formula. Such a substance would, among other things, be too heavy to evaporate at normal terrestrial temperatures. Its discovery would present the same problems as the simultaneous violation of Newton's Second Law and the law of gravity described in the last section. It would, that is, demonstrate the presence of fundamental errors in the chemical theory which gives meanings to compound names like Ή2Ο' and the unabbreviated form of 'XYZ'. Within the lexicon of modern chemistry, a world containing both our earth and Putnam's Twin Earth is lexically possible, but the composite statement that describes it is necessarily false. Only with a differently structured lexicon, one shaped to describe a very different sort of world, could one, without contradiction, describe the behavior of XYZ at all, and in that lexicon *H2O' might no longer refer to what we call 'water'. So much for the first part of Putnam's argument. In the second he applies it more concretely to the referential history of 'water', suggesting that "we roll the time back to 1750" and continuing: At that time chemistry was not developed on either Earth or Twin Earth. The typical Earthian speaker of English did not know water consisted of hydrogen and oxygen, and the typical Twin Earthian speaker of English did not know 'water' consisted of XYZ.... Yet the extension of the term 'water' was just as much H2O on Earth in 1750 as in 1950; and the extension of the term 'water' was just as much XYZ on Twin Earth in 1750 as in 1950.

In journeys through time as in those through space, Putnam suggests, it is chemical formula, not superficial qualities, that determines whether a given substance is water.

28

Th. S. K hn

For present purposes attention can be restricted to Earthian history, and on earth Putnam's argument for 'water' is the same as it was for 'gold'. The extension of 'water' is determined by the original sample together with the relation sameness-of-kind. That sample dates from before 1750 and the nature of its members has been stable. So has the relation sameness-of-kind, though explanations of what it is for two bodies to be of the same kind have varied widely. What matters, however, is not explanations but what gets picked out, and identifying samples of H2O is, according to causal theory, the best means yet found to pick out samples of the same kind as the original set. Give or take a few discrepancies at the margins, discrepancies due to refinement of technique or perhaps to change of interest, 'H2O' refers to the same samples that 'water' referred to in either 1750 or 1950. Apparently causal theory has rendered the referents of 'water' immune to changes in the concept of water, the theory of water, and the way samples of water are picked out. The parallel between causal theory's treatment of 'gold' and of 'water' seems complete. But in the case of water difficulties arise. Ή2Ο' picks out samples not only of water but also of ice and steam. H2O can exist in all three states of aggregation solid, liquid, and gaseous - and it is therefore not the same as water, at least not as picked out by the term 'water' in 1750. The difference in items referred to is, furthermore, by no means marginal, like that due to impurities for example. Whole categories of substance are involved, and their involvement is by no means accidental. In 1750 the primary differences between chemical species were the states of aggregation or modelled upon them. Water, in particular, was an elementary body of which liquidity was an essential property. For some chemists the term 'water' referrea to the generic liquid, and it had done so for many more only a few generations before. Not until the 1780's, in an episode long known as "The Chemical Revolution", was the taxonomy of chemistry transformed so that a chemical species might exist in all three states of aggregation. Thereafter, the distinction between solids, liquids, and gases became physical, not chemical. The discovery that liquid water was a compound of two gaseous substances, hydrogen and oxygen, was an integral part of that larger transformation and could not have been made without it. This is not to suggest that modern science is incapable of picking out the stuff that people in 1750 (and most people still) label 'water'. That term refers to liquid H2O. It should be describee! not simply as H2O but as close-packed H2O particles in rapid relative motion. Marginal differences again aside, samples answering that compound description are the ones picked out in 1750 and before by the term 'water'. But this modern description leads to a new network of difficulties, difficulties that may ultimately threaten the concept of natural kinds and that meanwhile must bar the automatic application of causal theory to them. Causal theory was initially developed with notable success for application to proper names. Its transfer from them to natural kind terms was facilitated perhaps made possible - by the fact that natural kinds, like single individual creatures, are denoted by short and apparently arbitrary names, names coexten-

Possible Worlds in History of Science

29

sive with those of the corresponding kind's single essential property. Our examples have been 'gold' paired with 'having atomic number 79' and 'water' paired with 'being H2O'. The latter member of each pair names a property, of course, as the name coupled with it does not. But so long as only a single essential property is required by each natural kind that difference is inconsequential. When two non-coextensive names are required, however, - 'H2O' and 'liquidity' in the case of water - then each name, if used alone, picks out a larger class than the pair does when conjoined, and the fact that they name properties becomes central. For if two properties are required, why not three or four? Are we not back to the standard set of problems that causal theory was intended to resolve: which properties are essential, which accidental; which properties belong to a kind by definition, which are only contingent? Has the transition to a developed scientific vocabulary really helped at all? I think it has not. The lexicon required to label attributes like being-H2O or being-close-packed-particles-in-rapid-relative-motion is rich and systematic. No one can use any of the terms that it contains without being able to use a great many. And given that vocabulary, the problems of choosing essential properties arise again, except that the properties involved can no longer be dismissed as superficial. Is deuterium hydrogen, for example, and is heavy water really water? And what may one say about a sample of close-packed particles of H2O in rapid relative motion at the critical point, under the conditions of temperature and pressure, that is, at which the liquid, solid, and gaseous states are indistinguishable? Is it really water? The use of theoretical rather than superficial properties offers great advantages, of course. There are fewer of the former; the relations between them are more systematic; and they permit both richer and more precise discriminations. But they come no closer to being essential or necessary properties than the superficial ones they appear to supplant. The problems of meaning and meaning variance are still in place. The inverse argument proves even more significant. The so-called superficial properties are no less necessary than their apparently essential successors. To say that water is liquid H2O is to locate it within an elaborate lexical and theoretical system. Given that system, as one must be in order to use the label, one can in principle predict the superficial properties of water (just as one could those of XY2), compute its boiling and freezing points, the optical wavelengths which it will transmit, and so on.29 If water is liquid H2O, then these properties are necessary to it. If they were not realized in practice, that would be a reason to doubt that water really was H2O.

29

Laypeople can, of course, say that water is H2O without controlling the fuller lexicon or the theory which it supports. But their ability to communicate by doing so depends upon the presence of experts in their society. The laity must be able to identify the experts and say something of the nature of the relevant expertise. And the experts must, in turn, command the lexicon, the theory, and the computations.

30

Th. S. Kühn

This last argument applies also to the case of gold, in which causal theory apparently succeeded. 'Atomic number' is a term from the lexicon of atomicmolecular theory. Like 'force' and 'mass', it must be learned together with other terms deployed in that theory, and the theory itself must play a role in the acquisition process. When the process is complete, one can replace the label 'gold' with 'atomic number 79', but one can then also replace the label 'hydrogen' with 'atomic number 1', Oxygen' with 'atomic number 8', and so on to a total well over a hundred. And one can do something more important as well. Invoking such other theoretical properties as electronic charge and mass, one can in principle, and to a considerable extent in fact, predict the superficial qualities - density, color, ductility, conductivity, and so on - that samples of the corresponding substance will possess at normal temperatures. Those properties are no more accidental than having-atomic-number-79. That color is a superficial property does not make it a contingent one. Furthermore, in a comparison of superficial and theoretical qualities, the former have a double priority. If the theory that posits the relevant theoretical properties could not predict these superficial qualities or some of them, there would be no reason to take it seriously. If gold were blue for a normal observer under normal conditions of illumination, its atomic number would not be 79. In addition, superficial properties are the ones called upon in those difficult cases of discrimination characteristically raised by new theories. Is deuterium really hydrogen, for example? Are viruses alive?30 What remains special about 'gold' is simply that, unlike 'water', only one of the underlying properties recognized by modern science - having atomic number 79 - need be called upon to pick out members of the sample to which the term has continued through history to refer.31 'Gold' is not the only term that possesses or closely approximates this characteristic. So do many of the 30

31

At issue, of course, is where to draw the boundary lines that delimit the referents of 'water', 'living thing', and so on, a problem which arises from and seems to threaten the notion of natural kinds. That notion is closely modelled on the concept of a biological species, and discussions of causal theory repeatedly invoke the relation between a particular gene-type and a corresponding species (often tigers) to illustrate the relation said to hold between a natural kind and its essence, between H2O and water, for example, or between atomic-number-79 and gold. But even individuals who are unproblematically members of the same species have differently constituted sets of genes. Which sets are compatible with membership in that species is a subject of continuing debate, both in principle and practice, and the subject of the argument is always which superficial properties (e. g., the ability to interbreed) the members of the species must share. Even for gold this generalization is not altogether correct. As mentioned above, scientific progress does result in marginal adjustments of the original samples of gold by virtue of "our increased ability to detect impurities". But what it is for gold to be pure is determined in pan by theory. If gold is the substance with atomic number 79, then even a single atom with a different atomic number constitutes an impurity. But if gold is, as it was in antiquity, a metal that ripens naturally in the earth, changing gradually from lead through iron and silver to gold in the process, then there is no single form of matter that is gold tout court. When the ancients applied the term 'gold' to samples from which we might withhold it, they were not always simply mistaken.

Possible Worlds in History of Science

31

basic-level referring terms used in everyday speech, including the everyday use of the term 'water'. But not all everyday terms are of this sort. 'Planet' and 'star' now categorize the world of celestial objects differently from the way they did before Copernicus, and the differences are not well-described by phrases like "marginal adjustment" or "zeroing in". Similar transitions have characterized the historical development of virtually all the referring terms of the sciences, including the most elementary: 'force', 'species', 'heat', 'element', 'temperature', and so on. Over the course of history, these and other scientific terms have participated, sometimes repeatedly, in the sorts of changes epitomized incompletely above by the change in the chemist's use of 'water' between 1750 and 1950. Such lexical transformations systematically split up and then regroup in new ways the members of sets to which terms in the lexicon refer. Usually the terms themselves remain the same through such transitions, though sometimes with strategic additions and deletions. So do many of the items referred to by those terms, which is why the terms endure. But the changes in membership of the sets of items to which these enduring terms refer are often massive, and they affect not just the referents of an individual term but of an interrelated set of terms between which the preexisting population is redistributed. Items previously regarded as quite unlike are grouped together after the transformation, while previously exemplary members of some single category are later divided between systematically different ones. It is lexical change of this sort that results in the apparent textual anomalies with which this paper began. Encountered by an historian in a text of the past, they strenuously resist elimination by any translation or paraphrase that uses the historian's own lexicon, the one he or she initially brought to the text. The phenomena described in those anomalous passages are stipulated neither as present nor as absent in any of the possible worlds to which that lexicon gives access, and the historian cannot therefore quite understand what the author of the text can be trying to say. Those phenomena belong to another set of possible worlds, one in which many of the same phenomena occur that occur in the historian's own, but in which things also occur that the historian, until reeducated, cannot imagine. Under such circumstances the only recourse is reeducation: the recovery of the older lexicon, its assimilation, and the exploration of the set of worlds to which it gives access. Causal theory provides no bridge across the divide, for the transworld voyages it envisages are limited to worlds in a single lexically possible set. And in the absence of the bridge that causal theory has sought to provide, there is no basis for talk of science's gradual elimination of all worlds excepting the single real one. That way of talking, neatly illustrated by the discussion of gold but not by that of water, has provided causal theory's version of what the tradition described as successively closer approximations to the truth, cutting the world closer to its joints, or just zeroing in. Such descriptions of scientific development can no longer be sustained. I know of only one other strategy available for their defense, and it seems to me

32

Th. S. Kühn

self-defeating, an artifice born of desperation. In the case of 'water' that strategy would be implemented as follows: until sometime after 1750 chemists were misled by superficial properties into believing that water is a natural kind, but it is not; what they called 'water' did not exist, any more than did phlogiston; both were chimerical, and the terms used to refer to them did not, in fact, refer at all.32 But that cannot be right. Putatively non-referring terms like 'water' can neither be isolated nor replaced by more primitive terms of indubitable referential status. If 'water' failed to refer, then so did other chemical terms like 'element', 'principle', 'earth', 'compound', and many others. Nor was referential failure restricted to chemistry. Terms like 'heat', 'motion', 'weight', and 'force' were equally empty; statements in which they appeared were about nothing. On this showing, the history of science is the history of developing vacuity, and from vacuity one cannot zero in. Some other explanation of the achievements of science is needed.

I take this to be the sort of response Putnam would have provided when the paper I have been discussing was written.

ARTHUR I. MILLER

Discussion of Thomas S. Kühn's paper "Possible Worlds in History of Science" The theme of Professor Thomas S. Kuhn's paper is that the history of science is comprised of a succession of theories that can be viewed as a series of separate possible worlds whose lexicons are incommensurable. The thrust of his paper is that the causal theory of reference "provides no bridge across the divide, for the transworld voyages it envisages are limited to worlds in a single lexically possible set" (p. 31). For illustration Kühn examines how the lexicon of Newtonian mechanics is learned by a scientific community. He focuses on two ways to teach the terms mass and weight. The lexicons of each method are open to the same community. One method stipulates Newton's second law with the gravitational law emerging as an empirical regularity, and vice versa. As long as no anomalies are found there are no lexical difficulties between the segments of the scientific community who have taken the two different paths to learning these terms. Suppose now, continues Kühn, anomalies are found that require revisions of both laws at once. Revisions may be attempted that include metaphorical extension of certain terms. Since the lexicon of neither route can withstand such changes, eventually a transition (discontinuous) is made to another theory (world) with a lexicon disjoint (incommensurable) with the former. Certain of the new theory's syntax, however, could look the same as the old theory.1 Kühn rightly points out the threat to realism incurred by transitions to new and incommensurable theories. He then turns to Hilary Putnam's causal theory of reference. Putnam tries to avoid the problem of meaning-incommensurability by lifting natural kind terms out of theories and describing their ever-evolving reality status independently of any theory. So, for example, according to Putnam the term electron is the natural kind term that was in the theories of Lorentz, Einstein, Bohr, quantum mechanics and modern quantum electrodynamics despite the incommensurability of certain of these theories. The reason is that the quantity electron in these theories can be referred back through a historical causal chain to a description of electricity responsible for certain For an analysis of concepts that have persisted almost unchanged through the history of science see G. Holton, Thematic Origins of Scientific Thought: Kepler to Einstein, Harvard University Press, 1973.

34

A. I. Miller

effects, that is, back to the "introducing event" for the natural kind term electron.2 According to Putnam's causal theory of reference, writes Kühn, the progress of science is a relentless reduction or zeroing in on the actual or real world. Kühn goes on to show that whereas the causal theory works well for a natural kind term like gold, historical case studies reveal the fallacies in Putnam's example of water on earth and twin-earth. Kühn further argues against the causal theory by pointing out that the history of science reveals that the change in meaning of terms such as planet and star cannot be described by phrases such as "zeroing in" or "marginal adjustment" (p. 31). Causal theory provides no bridge across the divide, for the transworld voyages are limited to worlds in a single lexical set (p. 31). Consequently, without a translation apparatus causal theory cannot "talk of science's gradual elimination of all worlds excepting the real one (p. 31) ... no such description of scientific development can, I think, be sustained" (p. 31). According to the causal theory of reference an entity x, say water, in another possible world is the same as the entity in our world if it agrees with the lexicon of modern chemistry in our world (p. 27).3 We may turn this statement around by saying that transworld journeys begin from the lexicon of our world. I see the principal problem in Kuhn's analysis to be his emphasis on discontinuous change from one theory (or world) to another. This leads him to explore, right from the start, a series of already developed theories which are incommensurable. While translation between already developed theories may not be possible, a problem of interest is how theoretical terms become incommensurable. A key point in my commentary will be that the translation problem should be explored by starting with the problem of the new theory's invention. We may summarize Kuhn's conclusion thus: How did science work as well as it did when certain essential terms in older theories failed to refer? It is overwhelmingly likely that the same is true of terms indispensible to current science. Why, then, does science work as well as it does? The tradition's answer to that question has characteristically invoked phrases like "zeroing in". "Such descriptions of scientific development can no longer be sustained... Some other explanation of the achievements of science is needed" (pp. 31-32). My commentary will focus on two points: (1) the movement between successive theories or, phrased otherwise, transworld journeys and the attendant translation apparatus (I emphasize, however, my agreement with Kühn that between already developed theories there are no Quinian translations). (2) I should like to suggest "a different sort of question." These two points are not unconnected.4 See H. Putnam, "Explanation and Reference," in H. Putnam, Mind Language and Reality: Philosophkai Papers, Vol. 2, Cambridge University Press, 1975, pp. 196-214, esp. p. 202. See Putnam, "Meaning of Meaning," ibid., pp. 215-271, esp. p. 232. The notion that representations of reality in physics could be construed to be different worlds was to some degree explored during the late 19th and early 20th centuries and were referred to as

Discussion of Thomas S. Kuhn's paper

35

I will construct my commentary using an actual episode from the history of ideas - namely, the development of atomic physics from!913-1927. Right from the beginning the basic problem in atomic physics during this era was how concepts from classical physics that are represented with images and language anchored in the context of sense perceptions could be transferred to a world beyond sense perceptions, a world that was counterintuitive.5 Prior to the intense research into atomic physics generated by Bohr's atomic theory, physicists had dealt with physical systems that with some justification were assumed to be amenable to their perceptions and for which the space and time pictures of classical physics were applicable. In the German-language scientific literature the visual imagery of physical processes was referred to with the often-used term "intuition." "Intuition" has a particularly deep connotation because this term is rooted in Kantian philosophy.6 By "intuition" physicists like Niels Bohr, Albert Einstein, Werner Heisenberg, and Erwin Schr dinger, among others, meant the mental imagery that is abstracted from phenomena that we have witnessed in the world of sense perceptions - for example, there is the intuition of light as a wave phenomenon whose properties are abstracted from those of water waves, and of electrons behaving like billiard balls. Intuition is the visual imagery that is imposed on the syntax of classical physics. Particularly in the late 19th and early 20th centuries this approach to constructing physical theories had been immensely successful. But straightaway in his 1913 seminal paper on atomic physics Bohr realized presciently the serious difficulties in extending this methodology into the submicroscopic domain. In 1913 Bohr invented a new and daring theory of the atom. It was daring because in order to characterize the known properties of atoms Bohr had to violate classical physics. The theory violated classical mechanics because an atom's electrons are permitted only certain positions in the atom, and it violated classical electrornagnetism by restricting the atomic electrons to emit radiation only when they move between allowed positions. But the syntax and semantics of "ordinary mechanics", wrote Bohr in 1913, was the only one available and it enabled Bohr to give the theory a "simple interpretation ... with the symbols

5 6

world-pictures (Weltbilder). The principal ones were the mechanical and electromagnetic worldpictures whose goals were to describe all matter in morion with die laws of mechanics or electrornagnetism, respectively. In modern philosophical parlance die goals of diese worldpictures were to formulate worlds in which natural kind terms like electron were commensurable widi diose in pre-Weltbilder dieories, but not necessarily terms like mass and substance. Ahhough neidier world-picture succeeded, the electromagnetic world-picture played a role in Einstein's thinking toward die special relativity theory of 1905. A discussion of world-pictures can be found in A. I. Miller, Albert Einstein's Special Theory of Relativity: Emergence (1905) and Early Interpretation (1905-1911), Addison-Wesley, 1981, esp. Chapters 1 and 2. For details see A. I. Miller, Imagery in Scientific Thought: Creating 20th-century Physics, Birkh user, 1984, ΜΓΤ Press, 1986, esp. Chapter 4. The German-language word is Anschauung.

36

A. I. Miller

from ordinary mechanics." These symbols allowed visualization or intuition of an atom as a minuscule solar system. Having identified the language problem in the transworld journey, Bohr offered a prescription for moving from the macroscopic to the submicroscopic world which, in 1920, he called the correspondence principle.7 The relation between natural kind terms like electron, light, position, and momentum between these worlds emerged as the central problem in atomic physics. Toward elucidation of this problem a correspondence principle was indispensible. Correspondence principles are the only means to make a transworld journey that of necessity starts from the scientific lexicon of the world of sense perceptions, which we have come to realize is essentially Newtonian rather than Aristotelian. It is difficult to imagine how science could be done in a world in which there are no correspondence principles. How would Einstein have invented special relativity without considerations based on Newtonian mechanics? How would Bohr have invented an atomic theory without extending classical theory into atomic dimensions? How would Heisenberg have invented quantum mechanics without moving from Bohr's theory? How would Heisenberg have sown the seeds of the renormauzation program without a correspondence principle from quantum mechanics into the density matrix formalism? And so on. A universe without correspondence principles is one in which there are forces so strong as to make it virtually impossible to cut the world into segments in which we can, for example, treat a two-body problem, or to separate forces into weak, electromagnetic, strong and gravitational. In such a universe there would be no science nor, for that matter, life itself. Throughout the genesis of atomic physics the correspondence principle provided the means for physicists to adjust Bohr's atomic theory whenever necessary. Such a situation occurred in 1924, owing primarily to new empirical data on the interaction of light and atoms. These data led physicists to realize that the natural kind term "atom" in Bohr's theory did not refer to the atom's depiction as a minuscule solar system, and that light could well have a dual representation as a wave and a particle, as Einstein had hypothesized in 1905. Most physicists, including Bohr, had rejected Einstein's proposal of a waveparticle duality for light because it was counterintuitive - that is, no mechanical model or visualization was available for how particles of light (light quanta) could produce the phenomenon of interference, whereas this phenomenon

The spacing between adjacent planetary electron orbits in Bohr's theory decreases with increasing distance from the nucleus. Consequently, when an electron is in an orbit very far from the nucleus the frequency of the emitted radiation for a transition between neighboring orbits is very nearly the frequency of the electron in either the initial or the final state, which is the result expected from classical physics. According to Bohr's correspondence principle, the laws of classical electrodynamics are adjusted for distant orbits to include the frequencies in Bohr's atomic theory. Without any rigorous justification the newly adjusted equations were assumed to be valid also for orbits close to the nucleus.

Discussion of Thomas S. Kuhn's paper

37

could be visualized using water waves. As one physicist commented pungently in 1916, how could something behave "as if it possessed at the same time the opposite properties of extension and localization." In response to these new data Bohr used the correspondence principle to formulate a version of his theory in which an electron in a particular state in an atom is replaced by as many harmonic oscillators as there are possible electronic transitions to other states. This permitted Bohr to interpret these new data as the interaction between wave radiation and atoms. But since there are many possible transitions, in the 1924 version of Bohr's theory the atomic electron became unvisualizable. The term electron in the 1924 version of Bohr's theory is incommensurable with the one from the 1913 theory. Although Bohr's 1924 theory ultimately failed, by June 1925 it provided Heisenberg with the clue for using the correspondence principle to invent the new atomic theory or quantum mechanics which was based on the measurable characteristics of unvisualizable particles. Almost immediately Heisenberg's mathematical formalism was rewritten so as to eliminate any vestiges of the correspondence principle that linked it to the old Bohr theory. But the umbilical cord between the new quantum mechanics and Bohr's theory with its classical intuition could not yet be cut because whenever statements of the new world's lexicon were set in sentences, they were ambiguous and led to paradoxes. Consequently, a possible world of the atom had been formulated in an ambiguously interpreted syntax (cf. Kühn, pp. 13-14). This was the central problem with which Bohr and Heisenberg would struggle during 1926-1927 at Copenhagen. Its resolution forms the core of the so-called Copenhagen interpretation which altered our view of physical reality in ways that are still not completely understood. In the published scientific literature of late 1925 through fall of 1927 key scientists expressed the desire for the return of some sort of visualization based on intuition of the atomic domain. For example, in late 1925, Heisenberg wrote of how the present theory "labored under the disadvantage that there can be no directly intuitive geometrical interpretation because the motion of electrons cannot be described in terms of the familiar concepts of space and time ... In the further development of the theory, an important task will lie in the closer investigation of the nature of the correspondence between classical and quantum mechanics and in the manner in which symbolic quantum geometry goes into intuitive classical geometry." It seemed reasonable that to establish a consistent and unambiguous lexicon for the new world required reaching back into the lexicon of the world of classical physics. The question was how this was to be established? Erwin Schrödinger thought he had a means to impose intuitive representations on atomic phenomena. Schrödinger based his version of atomic physics on a proposal that had been made in 1923 by Louis de Broglie of a wave-particle duality for matter. In early 1926 Schrödinger formulated a wave mechanics based on a wave imagery of electrons in which atomic transitions were contin-

38

A. I. Miller

uous and visualizable like the transitions between the vibrational modes in a drumhead. Schrödinger wrote in a March 1926 publication that he had formulated the wave mechanics because he "felt... repelled by the lack of visualizability" of Heisenberg's quantum mechanics. He went on to write that we should not approach atomic physics with a "theory of knowledge" in which we "suppress intuition." In a letter commenting on Schrödinger's work Heisenberg wrote that he considered what "Schrödinger writes on visualizability as trash." Heisenberg went on to demonstrate the incorrectness and inappropriateness of certain of Schrödinger's wave imagery, lauding only wave mechanics' calculational advantages over the quantum mechanics. During the brief period mid-1925 through fall of 1927 problem after problem that had resisted treatment in the old Bohr theory was solved, and several exciting and unexpected results emerged. Some of the new results were puzzling and misunderstood. The situation was rooted in the total failure of physicists to extend the mental imagery of classical physics, i. e., intuition with its perceptionladen language, into the atomic domain. Physicists were adrift with no anchor to the world of perceptions - that is, the concepts and language of classical physics were inappropriate in the atomic domain. Continued reliance on customary intuition was the stumbling block in Bohr's and Heisenberg's heroic struggles in Copenhagen during fall of 1926 into 1927 to interpret the syntax of the new quantum mechanics. Heisenberg often recalled the "despair" to which Bohr and himself were driven as a result of their considerations on the nature of physical reality in a realm where the language of tables and chairs from the workf of sense perceptions was not applicable, and yet it was the only language they had. We can glimpse the central point of their struggles from Heisenberg's first epistemological analysis of the quantum mechanics that he wrote in Copenhagen in November of 1926. He wrote, "Our customary intuition [leads us to attribute to electrons] the same sort of reality as the objects of our daily world [but this turned out to be false]. For the electron and the atom possess not any degree of direct physical reality as the objects of daily experience." One of Heisenberg's principal reasons for coming to this conclusion is the wave-particle duality of light and matter. In the world of customary intuition - that is, the world of classical physics - entities can be either continuous or discontinuous, but not both. Yet on the subatomic level entities seemed to be both continuous and discontinuous simultaneously. As Heisenberg wrote in a letter from Copenhagen during this period, "What the words wave and particle mean we know not any longer." This statement is marvelously close to the one from J. L. Austin that Kühn cites on p. 20 ("we don't know what to say. Words literally fail us."). See also Kuhn's p. 21 where, to change terminology, "Yes the entity is an electron (but to mean electron is not what it was)." Words literally failed Bohr and Heisenberg because the natural kind term electron in quantum mechanics did not refer to the electron in classical

Discussion of Thomas S. Kuhn's paper

39

physics. And, worse yet, the term "light quantum" never possessed an agreed upon introducing event, and so could not even be accorded the status of a natural kind term. Since the central problem for Bohr and Heisenberg concerned the nature of physical reality in a realm beyond sense perceptions, their resolution of this situation required an analysis that went beyond physics and even philosophy per se. In early 1927 Bohr and Heisenberg arrived at apparendy separate resolutions to the problem of interpreting the symbols of the new atomic physics. Bohr's insight, as he wrote in September 1927, involved his coming full circle back to a realization of 1913 that the restrictions imposed by language on our capacity to form images for scientific theories originated "above all in the fact that, so to say, every word in the language refers to our ordinary intuition." Bohr's dilemma was that whereas images from intuition had to be separated from the laws of atomic physics, we are forced to phrase these laws in a language tempered by sense perceptions (i.e., the world of classical physics) because it is the only language that we have. Bohr encompassed both horns of the dilemma with the principle of complementarity. This far-reaching principle has essentially two main parts: (1) In the atomic domain there is a difference between pictures or intuitive imagery and the actual development of atomic systems which is the province of physical laws that require a "departure from visualization in the usual sense." (2) Bohr's masterstroke was to realize that the wave-particle duality of light and matter was only paradoxical because of limitations in the atomic domain on our language. Rather, in the atomic domain both modes are connected. That is, a single fundamental constant of nature (Planck's constant) links quantities that characterize a localized entity like a particle (energy and momentum) with quantities that characterize an extended entity like a wave (frequency and wavelength). Bohr reasoned that just as the large value of the velocity of light had prevented our realizing the relativity of time, the minuteness of Planck's constant rendered paradoxical the wave-particle duality of matter and light. Since Planck's constant places restrictions on the use of our language in the atomic domain, then so too on our intuition or visual imagery, which enables us to describe only things that are either continuous or discontinuous but not both. Yet subatomic particles are simultaneously localized and extended. So, stressed Bohr, the wave and particle modes of light and matter are neither contradictory nor paradoxical, but complementary. Both modes or sides are required for a complete description of the atomic entity.8 8

I would like to mention here a similarity between possible worlds in art and science - namely, a similarity between cubist art and the complementarity description of subatomic matter. In their widely-read book of 1912 entitled Du Cubisme,]ta.n Metzinger and Albert Gleizes published a

40

A. L Miller

Unlike Bohr, Heisenberg had focused initially on his own quantum mechanics which described matter to be comprised of unvisualizable subatomic particles. In a March 1927 paper entitled "On the Intuitive Content of the QuantumTheoretical Kinematics and Mechanics," Heisenberg presented his own version of how the syntax of the quantum mechanics should be interpreted. The appearance of the word "intuitive" in the title of this classic paper in the history of ideas further underscores the importance of the concept of intuition for Heisenberg. In this paper Heisenberg resisted any imagery of atoms, and permitted the theory to decide restrictions on the meaning of symbols like v for velocity and χ for position, i.e., the uncertainty relations. These are among the symbols that are necessary holdovers from classical physics and so are linked to the world of perceptions; they are necessary for talking about the theory and for use in experimental observations. But at Bohr's insistence Heisenberg realized the necessity of the wave and particle representations of light and matter toward achieving a more consistent and general interpretation of the quantum theory than one with particles only. Although Heisenberg agreed with the complementarity principle's restrictions on metaphors from the world of sense perceptions, he remained wary of them owing to their previous disservices. As Heisenberg wrote in a letter of May 1927, "there are still differences of opinion between Bohr and myself over the term 'intuitive'." Through his subsequent research in the 1930's into nuclear physics and the interactions between matter and light, Heisenberg developed a more far-reaching form of mental imagery than the restricted metaphors of the complementarity principle - namely, that the mathematics of the quantum mechanics determines systematic exposition of cubist methods. Art historians regard Metzinger as a minor painter but as a major theorist of the cubist movement. A cubist painting, they wrote, represented a scene as if the observer were "moving around an object [in order to] seize it from several successive appearances ...". Cubists achieved this motif through the interpenetration of figure and space in order to free the artist from a single perspective in favor of multiple viewpoints. We know that Bohr had been much interested in cubism before 1927.1 recently learned that sometime after 1927 he acquired a cubist painting for his study which he took joy in explaining to his visitors. One might have expected Bohr to have displayed a painting by one of the masters of this genre. Instead Bohr hung Metzinger's L'Ecuyere which could mean that Bohr's interest in cubism went far beyond painting per se. I should like to propose the following line of speculation on how cubist theory could have affected Bohr's thinking toward complementarity. Mogens Anderson, a Danish artist and friend of Bohr, recollected Bohr's pleasure in giving "form to thoughts to an audience at first unable to see anything in [Metzinger's] painting - they came with a preconceived idea of what art should be." Such had been the case in 1913, when atomic physicists had a preconceived image of the atom that had been imposed on atomic theory owing to the representation of the theory's symbols. By 1925 atomic physicists had come to realize the inadequacy of visual perception, as had the cubists. In 1927 Bohr offered a motif for the world of the atom with striking parallels to the motif of multiple perspectives offered by cubism for glimpsing beyond and behind visual perceptions: According to complementarity the atomic entity has two sides - wave and particle - and depending on how you look at it, that is, what experimental arrangement is used, that is what it is.

Discussion of Thomas S. Kuhn's paper

41

also the visualizability of subatomic particles which was subsequently developed into the present-day Feynman diagrammatic methods. This result entailed Heisenberg's realization that in the atomic domain there is a distinction between visualization and visualizability. Visualization is the result of our cognition. Visualizability concerns the intrinsic properties of objects. In classical physics visualization and visualizability are synonymous. But there can be no visualization of atomic entities; Bohr's analysis of perceptions had made that abundantly clear. Yet atomic entities are visualizable because their intrinsic characteristics are revealed through the mathematics of the quantum mechanics; to that Heisenberg had been driven. Thus, Heisenberg's scientific research had led him to redefine the classical concept of intuition by linking it with the visualizability of subatomic matter that is achieved through a mathematical formalism and not through perceptions. Physics had come full circle back to the problem of knowledge representation that had been set by Plato in the Theaetetus 2,000 years before. In the developed and interpreted quantum physics the terms visualization and visualizability are incommensurable with the terms in classical physics because the visual images differ, among other reasons. So too are the terms electron, position and velocity incommensurable with those in classical physics. The main problem, it seems to me, is understanding the process by which such terms became incommensurable. Whereas deep inside the developed classical physics and quantum mechanics theoretical terms and their images are incommensurable, such is not the case at the boundaries - indeed, the boundaries of worlds of science are diffuse and intertwine with one another. Heisenberg, whose research strategy was rooted in correspondence principles, said it well when he entitled a book of essays describing his scientific research as, Across the frontiers.9 In summary: Describing scientific progress as discontinuous leads to comparison of already developed theories - or, I should say, the inner parts of different worlds which are disjunct. I maintain, however, that this does not describe adequately scientific progress, and perhaps a better description of scientific progress could result in understanding how theoretical terms are related in possible other worlds - that is, we are once again concerned with the problem of realism. Next I turn to Professor Kuhn's deep query of how can it be that science in the past was at all successful when most of its essential terms did not refer? But did not certain essential terms in the older theories refer to something that can be regarded as pre-scientific knowledge? For example, caloric referred to a fluid and the ether to a medium required to support light in transit just like water supports waves. In conclusion, Kühn is right - a different sort of "explanation" must be found. I suspect that it will be found only through expanding one's horizons to include philosophical systems and theories of art and literature that explore how things become the way they are. W. Heisenberg, Across the frontiers, translated by P. Heath, Harper, 1974.

TORE FRÄNGSMYR

Discussion of Thomas S. Kuhn's paper "Possible Worlds in History of Science" The task of the historian of science is to study the origins and historical course of science, both in its internal theoretical development and in its external relations with the society within which it has existed. In other words, the historian's work is not merely a matter of establishing certain facts about historical events or of confirming certain demonstrable causal connections, but also a process of entering into the intellectual world of a particular period, of "understanding", as Wilhelm Dilthey expressed it (verstehen). In the positivist tradition the ambition was primarily to "understand" science itself, its hypotheses and conditions; a more externalist approach attaches more importance to the "understanding" of history, and with it the part played by science during different epochs. The emphasis then shifts from science as we see it today to a particular historical period in which science was an important or unimportant factor. But if we are to be able to "understand" and analyse the function of science, we have to "understand" its origin and its structure. Theories do not arise by spontaneous generation - they develop in continuous interaction with other theories and the world of ideas, prejudices and policies which prevails at a particular time. The task of the historian of science is therefore not only to "understand" but also to an even greater degree to "reconstruct", to build up different "possible worlds". Such a possible world need not be global in extent, but the underlying philosophy must not be too restricted in scope, nor relevant only to certain parts of more complex theories or to theories explaining very limited phenomena.

In his paper Thomas Kühn has brought up important problems confronting the historian when he has to read and analyse an out-of-date scientific text. Kühn makes the point that between the old text and our modern understanding of it there is always a gap, which he explains with the aid of terms such as "incommensurability" and "untranslatability"; translation is in this respect not the same as language learning. All the well-intentioned theories of an Ur-sprache have unfortunately missed the mark. I also agree with the observation that the translating of an old scientific text has much in common with the translating of

Discussion of Thomas S. Kuhn's paper

43

fiction; the truth value of a scientific text is as tricky to bring out as the style and emotional tone of a literary one. As the starting point of my contribution, I should like to take Kuhn's comment that a word has to be seen in relation to its specific context: "knowing what a word means is knowing how to use it for communication with other members of the language community within which it is current." And he goes on to say: "At least in their application to historical development, the power and the utility of possible-world arguments appears to require their restriction to the worlds accessible with a given lexicon, the worlds conceivable by participants in a given language-community or culture." To illustrate his view, Kühn takes Newton's terms "force", "mass" and "weight", and imagines a situation where one has to teach students to understand the world of Newton's ideas. Kuhn's approach is largely abstract and 'theoretical'. I would like to introduce a complementary, more 'practical' perspective. In my opinion, the most usual task of the historian of science may not be to "translate" words and theories into our modern terminology and to our modern level of understanding, so much as to understand a system of thought in its given setting. In other words, it is often more difficult to understand an explanatory process and its various constituents as a self-contained "possible world" than to be abk to express the content of its key terms in the language of modem science. Such linguistic translation work is, as Kühn very rightly says, in most cases impossible. But that does not mean that it is impossible to "understand" theories, ideas or explanations concerning particular natural phenomena, provided that one knows the context in which they have developed. We may for example take the case of theories and explanations of natural processes, where the words and the terms are not difficult to understand and the chain of argument in its entirety has to be explained from its individual components. It follows that such a "possible world" can be understood by us, but that is not to say that it can be translated into the language of today. The commonest reason for this is of course that theories of this kind lack some explanatory factor which has emerged as a result of later research. If we are to avoid anachronisms in our historical perspective, which have often been a feature of the positivist approach, we must therefore accept a language and an explanatory model that have no counterpart in modern scientific argument. However, we need to familiarize ourselves thoroughly with the ideological and social context of the theory concerned. Perhaps one could even extend the compass of such a statement and say that most earlier scientific texts are of this nature. This particularly applies to empirical and inexact sciences such as biology, geology and chemistry (or at least to its alchemic tradition). Here a new stage of development has involved not only new interpretation of known facts but also the making of new empirical findings and observations. In this way an entirely new factor has been added.

44

T. Frängsmyr

Exact sciences, such as mathematics and physics, have probably not been so dependent on new empirical discoveries. I shall begin by giving some examples from the history of geology. As far back as the 1690s, the Swedish chemist and geologist Urban Hiärne discovered that the water level of the Baltic was changing. He saw that the water line had steadily become lower, but he did not know whether this was because in absolute terms the land was rising or the water was falling; however, he assumed the latter. He tried to explain this by stating that the level of the water of the Baltic was higher than that of the North Sea and that as the width of the strait between the two seas, the Sound, increased, more water was able to flow out into the North Sea; the two water surfaces were striving to find the same level. Hiärne's observation was correct, but his conclusion contained no very complex theory and it applied only to a phenomenon that he observed along the shores of the Baltic. However, this scientific question was of fundamental interest. It is no exaggeration to state that it was to dominate Swedish geological research for 200 years. Here we had empirical observations of a natural process, which was quite undeniable, but no tenable explanation of the cause of the observed phenomenon. Not until 1840 did it become clear - with the glacial theory of Louis Agassiz - that Scandinavia and the greater part of the northern hemisphere had been covered by a thick layer of ice, the weight of which had depressed the earth's crust. After the ice had melted, the crust of the earth had begun gradually to rise again to its original position, a process which is continuing today. It may also be mentioned that the theory or an ice age was not generally accepted by geologists until around 1875-1880. In the eighteenth century nothing was known about glaciation, but nevertheless scientists tried to explain the phenomenon. Moreover, many of them realized that the problem must concern not only the Baltic Sea but also a large part of the world, perhaps the whole globe. Here, in other words, we get a series of attempts to create "possible worlds" in the realms of science. This is of course a very familiar situation in the history of science. Scientists have observed a natural process, for which they want to give a rational scientific explanation. As they still lack the key to the true explanation (the glacial theory), they are compelled to speculate about various causal relationships. Naturally they concentrate on the factors with which they are working in their own scientific disciplines. We cannot pass judgment on these scientific attempts, for they were working under the particular conditions imposed by their time. But every interpretation, every model, sought to create a context, to give a consistent explanation of a phenomenon of general validity. Emanuel Swedenborg, scientist, philosopher and theologian, proceeded from Descartes' theory that the heavenly bodies were whirled around in space in ethereal vortices. He combined this theory with the old Christian conception of an ageing earth, mundus senescens. One of many indications that the earth was ageing was the fact that people no longer lived to be as old as those in the Bible:

Discussion of Thomas S. Kuhn's paper

45

Methuselah had lived 969 years, for example. But this could be interpreted as a mathematical problem. Formerly, believed Swedenborg, the earth had spun more rapidly on its axis, and the ageing process meant that it was now slowing down. But a year denoted one orbit of the sun, and Swedenborg assumed that the velocity of the earth had at first been up to eight times as great. In that case the stated age of Methuselah ought to be divided by eight, giving 121 years, which was quite respectable enough. If the earth had completed its orbit faster in earlier times, then the moon and its vortex must also have done so. Tidal ebb and flow would then have alternated more rapidly, perhaps only every two or three hours instead of six, which would mean that the water never had time to recede to its original position before a new tide came. For this reason the water line at the poles had been "some 100 ells higher than now", but as the velocity of the earth had decreased, the water had been able to fall further and assume a lower level. Swedenborg was, like many of his contemporaries, an eclectic; his theory was more complex than is shown by triis brief outline. He cited both modern astronomical principles and old Christian ideas, but nevertheless they resulted in a theory which was logically coherent. Naturally we may question his premisses, but based on those premisses he undoubtedly described a "possible world". Carl Linnaeus was a natural historian, first and foremost a botanist but also a zoologist. He assumed that the earth had originally been only a little island. By a slow process entailing both creation of land by siltation and the lowering of the water level, the level of the land rose and greater and greater landmasses were gradually exposed. Thus the dry land had eventually grown from a little island into a network of interconnected continents. The Bible told us that all the animals and plants had been present in paradise so that Adam could give them names. From his theory of an original island, Linnaeus could then explain the geographical dispersion of plants and animals; this was a scientific problem in Linnaeus' time, as it was necessary either to assume parallel acts of creation in different parts of the world or to explain how distribution had taken place over high mountains and great seas. To provide this explanation was the primary purpose of Linnaeus' theory. He did not attempt to establish the actual cause of the diminution of the water. He described in vague terms how the seas became deeper, while sand and mud increased the area of the land. Nor was he particularly interested in the causal connection; it was the cause of the distribution of the flora and fauna that he wished to describe. Like Swedenborg, he combined modern science with biblical conceptions. A third theory was advanced by Anders Celsius, the astronomer and mathematician. He took up Newton's comments on the capacity of the comets to provide the earth with humidity and water from their vapours. This idea had been particularly developed by the successor to Newton's professorship, William Whiston. Celsius had also gained knowledge of the comets from Halley. According to Celsius, all heavenly bodies went through a three-stage process:

46

T. Frängsmyr

inundation, "combustion" and an intermediate state, a sort of interlude between the two extremes. The earth is now, states Celsius, in this transient stage, but the continuing diminution of the waters shows that it has recently gone through a period of inundation. After a few thousand years the earth will have been deprived of its water and the air of its humidity. The combined action of the sun and the subterranean fires will cause the greater part of the earth's surface to begin to burn. If a comet passed close to the earth, it may cause a flood with its humidity: "If the air of the comet has much water in it, we might, as Mr Whiston believes, have a deluge of rain and a great flood over the earth." However, there was one essential difference between Whiston and Celsius. Whiston assumed that the great flood caused by the comet was identical with the Flood in the Bible, and that the fire represented the earth's final destruction. Celsius is of a different mind. He states clearly that his description does not relate to the flood of Noah or the fires consuming the earth on the last day, for according to the theologians these events cannot occur naturally; the implication is that the processes described by Celsius were natural ones and need not require supernatural intervention. Celsius thus takes a fully rational view of the question of the diminution of the waters. He himself took measurements and reached the conclusion that the water falls by roughly a centimetre per year, i.e. one metre per century, which is in fact very close to the true figure. Unlike Swedenborg and Linnaeus, Celsius did not make use of Christian mythology in his explanations, although we may of course feel that the speculations concerning the comets were on roughly the same level. We have looked at three examples of explanations arrived at from different starting points; the phenomenon to be explained is contemplated from the angle of the individual's own speciality. The theory that is formulated is indeed based on observation, but not on an empirical foundation of any breadth. Geology did not at that time work by systematic experimentation, and theory therefore has few points of contact with empirical reality. The three examples given illustrate the normal behaviour of scientists formulating theories: they believe that the possible world they have created is a part of reality. There are also cases where the scientist deliberately constructs a possible world, despite the fact that it does not correspond to reality. He wishes to use this possible world as a metaphor, a representation of reality; here we may recall the comparison with literature made by Kühn. The scientist may in other words paint pictures of reality in order to "understand" reality, but he knows that they are pictures and not factual descriptions. One example of this is the reaction of Jacob Berzelius to Dalton's atomic theory. Berzelius had been working on the question of chemical proportions for some time, when he heard of Dalton's theory. The news of it raised his hopes, for he thought that it would provide him with a theoretical explanation of the phenomenon of proportion. But Berzelius had also discovered that when two substances combined, the smallest possible unit of one of them had to be present in the compound in accordance with the formulae A+B, A+2B, A+3B etc. This

Discussion of Thomas S. Kuhn's paper

47

law of unity became very important to Berzelius, and put him on a collision course with Dalton. In the degrees of oxidation of various metals he had found the composition A + 1 VzB, and this did not accord with the atomic theory, which could not accept half-atoms. Nor could he think of expressing the relationship as 2A+3B, for this conflicted with the law of unity, which Berzelius considered he had proved empirically. After this, he criticized Dalton on several occasions. Dalton had put forward a hypothesis but argued as if it were an axiom. To be sure, a scientist had to make use of hypotheses, which might even be "bridges to the truth", but they must not be used as axioms to which the results had to be adjusted; in that event they might become "slippery slides to error and aberration". The criteria of a good theory were two in number: its practical applicability in the laboratory, and its comprehensiveness. Every theory must explain all the phenomena in the area to which it related, leaving none unaccounted for. Dalton's theory did not satisfy these requirements. Concerning its physical validity, we could know nothing, for it belonged to that area of chemistry that could be neither proved nor disproved. Purely empirically Dalton had not given a shred of evidence to show that the basic law of unity was incorrect. Berzelius had demonstrated it empirically, and Dalton had not refuted it. Berzelius often returned to the problem, but it became more urgent in 1816, when he was doing work on phosphorus, the phosphoric acids and the salts of phosphoric acids. His results did not agree with the theory of proportions, for he was obliged to assume the proportion 2:3 instead of 1:1 !/2 in certain isolated cases, and by so doing he denied the universality of the law of unity. It was with great reluctance that Berzelius made this concession, but once it was done it became easier to use Dalton's theory. He began to use it in his argumentation, but only as a model; he maintained that he did so "regardless of its possible truth". By the time the third volume of his book Lärbok i kernten ("Manual of Chemistry") came out in 1818, he had accepted the atomic theory "as the best conceivable hypothesis, which is not to say that it is true". Nor did this mean that other theories were incorrect, only that they were more limited and therefore not so applicable. He continued to cling to the law of unity, but he acknowledged that its validity was not comprehensive. In addition he maintained that Dalton had never tried to explain the forces acting between atoms. Berzelius never abandoned his basic position that the atomic theory was only a best-available hypothesis. He saw it as a model of reality, but he never believed that it described the true reality. The atomic theory had not been empirically verified, but it was very serviceable for the practical chemist. Berzelius plays with intellectual experiments, with "possible worlds", we might say. He was both an empirical scientist and a speculative theoretician, for he always sought an empirical foundation for his speculation. He was receptive to theoretical models, but he distinguished between them and descriptions of reality. In this sense Berzelius was a modern philosopher: a true description of reality was impossible, but a model of reality was of practical utility. Perhaps a discussion of possible

48

T. Frängsmyr

worlds also has to finish at this point, as far as the history of science and the understanding of early scientific texts are concerned. The most essential goal may not be to "translate" them into our own idiom, nor to apply them to the present positions of science. What is essential and difficult, may rather be to "understand" the reasoning of scientists at different periods and in different conditions, i. e. to put the theories in an intellectual, social and cultural context. Only then is it possible for us to understand possible worlds in the history of science.

THOMAS S. KÜHN Speaker's Reply I am most grateful to Professors Frängsmyr and Miller for their comments on my paper. Occasional misunderstandings aside (we are, for example, using the phrase "possible worlds" quite differently), I agree fully with what they have to say. Scientific development has numerous aspects besides the ones on which my paper focusses; their remarks complement mine by cogently illustrating other topics I might have discussed. On only one point is further response required. Early in his commentary, Professor Miller writes, "I see the principal problem in Kuhn's analysis to be his emphasis on discontinuous change from one theory (or world) to another." There is, however, no talk of discontinuous change in my paper, much less any emphasis on it. The contrast throughout is between the lexicons used at two widely separated times: nothing is said about the nature of the intervening process by which a transition between them is made. The point is worth elaborating: my past work has often invoked discontinuity, and my present paper points the way towards significant reformulation. In recent years I have increasingly recognized that my conception of the process by which scientists move forward has been too closely modelled on my experience with the process by which historians move into the past.1 For the historian, the period of wrestling with nonsense passages in out-of-date texts is ordinarily marked by episodes in which the sudden recovery of a long-forgotten way to use some still-familiar terms brings new understanding and coherence. In the sciences, similar "aha-experiences" mark the periods of frustration and puzzlement that ordinarily precede fundamental innovation and that often precede the understanding of innovation as well. The testimony of scientists to such experiences, together with my own experience as an historian, was the basis for my repeated reference to gestalt switches, conversion experiences, and the like. In many of the places in which such phrases appeared, their use was literal or very nearly so, and in those places I would use them again, though perhaps with more care for rhetorical overtones. In other places, however, a special characteristic of scientific development led me to use such terms metaphorically, often without quite recognizing the difference in use. The sciences are unique among creative descipfines in the For a particularly clear example ofthat modelling, see "What are Scientific Revolutions?", cited in the first footnote to my paper.

50

Th. S. Kühn

extent to which they cut themselves off from their past, substituting for it a systematic reconstruction. Few scientists read past scientific works; science libraries ordinarily displace the books and journals in which such work is recorded; scientific life knows no institutional equivalent for the art museum. Another symptom is presently more central. When reconceptualization occurs in a scientific field, displaced concepts rapidly vanish from professional view. Later practitioners reconstruct their predecessors' work in trie conceptual vocabulary they use themselves, a vocabulary incapable of representing what those predecessors actually did. Such reconstruction is a precondition for the cumulative image of scientific development familiar from science textbooks, but it badly misrepresents the past.2 No wonder that the historian, breaking through to that past, experiences the breakthrough as a gestalt switch. And, since what the historian breaks through to is not simply the concepts deployed by a single scientist but those of a once active community, it is natural to speak of the community itself as having undergone a gestalt switch when it displaced its previous conceptual vocabulary for a new one. The temptation to use 'gestalt switch' and related phrases in that way is particularly strong both because the interval in which the conceptual vocabulary shifted is usually short and because, during that interval, a number of individual scientists did experience gestalt switches. The transfer of terms like 'gestalt switch* from individuals to groups is, however, clearly metaphorical, and in this case the metaphor proves damaging. Insofar as the historian's gestalt switch provides the model, the magnitude of the conceptual transpositions characteristic of scientific development is exaggerated. Historians, working backwards, regularly experience as a single conceptual shift a transposition for which the developmental process required a series of stages. More important, treating groups or communities as though they were individuals-writ-large misrepresents the process of conceptual change. Communities do not have experiences, much less gestalt switches. As the conceptual vocabulary of a community changes, its members may undergo gestalt switches, but only some of them do and not all at the same time. Of those who do not, some cease to be members of the community; others acquire the new vocabulary in less dramatic ways. Meanwhile, communication goes on, however imperfectly, metaphor serving as a partial bridge across the divide between an old literal usage and a new one. To speak, as I repeatedly have, of a community's undergoing a gestalt switch is to compress an extended process of change into an instant, leaving no room for the microprocesses by which the change is achieved. Recognition of these difficulties opens two directions for further development. Trie first is the one for which Professor Miller calls and which his 2

On this subject see: "The Invisibility of Revolutions", in my Structure of Scientific Revolutions (2nd ed.; Chicago, 1970), pp. 136-143; "Comment on the Relations of Science and Art," in my Essential Tension (Chicago, 1977), pp. 340-351; and my "Revisiting Planck," Historical Studies in the Physical Sciences, 14 (1984), 231-252, esp. Part 4.

Speaker's Reply

51

comments illustrate: study of the microprocesses which occur within a community during periods of conceptual change. Excepting in its repeated references to metaphor, my paper has nothing to say about them, but its formulation, unlike that of my older work, is designed to leave room for their exploration.3 The second, which may prove even more important, is a systematic attempt to separate the concepts appropriate to the description of groups from those appropriate to the description of individuals. That attempt is currently among my central concerns, and one of its products plays a central, though mostly implicit, role in my paper.4 People may, I there insist, "use the same lexicon, refer to the same items with it, and yet pick out those items in different ways. Reference is a function of the shared structure of the lexicon but not of the varied feature spaces within which individuals represent that structure" (n. 25). A number of the classical problems of meaning, I am suggesting, may be seen as a product of the failure to distinguish between the lexicon as a shared property constitutive of community, on the one hand, and the lexicon as something carried by each individual member of the community, on the other.

3

4

The contrast is, of course, only with my older metahistorical work. As historian I have often dealt with the detail of the transition process. See, especially, Black-Body Theory and the Quantum Discontinuity, 1894-1912 (Oxford and New York, 1978), paperback edition (Chicago, 1987). Others are indicated in my "Scientific Knowledge as Historical Product," to appear in Synthese.

JAAKKO HINTIKKA Exploring Possible Worlds"" 1. Possible-worlds semantics and one of its problems

In its different forms, possible-worlds semantics has been by and large an eminently successful logico-semantical theory. It began its life as a theory of logical (conceptual) necessity and possibility, inspired by the old idea of necessity as truth in all possible worlds.1 Soon it was broadened into a framework for a general theory of meaning, based on the idea of meanings as functions from possible worlds to extensions.2 Possible-worlds semantics can also, and in my judgment much more naturally, be used to explicate the semantics of such propositional attitudes as knowledge, belief, perception, etc. For what could be a more natural explication of what someone knows than to spell it out in terms of the possible scenarios ("possible worlds") which the information he or she possesses enables him or her to disregard? This is nothing but the familiar idea of information as elimination of uncertainty. Of course, to spell out which scenarios are ruled out by what someone knows is ipso facto to specify the class of all the scenarios that are compatible with that someone's knowledge. And to specify this class (relative to each world, hence specifiable by means of a two-place "alternativeness relation" between worlds or scenarios) is precisely the PWS explication of knowledge (information). The explication of sucn notions as belief is equally natural. As far as perception is concerned, the possible-worlds treatment simply amounts to providing a conceptual frameAcknowledgement: Work on this paper was facilitated by NSF Grant No. 1ST 8310936 (Information Science and Technology). The history of possible-worlds semantics remains to be written. The crucial chapter of that history is the step from the earlier treatment of logical modalities in model-theoretical terms by Carnap, Tarski, McKinsey and others to "real" possible-worlds semantics. This step is much more philosophical than purely logical. It means adopting in some sense a realistic attitude of the other possible worlds. It does have a partial technical manifestation, however. This manifestion is the introduction of the idea of an alternativeness (accessibility) relation on the set of possible worlds. It is not known who was the first one to put forward that idea in print. Stig Kanger did so in 1957, but I do not know if he was anticipated. Kripke did not use the idea in print until 1962. This is of course the basic idea of Montague semantics. For it, see Richard Montague (ed. by Richmond Thomason), Formal Philosophy, Yale University Press, 1974, and David R. Dowty, Robert E. Wall and Stanley Peters, Introduction to Montague Semantics, D. Reidel, Dordrecht 1981 (both with bibliographies).

Exploring Possible Worlds

53

work for the informational approach to perception which is implicitly or explicitly (though not necessarily under this heading) favored by some of the best psychologists of perception.3 There is much still to be done to clarify the conceptual basis of possibleworlds semantics and even more to be done to apply this framework to different problems in philosophy, cognitive science and in other branches of science and scholarship. But what is there to be done in the logical theory of possible-worlds semantics itself? I shall look for an answer in the historical antecedents and presuppositions of PWS.

2. The parentage ofpossihle^worlds semantics In order to understand the prospects of possible-worlds semantics, it is useful to have a look at its historical background. In a sufficiently long perspective, the genesis of possible-worlds semantics is a part and parcel of a gradual switch from one philosophical position to another. I have called these two positions the conception of language as the universal medium and the conception of language as cakidus. The contrast between these two overall ways of looking at language and its logic, and of their interaction, is the most important and the most neglected general feature of the philosophy of language and philosophy of logic in the twentieth century. Their history mostly remains to be written, and certainly cannot be written adequately here. A few indications will have to suffice.4 By the conception of language as the universal medium, I mean the idea that we cannot in the last analysis escape our language and as it were look at it and its logic from the outside. As a consequence, the semantics of our language is inexpressible, and cannot be theorized about in language. As a further consequence, all realistic model theory is impossible (except as a mere technical device without deeper philosophical significance). This view has been represented, in

The relative neglect of the actual and potential applications of the PWS framework to the analysis of such concepts as knowledge, belief, memory and perception (known among philosophers as "propositional attitudes") is a puzzling phenomenon whose causes escape me. One partial reason is perhaps the prejudice, voiced among others by Noam Chomsky, that any PWS analysis of propositional attitudes is bound to lead to the problem of "logical omniscience". Yet this objection was shown to be inconclusive way back in 1975; see Veikko Rantala, "Urn Models", Journal of Philosophical Logic, vol. 4 (1975), pp. 455-474 and Jaakko Hintikka, "Impossible Possible Worlds Vindicated", ibid. pp. 475-484. Bits and pieces of the history of this grand contrast are found in the following publications: Jean van Heijenoort, "Logic as Language and Logic as Calculus", Synthesevol. 17 (1967), pp. 324-330; Warren Goldfarb, "Logic in uie^enues", Journal of Symbolic Logicvol. 44 (1979), pp. 351-368; Peter Hylton, "Russell's Substitutional Theory", Synthese vol. 45 (1980), pp. 1-31; Merrill B. Hintikka and Jaakko Hintikka, Investigating Wittgenstein, Basil Blackwell, Oxford, 1986, ch. 1; Jaakko Hintikka, "Is Truth Ineffable?", forthcoming.

54

J. Hintikka

different variants, by Frege, early Russell, Wittgenstein, the Vienna Circle (this is what their onetime exclusive reliance on "the formal mode speech" amounted to) and Quine. Outside the analytic tradition, the idea of language as the universal medium is an important part of the philosophical position of, among others, Martin Heidegger and Hans-Georg Gadamer.5 According to the opposite view, the view I have called the conception of language as calculus, you can so to speak stop your language and step off. In less metaphoric terms, you can discuss the semantics of your language and even vary systematically its interpretation. The term "language as calculus" ist not calculated to indicate that on this view language would be a meaningless jett de caracteres - that is not the idea at all. Rather, the operative word highlights the thesis that language is freely re-interpretable like a calculus. This view has played less the role of major philosophical assumption (or a presupposition) than that of a working hypothesis of many practicing logicians' research. In this sense, the tradition of language as calculus has been tacitly represented, e.g., by Boole, Peirce, Schröder, Löwenheim, Gödel, Tarski, Montague, etc. Not all of them have been prepared to apply the idea of language as calculus to the entire colloquial (natural) language, however. One of the most important consequences of the view of language as the universal medium is the uniqueness of our language and of its interpretation. All that language is good for on this view is to enable us to talk about this world. We cannot (unless a great deal of further explanation is given) use language to discuss other possible worlds. Hence all possible-worlds theory is impossible on the view of language as the universal medium. For instance, logical truths cannot be defined as propositions true in every possible world a la Leibniz or later Carnap. On this view, as Russell once put it, "logic is concerned with the real world just as zoology, though with its more general and abstract features." Thus the very possiblity of possible-worlds semantics depends on some version of the assumption of language as calulcus. A closer analysis of the historical origins of possible-worlds semantics would in fact confirm its place in the tradition of language as calculus. 3. A perspective on possible^warlds semantics But the contrast between the competing ideas of language as the universal medium and language as calculus does not only enable us to place possibleworlds semantics on the map of the history of twentieth-century philosophy. It

Long after I proposed the term "language as the universal medium" I discovered that this phrase had been used with essentially the same meaning by Gadamer in his Wahrheit und Methode (fourth ed.), Mohr, Tübingen, 1975 (first ed., 1961). The role of this assumption in Gadamer and Heidegger is being examined by Martin Kusch (forthcoming).

Exploring Possible Worlds

55

yields insights into the nature and potentialities of possible-worlds semantics itself. One of them results directly from the basic idea of language as calculus on which PWS was seen to rest. If we really are free to re-interpret our language, we can choose freely also the "universe of discourse" it is designed to apply to. This universe hence does not have to be an entire world in the commonplace sense of the word (i.e., a possible world history). It can be a "small world", that is, a relatively short course of local events in some nook or corner of the actual world. Hence the idea, which seems to enjoy considerable currency, that possible-worlds semantics somehow presupposes a Leibnizian framework where the alternative are entire grand universes,6 is not only mistaken but contrary to the way of thinking on which possible-worlds semantics is based. For instance, the alleged basic contrast between possible-worlds semantics and "situation semantics" is fictional.7 One reason why this observation is important is that several oi the prima facie problems that arise in possible-worlds semantics are easier in theory and in practice to deal with when the different worlds in question are merely alternative scenarios concerning some one medium-sized facet of the big world. The all-important problem of cross-identification is a case in point.8

4. Possible-worlds semantics and game-theoretical semantics Another consequence of the same historically motivated line of thought is also useful in developing further possible-worlds semantics. Part of the uniquenessof-interpretation idea which is inherent in the conception of language as the universal medium is that the interpretation of logical concepts, e. g., quantifiers, is fixed, mediated by certain eternal and immutable relations. For instance, on this view quantifiers are really second-order constants. Their extensions are the only things that really enter into discriminations of truth and falsity of different sentences. And these extensions are of course determined once and for all. This static view is unsatisfactory in several ways. Once again, antidotes are offered by the idea of language (and logic) as calculus. It suggests that the interpretation of logical concepts like quantifiers can also be systematically

6

7 8

Since this idea is a hangover from the unspoken assumption of language as the universal medium, it is difficult to establish its locus classicus or even to find a fully explicit statement of it. It is a mere prejudice, however. Cf. here my note, "Situations, Possible Worlds and Attitudes", Synthese vol. 54 (1983), pp. 153-162. For the cross-identification problem, cf. Jaakko Hintikka and Merrill B. Hintikka, "Towards a General Theory of Individuation and Identification", in W. Leinfellner et al., editors, Language and Ontology: Proceedings of the Sixth International Wittgenstein Symposium, Holder-Pichler-Tempsky, Vienna, 1982, pp. 137-150.

56

J. Hintikka

varied. This suggestion is taken up by the recently developed theory of abstract logics (abstract model theory). But it may be argued that such a variation should be based on a closer analysis of how quantifiers really operate. Now perceptive logicians realized long ago that the real subtlety and power of quantifiers is largely due to the interplay of quantifiers,9 more specifically, the way in which nested quantifiers depend on outer ones, for instance, the way in which the value of y depends on the value of in (4.1)

(vx)(3y) S[x,y].

The idea of logic as calculus suggests that these dependencies be spelled out explicitly so that they can be theorized about and maybe varied in suitable ways. This is precisely what happens in the theory I have called game-theoretical semantics (GTS).10 Its basic idea is very closely related to that of the so-called functional interpretations launched by Gödel.11 In such an interpretation, the choice functions involved in sentences like (4.1) (known in the trade as Skolem functions) are represented by explicit function variables and themselves quantified over. Thus, e.g., (4.1) has as its functional interpretation (4.2)

(3f)(Vx) S[x,f(x)]

This can also be taken to express the meaning GTS assigns to (4.1). Thus the suggestion yielded by the historical background of possible-worlds semantics is that it should join forces with game-theoretical semantics. Both are aspects of the same general approach, and hence can be expected to be well suited not only to complement each other but to join forces. Roughly speaking, the idea is this: PWS is the static theory of possible worlds, whereas GTS is the abstract theory of possible-world exploration. For the activities included in the moves in the semantical games of GTS are intuitively speaking of the nature of seekings and findings, in Ryle's phrase, of the nature of moves in a "game of exploring the world". What I shall do in this paper is to illustrate the prospects of the cooperation between PWS and GTS by showing what applications one crucial idea of GTS, viz. the idea of informational independence (in the sense of game theory), has in PWS.

Cf. here especially Warren Goldfarb, op. cit. (note 4 above). For GTS, see in the first place Jaakko Hintikka, The Game of Language, D. Reidel, Dordrecht, 1983 (with a bibliography). See here Kurt Gödel, "On a Hitherto Unexploited Extension of the Finitary Standpoint", Journal of Philosophical Logic, vol. 9 (1980), pp. 133-142 (with a bibliography). This is a translation of "Über eine bisher noch nicht benützte Erweiterung des finiten Standpunktes", Dialecticavol 12 (1958), pp. 280-287.

Exploring Possible Worlds

57

5. Game-theoretical semantics My main thesis in this paper is therefore that there is another approach to semantics that can usefully complement PWS for several purposes. This approach is game-theoretical semantics. Needless to say, I cannot here present a full exposition of the principles of GTS. I will have to assume that you are sufficiently acquainted with it, and merely remind you of its basic ideas. The merits of GTS are most clearly in evidence when it is used to develop a semantics for certain important aspects of natural languages, such as quantifiers, anaphoric pronouns, negation, etc. However, the basic ideas of GTS are explained most easily when they are used to provide a semantics for formal but interpreted first-order languages. This can be done as follows: Since we are dealing with an interpreted language L, we must be given a model M, i.e., a domain of individuals D(M) on which all the nonlogical constants of L are interpreted. This suffices to assign a truth-value to each negated or unnegated atomic sentence of L, for instance in virtue of the usual Tarski-type truth-definition for atomic sentences of L (or of any extension of L obtained by adding to it a finite number of names of the members of D(M)). The notions of truth and falsity are extended to complex sentences S of L by associating with each of them a two-person zero-sum game G(S), called a semantical game. The players may be called Myself and Nature. From the vantage point of Myself, this game is an attempt to verify S, while Nature is trying to falsify it. From this idea, the rules of G(S) are easily obtained: (G.v) The game G((Sj v 82)) begins with a choice by Myself of S; (i = 1 or 2). The rest of the game is G(S;). (G.&) The game G((St & S2)) begins with a choice by Nature of S; (i = 1 or 2). The rest of the game is G(S;). (G.E) The game G((3x)S[xj) begins with a choice of a member of D(M) by Myself. If a proper name of the chosen individual is "b", the rest of the game is G(S[b]). (G.U) The game G((Vx)S[x]) begins with a choice of a member of D(M) by Nature. If a proper name of the chosen individual is "b", the rest of the game isG(S[b]). In the last two rules, "b" can be either in L or a new name given to the chosen individual by the players. (G.~) The game G(~ S) is like G(S) except that the roles of the two players (as defined by these rules) have been exchanged. (G.A) If A is atomic, Myself wins G(A) iff A is true, otherwise Nature wins. These games serve their truth-defining purpose in virtue of the following stipulation:

58

J. Hintikka

(T) S is true iff there exists a winning strategy for Myself in G(S). S is false iff there exists a winning strategy in G(S) for Nature. This truth-definition is the true gist in the old idea that a sentence is true if it can in principle be verified. The way in which semantical games can be defined for natural-language sentences has been explained repeatedly elsewhere.12 In this paper, I shall typically deal with natural-language sentences via their translations into formal first-order or higher-order languages.

6. Possible-worlds semantics and game-theoretical semantics combined I shall first remind the reader how GTS can be extended to the situations dealt with by PWS. Indeed, this extension of GTS could not be more obvious. All that is needed is to recall the basic ideas of PWS adumbrated in the beginning of this paper. If necessary truth means truth in all alternative possible worlds, then the necessity-operator is like a universal quantifier, and has to be treated like one in GTS: it marks Nature's choice of an alternative possible world, and analogously for other notions. In order to implement this idea, a couple of small changes in the GTS framework are in order first: (i) A semantical game is not played on a single model, but on a space of models on which suitable alternativeness relations are defined. (ii) The game G(S0) connected with a sentence S0 starts from the world w0 in which S0 is being evaluated. At each move, the players are given not only a sentence Sj, but a world Wj. Now it is easy to see how the game rules for modal and intensional concepts are to be formulated. I shall formulate them for a simple version of epistemic language whose logical vocabulary consists ofthat of a first-order quantification theory (without function symbols) plus an operator "Ka" having (roughly) the force of "a knows that". (G.K) The game G(KaS), when starting from w0, begins with Nature's choice of an epistemic a-alternative Wj to w0. The rest of the game is G(S) starting from Wj.

Here epistemic a-alternatives to w0 are of course all the worlds compatible with what everything a knows in w0. (G.~K) The game G(~KaS), when starting from w0, begins with a choice by Myself of an epistemic a-alternative Wj to w0. The rest of the game is G(~ S) starting from Wj. 12

Besides Jaakko Hintikka, The Game of Language (op. cit.), see Esa Saarinen, editor, GameTheoretical Semantics, D. Reidel, Dordrecht, 1979.

Exploring Possible Worlds

59

For other modal and intensional operators, similar game rules can be formulated. It turns out, however, that combining quantifiers with intensional notions like knowledge necessitates changes in the quantifier rules and also in the treatment of negation. These important changes will not be discussed here. They are not crucial for my purposes in this paper. 7. Game-theoretical semantics and informational independence The proposed cooperation of combining PWS and GTS yields interesting new possibilities of theorizing, not the least in the direction of natural-language semantics. These advantages are not coincidental, either. GTS differs from most of the other recent truth-conditional approaches to semantics in that in it the evaluation of the truth-value of a sentence proceeds from outside in, while in approaches like Montague semantics which rely on Tarski-type truth definitions one proceeds from inside out. This implies that such approaches cannot take into account semantical context-dependencies, whereas GTS (and PWS combined with GTS) can do so.13 This opens to us a wide field of applications for PWS which could not be handled earlier. Perhaps the most important class of such applications is due to the possibility of informational independence between different moves in the semantical games GTS is based on. It is the role of informational independence in PWS that I shall study in this paper. I pointed out the presence of informationally independent quantifiers in natural languages in 1974.14 Lauri Carlson and Alice ter Meulen showed in 1979 that there are informational independencies between quantifiers and intensional operators.15 Yet I suspect that most linguists still tend to think of informational independence as a marginal phenomenon which does not have to be taken into account in systematic logico-linguistic theorizing. This is a mistake, I strongly believe, due to the fact that linguists have paid attention only to those cases where a branching-quantifier sentence (or a sentence involving other kinds of informational independence) does not have any translation into the old-fashioned linear (independence-free) notation. As I pointed out in my very first paper on the subject, such cases do not by any means exhaust the relevance of the phenomenon of informational independence. Rather, taking this phenomenon into account can help us to discover and to formulate important regularities which cannot be captured otherwise. This conjecture is oeing verified in a convincing manner by a closer study of those very areas of semantics which Cf. here the last essay in The Game of Language, op. cit. (note 10 above). Jaakko Hintikka, "Quantifiers vs. Quantification Theory", Linguistic Inquiry, vol. 5 (1974), pp. 153-177. Lauri Carlson and Alice ter Meulen, "Informational Independence in Intensional Context", in Esa Saarinen et al, editors, Essays in Honour of Jaakko Hintikka, D. Reidel, Dordrecht, 1979, pp. 61-72.

60

J. Hinrikka

PWS is calculated to deal with. What it means is that the combination of PWS and GTS really opens the door for important new theorizing in the large scale. The main aim of this paper is to illustrate this fact. But first I must remind you of the salient features of the phenomenon of informational independence. In its simplest form, this phenomenon mainfests itself in the form of partially ordered quantifiers. There a quantifier Q2 depends informationally on another quantifier Qj iff Qj > Q2 in the partial ordering. In the usual notation, this means that Qj is to the left of Q2 in some branch of the (partially ordered) quantifier prefix. This is illustrated by such sentences as (7.1)

(Vz)(3xW

or

(7.2)

Another way of spelling out dependencies and independencies among quantifiers is to translate each first-order sentence S into higher-order notation. This translation tr(S) simply says that there exists a winning strategy in the game G(S) connected with the original first-order sentence S (as far as quanrificational moves are concerned). All we need for the purpose are quantifiers ranging over choice functions (Skolem functions). Thus (7.1) translates as (7.3) (3f) (Vz) (Vy) S[f(z),y,z] and (7.2) translates as (7.4)

(3f)(3g)(Vx)(Vz)S[x,f(x),z,g(z)].

It is known that the logic of partially ordered quantifier sentences cannot be axiomatized.16 The examples I have given suffice to give the reader an idea of the notation for informationally independent quantifiers and of their semantics. 8. Possible-worlds semantics and informational independence How is this phenomenon of informational independence manifested in the realm of PWS, e. g., in the realm of the prepositional attitudes like the logic and semantics of "a knows that", in brief "Ka"? (For simplicity, I shall often drop in

In "Quantifiers vs. Quantification Theory" (note 14 above) I showed that the decision problem for the entire second-order logic reduces to that of the theory of branching quantifiers. This implies that no complete axiomatization of the theory of partially ordered quantifiers is possible.

Exploring Possible Worlds

61

the following the index indicating the knower.) For instance, what do branching expressions like the following mean?

(8.1) The import of (7.1) is immediately clear on the basis of GTS. In the game connected with (7.1), Myself chooses an individual, say b, and independently of that choice Nature chooses an epistemically a-alternative world. In the rest of the game, Myself will have to defend S[b]. But since Myself s choice is independent of Nature's, it might as well have been made before Nature's move. Hence (7.1) is equivalent with (8.2)

(3x)K a S[xJ.

Thus informational independence does not always give rise to structures which cannot be represented in conventional notation. But even when a non-linear notation - or some other equivalent notation - is not indispensable, it can be useful. In order to illustrate this, let me first extend my notation. Clearly, the interpretation of a proper name or a primitive (nonlogical) predicate can take place independently of an epistemic operator. Indeed, I have elsewhere argued at length that the interpretation of a proper name must take place in a semantical game by means of a special game rule which is subject to most of the usual ordering principles of GTS. An application of this rule may or may not be independent of the rules applying to certain other ingredients of a sentence. Likewise, the interpretation of other nonlogical constants can be thought of as being governed by a special rule which can be independent of applications of other rules. Then the independence of a concept of an intensional operator, e. g., of "K", can be expressed simply by writing "/K" after the symbol for this concept. Thus we can have expressions like the following: (8.3) (8.4) (8.5) (8.6) (8.7) (8.8)

K (3x/K)((A/K)(x) & A(x)) as compared with K (Vx/K)(A/K)(x) ID A(x)) K S[b/K] as distinguished from KS[b] K (Vx)(3y/K) S[x,y] as distinguished from (Vx)((3 2 )(x=2&(3y)KS[2,y]).

62

J. Hintikka

9. Independence and the logical form of wh-questions The expressions (8.3)-(8.8), and others like them, call for a number of general comments. For instance, (8.3) and (8.4) are the existential and the universal reading, respectively, of a knows + w^-construction.17 They are equivalent with the older formulations of these two readings, which are (9.1)

(3x)KA(x)

which can also be written as (9.2) (3x)(A(x)&KA(x)) and

(9.3) (Vx)(A(x) D (3z)(x=z & KA(z))). In the sense brought out by these equivalences, the use of informationally independent concepts does not give us anything in this case which would not be expressible in the usual notation. However, the new notation which allows informational independence is much closer to natural language than the customary one and hence brings out better the actual way in which natural languages like English operate semantically. For we can raise all sorts of awkward questions about (9.1) and especially about (9.3). Why is the second quantifier (3z) needed in (9.3), when there is no trace of it in the corresponding naturallanguage sentences like (9.4) It is known who has the property A and when it is not needed in the apparently parallel expression (9.1)? And even though an answer can be given in terms of the ways in which quantifiers operate in the context of epistemic operators, more particularly, in terms of the admissible substitution-values of variables by means of which we "quantify in" a context governed by epistemic operators, and even though sucn restrictions have a well-defined and well-motivated model-theoretical basis, even then the question arises how natural languages like English can get away with just one quantifier, if that feat is not possible (as it is not) in a model-theoretically backed formal language of the conventional sort. In fact, I suspect that the distance between my earlier analysis (9.3) of the universal-quantifier reading of subordinate whquestions and what we find in natural languages has been operative in discouraging linguists and philosophers from adopting my analysis, even though it is the only possible analysis model-theoretically and even though it is logically equivalent with the independent-quantifier formulation (8.4). All these questions are answered by noting the possibility of formulating the two readings as (8.3) and (8.4). For one thing, we can from them see how each of See here Jaakko Hintikka, The Semantics of Questions and the Question of Semantics (Acta Philosophica Fennica, vol. 28, no. 4), Societas Philosophica Fennica, Helsinki, 1976.

Exploring Possible Worlds

63

the two readings of subordinate wh-constructions can be expressed by means of just one quantifier. It is also instructive to have a look at the intuitive meaning of (8.3)-(8.4). Such a look shows how closely (8.3)-(8.4) match the intuitive idea which (8.4) is calculated to express. For what (8.4) says is literally that each actual instance of A is known to be so, and (8.3) says the same of some actual case of A. It is hard to see how one could bring one's formal language closer to our natural ways of thinking and speaking than this. And the possibility of expressing these intended ideas as simply as in (8.3) and (8.4) is due to the use of inrormationally independent quantifiers. In fact, there is a related a priori reason to think that the quantifiers, that interrogative wh-words in effect are, must normally be taken to be independent of K-operators. For what that independence implies is that the values of such quantifiers (or of variables bound to such quantifiers) be members of the actual world. These plainly are what our knowledge is about. In the simplest cases, this actuality can be captured simply by placing the quantifiers in question before the K-operator in a linear prefix. Fiowever, when such a wh-quantifier depends on another quantifier which in turn depends on the K-operator, no linear sequence of quantifiers and operators captures the intended meaning. This will in fact be the subject of the next section. 10. Questions with an outside quantifier In the case (9.4), its two readings (8.3) and (8.4) can after all be expressed without the use of independence assumptions as (9.1) and (9.3). However, in more complicated cases, there are readings of subordinate questions with "knows" as the main verb which cannot be expressed in the old independencefree notation. Consider the following example: (10.1)

It is known to whom everybody bears the relation S,

for instance, (10.2)

It is known whom everybody admires.

Here we are disregarding the reading on which the existential quantifier precedes the universal one. It can be expressed by (10.3)

(3y)K(Vx)S[x,y].

This leaves (8.8) as the only other reading that can be formulated in the old independence-free notation. But, as I have shown elsewhere, (8.8) is not the most natural reading of (10.1).18 Cf. here Jaakko Hintikka, "Questions With an Outside Quantifier", in R. Schneider, K. Tuite and R. Chametzky, editors, Papers from the Pamsession on Nondeclaratives, Chicago Linguistics Society, Chicago, 1982, pp. 83-92.

64

J. Hintikka

This is reflected by the higher-order translations of (8.7) and (8.8). The former is equivalent with (10.4)

(3f)K(Vx)S[x,f(x)].

This is in line with what has been said. To know whom everybody admires is to know what the function is which assigns to everybody a person he or she admires. This function can be known without knowing the individuals between whom it obtains. In contrast, (8.8) cannot be represented as (10.4). The reason is that when a function quantifier quantifies in an epistemic context, as in (10.4), its values are functions independent of the choice associated with "K", i.e., having the same value for the same argument in all the epistemic alternatives. This is the case for a suitable f if (8.7) is true, for the choice of the value of "y" is independent of the choice of an epistemic alternative associated with "K". But the truth of (8.8) guarantees only that a suitable function is defined in the actual world; it does not have to be invariant with respect to the change from one alternative to another. Hence (8.8) cannot be represented as (10.4). It has to be translated by something like (10.5)

(3f)(Vx)(3z)(3u)(x=z & u=f(x) & K S[z,u]).

From this representation it is seen that for (8.8) or (10.5) to be true, it has to be known who it is that each person admires in the sense of knowing the identity of each admirer and of at least one of his or her admired ones. This is not required by (10.4) or (on its most natural reading) (10.2). Collateral evidence is available to the effect that a reading like (8.7) or (10.4) is required to understand the behavior of questions in natural languages. However, in the existing discussions (by other logicians and linguists) of the different readings of sentences like (10.2), only the higher-order representation is used. It is not explained outside GTS how natural languages manage to express this second-order force by means of what looks and behaves like a first-order quantifier. An explanation is, of course, provided by the equivalence of (8.7) and (10.4). Thus in the case under scrutiny informationally independent quantifiers are required to represent the semantics of natural-language if we want to do it in the same way as in natural languages, i. e., without overt recourse to second-order quantifiers. As I have noted elsewhere,19 sentences like (8.7) cannot even be represented in a branching notation, but need a closed-loop configuration like the following: (10.6) K / \ (3y)-(Vx)

) S[x,y]. )

This is obviously equivalent with (8.7) and (10.4). "Questions With an Outside Quantifer", op, cit. (note 18 above).

Exploring Possible Worlds

65

11. De dicto vs. de re The contrast between (8.5) and (8.6) exemplifies another general semantical phenomenon. The oüstinction between (8.5) and (8.6) is easily generalized to all occurrences of a singular term like "b" in a context governed by one or more intensional operators. This contrast can be used to put into perspective one of the most confused subjects of recent logico-philosophical discussion. This subject is the distinction between the so-called de dicto and de re readings of ordinary-language sentences containing a noun phrase in a context governed by intensional words. At first sight, nothing could be simpler to explain than this distinction is. Take, for instance, the sentence (11.1)

Sherlock Holmes knows that the murderer is left-handed.

In my formal jargon, this can (on one of its readings) be abbreviated as (11.2) KshL(m) where L(x) is is left-handed-ana m = the murderer. The knowledge attributed to S.H. in (11.1) can pertain either to whoever the murderer is or may be, for all that the famous sleuth knows, or else to the person who in fact committed the crime, no matter whether Holmes is actually aware of his having done the deed or not. In the former case, the clues Holmes has might, e. g., be such as to leave no doubt of the sinister perpetrator's manual preferences while leaving his or her identity open. Something is true of a person, in other words, according to what Holmes knows, as soon as a certain dictum (here: "the murderer") applies to him or to her. In the latter case, Holmes's knowledge pertains to a certain res (here, an actual person) known to him, and the attribution of the murder to that person is made by the speaker, not by Holmes. Now this distinction is immediately and completely captured in PWS. In terms of the possible-worlds analysis of knowledge, the distinction is one between something's being true of the (possibly several) individuals which in different epistemic alternatives have the property being the murderer, and the same thing's being true in these worlds of that unique individual who in the actual world is the murderer. In other words, in the de dicto case we so to speak look at the individuals who in the several epistemic alternatives satisfy a condition ("dictum"), while in the de re case we follow the world line of the res who in the actual world satisfies this condition. This contrast is obviously captured in the usual old-fashioned PWS notation by the distinction between (11.2) and (11.3)

(3x)(x = m&K sh L(x)).

The surprising thing here is not that this analysis can be given so simply in PWS, or that it captures so completely the intended distinction. The astounding thing is that twenty years after the genesis of PWS, many philosophers are still

66

J. Hintikka

discussing this distinction as if there were something further to explain. Discussions of the de dicto - de re contrast have in my view been mostly an exercise in gross scholastic futility. All that is worth saying about the distinction seems to be capable of being said in two pages. Yet I have come to realize that philosophers who have been engaged in those boring and fruitless discussions have had something in mind which is worth serious attention. This something is not just to try to capture the distinction somehow in a formal language. Rather, the philosophers and linguists who have wasted their time on the distinction have been misled by a prevalent but mistaken methodological idea. This idea is the assumption that all our grammatical, semantical and even logical (in the sense of Spracblogik) concepts, distinctions and regularities must somehow be capable of being extracted from particular natural-language examples. This assumption is seldom propounded explicitly, but it has clearly been extraordinarily influential in different philosopnical traditions, including both Oxford-type "ordinary language" philosophy and the philosophers who derive their inspiration from contemporary theoretical linguistics. This assumption has done many kinds of damage, not the least in areas like PWS which is basically a theory of what language is about and whose basic concepts are not necessarily all reflected by explicit linguistic indicators. (An important case in point is the distinction between indexical and descriptive identification.)20 In the case of the de dicto - de re distinction, what has happened is that the linguistic mechanism through which it is expressed is far from obvious. Even though the underlying semantical situation is clear, as is shown in the light of hindsight by the contrast between (11.2) and (11.3), philosophers and linguists have been frustrated in their attempts to find the distinction somehow among the bag of tricks of actual natural languages. It is indeed impossible to find in an English sentence like (11.1) any traces of the quantifier occurring in the representation (11.3), even though this formula obviously captures one of its readings. In other words, while the contrast exemplified by (11.2)-(11.3) shows how the de dicto - de re distinction can be represented in the formal language of intensional logic, this representation does not explain how the distinction is handled in natural languages. Here the observations based on the phenomenon of informational independence again show their usefulness. It is immediately seen that (11.3) is logically equivalent with (11.4) KshL((m/K). Likewise, (8.5) is equivalent with (11.5) 20

(3x)(x=b&KS[x]).

This contrast is discussed in chapters 3-4 of Jaakko Hintikka, The Intentions of Intentionality, D. Reidel, Dordrecht, 1975.

Exploring Possible Worlds

67

Such equivalences show that the de re reading can be understood as an informational-independence phenomenon. More precisely expressed, the de re reading is the one in which the operative singular term is to be evaluated independently of the intensional operator within the scope of which it occurs. (This operator may be tacit. Typically, there is, e.g., an implicit sentence-initial epistemic operator in normal assertive utterances.) This treatment of the de dicto vs. de re distinction presupposes that the terms to which the distinction applies, including proper names, are dealt with by means of a game rule which assigns a value to such a term in the course a semantical game. This might at first seem surprising, but is in reality well backed up by evidence.21 My interpretation of the de re reading as an independence phenomenon is supported by the light it throws on a variety of other phenomena. First and foremost, we can understand why there is no trace of a quantifier in natural-language sentences like (11.1), in spite of the fact that they admit the reading (11.3). What they in reality admit of as their "logical form" is (11.4), which does not contain any quantifiers. The fact that the possible independence of a singular term (singular noun phrase) of an epistemic operator is not marked in natural language is an instance of a much wider uncertainty. While certain particular constructions in English require or at least strongly favor informational independence, in natural languages like English the informational independence of a singular NP or a quantifier of an epistemic operator is not marked in a uniform way. This explains largely the inconclusiveness of the earlier de dicto - de re discussions, assuming that their tacit agenda was to reveal the mechanism of the contrast in natural languages. There simply is not any uniform mechanism for the purpose in English.22 12. Backwards-looking operators vs. informational independence Here the discussion of the de dicto - de re contrast begins to merge with the study of another class of semantical phenomena. Their clearest exemplifications in natural languages are probably the so-called actuality-operators.23 An especially conspicuous case in point is probably the behavior of "now" in tense-logic. Suppose I say, (12.1) Ten years ago I could not have believed that I should now be living in Tallahassee. 21 22

23

See Jaakko Hintikka and Jack Kulas, Anaphora and Definite Descriptions, D. Reidel, Dordrecht, 1985, pp. 25, 59, 88-89, 159, 170-171, 187-188, and 209. This does not mean however, that certain particular English constructions do not strongly encourage the independence reading, perhaps even filter out others. This matter needs further investigation. For the entire complex of questions concerning actuality operators, backwards-looking operators, etc., see Esa Saarinen's own contribution to Saarinen, Game-Theoretical Semantics, op. cit. (note 12 above).

68

J. Hintikka

Then in the clause "I should now be living in Tallahassee" which describes my beliefs ten years ago, the word "now" does not refer to the time ten years ago, but tot the "now" of now - that is to say, to the moment of time with respect to which (12.1) is being evaluated, not to the time reached in the course of the evaluation process. This illustrates the nature of actuality operators in general. What is peculiar to such "actuality operators" can be explained concisely by reference to the interpretation process of a sentence S that the game G(S) in effect is (in the sense that appears from the earlier sections of this paper). In the usual old-fashioned intensional languages, this is always a one-way procedure. Each intensional operator marks a move from a world to one of its alternatives, but no rule enables us (or, rather, the players of a semantical game) to take a step in the opposite direction. Yet the peculiarity of an actuality operator is precisely that it forces us to move against the current. A sentence - say, Si - which contains an actuality operator and which is being considered at some stage of a semantical game at a world wt can only be evaluated by knowing the world w0 from which the entire game started. For instance, in the game connected with (12.1), the word "now" occurs at the later stages in subsentences which are being considered in relation to a time tj different from the time t0 at which (12.1) is being evaluated (verified or falsified) and from which the semantical game started. Yet the intended time is always t0 and not tj. Hence the players have to be able to return to time t0. In this tense-logical example, different moments of time play the role of different worlds. In general, the players of the semantical game (and a fortiori the hearer of the original sentence) must be able so to speak to buy a return ticket to the original "world" or at least to "remember" it. There have been two main attempts to account for the phenomena illustrated by actuality operators. One of them is, in its simplest form, to change the metatheoretical (semantical) framework so that the relevant rules refer, not only to the world at hand (so to speak), but also to the world from which the evaluation process started. This formulation pertains of course directly only to a game-theoretical formulation of truth-conditional rules, but the same idea can easily be formulated in other frameworks, too. Theories of this kind are often referred to as two-dimensional (or multi-dimensional) semantics of intensional logics. It has been argued that this treatment does not explain all that has to be explained. For instance, a proposition can have a reading, it has been claimed, on which the players of a semantical game have to "remember", not just the initial world, but also some of the intermediate worlds through which the game had passed before it reached the stage at which the operative phrase is evaluated. Be this as it may, such alleged impossibilities have provided inspiration and ammunition for another approach to the same problems. In it, the semantical framework is kept intact, as there indeed seems to be ample reasons to do. Instead, the object language itself is enriched by introducing what are usually called backwards-looking operators. What such an operator does is to undo a

Exploring Possible Worlds

69

specified earlier step from one world to another in a semantical game. For instance, DK will undo the transition prescribed by K. As an example, consider the following proposition: (12.2)

Thomas believes that there are hippogriffs, but at least one of them actually is not a hippogriff.

This has the following formal translation in terms of backwards-looking operators:

(12.3) B™omas (3x)(H(x) & DBlliomas ~H(x)) In terms of an actuality-operator AC it can also be written: (12.4) Rn^ (3x)(H(x) & AC~H(x)) Obviously, (12.3) and (12.4) are equivalent. The idea of backwards-looking operators is a neat one, and results in a much more satisfactory formal semantics than the multi-dimensional analysis. Unfortunately, there is a strong objection to backwards-looking operators as a part of the logical analysis of natural language: in natural languages like English, there are few actually expressed backwards-looking operators, and those that there are, cannot do the whole job for which backwards-looking operators were in the first place hired. Hence we have to postulate tacit operators if we want to save the backwards-looking operator theory. But such postulation is arbitrary, and unsupported by any collateral syntactical evidence. This is especially clear in examples where repeated uses of the backwards-looking operator idea would be needed to explain their meaning. In fact, the job of any one backwards-looking operator, say DK, as applied to a formula S, can equally well be done by making the evaluation of all the different nonlogical primitives of S independent of K. For instance, instead of (12.3) or (12.4) we can equivalently write

(12.5) BThomas (3x)(H(x) & ~ (H/BT^XX)) This obviously says the same as (12.3) or (12.4). This example can obviously be generalized. By means of independence assumptions, we can likewise handle all the other examples for the treatment of which backwards-looking operators have been evoked. For instance, take Saarinen's example (slightly modified): (12.6)

Ed believes that Larry kissed some girl last night. Ed believes that she is pretty.

Saarinen's translation of (12.6) into his formal notation is (mutatis mutandis) (12.7)

BEJ (3 x) (x is a girl & Larry kissed χ last night & DBE , Bj^ (x is pretty))

where Bgj corresponds to "Ed believes that". In terms of the independence notation, this can be written as

70

J. Hintikka

(12.8) BEd (3 x) (x is a girl & Larry kissed χ last night & (BE/BEcj)(x is pretty)). Likewise, Saarinen considers Geach's problem sentence (12.9) Hob thinks that a witch has blighted Bob's mare and Nob believes that she killed Cob's sow, and offers as one of its readings (with obvious alternatives) (12.10)

BHob (3x)(W(x) & Bl(x) & DBHob BNob Kl(x)).

This corresponds to the independence reading (12.11)

BHob (3x)(W(x) & Bl(x) & (BNob/BHob)Kl(x)).

Moreover, the independence notation is obviously superior to the backwardslooking operator notation as a semantical representation of what is actually happening in natural languages like English. For in English there is in most cases nothing whatsoever that syntactically corresponds to a backwards-looking operator. Postulating them has therefore no basis in the grammar of English, however natural it may be model-theoretically. In contrast, the fact that the independence of an operator is not signalled in (12.6) or in (12.9) is only a special case of the more general regularity which I noted above (to the effect that the independence of operators and quantifiers normally is not signalled in English). Hence it can be claimed that (12.8) and (12.11) match much closer the grammatical mechanism of the corresponding English sentences than do (12.7) and (12.10). 13. Informational independence and higher-order logic At this point, a general remark may be in order concerning the role of informational independence in natural languages. It used to be said by many linguists that all natural languages are equally complete by way of expressive power. Indeed, it was frequently said or implied that in some unspecified sense every full-fledged natural language is optimal in this regard. Now statements of this kind pose all sorts of general questions. For instance, for a logician there simply is not such a thing as optimally rich language - rich, that is to say, by way of expressive power. But even apart from such very abstract issues, there is the interesting fact that, prima facie at least, natural languages seem to be first order languages. Indeed, the classical approaches to natural languages by such logically oriented philosophers as Quine consisted essentially of an attempt to paraphrase natural language expressions in formal first-order languages.24 And, on the syntactical

24

Cf. W. V. Quine, Word and Object, MIT Press, Cambridge, 1960.

Exploring Possible Worlds

71

face of things, natural languages do seem to operate much more like manysorted first-order languages than higher-order ones. But if so, if natural languages are essentially first-order languages, then they are logically speaking very weak. And this weakness is not just a matter of fancy abstract theory. It shows up, among other places, in the inferential power relevant to the computational efficiency of different computer languages. On closer examination, it in fact turns out that natural languages do codify sundry higher-order ingredients. But how do they do this? At first sight, there are few, if any, ingredients in a natural language like English which could import second-order power into it. Even those ingredients which, on the second sight, might seem to do so, for instance, quantification over properties, typically turn out to be operating much more like ingredients in a many-sorted first-order theory and not to involve any of the real power of higher-order concepts. Indeed, the ways in which higher-order logic makes its entrance into a natural language like English are rather subtle. It seems to me that among these ways there are the following two: (i) plural quantifiers; (ii) informational independence. I am not dealing with plural quantifiers here, only with informational independence. This phenomenon derives its general interest partly from the fact that it is one of the main ways in which natural languages raise themselves above the level of first-order logic. Here we can thus see a vindication of my general thesis that informational independence is not a marginal phenomenon in natural languages. It turns out to have an important general function in the overall logicosemantical mode of operation of actual natural languages. In order to illustrate this point, it is in order to continue my line of thought in section 11 above. One interesting fact about the de dicto -de re contrast is that it can be extended to expressions other than singular terms. For instance, a natural-language statement about all P's (i. e., all individuals having the property P) may be interpretable either as a statement about the individuals which in fact are P's or about the individuals which in each of several different worlds are Ps, just as in the case of a singular term. For instance, a statement of what Howard believes about all football players may pertain either to the individuals who in fact are football players, or else to all the individuals who in each of Howard's belief worlds play this game. The distinction can be represented in analogy with (8.5) as distinguished from (8.6) only by means of higher-order quantifiers. Thus (13.1) John believes that all P's are Q's can be taken to mean either (13.2) BJohn (Vx)(P(x) D Q(x)) (13.3) (3X)((Vx)(P(x) ~X(x))

or else &BJohn (vx)(X(x) =

72

J. Hintikka

This contrast is obviously an instance of the de dicto vs. de re distinction. However, in this case the distinction cannot be represented in the usual independence-free first-order notation. It can nevertheless be represented easily by means of my independencefriendly notation. For instance, (13.2) and (13.3) are equivalent to (13.4) and (13.5), respectively: (13.4) (13.5)

BJohn (Vx)(P(x) => Q(x)) BJohn (Vx)((P/BJohn)(x) => Q(x)).

This shows how the independence-indicating notation can do part of the same job as higher-order quantifiers. (It can be argued that this notation can do almost the whole job of higher-order quantifiers.) It also shows how simple and natural this extension of the de dicto - ae re distinction is. It is also relevant to note that in (13.4) - (13.5) no notation is used which does not have a counterpart in English, except the independence indicator which typically is not signalled in English, anyway. 14. Other manifestations of informational independence The ubiquity of informational independence (and hence of phenomena studied in game-theoretical semantics) as a vital ingredient of the semantics of natural languages is not exhausted by the examples I have discussed. I shall mention three further examples but not discuss them in any detail. One of them is the behavior of the English quantificational expression "a certain".25 It used to be a part of logicians' and linguists' folklore that "a certain" is supposed to be an existential quantifier which always has the widest scope. I have shown elsewhere, however, that this is not the case. Rather, "a certain" is characterized by its always preceding clause-initial epistemic operators (or being independent or them). Not surprisingly, the resulting treatment of "a certain" turns on the possibility of informational independence and in fact uses logical structures not unlike (8.7) and (10.4) above. The second example is within the field of PWS. One important but largely neglected kind of modal logic is the relativised modal logic briefly studied by Risto Hilpinen.26 There the basic idea is that the alternatives to a given world w0 may differ from w0 only in certain respects, for instance, with respect to a certain individual or a finite set of individuals. Then it deals what is "possible for that individual or those individuals". This idea can be considered as a special case of

25

26

See here my paper, "On the Semantics of 'a certain"', Linguistic Inquiry, vol. 17 (1986), pp. 331-336. Risto Hilpinen, "An Analysis of Relativised Modalities", in J. W. Davis, D. J. Hockney and W. K. Wilson, editors, Philosophical Logic, D. Reidel, Dordrecht, 1969, pp. 181-193.

Exploring Possible Worlds

73

the idea of informational independence. Those nonlogical constants which are not allowed to vary when one steps from a world to one of its alternatives, prompted by an intensional operator (e. g., necessity-operator), are simply to be taken to be informationally independent of this operator. Relativised modalities promise interesting applications, among other areas in the study of why-questions, as witnessed by the semantical differences between the following differently stressed why-questions: (14.1) Why did Tom resign his deanship last year? (14.2) Why did Tom (and not Dick or Harry) resign his deanship last year? (14.3) Why did Tom resign his deanship (and not any of his other jobs) last year? (14.4) Why did Tom resign his deanship last year (and not, e.g., the year before) ? By this time, it should not be difficult for the reader to begin to see how these different questions (and their differences) can be handled by means of relativised modalities. Finally, along the same lines as the projected treatment of why-questions, it seems to be possible to construct a more satisfactory theory of counterfactuals than those that are found in the literature. The basic idea is to construe counterfactuals as (physically) necessary conditionals, but let only some of the nonlogical constants depend on the necessity-operator. In this way, e. g., all reliance on measures of closeness between different possible worlds could be avoided. The difficulties ("ifficulties", as Charles Stevenson called them) in trying to pin a sharp logical structure on natural-language counterfactuals would then be explained by the general failure of English to signal informational independence.

GUNNAR OLSSON

On Doughnutting Discussion of Jaakko Hintikka's paper "Exploring Possible Worlds"

Dealing with modal logic is like eating a doughnut from the inside. It is to chew away at fluffy possibility and to digest it into precise and agreed-upon knowledge. In the process, the black hole of Wittgenstein's Tractatus is gradually pushed to its outer limits, for on Professor Hintikka's way towards silence, there is "nothing but the familiar idea of information as elimination of uncertainty". As the explorer learns to speak exactly about states of affairs that are other than the actual, then the known becomes credible, the believed plausible. The question is this: Can the modal logician be trusted? In the paper at hand, Professor Hintikka merges his own formulation of possible-worlds semantics with a novel approach from game-theory. His problem is that the number and complexity of possible worlds is too large to be grasped and comprehended; the wfPs or theorems produced by the syntactic machinery leave the interpreter with too much to believe. Any sane person would go crazy for less, for as Wittgenstein once put it: "If I were sometime to see quite new surroundings from my window instead of the long familiar ones, if things, humans and animals were to behave as they never did before, then I should say something like Ί have gone mad'; but that would merely be an expression of giving up the attempt to know my way about."1 It was as a prophylactic against this type of madness that Hintikka originally developed his model system and the semantics which come with it.2 The idea was to have the theorems doubly anchored, not only within the formal system but without it as well. Not even this double medicine proved strong enough, though. There remains an embarras de richesse, for the ideas of one man are not necessarily those of another. It is in this situation that Hintikka now suggests that game-theoretical semantics be applied as a brake on run-away possibility. The players of his two-person zero-sum game are named 'Myself and 'Nature'. The strategy is to win atomic sentences, well noting that the players are dealt not one 1 2

Wittgenstein, Ludwig: Zettel. Berkeley: University of California Press, 1970, p. 393. Hintikka, Jaakko: "The Modes of Modality", Acta Philosophica Fennica, Vol. 16,1963.

On Doughnutring

75

card at a time but actually whole worlds of sentences. 'Myself' wins G(A) if A is true, otherwise 'Nature' wins. The point of the game is to discover the inconsistency of a sentence S, i. e. to elucidate the structure of logical (deductive) reasoning itself. Despite assertions to the contrary, I doubt whether game theory significantly will reduce the impossible number of possible worlds. Anyone who tries to evaluate them runs the risk of going mad. It is this potential schizophrenic that the Professor now prescribes therapeutic treatment in Monte Carlo and Las Vegas. My peculiar questions are not meant to divert interest away from the technical aspects of Hintikka's work. Indeed the opposite. For it is only by examining the minute details of the taken-for-granted that we can hope to unravel the governing techniques of social cohesion. And since logical reasoning is the most formalized and hence exaggerated and caricatured form of social practice, it will perhaps yield its secrets more easily than other types of discourse. In this regard, the concept of deduction is similar to the concept of money. What unites the languages of logic and money is that both of them are operationalizations of the concept of promise, hence of trust and hence of predictability. In their apparently different worlds, the Wall Street Journal and the Journal of Symbolic Logic are in fact serving the same dual function of evaluating the current value of past commitments. Indeed I cannot imagine a possible world which does not contain the notion of promising as one of its integral ingredients. My first problem concerns the desire of consistency and thereby the moving force of deductive reasoning. The accepted approach is of course to produce a chain of arguments in which the truths of the premises are preserved in the conclusions. One proposition is perfectly translated into another as everything which follows is parasitic on whatever went before. Understanding is consequently a matter of expressing the new in the language of the old for to know is to specify a translation function. If an error creeps into this reformulation process, then it shows up as a false conclusion; the innocent baby has been turned into a degenerate criminal, to be detained, treated and corrected. Who are the wardens in the prison hospital of crazy conclusions? How do I recognize its inmates? On this rendering, logical discourse becomes a species of moral discourse. Deduction proves itself to be an act of obedience, a submission under sets of institutionalized rules. The conclusions turn out to be predictable not because they are inherently true, but because they reflect the taken-for-granted of social practice. Intention has little to do with it, for it is well established that obeying a rule is to follow it blindly. There can never be surprises. And thus it is that truth and society come together in self-referential yet evolving forms of life. In Wittgenstein's words: "logic is not a field in which we express what we wish with the help of signs, but rather one in which the nature of the natural and

76

G. Olsson

inevitable sign speaks for itself. If we know the logical syntax of any signlanguage, then we have already been given all the propositions of logic".3 The immensity of possible worlds seems to be a function more of syntax and pragmatics than of semantics. At the same time, the habit of hooking formal reasoning to its premises appears as an insidious instance of a metaphysics of origin, a resort to the security of being and a flight from the fears of becoming. Logical deduction shows off as a kind of mythical narrative, proceeding from an axiomatic beginning to an inevitable end. The arrow of implication becomes a symbol of mimetic desire. It is not that the fantastics of counterfactuals and possible worlds are too much to believe,·* but that ΟgettingΟ rid of them leaves too L little to hang on to. No longer slowed down by the grapnel of causality, both individual and society would begin to drift like Flying Dutchmen on the open seas of imagination. Nietzsche was nevertheless correct: truths are illusions of which one has forgotten this is so. The gods of social cohesion always demand their sacrifices, in possible world semantics as well as in daily life. There is religion everywhere, for no society can survive without it. Even though truth-preservation is the stated goal of causal inference, practical inference is quite different. This is partly because the possibility of creating illusions, of cheating and deceiving is an integral part of language itself. And herein hides both the threat and the driving force to truth. A lie can be reasonable, the reasonable can be a lie. Plato knew it long ago. Hence his hatred of the theater. Derrida insists on it now: Truth is a woman, an umbrella on a dissection table. Temptation. Seduction. Who does what to whom, why, when and where? Jaakko Hintikka's performative readings of Descartes come to mind as well,4 for there are 'thoughtacts' just as there are 'speech-acts'. Are sayings like 'cogito, ergo sum' and 'to be or not to be' inferences or performances? In my mind they are both. The reason is that every utterance contains a crucial blending of self-reference and persuasion. There is indeed no description without performance, no thought without action. Telling truths is not enough. Being convincing is equally necessary. Even in the minimal semantics of a Donald Davidson or a Jacques Derrida, there remains a trace of speech act theory. Incredible truths are not truths at all, for truth is less an issue of what is the case and more of what is credible. Any narration is therefore a kind of thought experiment, an irresistable desire to produce possible worlds. The urge to express is linked with the urge to impress, the grasping of meaning with the evocation of meaning. Truth is not an immaculate conception, but stems from an unholy union of logic and rhetoric. Like everybody else, I would like not

3 4

Wittgenstein, Ludwig: Tractatus Logico-Philosophicus. London: Routledge & Keagan Paul, 1961, par. 6.124. Hintikka, Jaakko: "Cogito, Ergo Sum: Inference or Performance?" in W. Doney (ed.): Descartes. New York: Doubleday, 1967.

On Doughnutting

77

only to grasp truth but to be believed when I speak it. The trouble is that to be trustworthy is to be predictable. The deed and the doer are inseparable. Click in the lock. Imprisoned again behind the bendable walls of language I become uncertain about the distinctions between language as the universal medium on the one hand and language as calculus on the other. But high in the sky I glimpse a kite, seemingly free like a bird, yet in reality tied down with invisible strings of social convention. Once more, the human condition reveals itself as a struggle between certainty and ambiguity, necessity and possibility. Modal logic is entangled in it all; possible worlds are everywhere, not the least in the Sartrean concept of Nothingness. In turn, this raises the issue of whether it is possible to have an author who is authentic and legitimate at the same time. Should Professor Hintikka's exploration be read as a collective opinion issued by the High Court of modal logicians or as an individual's piece of postmodernist fiction? But even if a lion could talk, I could never understand him. And thus it is that the nature of consistency may show itself more clearly in tragedy than in logic. Already Sophocles seems to have known that tragedy concerns the relations between possible worlds semantics and ontology, between de dicto and de re. But he also knew that the dice of this risky game are loaded in favor of the touchable, the visible, the spatial. Groping in his blindness for his daughters' heads, he begs could I but touch them They would be mine again, as when I had my eyes. In tragedy as well as in logic, everything is right in the beginning. In the end there may be a difference. Both art forms are nevertheless social acts permeated by a dialectics of good intentions and obedient behavior. In reasoning there are axioms and logical constants, human artefacts solidified in social practice. In tragedy, it is all fate. How can I act otherwise? Is it dishonest to break the consistency of one's own rules? Is the modal logician a tragic hero? Is Sophocles the father of a valid theory of human action? "What therefore is truth", asked Nietzsche in his essay On Truth and Lie, and replied: "A flexible army of metaphors, metonymies, anthropomorphisms, in short, a sum of human relations, which have been poetically and rhetorically intensified, transformed, bejewelled, and which after long usage seem to a people to be fixed, canonical, and binding".5 "Too much", sounds the voice of possible-worlds semantics. "I shall show Myself to have the power of winning over Nature!" The game-theoretic approach raises issues about the techniques which Myself uses to convince Nature to change its mind and thereby lose its bet. Dangerous indeed. Since 'Nature' is not Nature but a pseudonym of 'Society', everything is Translated in Schacht, Richard: Nietzsche. London: Routledge Si Kegan Paul, 1983, p. 58.

78

G. Olsson

up for grabs. Not only the atomic sentence of Sj, but the whole world of Wj. Not only existential things, but individualized forms of life. Not only wife and children, but god himseE Consistency is the name ofthat game in which even consciousness must find a common language. But 'con-' and 'com-' are variants of the same prefix meaning 'together', i. e. 'social'. In this common context of conning, who is the croupier of game theory? How can he be trusted? Why play at all, if to be trusted is to be predictable? In every premise there is a promise of a just return. And in every promise there is a premise of proper conduct. Un coup de des, Mallarme! Champagne on the house! What is the strategy for winning: to be predictable or to be different? Is it more rewarding to mix genres than to stick to given categories? Is the optimal technique the Foucaldian one of reaching the heart of the matter by stretching the skin of its limits? Perhaps. For the act of convincing is never direct but always indirect. There is a connection yet radical difference between reality and the words in which it is caught and recreated. Although experience may sometimes be immediate, understanding is always one step removed, a presentation of a representation. Alienation. Could it have been this that Bertrand Russell had in mind, when he insisted that "every proposition and every belief must have an object other than itself'.6 Who knows what he believed? Perhaps the reply is that every word is in a sense forward-looking, waiting for the answer it prompts. For even in Tallahassee, Florida, Philosophy's Hall of Mirrors is now turned into a narcissistic echochamber, where all that is solid melts into air. As Max Black put it: "Perhaps every science must start with metaphor and end with algebra; and perhaps without metaphor there would never have been any algebra".7 Not surprisingly, W. V. Quine turns out to be an expert on metaphoric persuasion. Yet it was not a modal logician, but Stephen Dedalus, who "prove[d] by algebra that Hamlet's grandson is Shakespeare's grandfather and that he himself is the ghost of his own father".8 It is a function of metaphor to carry meaning from one level to another; to serve as a stereoscope of ideas; to create possible worlds; to be cognitive and creative at the same time. Theories are in fact metaphors of metaphors, all part and parcel of their own contexts. Kissing a Russian on the mouth is therefore not the same as licking the ass of a Swede. Moskovskaya is not Absolut. Indeed you choose your theory as little as your mate. Every discourse lives on its own boundaries where one word rubs its skin against another. Much written about metaphor. Much itself metaphoric. Deep roots in narration, therefore in metonymy. But stories evoke images and images are visible to 6 7 8

Russell, Bertrand: "Meinong's Theory of Complexes and Assumptions", Mind, Vol. 13, 1904, p. 204. Black, Max: Models and Metaphors. Ithaca, N. Y.: Cornell University Press, 1962, p. 242. Joyce, James: Ulysses. New York: Modern Library, 1934, p. 18.

On Doughnuttdng

79

the internal eye. One who saw was Georges Bataille, not the least in The Story of the Eye. Reading this marvel, it should be kept in mind that its leading character is not Simone or Marcelle, but the metaphor of the eye. In its search for meaning, this strange creature wanders from the socket of the face to the suck-it of the cunt. Tsch, tsch! What pushes it into novel forms is indeed form itself. The spheric eye is metamorphosed first into an egg and then into a testicle. And suddenly, "there in the sunlight, on Simone's seat, lay a white dish containing two balls, glands the size and shape of eggs, and of a pearly whiteness, faintly bloodshot like the globe of an eye".9 Immediately after the bullfight, under the sun of Seville, the narrator fuses together the killing of the bull, the loss of the toreador's eye and Simone's orgasm on the plate of testicles: "Thus, two globes of equal size and consistency had suddenly been propelled in opposite directions at once. One, the white ball of the bull, had been thrust into the 'pink and dark' cunt that Simone had bared in the crowd; the other, the human eye, had spurted from Grenaro's head with the same force as a bundle of innards from a belly. This coincidence, tied to death through a sort of urinary liquefaction of the sky, first brought us back to Marcelle in a moment that was so brief and almost substantial, yet so uneasily vivid that I stepped forward like a sleep-walker as though to touch her at eye level".10 In its own being, Bataille's story illustrates how possible worlds spring from a mating of precise words and ambiguous contexts. It seems significant that the Spanish term 'huevo' in its ovary carries the double meaning of 'egg' and 'ball'. Language which touches is neither blind nor insensitive for as words copulate, possible worlds are brought into being. The plot of the body. Nothing immaculate. Everything a self-referential play of ontological transformations. Words have their histories too. And thus it comes that the tremulous body is a means of meaning. The eye and the index finger become metaphors for grasping the distanciation inherent in all subject formation. My only contact with the world is through the holes of my body. It is through them that Myself is penetrated by the social norms of Nature. And vice versa. It is through them that you and I are made so obedient and so predictable. My body is my mode of existence. And yet: Most people remain proud of this little word T, that erect letter which denotes identity. But even greater than this, spoke Nietzsche in his Zarathustra, is your body which does not say T but performs T. Therein lies the power of Myself. Therein lies the double key to certainty-and-ambiguity, understanding-and-persuasion, thoughtand-action. My body is itself a metaphor.

9 10

Bataille, Georges: The Story of the Eye. Harmondsworth: Penguin, 1982, p. 51. Ibid., p. 54.

80

G. Olsson

Now I follow Lacan and put the signifier on top of the signified. This is my way of emphasizing that to be creative is not to have an idea that searches for its expression, but to have an expression that searches for its meaning. Built into this attitude is nevertheless a risk of thingification, of giving too much to existential quantifiers and too little to human relations. Nouns are made to win over verbs. The game is not fair. To illustrate, recall Hintikka's example (10.1) It is known to whom everybody bears the relation S, instantiated as (10.2) It is known whom everybody admires. In the formal translations, the focus is on the words of 'whom' and 'everybody'. What is highlighted is what is being related. Left in darkness is the relation itself. This practice of reification is deeply rooted, because relations are ontologically alien to the ruling ideology of presence. Relations are invisible, silences inaudible. Except in the drunkenness of the Saussurean Bar, the untouchable is pariah, the pariah untouchable. In society's interest of communication, there is a strong tendency to ignore fleeting relations by turning them to solid matter. How else could our words mean the same? In the process, we stare ourselves blind on what we are talking about and do not see the relations we are talking in. By taking subjects and objects for granted, we automatically do the same with predicates. Relations are not only relations between things but also between cultural words; not only words but concepts; not only concepts but meanings; not only meanings but other relations. Relations are always related to other relations, all twined into braids of strangely looping spirals. Self-reference is the term for this notion coiled at the center of current thought. The snake bites its own tail as modernism receives its postal prescript. In the meantime, it seems more important thatl admire, than whom I admire. Is my lover a fetish, my fetish a love? Is reification deification, deification reification? Modal logic is incredibly rich. Too much? Or too little? Jaakko Hintikka's repeated fireworks are as ingenious as they are illuminating. Hiding in the tivoli crowd, I have nibbled away at my doughnut. Smears in the face; crumbs in the moustache; licks on the lips. First, I wondered about consistency, obedience and predictability. Then, I asked how the world simultaneously is grasped and altered. In the end, I ended in self-reference. The riddle is why the doughnut did not give me indigestion. In search of an answer, I first went to Harry's Bar where I sipped a Montgomery of remembrance. I then moved on to the Gritti Palace. And in the breeze from the opened windows, belief returned to the knowledge it had never left. The Contessa is the

On Doughnutting

81

proof, for she knew at nineteen what the Colonel found out at fiftyone. Half a century of hard labor, across the river and into the trees. The lesson seems clear: To write is to walk on a pavement of citations. To read is to hear a hand whisper under a table. To explore possible worlds is to be a geographer with a mind that matters and a matter that minds. To gamble is to realize that every thought gives off a throw of dice.

WILLIAM R. SHEA Transworld Journeys Report on Session 1: Philosophy

Language-model epistemology (smuggled out of departments of philosophy and linguistics and lobbed fike grenades into unsuspecting departments of history of science) threatens to sever the tie between language and any reality external to language on the ground that language is the very structure of mental life, and no metalanguage can ever stand outside itself to observe a reality external to itself. Just as there are many possible linguistic worlds so, argues Thomas S. Kühn, there are many possible scientific worlds, and the problem of translation from one language to another is mirrored in the problem of interpreting one scientific world-view in terms of a different scientific world-view. The difficulty is compounded by the fact that, whereas members of one linguistic community generally recognize that other communities may have their own, equally valid, languages, the members of a given scientific tradition usually consider that theirs alone is genuinely scientific. Consider for instance, Sir Peter Medawar's scathing review of Teilhard de Chardin's Phenomenon of Man (that appeared with a laudatory preface by no less distinguished a scientist than Julian Huxley): Some reviewers hereabouts have called it the Book of the Year - one, the Book of the Century. Yet the greater part of it, I shall show, is nonsense, tricked out with a variety of tedious metaphysical conceits, and its author can be excused of dishonesty only on the grounds that before deceiving others he has taken great pains to deceive himself. The Phenomenon of Man cannot be read without a feeling of suffocation, a gasping and flailing around for sense. There is an argument in it, to be sure - a feeble argument, abominably expressed ...'

Underlying Medawar's intemperate outburst is the conviction that there is one scientific method, that he knows what it is, and that no one should dare to suggest there could be another one! Such statements, usually couched in a blander tone, were common in the heyday of logical positivism, the philosophy of science that dominated the scene from the eve of the Second World War to the early 1960s. Logical positivists recognized different languages but, like Medawar, they believed there was a clear demarcation between cognitively signifi1

Peter Medawar, Pluto's Republic. Oxford University Press, 1983, p. 242. The review originally appeared in Mind.

Transworld Journeys

83

cant and cognitively meaningless expressions. But locating the demarcation line soon proved difficult, and historical studies revealed that when found it had a way of shifting regardless of the pronouncements of logically-minded philosophers or philosophically-aspiring scientists. In his brillant The Structure of Scientific Revolution, published in 1962, Kühn dealt a blow to facile generalizations about the nature of science and ushered in a period of soul-searching that shows no sign of abating twenty-five years later. The quest for the scientific method that underpins all scientific research has proved as elusive as the search for a universal grammar that underlies all language. In the paper presented at this Symposium, Kühn does not disavow his belief that a scientific revolution marks a break between two "incommensurable" points of view, but he seeks a way of moving from the perspective of one group to that of a different group. Whereas a Gestalt switch was the analogy invoked in his earlier work, Kunn now favours a comparison with the acquisition of a foreign language by a culturally and socially sensitive anthropologist. Neither incommensurability nor untranslatability need debar us from the understanding of scientific texts if we have the required intelligence, determination (and modesty) to live with them and to learn from them. What Kühn will not grant is that understanding implies total comprehension. The point here is well taken since the constellation of theoretical concepts, practical insights, and mathematical techniques that cluster around a key notion in a given body of scientific knowledge cannot be fully evoked by even the best translation into a different scientific system. The case is analogous to that of poetry. A good French translation of Intimations of Immortality can capture most of Wordsworth's ideas. It may even recreate the atmosphere of the poem, but in order to do this it will have to forego literal translation for literary creation. Kühn stresses that we cannot translate an older scientific text simply by enriching the contemporary lexicon. A word alone, even a family of words, will not do. A scientific revolution is like a landslide: it moves whole layers of the lexicon to different places where they soon acquire their former deceptive naturalness and apparent permanence even though they no longer support the same superstructure. The delicate problem is the nature of the landslide: is it merely epistemological (i. e. a feature of our language about the world) or is it ontological (i. e. a feature of the structure of reality) as well? Kühn sometimes writes as though the structure of the world changes with each lexical shift, but he nonetheless maintains that we can use two different lexicons to describe the same phenomenal reality. It is difficult not to suspect that what Kühn is groping for is an up-dated version of the Kantian noumenon/phenomenon dichotomy, although he frames his discussion in terms of access to possible worlds. Members of various linguistic communities organize the world in ways that need not be identical, and Kühn even contemplates the abyss of saying that they need not overlap before withdrawing from an assertion that would preclude the possibility of the partial knowledge he wishes to defend. Kühn is no black-hole epistemologist. He does not believe we are sealed in a linguistic house of mirrors even if he chooses, at times, to use dazzling lights. What he conveys to us is a

84

W. R. Shea

vivid sense of the fact that the connection between the verbal signifier and the mental thing signified is more understandable and easier to describe than the connection of either with a world we revealingly qualify as "out there". Kühn has sometimes been branded as an anti-realist but it seems to me that he avoids this pitfall with the same kind of instinctive little lurch of faith that takes us out of bed every morning confident that the floor will be where we left it. Observations are never made in a cognitive void and even the most apparently factual report comes to us tinged with anticipations and shrouded in some conceptual garb. This theory-ladenness, however, is neither as permanent nor as objectionable as it may sound. To say that I cannot get at something without an instrument is not the same thing as to state that it cannot be reached. There is a kind of purity that is just another word for nakedness! Kühn himself gives an excellent account of various ways in which such terms as "force", "mass", and "weight" can be acquired, and the two discussants provide their own illuminating examples. The shifts that Arthur I. Miller describes in the development of atomic physics between 1913 and 1927 are complex and sometimes puzzling but he goes a long way toward making them intelligible, indeed he is even able to suggest where, in the broader cultural matrix, clues to certain changes should be sought. The worlds of science, art and philosophy are coterminous; several strands are intertwined and there is a constant interchange of information at the boundaries. Like Heisenberg we should all travel a little more Across the Frontiers. Tore Frängsmyr offers several illustrations of cross-fertilizations that may have been startling when they occurred or baffling to a later age but that make excellent sense when reconstructed with historical sensitivity. Emanuel Swedenborg's desire to explain the decrease of longevity since biblical times seems an unlikely stepping-stone to cosmological theories about the gradual slowing down of the axial rotation of the Earth, yet it stimulated research in unsuspected ways. The rough outline of what was later called the Kant-Laplace Cosmogony was formulated by Kant in his Allgemeine Naturgeschichte und Theorie des Himmels in 1755, a work in which Kant devotes several pages to the denizens of the planets of the solar system whose "natures become more and more perfect and complete in proportion to the remoteness of their dwellingplace from the Sun."2 Kant's interest in extraterrestrials may have been aroused by his reading of Swedenborg's Arcana Celestia, an eight-volume commentary on Genesis and Exodus that appeared between 1749 and 1756. The first volume of the commentary on Exodus was published in 1753 and contains the description of Swedenborg's communications with the inhabitants of the Moon, Mercury, Venus and Mars; the second volume published the following year deals with the inhabitants of Jupiter and Saturn. The discovery that the period of rotation of Jupiter is ten hours compared to the Earth's twenty-four was 2

Kant, Allgemeine Naturgeschichte und Theorie des Himmels, in Werke (edited by Wilhelm Weischedel). Wiesbaden: Insel-Verlag, 1960, Vol. I, p. 386.

Trans world Journeys

85

interpreted by Kant as an indication of the superior ability of thejovians: in five hours of daylight they achieve as much as earthlings in twelve!3 Kant's science fiction blends the possible biblical world of Swedenborg with the actual world of Newton in what for him, and many of his contemporaries, was a seamless robe. What Kühn, Miller and Frängsmyr illustrate is a cautionary tale about the perils of trying "to straighten out the facts" before "getting the facts straight", in other words of doing philosophy of science without history of science. From the vantage point of any particular moment in the development of science, what happened before the discovery of the current method is easily misunderstood. There is a natural tendency - conscious or unconscious - to mould great scientists of the past into the image of present-day scientists. Galileo is a particularly striking case of this kincfof attempt. The standard English translation of Galileo's Discorsi would have him say that he "discovered by experiment some properties of motion that are worth knowing and which have not hitherto been either observed or demonstrated".4 The words "by experiment", as Alexandre Koyre pointed out, are absent from the original Italian.5 The translators, Henry Crew and Alfonso de Salvio, obviously believed that by adding those two words they were merely making explicit what Galileo intended to convey. The result, of course, is to alter the very thrust of his argument, but the translators did not see this because they equated good science with experimentation, and they had no doubt that Galileo was a good scientist (i. e. a scientist, 1914 vintage, when the translation appeared). The rapid development of the experimental sciences led to a distortion of Galileo's views in his own century. This can be seen in a passage from the first English translation of Galileo's Dialogue on the Great World Systems by Thomas Salisbury in 1667. The content is the discussion of the path that a stone would follow if it were released from the mast of a moving ship. The Aristotelian Simplicio claims that the stone will not strike the deck at the root of the mast but some distance behind since the ship will have moved forward during the time the stone fell. Galileo's spokesman, Salviati, denies this and insists that it will strike the deck at the foot of the mast whether the ship be moving or at rest. When cross-examined, Salviata admits that he has not performed the experiment and Simplicio asks why he should believe him rather than the reputable authors who held the opposite view. Salviati's rejoinder is translated as, "I am assured that the effect will ensue as I tell 3

4

5

Ibid. p. 388. See William R. Shea, "Filled with Wonder: Kant's Cosmological Essay The Universal Natural History and Theory of the Heaven? in Robert E. Butts (ed.), Kant's Philosophy of Physical Science. Dordrecht: Reidel, 1986, pp. 95-124, esp. pp. 111-114. Galileo, Two New Sciences, trans, by Henry Crew and Alfonso de Salvio. New York: Dover, no date (reprint of 1914 edition), p. 213. The original reads, "symptomatum tarnen, quae complura et scitu digna insunt in eo, adhuc inobservata, necdum indemonstrata, comperio" (Discorsi, 3rd Day, in Galileo, Opere (edited by A. Favaro). 20 vols. Florence: Barbera, 1890-1910, Vol. VIII, p. 90). Alexandre Koyre, "Traduttore-traditore, ä propos de Copernic et de Galilee", Isis 34 (1943), pp. 209-210.

86

W. R. Shea

you; for it is necessary that it should." The original Italiant reads: "Ιο senza esperienza son sicuro ehe Peffetto seguir come vi dico".6 Salisbury, perhaps unwittingly, left out the crucial "without any experiment"! I. Bernard Cohen, who first spotted this interesting omission, points out that Salisbury, writing at the rime of the founding of the Royal Society, saw Galileo as a scientist for whom only experience counted.7 Two and a half centuries later, Crew and de Salvio, implicitly subscribing to the fashionable positivist interpretation of science, made Galileo think as they believed he must have. For my purposes in this paper, what is interesting is not so much the attempt to foster an empiricist philosophy of science on Galileo as the fact that both Koyre and Cohen were able to spot these occurrences, not because they were going over early translations of Galileo with a fine comb but because they were familiar enough with Galileo's thought processes to spot an incongruency when they came across it. Foreign languages can be learned; so can alien scientific methods. Just as a linguist who has mastered French will recognize a wrong gender, so an historian or science will pick out an anachronistic interpretation. He can never be sure that he has detected all the slips, anymore than the linguist can be certain that he has identified all the grammatical mistakes, but both practitioners, in their different ways, can become sufficiently adept to rule out gross misinterpretations. In other words, they get it right most of the time, and this is as much as can be hoped for in the realm of human communications. A deeper or, at least, a thornier problem is posed by the ambiguities that are almost always bound up with an early formulation of a new law. It is not only that there are many possible worlds, but that each world is open to several possible interpretations. Here again, the easy solution is the anachronistic one, the ascription to one man of a process that began long before him and was probably not completed until long after. A distinguished scientist and philosopher like Ernst Mach taught that Galileo, virtually single-handed, founded the new science of mechanics, created the notion of force, and discovered "the so-called law of inertia, according to which a body not under the influence of forces, i. e. of special circumstances that change motion, will retain forever its velocity (and direction)."8 Galileo scholarship has now swung the other way and some would argue that he is best understood as bringing to its culmination a long process that began in the Middle Ages.9 The realization that uniform rectilinear motion is a state and not a process is a seventeenth century achievement that cannot immediately be seen as having 6 7 8 9

Galileo, Dialogo sopra i due massimi sistemi del mondo, 2nd Day in Opere, Vol. Vm, p. 171. I. Bernard Cohen, "History and the Philosophy of Science" in Frederick Suppe (ed.), The Structure of Scientific Theories. Urbana: University of Illinois, 1974, p. 340. Ernst Mach, The Science of Mechanics, trans, by Thomas J. McCormack. La Salle, 111.: Open Court, 1960, p. 169. See, for instance, the books of William A. Wallace, notably Prelude to Galileo. Dordrecht: Reidel, 1981, and Galileo and His Source. Princeton: Princeton University Press, 1984.

Transworld Journeys

87

much in common with Aristotelian physics where motion in a straight line requires an external mover. But the principle of inertia did not spring Minervalike from a single scientific head. Between Aristotelian mechanics and Newtonian dynamics we find a transitional phase, the theory of impetus developed by such thinkers as John Buridan and Nicole Oresme in the fourteenth century. Still in the tradition of Aristotle's physics inasmuch as it looked for a cause of motion, the impetus theory moved in the direction of the modern view by making impetus an impressed (i. e. internalized) and incorporeal force, and by considering the speed and quantity of matter of a body as a measure of its strength. This theory encouraged a fresh approach to traditional problems by removing long-standing conceptual barriers. For instance, the Aristotelians had rejected outright the notion that the Earth could rotate on the grounds that a strong wind would be set up in the direction opposite to the Earth's motion. But if the air could receive an impetus and be carried around with the Earth, then the motion of the Earth itself became a distinct possibility in the world of science, and not merely a possibility in the world of science-fiction. Likewise by making the cause of motion an internal, impressed force, the impetus theory opened a new world to scientific speculation. Since air was no longer the cause of motion, as in the Platonic or Aristotelian account, motion in a void was no longer ruled out, and it became possible to think of the idealized case of a body moving in a perfect void, i. e. in the complete absence of any impeding force. Furthermore, by explaining all cases of motion in terms of one kind of cause, impetus, it removed the Aristotelian dichotomy between natural and constrained motion and provided the basis for a uniform interpretation of all motion, be it celestial or terrestrial. The fact that medieval scholars were able to question some of the fundamental tenets of the Aristotelian tradition in which they operated should be borne in mind. The Aristotelian-Scholastic cosmology was an intricate web of sophisticated concepts that accounted for physical reality in a way that discouraged thinking in certain other ways. The Void, for instance, appeared self-contradictory, and the motion of the Earth physically impossible, but the reasons for setting up these limitations were clearly stated in Aristotle, and they were explicitly recognized and criticized by writers like Oresme and Buridan. The word inertia in its technical sense was not introduced by Galileo but by Kepler who conceived of matter as characterized by a "sluggishness" or an innate tendency to rest.10 Any lump or piece of matter will come to rest unless acted upon by some force. An important consequence of this view is that a body will come to rest not only wheneverbut wherever a. force ceases to be applied to it. What moves the planets is the motive force emanating from the Sun. Were it to cease, the planets would come to a standstill. If we turn to Galileo we find statements that have a much more modern ring, such as, "Furthermore we may 10

See Alexandre Koyre, Etudes galileennes. Paris: Hermann, 1980 [reprint of 1939 edition], pp. 186-204.

88

W. R. Shea

note that any degree of speed found in a moving body is, by its nature, indelibly impressed fest in illo simple natura indelehiliter impressus] when the causes of acceleration or retardation are removed as is only the case on a horizontal plane"n,or, "Consider a body projected along a horizontal plane from which all impediments have been removed; it is clear, from what has been more fully stated in the preceding pages ffusius alibi], that this body will move along this plane with a motion that is uniform and perpetual, provided the plane extends to infinity".12 Mach took such statements to be identical with Newton's law of inertia. On closer inspection, however, we see that Galileo had not travelled that far. In Galileo's physics all horizontal planes are small sections of the circumference of the Earth and the motion that endures is not rectilinear but circular. The Newtonian analysis of planetary motion as compounded of a linear inertial component and a descent toward the centre is absent from Galileo's perspective because the belief that inertial motion is circular allowed him to claim that bodies on a rotating Earth would behave exactly like bodies on a stationary one. In his Letter on the Sunspots of 1613, he wrote: "all external impediments removed, a heavy body on a spherical surface concentric with the Earth will be indifferent to rest and to movement toward any part of the horizon. And it will maintain itself in that state in which it has once been placed... Thus a ship, for instance, having once received some impetus through the tranquil sea, would move continually around our globe without ever stopping".13 Galileo did not fly in the face of tradition, but he restated the common belief in the perennial nature of uniform circular motion in such a way that it invited consideration of motion along horizontal planes and further investigation of the concept of state of motion. What may nave been a passing remark in the text I have just quoted, became a general principle in Descartes' Principia Philosophiae where the "first law of nature" stipulates that uniform motion is conserved like rest because it is a state and not a process. A second and distinct "law of nature" states that the motion that is conserved is rectilinear as becomes "the simplicity and immutability of the operation whereby God conserves motion in matter".14 Newton encountered the concept of a state of motion in Descartes, and the continuing dialogue with his predecessor can be seen in the very title of his masterpiece, the Philosophiae Naturalis Principia Mathematica, which repeats the tide of Descartes' own work with two notable additions: the Principles are now said to be Mathematical, and the Philosophy Natural, namely what we now call

Galileo, Discorsi... intorno a due nuave scienzie in Open, Vol. VIII, p. 243. Ibid., p. 268. G^to,Istoriaedimostrazioniintomoalkmacchiesolanm Opere,Vol. V, pp. 134-135. Iquote from the translation by Stillman Drake in Discoveries and Opinions of Galileo. Garden City, N.Y.: Doubleday, pp. 113-114. Descartes, Principia Philosophiae, Part 2, art. 39, in CEuvres (edited by C. Adam and P. Tannery), new edition, 11 Vols. Paris: Vrin, 1966-1974, Vol. Vffl-I, p. 63.

Transworld Journeys

89

physics. The transformation of the tide is but a sign of the profound change that the Cartesian law of inertia underwent in Newton's hands. Descartes' two laws are fused into one, and inertia is seen as resulting from the nature of matter rather than stemming direcdy from a metaphysical attribute of God. Just as there is a continuous and intelligible path from Buridan to Galileo, so there is one from Galileo to Newton, but we must be wary of ascribing to Galileo insights that were arrived at only by working out implications of his assertions that he himself did not contemplate, let alone analyze. Four changes were necessary to convert Galileo's concept into Newton's First Law of Motion. The notion of intertia had to be: (1) recognized as playing a fundamental role in motion, (2) seen as implying rectilinearity, (3) extended from terrestrial to celestial phenomena, and (4) associated with quantity of matter or mass. The first three steps were taken by Descartes, the fourth awaited Newton. Of course, once the Principle of Inertia had been clearly formulated in the Principia Matbematica along with Newton's remark that Galileo had used it,15 no one could ever again turn to Two New Sciences without reading into Galileo's words the correct Newtonian implications. But for Galileo himself these allegedly obvious consequences had not yet entered the realm of possible worlds.

Newton, Principia Mathematics, scholium to Corollary VI of the Axioms on Laws of Motion (Newton, Principia, Motte's translation revised by Florian Cajori, 2 Vols. Berkeley: University of California Press, 1971, p. 21).

Session 2: Linguistics

BARBARA H. PARTEE

Possible Worlds in Model-Theoretic Semantics: A Linguistic Perspective /. Historical Background Possible worlds came into linguistic semantics from philosophy and logic. While the notion of possible worlds goes back at least to Leibniz, the foundations for contemporary possible worlds semantics were laid in the late 1950's and early 1960's by Rudolf Carnap, Stig Kanger, Jaakko Hintikka, and Saul Kripke. Alfred Tarski (1935) introduced the related notion of alternative models in the semantic explication of the notions of logical entailtnent and logical validity. On Tarski's account a sentence φ entails a sentence ψ if ψ is true in every model in which φ is true; and the sentence φ is valid, or logically true, if it is true in all possible models. Carnap (1947) came close to the notion of possible worlds with his state-descriptions, which nevertheless are crucially different from possible worlds in that state-descriptions are sets of sentences, i.e. linguistic objects, whereas the possible worlds of possible-world semantics are parts of the model structures in terms of which languages are interpreted. Quoting Carnap, "A class of sentences in S, which contains for every atomic sentence either this sentence or its negation, but not both, and no other sentences, is called a state-description in S„ because it obviously gives a complete description of a possible state of the universe of individuals with respect to all properties and relations expressed by predicates of the system. Thus the statedescriptions represent Leibniz' possible worlds or Wittgenstein's possible states of affairs." (p. 9)

Carnap systematized the intension-extension distinction, analyzing sameness of intension as sameness of extension in all state-descriptions, and linking the semantics of modal operators like "necessarily" and "possibly" to the intensions of the expressions they apply to. As Hintikka notes in his discussion of Carnap's contributions to the development of possible-worlds semantics, however, Carnap did not take the retrospectively natural further step of uniformly analyzing intensions as functions from possible worlds (or his state-descriptions) to the corresponding extensions, as Montague later did. Because of the ways in which state-descriptions differ from later conceptions of possible worlds, Carnap's approach applied smoothly only to the purely logical modalities, not to other philosophically important modal notions such as physical necessity or moral obligation. Indeed, even now the different notions of alternative models, of state-descriptions, and of possible worlds are not always

94

B. H. Partee

clearly distinguished. For some purposes, any one of the notions (or several others) can be made to serve equally well, but for other purposes the distinctions are crucial. The conflation is particularly easy to make, and likely to be benign, for purely extensional languages on finite domains, where the possibility that a certain predicate might have had a different extension can be taken to reflect either different ways the world might have been or different meanings the words in the language might have had. It was Kanger (1957) and Kripke (1959), (1963) who first emphasized the importance of distinguishing the notion of possible worlds from possible models and added possible worlds into the models as part of the semantic interpretation of various modal notions, particularly necessity and possibility. In the interpretation of modal prepositional logic, which is basically just like prepositional logic plus a necessity operator, written D, each model M will consist of a set W of possible worlds, an accessibility relation R between worlds, and an assigment of truth values to each atomic sentence in each possible world. Necessity is then interpreted as truth in all accessible possible worlds in the given model. So the basic clause for a formula Ο φ is given as follows: Ώφ is true with respect to M, w if and only if φ is true with respect to M, w' for all w' such that wRw' (" w' is accessible to w"). Or in the notation which we will use in discussing semantics in this paper, where [aj indicates the semantic value of an expression α and "1" represents the truth-value "True", the clause given above would be written as follows: ID^]M-W = 1 iff MM>W> = 1 for all w' s.t. wRw'. With this enrichment of the models Kripke and Kanger were able to show how different axiomatizations of modal logic corresponded to different accessibility relations among the possible worlds and hence how different competing accounts of the properties of necessity and possibility could all represent different reasonable notions that might correspond to different kinds of necessity such as logical necessity, deontic necessity, metaphysical necessity, etc. under different views about the nature of the possibilities involved. For example the axiom D φ - φ corresponds to the requirement that the accessibility relation be reflexive, i. e., that any given possible world is accessible to itself. The axiom Ο φ -* ΩΟ φ, a controversial one in early debates about modal logic, corresponds to the requirement that the accessibility relation be symmetric, i. e. that if wRw', then w'Rw. Hintikka (1962) showed the value of possible worlds for doxastic and epistemic logic with his well-known treatments of knowledge and belief. (Actually, he used "model sets" in his earliest work, possible worlds later.) The basic idea is to conceive of a certain agent's knowledge in a given possible world as dividing up the set of possible worlds into those compatible with what he knows in the given possible world and those that are not; similarly for belief. Again various axiomatizations of the notions of knowledge and belief can correspond to various accessibility relations that serve to define which are the agent's belief worlds from the perspective of a given possible world. So for example if we represent the sentence "a believes that φ" as Ba$o, we can indicate Hintikka's

Possible Worlds in Model-Theoretic Semantics

95

analysis of the semantics of such a sentence as follows: [Ba40]M>w = 1 iff iq0JM'w> = 1 for all w' such that wRjW' i. e. for all w' such that w' is compatible with what a believes in w. And similarly for knowledge, using an operator Ka. Both of these notions are also axiomatizable in various ways, analogous to modal logic. There are current applications of Hintikka-type analyses of knowledge and belief in the study of the logic of distributed systems in computation systems, consisting of multiple separate processors linked by potentially fallible communication channels. Montague contributed crucially to the development of formal semantics with his development of intensional logic, his combination of pragmatics with intensional logic (Montague 1968,1970a), and his applications of this work to natural language (Montague 1970b, 1970c, 1973), which gave rise to so-called "Montague Grammar". The intensional logic which he developed unified modal logic, tense logic and the logic of the prepositional attitudes; pragmatic aspects were introduced to treat the indexical character of such words as now, I, and here (this latter development represents joint ideas of Montague, Dana Scott, and Hans Kamp). Starting from the representation of propositions as functions from possible worlds to truth-values, properties as functions from possible worlds to sets, individual concepts as functions from possible worlds to individuals, Montague generalized these notions to a rich type theory with expressions of intensional types denoting functions from possible worlds to values of the corresponding extensional types for a broad range of types. The combination of a rich type theory, pervasive intensionality, and rich pragmatic features (particularly concerning indexicality) gave the languages that Montague developed a very rich expressive power. With his papers, Montague (1970b, 1970c, 1973), he articulated a general theory of syntax and semantics for both natural and formal languages and showed with several sample fragments how it might be applied to English as well as to the semantics of the intensional logic itself. One important factor in understanding why Montague's work made such an impact on linguistics is that Montague offered not just a richly expressive language such that one might be able to express in it every proposition expressible in English - for that matter, given a rich enough ontology, one could probably express everything expressible in Montague's intensional logic in first order predicate logic. What Montague also did and no previous logician or philosopher had done was to provide an explicit mapping from a reasonable English syntax into that semantics, embedded in a general theory treating syntax and semantics as algebras and requiring that the mapping between them be a homomorphism. That constraint can be viewed as a particularly strong version of the compositionality constraint, often referred to as Frege's principle, which states that the meaning of a complex expression is a function of the meanings of its parts and of their syntactic mode of combination (see Partee 1984). David Lewis, among his many important contributions to the development of possible worlds semantics and related issues in philosophy of language and

96

B. H. Partee

metaphysics, showed in his work on convention (Lewis 1969,1975) how one could distinguish the purely formal specification of a language from the empirical question of which language is the language of a given language community. He and Robert Stalnaker also showed the fruitfulness of possible worlds in the analysis of counterfactual conditionals (Lewis 1973, Stalnaker 1968): "If p had been the case, then q would have been the case", or as Lewis represents it, "p O q". For the Lewis-Stalnaker analysis, assume we have a measure of similarity among possible worlds so that we can speak of w' being closer to w than w" is: this relation is vague, but so are counterfactuals, arguably in the same way. Then the counterfactual above is analyzed as: [ρ Ο-* q]^>w = 1 iff [q]M'w> = 1 for that world w' which is the closest world to w such that [p]M,w = ι Thus, the counterfactual is true in the actual world if q is true in that world which is as much like the actual world as possible but with ρ true.! Lewis was also influential with his methodology; "In order to say what a meaning is, we may first ask what a meaning does, and then find something that does that" (Lewis 1970). Linguists working on semantics prior to Lewis and Montague had been working with what in hindsight could be seen as uninterpreted representational languages of unknown expressive power, often ad hoc modifications of first-order predicate logic, and not surprisingly had had at least as much difficulty in articulating what their formalisms were supposed to mean and what constraints there were on them as logicians had had in articulating the semantics of their languages before the advent of model theory. Lewis helped to clarify and extend the notion of meanings as functions, and showed how Montague's strategy for dealing with tense, modality, and pronouns might naturally be expected to extend to meanings as functions not just from possible worlds, times and speakers but from some generalized index containing all the parameters by which extension might vary, such as demonstrated objects (for the demonstratives this and that), degree of precision, degree of salience of various items in the context of discourse, etc. Cresswell (1973) later argued persuasively against Lewis's view of an index as a discretely specifiable "η-tuple" and in favor of a more generalized context parameter, arguing that there is no limit in principle to the aspects of context that may be relevant to interpretation. I believe that most linguists are inclined to agree with Cresswell. Cresswell also contributed a great many specific analyses of English constructions, including work on such thorny problems as the semantics of comparatives, of adverbial prepositional phrases and the mechanism of point-of-view in expressions of location and motion, of mass terms, of tense and aspect and its interaction with quantifier scope, and most tenaciously The formulation given here follows Stalnaker's analysis in presupposing the existence of a unique closest world satisfying the given conditions. Lewis's analysis allows the possibilty that there may be a multiplicity of closest worlds where ρ is true, and requires that q then be true in all of them for the counterfactual to count as true. I follow Stalnaker here for expository simplicity; nothing here will hang on the difference.

Possible Worlds in Model-Theoretic Semantics

97

on the semantics of propositional attitude reports, with his most recent views rejecting a pure possible worlds approach. (I will come back to the topic of propositional attitudes below.) Cresswell and Lewis were both concerned with the metaphysical foundations of possible worlds theory at least as much as with its application to semantics (I will return to the foundational questions in section IV.) With the work of Montague, Lewis, Cresswell, Thomason (1976) and others it began to be clear that intensionality, which in their theories was analyzed in terms of possible worlds, was the norm rather than the exception in natural languages. Let me illustrate what this means before going on with the historical sketch. Following Carnap (1956), we call a grammatical construction extensional, relative to a certain syntactic and semantic analysis, if the extension of the whole is a function of the extensions of the parts. And we will say that a grammatical construction is intensional, relative to a certain syntactic and semantic analysis, if the extension of the whole is a function of the intensions of one or more parts and the extensions of the remaining parts.2 Following the standard treatment of possible-worlds semantics, we take the extension of a sentence to be a truth value, and its intension to be a function from possible worlds to truth values (the intension of a sentence is called a proposition); the extension of a one-place predicate is taken to be a set of individuals, and its intension, called a property, is taken to be a function from possible worlds to sets of individuals. The matter of term phrases is made more complicated by the differences among proper names, definite and indefinite descriptions, and quantifier phrases, but for the simplest cases we can say that the extension of a term phrase is an individual and its intension is a function from possible worlds to individuals, often called an individual concept. (Cresswell coined the term hyperintensionalior constructions where not even the intensions and extensions of the parts suffice to determine the extension of the whole; whether any natural language constructions are in fact hyperintensional depends primarily on the correct analysis of propositional attitude sentences. We will assume that all natural language constructions are either extensional or intensional until we explicitly take up this question in sections ΙΠ and IV.) The clearest and best known test of whether a construction is intensional or extensional is Quine's test for referential opacity (Quine 1960): choose two singular definite descriptions which are contingently coreferential and test whether substituting one for another in a given construction necessarily preserves the extension of that construction: if it does, the construction is extenAn alternative way of viewing this, one more in keeping with Frege (1892) and the treatment of Montague (1973), is to require the extension of the whole always to be a function of the extensions of the parts, and characterize the intensional constructions (Frege's oblique or indirect contexts) as those in which we must assign as the extension of a given expression what would ordinarily (in direct contexts) be its intension.

98

B. H. Partee

sional, otherwise intensional.3 While the conclusions to be drawn from "failure of substitutivity" in such cases are of course somewhat theory-dependent, the following examples can at least plausibly be diagnosed as indicated (Partee 1974). (1) Verb plus direct object (y Np) can ^e ύ1*6*1^0113!; me classification depends on the verb. We give an example of each. (1.1)

Extensional: strike + NP Test: (i) The bullet strikes the owner of the red Volvo. (ii) The owner of the red Volvo is Mary's husband. (iii) (Therefore) The bullet strikes Mary's husband. Valid, so extensional.

(Note: I am using present tense despite unnaturalness to keep the context of evaluation constant in all three sentences and avoid complications of other tenses.) (1.2)

Intensional: seek + NP Test: (i) John seeks the owner of the red Volvo. (ii) The owner of the red Volvo is Mary's husband. (iii) (Therefore) John seeks Mary's husband. Invalid, (on at least one reading), therefore intensional.

Most transitive verbs are extensional; other intensional ones include want, need, hope for, wish for (and many combinations wither), imagine, picture, offer, owe. (2) Constructions with embedded thai-clauses. Verbs and adjectives that take ί^Λί-clauses as arguments are very often intensional. Tests parallel to those given above would show that all of the following examples involve the intension of the embedded sentence. (2.1) (2.2) (2.3)

Mary believes that the man in the raincoat is a spy. That the bank president works in the bank is obvious. It appears that a child has been writing on the walls.

Another frequently cited test for extensionality is the "existential generalization test", illustrated below, (i) John is sitting on a talking horse. (Therefore) There is a talking horse such that John is sitting on it. (ii) John is looking for a talking horse. (Therefore) There is a talking horse such that John is looking for it. The validity of (i) and invalidity of (ii) are taken as further evidence for their extensionality and intensionality respectively, supporting the evidence of (1.1) and (1.2) in the text. However, the results of these two tests may not always coincide and I believe it is best as well as more standard to take die substitutivity test as the criterial one.

Possible Worlds in Model-Theoretic Semantics

99

Among the few purely extensional sentence-embedding adjectives are true and false, as illustrated in (2.4); the rarity of this case suggests that the construction in general should be regarded as intensional, the extensionality of (2.4) following from the specific lexical meaning of true. (2.4) It is true that the bank president is a spy. (3) Verb-phrase adverb plus verb phrase, and sentential adverb plus sentence. Although it is not always easy to distinguish verb-phrase adverbs from sentential adverbs, one finds in general that sentential adverbs can create referential opacity in the sentences they modify and verb-phrase adverbs do the same in the verb phrases they modify. Examples of intensional sentence adverbs are given in (3.1-3.2), intensional verb-phrase adverbs in (3.3-3.4), extensional adverbs in (3.5-3.6) (3.1)

Smith's murderer is probably insane. (semantic analysis: probably interpreted as a function which applies to the proposition expressed by Smith's murder is insane.) (3.2) The president of trie country is necessarily the commander-in-chief of the armed forces. (3.3) Mary reluctantly bet on the best horse. (3.4) Sam willingly invited the robber into his house. (3.5) Smith's murderer is now insane. (3.6) Mary quickly bet on the best horse.

The fact that almost every English construction includes some instances which are intensional led Montague (1973) to make intensionality the general case, guaranteeing the extensionality of the extensional cases by means of meaning postulates which constrained the models that could be candidates for possible interpretations of English. For instance, a meaning postulate for the verb strike would limit the possible interpretations to those in which the extension in a given world of strike plus its direct object NP would depend only on the extensions of strike and of that NP in that world. For an intensional transitive verb like seek, on the other hand, there would be no such meaning postulate. By relaxing Montague's constraints on the syntactic-semantic mapping slightly, one can also make the distinction between extensional and intensional transitive verbs directly in their type structure, treating the extensional ones extensionally right from the start (Partee and Rooth 1983). The choice between these approaches is an empirical linguistic issue, and the treatment of the intensional cases is the same on both approaches, and in sharp contrast to the "sentential operator" approach to all intensionality, characteristic of treatments which just add certain sentential operators to otherwise first-order predicate logic. Making intensionality the general case, as Montague did, has the linguistic advantage over modified-first-order approaches of letting the semantic structure match natural language surface structure much more closely and hence making natural

100

B. H. Partee

languages look semantically much more reasonable than they look if one takes "logical form" to mean a basically first-order-predicate-logic form. We have already mentioned Montague's steps toward integrating certain aspects of pragmatics with possible worlds semantics. It was natural that the first aspect of pragmatics that he focused on was the treatment of indexicals, since, for instance, the integration of tense logic with modal logic leads directly to the consideration of the indexicality of the present tense and of words like now. It is then a small step from there to consideration of /, you, here, and other aspects of the context of a speech act that are directly involved in determining the truth conditions of expressed propositions - in fact in determining what proposition is in fact expressed, given that propositions are identified with their truth conditions on this classical possible worlds approach. Further attention to pragmatics was shown in the work of Hamblin (1970), David Kaplan (1977) and in the work of Robert Stalnaker (1974, 1976, 1978). Stalnaker introduced the important notion of common ground in a conversation: a set of propositions which all the participants in a conversation take to be shared background knowledge or uncontested background assumptions at a given point in the conversation. The propositions in the common ground'm a conversation at any stage determine a set of possible worlds, the context set, which the conversational participants take to be live options at that stage of the conversation, in the sense that as far as the information in the common ground goes, the actual world might be any one of the possible worlds in the context set. To gain information about the actual world, then, is to add propositions to the common ground that serve to eliminate some possible worlds from the context set, the set of live options. On this view, a primary purpose of making assertions (in the factual mode of discourse) is to cut down on the set of possible worlds in the context set, thereby narrowing down the range of uncertainty of the conversational participants as to which possible world is the actual one. Of course, since the common ground may include shared misinformation, there's no guarantee that the actual world is in fact included among those the participants take to be possible at any given stage. In fact, the notion of common ground in a conversation does not in any way actually require that there even be such a thing as an actual world, although it is more difficult to make sense of the purpose of much conversation if it is not taken to be a striving to converge on a more narrowly constrained set of possibilities as to which the actual world may be. Stalnaker emphasized the dynamic nature of pragmatics in stressing how what is said both depends on the common ground (and other aspects of context) and in turn affects the common ground. His pragmatic analysis of presupposition has been very successful, although much remains to be worked out - this is still a very lively field of inquiry. Stalnaker, building on earlier work of Montague, Kaplan, Kamp and Vlach, also did very important work showing how possible worlds have to be taken to enter the analysis of the interpretation of sentences twice - once as an aspect of context and once as an index at which a sentence is evaluated; this he took to be crucial for under-

Possible Worlds in Model-Theoretic Semantics

101

standing the functioning of demonstratives in prepositional attitude contexts, among other things. Extensions of Stalnaker's ideas about pragmatics in the interpretation of a variety of constructions by Kratzer (1978) and Heim (1982, 1983a) led to the suggestion by Heim that the integration of semantics and pragmatics has to be even closer than had been proposed in earlier work. Heim argued that for an explanatory linguistic theory, one wants to take the basic semantic value of a sentence to be not its truth conditions (a function from possible worlds to truth values), as in the classical models, or even a function from contexts of use to a function from possible worlds to truth values, as had been proposed by Stalnaker and by David Kaplan, but rather a function from contexts to contexts, where the notion of context is much like that of Stalnaker's context set, augmented at least by something like "discourse referents" representing the individuals being 'talked about' and available for anaphoric reference at a given point in a discourse. This may seem a small change, perhaps, given that contexts include possible worlds, but it is an important one in emphasizing a much more dynamic view of semantics and a greater inseparability of semantics from pragmatics and creating the potential for greater rapprochement between formal semantics and speech act theory, and incidentally also closer connections with work on the semantics of programming languages, where a dynamic view is common. While on this view the interpretation of a sentence is not simply as a set of possible worlds, possible worlds play as central a role in this picture as in its predecessors. In a theory like Heim's, a sentence is interpreted as a set of possible worlds in the special case where the context set is empty and the sentence itself has neither presuppositions nor contextual effects. The classical possible worlds theory is then viewed from this perspective as a theory applying to such context-independent cases, much like a theory of motion for frictionless planes. This brings us up approximately to the present in this brief sketch of the development of possible-worlds semantics. In the next section I will describe, also briefly, a sampling of alternative approaches to semantics, including alternatives within a basically possible-worlds tradition, plus non-possible-worlds approaches to semantics that are still within a model-theoretic tradition, and some alternatives that are not model-theoretic at all, so that the non-specialist may have at least a bit of perspective on possible-worlds semantics in relation to rival theories, and the specialist can see clearly where my own biases lie. Then in section ΠΙI offer my evaluation of the effects the possible-worlds approach has had on the development of linguistic semantics, and of what features of possible worlds theories have been most crucial in these developments. Finally in section IV, I come back to what I see to be some of the central foundational issues that are being confronted or will soon have to be confronted as the frontiers of research in semantics and pragmatics continue to be pushed back.

102

B. KRutee

Π. Α Sampling of Alternative Semantic Theories Semantics is currently the focus of a lot of lively research, much of it interdisciplinary, involving such fields as artificial intelligence, theoretical computer science, information science, and cognitive science, as well as linguistics and philosophy. It is not surprising that there are a lot of competing approaches, rapidly evolving theories, and considerable mutual influence among these diverse approaches. It is hard to draw a road map of the terrain; I can't pretend to be fully representative, but I would like to try to provide at least some perspective on the current scene.

A. Variants of Possible-Worlds Semantics First, let me mention some differences of approach within the possible-worlds tradition. On the philosophical side, there have been different degrees of metaphysical realism about possible worlds. There are "extreme realists", David Lewis the best known, who believe not only that other possible worlds do not differ in kind from the actual world, but that each possible world is just as actual from the point of view of its inhabitants as ours is from ours (Lewis 1973). Some philosophers and most linguists prefer to view possible-worlds talk as a useful instrument in theorizing about semantics and other enterprises, while taking a skeptical or an agnostic stand towards the metaphysical reality of possible worlds. Stalnaker in his recent book Inquiry presents a version of "moderate realism" in which there are other possible worlds in the sense of other ways things might have been, but these are not more things of the same kind as the actual world. Rather he takes them to be abstract entities existing in the actual world, "abstract objects whose existence is inferred or abstracted from the activities of rational agents." (Stalnaker 1984, p. 50.) While the debates about metaphysical realism do not translate in any simple way to differences in semantic theories, they are directly relevant to differences in broad conceptions of the nature and grounding of a semantic theory of natural language. Since attention to these questions is relatively recent, I will postpone further discussion of this issue to section IV, but note that until quite recently possible-worlds semantics has tended to be dominated by a realist picture such as Lewis's, creating some tension for linguists attracted by the technical merits of possible-worlds semantics but with strong leanings (derived from the Chomskyan perspective) towards a psychologistic basis for any branch of linguistic theory. A related difference within possible-worlds theories is whether possible worlds are taken to be primitive or constructed. An extreme realist will generally, perhaps necessarily, take possible worlds to be primitive, but a moderate realist or instrumentalist has a choice. Some conceive of possible worlds as built up out of individuals and properties - that then leads to questions about the nature of individuals and the nature of properties, questions on which there is at least

Possible Worlds in Model-Theoretic Semantics

103

as much debate as there is about possible worlds.4 From the perspective of possible-worlds semantics, one might view possible worlds as analogous to points in a theory of geometry, and simply not make any metaphysical claims as to whether they are to be viewed as primitive, as abstracted from conceptions of three dimensional space, or as possibly otherwise analyzable. Another dimension of variation, one which linguists tend to concern themselves with more directly, is on theories of syntax and of the relation of syntax to semantics and pragmatics. While most of these differences are beyond the scope of this paper, there are some that should be mentioned here. One important point is that while not all possible worlds semanticists advocate phrase-structurelike grammars it is probably fair to say that the development of model-theoretic semantics has done much to make the return of near-phrase-structure-grammars possible by providing tools which let one capture relations between sentences semantically in cases where previously the only way to show a systematic relation between two sentence-types was to derive them from a common syntactic source or deep structure via transformations. Under the heading of different theories of syntax and of the relation of syntax to semantics and pragmatics within the possible-worlds tradition, alternatives to mention include game-theoretic semantics (developed by Hintikka and his colleagues; see Saarinen, ed. (1979)), discourse representation theory (Kamp 1981), generalized phrase structure grammar (Gazdar, Klein, Pullum and Sag 1985), extended versions of categorial grammar (Bach 1984; Dowry 1982, Steedman, and others), head grammar (Pollard 1984), and other theories being developed in the Stanford area using ideas of unification as developed by Martin Kay and others. Early attempts to wed transformational grammar to possibleworlds semantics (Partee 1975 and others) have generally given way to attempts to use the power of the semantics to eliminate most transformation-like rules from the syntax; in particular there has been a tendency to eliminate essentially all "local transformations" (see Dowty 1978), but often to retain one or another special device for dealing with "unbounded transformations" such as whquestions or relative clause formation, as well as the representation of scope ambiguities among quantifiers. One major unresolved point of debate that divides theories within the model-theoretic tradition with respect to the relation between syntax and semantics concerns the existence of a distinct level of "logical form" mediating between natural language syntax and model-theoretic semantics. On Montague's original theory, the strong compositionality constraint required that any such level be in principle dispensable; the closest analogue to "logical form" in that theory is the "analysis tree" which displays the syntactic derivation of an

On such a view, the "building up" process may also have to be taken as a new primitive, usually modelled via standard set-theoretic approaches treating worlds as maximal consistent sets of propositions.

104

B. HPartee

expression.5 Since on Montague's theory there is a unique semantic interpretation rule for each syntactic rule, the analysis tree that displays the syntactic derivation of an expression corresponds simultaneously to the steps in the compositional semantic interpretation of the sentence. While it is a practical necessity to employ something like Montague's intensional logic as a metalanguage for representing the resulting model-theoretic interpretation, expressions in the intensional logic have no special status in the theory, and such a representational level is taken to be in principle dispensable.6 Not all theories in the possibleworlds tradition hold to such a strict version of compositionality, however, and the "representationalist" tradition in linguistics, discussed more below, leads linguists to continue to investigate the possibility that some sort of intermediate level of logical form or the like might play an important role in the mental representation of semantic interpretations both for individual languages and at the level of semantic universale. Empirical evidence for or against such a hypothesis has been very hard to come by so far, however. B. Non-Possible-Worlds Theories within Model-Theoretic Semantics A recent challenge to possible-worlds semantics from within the model-theoretic tradition has been raised by Barwise and Perry with their development of situation semantics (see Barwise and Perry 1983). One of their contentions is that early arguments by Frege, Church and others to the effect that the extension of a sentence must be a truth-value were flawed, and that the alternative intuition that a sentence describes a relatively "local" state-of-affairs can be defended and put on a sound formal footing. They thus take as primitive the notion of a situation, corresponding approximately to a limited piece of the actual world; with properties and individuals also taken as primitives, they build up the notion of a situation type. They emphasize the inherent partiality of situations and situation types as opposed to inherently "complete" possible worlds, although it has recently been argued by Stalnaker (ms. 1985b) that there is no obstacle in principle to identifying a situation with the set of all possible worlds that have that situation as a part. (However, problems do seem to arise with the introduction of situations that have other situations as parts, as proposed for the treatment of situations involving perception; see Landman (1985).) Situation semantics has not dealt explicitly with the semantics of the modal constructions such as possibility and necessity which motivated the possible worlds in the first Another way of putting the matter would be to say that any candidate level of "logical form" in strict Montague grammar must either be dispensable (like Montague's language of intensional logic) or must count as pan of the syntax (like Montague's syntactic rules for dealing with quantifier scope). Raymond Turner (personal communication) points out that one can question the special status accorded on this view to Zermelo-Frankel set theory, the metatheory for the underlying model theory, and it might be argued that the language of ZF set theory is less well suited than intensional logic as a metalanguage for semantics.

Possible Worlds in Model-Theoretic Semantics

105

place, and the theory is still very much in flux, so it is probably still too early to say whether the apparent eschewal of the notion of possible worlds represents a substantive difference in the theories or not. As a converse influence, the attention given to sentences reporting perceptions of events and other local situations within situation semantics is one of the influences that has led many possible worlds theorists to renewed interest in theories which include events along with or in place of times among the primitives in their model structures, as opposed to theories which reduce events to properties of instants or intervals of time as Montague had proposed. The idea that events, processes, and other "situation"-sized entities should be included in models as basic rather than treated reductionistically has surfaced in many places and with many motivations. I don't think one should necessarily see a fundamental opposition between possible situations and possible worlds, however; I think the broad notion of possible worlds, as alternative ways things might have been, is as fundamental in thinking about alternative possible courses of "local" events as in thinking about alternative possible global histories. I thus see the place of situations in a model structure as more closely analogous to that of individuals or events than to that of possible worlds. C Non-Model-Tbeoretic Semantics One could not do justice to the past and present development of semantics in linguistics without emphasizing the older and still thriving traditions that are not model-theoretic at all. Up until the early 1970's, when Montague's influence began to be felt in linguistics, virtually all linguistic theories of semantics were what we might characterize as representationalist in substance and conceptualist in foundation, and in many quarters this is still the dominant trend. By "representationalist", I mean the view that interpreting a sentence of a natural language is a matter of deriving a representation of it in some suitable language of semantic representation, such as Jerry Fodor's proposed "language of thought" (Fodor 1975,1981), Katz and Fodor's earlier language of "semantic markers" (Katz and Fodor, 1963), or recent versions of "logical form" advocated by May (1977, 1985), Higginbotham (1983), or others in Chomsky's "GB" framework (Chomsky 1981). The "conceptual dependency" representations or "semantic networks" used in some current artificial intelligence work are close relatives of this approach to semantics. Against the charge that just translating from one uninterpreted language to another gets one no closer to the first thing about the meaning of a sentence (a charge articulated by David Lewis (1970)), namely its truth-conditions, the responses are usually twofold, (i) If the semantic representation language is universal, as is often posited, so that the sentences of any natural language can be represented in it, then it can satisfy at least one goal of semantics, namely to provide a theoretical framework for studying how languages of the world with different syntactic structures can express the same meanings, and to provide a language-independent way of

106

B.H.Partee

characterizing these meanings (at least independent of any particular natural language.) (ii) Furthermore, it is still a matter of debate whether the way speakers of natural languages use semantics in speaking, understanding, and carrying out inferences is better modeled by model-theoretic or proof-theoretic means or some combination of the two. A strictly model-theoretic approach with no intermediate representational level is committed to denying that semantic competence involves in any systematic way the syntactic manipulation of expressions in some language of semantic representation, while a strictly representationalist account is committed to that being all there is to semantic competence, denying any systematic role to intrinsically model-theoretic operations. However, against this response it can be replied that model-theoretic approaches normally are accompanied by one or more representational languages, often with a proof theory as well as a model theory, while representationalist proposals often have neither. In the absence of an actual proof theory or other grounding, the import of proposed repesentational languages is difficult to judge. Actually, this characterization of representationalism reflects not only the representationalist position but the conceptualist position as well, by which I mean the position that as far as the semantics of natural languages goes, criteria of adequacy of a given theory derive ultimately from the attempt to account for what is "in the head" of the competent native speaker of a language. This position, a basically Chomskyan one, is directly at odds with the Fregean anti-psychologism inherited by much work in logic and formal semantics in philosophy. (See Partee (1979), (1980)) While representationalism and conceptualism often go together, particularly in the Chomskyan tradition, they are not inherently inseparable; and one could mention recent Katz (1981), as an example of a non-conceptualist representationalist, and Johnson-Laird (1983), as an example of a conceptualist who advocates a model-theoretic approach and has been seeking plausibly psychologically realizable versions of possible-world semantics. Conceptualism in linguistics is also related, I believe, to functionalism in the philosophy of mind and its influence on the foundations of cognitive science. For reasons that may be partly a matter of historical accident, cognitive science as it has been developing in the U. S. in the last decade or so has been dominated by a very syntactic "information-processing" model of semantics, influenced heavily by the computational metaphor, and it is only recently that new interest in the foundations of the semantics of computer languages has begun to bring serious interest in possible worlds semantics and other model-theoretic approaches to semantics into mainstream cognitive science. To put it very crudely, one might say that the position of conceptualists-functionalists is that all that really matters for semantics is what the language user thinks the sentence means and that in turn can be characterized by the state of his mind viewed as a machine with a lot of different internal states functionally related to each other; then, equally crudely, one could characterize the new interest in the semantics of

Possible Worlds in Model-Theoretic Semantics

107

knowledge representation languages used in artificial intelligence research as a concern with the question of how one can attribute informational content to internal states of a computer. It is no accident that this is in a sense a revival of the basic problem of intentionality - what it is for our mental states to have informational content. What is more surprising is that it has seemed possible to avoid coming to grips with that problem for this long in much of the forefront of research in cognitive science. The connections between possible worlds semantics and representationalist theories are bound to become much closer and much more intertwined in the coming years, I believe. ///. The Value of Possible Worlds for Linguistic Semantics As I tried to emphasize earlier, the value of possible worlds for linguistic semantics is largely independent of metaphysical issues. Linguists, qua linguists, tend to be instrumentalists about metaphysics, in the sense that they will tend to judge competing foundational theories more by their fruitfulness in helping to lead to insightful explanations of linguistic phenomena than by any other kinds of arguments. (In fact this is sometimes carried so far that not all linguists seem to believe that a theory of the semantics of natural language needs to be consistent although I think that in some cases this reflects confusion over where to deal with the fact that language users can have inconsistent beliefs.) So for example, in the case of questions about the nature of time - whether it is discrete or continuous, whether there is a first or a last moment, whether it it linearly ordered across possible worlds or whether one should rather posit a system with, say, "branching futures" but "linear pasts" - the primary concern of the working linguist is which set of assumptions will enable her to give the most satisfactory account of the semantics of tense, aspect and temporal adverbials in a given language and across languages. In fact, as Bach (to appear) has suggested, if one were to find evidence that different underlying assumptions of this sort gave optimal semantic accounts for different natural languages, one would have an important contribution to an enterprise he has dubbed "natural language metaphysics", possibly giving new respectability to the kinds of questions raised by Sapir and Whorf earlier in this century about the relation between language and "world view". I should also mention at the outset of this section that for the first decade or so of Montague Grammar, roughly the 70's, linguists working in Montague's possible worlds framework did not feel competent to tamper with the foundations, and tended to take the whole package of PTQ or at least Montague's Intensional Logic as a unit - including the type theory, the single set of possible worlds per model, linear time across all possible worlds, classical logic, functions that were always total, a single domain of possible entities across possible worlds, intensions abstracting simultaneously over worlds and times, a typed rather than an untyped λ-calculus, etc. There was even a tendency to regard

108

B.H.Partee

"model-theoretic semantics" and "possible-worlds semantics" as equivalent, and both as equivalent to Montague Grammar. Closer cooperation and mutual influence among linguists, philosophers and logicians has led to a much greater proliferation of theories within this general approach in recent years. So in assessing the value of possible worlds for linguistic semantics, it is not always easy to separate out the contribution of possible worlds proper from other aspects of possible-worlds semantics as developed by Montague and others. I would summarize the kinds of value possible-worlds theories have had for linguistic semantics as being of two kinds: (i) possible worlds as a technical tool have helped to provide an appropriate structure on the space of meanings; I will illustrate what I mean by that with examples below, (ii) As a further benefit, not so much within linguistics proper but of potential value for linking linguistics to other disciplines, the possible-worlds conception is of great help in relating linguistic meaning to other kinds of informational content, and I want to say a little bit about that before getting into specifically linguistic examples. As emphasized by Stalnaker (1984), possible worlds help to provide a general account of the intentionality of mental states that does not make all intentionality irreducibly linguistic. The reader may have noted the occurrence of both the words 'intentionality' and 'intensionality', and it is at this point that I should make the distinction explicit. 'Intentionality' as used in the contemporary philosophical literature can be said to be concerned primarily with the question of what it is for mental states to have informational content - sometimes expressed as the problem of 'aboutness'. 'Intension' is a technical term used in opposition to 'extension', analyzed in different ways in different theories. The term in its current usage has its roots in Carnap (1947) (although the term is older); Carnap suggested that individual terms, predicates, and sentences each had an intension as well as an extension, identifying these as follows: Ο

Expression sentence predicate individual term

C7

' JT

X

Intension proposition property individual concept

G>

Ο

Extension truth-value set individual

Montague unified and generalized Carnap's intension-extension distinction, developing a typed intensional logic in which intensions are uniformly analyzed as functions from possible worlds to corresponding extensions. The link between "intention" and "intension" can be argued to be the notion of possible world: both linguistic content and the content of mental states such as believing or perceiving involve discriminating among different possible worlds. To be capable of intentional states is on this view to be capable of in some sense entertaining different possibilities and in some sense discriminating among them. Such discrimination does not in principle require language nor does it require that there be some single absolute set of all possible worlds involved whenever we talk about discriminating among alternatives. It thus gives one a good handle on talking about the similarities and the differences between e. g. the

Possible Worlds in Model-Theoretic Semantics

109

content of an act of visual perception and the content of an uttered sentence; and also the similarities and differences between the content of the mental state of a dog which remembers where it buried a certain bone and its master who also knows where the bone is buried - an oft-cited case in which one problem is how to account for the sense in which they may be said to know the same thing without attributing to the dog a full set of mental states fully isomorphic to those of its master nor a sentence in the same "language of thought" as its master presumably thinks in. Even if the "language of thought" approach characteristic of representationalist views of semantics were able to match the model-theoretic approach in handling the kinds of linguistic examples discussed below, it might have to be at least supplemented by something like the possible-worlds approach in order to make appropriate connections between linguistic content and other obviously related cases of intentionality and content such as those just mentioned. Let us turn now to some specific examples of linguistic phenomena where possible worlds have played a role in the analysis, and try to flesh out the claim I made above that possible worlds help to provide an appropriate structure on the space of meanings. (i) Let us first consider the identification of a proposition with a set of possible worlds, that is with the set of possible worlds in which it is true. This is equivalent to identifying a proposition with its truth conditions or with a function from possible worlds to truth values, or with a method of discriminating among a set of possibilities. Ignoring the additional complications relating to pragmatics and context-dependence that I mentioned earlier, a proposition in this sense is taken to be the meaning of a sentence on the classical possibleworlds view. A first point to note about structure is that on this analysis, one can ascribe various logical relations to propositions without any necessary pointfor-point corresponding ascription of "logical form" to the sentences that express them. For example, consider sentences (l)-(4) below. (1) (2) (3) (4)

John and Bill are mortal. John is mortal and Bill is mortal. John is mortal. Bill is mortal.

In terms of overt syntactic form, (2) is a conjunction of (3) and (4), while (1) is not. Semantically, since (1) and (2) express the same proposition, (1) is just as much the conjunction of (3) and (4) as (2) is; on the possible-worlds account this is straightforward, since (1) and (2) pick out exactly the same set of possible worlds, a set which is the intersection of the sets of possible worlds picked out by (3) and (4). In a typical model-theoretic account such as that provided by Montague, (1) and (2) can be given different syntactic structures, analysis trees isomorphic to their respective surface structures, and these distinct syntactic structures are interpreted by distinct semantic rules: rules for noun phrase conjunction in the case of (1) and for sentence conjunction in the case or (2). If

110

B. RPartee

the rules are correctly formulated, they lead to the same truth-conditions, though via different compositional routes. On a purely model-theoretic account, there need be no "level" at which (1) and (2) have the same "semantic representation" or "logical form". To follow an analogy suggested by Stalnaker, in ordinary arithmetic the different expressions "7—2" and "1+4" determine the same number, but that need not mean that the number so determined is intrinsically a sum or intrinsically a difference, nor that the number 5 should be viewed as having the numbers 4 and 1 as "parts" just because it is their sum. Linguists originally tended to take a very syntactic view of semantics, focussing on the obvious relation of an overt sentential conjunction like (2) to its syntactic parts, (3) and (4), and tending to want to reduce the syntactically less obvious case of conjunction in (1) to that of (2). The difference in the two conceptions of conjunction, to summarize, is this: for a sentence A in a "logical form" to be the conjunction of two sentences B and C, A must contain B and C as parts with a conjunction sign between them; but for a proposition A (a set of worlds) to be a conjunction of propositions B and C, A must DC the intersection of B and C; and this latter does not require that B and C be in any sense "parts" of A. Thus the possible worlds account, but not a "logical form" account, can account directly for the fact that (1) expresses the conjunction of (3) and (4) just as much as (2) does. Propositions viewed as sets of possible worlds have a Boolean structure with respect to the Boolean connectives and, or, and not, The case of entailment is even clearer than the case of conjunction. In terms of possible worlds a proposition A entails a proposition B if the set of worlds in which A is true is a subset of the set of worlds in which B is true. In order to get a corresponding notion in a theory of logical form, one needs to articulate a set of axioms and of rules of inference and define a notion of consequence in terms of proof in such a system. Possible worlds semantics has made it possible to say a great deal about entailment relations among sentences of natural languages without having to enter into questions of choices among possible axiomatizations of natural language inference systems, choices which would probably be extremely difficult to make on principled grounds if possible at all. (ii) As mentioned earlier, the possible worlds analysis of intensionality has made it possible to appreciate and analyze the rampant intensionality of natural languages, especially by virtue of Montague's use or function-argument structure to interpret most basic syntactic constructions. The idea that for any possible extensional type of interpretation there is a corresponding intensional type has been a fruitful one. So, for instance, old puzzles about the behavior of "roledenoting" definite descriptions like the president can in many cases be analyzed by assuming the expression may denote a non-rigid individual concept, whose scope with respect to other operators in the sentence may be ambiguous. Example (5) is ambiguous in this way. (5)

The president is gaining power.

Possible Worlds in Model-Theoretic Semantics

111

Likewise, some old puzzles in the theory of actions can be seen to involve action-descriptions under the scope of intensional adverbs in cases where the descriptions have different intensions but identical extensions. In this manner the invalidity of an inference like that in (6) can be explained. (6) John raised his arm intentionally. In raising his arm, John set off the alarm. (John's raising of his arm = John's setting off of the alarm) Therefore John set off the alarm intentionally. (INVALID) Similar analyses have elucidated the semantics of intensional adjectives like good, skillful, strict and brought to light the hitherto unsuspected intensionality of the English progressive aspect, not to mention the expected applications to the semantics of modals, sentential adverbs, and many sentence-embedding constructions. In turn, as the limits of what can be explained by intensionality have been reached, the role of pragmatic factors in context-dependent interpretation has become clearer and the magnitude of the difficulty of what is probably the hardest problem in semantics, the interpretation of prepositional attitude constructions such as belief-sentences, has become more vividly apparent. (iii) The semantics of belief-sentences and other prepositional attitude constructions has been an interesting battleground for possible-worlds semantics. Hintikka's use of possible worlds to elucidate the analysis of knowledge and belief (Hintikka 1962) was an important advance in clarifying many of the issues around referential opacity in the context of belief-reports, distinctions between knowledge by description and knowledge by acquaintance, etc. But the very kinds of logical relations among propositions that I talked about as an advantage of the structure of possible worlds semantics in (i) above has been argued to be an insurmountable obstacle to an adequate treatment of propositional attitudes in terms of possible-worlds semantics. Let me first sketch the standard treatment of propositional attitude sentences in possible-worlds semantics before describing the problem. Consider sentences (7)-(9). (7) Martha believes that the president of the bank is honest. (8) The president of the bank is the man who bribed the judge. (9) Martha believes that the man who bribed the judge is honest. Clearly (9) does not follow from (7) and (8). This can be accounted for straightforwardly on the assumption that "believe" combines with the proposition expressed by the sentence in the i^wt-clause, i.e. with the whole intension of the sentence. Then the intensions and not merely the extensions of the parts can affect the truth-value of the whole containing sentence. Sentence (8) asserts that the two definite descriptions have the same extension, but clearly their intensions can differ. And since the intensions matter, (9) does not follow from (7) and (8). (This account can be seen to be essentially the

112

B. H. Partee

same as that proposed by Frege (1892): Montague took his work as formalizing and extending Frege's insights.) So ordinary referential opacity in belief-contexts of the sort discussed by Quine and others is handled in a possible worlds semantics in basically the same way as other modalities - in any intensional context, whether created by an adverb, a modal verb, a propositional attitude verb, or whatever, substitution of one expression for another with merely the same extension will not be expected to preserve truth. However, substitution of one expression for another with the same intension will be guaranteed to preserve truth in any intensional context, since in a possible world semantics, sameness of intension amounts to semantic identity. And herein lies the putative problem for the treatment of propositional attitudes, since the available data seems to give at least strong presumptive evidence that the substitution of co-intensional expressions in propositional attitudes contexts does not always preserve truth. That is, we seem in our ordinary sincere ^//^/attributions to be willing to ascribe to an agent belief in some proposition p without thereby ascribing to that agent belief in every proposition logically equivalent to p. Yet if propositions are analyzed as sets of possible worlds, all logically equivalent propositions pick out the very same set of possible worlds, so there's no way to interpret a sentence as ascribing to someone a belief in one proposition without thereby ascribing belief in the whole logical-equivalence family. There have been many reactions to this problem, and there is not space to do them justice here: besides, I suspect this issue will be a prominent one in the discussion of possible worlds in philosophy in this symposium, since it is a problem that has traditionally received more attention from philosophers than from linguists. Responses have ranged from outright rejection of the whole possible-worlds approach (but without, to my mind, any better account of propositional attitude reports being available in any other known framework), to enrichment with quotation-like devices to allow more fine-grained distinctions than can be gotten with possible worlds alone, to defense of the possibleworlds account of propositional attitude attributions with a bit of linguistic reanalysis of the sentences that seem to cause the trouble. Most recently there have begun to be attempts to combine aspects of representationalist accounts with aspects of possible-worlds model-theoretic accounts, as in discourse representation theory and some of the suggestions made within situation semantics. (iv) One of the earliest and most striking technical successes of possible worlds semantics was in the analysis of counterfactual conditionals of David Lewis (1973) and Robert Stalnaker (1968). This was mentioned briefly in section I: recall the semantic analysis given there, repeated below as (10). (10)

I i 0 0 ^ ] M ' W = l i f f [ ^ M , w ' =1

for that w' which is the closest world to w such that [p!M'w> = 1.

Possible Worlds in Model-Theoretic Semantics

113

This analysis is of general significance for possible-worlds semantics in that here it matters not just that sentences can get different truth-values in different possible worlds, but that some further structure be able to be imposed on the possible worlds, namely an ordering in terms of similarity to the actual world. This ordering is crucial in explaining some of the classic problems in the logic of counterfactual conditionals, such as the invalidity of "strengthening the antecedent", illustrated in (lla-d), which can all be true. (11) (a) If I had struck this match, it would have lit. (b) If I had struck this match and it had been wet, it would not have lit. (c) If I had struck this match and it had been wet and I had previously coated it with paraffin, it would have lit. (d) If I had struck this match and it had been wet and I had previously coated it with paraffin and the paraffin had in the meantime all worn off, it would not have lit. And so on. The possible continuing alternation of positive and negative consequents can be explained by the normal assumption that, for instance, worlds where the match is not wet are more similar to the actual world than the worlds where it is wet; worlds where I didn't coat it with paraffin are more like the actual world than the worlds where I did; and so on. As long as each successive "strengthening" of the antecedent takes us to worlds farther from the actual world than the ones we had previously had to consider, the closest worlds where the new antecedent is true will be different from the closest worlds where the previous antecedent was true, and hence the possibility arises that the consequent will take on a different truth-value there. More recent work on the semantics of counterfactual conditionals by Kratzer (198la, b, c) and others suggests that the full story may require having all of possible worlds, propositions, and situations available to appeal to, with each entering into an explanatory account in different ways. It should not be too surprising, however, if counterfactuals stretch the resources of our semantic theories to their limits, given the notorious interrelations among counterfactuals, lawlike statements, generic attributions, and the analysis of possibility and necessity. (v) The analysis of the semantics of questions and the pragmatics of questionanswer relations took a great leap forward with the advent of possible-world semantics, in my estimation, and has continued to develop and flourish. Karttunen (1977) made a very fruitful methodological move when he shifted attention away from the analysis of direct questions to the analysis of embedded questions ana framed the problem as one of finding an account of the semantics of embedded questions which could make an appropriate contribution to a compositional semantics for the full range of question-embedding constructions, ranging from ask who... to know where... to depend on, the last of which can take embedded questions for both its subject and its object, as in (12). (12) Whether John enjoys the party will depend on who is there.

114

B. H-Partee

In retrospect, it is easy to see why this shift of attention to embedded questions was so fruitful since in the case of direct questions it is much harder to distinguish the contribution of the semantics of the question from the contribution of the pragmatics of the speech act of asking (or any of the other speech acts that can employ question forms). There are still many competing analyses for the semantics of questions, and the more satisfactory ones are too complex to sketch very briefly. Let me just mention that one recent magnum opus on the topic, Groenendijk and Stokhof (1984), concludes that a satisfactory analysis of the semantics of questions must include a kind of double nesting of intensionality, or double layering of appeal to possible worlds. Expressed very informally, one might say that a question such as "Who Mary loves" is interpreted as a function which to each possible world assigns the proposition (set of possible worlds) that expresses the true and complete answer to that question in that world. The arguments for the necessity of such double-layering of possible worlds are complex and technical; but I think at the very least all would agree that the tools provided by possible-worlds theory have yielded enormous advances in the level of argumentation and the adequacy and explicitness of analyses of the semantics of questions. I think the developments in the analyses of questions have also been a good example of the value of simultaneously working to develop explicit theories of syntax, semantics, and pragmatics, and their interaction. The compositionality requirement, constraining the relation of syntax and semantics, forced careful attention to the semantics of embedded questions, which occur not only with verbs of asking but with verbs of knowing, doubting, wondering, and such miscellaneous constructions as the depend on construction. This in turn enabled researchers for the first time to begin to clearly distinguish between the semantic and pragmatic components that contribute to the interpretation of a direct question, which in turn has helped to sharpen our understanding of the relation of semantics to pragmatics more generally. (vi) As a last and quite different example, I will just mention the contribution of possible worlds to the development of pragmatics more generally, and in particular to the analysis of presupposition and of the role of context and context change in the interpretation of discourse. In fact I think it is fair to say that within the formal semantic tradition, possible worlds have been partly responsible for the shift of attention from the interpretation of sentences in isolation to the interpretation of discourses in context. This shift goes back to Stalnaker's work (Stalnaker 1974,1976,1978) and he has played a leading role in its development, with recent work by Kamp and especially by Heim (1982, 1983a, 1983b) giving it an added thrust in the linguistic community. I have already mentioned Stalnaker's notions of common ground and context set and alluded to Heim's extensions of them. The connection to the notion of presupposition is approximately as follows. From the time of Frege, and reinforced by Strawson, presuppositions of a sentence S were characterized as propositions whose truth was in some sense implied both by

Possible Worlds in Model-Theoretic Semantics

115

the truth of S and by the truth of not-S. So, for example, both (13) and (14) seem to imply the truth of (15). (13) John's children are asleep. (14) John's children are not asleep. (15) John has children. But it is difficult to incorporate such a notion of presupposition in the semantics without invoking some Kind of a 3-valued logic, and even with 3-valued logic, repeated attempts to work out a satisfactory theory of the computation of the presuppositions of a complex sentence from the presuppositions and semantic content of its parts had au been unsatisfactory in one way or another. Stalnaker (1974) had early on proposed that the phenomenon of presupposition shold be given a basically pragmatic analysis, in terms of requirements placed on the common ground in which an assertion is made. So for example, one can analyse (13) and (14) as being designed for use in discriminating among possible worlds in all of which (15) is true, and not felicitously usable in contexts in which (15) cannot be taken for granted, either by already being common knowledge or by the listeners' willingness to treat it as if it were already common knowledge. This proposal seemed plausible enough in the case of the presuppositions of simple sentences, but it aid not by itself satisfy the linguists' concern with the problem of computing the presuppositions of complex sentences; that problem seemed to require either a semantic solution or a much tighter integration, even sentence-internally, of pragmatic and semantic operations, which seemed contrary to common assumptions about the nature of pragmatics. The works of Stalnaker, Kaplan, Kamp, Kratzer and Heim mentioned earlier, and potentailly some of the ideas of Barwise and Perry for relating discourse situations, resource situations, and described situations, has gradually succeeded in convincing many scholars that semantics and pragmatics indeed have to be more closely integrated, or the boundaries between them redrawn, so that the very fabric of compositional semantic interpretation is set up not just to arrive at a set of possible worlds as the interpretation of a sentence, but as a complex function which (a) imposes constraints on the context on which it is interpreted, (b) contributes a proposition in the old sense if those constraints are met, and (c) effects certain changes in the context in the process. (This kind of approach is most clearly articulated in Heim (1982) and Heim (1983b).) There is more to context on this view than just a set of possible worlds determined by the common ground; but the possible worlds are indeed a crucial part of the context. In closing this section let me mention one possible challenge to the kinds of claims I have been making, which have by and large been claims about the beneficial effects of possible worlds theorizing on the development of linguistic semantics (and pragmatics). Aside from the last remarks about pragmatics, one might charge that all the advances I have discussed could be described in terms of

116

B. RPartee

analyses expressed in Montague's intensional logic or some variant of it, and the fact that that logic is interpreted in terms of possible worlds is inessential. Charges of this general form are sometimes leveled by representationalists who are willing to entertain something like intensional logic as a candidate language of semantic interpretation, but who remain unconvinced that the model theory is doing any real work here. This is a complex kind of charge to try to answer, though I think it is important to try. First of all, I believe the claim is only correct, if at all, for analyses carried out within a given paradigm such as Montague's intensional logic; whenever formal semanticists or philosophers argue for a need for a modification or extension of the framework, as has in fact been the case for many of the examples I have discussed here, the arguments have not been just for a new notation but for an explicit model-theoretic semantics as well - or sometimes for the latter directly without any new notation. Since Montague's intensional logic has no complete axiomatization, being explicit about the model theory is in many cases the only way to be explicit about the semantics at all. Secondly, if one looks closely at discussions of the semantics of propositional attitude sentences, one finds that notation has very little to do with the debates about interpretation. It is ironic that what seems to be the hardest problem in semantics is manifested by a syntactic construction that seems utterly straightforward and probably universal: a propositional attitude verb followed by an embedded sentence, with nothing tricky about the syntax of the embedding. In Montague's intensional logic, the semantic notation is equally straightforward: just put an intension operator on the sentence. In thinking about proposals for alternative treatments that might be more satisfactory, concern with notation plays a very subsidiary role compared to concern with what entailments one thinks should follow and where they come from - clearly semantic concerns in a model-theoretic sense. The same kinds of remarks could be made about the hard issues in the semantics of counterfactuals. So even if possible worlds turn out to be dispensable in the end, and even if appeal to possible worlds raises as many questions as it answers, I believe there is no denying that by the linguists' primary standard of fruitfulness in generating explanatory accounts of linguistic phenomena, the use of possible worlds in semantic theories has had a far-reaching positive impact, and that the role of possible worlds in that influence has been essential. IV. Foundational Issues

In this final section, I want to mention some of the important foundational issues I see facing the further development of possible-world semantics and pragmatics and the elucidation of the conception of possible worlds that can best underlie it.

Possible Worlds in Model-Theoretic Semantics

117

(1) Can possible worlds semantics be reconciled with conceptualism? If one reads just Montague or the logicians in whose tradition Montague was working, one sees semantics treated on its formal side as a branch of pure mathematics and on its empirical side as dealing with questions about the relation of language to the world, e.g. questions about the actual truth conditions of various sentences. None of this looks very inviting to a linguist who takes the study of semantics to be primarily a study of the nature of language users' semantic competence. Reading Stalnaker (1984), however, one sees more clearly that possibleworlds semantics can just as readily be viewed as a theory of the content of psychological states associated with the asserting or accepting of various sentences, though not directly as a theory of the mechanisms employed by the language users themselves. That is, it is in principle perfectly compatible with conceptualism if we take the possible worlds semantics as providing an "external" characterization of the relevant concepts and not an "internal" one. As Stalnaker has put it elsewhere (ms. 1985b), possible worlds semantics as he understands it is a framework for raising issues in philosophy, psychology, and linguistics, not a particular set of philosophical beliefs. In earlier writings I had raised some worries about the possible psychological reality of possible worlds which in retrospect I think arose partly from misguided concern with trying to fit possible-worlds theory "inside the speaker's head": but we don't try to fit the theory of gravitation "inside a falling apple". Nevertheless, linguists do care about mechanisms, and if possible-worlds semantics is a reasonable theory about the language user's semantic competence, there still have to exist some internal mechanisms partly by virtue of which the external theory is correct. Just to give one sample of the difference between external theory and internal mechanism as I now see the distinction, consider the question of how many possible worlds there might be. Even on a non-absolutist view where we don't take that question to have a single determinate answer, on many applications there will be at least non-denumerably many possible worlds. But that does not mean that a theory of internal mechanisms has to provide a way to represent each of non-denumerably many possible worlds: for our discriminations of possible worlds from each other are always at the level of discriminating certain sets of possible worlds from other sets of possible worlds. The sentences we use to express various propositions might themselves be candidates for representations oi sets of possible worlds! Though if it turned out that those were the only or the canonical representations of sets of possible worlds, one might well be inclined to wonder whether we weren't coming back full circle to a representationalist view after all. But if it is correct that sets of possible worlds serve to interpret the content of other psychological states such as perception, desire, etc., then it seems likely that internal mechanisms of representation, like the external theory, will employ some kind of medium less specifically linguistic than sentences, whether in English or in a language of thought. In any case, an internal mechanism might be classifying possible worlds into sets on just some

118

B.RPartee

finite number of parameters at the same time that a correct external description of the corresponding belief, assertion or whatever involved quantification over an infinite set of possible worlds. (2) A non-absolutist picture seems to fit linguistic semantics better than an absolutist one, where by absolutist I mean the position that there is one single maximal set (or class, if it's too big to be a set) of possible worlds. If a philosopher could find arguments that in the best metaphysical theory there is indeed a maximal set, I suspect that would for the linguist be further confirmation that his enterprise is not metaphysics, and I would doubt that such a maximal set would ever figure in a natural language semantics. As various people have noted, possible worlds are really not so different in this respect from entities: every model-theoretic semantic theory I'm familiar with takes entities to be among the primitives - but puzzles about the identity conditions of individuals and about whether there is a maximal set of all of them are just as problematic, and it is just as questionable whether semantic theory has to depend on settling such questions. As emphasized in section HI, it is the structure provided by the possible worlds theory that does the work, not the choice of particular possible worlds, if the latter makes sense at all.7 (3) "Conceivable worlds" are not the same as "possible worlds"; does concern with a conceptualist foundation for natural language semantics require that one work with the former instead of the latter? No, I don't think so. Some "conceivable" states of affairs may in fact be impossible, and some possible worlds may be beyond our powers of conception. I think that even the simple fact that we can conceive of there being possibilities we can't conceive of shows that such questions have to be approached with caution. The questions that have to be addressed in order to sort out these issues include both a further elucidation of the distinction between the external and the internal perspective on the human agent that I mentioned above, and questions about the nature of reflecting on one's own knowledge and beliefs, questions that are known to lead to deep philosophical and logical puzzles around which the scent of paradox is often hanging in the air. Furthermore, I'm not sure that the adjective "conceivable" should even be applied to individual worlds in the way that the adjective "possible" is. Given what we said above, discriminating single individual possible worlds from all other possible worlds may be entirely beyond our powers of conception, which may be limited to discriminating among sets of possible worlds. So the notions may be even more incomparable than one would have first supposed. In any case, I am convinced that it is possible worlds and not conceivable worlds that

7

This emphasis on the structures provided by the theory rather than the substance of its parts can be seen as a major motivation for the further step to "algebraic semantics" mentioned below as item (6).

Possible Worlds in Model-Theoretic Semantics

119

we should posit for must of natural language semantics, though I recognize that the relation between these notions needs much more investigation.8 (4) Current research on property theory as an alternative to set theory, such as reported in Dealer (1982), Chierchia (1982), Turner (1987), Chierchia and Turner (ms. 1985) and others, is bound to substantially reshape our thinking about possible worlds. Recall that intensions are treated in possible-worlds semantics as functions from possible worlds to extensions. But the logical equivalence problem suggests that this may not be a sufficiently intentional notion of intension. Part of the problem may lie in the fact that these functions from possible worlds to extensions are still functions in an extensional sense functions completely representable as sets of ordered pairs. What if we had a more thoroughly intensional property theory at the foundations of the whole theory, rather than an extensional set theory? If propositions were analyzed as properties of possible worlds rather than sets of possible worlds, the logical equivalence problem might not arise, at least not in the same form. On the other hand, if we had properties as primitive, and propositions as well (treating these as zero place properties), it is also less clear that we would need possible worlds as crucially as we do when we analyze propositions as sets of possible worlds and properties as functions from possible worlds to the sets of individuals that exemplify them. (5) The prepositional attitude problem, as I indicated above, is still to my mind the hardest problem in the foundations of semantics. Stalnaker (1984, ms. 1985a) has done quite a remarkable job of defending the classical possibleworlds analysis of prepositional attitude attributions against criticisms, but many linguists and philosophers remain unconvinced of the success of his attempts to explain away the logical equivalence problem. Yet repeated attempts to invoke a more "fine-grained" analysis of propositions have met considerable difficulties as well. Nevertheless, it does seem that progress is being made, so the problem should not be dismissed as utterly hopeless. One particular point to watch out for: if some other analysis of propositions, not based on possible worlds, turns out to do the best job for the semantics of propositional attitude attributions, the optimality or utility of possible worlds for other purposes will undoubtedly need to be subjected to very close scrutiny. (6) One alternative to possible-worlds semantics that I haven't discussed so far is algebraic semantics. I don't feel competent to discuss it with any authority, but as I understand the enterprise it might be viewed as a strategy or abstracting out the relevant structure in what might have started as a model-theoretic semantics, abstracting away from all properties of the original models which don't directly affect their algebraic structure. For instance, I suppose looking at propositional logic as a Boolean algebra is an example - once one sees the

But see the arguments for die contrary position in Thomas Kuhn's paper in diis volume. We bodi agree that trying to resolve diis issue takes us well beyond die scope of bodi our papers.

120

B. HLPartee

Boolean structure, it doesn't matter directly what the elements of the algebra are: all that matters are their algebraic connections to each other. Such concern with structure is quite congenial to the linguistic enterprise in general, so if possible-worlds semantics were to take such a turn I would imagine linguists would be perfectly happy to keep the structure of possible-worlds semantics while ignoring die possible worlds themselves for the most part. However, I would expect that as in any process of scientific modelling, when it came to making judgments about competing theories offering different abstractions of the relevant structure of the possible worlds, part of the basis for comparative judgments would be how faithful the abstractions were to the things being modelled, i.e. the possible worlds. But on the other hand, if the possible worlds themselves are, for semantics, just a useful theoretical posit (from the linguist's perspective at least), then a linguist's judgment would probably be based more on the predictions the theory made about judgments about various sentences than on what it claimed about the possible worlds. As far as the working linguist is concerned, possible worlds could conceivably end up going the way of caloric or phlogiston if the theories they figure in were supplanted. One might then be able to divide the "real linguists" from the "real philosophers" among those engaged in the currently very interdisciplinary work on formal semantics - the linguists would be those who could let the possible worlds go without a twinge, and without thereby denying their existence any more than current linguistic work in possible-worlds semantics depends on the metaphysical reality of possible worlds. Acknowledgements For valuable comments on the pre-symposium draft of this paper, I am particularly indebted to Raymond Turner and Fred Landman. I am also most appreciative of the comments of my two symposium discussants, John Perry and Hannes Rieser, but I have purposely refrained from making any substantive revisions in the light of their comments, saving my responses for the "speakers' replies" section. I have also benefitted greatly from stimulating conversations with Thomas Kühn on issues of common concern which the symposium gave us a delightful opportunity to discover and begin to explore. This work was supported in part by the System Development Foundation, Grant # 650 for Research on the Foundations of Semantics; their support is hereby gratefully acknowledged. References Bach, Emmon (1984), "Some generalizations of categorial grammars", in F. Landman and F. Veltman, eds., Varieties of Formal Semantics, Foris, Dordrecht (1-23).

Possible Worlds in Model-Theoretic Semantics

121

Bach, Emmon (to appear), "Natural language metaphysics", to appear in Proceedings of the 7th International Congress of Logic, Methodology and Philosophy of Science. Barwise, Jon and John Perry (1983), Situations and Attitudes, Bradford Books, Cambridge. Bealer, George (1982), Quality and Concept, Clarendon Press, Oxford. Carnap, Rudolf (1947), Meaning and Necessity: A Study in Semantics and Modal Logic, Univ. of Chicago Press, Chicago (enlarged ed. 1956). Chierchia, Gennaro (1982), "Nominalization and Montague grammar: a semantics without types for natural languages", Linguistics and Philosophy 5, 303-354. Chierchia, G. and R. Turner (ms. 1985), "Semantics and property theory", ms. Cornell University and University of Mass., Amherst. Chomsky, Noam (1981), Lectures on Government and Binding, Foris Publications, Dordrecht. Cresswell, M. J. (1973), Logics and Languages, Methuen, London. Dowry, David (1978), "Governed transformations as lexical rules in a Montague grammar", Linguistic Inquiry 9.3, pp. 393-426. Dowty, David (1982a), "Grammatical relations and Montague grammar", in P. Jacobson and G. Pullum (eds.), The Nature of Syntactic Representation, D. Reidel, Dordrecht (79-130). Fodor, Jerry A. (1975), The Language of Thought, Thomas Y. Crowell Co., New York. Fodor, Jerry A. (1981), Representations, Bradford Books/MIT Press, Cambridge. Frege, Gottlob (1892), "Über Sinn und Bedeutung," Zeitschrift für Philosophie und philosophische Kritik 100 (25-50); trans, by Max Black as "On sense and reference" in Translations from the Philosophical Writings of Gottlob Frege, ed. by P. Geach and M. Black, Basil Blackwell, Oxford, 1960 (56-78). Gazdar, G., E. Klein, G. Pullum, and I. Sag (1985), Generalized Phrase Structure Grammar, Harvard University Press, Cambridge. Groenendijk, J. and M. Stokhof (1984), Studies of the Semantics of Questions and the Pragmatics of Answers, published Ph.D. dissertation, University of Amsterdam. Hamblin, C. L. (1973), "Questions in Montague English", Foundations of Language 10, 41-53, reprinted in Partee, ed. (1976). Heim, Irene (1982), The Semantics of Definite and Indefinite Noun Phrases, unpublished Ph.D. dissertation, University of Massachusetts, Amherst. Heim, Irene (1983a), "File change semantics and the familiarity theory of definiteness" in R. Bäuerle, Ch. Schwarze and A. von Stechow, eds., Meaning, Use and Interpretation of Language, Walter de Gruyter, Berlin (164-189). Heim, Irene (1983b) "On the Projection Problem for Presuppositions" in M. Barlow, D. Flickinger, and M. Wescoat (eds.), Proceedings of WCCFL 2, Stanford, CA, 114-125. Higginbotham,James (1983), "Logical form, binding, and nominals", Linguistic Inquiry, 14,395-420. Hintikka, Jaakko (1962), Knowledge and Belief, Cornell Univ. Press, Ithaca. Hintikka, Jaakko (1973), "Carnap's heritage in logical semantics", Synthese 25, 372-397. Janssen, Theo M. V. (1984), "Individual concepts are useful" in Fred Landman and Frank Veltman (eds.) Varieties of Formal Semantics: Proceedings of the 4th Amsterdam Colloquium, September 1982,

Dordrecht: Foris Publications. Johnson-Laird, Philip N. (1983), Mental Models: Towards a Cognitive Science of Language, Inference, and Consciousness, Harvard University Press, Cambridge. Jubien, M. (ms. 1985), "Ist-Order Property Theory" manuscript, University of Massachusetts. Kamp, Hans (1981)," A theory of truth and semantic representation" in J. Groenendijk, Th. Janssen, and M. Stokhof, eds., Formal Methods in the Study of Language, Parti. Mathematisch Centrum, Amsterdam pp. 277-322. Kanger, Stig (1957), 'The Morning Star paradox", Theoria 23, 1-11. Kanger, Stig (1957), Probability in Logic, Stockholm Studies in Philosophy, vol. 1, Stockholm. Kaplan, David (1977), "Demonstratives" read in part at the March 1977 meeting of the Pacific Division of the American Philosophical Association; "Draft #2", ms. UCLA. Karttunen, Lauri (1977), "Syntax and semantics of questions", Linguistics and Philosophy 1, 3-44. Katz, Jerrold (1981), Language and Other Abstract Objects, Rowman and Littlefield, Totowa, New Jersey

122

B. H. Partee

Katz,Jerrold and Jerry Fodor (1963), "The structure of a semantic theory", Language^*), 170-210. Keenan, Edward L. and Leonard M. Faltz (1985), Boolean Semantics for Natural Languages, Dordrecht: Reidel. Kratzer, Angelika (1978), Semantik der Rede. Kontexttheorie, Modalwörter, Konditionalsätze, Scriptor, Königstein. Kratzer, Angelika (l 981 a), "The notional category of modality", in H. Eikmeyer and H. Rieser (eds.), Words, Worlds, and contexts: New Approaches in Word Semantics, de Gruyter, Berlin. Kratzer, Angelika (1981b), "Blurred conditionals", in W. Klein and W. Levelt, eds., Crossing the Boundaries in Linguistics. Dordrecht, Reidel (201-209). Kratzer, Angelika (1981c), "Partition and revision: the semantics of counterfactuals", Journal of Philosophical Logic 10, 201-216. Kripke, Saul (1959), "A completeness theorem in modal logic", Journal of Symbolic Logic 24,1-14. Kripke, Saul (1963), "Semantical considerations on modal logics", Acta Philosophica Fennica 16, 83-94. Kühn, Thomas (to appear), "Possible worlds in history of science", in S. Allen, ed., Possible Worlds in Humanities, Arts, and Sciences: Nobel Symposium 65, Walter de Gruyter, Berlin, [this volume] Landman, Fred (1985), "The realist dieory of meaning", Linguistics and Philosophy 8, 35-51. Lewis, David (1969), Convention: A Philosophical Study, Harvard Univ. Press, Cambridge, Mass. Lewis, David (1970), "General semantics", Synthese 21; reprinted in D. Davidson and G. Harman, eds., Semantics of Natural Language, D. Reidel, Dordrecht, 169-218. Lewis, David (1973), Counterfactuals, Harvard University Press, Cambridge, Mass. Lewis, David (1975), "Languages and language", in K. Gunderson (ed.), Language, Mind, and Knowledge, University of Minnesota Press, Minneapolis. May, Robert (1977), The Grammar of Quantification, Ph.D. dissertation, MIT, Cambridge. May, Robert (1985) Logical Form: Its Structure and Denvation, Press, Cambridge. Montague, Richard (1968), "Pragmatics", in R. Klibansky (ed.), Contemporary Philosophy: A Survey. La Nuova Italia Editrice, Florence, pp. 102-122. Reprinted in Montague (1974). Montague, Richard (1970a), "Pragmatics and intensional logic", Synthese 22, 68-94. Reprinted in Montague (1974). Montague, Richard (1970b), "English as a formal language", in B. Visentini et al, eds., LinguaggiNeUa Societa e NelL· Tecnica, Edizioni di Comunita, Milan. Reprinted in Montague (1974). Montague, Richard (1970c), "Universal grammar", Theoria 36, 373-98. Reprinted in Montague (1974). Montague, Richard (1973), "The proper treatment of quantification in ordinary English", in Hintikka et al, eds., Approaches to Natural Language, D. Reidel, Dordrecht. Reprinted in Montague (1974). Montague, Richard (1974), Formal Philosophy: Selected Papers of Richard Montague, edited and widi an introduction by Richmond Thomason, Yale University Press, New Haven. Partee, Barbara (1974), "Opacity and scope", in Milton Munitz and Peter Unger, eds., Semantics and Philosophy, New York University Press, New York, 81-101. Partee, Barbara (1975), "Montague grammar and transformational grammar", Linguistic Inquiry 6, 203-300. Partee, Barbara, ed. (1976), Montague Grammar, Academic Press, New York. Partee, Barbara (1979), "Semantics - Madiematics or Psychology?", in R. Bäuerle, U. Egli, and A. von Stechow (eds.), Semantics from different points of view, Springer-Verlag, Berlin. Partee, Barbara (1980), "Montague Grammar, mental representation and reality", Philosophy and Grammar ed. by S. Kanger and S. Öhman, D. Reidel, Dordrecht: Boston. Partee, Barbara H. (1984), "Compositionality", in F. Landman and F. Veltman, eds., Varieties of Formal Semantics. Proceedings of the 4th Amsterdam Colloquium, September 1982, Foris, Dordrecht. Partee, Barbara and Mats Rooth (1983), "Generalized conjunction and type ambiguity," in R. Bäuerle, C. Schwarze, and A. von Stechow (eds.), Meaning, Use and Interpretation of Language, Walter de Gruyter, Berlin, 361-383. Pollard, C. (1984), Generalized Phrase Structure Grammar, Head Grammar, and Natural Language, unpublished Ph.D. dissertation, Stanford University.

Possible Worlds in Model-Theoretic Semantics

123

Quine, W.V.O. (1960), 'Word and Object, MIT Press, Cambridge. Saarinen, Esa (ed.) (1979), Game-Theoretical Semantics, Reidel, Dordrecht. Stalnaker, Robert (1968), "A theory of conditionals", in N. Kescher, ed., Studies in Logical Theory, Blackwell, Oxford, 98-112. Stalnaker, Robert (1974), "Pragmatic presuppositions" in M. Munitz and P. Unger, eds., Semantics & Philosophy NYU Press, New York, 197-213. Stalnaker, Robert (1976) "Propositions", in A. Mackey and D. Merrill, eds., Issues in the Philosophy of Language, Yale Univ. Press, New Haven, 79-91. Stalnaker, Robert (1978), "Assertion", in Peter Cole, ed., Syntax and Semantics, Vol. 9: Pragmatics, Academic Press, N.Y., 315-332. Stalnaker, Robert C. (1984), Inquiry, Bradford Books/ΜΓΓ Press, Cambridge. Stalnaker, Robert (ms. 1985a), "Belief attribution and context", paper read at Oberlin Colloquium in Philosophy 1985. Stalnaker, Robert (ms. 1985b), "Possible worlds and situations", paper read at symposium on possible worlds, meeting of the Assn. for Symbolic Logic, Stanford, July 1985. Tarski, Alfred (1935), "Der Wahrheitsbegriff in den formalisierten Sprachen", Studia Philosophica 1; English translation in Tarski (1956). Tarski, Alfred (1956), Logic, Semantics and Metamathematics, Oxford, Clarendon Press. Thomason, R. (1976), "Some extensions of Montague grammar", in B. Partee, ed., Montague Grammar, Academic Press, New York, 77-117. Turner, R. (1987), "A theory of properties", Journal of Symbolic Logic 52.2. Turner, R. (to appear), "Formal semantics and type-free theories", to appear in J. Groenendijk et al., Information, Interpretation, Inference, GRASS 7, Foris, Dordrecht.

JOHN PERRY

Possible Worlds and Subject Matter Discussion of Barbara H. Partee's paper "Possible Worlds in Model-Theoretic Semantics: A Linguistic Perspective"

1. Introduction

Professor Partee emphasizes two contributions possible-worlds semantics has made to linguistics.1 First, possible worlds have helped to provide an appropriate structure on the space of meanings. Second, the possible-worlds conception helps relate linguistic meaning to other kinds of informational content.2 (108)3 She believes that these contributions are 'largely independent of metaphysical issues'. (107) I agree that possible worlds have helped to provide a MORE appropriate structure on the space of meanings than that provided by extensional semantics. But I do not think the structure it provides is fully adequate to the needs of a The research reported in this paper was supported by a grant from the System Development Foundation to the Center for the Study of Language and Information. Many at the center have contributed to the paper through discussion on possible worlds and related topics. Of course my views about possible worlds have been heavily influenced by Jon Barwise. John Etchemendy, David Israel, Ken Olson, and Dagfinn Follesdal also provided many ideas and useful suggestions. I have one reservation about how Partee puts the point about the benefits for linguistics of a general account of meaning and information. She says this benefit has not so much to do with linguistics proper, but with the linking of linguistics to other disciplines. But this general conception of meaning and information is needed within linguistics proper, too, in at least two ways. First, the concept of meaning that is used by formal semanticists should connect up with the concepts of meaning used in the rest of linguistics - in the study of meaning change, for example. Second, it seems plausible that die information content that phonological, morphological, and syntactic aspects of an utterance contain about each other should be understood widiin the same general framework of meaning and information as the content of the utterance as a whole. Thus, a substantive theory of meaning and information should be useful not only for understanding how linguistics connects with other disciplines, but for understanding how the parts of linguistics connect with each other. Numbers in parentheses refer to [Partee 88]. I think it should be mentioned that Barbara Partee deserves considerable credit for seeing the importance of Montague's theories, helping her students and colleagues to understand the techniques necessary to appreciate his work, and contributing a steady stream of ideas of her own to the development of possible-worlds semantics. Without Barbara Partee's work the impact of Montague's work and model-theoretic techniques on linguistics would have been much less than it had been.

Possible Worlds and Subject Matter

125

general theory of informational and intentional content. The basic problem is that the notion of a proposition that is suitable for the special case of modality is not suitable for a general theory of intentionality. In this paper, I argue that versions of possible-worlds semantics that limit themselves to the structure of intensions needed for the semantics of modality face serious problems, and that the assumptions at the root of these problems are metaphysical. The problems I list are instances of what I shall call "the subject-matter problem." These problems are not new. They all are instances of a category of problems that Partee describes as the crucial ones facing possible-worlds semantics. (119) These problems are presented by contexts in which substitution of cointensional statements does not preserve truth of statements in which they are embedded. I think the force of the subject-matter problem has not been fully appreciated, however, because it has been assimilated to another class of problems in the same category, which are not specific to strong possible-worlds theory. This is the "mode-of-presentation" or "Cicero-Tully" problem. I argue that the subject-matter problem is distinct, and related to the metaphysical principles characteristic of strong possible-worlds theory. I end by arguing that there is no very strong motivation for the metaphysical assumptions that underlie possible-worlds semantics, and hence every reason to explore alternative frameworks that do not rest on these assumptions and face the difficulties they bring. 2. Strong versions of possible-worlds theory The versions I have in mind share the following assumptions: 1. Possible worlds are total possiblities. 2. Intensions may be set-theoretically defined as functions from possible worlds to extensions. 3. Possible worlds and the intensions definable from them provide us all that we need for a theory of informational and intentional content. David Lewis [Lewis 73], and Robert Stalnaker [Stalnaker 84, Stalnaker 85] are among those who hold strong versions of possible-worlds theory. But these two theorists are in radical disagreement as to the metaphysical nature of possible worlds. Lewis asserts the existence of an infinity of possible worlds, of which our world, the one we call 'the actual world', has no special status other than our presence. In Lewis's theory, truth is relative to possible worlds in just the way it is relative to times and places and inertial frames. Possible worlds are simply an often unnoticed parameter of all empirical relations. Stalnaker doesn't believe in these other real worlds, but he does believe in other ways things might have gone, other ways the one actual world might have been. These are what be calls possible worlds.

126

J. Perry

Principles l)-3) do not, then, really delineate a metaphysical doctrine, but only a metaphysical schema - something that provides a doctrine, once we are told what a possible world is. Even so, it may be the case that the various strong versions have enough in common for there to be common problems for them. I shall argue that this is the case. Before considering the objections, however, let us say a bit more about what principles l)-3) come to. 2. /. Possible worlds are total possibilities Both actualists and possibilists hold that possible worlds are 'total' or 'complete', but actualists have a little more flexibility here than do possibilists. Let us say that for any η-place relation and sequence of n appropriate objects there is an issue, which has one and only one answer (or one per world, at any rate), Yes or No. (This is just an apparatus for discussing things; neither the reader nor the scholars discussed will DC assumed to be committed to it without warning.) Possible worlds supply answers to issues, somehow or other, depending on the conception of what a possible world is. By saying that possible worlds are total, I mean that each possible world provides an answer to each issue in the relevant universe of issues. I distinguish being total from being comprehensive, which is to provide an answer to every issue, period. Actualists may deny that worlds are comprehensive, while maintaining that they are total. The idea is that for a given analytical project, only some issues may be relevant; besides, there may not be a complete set of issues. [Stalnaker 84] So, each possible world in a 'partitioning of the space of possibility' will provide answers to all of the issues that the others in the same partitioning do. 2.2. Intensions may be set-theoretically defined in terms of possible worlds and individuals This may seem like a truism, for 'intension' is now widely used to mean just a function from possible worlds to extensions. But the word 'intension', and the view that language was basically intensional, is older than possible-worlds semantics. Basically, intensions are entities that provide some principle of classification, and that have an identity, independently of the objects so classified. So the intension of a term, which determines whether is describes an object or not, is contrasted with its extension, the set of objects it describes. Properties are intensions, for we can classify objects into those that have them and those that do not. Concepts are intensions, for we can classify objects into those they fit and those they do not. Words are intensions, when we consider them with their meanings intact, for we can classify objects into those they describe or are true of, and those they do not describe or are not true of. The notion of an intension can be found in a wide variety of theories; we might say, for example, that the

Possible Worlds and Subject Matter

127

classical debate between nominalists, conceptualise, and realists pertained to which intensions were needed. The idea that most discourse was 'intensional', in the sense that intensions, and not just the actual extensions determined by them, were relevant to truth, is quite old. In the introduction to the first edition of Prindpia Matbematica, for example, Russell holds that it is the extensionality of mathematical discourse that requires explanation. The thought that all intensions are set-thoretical constructions out of possible worlds is thus a substantive thesis. It is one that many philosophers reject who nevertheless find possible worlds good entities to have when thinking about theological, metaphysical, and semantical issues connected with necessity. Adams [Adams 74] and Plantinga [Plantinga 74], for example, fall into this category. Thus they are not adherents of strong possible-world semantics, in my sense. 2.3. Possible worlds and the intensions definable from them provide us all that we need for a theory of informational and intentional content Informational and intentional notions are often expressed by constructions that embed statements, as in "Harold said that Mary was crying," "Tom knows that Mary wasn't crying," and "The tree rings indicate that the tree is over one hundred years old." Philosophers commonly suppose that such statements express propositions, and that the propositions expressed by the embedded statements are the objects of belief, assertion, knowledge, etc. From this point of view, providing a satisfactory notion of a proposition is a key test for strong possible-worlds semantics. Strong possible-worlds theorists take propositions to be the intensions they assign to statements, functions from possible worlds to truth-values. I shall argue that this notion of a propostion is inadequate, and that its inadequacies are related to the metaphysics that leads to strong possible-worlds theory. 3. Two problems Functions from possible worlds to truth-values surely provide a more appropriate structure for the space of meanings of statements than do truth-values by themselves. By taking propositions to DC such functions, and to be the semantic values of statements embedded in modal and intentional contexts, we can explain the nonextensionality of those contexts. Still, one might well suspect that such propositions, motivated by the needs of modal contexts, might prove less than completely successful for intentional contexts generally. In particular, the possible worlds treatment of propositions requires that propositions that are necessarily equivalent are identical, and one might suspect that this notion is too coarse-grained for many purposes. It is

128

J. Perry

natural to suppose that propositions that are true in the same possible worlds have their modal properties in common, so that exchange of logically or necessarily equivalent statements in modal contexts preserve truth. But it is not at all obvious that propositions true in the same possible worlds have all properties in common. So it seems that this way of individuating propositions might give us problems, in dealing with nonmodal contexts. A variety of cases, in which substitution of cointensional statements within an informational or intentional context does not seem to preserve truth-value for the embedding statement, appear to justify this worry. Partee identifies such cases as the crucial problem facing possible-worlds semantics. (119) I maintain that me problems in this category fall into two quite different subcategories. Problems in the first category are not specific to strong possibleworlds theory, while those in the second category are. 3.1. The mode-of-presentation problem The first kind of problem involves statements that differ only in the proper names used to designate some individual, such as "Tully was a Roman orator" and "Cicero was a Roman orator." Such statements will be cointensional. Given that Tully and Cicero are the same person, the possible worlds in which Tully was a famous orator will be just those in which Cicero was a famous orator. But substitution of one of these statements for the other in reports of belief, knowledge, assertion and the like doesn't always preserve truth-value. In certain circumstances, we would be very reluctant to infer from "Smith believes that Cicero was a Roman orator" to "Smith believes that Tully was a Roman orator." I call this the "mode-of-presentation problem." The change from "Cicero" to "Tully" does not alter who we are talking about, only how that person is presented. This problem is not specific to possible-worlds theories. Any semantic theory that takes propositions to be nonlinguistic intensions, so that the interchange of two codesignational names determine the same proposition, has the same problem, whether propositions are taken to be states of affairs or worlds, sets or functions, primitive or constructed, partial or total. A natural reaction to this problem is to suppose that somehow language is relevant to our reluctance to allow substitution in these contexts. After all, language is explicitly involved in assertion, and is intimately associated with the higher cognitive states and activities reported by "believes" and "knows." One might suppose that the objects of these attitudes are not propositions at all, but some linguistic entity or a hybrid. Or, taking the assertion case as one's model, one might suppose that there is an implicit parameter for linguistic entities. One does not simply say that so-and-so, one does so by uttering a certain statement in a certain context. Our reluctance to substitute one name for another might be due to the effect on the implicit parameter, rather than on the proposition.4 See [Barwise and Perry 83], pp. 262 if.

Possible Worlds and Subject Matter

129

While no solution along any of these lines has won universal acclaim, the line of attack is plausible enough to make it unreasonable to reject the possible worlds notion of a proposition because of the mode-of-presentation problem. The second kind of example presents a problem that is specific to strong possible-worlds semantics, however. I call this the "subject-matter problem." 3.2. The subject-matter problem We have before us a fairly coherent and intuitive notion of what things various propositions are about. If you ask me to tell you about Martha, and I say, "The sea is salty" or "I really like macadamia nuts", you will think I am dotty, or run and reread Grice on conversational implicature in hopes of finding some clue as to what I am getting at. The propositions that the sea is salty and that I like macadamia nuts are not about Martha, even though, if they are true, Martha lives in a world in which the sea is salty and in which I like macadamia nuts. The notion of aboutness may not be as crucial as belief or necessity, but it's quite useful and we have pretty clear intuitions about it. It think the following informally stated conditions could be considered the beginnings of a theory of aboutness: a) A simple statement that contains a name for an individual expresses a proposition that is about the individual named. For example "David is clever" expresses a proposition about David; "Ruth threw Paul over the fence" expresses a proposition about Ruth and Paul. b) If P is about a, then not-P is about a,. c) If P is a proposition about a and Q is about b, then P and Q is about a and about b. So, "David is clever and Ruth threw Paul over the fence" expresses a proposition that is about Ruth and Paul and David. a) If P is about a and Q is about a, then PorQ'is about a. So, "David is clever or Ruth threw David over the fence" expresses a proposition that is about David, whether or not it is about Ruth. e) If P is about a, and P= Q, then Q is about a. Item d) is simply an application of the indiscernibility of identicals. But it creates a problem for strong possible-worlds theory. For, given the coarsegrained way in which possible-worlds theory individuates propositions, we get the consequence that every proposition Q is about every object that any proposition P is about. (1) Assume that P is about a. (2) P or not-P is about a, by b) and d) (3) Q and (P or not-P) is about a, by c) (4) Q = Q and (P or not-P), by the possible-worlds theory of propositions. (5) Q is about a, by e). Which is to say, strong possible-worlds semantics cannot capture the notion of a proposition's being about an object.

130

J. Perry

The problem can be traced to the mechanism, in strong possible-worlds semantics, of obtaining propositions that focus on any particular matter of fact, such as whether Peter is picking peppers. Each individual possible world provides an answer to every such issue. One focuses on a particular issue, by canceling out answers to other issues. The proposition that Peter is picking peppers is the set of possible worlds that give an affirmative answer to that issue, and cancel out each others' answers to every other issue. One possible world has Sally selling seashells by the seashore; another one has her selling peaches instead, and so forth. The only thing that they agree about, is the target issue, that Peter is picking peppers. That, at any rate, is the strategy, but it doesn't quite work. Some issues, such as whether 7 and 5 add up to 12, have only one possible answer. And other complex things, such as that Sally is either selling seashells by the seashore or she is not, will be true throughout the set of worlds. Since they are true in all worlds, they cannot be separated one from another; since they filter out no worlds, the conjunction of statements that express them with simple empirical statements are true in the same worlds as those simple empirical statements by themselves. A natural reaction to this problem is to suppose that the real culprit is not the possible-worlds notion of a proposition, but, as with the mode-oi-presentation problem, some implicit linguistic parameter. One might suppose that propositions are really not about anything in and of themselves (just as the possibleworlds theory predicts), but only relative to some statement by which they are expressed, and that when we work out the nature of this relativity to language, the coarse-grained nature of the possible-worlds propositions will be no problem. This approach is exemplified in Stalnaker's treatment of attitudes towards mathematical truths in Inquiry. However, I do not think that this defense works. We have basically two paradigms of nonextensional, statement embedding contexts. The first, modal contexts, are insensitive to shifts in mode-of-presentation and shifts in subject matter. The second, the cognitive attitudes, are sensitive to both. These latter are intimately involved with language, which suggest that the sensitivity may be due to an implicit linguisticparameter, rather than with the coarseness of possible-worlds propositions. But a great many informational and intentional concepts fit neither of these paradigms. They are insensitive to shifts in mode-of-presentation, but sensitive to shirts in subject matter. These concepts typically do not have much of anything to do with language, so that the strategy of protecting coarse propositions, by finding an implicit linguistic parameter, is not very promising. I now turn to a number of examples of such concepts. 3.3. Mare problems These examples canvass a broader range of states and activites that are usually considered in discussions of intentionaüty. One the one hand, a number of the examples have to do with information, what an inanimate object indicates or

Possible Worlds and Subject Matter

131

shows. On the other, a number have to do with action, what an intelligent being does in virtue of the way it moves. The important connection between intentional notions and informational notions has been stressed by Dretske, and that between intentional notions and action is part of the functionalist perspective, whose relevance to semantical issues Stalnaker has stressed. 3.3.1. Causation ana action Caesar brought it about that Tully fell out of bed. Caesar made Tully fall out of bed. The collapse of a defective bedpost caused Tully to fall out of bed. In each case, the substitution of "Cicero" for "Tully" will not have any conceivable effect on the truth-value of the embedded statement or clause. A man made to fall out of bed by any other name is still a man made to fall out of bed. Note here that we are not talking of what a man tried to do or thought he was doing or formed an intention to do, but just what he is fact did. It seems quite odd to suppose, say, that just because Caesar made Tully fall out of bed, he made Tully fall out of bed and7+5=l2, or that he made Tully fall out of bed and Peter pick or not pick his peck of pickled peppers. It is true that when a person performs an action the whole world, in some sense, changes. One might then think of modeling the results of an action as the set of worlds that might be actual given that it is done. I think it is clear, however, that some more incremental notion underlies the way we actually think about the effects of action. We are interested in those things about the world that would not have been as they were, without the action: the issues, on whose answers the action had some effect. That 7+5=12, that Peter did or did not pick his peck of peppers, and much else about the world is unaffected by Caesar's overturning of Cicero's bed. 3.3.2. Perception ana memory Harold saw Mary cry. Harold remembers Mary crying. We use a variety of constructions for perception and memory, with interestingly different properties. The statement that Harold saw that Mary was crying, seems sensitive to exchange of names, as does the statement that he remembers that she was crying. There are many contexts in which we would use these locutions, in which the information Harold sees, or needs to produce, is linguistic in form. In these cases, the exchange of names seems to imply a difference in abilities to interpret or produce information. But the constructions we have used do not seem sensitive in this way. Any inclination to say that Harold saw or remembers 'under a description', seem to disappear once we have reminded ourselves that we are speaking of seeing and remembering, not seeing-that and remembering-that.

132

J. Perry

Yet it would be strange indeed, to reason from the fact that Harold saw Mary cry, to the conclusion that he saw her cry and Peter pick his peppers or not. After all, if Harold saw Mary cry and Peter pick his peppers or not, then he saw Peter pick his peppers or not pick them. And if Harold saw Peter pick his peppers or Peter not pick his peppers, then he saw one or the other. But we cannot conclude, from the fact that Harold saw Mary cry, that he either saw Peter pick his peppers or not pick them.5 Analogous reasoning applies to "remembers". 3.3.3. Information and meaning These tree rings show that this tree is more than one hundred years old. The height of the mercury in the thermometer indicates that Freddy has a fever. In the old days, "Ken is starving" meant that Ken was dieting. "IV + ΠΙ = VIF means that four plus three equals seven. The principle, that if X shows that P & Q, then X shows that P and X shows that Q, seems pretty plausible. So, given the possible-worlds notion of a proposition, everything shows that R, for any necessary truth R. One might respond, "Why not? It seems pretty harmless to let tree rings show 'the necessary proposition'." The problem again is that our ordinary devices for dealing with information, and for dealing with devices, like ourselves, that deal with information, are interested in increments. In the typical case we have a system that can vary along a certain dimension or dimensions; in this case, the number of rings the tree has. This systematically depends on, and so carries information about, some other dimension of variation of the system or its environment. Our concept of information, of what is shown, wants to focus on these variations and dependencies. Our ordinary notion of a proposition, what is said or shown or indicated, seems to have built into it a notion of subject matter, of focus on definite objects and issues about them. This seems most clear in the case of meaning, a notion with which possibleworlds semantics must come to grips, if it is to provide a framework in which the activities of the natural-language semanticist cohere with empirical studies about meaning conducted in other parts of linguistics. We are interested in the increment the truth of a statement makes to the totality of what is true, not that totality. It is not correct that in the old days "Ken is starving" meant that Ken was dieting and 7+5=12.

See [Barwise 81, Barwise and Perry 83].

Possible Worlds and Subject Matter

133

3.3.4. Necessary truths and conditionals

The principle that there is only one necessary proposition causes havoc with conditionals. If this is Stockholm, we are in Sweden. If Nixon bet on ODD and won when the ball landed on 7, then 7 is an odd number. If 387 is a prime number, Jack owes Jill $5. Again, there is no way that exchange of codesignative names could affect the truth-value of these statements. But, at least to untutored intuitions, changing the subject matter makes quite a difference, even if the resulting proposition is necessarily equivalent to the original. Suppose the first statement is made when a plane lands in Oslo. The antecedent is then false, and necessarily so. But in possible-worlds semantics, there is but one necessarily false proposition, so we should be able to exchange this way of expressing it for any other. But this seems clearly wrong; the statement in question seems true, while, say, "If 7+5=13, then we are in Sockholm" does not seem true at all. One might point out that, on most possible-worlds analyses of conditionals, conditionals with necessarily false antecedents are true, so this second conditional would also be true. But this latter doctrine seems just more reason to look for some other notion of a proposition. The second conditional seems true enough, out if we replace the consequent with, say, "41 is a prime number," the conditional doesn't strike me as true at all. The third conditional seems like the sort ofthing that becomes true when Jill bets Jack that 387 is a prime number. But, given that 387 is not a prime number, so that the antecedent is a necessarily false proposition, the conditional is true on the standard possible-worlds analyses, quite independently of any wager between Jack and Jill, which seems quite implausible. It can seem odd that things might depend on one necessary truth, but not depend on others. Odd or not, people talk this way all the time. Logicians who teach less mathematically oriented philosophers are always saying things like, "That depends on the fact that addition is associative." This is not confused, although it can be confusing. As Jon Barwise has emphasized [Barwise 85], use of conditionals in mathematical pursuits is not an isolated language game, but continuous with their use for other purposes. The notion of prepositional content bequeathed to us by strong possible-worlds semantics is simply not flexible enough to cope with such conditionals. All of the problems I have mentioned are but putative problems; that is, one way or the other of treating them within the framework of strong possibleworlds semantics may, in the end, when all the costs and benefits of the approach are summed, be reasonable. My point is that it is not semantics, or possible worlds, but only the strong view that creates the problems in the first place, so the approach needs to have some benefits to counterbalance the costs.

134

J. Perry

4. Towards intensional tolerance The source of these problems is the basic thesis of strong possible-worlds semantics, that all intensions are to be built up from total possible worlds. If we drop this thesis, we could take propositions to be more 'fine-grained' entities. There are a number of ways of doing this available to the tolerant intensionalist. Here is a simple one, designed to keep as close as possible to the possible-worlds idea of taking propositions to be sets of possibilities. We start by taking relations and objects as primitives, rather than possible worlds. From these we can build up 'issues'; the answers can be 0 and 1 or any suitable set-theoretic entities. We can introduce worlds as further primitive entities, associating with each total functions from issues (whatever universe of issues is relevant to our linguistic or philosophical project) to answers. Call this the world's 'issue profile'. Then we can focus on simple facts by subtraction rather than addition. That is, augment the total functions from issues to answers with partial functions, which only provide answers for some of the issues. The limiting case would be a function defined only on a single issue, say whether Peter was picking peppers. We introduce entities, perhaps 'partial worlds' or 'partial ways' with these partial functions from issues to answers as their profiles. Then, we take propositions to be functions from the worlds plus these new partial entities to truth-values.6 Where an issue consists of the relation R and the sequence of objects a l v . .,an, I shall say that the members of the sequence are constituents of the issue. The constitutents of a (partial or total) world w are the constituents of the issues on which the associated function is defined. A proposition is about those objects that are constituents of every (partial or total) world that is a member of the proposition. So, the proposition expressed by "Peter is picking peppers" will consist of all (partial or total) worlds that provide the affirmative answer to this issue, including the limiting case, the one defined only on this issue. The proposition that Peter is picking peppers will be about Peter, but not about anyone else. This suggestion fits with the intuitive principles about aboutness that we noted above. A conjunction will be about the objects the conjuncts are about, since the proposition expressed by the conjunction will be the intersection of those expressed by the conjuncts. But a disjunction may not be about the objects

Partee claims (p. 104) that Stalnaker shows in the ms. of "Possible Worlds and Situations" that "there is no obstacle in principle to identifying a situation with the set of all possible worlds that have that situation as a part." The difficulties in question stem from the basic strategy of modeling something partial with a set of totalities. Some of them were spelled out in "The Situation Underground." Tore Langholm is exploring the differences in logics that result from dealing directly and indirectly with partiality in his dissertation project at Stanford. For more on the relation between situations and partial possible worlds see [Perry 86].

Possible Worlds and Subject Matter

135

the disjuncts are about, since the ways in the union of two propositions may not all be defined on issues with the constituents the original propositions were about. The more fine-grained notion of a proposition we have just considered does nothing directly to solve the problems connected with the mode-of-presentation problem. But it does solve the problem of subject matter, in that it blocks the inferences which depended on the strong possible-worlds treatment of propositions. We need not infer that Caesar made it the case that Peter was or wasn't picking peppers, simply because he played a trick on Tully, or the tree rings indicate all of the truths of logic and mathematics, simply because they indicate the age of the tree. Conditionals can be taken to express relationships between partial worlds, and not simply total worlds (see [Barwise 85]). It is, then, a partial solution in two senses, to some putative problems for possible-worlds semantics. I am not claiming, of course, that the apparatus just sketched is one that will meet all of our needs. My point is simply to suggest how limiting the strong possible-worlds position is. The problems that derive from the identity of necessarily equivalent propositions are not inevitable or based on any principle of technique or methodology; they are simply the results of a metaphysical commitment to reduce all intensions to functions on the domain of total possible worlds.

5. Motivations for strong possible-worlds semantics So what motivation is there for strong possible-worlds semantics? Here, again, there seems to be a large difference between the possiblilist and the actualist. Although possiblilism is often characterized as some sort of extreme modal realism, there is a pretty clear sense in which the possibilist is not, or need not be, a realist at all. According to Lewis, for example, possible worlds are not properties or relations or universale or abstract entities of any kind. So possible worlds are not intensions. Traditional intensions, or replacements for them, are built out of possible worlds and other nonintensions. These 'new' intensions are all just sets of various sorts. As far as I can see, there is nothing in the ontology of Lewis' theory that someone like Quine should object to, except the possible worlds. Each of them is a desert; it just turns out that there are more deserts than one counted on. From the perspective of someone who likes desert landscapes, but is not put off by needing a lot of them, strong possible-worlds semantics is perfectly natural. The problem of change of subject matter is difficult to avoid, because the only intensions he can fairly avail himself of are those built up out of possible worlds. But these difficulties might be fair price to pay for a theory that can be at once so scrupulous about abstract entities and generous in providing interpretations for intensional constructions. It all depends on how one feels

136

J. Perry

about alternative concrete realities, about living in a possibluum of equally intrinsically real worlds as well as a continuum of equally real locations.7 For the possibilist, then, possible worlds are not intensions at all, but concrete objects. For an actualist, possible worlds are intensions of a special kind, sort of like a property of the entire universe. There is clear motivation for recognizing this new sort of intension, for they deal well with necessity and possibility. But what is the motivation for an actualist to adopt strong possible-worlds semantics, forsaking all intensions but these? A possible reason is that one believes that possible worlds are in some metaphysical sense the fundamental intensions. Stalnaker disavows this reason. He points out that taking possible worlds as primitive and other intensions as derivative, in one's formal semantics, does not preclude thinking that possible worlds are definable or analyzable in terms of other sorts of entities, and that the possible worlds needed for various analytical enterprises might be analyzable in various ways (see [Stalnaker 85]). A reason Stalnaker does give, is that by working in a framework basically the same as that of the possibilist, technical points can be debated without getting enmeshed in philosophical issues. This point is not compelling. Lewis and Stalnaker agree that conditionals describe an abstract relation between total worlds. One can hardly compare their theories in a system that has no total worlds, but there seems to be no motivation for excluding partial worlds. Stalnaker's most important reason, which he has advanced and defended in a number of places, has to do with intentionality and the nature of informational content: The picture is this: to have the capacity to represent, an organism must be capable of being in each of a set of alternative states ... which will, under ideal conditions, reflect corresponding states of the enviornment... the state of the environment tends to cause the organism to be in its state... Now we can ask, what must informational content be if this is what it i.s for an organism to be in states that have informational content? What is essential to contentful states is that they distinguish, in some way or other, between alternative possible states of the world ... what any representer must do - what it is to represent - is to locate the world in a space of alternative possible states of the world. It is appropriate to begin with possibilities because that is the level of abstraction that captures what is essential to representation. [Stalnaker 86]

This argument seems to beg the question. If we use total possible worlds to model environmental states, then this picture will lead us to use total possible worlds to model intentional states. But if we use some partial technique for the former, we may also do so for the latter. Consider a simple case. The waiter brings it about that there is a milk shake in front of me. I see that there is a milk shake in front of me, believethat there is, and drink the milk shake. We can represent what the waiter brings about as the set of total possible worlds, whose issue profiles disagree about everything except I would feel better about rejecting this out of hand, if I understood better how equally real I think the past and future are, and why.

Possible Worlds and Subject Matter

137

necessary truths and the relative position of me and the milk shake. Or we can represent it by the set of partial worlds, whose issue profiles have the milk shake in front of me. Whichever we choose to model the state of my environment, we can use to model my intentional states. Choosing sets of total worlds, the route of strong possible-worlds semantics, will enmesh us in the putative problems listed above. Choosing to use partial objects, will mean that our theory of informational content cannot simply rely on the structures bequeathed to it by the semantics of modal logic. The picture sketched by Stamaker does not provide an argument for the first choice over the second. References Adams, Robert M. 1974, Theories of Actuality. Nous 5. Barwise, Jon. 1981, Scenes and Other Situations. The Journal of Philosophy LXXVTII(7):369-97. Barwise, Jon. 1985, The Situation in Logic-II: Conditionals and Conditional Information, Report CSLI-85-21,CSLI. Barwise, Jon, and Perry, John. 1983, Situations and Attitudes. MIT/Bradford Books, Boston. Carnap, Rudolf. 1947, Meaning and Necessity. University of Chicago Press. Lewis, David K. 1973, Counterfactuak. Harvard University Press. Partee, Barbara H. 1988, Possible Worlds in Model-Theoretic Semantics: A Linguistic Perspective. (In this volume.) Perry, John. 1986, From Worlds to Situations. Journal of Philosophical Logic 15:83-107. Plantinga, Alvin. 1974, The Nature of Necessity. Oxford University Press, Oxford. Quine, W. V. O. 1953, From a Logical Point of View. Harvard University Press. Stalnaker, Robert. 1984, Inquiry, MIT/Bradford Books, Cambridge, Mass. Stalnaker, Robert. 1985, Propositions. In Martinich, A. P. (editor), The Philosophy of Language, pages 373-80. Oxford University Press. Stalnaker, Robert. 1986, Possible Worlds and Situations. Journal of Philosophical Logic 15.

HANNES RIESER

Some Caveats with Respect to Possible Worlds Discussion of Barbara H. Partee's paper "Possible Worlds in Model-Theoretic Semantics: A Linguistic Perspective" 0. Preliminaries Professor Partee's paper contains four chapters, I. Historical Background, Π. A Sampling of Alternative Semantic Theories, HI. The Value of Possible Worlds for Linguistic Semantics, IV. Foundational Issues. In chapter I the development of possible-worlds semantics from Carnap to the present day is sketched and the predicates "extensional" and "intensional" are introduced. Chapter Π describes variants of possible-worlds semantics from the point of view of their foundation and their role with respect to syntax, semantics and pragmatics. Concerning non-model-theoretic semantics, representationalistic and conceptualistic approaches as well as some of their problems with respect to linguistic explanation are discussed. Chapter ΙΠ starts from the assumption that possible-worlds semantics is important for two reasons, as a technical tool and as a means of linking linguistics to other disciplines. With respect to the first reason it is shown that possible worlds yield an interesting structure on the space of meanings. This is done by discussing various phenomena like term-conjunction, entailment, belief-sentences, counterfactuals etc. In order to back up the second reason intensional phenomena are presented which have been discussed in philosophy of language or in action theory like intentionality or questions of validity. Chapter IV reviews several foundational issues like the relation of possibleworlds semantics to conceptualism, the question how many possible worlds there might be, whether conceivable worlds should be identified with possible worlds or the consequences of supplanting set theory by property theory. Professor Partee's estimation concerning the role of possible worlds in linguistic semantics is, I think, laid down in nuce in the following quotation (p. 116):

Some Caveats with Respect to Possible Worlds

139

"[...] even if possible worlds turn out to be dispensable in the end, and even if appeal to possible worlds raises as many questions as it answers, I believe there is no denying that by the linguists' primary standard of fruitfulness in generating explanatory accounts of linguistic phenomena, the use of possible worlds in semantic theories has had a far-reaching positive impact, and that the role of possible worlds in that influence has been essential."

I do not want to argue against this evalution, since I think it basically correct. Mutatis mutandis a similar judgement could be made with respect to many concepts and tools in the history of linguistics such as e.g. the Saussurean dichotomy, the introduction of categorial grammars by Ajdukiewicz etc. But at the same time I want to maintain that possible-worlds semantics has several serious shortcomings. My own prejudices cannot be easily subsumed under the labels used by Professor Partee in her section Π, where methodological points of view are discussed. There conceptualism, representationalism and functionalism are placed in opposition to various possible worlds theories. Roughly, I think that linguistic theories have to deal with two things, the mental activities of speakerhearers of a certain langugage variety and the interactive control, cooperation etc. going on in (oral) communication. At least in Europe this kind of position is quite common among scholars interested in natural discourse and cognitive science. At the same time (believing in 'branching pasts') I regard methodological positions as contingent results of particular historical constellations and developments which, hence, can be understood and criticized from the historical point of view. As we shall see, e.g. the Frege-tradition is not sacrosanct to me, despite the "Begriffsschrift". My caveats will belong roughly to two domains, the methodological, foundational and empirical one on the one hand and the historical one on the other hand. I will shortly summarize them here for the quick reader. The first caveat (in "How avoid explications per obscurat"} is simple and commonplace. It says: We do not know what kind of entities possible worlds are, hence we cannot use them in linguistic explanations if we want to be on firm and solid ground. The second caveat (in "Foundational problems") points out that the notion which seems to be most amenable to explication by possible worlds, namely 'semantic competence', is itself of questionable standing. The third caveat (in "A dynamic view concerning syntax and semantics is needed") argues for models representing the dynamics of speech production and speech analysis including the interactive goings-on in discourse. A small example is provided, illustrating the relation between manifest mental activity and interaction. Possible worlds do not seem to capture the dynamics of speech events. These should be modelled by suitable dynamic or procedural models. The fourth caveat (in "Two Frege-burdens of current linguistic semantics") is historically oriented and proceeds in two paths. The first shows, by providing quotations from Frege's "Sinn und Bedeutung", that Frege's notions of 'Sinn',

140

H. Rieser

'Bedeutung' and 'Vorstellung' are quite congenial to matters of speech, even if they result from heavy 19th-century prejudice concerning natural language. It is pointed out that Frege allows 'Sinn' to vary from speaker to speaker. I will argue that 'Vorstellung' (dismissed by Frege) cannot be separated from 'Sinn'. Explications of Frege using the intension-extension dichotomy do not seem to be aware of this difficulty. Besides that they are even narrower than Frege and provide only a cryptical notion of meaning for natural languages. Possible-worlds semantics is clearly of this sort, hence etc. The second path tries to show that by setting up rigid standards for logics Frege cut us totally off from interesting pre-Fregean traditions, especially (nonHegelian) dialectics. The transferability and fruitfulness of the post-Fregean notions 'truth' and 'validity' is questioned to some extent. The firth caveat (in "Sea contra'} maintains that notwithstanding the caveats mentioned, possibleworlds semantics is a valuable challenge for linguistic semantics. 1. Haw avoid explications per obscuraf* Possible worlds have played an important role in semantic theories developed in the last fifteen years or so. It seems to me that whether one can make friends with them depends very much on general decisions made with respect to one's research. The decisions concern e.g. one's methodological position, the data one wants to describe, the phenomena one wants to explain, the depth of explication aimed at, and the tradition of problem-handling and problem-solution one wants to work in. Thus possible worlds may be all right with respect to some interests, but they are certainly no universal remedy, since they are intimately bound to heavily regimented data, to a specific logical tradition, and above all, to a particular, namely Fregean explication of the dark term 'meaning', a particular explication of the notion of language (usually a not very intuitive one) and the idea that describing objects consists in mapping them onto structures of a certain kind, in opposition to e.g. mapping them onto procedures. Let it be granted then that possible worlds suit certain interests of this sort. Nevertheless, I do not think that the notion of possible worlds is clear altogether. We should know which kinds of entities they are. This amounts to saying which properties they have and how their identity conditions can be given. Given a certain object, we certainly want to tell whether the predicate "possible world" applies to it and if not, how it should be classified. This is first of all a problem to be dealt with by the metaphysics of logic in much the same sense as parts of Aristotle's "Metaphysics" analyse notions necessary in order to understand the two "Analytics". But even if these problems are of a metaphysical kind, this does of course not imply that linguists need not or should not care about them. Since Frege and Russell, at least, logic is free of empirical claims, but linguistic theories are usually bound up with empirical objects, human language

Some Caveats with Respect to Possible Worlds

141

(by Frege somewhat derogatorily termed 'Volkssprache'), human reasoning etc., hence linguistic theories will make empirical claims concerning these objects. Otherwise there would be not much difference between linguistics and logics. We certainly need an answer to the metaphysical questions in order to determine which role possible worlds can play in linguistic descriptions or explanations. For example - and these are the hard problems, I think - we might want to know such things as - whether possible worlds are related to the speaker's knowledge and beliefs with respect to situations, - whether possible worlds can change during a conversation, - how large possible worlds are, - how possible worlds are searched through, - what the relation of possible worlds to different sorts of memories is, - whether possible worlds are globally or locally consistent, - whether speaker-hearers have access to all of the possible worlds assumed to exist, - what the relation between contexts and possible worlds is, etc. I do not know of answers concerning these problems, hence I regard possible worlds as - perhaps suitable - fictions. What is lacking is, I think, a connexion between empirical research and possible-worlds semantics. 2. Foundational problems Many scholars working in the field of formal semantics seem to think that the possible world concept can be used to explicate the notion of semantic competence. As far as I can see, this is also the point of view taken in Professor Partee's paper. As can be gathered from my caveat one I do not think this is without problems. But even if we neglect the metaphysical and empirical provisos stated there, we get the new issue that even the explicandum 'semantic competence' is not understood well enough in order to be amenable to an explication by the notion 'possible world'. Semantic competence is usually used as a default for the intuitions of a linguist or a philosopher working within a particular paradigm. Since there is no generally accepted independent explicandum 'semantic competence' thisprocedere verges on circularity. To illustrate: A mainstream Fregean will usually want to express his intuitions via the concepts intension, extension and the whole apparatus he needs to base them upon. And this will, at least to some extent, filter out properties of natural language or discourse which cannot be easily understood in terms of the presupposed concepts. I will trace some of the Fregean biases of current semantics later on. Here I want to enlarge upon another difficulty connected with the use of terms like 'semantic competence' or 'linguistic competence' in general, which will directly

142

H. RJeser

lead to my caveat number three concerning language, possible worlds, and dynamics. Notions like 'competence' or the time-honoured Saussurean concept of 'langue' are not only poorly understood, as a rule they are connected with a very restrictive notion of language which has led to the description of heavily regimented data. From the historical point of view, I think that didactic grammar of the Anglo-European tradition, the discussion of examples in logics (by Frege, Husserl, Russell, Carnap, Strawson etc.) and, above all, the written mode of communication have deeply influenced the concepts of competence. Especially 'semantic competence' has been fixed via logical notions like 'truth', 'validity', 'entailment', 'presupposition', 'contradiction', 'intension', 'extension' and the corresponding prototypical examples. Possible-worlds semantics has stabilized the trend in linguistics to shift foundational problems to neighbouring disciplines. For the description of live discourse, e.g. the notions borrowed from logics are too static and too narrow: Models, i.e. possible worlds and domains of individuals, are given in advance, or meanings reconstructed as functions remain fixed once and for all. In non-regimented discourse information is developed and shared by the participants, individuals are introduced one after the other, interpretations of words are changed, metaphors and hyperbolic expressions are used, ad-hocwords created, the boundaries between literal and metaphorical meaning is blurred, self-monitoring and monitoring by others occurs etc. Talking in paradigms, the whole situation resembles pre-Hegelian dialectics (i.e. logic in dialogue-form) more closely than objects depicted by Kripke-models, whatever these may mirror from the empirical point of view. I want to enlarge on these questions a bit further, hence my caveat number three. 3. A dynamic view concerning syntax and semantics is needed^ If we look at the linguistic behaviour of the participants in natural discourse, we find that at least the following phenomena are relevant and should be accounted for in linguistic descriptions: Language production and language reception is dependent on such things as -

planning and self-monitoring 'incrementality': succession of acts involved structured but finite memories of different sorts methods for focusing and searching monitoring by others

1

This caveat'is based on joint work of H.-J. Eikmeyer and me. See Eikmeyer (1983) and Eikmeyer and Rieser (1985).

Some Caveats with Respect to Possible Worlds

143

- principled interaction and negotiation - turn-taking systems and discourse structures How exactly planning and reception processes are organized and how they interact is largely unknown at the time being, although there has been quite a lot of research in these areas, especially from psycholinguistics (vide e.g. Levelt 1983, Levelt and Cutler 1983) and ethnomethodology. As an example for goings-on of the sort mentioned above I will give an example for name-searching taken from an article of ScheglofPs (Schegloff 1979): B: Uh she asked me to stop by, she bought a chest of drawers from uhm (4.0) B: what's the gal's name? Just went back to Michigan (2.0)

B: Helen uhm A: Oh I know who you mean, (1.0) A: Brady--Brady. B: Yeah! Helen Brady. A: Mm hm B: And she - she says she's uh never ... B's search for a name is placed in the middle of an utterance, which is interrupted for the search. B indicates the start of the search by 'uhm' and then she provides two search-patterns: 'what's the gal's name?' and '[the girl who] just went back to Michigan'. Obviously, B marks the result of her search concerning her own knowledge base: 'Helen'. The continuation of the search is signalled as well: 'uhm'. A expresses her willingness to cooperate: Oh I know who you mean'. The result of her search is first tentatively given and then confirmed: 'BradyBrady'. B ratifies the result: 'Yeah! Helen Brady'. Now B can go on with her (original) utterance, perhaps using a syntactical restart. This piece of discourse shows many of the features mentioned above such as the succession of interdependent linguistic acts, turn-taking, principled interaction, activation of knowledge systems etc. From the semantical point of view it is interesting that an individual is implicitly introduced by two interrupted prepositional phrases ('... to stop by', 'she bought a chest of drawers from ...') and named. Mutual knowledge concerning a name-individual relation is gradually established. If one is interested in these things, and I think that this interest is legitimate, then possible-worlds semantics is not of much help. The reason is, as already hinted at, that the dynamics of discourse development, introduction of individuals, focusing on entities, building up mutual knowledge etc. do not seem to be describable in one of the possible-worlds frameworks at all.

144

H. Rieser

I do not want to pretend that the systematic description of the aforementioned phenomena has made much progress in recent years, but still my impression is that the syntactic and semantic phenomena involved provide a principal obstacle to non-procedural tools. In order to model goings-on of this sort we need e.g. above all routines which regulate the interplay between untroubled production, indication of the trouble spot, switch to the problemsolving strategy (a search of some kind), activation of the strategy, evaluation of the result achieved, and blending back into normal talk again. Thus, these routines have to account for the building-up of one Gestalt within another. Higher programming languages are procedural tools which have as a rule enough facilities to model interactions of this sort (at least prototypical cases thereof), especially mechanisms concerning the flow of control like loops or alternative decisions. As a side-effect it may be of some use that a programming language is usually clearer and easier to control than a metalanguage using a possible-worlds concept. This is true at least for cases where the linguistic regularities to be captured need lots of parameters, scales etc. or where some sort of regimented natural language would have to be used as a metalanguage. Even if it is accepted that there are such data and that it is worthwhile to describe them, couldn't it be that paradigms using possible-worlds semantics describe data that only show the necessary amount of abstraction and idealization in order to make systematic work possible at all? Couldn't it be that if we strip these data of all crude things like repairs, corrections, phonetic irregularities, syntactic restarts etc. we would get tractable data which could be considered idealizations of empirical findings? I do not think so, one reason being that the notions of abstraction and idealization should be used in connection with the setting-up of suitable models. They should by no means be taken as a metaphor for data pruning. Another reason is that the dynamics of discourse, the adaptation of speakerhearers to new constellations and the possibilities for interactive control are easily under-estimated. Language is a vague, contingent matter open to longerlasting as well as to ad-hoc modifications, especially in the realm of semantics. Favouring the idea that semantic competence somehow represents the intersection of the individual speaker's semantic abilities of some externally defined group is certainly inadequate. Rather semantic competence should comprise the metarules or semantic tools for dealing with vague concepts, using prototypical meaning components, going back to moulds like dictionaries, conduct interactive control etc. In natural discourse it can be frequently found that even word meanings and their use in situations are open to negotiation. Meanings are rather vague entities liable to change. They can be adapted to context. In other words, they are far more indexical than one would assume at first sight. However, and here the possible counter-argument enters, it is certainly true that people do not negotiate meanings all the time. If this were the case, they would be busy setting up conventions of language use and would not have enough time to communicate.

Some Caveats with Respect to Possible Worlds

145

This, however, does not entail that the meanings used by different speakers are the same or nearly the same. The speaker-hearers' behaviour can be explained by the assumption that they presuppose sufficient meaning similarity, that they are content with some superficial understanding or that they are simply uninterested since nothing is at stake for them. Some such least effort principle or means-end maxim is, I think, central in order to account for the behaviour of speakers, especially their varying engagement in negotiation. Negotiation concerning matters of understanding, meaning and interpretation frequently arises at points of high mutual interest. Matters which seem to be settlecf in discourse can be re-opened again. In short, there is no canonical one-and-for-all interpretation for utterances or parts of them. Above all, and that is quite characteristic for natural discourse, utterances looking like declarative sentences need not be used in the mode of true assertion. They can serve as starting point for negotiation. Thus, e.g. people demonstrating in these days with banners saying "Tschernobyl ist überall" ('Tschernobyl is everywhere') do not mean to imply that we all live in Tschernobyl or that Tschernobyl is scattered all over the earth. 4. Two Frege-burdens of current linguistic semantics Possible worlds-semantics is deeply connected with the Fregean tradition of logics and philosophy in roughly two ways. The first way is bound to the solutions Frege suggested, the second is bound to the traditions Frege turned down, namely the whole domain of pre-Fregean logics and dialectics. I consider the first as one of the Frege-burdens of current linguistic semantics and will now elaborate on it. Adherents of possible-worlds semantics are working within the boundaries of the Fregean distinction of 'Sinn' and 'Bedeutung'. Some even try to explicate these notions and Frege's application of them concerning e.g. subordinate clauses. Seen from the linguist's side, there is of course no guarantee that meaning in natural language is exhausted by 'Sinn' and 'Bedeutung' (I take these terms in the sense of intension and extension respectively), even if we stick to so-called cognitive meaning. However, my point here is somewhat different and crudely expressed as follows: (i) Frege's introduction of 'Sinn' rests on specific assumptions concerning speakers and natural language which receive no further explanation. (ii) Nevertheless his notion of 'Sinn' is more congenial to natural language than the technical notion of intension, because he admits that 'Sinn' may vary. (iii) It is frequently neglected that Frege introduced a third concept, 'Vorstellung' ('associated idea'). If the Frege paradigm is evaluated as a whole, these three distinctions should be taken into consideration, otherwise Fregean concepts or their explication will not have the power needed for linguistic reconstruction.

146

H. Rieser

Ad (i): Concerning 'Sinn' (sense) Frege assumes2: 'Der Sinn eines Eigennamens wird von jedem erfaßt, der die Sprache oder das Ganze von Bezeichnungen hinreichend kennt, der er angehört...' Here 'Sinn' is introduced via 'sufficient knowledge of language', 'sufficient knowledge of the totality of referring expressions', and 'belonging to a language'. All these expressions used are not uncontroversial, especially if seen in the light of caveat three, where clearly the notion of 'a language' is called into question. Ad (ii) we have: 'Gewiß sollte in einem vollkommenen Ganzen von Zeichen jedem Ausdruck ein bestimmter Sinn entsprechen, aber die Volkssprachen erfüllen diese Forderung vielfach nicht, und man muß zufrieden sein, wenn nur in demselben Zusammenhange dasselbe Wort immer denselben Sinn hat.' Here Frege observes that in natural language there is frequently no definite 'Sinn' for expressions. What can be expected is constancy of 'Sinn' within one context or in contexts of the same kind (the German wording is ambiguous between these two readings). This remark clearly presupposes that this may well not be the case. In a footnote he further notes that the 'Sinn' of a proper name like 'Aristotle' could be different for different people. He does not think that this is bad, given that '[solange] nur die Bedeutung gleich bleibt, lassen sich diese Schwankungen des Sinnes ertragen, wiewohl auch sie in dem Lehrgebäude einer beweisenden Wissenschaft zu vermeiden sind und in einer vollkommenen Sprache nicht vorkommen dürfen.' In this footnote ideal language and formal theory ('vollkommene Sprache', 'Lehrgebäude einer beweisenden Wissenschaft') are contrasted to normal language. The former should not show variation of sense. Two things need to be added here: What holds for proper names will hold for verbs, adjectives etc. at least to the same degree. And, what is admitted by Frege for proper names, may be generalized: Variation of a constituent's 'Sinn' will lead to different senses of the containing expressions. Let us have a look at (iii), concerning 'Vorstellung': 'Von der Bedeutung und dem Sinn eines Zeichens ist die mit ihm verknüpfte Vorstellung zu unterscheiden. Wenn die Bedeutung eines Zeichens ein sinnlich wahrnehmbarer Gegenstand ist, so ist meine Vorstellung davon ein aus Erinnerungen von Sinneseindrücken, die ich gehabt habe, und von Tätigkeiten, inneren sowohl wie äußeren, die ich ausgeübt habe, entstandenes inneres Bild. Dieses ist oft mit Gefühlen getränkt; die Deutlichkeit seiner einzelnen Teile ist verschieden und schwankend.'

I quote the German text, because the translation by Max Black does not seem to be correct. I restrict the discussion of the Frege notions to proper names, but I do not think that this will impair generality.

Some Caveats with Respect to Possible Worlds

147

I do not want to quote the passage in full length here, suffice it to say that 'Vorstellung' is regarded as subjective in opposition to (more) objective 'Sinn'. With one 'Sinn' different 'Vorstellungen' can go together, even for one speaker. 'Sinn' can be used without further qualification, 'Vorstellungen' have to be specified according to speaker and time. This assumption of Frege's clearly contradicts the sense-variation passage in the quoted footnote. With respect to 'Vorstellung' Frege asks whether we could not say that people connect different 'Sinne' with words just as we can say that they connect different 'Vorstellungen' with them. Frege gets himself out of this difficult question by postulating a difference in the mode of connexion obtaining between word and 'Vorstellung' on the one hand and word and 'Sinn' on the other hand. He goes on, varying his argument: 'Sinn' is accessible to every speaker, 'Vorstellung' is not. However, in connection with words, expressions and sentences three types of difference can be recognized according to Frege: We may have difference with respect to 'Vorstellung', 'Sinn' or 'Bedeutung'. For the logical tradition, especially Carnap and his followers, only 'Sinn' and 'Bedeutung' have become relevant. But how can we solve the problem Frege evaded? How can we separate Frege's 'Sinn' from Frege's 'Vorstellung' without presupposing the very distinction by one of the current logical explications? How do we know that e.g. in connexion with intuitions concerning prepositional attitudes only 'Sinn' and not also 'Vorstellung' is relevant? I do not think that we know, and I do not see that a trivial separation into 'Sinn' and 'Vorstellung' can be given. The reason is that a speaker's meaning of a word will be intimately connected with the history of its acquisition and prototypical contexts of application. As I said above, the idea of a common intersection of the meanings in a language community is too strong and has to be replaced by the speaker's assumption of sufficient similarity and his being able to carry out interactive control. There would be a lot more to say on Frege's ideas and Frege-reconstruction, but Frege is not at stake here, rather the tradition of Fregean notions. Without going into more details I think that dichotomies comparable to 'Sinn' and 'Bedeutung' and notions built upon them like 'truth' and 'validity' are not easily to be transferred to linguistic matters and the independent linguistic motivation for them has yet to be provided. On the whole, I think, perhaps Putnam's notion of stereotype (vide Putnam 1975), which I take to be some generic meaning feature, comes fairly close to word meaning in the normal, nonreconstructivistic linguistic sense. It would be worthwhile to provide explications of this notion and to formulate combinatory principles for the fitting together of stereotypes. The canonical interpretation of 'Sinn' and 'Bedeutung' is one burden, I think, for formal semantics connected with empirical claims. A logician is, of course,

148

H. Rieser

free to choose his concepts, since he need not make empirical assumptions about language or speaker's acts etc. He may select his primitive terms as he likes, since in Frege's words he is working within the confines of a Vollkommene Sprache' and a 'beweisende Wissenschaft'. The first Frege-burden is perhaps obvious for anyone having had a look into the history of formal semantics from Frege onwards. The second Frege-burden is more difficult to detect, since it is connected with 'branching pasts in logics', i.e. with the logical paradigm in pre-Fregean days. Roughly, I believe that the Frege-tradition cut off linguistics from interesting and fruitful pre-Fregean developments in logics, i.e. in dialectics. I do not of course want to minimize the contribution of Frege to the development of formal logic and philosophy of language. But logic, i.e. dialectics in pre-Fregean days, hacf more direct links with natural language than the Frege approach which was developed in order to overcome the vagaries of non-regimented natural language. I take it for granted that at least up to Leibniz the paradigm of scholastic logic, i.e. terministic as well as Aristotelian logic was the representative paradigm not too different from William of Sherwood's "Introductiones in logicam" or Peter of Spain's "Summulae logicales". This means, in short, that in pre-Fregean times we still have the triad grammar, logics/dialectics and rhetoric with vague boundaries but still some division of labour existing among them. Concerning logics this implies that it included dialectics (hence logics frequently was called dialectics by pars pro toto) and encompassed the whole of the Aristotelian Organon" including the "Topics" and the "Sophistical Refutations". The commentaries to the "Topics" as laid down in scholastic logic books treat many patterns of argument (called loci or topoi) which are also familiar from everyday argumentaion, for example the "topos of authority", the "topos of more or less", or the "topos of similarity". (These topoi are usually given reductions to the Aristotelian syllogisms but that need not concern us here.) An example of the "topos of authority" is e.g.: If Professor Lewis say s possible worlds exist, then they exist.

An example of the "topos of more or less" is:

If Professor Cresswell does not know what possible worlds are, how should I? Concerning the first example, who would deny that the topos of authority, perhaps in some milder form, is a prominent and successful pattern of argument in the humanities? The argumentative background of the second example is of course that Professor Cresswell occupies a higher degree of experthood on the experts-on-possible-worlds scale than I do, hence, if he does not know etc. There is a very nice example of the "topos of similarity" in Professor Partees paper which I want to quote here: 'In earlier writings I had raised some worries about the possible psychological reality of possible worlds which in retrospect I think arose partly from mis-

Some Caveats with Respect to Possible Worlds

149

guided concern with trying to fit possible-worlds theory "inside the speaker's head": but we don't try to fit the theory of gravitation "inside a falling apple".' One of the arguments here is roughly (1) (2) (3) (4)

Theory of gravitation and possible-worlds theory are similar. A falling apple and a speaker's head are similar. Of two similar things, what holds for the one also holds for the other. The theory of gravitation is not put inside a falling apple.

(5) The possible-worlds theory is not put inside a speaker's head. Although there is some truth to (2), it would be a good point for the opponent to start a counterargument there, asking whether an apple and a speaker's head are similar concerning the right modus. It is certainly not the case that everyday argumentation as encountered in written or oral discourse follows the norms of Aristotelian argumentation. Nevertheless it remains true that the topoi used in this tradition which is concerned with widely accepted or successful argumentation are among the most frequently encountered mechanisms in discourse. Hence we can use classical topoi for matters of investigation. The crucial point is, of course, that we cannot work with truth and validity here as the above examples show. What is needed are concepts like "proposed to be true", "force of an argument" or "accepted argument" in order to model these cases. It is difficult to see how these forms of argument are connected with those investigated in the post-Frege tradition, especially in possible-worlds theory. As far as I can see, language philosophy in the Frege tradition brought a narrowing down of the concept of meaning (to intension and extension) and a reduction of the (empirically) interesting forms of argument to the logically valid ones. With respect to pragmatics, if pragmatics is concerned with language use in normal situations, I believe that possible-worlds semantics is bound to fail, since its very fundamentals are too narrow, notwithstanding the fact that it may be enlarged in some interesting way. 5. Sed contra Nevertheless, as I already maintained in the preliminaries, I do not think that possible-worlds semantics is entirely without merits. There is of course a lot to the analyses given for different phenomena labelled 'intensional' as one can see from Professor Partee's ΙΠ, 'The Value of Possible Worlds for Linguistic Semantics'. I want to admit this, even if I think that analyses should be carried out in general with respect to the knowledge, beliefs and the language-variety of the subjects involved. Let me just illustrate this by the famous argument concerning intensionality:

150

H. Rieser

(1) Erna is looking for the dean. (2) The dean = Mr. Miller. (3) Erna is looking for Mr. Miller. It is usually assumed that Erna, knowing the reference of 'the dean', is not aware of (2). Hence (3) is not valid. But the terms used could be exclusively the speaker's, the dean could be known to Erna by a different description, which is also familiar to the speaker. Then (3) would hold with respect to the speaker's use of language. I find this argument difficult to neglect, even in extensional cases. Possible-worlds semantics in linguistics has clearly been dominated by the paradigm of logics. Nevertheless it has brought to the fore notions which are of considerable help to linguistic semantics, since they correspond (in a nottoo-well-understood sense) to intuitive concepts used by speakers. Among these is, above all, the concept of alternative. People clearly employ such a concept in order to anticipate e.g. possible outcomes of actions or events in which they are involved. They are able to set an existing situation in opposition to an as yet nonexisting one. In addition, they may envisage that the present situation could develop into one out of a set of equally conceivable alternatives. This ability of conceiving of alternatives is at die base of many arguments, actions or explanations. It presupposes notions like 'situation of the same kind in such a respect' or 'situation sufficiently similar with respect to such and such'. Clearly, the use of counterfactuals, tenses or of natural language modals depends on these notions. There will, of course, be different ways to capture them and one will not necessarily need possible worlds for that, nevertheless it is to be admitted that they have been reconstructed in possible-worlds frameworks first.

Bibliography AllwoodJ.: 1986, Some Perspectives on Understanding in Spoken Interaction. In: Furberg, M. etal. (eds), Logic and Abstraction. Essays dedicated to Per Lindström on his fiftieth birthday. Aristoteles: 1968, Aristotelis Analytica priora et posteriora, ed. by W. D. Ross. Clarendon Press: Oxford. Aristoteles: 1970, Aristotelis Topica et Sopbistidelenchi, ed. by W. D. Ross. Clarendon Press: Oxford. Aristoteles: 1970, Metaphysics/Aristoteles, ed. by W. D. Ross. Clarendon Press: Oxford. Ballmer, Th. T.: 1978, Logical Grammar with Special Considerations of Topics in Context Change. North-Holland: Amsterdam [Cicero]: 1977, Ad C. Herennium. De ratione dicendi. Ed. and with Engl. transl. by H. Caplan. Harvard Univ. Press: London. Clocksin, W. F., Mellish, C. S.: 1984, Programming in Prolog. Springer-Verlag: Berlin. Cresswell, M. J.: 1973, Logics and Languages. Methuen & Co: London. Eikmeyer, H.-J.: 1983, Procedural Analysis of Discourse. In: Text, Vol. 1, pp. 11-38. Eikmeyer, H.-J. and Rieser, H: 1985, Procedural Grammar for a Fragment of Black English Discourse. In: Ballmer, Th. (ed.), Linguistic Dynamics, de Gruyter: Berlin. Feyerabend, P. K.: 1975, Against Method. Humanities Press: London.

Some Caveats with Respect to Possible Worlds

151

Frege, G.: 1980, Über Sinn und Bedeutung. In: Patzing, G. (ed.): Gottlob Frege. Funktion, Begriff, Bedeutung. Vandenhoek & Ruprecht: Göttingen, pp. 40-65. Translated by Black, M. as On Sense

and Reference' in Geach, P. and Black, M. (eds): Translations from the Philosophical Writings of Gottlob Frege. Basil Blackwell: Oxford, pp. 56-78. Grabmann, M.: 1937, Die Introductiones m logicam des Wilhelm von Shyreswood. Verlag der Bayerischen Akademie der Wissenschaften: München. (= Sitzungsberichte der Bayerischen Akademie der Wissenschaften, Jg. 1937, Heft 10). Guilelmus de Shyreswode: 1975, William of Sherwood's Introduction to Logic, transl. with an introd. and notes by N. Kretzmann. Greenwood: Westport, Conn. Hughes, G. E., Cresswell, M. J.: 1968, An Introduction to Modal Logic. Methuen & Co: London. Kneale, W. and Kneale, M: 1971, The Development of Logic. Clarendon Press: Oxford. Kretzmann, N. et al. (eds.): 1982, The Cambridge History of Later Medieval Philosophy. CUP: Cambridge. Levelt, W. J. M.: 1983, Monitoring and self-repair in speech. In: Cognition 14, pp. 41-104. Levelt, W. J. M. and Cutler, A.: 1983, Prosodic marking in speech repair. In: Journal of Semantics. Vol. , No. 2, pp. 205-217. Lewis, D.: 1972, General Semantics. In: Davidson, D. and Harman, G. (eds), Semantics of Natural Language. D. Reidel: Dordrecht-Holland, pp. 169-219. Montague, R.: 1974, Formal Philosophy. Selected Papers ofR. Montague. Ed. and with an introd. by R. H. Thomason. YUP: New Haven and London. Petrus Hispanus: 1947, Summulae Logiades. ed. by I. M. Bochenski O. P. Domus Editorialis Marietti. Putnam, H.: 1975, The meaning of 'meaning*. In: Mind, Language and Reality. Cambridge: CUP. (= Philos. Papers, Vol. 2) pp. 215-271. Rieser, H.: 1985, Procedural Semantics and the Flow of Discourse. In: Quademi di semantica, Vol. VI, pp. 102-111. Schegloff, . .: 1979, The Relevance of Repair to Syntax-for-Conversation. In: Syntax and Semantics, Volume 12, pp. 261-286.

BARBARA H. PARTEE Speaker's Reply I appreciate the thoughtful comments of both of my discussants, and I welcome the chance to respond to some of the issues they have raised. I consider the foundational questions in this area to be very difficult ones and the answers far from settled. It is probably also clear to the reader by now that background assumptions, terminology, goals, and criteria for a successful theory are far from standardized across the various overlapping communities of researchers with an interest in semantics. Nevertheless my discussants and I certainly share enough common background and common goals that one can find a good basis for fruitful debate on substantive issues. These remarks are not intended as the "final word", then but simply as a next move in a continuing dialogue. I will respond first to John Perry's comments, then to Hannes Rieser's, since they focus on largely distinct sets of concerns. I will also bring in some issues raised in the discussion by other participants in the symposium, particularly Jaakko Hintikka, David Kaplan, and Thomas Kühn. /. Reply to Perry John Perry's points center on the "subject matter problem" or the "problem of aboutness", and its implications for theories which consider only total possible worlds rather than countenancing more partial world-like entities as well so as to permit more fine-grained individuation of propositions. All the crucial examples he discusses involve tautologies or contradictions and revolve around the fact that in possible-worlds semantics there is just one tautological proposition (modelled as the set of all possible worlds) and just one contradictory proposition (modelled as the empty set of worlds), whereas our linguistic intuitions tell us that different tautologies are about different things and likewise for different contradictions. As Perry presents it, the "problem of subject matter" is distinct from the "problem or modes of presentation" manifested by the apparent failure of substitutability of cointensional atomic expression (Cicero - Tully, oculist - eye doctor, etc.) in propositional attitude contexts; in the terms of my own presentation, both would be subcases of the "problem of logical equivalence". Perry considers the "problem of modes of presentation" to be one shared by "any semantic theory that takes propositions to be non-linguistic intensions", but the

Speaker's Reply

153

problem of subject matter to be a more pervasive problem, in the sense of not being limited to prepositional attitude contexts, and one that is an artifact of theories that analyze propositions in terms of total possible worlds only. (Perry and I agree on distinguishing these problems from the issue of 'comprehensiveness': total possible worlds may be totally specified alternatives within some very small domain of possibilities, such as the distinct possible outcomes of a sequence of three coin-flippings.) I take issue on two counts with Perry's claim that strong possible-worlds semantics "cannot capture the notion of a proposition's being about an object": (i) it can capture some such semantic notion of aboutness; and (ii) Perry's claims notwithstanding, our pretheoretic intuitions do not seem to delimit a unique notion of aboutness in a clear and distinct way, and I find myself quite skeptical about the possiblity of turning Perry's sketches into genuine counterexamples. Let me spell out my reasoning on points (i) and (ii); then I can come back to broader questions about whether tne problem of subject matter can bear the weight Perry suggests it can in helping to decide among semantic theories. Ad (i): One can identify a semantic notion of aboutness in possible world semantics, though not in a way that would make any tautology or contradiction be about anything in particular. An empirical proposition can be said to be about a particular individual (or property, etc.) if the truth of that proposition varies systematically with the properties ofthat individual (or property, etc.).1 Since in classical possible worlds semantics, a proposition is analyzed as a function from possible worlds to truth-values (or equivalently as a set of possible worlds, namely those in which it is true), the semantic characterization of aboutness given above - let me call it "pw-aboutness" - cashes out as follows: Let the worlds in which proposition p is true be p's T-set and those in which p is false be p's F-set. Then a proposition p is pw-about Jones if there is some systematic difference between the properties Jones has in the worlds in p's T-set and the properties Jones has in the worlds in p's F-set. This will be the case for the proposition expressed by the sentence 'Jones is bald' but not for the proposition expressed by 'Smith is bald'; Jones's properties in various worlds will presumably not have any systematic correlation with whether those worlds are in the T-set for 'Smith is bald'. This conception of pw-aboutness is very definitely a semantic notion of aboutness and clearly distinct from any syntactic or structural notion that might relate to being the subject, or topic, or focus of a sentence; the entities and properties a proposition is "pw-about" are not in any sense taken to be constituents of the proposition. Indeed it is worth emphasizing that on the pure I don't know whether this proposal has been made in the literature or not. On the one hand it seems a natural and familiar idea and I'm sure I have encountered it at least orally before; on the other hand, I wouldn't be surprised if it hasn't been spelled out in print because such notions usually get spelled out formally only when there is some further role for them to play which depend on having an analysis of "aboutness" to build on.

154

B.KPartee

classical possible-worlds approach where a proposition is analyzed as a set of worlds, propositions do not have constituents at all. (This is one of the chief differences I was getting at in my paper in contrasting pure possible-worlds approaches with approaches based on or incorporating more syntactic notions of "logical form".) The proposition expressed by a given sentence may be pw-about several different entities and properties, including ones not denoted by any expression in the sentence at all. Consider sentence (1), for instance: (1) Jones is over 40. This sentence is not only pw-about Jones but also pw-about Jones's age, since there is a uniform difference in what Jones's age is in the T-set vs. in the F-set of the proposition. And this matches our ordinary intuitions - sentence (1) is indeed about Jones's age. Yet there is no constituent in the sentence corresponding to the concept "Jones's age"; so the alternative approach to aboutness which Perry advocates, building up "issues" from relations and sequences of objects and defining aboutness in terms of constituents of such issues, seems in danger of leading to a too syntactic notion of aboutness, one on which it will not turn out that (1) is about Jones's age. Now of course on the pw-aboutness account that I have offered, no tautology is ever pw-about any particular individual, nor is any contradiction. This follows straightforwardly from the fact that every possible world is in the T-set of any tautology, and none in the T-set of any contradiction. This is what Perry considers a shortcoming of the theory and what he is suggesting an alternative approach to aboutness for. But it seems to me that insofar as the beginnings of this approach are spelled out, it looks like a relatively syntactic notion of aboutness rather than a semantic one, as I suggested in the previous paragraph, and it may well be that there isn't any way to analyze a tatology as being about some particular individual without a notion of aboutness connected in some intimate way to the syntactic structure of the sentence expressing the particular tautology. This brings me to point (ii): is there really a relatively clear and robust pre-theoretic notion of aboutness that a semantic theory should enable us to capture? Perry clearly thinks so: "it's quite useful and we have pretty clear intuitions about it". I don't agree. There is a large literature on such theory-laden concepts as topic, focus, theme and rheme, presupposition, structured propositions, "relevance" in the sense of relevance logic, overlapping concepts which all have something to do with "aboutness". The data in these domains are still much more controversial than the data underlying the concept of entailment; it's not that entailment is a perfectly sharp notion pre-theoretically, but I think it has proved to be much more robust with a much more solid core of clear data than any notion of aboutness can be said to be at this time. I agree with Perry that some notion(s) of aboutness is\are very important, and I am also perfectly ready to agree that strong possible worlds semantics does not by itself answer all the questions one may want to ask about such a notion. But I do not agree that there

Speaker's Reply

155

is any coherent set of data of the sort Perry suggests which will support the idea that a proper semantic notion of aboutness can motivate the use of partial worlds or situations rather than total ones. Note, for instance, the clearly syntactic nature of the first three of the four conditions that Perry offers as "the beginnings of a theory of aboutness", and note how much less clear our intuitions become if we choose examples for which Perry's aboutness and pw-aboutness differ. "(a): a simple sentence that contains a name for an individual expresses a proposition that is about the individual named." But does this really extend to contradictions such as "Jones is not Jones" (taken on a literal reading; of course if we reinterpret the sentence metaphorically we can get it to be about Jones)? Similarly, principles (b) and (c) concerning conjunction and disjunction are made to look uncontroversial by being illustrated with contingent propositions where Perry's aboutness claims agree with those obtained on the pw-aooutness account. But on tautologies and contradictions the approaches disagree, and I don't know of any way of marshalling intuitions which are both pre-theoretic and independent of syntactic structure to support the claim that the proposition expressed by (2) is clearly about Jones. (2) Jones is over 40 or Jones is not over 40. Of course sentence (2) mentions Jones; but it begs precisely the questions at issue to stipulate that the proposition expressed by (2) is about Jones. None of Jones's properties affect its truth. It doesn't tellus anything about Jones. (Probably the name "Jones" must have a referent in order for the sentence to express a proposition at all, but whether that referent must exist for the proposition to have a truth-value in a given world is a matter on which theories disagree. There are long-standing problems with examples such as Zeus, Santa Glaus, and Sherlock Holmes.) I also agree with the opinion expressed by Jaakko Hintikka in the discussion period that none of Perry's examples not involving prepositional attitudes contexts are convincing in showing the need for distinguishing among propositions considered equivalent in strong possible worlds semantics. In any case, semantics has progressed to the point where arguments are normally not carried out over isolated examples, which would be quite endless and fruitless; one normally debates the overall explanatory force of competing combinations of syntax, semantics, pragmatics, logic, and metaphysics. And since neither Perry nor I are offering explicit full accounts of the debated linguistic constructions here, there's not much point in my trying to do more here than register my claim that his arguments don't make his case and my belief that they cannot be strengthened so as to do so. In the end I find no valid challenges to strong possible worlds theories in his remarks that I would not subsume under the challenges from prepositional attitudes that I had already discussed in my paper. To me, the most interesting and challenging problem of aboutness is an entirely different one, one which I touched on in my paper: a version of the problem of

156

B. KPartee

aboutness on which it is the fundamental problem of intentionality (with a "t") and on which the central locutions are "thinking about", "talking about", "wondering about", etc., i.e. the paradigmatic prepositional attitude constructions. The question of whether the possible-worlds-based concept of intensionality adequately accounts for the main properties of intentionality is to my mind the most important question to raise about strong possible-worlds semantics. The problems in this area are the ones that have led various researchers to advocate such enrichment or alternatives to the theory as structured meanings, or intermediate levels with more syntactic or representationalist properties, or using property theory as the metatheoretical foundation and replacing "sets of possible worlds" with "properties of possible worlds". These sorts of problems with prepositional attitudes affect the kind of theory Perry advocates just as much as they affect possible worlds semantics, and will undoubtedly be the focus of much work for some time to come. Π. Reply to Rieser My principal disagreement with Hannes Rieser seems to be methodological. He and I are in agreement that possible worlds have provided valuable concepts and tools for the enrichment of linguistic theories of semantics, and also that the development and application of possible-worlds theories of semantics have involved considerable regimentation and idealization of the data to be accounted for. Where we seem to differ most is over whether such idealization and regimentation is to be regarded as a strength or a weakness. This comes out most clearly in Rieser's second caveat: he regards the notion of "semantic competence" which possible worlds semantics offers an explication of as a notion of questionable standing because of its narrowness, its restriction to logical properties such as truth-conditions, entailment relations, etc., and its omission of many aspects of the dynamics and fluidity of actual communicational interaction. Similar charges are levelled both at Frege and at neo-Fregeans such as Montague, in developing logics which ignored many parts of pre-Fregean traditions of rhetoric and dialectics. I believe on the contrary that the identification of a suitably delimited and idealized set of data is often of absolutely crucial importance for theoretical advance. Of course one can always challenge another scientists's selection of data to be explained and try to build a better theory with a different way of carving up the data, and of course it's valuable for some scholars to be looking at very large domains of data and asking very broad questions about the fit among different theories that are trying to jointly explain a wide range of phenomena. But in the particular cases under discussion I believe the narrowing and idealization of data in question were not only defensible but superb choices that put first logic and then semantics onto an explosively fruitful path of development.

Speaker's Reply

157

Logic before Frege had made rather modest advances since Aristotle; after Frege it underwent rapid and immensely fruitful development. Semantics before Montague was a fascinating but rather "mushy" field, particularly in comparison with syntax as it had been developing since the Cnomskyan revolution. I would argue that one of Chomsky's most important ideas was to take the basic data for syntax to be grammaticality judgments and to define the lowest level of adequacy for a grammar, "observational adequacy", to be that it generate all and only the grammatical sentences of a language. That data is both extremely narrow and impossible to pin down precisely; yet that conception of what the basic task of syntax is, what syntax is about, was absolutely central in turning syntax into the robust centerpiece of linguistics and one of the seeds from which cognitive science has grown. All this in spite of the fact that Chomsky would no longer define the goals of syntax in that way; but it wouldn't have been very helpful to try to define syntax as a mapping between sound and meaning in the 1950's. When linguists such as Katz, Fodor, Weinreich, Lakoff, and Jackendoff started in the 1960's to try to build theories of semantics to go with generative theories of grammar, the initial choices of data to be taken as central were semantic ambiguity, semantic anomaly, and synonymy. Synonymy proved to be rather elusive, and most real work involved "getting the right number of readings" for a given sentence: none if it was anomalous, one if it was semantically well-formed and unambiguous, and n if it was «-ways ambiguous. Truthconditions and entailment were not mentioned. A lot of good work was done, for instance on scope ambiguity, but the field lacked solid ground under it: there were really insufficient constraints on what sort of thing a meaning should be taken to be, what data meanings were supposed to be accountable for. Montague's proposal, coming straight out of the Fregean tradition, that the core observational data should be truth-conditions and entailment relations, was in my opinion as crucial a factor in the fruitfulness of the formal semantic enterprise as the choice of grammaticality data was for the Chomskyan revolution. Of course in neither case would the choice of how to regiment and idealize the data have attracted so many followers if it had not been accompanied with powerful theoretical proposals and insightful applications to interesting and difficult problems, but I'm inclined to believe that good ideas about data selection are just as important and difficult to achieve as good ideas about theory-building and problem-solving. As for the other interesting kinds of data that are not handled by possible worlds semantics per sey some probably properly belong to other theories that may stand in various relations to possible-worlds semantic theories, and some may indeed call for a revision of the basic foundations of semantic theory; but one doesn't know in advance which is which. I'm inclined to think that the vagueness of the line between Fregean Sinn and Vorstellung does not weaken the importance of the conceptual distinction between them any more than the vagueness of the line between night and day (especially in polar regions!)

158

B. KPartee

challenges the importance of distinguishing between them in explaining the behavior of various species. On the other hand the need for a more dynamic and less static conception of meaning and for greater integration of contextdependence and context-change into the workings of semantics is becoming widely accepted and is leading to foundational alternatives, some of which I mentioned in my paper. Nevertheless the role of possible worlds in theories like Heim's and Kamp's is as strong as ever. (As I noted in my paper, Stalnaker in his work on the dynamics of "common ground" was already quite explicit about the valuable role possible worlds could play in explaining the dynamic properties of conversational dialogue.) Returning last to Rieser's first caveat, concerning explications per ohscura, I heartily resist the contention that a linguist should know what possible worlds are in order to use them. Resorting once more to a "topos of similarity": early geneticists were not in any position to know what genes were, but some basic structural assumptions about their distribution and behavior were sufficient for a lot of fruitful development of genetic theories. Linguists' theories of semantics can indeed help to determine what possible worlds are, or at least what the possible worlds of possible-worlds semantics are. I see this whole conference as a fascinating exercise in intentionality - we are all debating whether there are possible worlds and what they can help us to explain while simultaneously trying to figure out if we mean anything in common by the term "possible world". I gather from the discussion that many participants are inclined to think of possible worlds as something like sets of sentences or sets of propositions or in some other way as constructed out of linguistic or conceptual constituents. (This is indeed one of the traditions in the philosophical literature as well.) I follow the Montague-Kripke-Lewis-Stalnaker tradition in making a clear distinction between possible worlds as alternative ways things might have been vs. descriptions of possible worlds; I don't take the actual world to be a set of sentences, and I don't take other possible worlds to be sets of sentences either, even though my access to possible worlds (including the actual one) may require the mediation of sentences or other mental constructs. In any case I will reiterate that as a linguist I am an instrumentalist about possible worlds: as a linguist I don't care a bit what they are like or what they are made of beyond the properties that play a role in my semantic theories. And I don't believe I need to or should care. (But I'm also a philosopher of language, and as a philosopher of language I do care. But that doesn't slow down my linguistics.) I don't think there is any absolute answer to which properties of possible worlds a linguist may have to worry about, since one doesn't know in advance which properties will play a role in linguistic semantics. Certainly it could matter whether possible worlds should be constructed as total or partial in the sense describee! by Perry; whether time is discrete or continuous could matter; many metaphysical questions could turn out to matter. But I wouldn't recommend that linguists ask for all such possible issues to be settled first before trying out

Speaker's Reply

159

theories that use possible worlds in some way. Maybe possible worlds are not even a ladder we are now climbing up and will eventually kick away; maybe they are a mass hallucination that many of us are sharing while climbing up some totally different ladder that we can't see. That should make interesting history of science someday! HI. Miscellaneous final remarks (i) David Kaplan made some useful remarks in the discussion period that I would like to pass on. One of his points was to emphasize the role played in later developments by Frege's "peculiar" choice of Bedeutung (reference) for sentences, namely that the reference of a sentence is its truth-value. Here the big line is between Frege and Russell, Russell taking a much more highly structured object as the denotation of a sentence. Carnap, Montague, and others followed Frege; current developments would undoubtedly be very different if the Russellian direction had been taken. Kaplan also pointed out that an important distinction had not been made sufficiently clear, namely that between the use of possible worlds in developing a theory of meaning and their use in developing a theory of necessity and possibility or more generally a theory of modality. Kaplan suggested that these two diiferent enterprises had frequently been confused in the past even by major contributors to them, but that it was important to keep them separate because they may require different notions of possible worlds. In particular, Kaplan argued that for a theory of meaning, one does not want to impose metaphysical constraints on which possible worlds there are since one wants to be able to distinguish any two sentences that are not logically equivalent to each other (by having some world in which one is true and the other false). On the other hand, in the branch of metaphysics concerned with theories of modality it is appropriate to offer metaphysical arguments about what possible worlds there are. (For example, Kripke has argued that if χ is a child of y, then χ is necessarily a child of y: i.e., that there are no possible worlds in which you have different parents than you in fact have.) Thus one needs different possible worlds for the theory of meaning than for the theory of metaphysical modality; and according to Kaplan, locutions such as "other concrete realities" belong primarily to the domain of modality. While I have undoubtedly been guilty of blurring the distinction myself, I am glad to have what I take to be support for my hope that linguists should be able to help develop theories of possible worlds semantics appropriate for the study of meaning without being obliged to become involved in metaphysical debates (or at least no more so than across any neighboring disciplines.) (ii) Also in the discussion Jaakko Fiintikka made an interesting challenge to John Perry's claim that possible worlds play no causal role in actual-world events. According to Hintikka, insofar as we agree that the possiblity of rain

160

B.KPartee

influences whether we carry an umbrella or not, then the possible worlds in which it does rain are playing a causal role in our umbrella-carrying behavior regardless of whether the actual worlds turns out to be one of them. I report that here because I found it interesting; I have no opinion on the matter myself. In fact that seems to me a perfect example of a metaphysical issue that a theory of meaning probably wouldn't and probably shouldn't resolve. (iii) Last but by no means least, Thomas Kühn challenged my reasons for holding that a theory of meaning should be based on possible worlds not limited to conceivable worlds. In our two papers for the conference we found many points of agreement and common interests, and we had many stimulating further conversations at the conference and afterward; but as we continued to probe at the grounds for our apparently unbridgeable difference on this one point, we found that we had more differences about our conceptions of possible worlds than we had imagined. This is an issue which is not much discussed within the formal semantics and modal logic communities except in the classroom, where it always comes up at some point and the students get "straightened out". (I remember being puzzled and then straightened out as part of my philosophical education at UCLA under the guidance of David Lewis, Richard Montague, and David Kaplan.) But it is certainly a crucial issue for the question of whether a theory like Montague's can possibly be appropriate as a theory of human semantic competence; whether, as Hannes Rieser put it in discussion, a theory like Montague's could only serve as a theory of language for infinite minds, not for creatures with finite minds. Here is one example Tom Kühn and I debated. Consider a sentence like (3). (3) If something had caused the boat to rock just then, we would have sunk. I claimed that in uttering such a counterfactual conditional, with such an open-ended word as "something" in it, one is making a different and more inclusive assertion than would be made in the same context by (4): (4) If something I can conceive of had caused the boat to rock just then, we would have sunk. I claim that (3) quantifies over all possible worlds in which anything causes the boat to rock, not limited to the kinds of forces and agents I can conceive of and hence a fortiori not limited to possible worlds I can conceive of. Kühn made the interesting response that the verb cause in (3) and (4) has a meaning that is limited by our conceptions of possible causing-events or causal agents, and that if we were to expand the domain of possible causers we would have to correspondingly change the meaning of the word cause as well. I responded by changing the examples to (3') and (4'), replacing the possibly theory-dependent verb cause by the similar but more colloquial and more "pre-theoretic" verb make: (3') If something had made the boat rock just then,... (4') If something I can conceive of had made the boat rock just then,...

Speaker's Reply

161

Kühn expressed agreement with me on the importance of articulating a semantic theory which makes room for our awareness of our own ignorance, allows us to quantify over entities beyond our acquaintance, etc., but we evidently disagreed about the application of that point to examples like these. I think that we don't have to believe that we know what kinds of things can make boats rock to use (3) or (3'), and that there really need be no further implicit sortal term with the "something" in (3) or (3'), and that when we use a sentence like (3) or (3'), we have committed ourselves to a claim whose truth depends on what happens in a certain (vague) range of physically possible worlds, a set of worlds which may or may not include lands of things that can make boats rock whose existence we have no conception of. The sincere, felicitous, and justifiable assertion of a statement does not require that we know beyond a doubt that it is true - thank goodness! I think the met that our language's reach can exceed our mind's grasp is a very welcome fact indeed, and part of what helps us communicate across cultures and centuries. At the same time I am sensitive to and sympathetic with Kuhn's concerns about the theory-dependence of meaning and the influence of our networks of beliefs on the meanings of our terms. That was in fact why I looked to such unloaded terms as "something" to hope for a most convincing case. But I think in subsequent discussion we found another difference that was affecting our dispute. I mentioned earlier that I don't regard possible worlds as sets of sentences, and in fact I don't regard them as constructed at all. Kühn does, it turns out; if I have understood him rightly, he wants to take them to be mental constructs, constructed from entities, properties, and relations. Now I'm not sure, but it seems likely that on such a view one could indeed arrive only at conceivable worlds, at least potentially conceivable, if the building blocks have to be conceivable entities, properties, and relations. Of course with sufficiently rich and higher-order properties and relations included, it won't be trivial to find an argument that this doesn't give one enough possible worlds to do semantics with adequately. I don't think it does and I think the contrasts between (3) and (4) or (3') and (4') are relevant evidence; but I consider the whole issue worth much more discussion, since it obviously touches on extremely important issues for the foundations of semantics and trie tensions between the modal logic possibleworlds tradition and the linguist's concern for an account of human semantic competence. I actually think it's more exciting if our language does not limit us to talking about conceivable worlds, but that's not an argument. In closing, let me just take this as one more opportunity to thank the organizers and hosts of the symposium and all my fellow symposiasts for an extremely stimulating, educational, and thought-provoking experience.

NILS ERIK ENKVIST

Connexity, Interpretability, Universes of Discourse, and Text Worlds*

The first of my many queries is whether a practically oriented student of authentic natural-language discourse ought to speak about 'possible worlds' at all. As a technical concept, possible worlds were first postulated by philosophers and modal logicians, in formal and theoretical frames more hospitable to precise conceptual limitation. Lest some of my theory-minded hearers should feel that 'possible worlds' ought to be patented for their own exclusive use - despite the tide of the present interdisciplinary symposium which generously offers 'possible worlds' to all arts and sciences - I shall here call my own constructs 'universes of discourse' and 'text worlds'. To what extent my worlds are 'possible' or something else is a problem I must leave to others. So, the purpose of my paper is to give a wide-ranging overview, with a medley of concrete examples for mainly inductive use, of phenomena and problems related in the first place to textual connexity. What these examples are supposed to show is that the coherence, connexity, and hence interpretability of authentic natural-language discourse are not matters of syntax alone. Rather, coherence and connexity depend on our ability of specifying the worlds evoked by, and surrounding, a text. In Section 2 I shall give some introductory examples of a gamut of problems that have given concern to linguists over the past several years and which raise problems of coherence. Section 3 should be read as a set of notes towards the definition of textual interpretability. Section 4 is devoted to text strategies as principles guiding text composition and interpretation. Section 5 presents a few examples of texts which illustrate where run the borders and confines of interpretability. And Section 6 attempts a summary.

I am grateful to Wolfgang Heydrich and Ulf Teleman for several improvements prompted by their comments on the preliminary sketch towards this paper. Jaakko Hintikka, Hannes Rieser und Roger Sell also gave me the benefit of their knowledge and wisdom when I was preparing my contribution. Had I been able to follow all of their suggestions, many of the remaining weaknesses too would have disappeared; for these blemishes I alone remain responsible.

Connexity, Interpretability, Universes of Discourse

163

When linguists began worrying about the study of cohesion between the sentences of a text, they usually, and sensibly, started out by extending their traditional grammatical apparatus to the study of intersentential linkage. A good example of this is Halliday and Hasan 1974. Indeed a number of devices such as 'phoric' reference, substitution, ellipsis, and conjunction, all well known from traditional grammar and rhetoric, proved applicable to the description of linkage in simple texts and pieces of discourse1 such as (1) A: "Where is John?" B: "I haven't seen him." (2) A: "Give me two of your books, please." B: "Which ones do you want?" (3) A: "Susie was reading Shakespeare." B: "And John War and Peace" But it soon appeared that textual well-formedness, as intuitively conceived by competent informants, did not follow only from the kind of cohesion that was overtly marked by elements denned within traditional syntax. Even a process such as anaphor resolution seemed in some instances to depend, not on morphological congruence or syntactic rules (such as those involving precedence and command) but on pragmatic judgements of plausibility. Compare (4)

(a) Susie gave Betty an aspirin because she had a headache. (b) Susie gave Betty an aspirin because she had just been to the drugstore. (Enkvist 1983: 303).

In (4a) we presume that she refers to Betty rather than to Susie because we know people are likely to give aspirins to friends with a headache. On the contrary, in (4b) we tend to identify she with Susie because we know that people who have been to the drugstore are likely to have aspirins to give to suffering friends. (Note that because-clzuses are linked to their surroundings more by causal plausibility than by syntactically definable formal constraints: this is why they provide such nice examples of pragmatic anaphor resolution.) There are many other types of instances in which the links between sentences resist traditional syntactic description. For instance

A terminological note. For the duration of this paper I shall join those who define text and discourse with the equation 'discourse=text + context' and hence 'text=discourse - context'. As a corollary, 'text linguistics' studies the linguistic structure of texts as such, whereas 'discourse linguistics' (or 'discourse analysis' to use a well-established but perhaps misleading term) views texts not only in terms of their internal structure but also in relation to their contextual and situational ambience.

164

N. E. Enkvist

(5) George's high pass was headed to the right. The forward shot at once without dribbling and made a goal. The referee declared the kick offside. (Cited from Enkvist 1978a: 111.) is a perfectly good text to all who know the first thing about football, though it contains no syntactically definable intersentential links. Conversely we can fabricate sentence sequences such as (6) I just bought a car. The car in which President Wilson rode down the Champs Elysees was black. Black English has been widely discussed. The discussions between the parties ended last week. A week has seven days. Every day I feed my cat. Cats have four legs. The cat is on the mat. Mat has three letters. (Ibid.) which contain repetitions of words apparently linking the sentences to each other and thus giving the text a spurious pseudo-coherence. But our intuitive sense of coherence will all the same suggest that (6) is a far worse text than (5), if indeed it qualifies as a text at all (which depends on which of the many definitions of 'text' we opt for). It follows that coherence does not depend on syntactically definable surface marking. Nor do we get very far if we try to define constraints for conjunctions in stringent syntactic or semantic terms (cf. Robin Lakoff 1971). A sentence such as (7) (a) Susie has the mumps and the percentage of Moslems in Guinea-Bissau is thirty-eight. will seem distinctly odd until we provide it with a contextual motive: (7) (b) A: "Where is Susie? She was supposed to tell me how many Moslems there are in Guinea-Bissau." B: "Susie has the mumps and the percentage of Moslems in GuineaBissau is thirty-eight." Those who find B's contextualized reply in (7b) less bizarre than the uncontextualized sentence in (7a) will note the difference. The question posed by speaker A in (7b) provides a contextual justification for talking about Susie's mumps and Moslems in Guinea-Bissau in the same sentence in two clauses connected by and. Thus the conditions for the use of and cannot be defined through syntax alone. (There is, by the way, a parlour-game here: try producing bizarre conjunctions and ask your friends to justify them by context!) Arising from considerations such as these, a common stance among text linguists has been to distinguish between cohesion, meaning syntactically definable linkage between elements in a piece of discourse, and coherence, meaning the kind of connexity that relies on factors other than syntactically definable linkage. At first blush this distinction may seem neat enough. Still it leads to problems. Thus 'syntactically definable' is not an absolute term: what is 'syntactically definable' will of course depend on what kind of syntax we have at our disposal.

Connexity, Interpretability, Universes of Discourse

165

If, for instance, we have a syntax which contains a lexicon which marks the words pass, forward, dribble, goal, referee, kick, and offside with a shared feature, say < +football >, we can assert that example (5) has syntactically definable cohesion. And if we can similarly build in the connection between a superordinate concept such as flower and its hyponyms such as tulip into a syntax, instances such as (8)

The flowers were terribly expensive. One tulip was all I could afford,

or even (9)

My outboard motor stopped. The plugs were wet.

can be reduced to syntax. In practice, however, it is hard to imagine a practicable syntax with a lexical component capable of cataloguing the infinitely complex relations that obtain between words, their semantic frames, wholes and parts including all the world's machines and the processes of their running and maintenance, and so on ad infinitum. For the practical linguist it seems wiser to refer coherence to the world picture of a linguistically and pragmatically competent person. Besides, all syntaxes and lexicons will ultimately build on what competent informants, such as the grammarian himself or other trustworthy people, have intuited about their language, or put into bodies of text on whose basis grammars and dictionaries are written. So, why not trust relevant intuitions directly? Coherence is obviously not a structural property of a single sentence. And if coherence cannot in practice be described exclusively in terms of syntactic structures even of successive sentences, it follows that we should start worrying about the processes involved in incremental text comprehension. Exemplifying such processes in detail readily turns into a space-consuming job, and I shall therefore only offer a brief sketch around the beginning of a simplistic, fabricated text, namely (10)

The tablecloth was dirty. Susie called a waitress, but she refused to change it. Therefore she had to do without a tip.

To understand (10) we must know something about what goes on in restaurants, or about the 'restaurant script'. The hearer or reader of (10) begins from the start to build up a text world which ought to tally with the meanings he extracts out of linguistic forms in the text, conform to his previous knowledge, and be internally consistent in terms of certain consistency criteria (and I shall here bypass the interesting question what to do with 'counterfactually non-consistent' texts such as the nude girls took their clothes off or the melted ice made a crystalline tinkle in my dry martini). Thus the tablecloth evokes a universe of discourse which has tablecloths and what goes with tablecloths (tables, textiles, furniture, and so on). The tablecloth was dirty adds the existence of dirt to the universe of discourse and describes a text world with a specific state of affairs, namely a dirty tablecloth (presumably known or inferrable to the receptor of the

166

N. E. Enkvist

text because introduced text-initially with the). Susie called a waitress evokes a universe of discourse where Susie and several waitresses exist, and a text world in which Susie called a waitress. From our knowledge of universes of discourse we identify the scene as a restaurant, because that is where one calls on waitresses: the text world is the specific restaurant where Susie was confronted with a dirty tablecloth and called upon a waitress, one not known to us as signalled by the indefinite article. That Susie was in fact a customer and that she was annoyed with the waitress's neglect of her duties becomes clear in the remainder of the text, which the reader is invited to look at on his own. Thus when we are exposed to an emerging text, certain elements and their collocations in the text activate references to a semantic universe of discourse, definable as a conceptually organized and retrievable system of models of reality, and lead us to a specific text world characterized by a highly constrained, specific set of states of affairs. Initially the receptor of an uncontextualized text must go from text to universe and world: the process might be characterized as bottom-up or as text-driven. But then his knowledge of universes is brought to bear on the text and its world through top-down, knowledge-driven processing, which enables the receptor to infer matters not explicitly mentioned in the text and to anticipate ways in which the text might go on. The processes are further complicated by the iact that universes of discourse are added to, and new frames and schemata and scripts added to them, from evidence drawn out of the texts we process. Text comprehension and interpretation can thus be seen as a highly complex, incremental process involving the interplay of bottom-up and topdown processing, as well as zigzagging between the text, the universe of discourse meaning the universe at large within which the text can be placed, and the specific world of the text with its specific, usually highly constrained states of affairs. Those wishing to model text comprehension in these terms will therefore need both a universe of discourse containing knowledge about the world, and a text world explicating the specific, single, constrained projections into states of affairs that are described in the particular text. Whether other levels and worlds are indicated depends on what type of model of the processes of comprehension and interpretation we are interested in setting up. In current semantics and artificial intelligence, what I have called universes of discourse are usually modelled as frames, schemata, scripts and memory-operating packages (see e. g. de Beaugrande and Dressier 1981, Schank and Abelson 1977, Schank 1982, and Schallen 1982), which are often stored as associative networks (Findler 1979). One avenue towards modelling the incremental construction of the specific text world might be through a sequence of successive constraints achieved by the elimination of alternatives: I shall return to this in Section 5 in my discussion of text strategies. Meanwhile, however, a word about inferencing, which often plays a crucial part in these incremental processes. Inferencing can be defined in different ways, for instance - and roughly - as the introduction of specifying elements from a

Connexity, Interpretability, Universes of Discourse

167

universe of discourse into an emerging text world, even when they have not been explicitly mentioned in the text. Often such inferred elements are used to build bridges between utterances or their elements. Inferencing is what makes possible dialogues such as the often-cited example (11) John: "Mary! The telephone!" Mary: Tm in the bath." A Martian innocent of our culture might need a long series of footnotes to spell out the inferences we take for granted. We all know that what Mary said was short for something like (12) Yes, John, I heard you and I know that when you said "Mary! The telephone!" you meant that I should answer it because otherwise you would have answered it yourself instead of calling for me, but I don't want to answer it because I am having a bath and, as you know, when people are having a bath they do not like to get out of the tub and go into the hall to answer the telephone because they are wet and soapy and get cold and produce puddles on the floor and the bathwater cools o f f . . . Another, authentic instance. One day when I came home from the office the following dialogue took place: (13) Tua: "Peckie refused." Nils: "Where are the gloves?" Peckie was our pugnacious tom-cat whose fighting wounds had infected and who was on penicillin. After one try, Tua decided that her husband was the right person to persuade Peckie to take his pills, and similarly after one try Nils discovered that a pair of thick gloves were of decisive help. Note that had Tua needed the aid of a stranger sne would have had to embark on a long and elaborate explanation. The reliance on inferencing is of course a simple instance of receptor adaptation of the message: as the Gricean maxims tell us, we should give enough but not too much information. And in deciding what is enough we should reckon with the receptor's inferencing capacities. I shall return to this point too in Section 4 on text strategies. Meanwhile let me note that for communicative success, the receptor should be capable of construing a text world which is isomorphic with the text-producer's world in all relevant respects. The relevant features of the isomorphy can be explicitly specified, or inferred by a competent receptor. A special instance of inferencing is allusion. Allusions add a new 'vertical context', bringing in a new universe of discourse to bear on the text world (Schaar 1975). Thus if a man falls in the street and I remind a literate conversation partner of Satan's plight, he will add to his text world some features he finds relevant in Paradise Lost, perhaps mainly Book I. Along similar lines, allegory could be defined in terms of a text world which is a simultaneous, sustained projection of two universes of discourse.

168

N. E. Enkvist

Speaking about inference and allusion, it has been suggested, with the support of empirical data from children, that there are cognitive differences in the way receptors process writing and speech. Readers, it has been said, are more literal-minded and keep more to what has been actually specified in the text, whereas listeners treat what they hear more in the way of clues to what is actually meant. In resolving uncertainties, readers would then create their own worlds, whereas listeners try to enforce consistencies with the known world they share with their interlocutors. Such differences seem plausible enough as long as we are contrasting, say, storytelling or literary fiction or history with familiar small talk face to face, where the former text types must necessarily rely on evoking an unfamiliar world and the latter can refer to a shared and familiar world. To what extent such differences exist between, say, familiar letters and familiar conversation is another matter. (See Hildyard and Olson 1982.) In simulating natural-language comprehension by computer too, the importance of top-down processing and inference from text universes back to text worlds becomes obvious. One of the reasons why the early, simplistic attempts at machine translation failed, even to the point of despair, was that they could not manage the semantic and pragmatic aspects of language processing. "We came to realize," says David L. Waltz in a state-of-the-art paper, "during the 1970's that we needed to endow natural-language programs with 'common sense', which can only be based upon a body ofknowfedge of the outside world. In understanding language, people bring a large amount of information to bear that cannot be deduced from language itself. A sentence is never a formula or a program that is complete in and of itself." (Waltz 1982:14.) Indeed we can say that the degree to which we succeed in building knowledge of the world into a computer, and can retrieve and use it, will determine how well the computer will simulate human linguistic behaviour. In machine-aided translation for instance, we already have operational systems that save considerable amounts of human effort, provided however that the translation deals with a limited range of texts (such as medical ones), limited combinations of languages (such as Englishinto-Spanish and vice versa), and is routinely post-edited by skilled editors (Vasconcellos 1985). We can today build operationally satisfactory models of certain strictly limited universes or discourse for so-called expert systems, but we cannot set up detailed replicas of a complete individual human mind. (For further discussions of concrete evidence on text interpretation and related matters, see e. g. Sanford and Garrod 1981, van Dijk and Kintsch 1983, Bower and Cirilo 1985, and collections such as Flores d'Arcais and Jarvella 1983, Longuet-Higgins et al. 1981, Lehnert and Ringle 1982, Joshi, Webber and Sag 1981, MandVStein and Trabasso 1984, Sözer 1985, etc.). Another topical task in many ongoing projects which ultimately involves universes of discourse and text worlds is the automatic tagging of computer corpora. By tagging we mean the adding of grammatical labels to elements of a text, beginning with the identification of word classes. But as ambitious tagging tends to expand to include more and more syntactic information, it gradually

Connexity, Interpretability, Universes of Discourse

169

merges into parsing. And - what is more - any algorithm for tagging must reckon with some syntactic, and even discoursal, information. Should we for instance wish to tag a telegraphic or headline-type sentence such as the classic Ship Sails Today we must go beyond that sentence to decide which of its meanings is most likely in the given context and world. And even syntactically unambiguous sentences such as Constance can draw, can only be tagged with the aid of syntactic and collocational clues. Is Constance here a proper noun, and then the name of a woman or of a city (Konstanz), or is it a common noun signifying one of the virtues? Is can a modal auxiliary or a substantive? Is draw a verb or a substantive? In assigning grammatical tags as well as meanings to ambiguous elements and structures, human beings presumably proceed by a best-guess strategy. They try to maximize relevance by choosing the most probable and least surprising (and in that sense the least informative) meaning: the least meaning is the best meaning, to cite the maxim referred to as Joos's Law (Joos 1972). Philosophers have similarly spoken about the 'principle of charity' for the efforts we make to minimize disagreement and thus to extract the best possible sense out of an utterance (Davidson 1984). If we wish to imitate such best-guess strategies in automatic tagging by computer we can choose between two methods. Either we parse the sentence before tagging it (which may involve risks of circularity, as in some models of language processing laggings can be regarded as prerequisites for parsing), or then we opt for best guesses as indicated by collocational probabilities that can be computed by Markov models out of large bodies of text. The point I wish to emphasize is that collocational probabilities must come from somewhere: in algorithms for tagging they are obtained by counting collocations in a corpus of authentic text. And in authentic texts, collocations necessarily reflect the structure of real, authentic universes of discourse. Even if we opt for Markovian strategies of tagging through best guesses based on collocational probabilities, the tagging procedure comes partly to rely on universes of discourse, because the probabilities come from texts that reflect such universes. 3

This array of heterogeneous examples ought to reveal their common denominator. Ultimately they all raise the classic question what interpretability consists of. The interpretability of a text is one kind of functional well-formedness. To do its job in communication a text must be interpretable, and interpretable in the right way. Therefore interpretability should be viewed in relation to the other kinds of well-formedness that linguists have worried about. As the concepts and terms related to well-formedness have been variously used by different linguists, a brief rehearsal of one of the relevant sets of definitions may be in order.

170

N.E.Enkvist

Grammaticality, then, can be defined as indicating the conformity of an expression, or of part of one, to an already existing grammatical description of the language. Thus one and the same expression may be grammatical in relation to one grammar which approves of it, but ungrammaucal in relation to another which condemns it. Acceptability relates an expression or utterance to receptor judgements. To study acceptability we have to ask suitably selected informants (or, in the extreme instance, ourselves) to what extent they are prepared to approve of a certain structure or expression. Acceptability can also be measured indirectly by asking informants to manipulate an utterance and by measuring their success, failure and compliance (Greenbaum and Quirk 1970). Finally, appropriateness is a measure of acceptability in a specific context and situation (on these concepts, see Enkvist 1982). Appropriateness expresses degrees of conformity of an expression or utterance to certain social expectations and stylistic standards or norms. But as the examples in the previous sections should have shown, there is yet another family of concepts that intrudes into judgements of textual function, namely that involving cohesion (still in the sense of syntactically describable intersentential linkage) and coherence (correspondingly, in the sense of connexity devices not susceptible to syntactic description). From the point of view of the receptor, connexity (in the sense of cohesion plus coherence) and interpretability are closely related concepts. Connexity refers to a characteristic of the text, whereas interpretability suggests communicative success. Connexity can be regarded as the sum of the qualities of a text which make it interpretable. Interpretability is not exclusively a syntactic concept. Beside syntax it involves semantic and pragmatic components (allowing for the overlaps between these three fuzzy-edged concepts). As interpretation involves universes of discourse as well as text worlds (in the sense defined above in Section 2), we might also try to regard the establishment of relevant universes of discourse as a semantic component in interpretation, and the building up of text worlds as its pragmatic projection. Universes of discourse indeed come from a general semantic knowledge of the world, whereas text worlds are characterized by sets of specific states of affairs constrained by the specifications given in individual texts, and the relevant inferences: they are pragmatic in nature insofar as they reflect the use of language to describe one specific world. A student of text interpretation should also take an interest in the variables that affect the interpretive process. One variable is individual: a given text may be interpretable to one person but not to another. Also, different persons may interpret the same text in different ways. Then, the situation and context of a piece of discourse will affect its interpretation: the same utterance and text can be interpreted by the same person but in different ways depending on the situation. (A classic example is the lecturer's lights please! winch may mean both 'please turn off the lights' or 'please turn on the lights'.) The interpreter's purpose and motivation are yet other variables that affect interpretation. A linguist may for instance read Pope's poems for pleasure, but he may also look at them on a hunt

Connexity, Interpretability, Universes of Discourse

171

for-rhymes as evidence of 1 Sth-century pronunciation, or as data on the effects of a strict metre on word order. We should also reckon with the receptor's knowledge of the speaker or writer and of his presumed motives. There are, for instance, kinds of irony whose spotting requires an intimate knowledge of the speaker's or writer's person, his knowledge and his verbal habits. And there are poems which mean more to us if we know by what poet they were written, and in what emotional plight. For one more consideration showing how interpretations are affected, not only by the text as such but by the situation, consider the following anecdotal but authentic example. At a conference where several hundred people were indulging in malt whisky beside the lawn of a Scots castle, with the local police bagpipe band marching up and down on the green, Mrs A mistook the German Professor X for the Finnish Professor Y, and with the best intentions addressed him in Finnish. When Professor X looked puzzled Mrs A articulated her message more directly into his ear, trying to outdo the bagpipes. After the denouement Professor X confessed that he had for a moment entertained grave doubts about Mrs A's resistance to the local produce. But was Mrs A's formidable gibberish (in fact, friendly but loud and insistent Finnish) meaningless and uninterpretable? Referentially, yes: Professor X did not understand what Mrs A was trying to get across, he could not build a world around her Finnish text. But situationally and pragmatically Mrs A's behaviour was far from meaningless. It proved her sociability, while also suggesting something of her condition: it had symptomatic meaning. Or, to use Deborah Tannen's term (Tannen 1986), the metamessage seemed loud and clear even where the referential message proper failed to reach its target. Thus in authentic communication, and not least in face-to-face situations, a message is also interpreted as a symptom of its producer's situation, moods and attitudes, and not only referentially in relation to the text world and universe of discourse referred to within the text itself. In psychiatry, texts are in fact often read for their symptomatic value rather than for referential content (cf. e. g. Rosenbaum and Sonne 1983). Symptoms often creep into texts unbeknownst to their producer or even against the producer's will. In the light of all these considerations and variables, how could we try to define interpretability? We must obviously take recourse to pragmatic concepts related to communicative success and not only look at text-internal structures or formal abstractions of such structures. We might therefore posit that a text is interpretable to a certain individual receptor in a situation where that receptor is capable of constructing around that text a text world, and place the text world within a universe of discourse within which it 'makes sense' (or 'seems plausible'). Conversely, a text remains uninterpretable to those who cannot build a world around it and place it plausibly within a universe. 'Makes sense' and 'seems plausible' are, however, vague, and here circular, expressions crying out for further specification. We might proceed to explicate them along various avenues. One such avenue involves truth-conditional

172

N. E. Enkvist

semantics: we might suggest that a text is interpretable to those who can build around it a world in which that text might be true ('might be' rather than 'is' to allow for fiction and fantasy). And another avenue leads us to pragmatic concepts such as the Gricean maxims (Grice 1975) or the Sperber-Wilson theory of relevance (Sperber and Wilson 1986). Those who prefer this route can define texts as interpretable when they can be placed in a world and universe in which Gricean or Sperber-Wilsonian principles hold. I must here desist from further discussion of what could be viewed as two different kinds of interpretability, truth-conditional and pragmatic, and go on to a brief discussion of mechanisms of text interpretation in relation to text worlds and universes of discourse. To many practical linguists, the mechanisms and stages involved in text interpretation are of more immediate interest than general, speculative definitions of interpretability which have not been backed up by empirical data. One way of classifying stages in text interpretation is in terms of 'intelligibility', 'comprehension' and 'contextualization'. In such a view, a piece of discourse is intelligible to those who can successfully segment it into its phonemes and group these into morphemes, phrases, clauses and sentences. It is comprehensible to those who can assign basic, canonical meanings to the intelligible structures that have been identified by segmentation and grouping. Comprehensibility thus involves relating the text to semantic universes of discourse. And it is contextualizable to those who can assign to it a situation-bound, communicative meaning by surrounding it with a specific text world where specific states of affairs prevail. Thus - to take up yet another classic example - the sentence It is hot in here, is intelligible to those who can identify its phonemes, words and syntactic structures. It is comprehensible to those who understand that it can be paraphrased as 'the temperature in this room is high' or translated into, say, the Swedish detärhetthärinne. And it is properly contextualized by those who also capture the speaker's possible purpose or appeal when it is uttered in a hot room with closed windows beyond the speaker's reach: 'will somebody please open a window'. These stages of interpretation do not, however, take place in neatly discrete order, one after the other. We can cruise from one to the other just as we zigzag between the intelligible text, the universe of discourse which reflects canonical meanings, and the text world which reflects specifically constrained states of affairs. For instance, we all know how a situation may help us to hear utterances correctly despite noise: if so, context becomes an aid to comprehension and intelligibility. We can also relate intelligibility to syntax, comprehension to semantics and contextualization to pragmatics, though with the usual fuzziness and overlap which tend to obfuscate such global concepts. Altogether, the view of interpretation as a resultant of intelligibility, comprehension and contextualization is wholly compatible with our discussion of interpretability, universes of discourse, and text worlds. In traditional linguistic terms we might go on to analyse the processes involved in text interpretation as involving phonetic or visual perception,

Connexity, Interpretability, Universes of Discourse

173

phonemic or graphemic analysis, morphemic analysis, lexical lookup, choice between meanings through projections within the sentence and between sentences, and choice between potential pragmatic and communicative function through projections between discourse and situation. This is what linguistics and pragmatics are all about. In terms of process linguistics we might further inquire into the nature of the basic processes involved in these linguistic and pragmatic operations. We might for instance ask ourselves whether all these operations could be modelled in terms of a small set of fundamental processes: element or pattern identification (answering the question: are χ and y identical or different in terms of certain specific criteria?), transformation (can χ be changed into y ?), and projection (is there a correspondence between the one structure or system with that of the same, or another, element or element combination in another structure or system?). Here I can only list such problems in passing because they show what kinds of approaches we must adopt if we are to build linguistic, or processual, models of text interpretation. Should we wish to study what actually happens when a person is faced with a piece of discourse, we can proceed in two ways, as always when collecting behavioural data. We can either try introspection and use that ubiquitous and easily accessible, though often unreliable, informant: ourselves. This is the classic method of literary critics. Or we can elicit data from suitably selected informants by some suitable experimental technique, for instance from protocols documenting stages in incremental text comprehension. The latter approach involves all the problems inherent in protocol studies (e.g. McHoul 1982, giving ethnomethodological data on coherence in a poetic text and on interpretations of sequencing in narrative; and Enkvist and Leppiniemi, in preparation). Elicitation methods might for instance be based on asking informants to describe the world and universe in which that text might be true. Another technique is to ask informants to summarize a text, assuming that summarizing must be based on a construction of a plausible and consistent universe of discourse around the text (see e.g. under 'macrorules' in van Dijk 1980). 4

In classic structural linguistics, the text was seen as mediating information from speaker to hearer: the formula used to be S(peaker) — T(ext) - H(earer). Such a tripartite formula is of course an oversimplification. In actual communication, a person who has decided to speak or write (a decision which is itself of interest) will first consider what his receptor already knows; under what circumstances and in what context the receptor will process the message; and what the receptor's message-processing capacity is likely to be under the circumstances. Only then will he produce a text. One of the proofs of this is that the distinction between information a speaker thinks is known to the hearer, and information he thinks is new, is so fundamental that it guides text organization at many

174

N. E.Enkvist

levels, including the choice of words and syntactic structures. For instance the choice between Susie is Peter's daughter or Peter is Susie's father will depend precisely on such considerations. And compare the active/passive distinction in (14) (a) A: "What did John do?" B: "John ate the sandwiches." But not * "The sandwiches were eaten by John." (b) A: "What happened to the sandwiches and the apples I left on the table?" B: "The sandwiches were eaten by John. And the apples Susie took with her." Here the strategic choice of syntactic form depends on what the speaker regards as old, and what as new, information for the hearer. I shall return to this in connection with text strategies in Section 5. I have suggested that interpretability depends on a receptor's ability of building a specific world around a text, using a universe of discourse as an aid. Now "build" can be regarded as a metaphor. If so, what are the bricks and what is the mortar we use in building worlds around texts? To answer this question, let us first consider what information is. Information, so the theorists tell us, is certainty. And certainty is arrived at through the elimination of alternatives. But as alternatives can be eliminated in different order, a text producer must order his text according to the sequence in which he wants to eliminate alternatives, and in which he wishes incrementally to specify the text world. In a somewhat different perspective I already illustrated this above with (10). To take yet another brief example, if I say, (15) In Belfast, two bombs exploded last night. in Belfast eliminates all other locations, two bombs all other noun phrases from the relevant slot, exploded dk other verbs, and last night all other possible times. By eliminating alternatives I have thus specified a text world, and I have done so in a specific incremental order, by beginning with in Belfast and ending with last night. Each exclusion of alternatives can be compared to a brick because it adds a unit, or 'chunk', of meaning to the specification of the universe. And the way in which units of meaning are strung and cemented together by a specific strategic order (though one constrained by syntax) might be said to correspond to the mortar. This ordering is one of the domains of text strategy, and has to do with optimizing the communicative interplay between old and new information under the constraints imposed by the canonical principles of English syntax. To anticipate, let me note that text strategies also regulate the ways in which the discourse is linked to situation and context, and in which the density of information is optimized by dilution or by concentration so as to trade off economy against explicitness in the desired fashion. Unless we are satisfied with using the term 'strategy' as a loose metaphor, we shall need a process model defining it in terms of decision-making. A simple

Connexity, Interpretability, Universes of Discourse

175

version of such a model could be sketched as follows. Whenever we find ourselves at a point where we must choose between alternative actions, we must make a decision. Many such decisions, in syntax for instance, are made intuitively, subconsciously and with apparent automatic ease. The decision-maker then remains unable to retrieve the alternatives or explain his actions, precisely as he is unable to explain other aspects of his syntactic or lexical behaviour. Other decisions may be more conscious. Still, whenever we make a decision we make it by evaluating the factors that affect the decision and then by choosing the alternative that best fits our purpose. (At least we assume that man's actions can be thus rationally motivated.) And the choice of the best alternative is made by referring the decision-affecting factors to a strategy, definable as a goal-directed evaluation system of decision-affecting factors. In a particular situation, certain considerations lead to a weighting of those factors which seem more important than others. They are then assigned priorities by the strategy, at the expense of those other factors that conflict with them, and the decision is made accordingly. So, in choosing among possible strategies - which in turn determine the shape of the text, including its style - one fundamental question is, in what order shall I set out to eliminate my communication partner's uncertainties to lead him to the appropriate text world? Which uncertainties must I eliminate explicitly, and which of them can I trust him to eliminate on his own, by inferencing? More than thirty years ago Dwight Bolinger pioneered the discussion of the 'linear modification' underlying what I have called text strategies: (16) Let us consider what happens when elements - call them words, for convenience - are laid end-to-end to form a phrase. Before the speaker begins, the possibilities of what he will communicate are practically infinite, or, if his utterance is bound within a discourse, they are at least enormously large. When the first word appears, the possibilities are vastly reduced, but that first word has, in communicative value for the hearer, its fullest possible semantic range. The second word follows, narrowing the range, the third comes to narrow it still further, and finally the end is reached at which point the sentence presumably focuses on an event. (Bolinger 1952/1972: 32.)

In fact, as every cloze-tester knows, such a model of incremental world-building around a text is not altogether accurate, because the reduction of possibilities does not increase in steady linear progression but rather stepwise. There are points where what comes next is wholly predictable, and other points where it is impossible to guess what will follow. Still a renewed contemplation (cf. example 15) of sentences such as (17) Two bombs exploded in Belfast last night. In Belfast two bombs exploded last night. Last night two bombs exploded in Belfast. In Belfast last night two bombs exploded. Two bombs exploded last night in Belfast. and of the various contexts in which each of them seems most natural shows that each of the permutations could have its own use, its own 'textual fit', its own

176

N.E.Enkvist

preferred context. The linearization of syntactic hierarchies into sequences of symbols on the textual surface is a process often discussed (cf. Enkvist 1978, Levelt 1981) but not always in terms of overall text strategies. Yet such strategies are crucial because they give each sentence its textual fit and make it serve the overall communicative design. To borrow a term from Jan Firbas, linearization is one of the important devices patterning the 'communicative dynamism' (or CD for short) of the text (Firbas 1981). In the unmarked instance, in written and in many spoken texts, the most common policy is to begin clauses and sentences with old information and then go on to new. In the terminologies common among linguists, themes precede rhemes or topics precede comments. However, if we depart from this unmarked pattern by putting new information before old, we must mark what we are doing to caution the receptor, by intonation (a high-falling tone) on the element carrying new information early in the sentence, or by cleft (Enkvist 1980). The general strategy of putting Old Information First (OIF) makes good teleological sense. As Bolinger implied, on the whole guessing what comes next becomes easier, the more we have heard or seen of a sentence. Conversely, guessing is most difficult at the beginning of the sentence. Therefore if we want to ease the receptor's processing load it seems sensible to begin a sentence or clause with old, and therefore more predictable, information. This spreads the onus of processing more evenly over the sentence. And if we assume that sentences are interpreted by reference to universes of discourse as modelled in frames, schemata and scripts, getting to know the theme of a sentence makes it possible to retrieve some relevant frames, schemata and scripts at once, and then to put them to use in the interpretation of the remainder of the sentence. But there are texts and discourses in which we see strategies other than OIF. In normal impromptu dialogue, exchanges such as this are common: (18) A: "Where did John go?" B: "(To) Paris." A: "When?" B: "Day before yesterday." Such exchanges have raised problems for students of information structure, theme and rheme or topic and comment. Utterances such as Pans, and Day before yesterday, might be said to consist of themeless rhemes or topicless comments, the theme or topic being recoverable from the previous question. Indeed in informal dialogues such as (18), answers such as He went the day before yesterday, might seem artificial and remind us of language classes where we were told "to answer with a complete sentence". One solution, then, is to regard B's utterances in (18) as elliptic, the deleted element being recoverable from context. But another solution is to say that such structures arise, not from a process where an underlying complete sentence has been ellipted, but from a different general strategy which we might label as Crucial Information First (GIF), and whose variant in (18) could be called

Connexity, Interpretability, Universes of Discourse

177

Crucial Information Only (CIO). There are situations, so runs the argument, in which it is unnecessary to waste time and energy on old information, givens, themes and topics, because the receptor is certain to identify it and fill it in anyway. So, the speaker can opt for GIF or CIO. For obvious reasons the GIF strategies are likely to occur in the kinds of communication where the partners share a situational context and where immediate feedback can be relied upon to set things right if misunderstandings arise. I presume one reason why linguists have so far concentrated on OIF strategies is, simply, written-language bias (cf. Linell 1982). Granted, OEF is the dominant pattern, in many and perhaps the majority of spoken texts as well, but GIF and CIO still deserve attention too. The GIF and CIO strategies are at a premium especially where a speaker must strive for maximal economy. This happens for instance in radio reports of sports events such as boxing matches, where the speaker has to adopt a machine-gun tempo to catch up with climactic events in the ring. The following sample comes from one of the unprinted texts of the London-Lund Corpus, with grateful acknowledgement to Sir Randolph Quirk and Professor Jan Svartvik. (19) (bell) round three now they come out a little faster this time at least Miteff comes out as quickly as Cooper nice left jab by Cooper there right bang on the spatulate nose of the Argentinian (Text S. 10. 3., computer printout, lines 617-626.) and a tremendous left hook to the head there by Miteff catches Cooper... (Ibid., 630-632) one two thunderous punches to the body there by Miteff... (662-664) Cooper first to land with a long light left to the body now a new style right hand up almost over his right ear Miteff left hand down about chest level Cooper jabs it past it... (809-816) Miteff suffers no real answer long left to the body long left to the chin right and left by Cooper... (891-895)

In this excited commentary, the commentator - obviously a skilled master of his craft - time after time mentions action before actor: nice left jab by Cooper, a

178

N. E.Enkvist

tremendous left hook to the head there by Miteff, long left to the body long left to the chin right and left by Cooper. On the radio, where the listener does not see the fight (as he would on tv), the strategy adopted was to report on events at once, and then fill in the actors for clarity when necessary, and with an agentive phrase (by Cooper, by Miteff), with verb deletion. The style is highly nominal, consisting of nominalizations and noun phrases and preposition-phrase adverbials, often given as fragments without a complete SVO structure, subjects being identified through postposed agents. Novelists often try to catch salient features of spoken dialogue. Let me therefore cite a GIF example from fiction: (20) "Your're early about the house," he said without surprise. There was still sleep-dirt in his gummed-up eyes and his hair was knotted and tangled, not combed this morning. He yawned hugely, so hugely she noticed a decaying molar. "Would you like some tea? I hope it was all right. I mean, to make tea." "Oh, yes, at this time of day. Lashings of tea, I'd like, and three sugars" (Angela Carter, The Magic Toyshop. London: Virago Press, 1981, p. 61.)

"Lashings of tea I'd like" - lashings of tea is crucial, I'd like is in the nature of a polite afterthought. There are various devices enabling us to linearize a text and adjust the textual fit of a sentence in the way required by the strategy. There are for instance the choice between lexical converses, the adjustment of syntactic constructions (of which example (14) already gave us a sample), and topicalization (fronting without change in syntactic relations between constituents), which (as we noted above) serve to put old and new information into the desired place, as in (21)

(a) John sold the car to Peter. Peter bought the car from John. (b) What did John do? He ate the sandwiches. What happened to the sandwiches? They were eaten by John. But not What did John do? * The sandwiches were eaten by him. (cf. example 14). (c) A: (pointing at one of several cars in a garage) "This car I drove to New York yesterday."

And high-falls and clefts can be used to mark new information when it comes early in the sentence, as in (21)

(d) It was JOHN who came to Stockholm.

But another subject so far neglected by linguists and rhetoricians has been the influence of text strategies on certain word-order patterns, especially those involving the strategic placement of adverbials. To avoid long quotations I shall illustrate what I mean by fabricated, prototypical examples, whose verisimilitude is easy to check from countless exponents of the relevant text type. Thus

Connexity, Interpretability, Universes of Discourse

179

the locative strategy of a travel guide favours beginning sentences with topicalized adverbials of place, as in (22)

Opposite the entrance is the main altar. Behind the altar is the tomb of X. On the right wall is a plaque commemorating Y.

A chronicle often topicalizes temporal adverbials: (23) In 1889, X took place. Three years later A did B. And in 1908,... Narrative texts may contain sequences with a thematized hero: (24)

Susie came home. She was tired. She wondered which of her dresses she ought to take for the banquet. She went to her clothes closet...

Such sequences dominated by a locative, temporal or agentive strategy might be called single-strategy sequences. But underlying the linearity of language, which after all evolves over time, there is always a certain temporality. Thus in the locative-dominated text in (22) it is understood that the visitor wul see the sights in the temporal order reflected by the order in the text. And the same is true of (24). In the unmarked instance, the linear order in which matters are presented will be taken to stand for the temporal order in which the concomitant events took place or operations should be carried out. Thus the two sentences (25)

(a) John and Mary got married and had a baby, (b) John and Mary had a baby and got married.

actually suggest different sequences of events. And when a cookery-book writer topicalizes a place adverbial even with the verb put (which needs strong motivation because of the strong valency bonds between put and its locative), as in (26)

Into a champagne glass put three lumps of sugar.

this is a pregnant abbreviation standing for 'first take a champagne glass and then put into it three lumps of sugar'. What strategies such as those illustrated in (22-26) have in common is that their linearity signals the ordering of experience, iconically, and also indicates the strategy that holds the text together and gives it one of its major macropatterns. We can speak of experiential iconicity (Enkvist 1981) or, with an older term, of ordo naturalis (which was an ambiguous term also applied to canonical word-order patterns in syntax). These examples of texts with locative, temporal, and agentive strategies of course do not cover more than a fragment of the text-typological spectrum. There are other strategies, for instance one we might call taxonomic which classifies parts of a whole (and may make use of locative and temporal organizing principles), the argumentative (whose categories have been analysed e.g. in Tirkkonen-Condit 1985), and so forth. In the real world of discourse, consistently single-strategy texts are rare and occur in their purest form in, say, travel guides and chronicle-type lists of events and biographical articles or curricula vitae. More common are dual-strategy or multiple-strategy texts, in which we

180

N. E. Enkvist

may often find a dominant strategy to which other strategies are hierarchically subordinate. A biography for instance may have a superordinate agentive strategy, whose chapters - each dealing with one type of activity - may be organized chronologically. But a biography may also have locatively organized sections, taking up what the biographee did in one specific location. In such instances the text-producer can choose between various alternative strategies. All the same these concepts may help us to grasp concretely the strategies that make up the total disposition of a piece of discourse or text, and also to analyse the exponents of such strategies on the textual surface. For instance the fronting of adverbials serves to signal what strategy the speaker or writer has opted for in the relevant part of the text. At the same time, such strategy markers also indicate the division of a text into sub-units. In a locative-strategy text for instance, a new fronted adverbial of place serves to mark a shift to a new location (or perhaps a new subpart of a previously specified location) and thus a new text unit. And in a temporaldominated strategy a new adverbial of time indicates a shift to a new point in time, and in that sense to a new, temporally definable text unit. The number and kind of interwoven text strategies, and the number of levels and sublevels within each strategy, may give us an indication of the complexity of the text structure and of the accuracy of the picture the text is trying to present of its world and universe of discourse. The purpose of these brief hints at certain fundamental concepts in the study of text strategies - the choice of lexical and syntactic patterns to carry out OIF or GIF principles in text strategy; the ways in which linearity is made use of to signal temporality; the use of locative strategies in guidebooks and temporal strategies in chronicles and agentive strategies in hero-dominated narrative; the typological distinction between single-strategy texts or passages dominated by one such strategy, and between dual-strategy and multiple-strategy texts which blend strategies - has been to remind us of the ways in which text organization steers and guides the incremental building-up of a text world around the text. Worlds are built up, and discourses comprehended and texts interpreted, through the successive increase of information through the elimination of uncertainties. And the order in which uncertainties are eliminated is one of the things text strategies are supposed to regulate.

What, then, about texts that raise interpretability problems? What happens when we are faced with a text which in some way flouts the normal, canonical patterns of language? Where run the borders of interpretability? Referential interpretability may be seen as a threshold. Either an utterance is referentially interpretable, or it is not. But once an utterance is referentially interpretable, there can be degrees of interpretability depending on how much

Connexity, Imerpretability, Universes of Discourse

181

information a given receptor can extract from it in his specific situation. Thus a statement of the results of a leucocyte count may be interpreted by a layman as something that has to do with medicine, by a more knowledgeable layman as an indication of the condition of the blood, and by a doctor as a warning that the patient's state is serious. One of the remarkable qualities of human beings is that they always assume that texts they hear or see should make sense. They work hard to find appropriate universes of discourse and text worlds in which to place the text. If they did not, figurative language would be impossible and irony would misfire. If they fail, despite their efforts, they can either blame the text producer (as many people do when faced with modern poetry) or themselves (as one ought to do when challenged by scientific texts one cannot follow). Still there are sequences such as (27) the a some but thing a for which it is hard or impossible to find a universe of discourse and text world. And texts in an unknown language may make no sense to us. Another type of text failure results from purposeless misleading: we cannot say Jack kicked Jill. with impunity if what we wanted to say was Jill kicked Jack. It is therefore interesting to test the interpretive abilities of human beings by presenting to them texts which are in some respect unusual. One category of such texts comes from modern poetry. Take for instance the following: (28) more to going posthumous appeared editorship of little himself taste boy Diamond Ritz both than did "Not a good poem," said one discerning informant, "but it does make me think of Scott Fitzgerald." Indeed she was right. The text was composed by taking, at random, every fifth word from a passage in Carlos Baker's Ernest Hemingway (Pelican Books 1972, p. 561) which dealt with Fitzgerald. In our normal, relentless quest for meanings and references to build universes of discourse, we clutch at every straw. To the conoscenti, Diamond Ritz at once recalls Fitzgerald's short story, "A Diamond as Big as the Ritz". And once this has helped us to build a Fitzgerald world around the text, we can top-down-process new meanings out of it: more to going posthumous may thus remind us of the Fitzgerald revival ot the late 'forties and the 'fifties; editorship of little might stand for Fitzgerald's contacts with other writers, some of whom edited little magazines; himself taste hoy bears a whiff of the decadence associated with Fitzgerald's merry 'twenties. More seriously, a poem by e. e. cummings, a fine representative of the school of poets who liked to play and experiment with language (including Mayakovsky, Christian Morgenstern, Gunnar Björling and a host of others).

182

N. E.Enkvist

(29) quick i the death of thing glimpsed (and on every side swoop mountains flimsying become if who'd) me under a opens (of petals of silence) hole bigger than never to have been what above did was always fall (yes but behind yes) without or until no atom couldn't die (how and am quick i they'll all not conceive less who than love)

(cited from Fairley 1975:44.)

In her study E. E. Cummings and Ungrammar, Irene Fairley has given us her own interpretation of this poem. Her method was to 'reconstruct', that is, to bring Cummings's text closer to normal non-deviant syntax, for instance through 'shifts' or reversals such as me under a opens — under me opens a. This particular poem, says Fairley, makes use of three major types of shift: dislocation (a shift away from standard position), inversion, and intercalation (placing one constituent within another or interlacing parts of separate clauses or sentences). When such shifts are reconstructed, the poem becomes interpretable. (30) In my deliberations over this poem, I was struck by the possibility of two voices corresponding to the two "i's". Very likely the result of a combination of elements in the poem, "atom" and "above" when considered in relation to the initial description of impending doom ("fail" and later "die") and the final note of love and possible redemption - No! Atom couldn't die - suggested to me one of Cummings' favourite themes - man's smallness, God's omnipotence... the importance of love and faith. It is a devotional poem! (Fairley 1975: 44-45.)

Explications through reconstruction of course always raise a problem. If Cummings is interpretable only after reconstruction, what then is the purpose and meaning of his dislocations, inversions and intercalations? Are they merely an intellectual game, whose results must be subtracted from the meaning and left out without trace from the interpretation? Or do they have a function, for instance in drawing attention to those structures that we can only recover after considerable effort such as reconstruction? Are we expected to delight in them like children who find hidden Easter eggs? Or are there some arcane subtleties in the shifts which in fact get lost in reconstruction? Another interesting problem is to what extent such reconstructions can be based on mere mechanical syntactic permutation, and to what extent they are in fact built on interpretation. Is perhaps the text first read until it suggests a tentative universe of discourse, despite - or perhaps thanks to - its oddities; and is the poem then reconstructed through operations suggested by top-down processes, to bring it into harmony with the

Connexity, Interpretability, Universes of Discourse

183

universe of discourse? And to what extent do different readers interpret such poems with different degrees of success? We all know there are people who like modern, syntactically deviant poetry, and there are others who scoff at it. Perhaps such attitudes correlate with different interpretive skills. Those who find the poem interpretable, and have been willing to go through the pains and troubles involved, may like it. And those who do not may shrug it away. Yet another type of interpretation problem arises in science-fiction novels, which must rationalize a counterfactual universe of discourse in order to become interpretable. An example. "Everyone now knows," writes Kurt Vonnegut to start off The Sirens of Titan, (31) how to find the meaning of life within himself. But mankind wasn't always so lucky. Less than a century ago men and women did not have easy access to the puzzle boxes within them. They could not name even one of the fifty-three portals to the soul. (New York: Dell, 1970, p. 7.)

If somebody came into a professor's office and uttered this, the professor would be justified in interpreting the text as a symptom. But its ambiance in a book like the ambiance of a beer-can on a pedestal in an art exhibition - suggests a different interpretation in terms of a counterfactual, science-fiction world. By implication, men and women "now" (at the pretended moment of writing, some time in the future) have easy access to their puzzle-boxes and can readily name at least some of the portals to the soul. Which we cannot do today. A science-fiction writer's problem is how best to rationalize a counterfactual world. Unless the universe of discourse is counterfactual, the book does not qualify as science fiction. But if it is counterfactual it must still remain rational enough to be interpretable. For interpretability, the counterfactual elements must be rationalized as extrapolations of today's scientific knowledge, extrapolations that might become true (like sending men to the moon) or might not. To avoid longer quotations out of authentic science-fiction texts I shall give a fabricated example to show what I have in mind: (32)

My urge to see Amy again became so irresistible that I got myself a De-Re-Composition Permit from London to Sydney. I went to the Decomposition Station at Highgate Hill, and in due course, after a completely painless anaesthetic, recovered in the Recomposition Centre (as the Australians call it) at Katoomba, New South Wales. Two toes were missing from my left foot, which the Aussie Recomposers explained as a result of a slight malfunction of the atomic scanner on Highgate Hill.

I presume it did not take the reader long to interpret this text. Its writer had in mind a universe of discourse where people can travel by energy transfer, being decomposed, molecule by molecule and atom by atom, and then recomposed. Completely and without residue, if the trip was a success.

184

N.E.Enkvist

6

In this paper I have tried to view man as a processor of information through natural language. Information is obtained through the reduction of uncertainties by the elimination of alternatives. In text processing we eliminate alternatives with the aid of specifications given in, and suggested by, the text; we relate these specifications to what we know about the universe, and provide the text with a specific, constrained world in which the specified states of affairs obtain. If the receptor's text world is adequately isomorphic with that of the text producer with respect to those specified or inferrable features that are relevant in that particular instance, the communication is a success. A text should thus be seen as a trigger releasing a number of complex, interrelated processes of interpretation, which are steered by a text strategy. The strategy is chosen by the text producer with due regard to the real or assumed receptor's previous knowledge and text-processing capacity. Text strategies are manifested in macrostructural principles of text organization and, at the syntactic level, in the patterning and signalling of the distinctions between information a speaker or writer thinks the receptor already knows and has activated in his mind, and information that he presumes is new to his receptor. These strategies involve, among other things, the manipulation of lexis and syntax so that references to old information can be placed before references to new information, or, in some instances, so that new information comes first, old information sometimes being omitted altogether. Texts and portions of texts can be organized for instance by agent, by time, or by place; most texts will show a blend of strategies, or embeddings where passages dominated by one strategy occur within passages linked to each other by another strategy. All my various examples should bear on one basic point: text comprehension and interpretation are processes that do not involve formal syntax alone. Nor can textual connexity be explicated in terms of overt text-internal structures alone. Connexity and interpretability are two sides of the same coin: to be interpretable, a text must have connexity, and to have connexity, a text must be interpretable. And to interpret a text we must conceive a world. This was elegantly summarized by George Miller, as cited by Haj Ross (Ross 1983): "In order to understand what another person is saying, you must assume it is true, and try to imagine what it might be true of." What it might be true of is a text world. And to construe text worlds we must know universes of discourse.

Bibliography de Beaugrande, Robert, and Wolfgang Dressier, 1981. Introduction to Text Linguistics. London: Longman.

Connexity, Interpretability, Universes of Discourse

185

Bolinger, Dwight, 1952. "Linear Modification." Publications of the Modern Language Association of America, 67: 1117-1144. Reprinted in Fred W. Householder, ed., Syntactic Theory 1. Structuralist, Harmondsworth: Penguin Books 1972, pp. 31-50. Bower, Gordon H., and Randolph K. Cirilo, "Cognitive Psychology and Text Processing." In van Dijk 1985, 1:71-105. Davidson, Donald, 1984. Inquiries into Truth and Interpretation. Oxford: Clarendon Press. van Dijk, Teun A., 1980. Macrostructures. Hillsdale, New Jersey: Lawrence Erlbaum Associates. - 1985. Handbook of Discourse Analysis. New York etc.: Academic Press. 4 vols. van Dijk, Teun A., and Walter Kintsch, 1983. Strategies of Discourse and Comprehension. New York etc.: Academic Press. Enkvist, Nils Erik, 1976. "Notes on Valency, Semantic Scope, and Thematic Perspective as Parameters of Adverbial Placement in English." In Nils Erik Enkvist and Viljo Kohonen, eds., Approaches to Word Order (- Publications of the Research Institute of the Abo Akademi Foundation 8), pp. 51-74. Abo Akademi. - 1978a. "Coherence, Pseudo-Coherence, and Non-Coherence." In Jan-Ola Östman, ed., Semantics and Cohesion (- Publications of the Research Institute of the Abo Akademi Foundation 41), pp. 109-128. Abo Akademi. - 1978b. "Linearity and Text Strategy". In John Weinstock, ed., The Nordic Languages and Modem Linguistics 3, pp. 159-172. Austin: The University of Texas. - 1980. "Marked Focus: Functions and Constraints." In Sidney Greenbaum, Geoffrey Leech and Jan Svartvik, eds., Studies in English Linguistics for Randolph Quirk, pp. 134-152. London: Longman. - 1981. "Experiential Iconicism in Text Strategy." Text 1, pp. 77-111. The Hague: Mouton Publishers. - 1982. "Categories of Situational Context from the Perspective of Stylistics." In Valerie Kinsella, ed., Language Teaching Surveys 1, pp. 58-79. Cambridge etc.: Cambridge University Press. - 1983. "En not over multipla kriterier for korrelatbestämning." In: Struktur och variation. Festskrift till Bengt Loman (- Publications of the Research Institute of the Abo Akademi Foundation, 85), pp. 301-307. Abo Akademi. - 1984. "Contrastive Linguistics and Text Linguistics." In Jacek Fisiak, ed., Contrastive Linguistics. Prospects and Problems. Berlin etc.: Mouton Publishers. - In press. "From 'Text' to 'Interpretability'." In W. Heydrich, F. Neubauer, J. S. Petöfi and E. Sözer, eds., Connexity and Coherence. Berlin and New York: W. de Gruyter. Enkvist, Nils Erik, and Gun Leppiniemi, under preparation. "A Protocol Study of Incremental Text Comprehension: Auden's 'Gare du Midi'." Fairley, Irene, 1975. E. E. Cummings and Ungrammar. New York: Watermill Publishers. Findler, Nicholas V., ed., 1979. Associative Networks. New York etc.: Academic Press. Firbas, Jan, 1981. "Scene and Perspective." Brno Studies in English 14: 37-79. Flores d'Arcais, G. B., and R. J. Jarvella, 1983. The Process of Language Understanding. Chichester etc.: John Wiley and Sons. Givon, Talmy, 1979. On Understanding Grammar. New York etc.: Academic Press. Greenbaum, Sidney, and Randolph Quirk, 1970. Elicitation Experiments in English. London: Longman. Grice, H. P., 1975. "Logic and Conversation." In P. Cole and J. Morgan, eds., Syntax and Semantics 9: Pragmatics. New York etc.: Academic Press. Halliday, M. A. K., and Ruqaiya Hasan, 1976. Cohesion in English. London: Longmans. Hildyard, Angela, and David R. Olson, 1982. "On the Comprehension and Memory of Oral vs. Written Discourse", pp. 19-34 in Deborah Tannen, ed., Spoken and Written Language. Norwood, New Jersey: Ablex. Joos, Martin, 1972. "Semantic Axiom Number One." Language 48: 257-265. Kieras, David E., and Marcel A. Just, eds., 1984. New Methods in Reading Comprehension Research. Hillsdale, New Jersey: Lawrence Erlbaum Associates.

186

N. E.Enkvist

Kintsch, Walter, 1985. "Text Processing: A Psychological Model." In van Dijk 1985, 2: 231-243. (Concrete examples of the building of macrostructures out of text bases.) Lakoff, George, and Mark Johnson, 1980. Metaphors We Live By. Chicago and London: University of Chicago Press. Lakoff, Robin, 1971. "If s, And's and But's about Conjunction." In Charles J. Rllmore and D. Terence Langendoen, eds., Studies in Linguistic Semantics. New York: Holt, Rinehart & Winston, 115-50. Lehnert, Wendy G-, and Martin H. Ringle, eds., 1982. Strategies for Natural Language Processing. Hillsdale, New Jersey, etc.: Lawrence Erlbaum Associates. Levelt, W. J. M., 1981. "The Speaker's Linearization Problem." Philosophical Transactions of the Royal Society of London B 295:305-315. Also in Longuet-Higgins 1981: 91-100. Linell, Per, 1982. The Written Language Bias in Linguistics (- University of Linköping Studies in Communication 2). University of Linköping: Department of Communication Studies. Longuet-Higgins, H. C., et al., 1981. The Psychological Mechanisms of Language. London: The Royal Society and the British Academy. Mandl, Heinz, Nancy L. Stein and Tom Trabasso, 1984. Learning and Comprehension of Text. Hillsdale, New Jersey: Lawrence Erlbaum Associates. McHoul, A. W., 1982. Telling How Texts Talk: Essays on Reading and Ethnomethodology. London: Routledge and Kegan Paul. Meyer, Bonnie J. F., and G. Elizabeth Rice, 1982. "The Interaction of Reader Strategies and the Organization of Text." Text2: 155-192. Rosenbaum, B., and H. Sonne, 1983. "Crazy Talk: A Study of the Discourse of Schizophrenic Speakers. "Journal of Pragmatics 7: 207-217.

Ross, Haj, 1983. "Human Linguistics." Forum Linguisticum 7/3: 211-230. Sanford, Anthony J., and Simon C.Garrod, 1981. Understanding Written Language. Chichesteretc.: John Wiley and Sons. Schaar, Claes, 1975. "Vertical Context Systems." In Hakan Ringbom et al., eds., Style and Text. Studies Presented to Nils Erik Enkvist, pp. 146-157. Stockholm: Skriptor. Schallert, Diane L., 1982. "The Significance of Knowledge: A Synthesis of Research Related to Schema Theory" in Wayne Otto and Sandra White, Reading Expository Material, pp. 13-48. New York etc.: Academic Press. Schank, Roger C., "Reminding and Memory Organization: An Introduction to MOPs." Pp. 455-493 in Lehnert and Ringle 1982. Schank, Roger C., and Robert Abelson, 1977. Scripts, Plans, Coals and Understanding. Hillsdale, New Jersey: Lawrence Erlbaum Associates. Sözer, Emel, ed., 1985. Text Connexity, Text Coherence (- Papiere zur Textlinguistik, 49). Hamburg: Helmut Buske Verlag. Sperber, Dan, and Deirdre Wilson, 1986. Relevance. Communication and Cognition. Oxford: Basil Blackwell. Tannen, Deborah, 1986. That's Not What I Meant! New York: William Morrow and Company. Tirkkonen-Condit, Sonja, 1985. Argumentative Text Structure and Translation (- Studia Philologica Jyväskyläensia 18). University of Jyväskylä. Vasconcellos, Muriel, and Marjorie Leon, 1985. "SPANAM and ENGSPAN: Machine translation at the Pan American Health Organization." Computational Linguistics, 11. 2-3:122-136. [Status as of 1984: updated version available on stencil at the PAHO, Washington, D. C.] Waltz, David A., "The State of the Art in Natural-Language Understanding." Pp. 3-32 in Lehnert and Ringle 1982.

WOLFGANG HEYDRICH

Possible Worlds and Enkvist's Worlds Discussion of Nils Erik Enkvist's paper "Connexity, Interpretability, Universes of Discourse, and Text Worlds"

The development of scientific terminologies very often requires modifications of the common use of language. Within zoology it turns out to be not very convenient to call dolphins and whales fish and convenient to classify penguins as birds, although the former swim in water like sharks and salmons whereas the latter do not fly through the air like ravens and blackbirds. According to astronomy the sun is a star although it does not appear as a small, brilliant point in the sky at night, whereas its planets are not although some of them do. In logic a certain connective is known as material implication and canonically read "if-then", although it combines sentences to form a true complex sentence in cases, where, according to innocent common usage, the antecedent cannot be said to imply the consequent and where the if-then-reading should be considered as nonsensical or even false. Sometimes these deviances from common usage are only slight as in the zoological examples given above; sometimes they are more radical as in the case of the logician's "if-then" or in the case of the mathematician's "group", "family", or "lattice". Whether there is any serious risk of confusion by confounding the new reading with the old one seems to depend on how explicit the distinction has been made and how easy it is to keep it in mind. Virtually no one gets a family in the sense of set-theory mixed up with mom, dad, and the kids just as practically no one expects a mathematician who deals with Abel-groups to say something about Cain's brother. The astronomical and zoological examples are equally harmless. Interestingly, however, there is a strong inclination towards confusion in the case of the logician's "if-then", an issue I do not want to go into here. In sum: scientific terminologies are very often parasitic on pre-established or pre-entrenched uses of language and harmlessly so if the deviance has been made explicit and can easily be kept in mind. Pre-established or pre-entrenched usage does not mean common or ordinary usage in all cases. Sometimes a technical term in one discipline derives from the technical use ofthat term in another field. The term "possible worlds" as used in recent linguistics is a good example.

188

W.Heydrich

2.

The term was used in traditional philosophy at least since Leibniz and has been revitalized by those (mainly Saul Kripke) who, in the late fifties, developed model-theory for modal logic, Kripke /19637 defined model-structures for modal propositional languages to be triples , where G e K and R c Κ χ Κ. Truth-values are assigned to propositional variables relative to members of K. Recursively, this assignment 0 is extended to all formulas of the language, where the crucial clause runs: If 0 (B, H') = T for every H' such that HRH', then 0 (DB, H) = T; otherwise 0 (DB, H) = F. The elements of K were called "possible worlds". Although, obviously, this term was chosen in allusion to Leibniz' concept, it had no metaphysical import: what matters technically is, simply, that K is some arbitrary non-empty set. The terminological choice here may be compared with the mathematician's use of "group". Practically nothing hinges on the term. All that matters is that "group" is to refer to certain algebraic structures satisfying the relevant axioms. Model-theory for modal logic is not concerned with any deep analysis of the metaphysical concept of possible worlds. Those who are convinced that the concept of possible worlds is indispensable for the proper analysis of central notions (eg. modality) will not find a definite clarification of this concept in the sophistically worked out metalogical systems for modal logic. These systems simply do not say what possible worlds are and which relations obtain among them. This point has been stressed recently by one of the most enthusiastic friends of possible worlds: David Lewis. With respect to the analysis of modality, he writes: "[The] metalogical investigations seemed to cast light on the status of the controversial axioms. Maybe we didn't yet know whether the axioms were to be accepted, but at least we now knew what was at issue. Old questions could give way to new. Instead of asking the baffling question whether whatever is actual is necessarily possible, we could try asking: is the relation R symmetric? But in truth the metalogical results, just by themselves, cast no light at all. If the modal operators can be correctly interpreted as quantifiers over the indices of some or other frame, restricted by the relation of that frame, then we have found out where to look for illumination about controversial axioms. If not, not. To apply the results, you have to incur a commitment to some substantive analysis of modality." (Lewis /19867, p. 19)

One of the merits of Lewis' book is that it provides for a thorough and extensive discussion of different alternatives concerning the substantive analysis of possible worlds. This is not the place to go into details. Let me only note that Lewis contrasts his own (radical and extreme) view - called "modal realism" - with several versions of Ersatzism. According to modal realism worlds are chunks of reality, maximal individuals, spatio-temporally and causally unconnected with each other. Ontologically, all of them are on a par. No actual world is separated from merely possible ones; with actuality being essentially an indexical concept each world is the actual one according to itself. Ersatzism takes a quite different

Possible Worlds and Enkvist's Worlds

189

view of possible worlds. Here, there is one and only one actual world, actuality being not an indexical concept. Strictly speaking, no possible world is actual: They are abstract representations of the ways the world might have been or actually is. All of them - except, presumably, one - misrepresent the world, but they are supposed to do so consistently and in a complete way. Lewis grants that most versions of Ersatzism have their merits with respect to limited analyses of modality. His point is that modal realism - despite its counter-intuitiveness and disagreement with pre-theoretical common opinion - recommends itself because it provides an unrestricted overall analysis of modality. He even concedes that none of his arguments will convince a hard-line actualist who rejects any sort of quantification over possibilia - real or ersatz. Let me conclude this brief discussion of the notion of possible worlds in recent philosophy of language, logic, and metaphysics by one final remark. Although - as pointed out above - model theory of modal logic is, strictly speaking, not committed to any speafic concept of possible worlds, there seems to be some tacit ideology behind the construction of metalogical systems, a basic orientation that does not concern the letter, but rather the spirit or intuition behind it. This basic intuition is about the nature of modality, and it consists in the leading idea that possibility amounts to a pre-fabricated space of all the consistent and complete ways the world is or might have been. Even if the specific nature of these ways is left open, the space itself is construed as a set whose elements can be referred to and quantified over in the semantic metatheory. The ideology of possible worlds involves reification of possibilia and deals with the set of all possibilia in an unconstructive way. For each interpretation of a modal language, the modal space is assumed to be metatheoretically given in advance.

3. The sketch, in the preceding section, of the use of the term "possible worlds" in certain academic areas has been included here to indicate from where the notion intruded into linguistic theory. In some linguistic fields (mainly so-called model-theoretic semantics, Montague-Grammar, and related theories) the concept has been used in just the same way as in philosophical logic, metalogic for modal systems, and related areas, namely with full imprecision. The slogan seemed to be: take advantage of the suggestive term to convince people of its usefulness, but don't bother about what possible worlds really are! This strategy was not without success. Especially with respect to certain so-called non-extensional contexts which do not allow for an unrestricted principle of substitution salva veritate things went fine. The approach was impressingly successful with respect to the analysis of modal particles, certain adverbs, some sentence-embedding constructions, and tense. A partial reconstruction of the sense/reference distinction became possible, be-

190

W. Heydrich

cause strict equivalence (material equivalence under the necessity-operator) provided an interesting quasi-synonymy-relation. The success, however, had its limits. In itself and alone the account cannot discriminate semantically between logically equivalent sentences (a topic which has recently been taken up emphatically by Situation Semantics). In itself and alone it does not provide a straightforward analysis of so-called prepositional attitudes indicated by verbs like "believe", "hope", "fear", etc. (according to Cresswell [1985], some recourse to structured meanings is necessary here in addition to possible worlds). Finally the logic offictionposes special problems. I do not claim that these difficulties cannot be solved by possible-world-semantics in principle. However, it seems to me that the three problem-areas indicated above should provide enough motivation for possible-world-linguists to endow their hitherto rather empty and bloodless favourite notion with content, and to investigate the very nature of possible worlds. Besides model-theoretic semantics, there is, however, at least one additional linguistic field within which the concept of possible worlds has gained some popularity: text-linguistics and text-theory. Since Aristotle's claim that poetry is superior to history (because it deals not only with factual events but also - more generally - with non-factual ones thereby contributing to the investigation of necessity), it has often been assumed that the realm of the ways things might have been is relevant for the theory of texts, especially literary texts. As a matter of fact, Leibniz' concept of possible worlds had already been used in 17th and 18th century aesthetics and theory of literature. Wolff, Bodmer, and Breitinger - to mention only three authors1 took up Leibniz' ideas and used his concept more or less parasitically for their own purposes. Thus, it is not astonishing that the philosophical renaissance of the traditional notion of possible worlds in the second half of this century found its echo in the various fields of literary theory, and in the new area of text research. However, I am somewhat doubtful, whether this recent taking-over of a suggestive and - at first sight - attractive notion has always been accompanied by due caution and reflectiveness. Perhaps, some more or less theory-minded text-linguists and some of those who take so much pride in their practical orientation simply overestimated the extent of conceptual elaboration which the champions of philosophical logic devoted to the concept of possible worlds. Did the colleagues in the field of text-linguistics and theory of literature always realize that in the domain of its most successful application (the completeness-proofs for modal calculi) the term "possible worlds" was a label practically devoid of content? And that wherever the very nature of possible worlds was under discussion there was (and is) more quarrel and controversy among philosophers than agreement and consensus? Cf. Herrmann (1970) and Scholz (1984)

Possible Worlds and Enkvist's Worlds

191

(I am not dealing here with the question, whether some people are authorized to "patent" the term "for their own exclusive use" [Enkvist, p. 162] or whether "possible worlds" should better be "generously" offered "to all arts and sciences" [ibid.] What I'm saying here is, simply, that I have nothing against some more people trying to be more precise about what they mean by the words they adopt; there is nothing wrong with re-coining and giving new usage to a term already entrenched in a different context. But if confusion is to be avoided [I do not say that confusion is to be avoided - sometimes it is inspiring, sometimes funny, and sometimes ... - very much depends on the text-type here] - so: if confusion is to be avoided, then a deviating use should be made as explicit as possible.) My suspicion is that philosophical logic and text-theory diverge already with respect to the implicit and basic intuitions behind the supposed attractiveness of the notion of possible worlds. The modal logician's tacit assumption is - as pointed out above - that modality is rooted in a pre-existing realm of given possibilia somehow corresponding to completely and consistently specified ways the world might have been or actually is. The text-theorist, however, is not interested in fixed possibilia, pre-established and given once and for all; rather, he (or she) is concerned with linguistic modes of establishing meaningful structures, patterns of content or sense. He (or she) is interested in the ways these structures are composed by means of utterances, intelligent interpretation, co-operative text-production, and text-understanding. The linguist is interested in the variable degrees and kinds of stability these structures acquire. Some of them are very instable, ceasing to exist like soap-bubbles after a short life-time, while others endure for a long time. Some become fixed patterns to be retrieved and processed again and again; they can be used to form new structures; they become mutual 'knowledge' in the game of verbal communication, they can be presupposed and alluded to. There are patterns of content associated with casual remarks, there are stories, myths, archetypes, theories, ideologies, religions, common sense wisdom, and what not. Our intellectual and emotional cosmos abounds with more or less enduring patterns of meaning. Some are almost private, some belong to certain groups, some are the property of language-communities, some are common among cultures with a variety of languages. Some depend on the very wording of the utterances used to produce them, some can easily be rendered in different terms or languages. Some do not even depend on language and can be reproduced in various symbolic systems. (Myths e. g., originally produced by means of natural language, have often been represented by pictures.) These interests, intuitions, and tacit ideologies of text-theorists do not fit in very well, I think, with the modal logician's basic assumptions concerning the space of possibilia required for the logical analysis of modality: The modal logician needs possible worlds as something fixed to quantify over (he (or she) thinks of the modal operators as hidden quantifiers). If text-theorists, however, use the notion of worlds, they look for something flexible like content-patterns

192

W. Heydrich

that can be constructed, transformed and reorganized in a variety of ways. The error lies with those text-linguists who think that what they want is essentially the kind of thing - take or leave cryptic formalism - the modal logician has to offer. It is not. Apart from the point stressed in this section that possible worlds are conceived of as fixed and given whereas content-patterns are flexible and open to reconstruction, three additional points should be mentioned. First, possible worlds are complete in the sense that for each sentence p and each possible world w: either p is true in w or not-p is true in w. The semantic content of a text, however, is essentially incomplete. Regarding many objects mentioned in a text it is simply irrelevant whether they have certain properties or not. Think of Susie, who did not give a tip, in Enkvists's example (10). Is she married or not? This is irrelevant. By this, I do not want to suggest that according to the content-pattern established by (10) she is either way, and that we - as readers do not have enough information to decide the question. What I mean is that incompleteness is a property of the content-patterns established by (10) itself and not a property of our knowledge of this pattern. Second, possible worlds are consistent: for each set of sentences s there is a possible world w such that each element of s is true in w, if and only if s is consistent. Think of Ο. Ε. Dipus, the hero of a science-fiction story. Using his time-machine he travels back into the past, kills his father und marries his mother while she is still a virgin. According to this story, it seems to me, the victim begot his murderer and did not beget him2 - something that cannot happen in any possible world. Finally, possible worlds are closed under logical consequence: If all the elements of a given set of sentences s are true in w und p logically follows from s, then p is true in w. Presupposing the classical notion of logical consequence (which does not take the preposition in "follows logically from" very seriously3) we get as a corollary, that all logically true sentences are true in all possible worlds. However, given the semantic content of Enkvist's story about Susie, we are not inclined to admit that it involves some farfetched theorem of Principia Mathematica. Given all this, I think, the difference in orientation between the philosophical logician interested in modality and the text-theorist becomes clear. The former, when confronted with a set or sequence of sentences looks for conditions of consistency asking whether all sentences in the set might be true and hopes to explain what he is after by resorting to possible worlds (whatever they may be), Or, perhaps, I should rather say that, according to the story the victim begot his murderer and that, according to the story, the victim did not beget his murderer. The question is, whether the story is strongly inconsistent such that, according to it, a contradiction is true or only weakly inconsistent, such that two sentences are true according to it which contradict each other. For different notions of inconsistency, cf. Rescher/Brandom (1980). Cf. Anderson/Belnap (1975) p. 17f. and passim.

Possible Worlds and Enkvist's Worlds

193

whereas the text-theorist looks for conditions that form a text out of a mere set or sequence of sentences; he is interested in textuality or coherence instead of consistency. Consistency, however, is neither necessary nor sufficient for coherence. It is not sufficient because, obviously, there are incoherent sequences of semantically heterogeneous sentences, which are perfectly compatible with each other. It is not necessary, because even inconsistent stories may have coherent contents.

4. If I am right, the theoretical aims for the philosophical analysis of modality and the linguistic analysis of textuality are rather different. The tools appropriate to the former (possible worlds) are not of much help to the latter. Unfortunately, this insight is no solution for the theoretical problems of text-analysis. Rejecting an inappropriate tool does not mean having found an appropriate one. Note that the account given above of the explicational interests of the text-theorist is not meant as a solution to his problems. I just tried to formulate the basic intuitions that underly his approach. Tacit ideology is no substitute for explicit theory. Rejecting one term ("possible world") and coining another ("text world") is in itself no solution to conceptual problems. The task is to find a reliable basis for the description and explication of the establishment, modification, and reorganization or what I informally have called "meaningful structure" or "pattern of content". First and foremost, a theory is required to explain where the assumed relative stability of these structures comes from. What makes the parts and parcels of meaningful structures stick together? In one word: what makes content cohere? In his paper presented at the Stockholm conference, Enkvist proposed to explain coherence (or - as he puts it - interpretability meaning coherence relative to certain pragmatic factors) by means of possible worlds. "A text is interpretable to a receptor in a situation", he maintained, "where that receptor is capable of building around that text a universe of discourse (or text world or possible world) in which that text might be true." At that time Enkvist treated "universe of discourse", "text world", and "possible world" roughly as synonyms. In my comments I urged (apart from a list of minor points) that this explication wipes out the decisive difference between textual coherence and mere consistency. In the revised version of his contribution Enkvist has changed his position without much ado. He withdraws all reference to possible worlds and differentiates now between "text worlds" and "universes of discourse": "A text is interpretable to a certain individual receptor in a situation where that receptor is capable of constructing around that text a text world, and place that text world within a universe of discourse within which it 'makes sense' (or 'seems plausible')" (p. 171). The strategy is simple: if text worlds and universes of discourse do not do what they are supposed to do, as long as they are identified with possible

194

W. Heydrich

worlds, let us simply no longer insist that they are possible worlds. The terms are nice enough to suggest that whatever they refer to will guarantee the coherence of the text around which these marvellous constructions are built. With respect to "semantic universe of discourse", Enkvist says, it is "definable as a conceptually organized and retrievable system of models of reality", and about text worlds we learn that they are "characterized by a highly constrained, specific set of states of affairs" (p. 166), a remark which is later on referred to as a definition. Very well said! As sympathetic readers we get a feel for what is intended by the choice of the terms "universe of discourse" and "text world" (although we knew that already). As explications in a more regimented and rigid sense, however, Enkvist's remarks are not very useful. Surely, "conceptually organized and retrievable system of models of reality" and "highly constrained, specific set of states of affairs" are as much in need of explication as "universe of discourse" and "text world". Perhaps Enkvist's leading idea amounts to something like this: to each text we may, in a first step, assign a rudimentary semantic-pragmatic core. By doing so we may keep very closely to what is semantically contained in the text, explicitly and literally. This is where the emerging of a text world begins. But since the result of this procedure will normally provide a fragmentary and unconnected structure, additional steps are required to fill the gaps, to make sense of the text, and to produce coherent meaning. The missing information comes from relatively stable patterns of content or sense ("universes of discourse") which are somehow "evoked" (p. 165), "activated" (p. 166) or "established" (p. 170) by words or phrases in the text. The universes are known to the individual receptors (cf. p. 165f.) and they contain knowledge about the world (sic!) (p. 166). Universes of discourse are semantic (p. 166) in character and are "usually modelled as frames, schemata, scripts and memory-operating packages" (p. 166). "Text comprehension and interpretation" thus consists in "a highly complex, incremental process involving the interplay of bottom-up and top-down processing, as well as zigzagging between the text, the universe of discourse meaning the universe at large within which the text can be placed, and the specific world of the text with its specific, usually highly constrained states of affairs." (p. 166) So far so good. Text comprehension is highly complex: bottom-up, topdown and zigzagging, but finally it leads to a coherent text-world "explicating the specific, single, constrained projections into states of affairs that are described in the particular text" (p. 166) It does so in virtue of knowledge codifiable in frames, schemata or scripts, i.e. certain entities (texts?) whose coherence is already assumed. The crux with this explication of interpretability (coherence) is that it makes use of text worlds consisting of highly constrained collections of states of affairs and universes of discourse conceived of as appropriately constrained knowledge of the world. But assuming highly constrained collections of states of affairs and appropriately constrained knowledge of the world does not in itself explain what actually constrains them. As a matter of fact, the concept of being constrained intended here simply amounts to coherence, and we are

Possible Worlds and Enkvist's Worlds

195

back where we started: If text-worlds are coherent collections of states of affairs and universes of discourse are coherent accounts of the world, they can hardly be used for explaining the concept of coherence; instead, reference to text worlds and universes of discourse presupposes an explication of coherence. Enkvist's attempt to derive the coherence or texts from text worlds and universes of discourse shifts the problem of explaining the concept of coherence from one level to another. Instead of dealing with the problem of circumstances under which sequences of sentences may be said to cohere, we have to face the question of what makes the parts or components of text worlds cohere. If, in a next step, we are told that text worlds are coherent inasmuch as they can be placed coherently within a universe of discourse which in turn is coherent because it is usually represented by a coherent text (a frame, a script or what not), the explicational profit equals zero. 5.

Let me join here some general remarks concerning the theory of coherence. If we want to understand under which conditions a collection of things forms a coherent whole in contradistinction to an arbitrary aggregate, in other words: a relevant kind4 in contrast to a mere set, we have to realize two points. First, which things are to be coherently aggregated, which conditions of identity are to obtain for them. Second, which principles constrain the aggregating, which conditions of comprehension are in operation. With respect to Enkvist's text worlds, which are conceived of as sets of states of affairs, the first point implies that we need an account of these states of affairs. Enkvist does not say anything in relation to this point. Perhaps he thinks the question is trivial or hopelessly complicated or irrelevant for the "practically oriented student of authentic . . . discourse" (p. 162). However, with respect to this question, theoretical linguistics and logic have developed promising approaches which provide at least partial answers. States of affairs may be construed e. g. as persistent sets of situations in the sense of Barwise and Perry's situation semantics. Alternatively, and, perhaps, equivalently, states of affairs may be thought of as sets of facts in van Fraassen's (1969) sense. In accordance with the intuitions supporting these approaches we get some interesting conditions of identity and non-identity. Let /p/, /q/, /r/ be the states of affairs described by the sentences p, q, and r. Then, /p/ = /not not p/ = /p and p/ = /p or p/ /p and q/ = /q and p/ This is Goodman's (1978, p. 10) term. He avoids "natural" which "suggests some absolute categorical or psychological priority". In (1955, p. 122f.) he uses the term "genuine" as opposed to "artificial kinds".

196

W.Heydrich

/P or q/ = /q or p/ /(p and q) and r/ = /p and (q and r)/ /(p or q) or r/ = /p or (q or r)/ /p and (q or r)/ = /(p and q) or (p and r)/ /p or (q and r)/ = /(p or q) and (p or r)/ /not (p and q)/ = /not p or not q/ /not (p or q)7 = /not p and not q/ However, /not p or p/ =£ /not q or q/ /not p and p/ =£ /not q and q/ /p/ =£ /p and (not q or q)/

If, e. g., a sentence describes the state of affairs that John is a horse, it is not bound to describe the state of affairs that John is a horse and Harvey is either a rabbit or no rabbit. Consequently, representing the same state of affairs is an equivalence relation stronger than expressing the same proposition and weaker than intensional isomorphism. Taken for granted that there is a logically sound and intuitively satisfying solution for the task of introducing states of affairs, we have to approach the second task. How can we combine states of affairs to form coherent collections, which may be legitimately considered as text worlds? Simple sets of states of affairs will not do. Being arbitrary collections they are not sufficiently constrained. Imposing additional restrictions may be useful. E. g. we may demand that each collection of states of affairs must include the state of affairs /p and q/, provided it includes the states of affairs /p/ and /q/. But obviously restrictions of this kind are not sufficient for establishing coherence. It may be doubted whether there is a purely formal, semantic or truthconditional way of defining coherence. Whether some state of affairs fits into a collection coherently is not a formal question. It depends on how users of language are accustomed to combine states of affairs or types of states of affairs, which schemata of forming collections or principles of comprehension have already become established, entrenched, or sanctioned by habit. The problem of coherence, I daresay, is essentially a problem of projection. Given a collection of states of affairs we are confronted with the question of how to project it, i. e. how to integrate new states of affairs in order to maintain, increase or establish its coherence. I therefore propose to view the concept of coherence in analogy to Wittgenstein's problem or following a rule and Goodman's /1955/ problem of pathological predicates. Wittgenstein asks,* what does f^ it mean to continue a sequence of numbers, say, 2, 4, 6, 8, 10,... in a manner already indicated by the initial segment. Goodman's question is, how do we come to project a predicate, which in the past has been applied exclusively to green things, only to green things in the future, and not to grue ones. The connections between Wittgenstein's and Goodman's questions have recently been emphasized by several authors (Elgin /1983/, Putnam /1983/,

Possible Worlds and Enkvist's Worlds

197

Kripke /19827). Their relations to the problems of counter-factuals, dispositional terms, law-likeness, natural kinds, and the notion of similarity are wellknown (Goodman /1955/, Quine /19697). Far from being able to argue for a solution in this paper I want to propose that the concept of coherence should be dealt with within the context of this family of epistemological puzzles (cf. Heydrich 719847). 6.

I have no theory of coherence to offer. Nor has Enkvist. My point is that the problem of coherence cannot be reduced to the problem of modality and that possible worlds provide no means for improving our understanding of coherent structures. This seemed to be a matter of conflict between Enkvist and me as his commentator at the time of the Stockholm conference. If I understand him correctly, he has, meanwhile, implicitly conceded the point in question. Perhaps my main concern is not the same as Enkvist's. Perhaps he feels that, given the current situation in discourse analysis and text linguistics, the development of precise terminologies and formalized theories as well as the demand for explicit formulation of empirical hypotheses is premature, and we are still in a phase of heuristic reflections, tentative classification of phenomena, and deliberate pondering of terminological proposals. Accordingly, he may feel that we need not be too much concerned with terminological questions, definitions, connections between theoretical concepts etc. and may leave in the penumbra the precise nature of the relations between what he labels coherence and interpretabüity on the one hand and text worlds, universes of discourse, and states or affairs on the other. Perhaps the main concern in Enkvist's contribution "is to give a wideranging overview, with a medley of concrete examples for mainly inductive use" (p. 162). In fact, Enkvist has brought together a variety of issues currently discussed in the fields of text-linguistics and discourse analysis and he has organized them perspicuously under the perspectives indicated by the title of his paper. For this he has my admiration. With respect, however, to the task of improving our understanding of those concepts which form the core of text-theory (coherence and textuality) Enkvist's juggling with worlds is not very helpful. In my comments on his original paper, I maintained that "either Enkvist's reference to possible worlds or text worlds or universes of discourse leads to quite unacceptable results - identifying coherence (of texts) with mere consistency - or it is uninformative." By withdrawing his reference to possible worlds, Enkvist now takes measures against the unacceptable results. Accordingly, the first disjunct of my claim has to be negated. What remains, can be brought out by applying Disjunctive Syllogism.

198

W.Heydrich

References Anderson, A. R. and N. D. Belnap, Entailment. The Logic of Relevance and Necessity. Vol. I, London

1975 Barwise, J. and J. Perry, Situations and Attitudes, Cambridge/London 1983 Cresswell, M. J., Structured Meanings, Cambridge/London 1985 Elgin, C. Z., With Reference to Reference, Indianapolis/Cambridge 1983. van Fraassen, B. C., "Facts and Tautological Entailments", in: Journal of Philosophy, 66, 1969, pp. 477-87. Goodman, N., Fact, Fiction, and Forecast, (1955), Cambridge 19834. Goodman, N., Ways of Worldmaking, Hassocks, 1978 Heydrich, W., "Relevant Objects and Situations", in Conte, E., Petöfi, J. S., and Sözer, E. (eds.), Text and Discourse, Amsterdam (forthcoming). Herrmann, H.-P., Natumachahmung und Einbildungskraft. Zur Entwicklung der deutschen Poetik von 1670 bis 1740, Bad Homburg v.d.H./Berlin/Zürich 1970 Kripke, S. A., "Semantical Considerations on Modal Logic", in: Acta Philosophica Fennica 16, 1963, pp. 83-94 Kripke, S. A., Wittgenstein on Rules and Private Language, Oxford 1982. Lewis, D., On the Plurality of Worlds, Oxford/New York 1986 Putnam, H., "Foreword to the Fourth Edition", in: Goodman, N., Fact, Fiction, and Forecast, op. cit. Quine, W. V. O., Ontological Relativity and Other Essays, New York/London 1969. Rescher, N. and R. Brandom, The Logic of Inconsistency. A Study in Non-Standard Possible-World Semantics and Ontology, Oxford 1980 Scholz, O. R., "Fiktionale Welten, Mögliche Welten und Wege der Referenz", in: Finke, P. and S. J. Schmidt (eds.), Analytische Literaturwissenschaft, Braunschweig/Wiesbaden 1984

ULF TELEMAN

The World of Words - and Pictures Discussion of Nils Erik Enkvist's paper "Connexity, Interpretability, Universes of Discourse, and Text Worlds" 1. The concept "possible world", or rather the expression "possible world", is used within many scientific cultures nowadays, and its meaning is often rather vague and ambiguous. In linguistics theoretical grammarians use the term in a way which is close to the original philosophical meaning, outlined by Leibniz, I believe, and then reintroduced in modern times by Kripke (1959). The logic of possible worlds has been applied to the syntax and semantics of modal verbs, counterfactual conditional clauses, f/wi-clauses after so-called world-creating predicates (like believe, want), and to understand the tense systems of natural languages. In text analysis, i. e. the study of verbal communication, on the other hand, the expression "possible world" is used in a broader sense. Text linguists often use other words for "world", such as "universe". This tradition is ably reviewed by Enkvist (in this volume). Enkvist's paper will be the background and starting-point for the following remarks. I shall use the expression "possible world" in the broad sense which is common in text linguistic writings. Apart from the physical world that we live in and are part of, there are three kinds of worlds that have to be distinguished in the analysis of text structure and text meaning: (1) Possible worlds (Weltbild, Weltanschauung, general knowledge etc.) are generated with possible world rules, i. e. rules which define what a possible or conceivable world is according to the internalized culture of the language user. The rules are generic and most of them have the form of implications ("rules of compossibilities"): if p then q. Possible world rules include normality rules (how things must or may be) as well as norm rules (how things ought to be, what is right or wrong, good or bad). For the individual to learn possible world rules is to be socialized into a culture. This process is accompanied by the acquisition of one's mother tongue, and even if a possible world is not the same as a world that can be described in one's language it is obvious that possible world rules have much in common with lexical and grammatical rules and may be difficult to separate from them (cf. Kuhn's paper in this volume). Since in some respects linguistic rules are more powerful than possible world rules it seems reasonable to assume that language plays a crucial role when the possible world rules of an individual are changed or extended.

200

U. Teleman

(2) Real world (specific knowledge, specific experience). This is the language user's memory of specific, historical facts that he has experienced directly or indirectly. This kind of knowledge interacts with (1) since sensory experience is always interpreted by way of the possible world rules already acquired, and since specific experience might bring about restructurings of these rules. The real world of the language user has to agree with his possible world rules: if they differ, the possible world rules have to be changed or the actual experience of the real world has to be judged anomalous. The real world is then a possible world that the individual assumes to be true of the physical world he lives in. (3) Text world (discourse universe). This is the world that the text recipient is supposed to construct by means of the text and his previous knowledge (1) and (2). The text world is either factive (i.e. it is intended to be a representation of part of the speaker's real world) or it is fictive (where it has only to agree with the possible world rules). Notice the interaction of (1) and (2) with (3): the text world is created and reconstructed on the basis of (1) and (2), but when this is done and when (3) is accepted by the listener (and the speaker) as plausible, it counts as experience, and it may be integrated into the listener's real world and bring about changes in his possible world rules, if necessary. An important characteristic of possible world rules is that they can generate possible worlds embedded in each other. Just as a text world is always embedded in the real world (i. e. the highest possible world) of the person who builds it up (through the understanding of a text which he produces or interprets), a text world can be embedded in another text world and so on until the system reaches the top of the hierarchy: the world that is considered to be real. Lower worlds in the hierarchy are always parasitic upon higher worlds in which they are embedded, and especially upon the real world. Not only are the possible world rules partly generalizations of the real world, but a text world is only fragmentarily described in the text, since it is assumed that a text world is always identical to the real world if the text does not say otherwise (the principle of isomorphism). Text economy then has two sources: the possible world rules allow the interpreter to infer (and the speaker/writer knows that) what can be understood from what is explicitly said, and the principle of isomorphism makes the speaker/writer free to state only relevant distinctions between the text world and the real world (or the next higher text world). Using these concepts of worlds and their ability to be embedded in each other we can distinguish three aspects of text meaning: propositionality, mood and deixis.1 Propositionality is the text's representation of relations within the text world itself, mood is a relation from the speaker into the text world and deixis is a relation from the text world into the real world: 1

My "propositionality" and "mood" resemble Halliday's "ideational" and "interpersonal" functions (Halliday 1970).

The World of Words - and Pictures Propositionality

Mood

201

Deixis

Propositionality is the body of the text content; it is the picture the speaker/ writer has drawn, as it were, of a factive or fictive world. Mood is the communicative dimension of the text; mood connects the speaker/writer with his own text: his intentions with the picture he has drawn, his attitudes towards it, and so on. The deixis machinery like the mood transcends the text world itself but in another way: it connects elements of the prepositional picture with elements of the real world: the speaker points to real persons, institutions, places, times etc., in order to let the listener know what in the real world his text is about. Propositionality, mood and deixis are typical of verbal texts, but they are important also in other kinds of communication. Various message media may be more or less explicit, though, relying more or less on implications where the listener exploits the possible world rules to arrive at a plausible interpretation of what is the propositionality, the deixis and the mood of the text. (On inference in text interpretation, cf. Enkvist in his volume.) I shall illustrate how possible world rules have to be utilized differently in two different media: verbal texts and pictures. I shall start with mood. 2. Mood is what the originator wants the message to do for him (e. g. question, request, assert etc.), what he thinks of its truth (true, probable, false etc.) and how he evaluates the state of affairs referred to (good, bad etc.). It is obvious that verbal language makes it possible to be rather explicit about the mood of the text, whereas pictures have to be implicit. If the originator of a picture is not allowed to comment upon his picture verbally, his intention with it or his attitudes to it have to remain implicit, i. e. the viewer has to infer them via possible world rules. Verbal texts can easily distinguish between the author's questions and assertions, while pictures have to be ambiguous in this respect. (And indeed, it is very difficult to make questions with pictures; how would you draw "Is there any milk in the fridge?") An author can also specify which degree of truth or probability he assigns to a description of a state of affairs. By the remarkable instrument of negation he can even say it is false: it is as easy to say that it is not raining as it is to say that it is raining. For a picture to tell that it is not raining is next to impossible: too many things are not happening at the same time. It is also difficult for a photographer or a painter to tell the recipient - explicitly - about his feelings towards what is on his picture. It is not impossible, though, but

202

U. Teleman

where it is done, the originator of the picture has to depend on the possible world rules as a basis for the viewer's inferences. One might ask whether mood is part of the message of a picture or if it is something that the "reader" of the picture is totally responsible for. The fate of many painters and photographers indicates that pictures are taken to be assertions just like verbal texts. Pictures are forbidden and artists are censored or sent to prison because it is easy to establish what they "assert" in their pictures. The originator of verbal texts, too, has to rely on his ability to elicit the listener's inferences concerning the mood of the message. Common maxims for conversation (Grice 1975) make possible very indirect expressions: questions may function as assertions, assertions as questions, questions as requests etc., and the listener can still figure out what the speaker's intention is. In one important aspect the verbal text is always as implicit as the picture: it is seldom stated directly whether the text is meant to be fictive or factive, or where a text is fictive and where it is factive. The recipient has to infer it, comparing the text world with the real world and other possible worlds. He is normally also guided by non-verbal circumstances embedding the text: how the text is arranged, how it is ornamented, where it is printed etc. The parallelism between verbal texts and pictures is here complete.2 3. The prepositional aspect of the text is its worldcreating function: the way it guides the recipient in reconstructing a mental universe around the text. Normally, but not necessarily, this text world should be as similar as possible to the world the originator had in mind when he wrote or spoke the text. Since the text world has to agree with the possible world rules, the text originator can omit anything that can be inferred by the recipient via general and specific knowledge. Possible world rules make text production more economic: in a novel or in a newspaper article it need not be said that people each have two legs and one head until somebody turns up who does not fit the normality rules. The construction of fictive text worlds is restricted by possible world rules only (and sometimes even these can be suspended). The factive text world has to agree also with the real world of the originator. To produce fictive and factive text worlds are then quite different undertakings. To interpret such texts, however, is the same kind of operation: a text world is in both cases reconstructed by way of the text and the possible world rules. The difference lies in the use the recipient makes of his interpretations. If a factive text is accepted as such, its world is taken over and integrated with the recipient's own real world. If a fictive text is accepted as plausible, the recipient may draw some general conclusions which are added to his system of possible world rules.3 To convey the idea that the main text is fictive the author can embed it into a frame story. But remember that the fictivity of die top frame cannot be explicidy stated. It is a reasonable guess diat large parts of die possible world and real world (experience) of die speaker are stored in verbal form, i. e. the words of the language user's lexicon carry bodi the

The World of Words - and Pictures

203

Pictures are a way of creating "text" worlds, too. It goes without saying that verbal texts and pictures are both interpretations of reality, not mechanical imprints of it. The interpretation aspect of verbal texts is much more obvious, though. The very essence of verbal language is that it forces its user to typify the individuals, states and processes spoken of, and it invites its user to specify the logical relation between states of affairs by way of a more or less direct reference to the possible world rules. With pictures it is the other way round. The picture is, as it were, an example of something, its world is a token of a type, which is not explicitly specified, and the objects represented in the picture are specific members of secret sets. A verbal text says, for example, that "a boy is sitting on a chair", but the picture has to tell what the boy looks like, how he is sitting, what the chair is like etc. The verbal text is obliged to use the standard interpretations of the lexicon Weltbild or ideology to build its prepositional picture. Words are necessarily far away from the level of sensory impression which pictures can represent. Notice that such sensory impressions are always more or less ambiguous relative to the possible world rules and the classifications they make: A Volvo is a car but it is also a machine, an artifact, something which needs parking space, an expense, something you can make money on, a threat to the environment etc. The text originator can and sometimes must choose among these interpretations. The obligation to typify and to interpret along culturally determined lines is what constitutes verbal communication. It is at the same time a consequence and a prerequisite of human society. To some observers words hide reality: the essence of things is abstracted. Nietzsche is one of the thinkers who have been fascinated by this important aspect of human communication. I would like to quote the following two passages from Menschliches, Allzumenschliches (1878) and Die fröhliche Wissenschaft (1882): . . . denn allein dieses bewusste Denken geschieht in Worten, das heisst in Mittheilttngszeichen, womit sich die Herkunft des Bewusstseins selber aufdeckt.... Mein Gedanke ist, wie man sieht: dass das Bewusstsein, nicht eigentlich zur Individual-Existenz des Menschen gehört, vielmehr zu dem, was an ihm Gemeinschafts- und Heerden-Natur ist; ... dass unser Gedanke selbst fortwährend durch den Charakter des Bewusstseins - durch den in ihm gebietenden "Genius der Gattung" - gleichsam majorisirt und in die Heerden-Perspektive zurück-übersetzt wird. (Nietzsche Bd 3:592.) Die Bedeutung der Sprache für die Entwickelung der Cultur liegt darin, dass in ihr der Mensch eine eigene Welt neben die andere stellte, einen Ort, welchen er für so fest hielt, um von ihm aus

general and the specific knowledge of the individual. Such knowledge is activated when speakers or writers search for words. The words they find thus always have greater potential meaning than what the speakers or writers looked for. This has important consequences for the production and perception of texts: (1) The words found by the speaker/writer are generative. They give rise to thoughts he was not aware of before he found the words. (2) The text contains more than the speaker has put into it: it is interpreted with the recipient's word and world knowledge, not with the originator's.

204

U. Teleman

die übrige Welt aus den Angeln zu heben und sich zum Herrn derselben zu machen. ... Der Sprachbildner war nicht so bescheiden zu glauben, dass er den Dingen oben nur Bezeichnungen gebe, er drückte vielmehr, wie er wähnte, das höchste Wissen über die Dinge mit den Worten aus; in der That ist die Sprache die erste Stufe der Bemühung um die Wissenschaft. Der Glaube an die gefundene Wahrheit ist es auch hier, aus dem die mächtigsten Kraftquellen geflossen sind. Sehr nachträglich - jetzt erst - dämmert es den Menschen auf, dass sie einen ungeheuren Irrthum in ihrem Glauben an die Sprache propagirt haben. (Nietzsche Bd 2:30f.)

To read a verbal text and to "read" a picture are then quite different activities. The recipient of the verbal text has to amplify, he has to utilize his possible world rules and his real world experience to make the text world "sinnlich", i.e. to regain the concreteness which the text necessarily abstracts from. The recipient of a picture is under the opposite obligation. His job is not amplify but to interpret and reduce the information of the picture. Since a picture which has only itself as its meaning is practically meaningless, a system of implications must be found by which the recipient can construct a coherent world on an interesting level of generalization. This interpretation or reduction process is a very open and a very risky one, and the originator of the picture can control it only to a limited extent. The picture is a combination of so many ambiguities that higher level interpretations easily become infinitely many - unless they are restricted by verbal comments such as titles or captions. Illustrative examples of the difference between amplification and reduction in interpretation are the worlds we create when we read a book before we see the film and vice versa. The worlds of verbal texts can be embedded in each other and the world of the lower text can be commented upon in the next higher text. It is also possible to tell in words how various (parts of) text worlds are related to each other, i. e. causal, conditional, disjunctive etc., relations can be stated, relations which reflect more or less directly the possible world rules of the speaker/writer. With pictures the situation is different. Just as they cannot be embedded in each other, they cannot explicitly state the logical relation between various pictorially represented worlds. The reader of a picture again has to find out about causes and conditions all by himself, the originator of the picture does not spell them out for him. The risks and the advantages of both types of messages are easy to see. The reader or listener is made aware of general phenomena, the text world is structured for him. He is presented not only with a picture but also with an interpretation and he has to respond to both. The risk is that he accepts the author's view and is deterred from drawing his own conclusions. Another risk is that he may not be able to amplify the text world, which is often necessary to check the claims of a factive text against real world experience. The recipient of a picture is threatened in another way. His fascination with the pictorial surface may prevent him from undertaking the analytical work which may be needed really to understand the meaning of the picture. He sees the picture just as many viewers experience reality, seeing the surface but not the generalizations that can be based on it. (This is the point where a discussion of textual and pictorial mass media could start.)

The World of Words - and Pictures

205

However: the differences between verbal texts and pictures in the matter of their propositionality should not be exaggerated. Some verbal texts are similar to pictures, inasmuch as they report sensory data on a rather concrete level. For such texts the interpretation process is very much like the reading of pictures. Some pictures can DC very abstract in an explicit way. This is obvious where conventionalized pictorial languages are used, as in scientific graphs, but it may be the case also in other kinds of pictures. Comic strips, for instance, may be constructed as rather abstract representations of their text worlds. One interesting aspect of the relationship between the verbal text and its text world is that two or more distinct text worlds may be simultaneously constructed for the text under interpretation. Enkvist (in this volume) mentions some of the cases, e. g. the two kinds of "context". In poetical texts such parallel text worlds may struggle for dominance in the mind of the reader during the reading. For some texts like jokes or detective stories the switch between the worlds is what constitutes the text genre. Such texts create two universes, one dominant to which the reader's attention is drawn, and one less dominant, which is kept secret until the punch line, where the recipient is violently made aware of trie formerly secret interpretation, while the other interpretation is definitively discarded. 4. Finally the deixis aspect of texts. Verbal texts can not only describe or make a picture of the real world, but they can also point at it, tie some of its elements to elements of reality. For this purpose some very useful verbal tools are available: proper names, deictic pronouns (I, you, this...) and adverbs (now, soon, here...), conventionalized names of points of time etc. Via such deictic anchor points other elements of the text can also be indirectly connected with the real world. By way of "Γ we can also designate uniquely "my brother" or "the house where I was born". Deictically fixed times like the birth of Christ in combination with measures of time duration can point to any particular point of time in history. Deixis is indispensable in factive texts. Schizophrenic patients have problems with such texts, because their deixis machinery is out of order. Their texts may look perfectly natural from other angles, but they are not anchored in reality. A schizophrenic text is in this respect similar to a picture, but the reason for the lack of deixis is different: in pictures it is a characteristic of the medium itself. (One might expect that schizophrenic pictures would be quite regular, and some would say they are.) Pictures cannot point to the real world any more than they can specify the mood of the message in an explicit way. Pictures cannot designate times, places, individuals, institutions etc. Such referents can be represented visually sometimes, and the recipient might figure out who or what is meant if he knows what they look like (Hitler, Napoleon etc.). Normally, however, this method is not available. (How does an artist uniquely designate EEC, Berlin, the ninth century, if he is not allowed to use verbal language?)

206

U. Teleman

Some of the deictic instruments of verbal texts can be utilized by the author to point not only at elements of the real world but also at elements of the world that is being built around the text itself. By "this chair" he can point at a specific chair in the nonlinguistic context of the utterance, but he can also refer to a chair mentioned previously in the text, i. e. a chair which is located in the text world of the text already written. (Historically anaphoric devices such as the, this, he seem to have developed out of deictic expressions.) Compare again pictorial texts like comic strips with verbal texts. The comic strip - if we forget about the verbal balloons - lacks anaphoric elements. The artist cannot point at specific persons, he has to draw them fully specified again and again, whereas the corresponding verbal text needs to mention their characteristics only now and then, since it is enough to point at them with he or she or it or with their proper names. Place descriptions, too, have to be repeated (to a certain extent) in the succession of comic squares; the verbal text has to mention the scenery only when it changes. As for time, the comic strip can use partly the same technique as the narrative verbal text: the time flow is presupposed iconically. But if there is a jump in the temporal sequence of actions (or if an action told later happened earlier), it has to be stated in words, in the comic strip, too. 5. The words of a verbal text constitute a one-dimensional track which the interpreter has to follow if he wants to (re-)constructthe world of the text: verbal messages are linearized. What the interpreter meets first determines his expectations as to what comes next, determines his understanding of it. In contrast to this, pictures are two-dimensional and the viewer is given a great deal of freedom to organize his interpretation process temporally. Where the originator of the verbal text chooses for his interpreter whether background should precede foreground in the world-building process, the originator of the picture cannot restrict the interpreter in this respect: the viewer is free to begin where he wants to. Sentences are verbal mini-texts and Enkvist (in this volume) has discussed various linearization strategies determining the interpretation process of the listener/reader. He suggests that the strategy to put background (old) information first (OEF) in the sentence is characteristic of written language, while it should be more typical of spoken language to start with foreground (crucial) information (GIF) in the sentence. The OIF strategy is well-known: it has been studied empirically and it is easy to explain as an effect of general conditions on perception and production. So far we have only anecdotal evidence for the GIF-strategy, and I doubt that it is a spoken language feature in general. In Swedish it is my impression that GIF is avoided in spoken language. The normal stress pattern in Swedish is that heavy stress, indicating crucial information, is found towards the end of the sentence. In spoken language semantically empty expressions are often used to fill the grammatical slots at the beginning of the sentence. In a large corpus of spoken and written Swedish texts the fre-

The World of Words - and Pictures

207

quency of such an empty morpheme det ('it', 'there') in spoken texts is nearly three times as high as in written texts. It may well be that different languages, even if they are structurally similar, invite different constituent order strategies. In Swedish two different Jet-constructions can be used to avoid GIF-order, where this is the normal English alternative. If the question is "Who took my car?" the Englishman answers: "John did." In Swedish the answer begins with det: Det gjorde John. (Lit.: That did John.) or

Det var John (som gjorde det). (Lit.: It was John (who did it).) (Of course it is possible in Swedish as well as in English to answer only "John", i.e. with cruciaf information only. This is typically the case in dialogues where the background information is presented in identical form by the previous speaker. In cases like these it seems not very interesting to distinguish between GIF and OIF.) Theoretically the viewer of a picture is free to start his decoding where he chooses to. Nervertheless it is a well established fact that he directs his gaze first to the salient material of the picture, i. e. to the mental foreground of the message (e.g. Bus well 1935). I would think that written language texts are more like pictures than spoken messages in this respect. Both are normally monologues and both can be read and reread until the interpreter has reached a satisfactory level of understanding, i.e. has built up a text world that seems reasonable around the text. The interpretation of auditory texts depends much more on the temporal organization chosen by the speaker. Functional considerations favour the idea that OIF is more effective in spoken language than in written language. My little exploratory investigation has come to an end. I have tried to show how important aspects of text meaning can be derived from the concept of possible worlds and possible world rules. It turned out that pictures are necessarily implicit concerning the relations between the text world itself and the real world where it is embedded (deixis, mood). The interpretation of verbal and pictorial texts differs in the use of possible world rules for inference: in the first case the interpreter has to amplify, in the second to reduce. In the linear process of interpretation I have suggested some similarity between the two kinds of visual messages as against spoken language texts.

References Buswell, G. T. 1935. Haw people look at pictures. Chicago. Einarsson, Jan. 1978. Talad och skriven svenska. Lund: Ekstrand. Grice, H. P. 1975. Logic and Conversation. In: P. Cole &J. Morgan (eds.), Syntax and Semantics 3: Speech Acts. New York: Academic Press. 41-58.

208

U. Teleman

Halliday, M. A. K. 1970. Language Structure and Language Function. In: J. Lyons (ed.), New Horizons in Linguistics. Penguin. 140-165. Kripke, Saul. 1959. A completeness theorem in modal logic. In: Journal of SymbolicLogic24,1-14. Nietzsche, Friedrich. 1980. Sämtliche Werke. Kritische Studienausgabe. Hrsg. von G. Colli und M. Montinari. Berlin: de Gruyter.

JÄNOS S. PETÖFI

Possible Worlds - Text Worlds Quo vadis Linguistica? Report on Session 2: Linguistics 1. In the second half of this century linguistics has become increasingly an interdisciplinary branch of humanities. This development, however, is not limited to linguistics alone; the changes which its object domain and its methods have undergone have also had an impact on all those disciplines whose methods and results have become a part of its strategy. Figure 1 (p. 210) represents those disciplines among which this mutual effect has become most evident. To begin with I want to make some comments on these disciplines and their mutual interrelations. Whatever one's personal attitude towards the Chomskyan generative transformational theory of language (which became known to a larger linguistic public in 1957) may be, one has to admit that it initiated a methodological discussion which has had a considerable influence on the further development not only of linguistics but also of a number of other disciplines. The basic component of generative transformational grammar, syntax, has been elaborated in analogy with an algebraic (formal logic, automaton-theoretical) model. This basic component has been complemented by two interpretative components: the phonological and the sense-semantic components. From these two components the latter has evoked intensive methodological discussions. The controversy between referential semantics vs. sense-semantics (extensional semantics vs. intensional semantics) has finally been solved by Montague, who constructed a so-called universal grammar which has both an intensional and an extensional semantic component. The underlying model has again been algebraic. the three components (the syntactic and the two semantic components) of Montague's grammar are homomorphous algebras. It should be mentioned that Montague's universal grammar also integrates formal pragmatic entities - as a methodological answer to the increasingly intensive pragmatic discussion. Speech act theory, which belongs both to the object domain of linguistics and to that of philosophy, arose as an attempt at correcting Chomsky's conception, however with a different aim than that of Montague. It has had a considerable influence on the further development of linguistics. Linguistic-dominated text research started in the sixties, parallel with the establishment of sentence (or utterance-of-sentence-size) centered theories. The

210

J. S.Petöfi

Sociology Ethnomethodology

Artificial Intelligence Research

Figure 1

roots of text research are mainly in Europe, even if text investigation has also had its devotees in the States, primarily within the school of Pike. At the beginning of text-linguistic research the relation between text linguistics and generative linguistics was bidirectional. While schools of text linguistics have tried to apply generative grammar as a model, generative linguistics has attempted to offer a solution to the problems shown as genuine tasks for text linguistics within the generative framework. (A similar situation occurred later between extensional semantic theories and speech act theory on one hand, and text research on the other hand.) It has to be emphasized, in addition, that text research also provided a new impetus for the analysis of poetic, stylistic and rhetorical questions, thus reactivating and reinforcing the interrelation between poetics, stylistics, rhetoric, and (text) linguistics. Regarding the psychological presupposition of generative linguistics, psychologists interested in the functioning of language found motivation for intensifying cognitive psychological research concerning verbal communication parallel with the development of generative language theoretical research.

Possible Worlds - Text Worlds

211

Linguistic investigation of speech-act theoretical (and, in general, pragmatic) questions found its parallel research branch in sociological/ethnometkodological research directed towards language and language communication. The interrelation between philosophy and linguistics has many layers: (i) it is unthinkable to discuss methodological questions of linguistics without taking into account the general principles of the philosophy of science; the philosophy of science, on the other hand, has to reflect the methodological questions arising within the framework of linguistics; (ii) the philosophy of language is substantially a philosophical discipline directed towards language; (iu) in fact, most of the philosophical logics can be considered as logics of different language aspects (language expressions); finally (iv) the investigation of meaning (first of all text meaning), especially by the recognition that meaning is not a research object limited to linguistics only, increasingly strengthens the interrelation between the philosophy of interpretation and (text) linguistics. The relation between artificial intelligence research and linguistics must also be mentioned here. In the sixties, the central interest of computational linguistics was directed towards questions of machine translation; in the seventies this central place has been taken over by the simulation of text understanding. This development has entailed a close relationship between artificial intelligence research, (text) linguistics and also cognitive psychology. My above remarks were mainly concerned with the (inter)relationship(s) between linguistics and other disciplines, I did not enter into showing the interrelationships between the other disciplines mentioned. In conclusion I mention some basic points: (i) in all disciplines presented in Figure 1 a central interest in the sentence (or utterance of the size of a sentence) arose first which has gradually been extended to include texts; (ii) by this extension of interest it became plausible within all disciplines that aspects of text - especially text meaning and text understanding - cannot be investigated adequately by merely using the apparatus and methodology of any of the disciplines alone. This outline is intended to serve as a multi- and interdisciplinary framework for placing the topics presented in the section Linguistics of this Symposium. 2. In the previous section I outlined the inter- and multidisciplinary context of sentence and text research. Here I want to elucidate some basic questions of research into meaning (text meaning) by means of some examples. Let us consider the following text-parts: (1) Twas brillig, and the slithy toves Did gyre and gimble in die wabe: All mimsy were die borogoves, And die mome radis outgrabe. (Lewis Carroll: "Through die Looking-Glass", Ch. I.) (2) 'Come, diere's no use in crying like diat!' said Alice to herself rather sharply. Ί advise you to leave off this minute!' (Lewis Carroll: "Alice's Adventures in Wonderland", Ch. I.)

212

J. S. Petöfi

(3) The Fish-Footman began by producing from under his arm a great letter, nearly as large as himself, and this he handed over to the other, saying, in a solemn tone, 'For the Duchess. An invitation from the Queen to play croquet.' The Frog-Footman repeated, in the same solemn tone, only changing the order of the words a little, 'From the Queen. An invitation for the Duchess to play croquet.' (Lewis Carroll: "Alice's Adventures in Wonderland", Ch. VI.) (4) 'How are you getting on?' said the Cat, as soon as there was mouth enough for it to speak with. Alice waited till the eyes appeared, and then nodded. 'It's no use speaking to it,' she thought, 'till its ears have come, or at least one of them.' In another minute the whole head appeared, and then Alice put down her flamingo, and began an account of the game, feeling very glad she had some one to listen to her. The Cat seemed to think that there was enough of it now in sight, and no more of it appeared. (Lewis Carroll: "Alice's Adventures in Wonderland", Ch. Vffl.)

When reading (1) (the first verse of the poem "Jabberwocky") we may well have the same impression as Alice had: "'Somehow it seems to fill my head with ideas - only I don't exactly know what they are! (...)'". - "To a large extent," comments Hockett on Alice's statement - "Alice's unrecognizable "ideas" were just the abstract framework of relationships (...). This framework itself asserts nothing about anything, but is familiar to Alice, and to us, because as we speak English we constantly use bits of it in utterances which do purport to deal with the world around us." (Hockett 1964:263). In other words we might as well say that the syntactic organization of (1) is well-formed, however, we cannot assign a sense to (1), because we do not know the sense of most of the words constituting it. Alice also understands the sense of this verse when Humpty Dumpty explicates the words unknown to her. These Explications concerning the two first lines are the following: "brillig"means four o'clock in the afternoon - the time when you begin broiling things for dinner; "slithy"means "lithe and slimy"; "lithe" is the same as "active" (...) it's like a portmanteau - there are two meanings packed up into one word; "toves "are something like badgers - they're something like lizards - and they're something like corkscrews; (...) also they make their nests under sun-dials also they live on cheese; to "gyre"is to go round and round like a gyroscope; to "gimhle"is to make holes like a gimblet; Alice also took part in the explication of "wabe" guessing its basic component. This explication is the following: "the wabe"is the grass-plot round a sun-dial; it's called "wabe" because it goes a long way before it, and a long way behind it - and a long way beyond it on each side. Taking these explications into consideration, Alice could assign the following sense to the first two lines of (1):

Possible Worlds - Text Worlds

213

it was four o'clock in the afternoon - the time when one begins broiling things for dinner, and the active and slimy creatures which are something like badgers, lizards and corkscrews, and make their nests under sun-dials and live on cheese, did go round and round like a gyroscope and did make holes like a gimblet in the grass-plot round a sun-dial. It should be mentioned that had Alice read this verse in a family magazine and had she used the explications found there when constructing the sense of the above lines, she would have obtained a different sense, even if all words except "wabe" had been explicated there in the same way as Humpty Dumpty had done, namely "wabe" has been explicated there as follows: "(derived from the verb to swab or soak} the side of a hill (from its being soaked by the rain)." There are linguists who think that the meaning of the first two lines of (1) is nothing else but the above constructed sensus. Others, however, believe that the disclosure/construction of meaning (also) requires the formulation of the conditions which a world-fragment has to meet in order that the expression in question can be true. The investigation of the nature and constitution of meaning implies many questions. Let us consider some of them: (i) What is the meaning-constitutive role of the language elements belonging to different parts of speech (and/or to their subclasses)? - Let us think here, for example, of the different meaning-constitutive role of the expressions like "Lewis Carroll", "Alice", "the slithy toves", "the wabe",...; "gyre", "gymble",...; "all", "the",... (ii) How can simple meaning-carrying elements be combined to form compound/complex meaning-carrying elements? - Let us think here, for example, of the "ancr-s and of the colon in (1); of the quotation marks, the verb "said", and the fact that "Alice", "herself", T' and "you" refer to one and the same person (and this person is also involved in the expressions "come", "crying", and "leave off') as the carriers of connectedness in (2). (iii) When can two statements be said to have the same meaning? - Can the following statements in (3), for instance, be said to have the same meaning?: "For the Duchess. An invitation from the Queen to play croquet" and "From the Queen. An invitation for the Duchess to play croquet". (iv) Is it possible to assign a meaning to all expressions of language of the size of a sentence or a text? - Has, for example, (4) a sense or is it non-sensical? Even intuitively obvious questions are formulated in semantic research as follows: (i) What is meaning? What is expedient (and/or necessary) to consider as meaning?, (ii) What are the meaning-determining factors?, (iii) What are the theoretical methods of meaning-research?, (iv) What meaning-theories can be constructed?, (v) How is the meaning of a language expression to be constructed within a meaning-theory?, (vi) Does the complexity of a language expression have any influence on how its meaning is to be defined and how to assign a

214

J.S.Petöfi

meaning to it? The reports presented in the Session 'Linguistics* of the Symposium are directly or indirectly connected with these questions. 3. In the Linguistics Session of the Symposium two reports have been discussed: the sentence-centered report of Barbara Partee on "Possible Worlds in ModelTheoretic Semantics. A Linguistic Perspective", and the discourse-centered report of Nils Erik Enkvist on "Connexity, Interpretability, Universes of Discourse, and Text Worlds". 3.1 In her report, Barbara Partee provided a comprehensive survey of both the model-theoretical and the non-model-theoretical trends in modern semantic research, illustrating their main characteristics by means of a number of examples. The so-called possible-worlds semantics is one of the main trends in model-theoretical semantics. To the non-possible-worlds theories of modeltheoretical semantics belongs, among others, the so-called situation semantics initiated and constructed by Barwise and Perry. Partee also ranks the so-called representationalists among the representatives of non-model-theoretical semantics and those who consider the construction of semantic networks as the main task of semantic research. The way in which Partee presented these directions is clear and impressive, pointing out all of the positive and negative aspects of these trends (and of their variants). She of course treated, in accordance with her own task and main interest, the possible-worlds semantics in detail. Within this direction she stands for the opinion that a linguist does not have to be concerned with the metaphysical/ontological problems of possible worlds (i.e. with questions like what a possible world is at all); for her, a possible world is a theoretical construct which may turn out to be an adequate means for handling many of the basic problems of meaning. To formulate it in another way: if this function is regarded as the basic one, this construct might as well be called anything else; as far as the understanding of what is the essence of possible-world semantics, any association suggested by the term 'possible world' in a natural way may be misleading. Among the discussants, Perry remains within the framework of model-theoretical semantics and approaches some for him relevant aspects of possible-worlds semantics from this angle. His paper is in contrast to Rieser, whose approach as an outsider to possible worlds semantics (with main interest in conversational analysis, procedural lingustics, and the non-Fregean tradition of logic) resulted in different methodological questions he addressed to Partee. Partee's reply involved, in accordance with her interest in some of the aspects, going into detail on some of the questions and also leaving some of them out of consideration or explicitly rejecting them. To elucidate Partee's standpoint, let me quote one of her statements of the character of a 'self-confession': "as a linguist I am an instrumentalist about possible worlds: as a linguist I don't care a bit what they are like or what they are made of beyond the properties that play a role in my semantic theories. And I don't believe I need to or should care. (But

Possible Worlds - Text Worlds

215

I'm also a philosopher of language, and as a philosopher of language I do care. But that doesn't slow down my linguistics.)" 3.2 From Nils Erik Enkvist's report we obtain a comprehensive picture of the various aspects of discourse/text research. What he has been mainly concerned with, is not a presentation of the individual schools/trends; he mainly stressed the main questions arising necessarily when discourses/texts are being analysed. These main questions are the following: what factors guarantee intersentential connectedness, what factors play a role in the interpretation of a discourse/text, and what strategy may an interpreter pursue in the course of interpretation in general and what are the main problems arising in the course of interpretation. Enkvist has not dealt with the problems of possible-world semantics. Although he used the term "possible world" in the version of his report presented at the Symposium, this was not an indicator of a formal construct as possibleworld semanticians - also Partee - use this term. He uses the term 'possible world' to refer to the state-of-affairs configuration which is constructed (reconstructed) by the interpreter when interpreting a text. In the final version of his report published here, when he treats the strategy of interpretation, he pleads that "interpretability depends on the receptor's ability of building a specific world around the text, using a universe of discourse as an aid" (my italics - J. S. P.). In this sense this (possible) world - in the framework of a (possible) interpretation process - stands, of course, nearer to the intuitive notion of world than the 'possible world' notion of the instrumentalists, however, the theoretical conceptions concerning (possible) worlds conceived in this manner can at best be considered as 'preliminary' at present. Among the discussants, Teleman tried to extend some of Enkvist's notions to pictorial communication. His arguments are not only relevant with respect to this communication type, they are also significant for interpreting the verbal semantics, if we interpret the semantics of the verbal language not only in its narrow sense. In his discussion, following Enkvist's report Heydrich formulated some basic methodological questions from the angle of text theory and formal semantics, seeking the possibility of a dialogue between possible-world semantics and text theory. 3.3 I wish to make three remarks in connection with the two reports at the Symposium. Both reports touched upon the connection between semantics and pragmatics. In this connection two notions are of crucial importance: that of speech acts and that of context/contextualization. The two reports have greatly contributed to the understanding/application (instrumentalization) of these two notions, - but also to understanding and clarifying the (inter)relationship(s) between semantics and pragmatics, even if the underlying conceptions are different both as to their research object and their methods. Both reports emphasize the importance of the interdisciplinary context (philosophy, cognitive psychology, artificial intelligence research) which neces-

216

J. S.Petöfi

sarily has to be taken into account if the linguistic problems concerned are to find an adequate solution. Although there were no representatives of cognitive psychology and artificial intelligence research among the participants in the Symposium, I hope the discussion pursued here will find its way to them, and have an echo. One remark should also be made concerning the connection between sentence-centered and discourse-centered linguistic research. Partee even explicitly refers to the fact that "it is fair to say that within the formal semantic tradition, possible worlds have been partly responsible for the shift of attention from the interpretation of sentences in isolation to the interpretation of discourse context." Even if the expression "interpretation of discourses in context" is not used in the same way as text linguists or text theoreticians use this expression, it indicates the pragmatization of formal semantics in the sense without which problems of texts could not be handled. The results of formal semantics are not indifferent to text linguists - this is as clear to Enkvist as to many other researchers dealing with text problems. The basic problem is, however, whether it is possible, and if so, how to bring the two research directions closer to one another. 4. In the last point of my summarizing report I wish to make some comments on the picture we gave of linguistic research, especially about the investigation of the aspects of meaning, on the basis of the reports and the discussions in this Symposium. 4.1 First of all I want to emphasize that when we investigate the aspects, perspectives, and development of linguistic research, we must be aware of the fact that the object of linguistics is not given in the way it is given, for instance, in the case of, say, chemical research. Linguists are themselves the ones, who define/choose the object of their research - in many cases depending on the goal to be reached or the method to be applied. One may say that there are as many linguistics (linguistic directions/schools) as it is possible to define configurations of object+goal+metriod for linguistic research. A further fact complicating this basic situation is that in the three-tuple object+goal+method different dominance relations may exist. If we understand the above basic fact about linguistics, the main point of view when judging the different linguistic research directions is whether they are as to the configuration object+goal+method they chose coherent and consistent. This is the only justifiable starting point for criticism. This way of applying criticism, however, does not dispute the justification of the question whether the different three-tuples object+goal+method may be evaluated on the basis of some other external criterion. Such external criteria are not only imaginable, they are existent, as is proved by the development of formal linguistics (including model-theoretical semantics): How could we understand, for example, the pragmatization of (formal) semantics, if not under the condition of assuming certain external criteria.

Possible Worlds - Text Worlds

217

In connection with linguistic research, the determining external factor is the static and dynamic character of natural language organization. This factor is being taken into consideration by most of the directions/schools as seriously as possible. The border lines of the potential for taking into consideration these possibilities are in general determined by the methods applied by these directions/ schools. 4.2 The picture of semantic research presented in the papers of the Linguistics Session is an asymmetric one: Partee discussed in her report aspects of a sentence/utterance-centered semantics dominated by the method (aspects of model-theoretical semantics, and within this those of possible-worlds-semantics). Enkvist, on the other hand, focussed his attention on the aspects of linguistic research dominated by the object (= text/discourse). In order to make this picture complete, it would be necessary to complement Partee's report by a sentence/utterance-centered research report in which the aspects of the object are in the foreground, and Enkvist's report ought to be complemented by a report discussing text/discourse research mainly concerned with the methods relatable to the methods of possible-worlds semantics. Only such a balanced presentation would guarantee that questions like the following could be dealt with objectively: which method appears to be more effective for sentence/ utterance-centered and which for text/discourse-centered semantic research, and in what way could these research directions approach one another - if the methods appearing to be effective render such an approach possible at all. Since these complementary reports are missing, I will confine myself in what follows to the fact already emphasized previously: that pragmatization (and interdisciplinarity which is partly dependent on it) is a dominant characteristic both within sentence-centered and text-centered research. 4.3 If we accept the claim to pragmatize as a fact, the question which follows next is whether this pragmatization has objectively establishable limits. This question can DC formulated from the angle of the object as follows: Can the/«//static and dynamic organization of natural-language communication be the object of linguistic research? If the answer is negative, the question to be clarified is, what kind of division of labor can/should be defined among the disciplines represented in Figure 1 so that interdisciplinary co-operation remains guaranteed? From the aspect of the goal, the question is the following: Can the full stone and dynamic organization of natural language communication be described within one single integrative framework? If the answer is affirmative, the next question is whether this integrative framework can be considered as a linguistic framework, and should it turn out that the answer must be negative, which discipline may claim to offer such an integrative, discipline-specific framework? In the case of a negative answer the question is what kind of theoretical modules must be constructed for such a full description, which disciplines have the task of

218

J.S.Petöfi

constructing the individual modules, and how can the co-operation of these modules be guaranteed? Finally, from the angle of the method, the question is what kind of criteria formulated in terms of philosophy of science can (must) a theory fulfill which aims at describing the full static and dynamic organization of natural-language communication within one single integrative theoretical framework? With regard to the topic discussed in the Linguistics Session of the Symposium, this question may be translated into the following concrete question: Is it imaginable to extend the application of the methods of possible-worlds semantics to such an extent that the aspects of the meaning of texts of any lengths can also be accounted for, so that it corresponds to our intuition concerning the meaning of the text in question? Should this, for whatever reason, not be imaginable, what method with what degree of exactness can/must be applied? 4.4 When formulating the above questions, I started out from the assumption that it is possible to describe the full static and dynamic organization of naturallanguage communication within one single integrative framework. To bolster the plausibility of this assumption let me only refer to the fact that we are capable of communicating by producing and receiving texts, i.e. we are capable of applying all pieces of knowledge necessary for fulfilling this function in an integrative way. The time has not yet come to be able to answer the above formulated questions. However, I am convinced that research into natural language (in the interdisciplinary way becoming increasingly evident) will make it inevitable to deal intensively with and find answers to these questions. The discussion pursued in the Linguistics Session of this Symposium will certainly have to be considered as an important milestone on tnis road. Bibliography

Carroll, Lewis. 1982. The Complete Illustrated Works of Lewis Carroll, London: Chancellor Press. Chomsky, Noam. 1957. Syntactic Structures, The Hague - Paris - New York: Mouton Publishers. Hockett, Charles F. 1964. A Course in Modem Linguistics, New York: The Macmillan Company. Montague, Richard. 1974. Formal Philosophy. Selected Papersoi Richard Montague. Edited and with an Introduction by Richmond H. Thomason, New Haven - London: Yale University Press. Petöfi, Janos S. 1986a. "Text, Discourse", in: Thomas A. Sebeok (ed.), Encyclopedic Dictionary of Semiotics, Berlin - New York - Amsterdam: Mouton, de Gruyter, pp. 1080-1087. - 1986b. "Report: European Research in Semiotic Textology. A historical, thematic, and bibliographical guide", Folia Linguistica, XX, pp. 545-571. - 1988. "Language as a Written Medium: Text", in: Neville E. Collinge (ed.), Encyclopaedia of Language, Beckenham: Croom Helm Ltd Publishers (in press). Pike, Kenneth L. 1967. Language in Relation to a Unified Theory of the Structure of Human Behavior, The Hague: Mouton. Searle, John R. 1969. Speech Acts. An Essay in the Philosophy of Language, London: Cambridge University Press.

Session 3: Literature and Arts

LUBOMiR DOLE2EL

Possible Worlds and Literary Fictions During the 20th century, the theoretical discussion of the problem of fictionality has been conducted primarily in philosophy and logic (formal semantics). Appropriately, two issues have dominated the debate: a) What is the ontologjcal status of nonexistent fictional objects and individuals (golden mountain, wonderland; the present king of France, Odysseus), b) What is the logical status of fictional representations, i. e., of expressions designating nonexistent entities (the problem of fictional reference) and of sentences (statements) about such entities (the problem of truth-valuation of fictional sentences). The inconclusiveness of the philosophical debate is due only partly to the complexity of these issues. Its main reason is the speculative and atomistic method characteristic of much of contemporary analytic philosophy. Philosophers of fictionality have, as a rule, ignored the empirical theories of anthropologists, literary theorists, art historians, etc. Focusing on isolated, constantly repeated and contrived specimens, they have practiced a dated, prestructural (presystemic) method. With a few notable exceptions, they have not perceived fiction-making as a cultural activity; modern theories of culture, such as cultural semiotics, have been ignored. On the other hand, literary and art theorists dealing with the problem or fictionality have almost totally neglected the solutions proposed by logicians and analytic philosophers.1 They have not felt the need to cultivate this contact because they are, as a rule, satisfied with a relatively low level of rigour and explicitness. This symposium gives us an opportunity to start an interdisciplinary exchange around the problem of fictionality. This long overdue initiative will be, hopefully, extended to all concerns common to philosophical aesthetics and to empirical art theory and poetics. There is no doubt that such exchange would be most beneficial to both sides by challenging our parochial ways of thinking and our stereotyped habits of formulating and solving problems. With the view of contributing to this exchange, I venture into a high-risk enterprise: to bring together in comparison and confrontation what has been developed in isolation by philosophers, linguists and literary theoreticians - the diverse approaches to 1

In a different philosophical tradition, in phenomenology, such an exchange has been initiated, as demonstrated especially by Ricoeur's work (cf. Ricoeur, 1983; 1985). Undoubtedly, the popularity of phenomenology in modern art theory and literary criticism is due, among other things, to the active participation of philosophers in the dialogue with dieoreticians of art and literature.

222

L. Dolezel

the problem of fictionality. This confrontation will lead us not only to a reassessment of the extant solutions, but also to the development of a new approach to the theory of fictionality. /. One-world model frame The most popular philosophical and literary critical solutions of the problem of fictionality are based on the assumption that the sole legitimate mode of existence is actual existence; the actual (real) world is the one and only universe of discourse. These semantics of fictionality can be said to be located within a one-world model frame. In philosophical logic, two well-known one-world semantics of fictionality were proposed, the Fregean and the Russellian. Within their common assumption, Russell's semantics seems to me more radical than Frege's; therefore, in presenting the two philosophical versions of the one-world semantics of fictionality, I will reverse tneir chronological order. 1. Russell: empty terms. Russell's view is to be placed at the extreme pole of the gamut of one-world theories of fictionality because he was absolutely consistent in his "robust" realism: 'There is only one world, the "real" world... It is the very essence of fiction that only the thoughts, feelings, etc. in Shakespeare and his readers are real, and that there is not, in addition to them, an objective Hamlet' (Russell, 1919: 169).2 Russell recognized that by accepting the actual world as the sole universe of existents, we are committed to the claim that fictional entities do not exist, fictional terms lack reference (are 'empty') and fictional sentences are false. In Russell's semantics fictional entities are part and parcel of an undifferentiated set of nonexistent objects, lumped together with impossible objects (square circle), empty mathematical terms (the even prime other than 2), erroneous beliefs (the present king of France), discarded scientific concepts (phlogiston) etc. Therefore, fictional entities have to share the ontological and logical fate of the whole set. The core of Russell's semantics of nonexistent entities is his well-known theory of descriptions. We do not have to rehearse the details of Russell's analysis which - he believed - eliminates empty terms from logical representations. What is important for our concern, are the consequences of Russell's semantics for the treatment of fictional entities. Within this semantics, both Emma Bovary committed suicide and Emma Bovary died of tuberculosis have one and the same truth-value; both are false sentences. Russellian semantics does not allow us to decide which properties belong and which do not belong to fictional entities, what kind of actions the fictional agents are or are not engaged in, what

2

It is a matter of common knowledge that Russell formulated his rigid semantics in a polemic with Meinong, refuting his own earlier views.

Possible Worlds and Literary Fictions

223

does and what does not constitute the setting of their activity, etc.3 At one point, Russell perceived the implausible consequences of his purely negative interpretation. He admitted that different fictional terms, such as unicorn and sea-serpent, while having the same ('zero') reference, differ in meaning (signification). He explained this difference by accepting a fictional universal ('concept') into his semantics: '"I met a unicorn" or "I met a sea-serpent" is a perfectly significant assertion, if we know what it would be to be a unicorn or a sea-serpent, i.e., what is the definition of these fabulous monsters' (ib.: 168).4 Russell does not explain where the knowledge of fictional concepts is derived from. However, by acknowledging that fictional terms, while lacking reference, do not lack signification, he admitted unwittingly the limits of his rigid system. In fact, he switched to a softer version of the one-world semantics of fictionality, the Fregean alternative. 2. Frege: pure sense. Frege's semantic treatment of fictions rests on his wellknown differentiation of two dimensions of meaning, reference ('Bedeutung') and sense ('Sinn') (Frege, 1892). While denying existence to fictional entities and reference to fictional terms, Frege is able to stipulate that fictional terms carry sense. At the same time, his semantics allowed to formulate softer truthconditions for fictional sentences: lacking reference, they lack truth-value, i.e., they are neither true, nor false.5 In such a way, fictional terms and sentences despite their lack of reference, are made meaningful; their meaning is equivalent to, and exhausted by, their sense. Frege postulates the existence of a pure-sense language and identifies this language with poetry ('Dichtung'). Poetic language has to be liberated from the constraints of referentiality and truth-valuation, since its aim is to pursue 'aesthetic delight' rather than knowledge. The radical split between cognitive (referential) and poetic (pure-sense) language leads us to designate Frege's semantics as a one-world two-language-model.6

3

4

5

6

Grossmann takes this failure to be the 'essence' of the system: Ά nonexistent golden mountain' 'neither has the property of being golden nor has the property of being not golden' (Grossmann, 1983: 410). The argument is based on Russell's analysis of the indefinite description: 'Since it is significant (though false) to say "I met a unicorn", it is clear that this proposition, rightly analyzed, does not contain a constituent "a unicorn", though it does contain the concept "unicorn"' (Russell, 1919: 168). Frege's principle is often interpreted as requiring a three-value logic. In fact, Frege was quite unequivocal in expressing his opinion on the matter: 'By die truth value of a sentence I understand the circumstance diat it is true or false. There are no further trudi values' (Frege, 1892: 46). Fregean logic of fiction is a logic of truth-value gaps, as conceived by van Fraassen (1966; 1969; cf. Haack, 1974: 47, 55 ff.; Pelc, 1977: 245). The idea of two functionally differentiated languages anticipates modern functional linguistics which has postulated polyfunctionality as a basic and necessary property of language. According to Holenstein, polyfunctionality is one of the major discoveries of die modern science of language (Holenstein, 1981). However, many philosophers of language have been unwilling to acknowl-

224

L. Dolezel

Frege's distinctions have strongly inspired modern logical semantics (see, for example, Dummett, 1973; Sluga, 1980; Schirn, ed., 1976) and, more recently, have Been also taken into account in literary semantics (Gabriel, 1975; Dolezel, 1982). It has to be borne in mind, however, that Frege's idea of a pure-sense language is an account of poeticity, rather than fictionality. Therefore, Fregean semantics will explain only those aspects of fictionality which can be derived from the non-referential, pure-sense use of language. With respect to all other aspects (ontological status of fictional entities, reference of fictional terms) a Fregean semantics of fictionality has to accept Russell's negative formulations. This essential compatibility is not surprising if we remember the common foundation of Frege's and Russell's semantics - the one-world model frame. Frege's poetic semantics hinges on the concept of non-referential, pure-sense language. However, if sense is defined on the basis of reference, i.e., as the referent's 'mode of presentation', then it seems impossible to speak about the sense of terms which lack reference. This major difficulty of Fregean poetic semantics has been pointed out by Evans: 'It is really not clear how there can be a mode of presentation associated with some term when there is no object to be presented' (Evans, 1982: 22). We have to emphasize, however, that this objection is valid only with regard to the standard interpretation of the concept of sense. The idea of pure-sense language becomes theoretically sound if sense is made logically independent of reference. Such a grounding of sense was theoretically justified in Saussure's account of meaning in language. 3. Saussure: 'self-referentiality'. Saussure eliminated the concept of reference from linguistic semantics by rejecting the conception of language as nomenclature. In this naive conception language appears as 'a list of terms each corresponding to a thing' (Saussure, 1916: 97; 65). A different relationship has to be identified as the basis of meaning. 'The linguistic sign unites not a thing and a name, but a concept and an acoustic image' (ib.: 98; 65f.). Both terms of the semantic relationship are now placed in one and the same, i. e., mental, domain. In Saussure's next step, the transformation from a nomenclatural to a semiotic semantics is completed: the terms of the semantic relationship are named by purely semiotic rather than psychological designations: 'We propose to retain the word sign ("signe") to designate the whole and to replace concept and acoustic image respectively by signified ("signifie") and sigmfier ("signinant")' (ib.: 99; 67). Thanks to this replacement, the linguistic sign is lifted out of extralinguistic relationships; meaning becomes a purely immanent phenomenon located in the signifiant-signifie nexus. Since both the signifiant and the signifie are language-

edge that under different functional conditions language (or language use) might be governed by different logjco-semantic principles. Evans, for example, characterized Frege's exemption of poetic language from the truth-conditions of the ordinary language as 'barely intelligible' (Evans, 1982: 23).

Possible Worlds and Literary Fictions

225

inherent structures, independent of the structures of the world, meaning is not determined by reference. Saussurean langue is a system of formally structured meanings which can, but need not be, used to refer. Saussurean semantics is complete without incorporating the concept of reference. A reinterpretation of Frege's sense is now possible: sense (signifie) is not determined by reference, but by the form of expression (signifiant). Or, to put it in Fregean terms, sense, the mode of presenting the referent, is determined by the structure of the referring linguistic expression.7 Saussure's linguistic theory has had a tremendous impact on contemporary thinking about literature. It inspired a non-referential poetic semantics which explains poetic meaning by manipulations of the signifiant-signifie nexus, with the signifiant taking a more or less active role in the meaning production.This semantics was enunciated by Saussure himself in his anagramatic studies (cf. Dolezel, forthcoming); its contemporary version is epitomized in the concept of self-referentiality. A lucid theoretical justification of self-referentiality is offered by Riffaterre: While in ordinary language reference proceeds along the Vertical' axis linking signifiant - signifie - referent, in poetic language 'the axis of significations is horizontal. The referential function in poetry is carried out from signifier to signifier' (Riffaterre, 1983: 26, 35f.).8 Poetic language is 'selfreferential' because the reference relation is immanent, linking signs with signs. The introduction of the concept of reference into the Saussurean framework creates confusion. While reference along the 'vertical' axis (in ordinary language) corresponds to the classical referential relationship (sign to world), reference along the 'horizontal' axis (in poetic language) is a different, intrasemic relationship (sign to sign). Intrasemic relationships, are, certainly, powerful generators of poetic meaning; but this fact cannot justify the claim of 'self-referentiality'.9 Whatever our opinion about 'self-referentiality', Saussurean semantics, just as its Fregean antecedent, is a theory of poeticity rather than fictionality. It is typical

A Saussurean interpretation of Frege's sense, without mentioning Saussure, has been formulated by Birjukov: 'The sense of a complex name is determined by the sense of its parts and the character of the rules according to which it is constructed; its sense changes when the sense of any of its parts changes. If we consider that the sense of each pan is determined by its linguistic character and that the rules according to which the name is composed are fixed in the grammatical structure of the name, then it becomes clear that the sense of the name is expressed by means of language and only thus' (Birjukov, 1964: 62). We have to emphasize that Riffaterre's 'horizontal' and 'vertical' axes of reference have nothing in common with Saussure's (and Jakobson's) paradigmatic and syntagmatic axes defining the structure of language. It is noteworthy that 'reference' is for Riffaterre synonymous with 'reference to reality' (op. cit. 17 et passim). The one-world model frame of Saussurean semantics is reconfirmed. Kerbrat-Orecchioni considers the very term 'self-reference' inappropriate. Her criticism of self-referentiality of poetic texts is, in fact, a criticism of its Saussurean foundation: 'Le referent ne disparait qu'avec la disparition du sens - mais ä Pinverse, toute emergence d'un sens impose la representation correlative d'un referent' (Kerbrat-Orecchioni, 1982: 31).

226

L. Dolezel

for Saussurean poetics that it has been only marginally or not at all concerned with clarifying the status of literary fictions.10 It appears that the one-world model frame is not a propitious ground for developing a semantics of fictionality. Russell's faceless non-entities are theoretically explained by being eliminated. In Frege's and Saussure's semantics the issue of fictionality is backgrounded: lack of reference is characteristic of poetry (Frege) or of langue (Saussure). Before we lose faith in the one-world model frame, however, we have to deal with the most ancient and authoritative version of fictional semantics, with the theory of mimesis. Despite its philosophical origins, mimesis has been cultivated primarily in aesthetics and literary criticism; unlike the logical and linguistic semantics dealt with so far, it was expressly designed as a theory of fictional representations. 4. Mimesis: actual prototypes. Mimetic theory has been usually formulated by defining the mimetic predicate. It should not surprise us that during its long history the interpretation of this predicate underwent substantial shifts (from "passive copying" to "creative representation").11 Focusing on the interpretation of the mimetic predicate, the theory has neglected the terms of the mimetic relationship - fictions and the world -, an aspect of mimesis which is crucial for fictional semantics. I shall recover the principles of this semantics by examining the interpretive practice of modern mimetic criticism. The basic and most popular method of mimetic criticism is the identification of an actualprototype of a fictional character, event, place, etc. Thus, for example, in The Discovery of King Arthur (1984) Geoffrey Ashe claims that the historical prototype of the legendary king is a fifth-century 'high king' of the Britons known as Riothamus. In Robin Hood. An Historical Enquiry (1985) John G. Bellamy confirms a 19th century hypothesis according to which the prototype of the legendary outlaw was a valet to Edward Π by the name of Robyn Hode. One could report many such 'discoveries', but the examples given should suffice for our purpose. They demonstrate that the basic interpretative move of mimetic criticism is the matching of fictional particulars (representations) with actual particulars (objects of representation). The crux of this semantics is a function which shall be called mimetic function: Fictional particular Pp represents actual particular PA. The scope of this function is, obviously, restricted to the set of fictional particulars which have, or can be assumed to have, actual prototypes. How do we explain, however, the large majority of fictional particulars which do not have actual particulars as their objects of representation? It would be absurd to claim that, say, the fictional Raskolnikov is a mimetic representation of an actual 10

11

This is true already of the 'classical' version of Saussurean poetics elaborated in the Prague school (cf. Dolezel, 1982: 291). The history of the changing conceptions of mimesis is sketched in Spariousu, ed., 1984 and Dolezel, forthcoming.

Possible Worlds and Literary Fictions

227

individual who lived in St. Petersburg around the middle of the 19th century. No historical search, however meticulous, would discover such a person. The impossibility of identifying an actual particular behind every fictional representation has forced mimetic criticism into an interpretive detour: fictional particulars are interpreted as representations of actual universals (psychological types, social groups, historical conditions, ideological positions, etc.). The mimetic function is radically altered: Fictional particular Pp represents actual universal UA· This function dominates a venerable version of mimetic semantics from Aristotle to Auerbach. Let me give just three examples of this kind of interpretation from Auerbach's popular book Mimesis. The Representation of Reality in Western Literature (German original, 1946; English translation, 1957):12 'We are confronted in their boredom (in the boredom of De la Mole's guests in Stendhal's Le rouge et le noir - L. D.) by a political and ideological phenomenon of the Restoration period' (401; 456). 'The novel (Madame Bovary - L. D.) is the representation of an entire human existence which has no issue' (434; 488). 'There are passages in it (Zola's Germinal - L.D.) which . . . depict with exemplary clarity and simplicity the situation and awakening of the fourth estate ...' (455; 512). If fictional particulars are interpreted as representing actual universals, mimetic theory becomes 'a language without particulars' (Strawson, 1959), a universalist semantics. Fictions are explained as representations of categorized reality.13 Fictional particulars disappear from mimetic representations. Is it possible for mimetic semantics to avoid translating fictional particulars into actual universals? The answer to this question is provided, in an instructive twist, in another best seller of mimetic criticism, Ian Watt's The Rise of the Novel (1957). In Watt's text, formulations of the following type abound: 'Defoe ... portrays the personal relationships of Moll Flanders' (111). 'We are given (by Richardson - L.D.) a highly detailed description of Grandison Hall' (26). 'Fielding ... lets us into BHfil's mind' (263). These statements preserve fictional particulars (Molly's personal relationships, Grandison Hall, Blifil's mind), but do not match them with represented actual objects or categories; rather the source of representation (its author) is identified.14 We are told who portrays, gives us a description of, lets us into the mind of, a fictional particular. These statements are instances of a new function: Actual source S A represents (pro-

This text is especially suited for our purpose. Avoiding deliberately any theoretical reflection (see 'Epilogue'), Auerbach demonstrates a spontaneous practice of mimetic criticism. Let us note that a universalist critic has to perform a double operation. First, he transcribes reality in abstract categories of an ideological, or sociological, or psychoanalytic, etc. conceptual framework; second, he matches the fictional particulars with the postulated categories of interpreted reality. Since one and the same person performs both operations, we should not be surprised by the high ratio of success of universalist interpretations. In a variant of this function, a text or textual device is given as the source of representation (cf. Dorrit Cohn, Transparent minds, 1978).

228

L. Dolezel

vides a representation of) fictional particular PF. By an insidious semantic shift of the mimetic predicate, mimesis is replaced by pseudomimesis. I speak about pseudomimesis because statements of a Wattian critic seem to express the mimetic function while, in fact, they do not. They do not derive fictional particulars from actual prototypes; rather, they presuppose that fictional entities somehow pre-exist the act of representation. There are Moll's personal relationships, Grandison Hall, Blifil's mind and Defoe, Richardson, Fielding, having a privileged access to them, report about them, describe them, share their knowledge with us. A fiction writer is a historian of fictional realms. The existence of these realms is assumed without being explained. As a theory of fictionality pseudomimesis is vacuous. Our analysis of the practice of mimetic criticism reveals the nature and the limits of the mimetic semantics of fictionality. This semantics does not provide a theoretically consistent account of fictional particulars. In the paradigm case which, however, is rather restricted, fictional particulars are matched with actual particulars-prototypes. In the most common version of mimesis, in universalist semantics, fictional particulars are explained by being transformed into actual universals. Finally, pseudomimesis, attributing fictional particulars to sources of representation, leaves them semantically unexplained. The theoretical inconsistency of mimetic semantics is not fortuitous; it is a necessary consequence of its one-world model frame. There is no escape from Russell's 'law': the actual world cannot be domicile of fictional particulars. Mimetic semantics in its universalist version has been perennially attractive because it provides a procedure for by-passing this law. Fictional particulars are integrated into the actual world by being interpreted as representations of abstract categories. Now we can see why the mimetic and the Russellian semantics of fictionality are not only compatible, but also complementary. Russell provided a logical procedure (theory of descriptions) for eliminating fictional particulars from the actual world; mimetic semantics offers a hermeneutic procedure (universalist interpretation) for naturalizing fictional particulars in the actual world. In either case, fictional particulars are theoretically explained by being deleted.

II. Possible-worlds model frame Fictional particulars are neccessary and indispensable constituents of literary fictions. Literature deals with concrete fictional persons in specific spatial and temporal settings, bound by peculiar relationships and engaged in unique struggles, quests, frustrations.15 A model frame which does not accommodate According to Martinez-Bonati, 'a world of individuals' is 'the fundamental compass of narrative ... Cervantes' Don Quixote is not basically a type or symbol, but an individual' (MartinezBonati, 1981:24).

Possible Worlds and Literary Fictions

229

the concept of fictional particular cannot be an adequate theoretical base of fictional semantics.16 This serious deficiency of the one-world model leads us to explore the potentials of a radically different theoretical foundation of fictional semantics, the model frame of multiple, possible worlds. This exploration is stimulated by the rapid development of possible-worlds semantics in logic and philosophy of language. Since Kripke (1963) suggested the 'classical' Leibnizian concept as interpretant of an axiomatic set-theoretic model of logical modalities, the entire system of formal logic was reinterpreted on the assumption that Our actual world is surrounded by an infinity of other possible worlds' (Bradley, Swartz, 1979: 2).17 During the 1970's first attempts were made to interpret fictional worlds of literature in terms of possible-worlds semantics (van Dijk, 1974/75; Pavel, 1975/76; Eco, 1979; Dolezel, 1979).18 It has to be emphasized, however, that a comprehensive theory of literary fictions will not emerge from a mechanical appropriation of this conceptual system. Fictional worlds of literature have a specific character by being embodied in literary texts and by functioning as cultural artefacts. A comprehensive theory of literary fictions will arise A clear sign of the impossibility of developing a semantics of fictionality within the one-world model frame is the recent popularity of pragmatic theories. Since the one-world model gives no possibility of locating fictionality on the 'language-world' axis, it has to be placed in the 'language-user' nexus. Fictionality is explained as a speech-act convention. The paradigm version of the pragmatics of fictionality, one that treats fictional speech acts as pretended assertions, emerged spontaneously within both the Fregean, and the Russellian framework. Frege himself gave a stimulus for this development in his later writings, especially in his often quoted analogy: 'As stage thunder is only apparent thunder and a stage fight only an apparent fight, so stage assertion is only apparent assertion. It is only play, only poetry' (Frege, 1918/19: 36; cf. Pollard, 1973: 58f; Gabriel, 1975; Dolezel, forthcoming). In the Russellian tradition, a recourse to pragmatics was suggested by Russell (1970: 277) and formulated by Evans: All participants of fictional communication (the story-teller, the audience, the critics) are engaged in the game of pretence: 'People are engaged in ... serious exploitation of pretence . . . by knowingly using empty singular terms'. Moreover, for the audience die linguistic game might turn into an ontological game: 'He (the receiver offictionalinformation - L.D.) pretends there are such-and-such people and within the scope of the pretence he admires or loathes them' (Evans, 1982:364 ff.). For a most cogent presentation of die pragmatic theory of fictionality, see Searle, 1974/75; for its criticism, Cohen, 1980: 162f. It should be noted that as a formal model the possible-worlds frame does not require any ontological commitment. Pointing specifically to Hintikka's and Kripke's proposals, a Soviet logician has emphasized that they should be taken 'simply as mathematical models of the corresponding logical calculi, without any philosophical interpretation' (Slinin, 1967: 137). Outside formal logic, however, the model cannot preserve ontological innocence. Adams has recognized the ontological split by differentiating between the 'actualist' and the 'possibilist' versions of possible-worlds semantics. Possibilism treats all possible worlds as ontologically uniform; in actualism (which is a possible-worlds form of realism) the actual world has the privilege of empirical existence while die other worlds are its possible alternates (Adams, 1974). It seems that die actualist position is inscribed in Kripke's original model structure where one set (G) is singled out from die set of sets K (Kripke, 1963: 804). It is symptomatic diat during die domination of die one-world model, a Leibnizian possibleworlds semantics of fictionality, oudined in die 18di century by Baumgarten, Breitinger and Bodmer (cf. Abrams, 1953: 278f.; Dolezel, forthcoming), was practically forgotten.

230

L. Dolezel

from a fusion of possible-worlds semantics with text theory. I want to prepare ground for such a fusion by taking possible-worlds semantics both as a theoretical foundation of the semantics of fictionality, and as a theoretical background against which the specific properties of literary fictions can be grasped. Let me first formulate three fundamental theses of literary fictional semantics which can be derived from the possible-worlds model frame: 1. Fictional worlds are sets of possible states of affairs. The most important feature of the possible-worlds model frame is its legitimation of non-actualized possibles (individuals, attributes, events, states of affairs etc.) (cf. Bradley, Swartz, op. cit.: 7f.). A fictional semantics derived from this model will accept the concept of fictional particular without reservations. While Hamlet is not an actual man, he is a possible individual inhabiting the fictional world of Shakespeare's play. Consequently, the name Hamlet does not lack reference; it refers to this possible individual. Because fictional individuals possess sets of determinate properties, we can differentiate between false and true statements about Shakespeare's Hamlet. Unlike the one-world semantics of fictionality, the possible-worlds model frame allows one to formulate characteristics of fictional particulars. Rather than being deleted in the process of semantic interpretation, fictional particulars can be described and specified in their diverse aspects. If fictional particulars are interpreted as nonactualized possibles, the difference between fictional and actual persons, events, places, etc. becomes obvious. Everybody would agree that fictional characters cannot meet, interact with, communicate with, actual people (cf. Walton, 1978/79: 17). In the fictional semantics of the one-world model frame, however, this distinction is often blurred on the account of shared proper names. Possible-worlds semantics correctly insists that fictional individuals cannot be identified with actual individuals of the same name (cf. Ishiguro, 1981:75). Tolstoy's Napoleon or Dickens's London are not identical with the historical Napoleon or the geographical London. Fictional individuals are not dependent for their existence and properties on actual prototypes. It is irrelevant for the fictional Robin Hood whether a historical Robin Hood existed or not. To be sure, a relationship between the historical Napoleon and all the possible fictional Napoleons has to be postulated; this relationship, however, reaches over world boundaries and requires cross-world identification.^ The identity of fictional individuals is protected by the boundary between the actual and the possible worlds. As non-actualized possibles all fictional entities are ontologically homogeneous. Tolstoy's Napoleon is no less fictional than his Pierre Bezuchov and Dickens's London is no more actual than Lewis's Wonderland. The principle of ontological homogeneity is a necessary condition of the coexistence and comHintikka's 'individuation function* is a formal tool of cross-identification. It 'picks out from several possible worlds a member of their domains as the "embodiment" of that individual in this possible world or perhaps rather as the role which that individual plays under a given course of events' (Hintikka, 1975: 30).

Possible Worlds and Literary Fictions

231

possibility of fictional particulars; it explains why fictional individuals can interact and communicate with one another. A naive view which presents fictional individuals as a mixed bag of 'real' people and 'purely fictitious' characters is explicitly refuted.20 Ontological homogeneity epitomizes the sovereignty of fictional worlds. 2. The set of fictional worlds is unlimited and maximally varied. If fictional worlds are interpreted as possible worlds, literature is not restricted to the imitations of the actual world. "The possible is wider than the actual' (Russell, 19372: 66; cf. Plantinga, 1977: 245). To be sure, possible-worlds semantics does not exclude from its scope fictional worlds similar or analogous to the actual world; at the same time, it includes without difficulty the most fantastic worlds, far removed from, or contradictory to, 'reality'. The whole gamut of fictions is covered by one and the same semantics. There is no justification for a double semantics of fictionality, one for fictions of the 'realistic' type, another one for 'fantastic' fictions. The worlds of realistic literature are no less fictional than the worlds of fairy-tale or science fiction.21 It is well-known that Leibniz imposed a restriction on possible worlds, but this restriction is purely logical: Possible worlds have to be free of contradictions (Leibniz 3: 574; Loemker 2: 833). Worlds which imply contradictions are impossible, unthinkable, 'empty'. Do we have to accept this restriction into fictional semantics? I will deal with this question in the last section of this paper. At the moment, let me just remark that even if the possible-worlds model is restricted to the Leibnizian universe it provides a much larger space for literary fictions than the one-world model frame.22 While imposing a logical restriction on possible worlds, Leibniz left wide open the variety of their designs. He stipulated different 'laws' ('general order') for different possible worlds. The laws of nature are just a special case of possible orders, valid in the actual and the 'physically possible' worlds (cf. Bradley, Swartz, op. cit.: 6).23 A general order determines a possible world by operating as a constraint on admissibility: only such possible entities are admitted into the

20

21

22

23

The 'mixed-bag' conception requires a double semantics for fictional texts, one for sentences about Pierre Bezuchov, another one for sentences about Napoleon (cf. Pollard, 1973: 61; Pelc, 1977: 266). In reading fictional texts we are expected to switch constantly from one mode of interpretation to another. It has been observed that the same principle is valid from the point of view of the reader: 'For the reader it is not easier to create and believe in the well-documented world of Zola than it is for him to imagine hobbits or elves; the imagination leap into the novel's world of time and space must be made in both cases' (Hutcheon, 1980: 78). It might, in fact, appear that the possible-worlds model frame is so broad as to be of little interest for empirical study. We should bear in mind, however, that for a majority of semantic problems a restricted set of relevant possible worlds can be determined (cf. Hintikka, 1975: 83). 'Worlds may differ from the actual world not only in number and quantity (of their elements L. D.), but in quality. Other worlds might have other laws of motion ... Every causal law, in fact, (though not Causality itself) might have been different' (Russell, 19372: 68).

232

L. Dolezel

world that comply with its general order. In such a way, the set of all possibles is split into 'many different combinations of compossibles' (Leibniz 3: 572-576; Loemker 2: 1075f.). In this perspective, a fictional world appears as a set of compossible fictional particulars characterized by its own global, macrostructural organization. Structure and specificity are complementary aspects of the worlds' incuviduation. The macrostructural conception of fictional worlds has proved most fruitful for literary semantics (cf. Dolezel, 1985). Here, I cannot go into the identification of the diverse global constraints which can be imposed on fictional worlds, nor into the description of the resulting variety of world structures. Let me just give one illustrative example. It has been pointed out that modalities (modal systems) can operate as macrogenerators of fictional worlds (Greimas, 1966; 1970; Dolezel, 1976). By considering alethic modalities (the system of possibility, impossibility and necessity) in this role, we can generate not only the well-known natural und supernatural worlds, but also the up-to-now unnoticed hybrid world.24 This example indicates how fictional world defined as macrostructure of compossible fictional particulars becomes an operational concept of literary analysis. 3. Fictional worlds are accessible from the actual world. Possible-worlds semantics legitimates the sovereignty of fictional worlds vis-ä-vis the actual world; at the same time, however, its notion of accessibility offers an explanation of our contacts with fictional worlds. The access requires crossing of world boundaries, transit from the realm of actual existents into the realm of fictional possibles. Under this condition, physical access is impossible. Fictional worlds are accessible from the actual world only through semiotic channels by means of information processing. In the formation of fictional worlds, the actual world participates by providing models of its structure (including the author's experience), by anchoring the fictional story to a historical event (Wolterstorff, 1980:189), by transmitting 'brute facts' or cultural 'realemes' (Even-Zohar, 1980), etc. In these information transfers the actual-world 'material' enters into the structuring of fictional worlds. Literary scholars have studied intensively the participation of 'reality' in the genesis of fictions. Possible-worlds semantics makes us aware that the actual material has to undergo a substantial transformation at the world boundary: it has to be converted into non-actual possibles, with all the ontological, logical and 24

It has been proposed (in Dolezel, 1984) that the world of some of Kafka's fictions (such as 'The Metamorphosis' or 'A Country Doctor') is a hybrid world. It is interesting to note in this connection Austin's opinion as recorded by Berlin. When asked whether the hero of 'The Metamorphosis' should be spoken of 'as a man with the body of a cockroach, or as a cockroach with the memories and consciousness of man', Austin replied: 'Neither ... In such cases we should not know what to say. This is when we say "words fail us", and mean this literally. We should need new word. The old one just would not fit' (Berlin, 1973:11). Austin did not notice that the old word hybrid fas this new case of problematic identity.

Possible Worlds and Literary Fictions

233

semantical consequences. We have already noted this conversion in the special case of fictional individuals; actual-world (historical) persons are allowed to enter a fictional world only if they assume the status or possible alternates. In the reception of fictional worlds the access is provided through literary texts which are read and interpreted by actual readers. The reading and interpretation involves many different procedures and depends on many variables, such as the type of the reader, his style of reading, the purpose of his reading, etc. The details of the access procedures will be revealed only by studying actual reading and interpretation activities. Here, let us just note that thanks to semiotic mediation, an actual reader can Observe' fictional worlds and make them a source of his experience, just as he observes and experientially appropriates the actual world.25 When speaking about semiotic textual mediation, we are already invoking a specific feature of literary fictions which points outside possible-worlds semantics. I do not claim to have exhausted the theoretical potentials of this model. It seems, however, that at this point we have reached the limits of its utility.

El. Specific features of fictional worlds of literature I have suggested earlier that the possible-worlds model frame is appropriate for providing the foundations of the theory of literary fictions, but that it cannot substitute for such a theory. If we do not want to turn the possible-worlds model into a collection of theoretically useless metaphors, we have to perceive the limits of its explanatory power with respect to cultural artefacts. Some fundamental features of fictional worlds of literature cannot be derived from the possible-worlds model of formal semantics; at the same time, however, they can be identified only against the background of this model frame. Let me point out just three such features: 1. Fictional worlds of literature are incomplete. This property of fictional worlds has been widely recognized (Lewis, 1978: 42; Heintz, 1979: 90 ff.; Howell, 1979: 134f.; Parsons, 1980: 182-185; Wolterstorff, 1980: 131-134). Incompleteness is a manifestation of the specific character of literary fictions since the possible worlds of the model frame (including the actual world) are assumed to be complete ('Carnapian') logical structures. The property of incompleteness implies that many conceivable statements about literary fictional 23

Using the example of dramatic performance, Wolterstorff makes the following claim: 'To regard us as watching the dramatis personae - that is just confusion... It is not the case that I saw Hedda shoot herself, since in that world (in the world of the drama - L.D.) I don't even exist, and so can't see Hedda . . . What I can see is someone playing the role of Hedda' (Wolterstorff, 1980: 111 f.). Wolterstorff denies the actual spectator access to fictional worlds precisely because he does not consider actors as semiotic mediators. Walton, again ignoring semiotic mediation, has to make the implausible assumption that the reader/spectator is both actual and fictional (Walton, 1978/79: 21 f).

234

L. Dolezel

worlds are undecidable. Using the popular issue of the number of Lady Macbeth's children as his example, Wolterstorff has succinctly justified this interpretive principle: 'We shall never know how many children had Lady Macbeth in the worlds of Macbeth. That is not because to know this would require knowledge beyond the capacity of human beings. It is because there is nothing of the sort to know' (Wolterstorff, op. cit.: 133; cf. Heintz, op. cit.: 94).26 If incompleteness is a logical 'deficiency' of fictional worlds, it is an important factor of their aesthetic efficiency. 'Empty' domains are constituents of the fictional world's structure no less than are the 'filled' domains. The distribution of filled and empty domains is governed by aesthetic principles, i. e., by a writer's style, by period or genre conventions, etc. Several recent studies in literary semantics have revealed fictional world's incompleteness as a basic aspect of literary styles or types (cf. Dolezel, 1980a; Pavel, 1983; Ryan, 1984).27 2. Many fictional worlds of literature are semantically unhomogeneous. We have claimed that fictional worlds are formed by macrostructural constraints which determine the set of their compossible constituents. At the same time, however, we can easily discover that many fictional worlds manifest a complex 'inner' semantic structuring. Such worlds are sets of semantically diversified domains integrated into a structural whole by the formative macroconstraints. The semantic unhomogeneity is especially prominent in the fictional worlds of narrative literature. A prime example of the semantic partitioning of narrative worlds is provided by agential domains. Every fictional agent forms his own domain constituted by his property set, his relation network, his belief set, his action scope, etc. (cf. Pavel, 1980). If there is just one agent in the world - as in Hemingway's story 'Big Two-Hearted River' - then this agent's domain is equivalent to the fictional world. In a more common case, in the case of multi-agent worlds, the fictional world is a set of agential domains held together by the macrostructural conditions of the agents' compossibility. I have already mentioned that modalities represent an important formative macroconstraint of narrative worlds. Modal structuring yields a variety of homogeneous, as well as unhomogeneous, narrative worlds. Thus, for example, the world of realistic fiction is an alethically homogeneous, i. e. natural (physically possible) world; on the opposite pole, an alethically homogeneous supernatural (physically impossible) world (the world of deities, demons, etc.) can be conceived. A mythological world, however, is a semantically unhomogeneous structure constituted by the coexistence of natural and supernatural domains. 26 27

According to Lewis, answers to such 'silly questions' as 'what is inspector Lestarde's blood type' would doubtless fall into the category of neither true nor false statements (Lewis, 1978: 43). If the 'filling-in of gaps', postulated by phenomenological dieories of reading (cf. Iser, 1978), applies to the 'empty' domains, then it is a reductionist procedure. Fictional-world structures, richly diversified in their incompletness, are reduced to a uniform structure of the complete (Carnapian) world.

Possible Worlds and Literary Fictions

235

The domains are divided by sharp boundaries but, at the same time, are linked by the possibility of cross-boundary contacts. The case of the mythological world demonstrates that semantic unhomogeneity is a primordial feature of narrative-world formation. A fictional world of narrative has to be a complex set of diversified domains in order to accommodate many different possible individuals, states of affairs, events, actions, etc. This semantic complexity leads some critics to view narrative fictional worlds as miniature models of the actual world. Such a view, however, is misleading. Semantic complexity is a prime manifestation of the structural self-sufficiency of fictional worlds. 3. Fictional worlds of literature are constructs of textual activity. Having characterized fictional worlds as sets of non-actualized possibles, we have identified their general ontological base. We have left unspecified the characteristics which differentiate fictional entities from other non-actualized possibles. Hamlet is a different kind of possible individual than the present king of France.28 We have to assume that a special operation is needed to transmute non-actual possibles into fictional entities, to assign fictional existence to possible worlds. Already the first, Leibnizian version of the possible-worlds semantics of fictionality has suggested a solution of this problem. According to this view, possible worlds acquire fictional existence by being discovered (cf. Dolezel, forthcoming). This explanation is based on Leibniz's assumption that all possible worlds have a transcendental existence (in the divine mind) (cf. Stalnaker, 1976: 65). The poet, thanks to the power of his imagination, gains a privileged access to these worlds, similarly as the scientist, using the power of his microscope, gains access to the invisible microworld. Existing as non-actualized possibles in transcendental obscurity, fictional worlds are publicly displayed in poet's descriptions.29

28

29

The difference is revealed by Linski's 'test': 'Though we can ask whether Mr. Pickwick was married or not, we cannot sensibly ask whether the present king of France is bald or not' (Linski, 1962: 231; cf. Woods, 1974: 14). Of course, nothing could prevent the present king of France from becoming a fictional individual if he was transferred from logical examples into fictional texts. Even this brief summary indicates that Leibnizian poetics takes, in fact, a mimetic view of the emergence of fictional worlds. The scope of imitation/representation is extended to cover not only the actual, but also the possible worlds. Fictions are descriptions of possible worlds which pre-exist the act of description. If I understand it well, WolterstorfP s concept of world-projection is a surprising return to Leibnizian poetics: 'The world of Dead Souls existed apart from Gogol's activity as writer; and neither he nor anyone else has made it occur. The artist's activity consists in projecting an already existent but normally non-occurrent state of affairs by way of indicating certain states of affairs... That conjunctive state of affairs which is in fact the world associated with the text of Dead Souls existed, awaiting Gogol's selection, waiting to become the world of his work. The text is made and associated with the world rather than the world made and associated with the text. World-projection is a mode of selection, not a mode of creation' (Wolterstorff, 1980: 129f.).

236

L. Dolezel

Contemporary thinking about the origins of possible worlds is not bound by the metaphysical assumptions of Leibniz's philosophy. Possible worlds are not discovered in some remote, invisible or transcendent depositories, they are constructed by human minds and hands. This explanation has been explicitly given by Kripke: One stipulates possible worlds, one does not discover them by powerful microscopes' (Kripke, 1972:267; cf. Bradley, Swartz, 1979:63f.). As pointed out by Stalnaker, possible worlds arise from diverse activities of human agents (Stalnaker, 1984: 50f.) (For a contrary view, see Lewis, 1986: 3). The construction offictionalpossible worlds occurs, primarily, in cultural activities poetry and music composition, mythology and story telling, painting and sculpting, theater and cfance performance, film making, etc. Many semiotic systems - language, gestures, motions, colours, shapes, tones, etc. - serve as media of fictional world-construction. Literary fictions are constructed in the creative act of the poetic imagination, in the activity of poiesis. Literary text is the medium of this activity. Employing the semiotic potentials of the literary text, the poet brings into fictional existence a possible world which has not existed prior to his poietic act. In this explanation of the origins of fictional worlds, constructional texts are sharply differentiated from descriptive texts. Descriptive texts are representations of the actual world, of a world existing prior to any textual activity. In contrast, constructional texts are prior to their worlds; fictional worlds are dependent on, and determined by, constructional texts. As textually determined constructs fictional worlds cannot be altered or cancelled, while the versions of the actual world provided by descriptive texts are subject to constant modifications and refutations.30 We have emphasized the crucial role of the poet's imagination in the construction of fictional worlds of literature. Literary semantics, however, is concerned primarily with the semiotic medium of world construction, with the literary text. Constructional texts can be called fictional texts in the functional sense: they are actual texts with the potential of constructing fictional worlds. However, the role of the fictional text does not end in its serving as the medium of the poet's constructive activity; the literary text is also the semiotic means for storing and transmitting fictional worlds. We have already mentioned that fictional worlds are publicly and permanently available in fictional texts. As long as the text exists, its world can be at any time reconstructed in reading and interpretive activities of potential receivers. From the viewpoint of the reader the fictional text can be characterized as a set of instructions according to which the fictional world is to be recovered and reassembled. The crucial link between fictional semantics and text theory is now becoming obvious. The genesis, the preservation and the reception of fictional worlds 30

Radical constructivism obliterates the distinction between world description and world construction proclaiming all texts world-constructing and all worlds dependent on texts (cf. Goodman, 1978; Schmidt, 1984). For a criticism of this 'semiotic idealism', see Savan, 1983.

Possible Worlds and Literary Fictions

237

depend on special semiotic capacities of fictional texts. For a theory of literary fictions it is especially important to pinpoint the textual capacity which can be credited with the genesis of fictional worlds. I have suggested earlier (Dolezel, 1980b) that this world-constructing capacity can be identified if literary texts are interpreted in terms of the Austinian theory of performative speech acts.31 Austin has stipulated that performative speech acts carry a special illocutionary force; thanks to this force the utterance of a performative speech act under appropriate felicity conditions (given by extralinguistic conventions) brings about a change in the world (Austin, 1962; 1971; cf. Searle, 1979: 16rf.; Urmson, 1979). Genesis of fictional worlds can be seen as an extreme case of world-change, a change from nonexistence into fictional existence. The special illocutionary force of literary speech acts which brings about this change has been called (in Dolezel, op. cit.) the force of authentication. A non-actualized possible state of affairs becomes a fictional existent by being authenticated in a felicitously uttered literary speech act.32 To exist fictionally means to exist as a textually authenticated possible. The theory of authentication assumes that the force of authentication is exercised differently in different literary text types (genres). In the particular case of the narrative text type (which was studied in Dolezel, 1980b), the force of authentication is assigned to the speech acts which originate with the so-called narrator. The narrator's authority to issue authenticating speech acts is given by the conventions of the narrative genre.33 The mechanism of authentication is best manifest in the case of the Omniscient', 'reliable', authoritative Er-form narrator. Whatever is uttered by this source, becomes automatically a fictional existent. Other types of narrators, such as the 'unreliable', 'subjective' Ich-form narrator, are sources with a lower degree of authentication authority. Since fictional existence depends on the act of authentication, its character is, ultimately, determined by the degree of authority of the authenticating source. The theory of authentication leads us to acknowledge different modes of fictional existence correlated with the different degrees of the text's force of authentication. In such a way, fictional existence is not only constructed, but also manipulated by the authenticating narrative act. 31

32 33

A link between literature and performatives has been perceived already by Barthes: 'Writing (in the sense of ecriture - L.D.) can no longer designate an operation of recording, notation, representation, "depiction" (as the Classics would say); rather, it designates exactly what linguists, referring to Oxford philosophy, call a performative' (Barthes, 1977:145.). Barthes did not got beyond this passing remark. If we want to express the authenticating illocutionary act by an explicit performative formula, then we could suggest the prefix: Let it be. It is true that the actual source of all speech acts in narrative is its author. In the narrative text itself, however, there is no author's discourse. By a generic convention, discourses of the narrative text are assigned to different fictional sources (die narrator, the acting characters). It is a commonly accepted principle of narrative theory that the narrator cannot be identified with the author. For this reason, the so-called sayso pragmatics of fictionality which relies on the author's authority (Woods, 1974: 24 f.) is misdirected.

238

L. Dolezel

IV. Self-voiding texts and impossible fictional worlds The illocutionary force of a performative speech act is activated only if its particular felicity conditions are met. If a breach of any of these conditions occurs, the act is 'null and void'; no change in the world is brought about. A different kind of failure of the performative act is 'self-voiding'. A performative utterance is said to be self-voiding if it is 'abused', if, for example, it is issued 'insincerely' (Austin, 1971: 14f.). Just as the breach of felicity conditions does, self-voiding deprives the speech act of its performative force. The concept of self-voiding is, in my opinion, highly important for the theory of fictional narratives. It offers an explanation of diverse non-standard narratives of modern literature which arise, in fact, from a cultivation of performative failure. The authenticating act of fictional narratives is abused in many different ways by not being performed 'seriously'. Let me give two examples of such 'abuse': a) In skaz narrative the authenticating act is abused by being performed with irony. The skaz narrator engages in a non-binding game of story-telling, moving freely from the third to the first person, from a lofty to a colloquial style, from the Omniscient' to the 'limited knowledge' posture. Skaz has been extremely popular in Russian fiction, especially since it was established by Gogol (cf. Ejchenbaum, 1979). b) In self-disclosing narrative ('metafiction') the authenticating act is voided by being 'laid bare'. All procedures of fiction-making, and particularly the authentication of fictional worlds, are practiced overtly as literary conventions. Selfdisclosing narrative has achieved great popularity in Postmodernism (the 'nouveau roman', John Barth, John Fowles) and has been very attractive to critics (cf. Hutcheon, 1980; Christensen, 1981). It is a radical manifestation of literature's power to generate aesthetic effects by flaunting its hidden foundations. Both the skaz and the self-disclosing narrative are self-voiding; in both the authenticating act loses its performative force. Fictional worlds constructed by self-voiding narratives lack authenticity. They are introduced, presented, but their fictional existence is not definitely established. Self-voiding narratives are games with fictional existence. On the one hand, possible entities seem to be brought into fictional existence since conventional authentication procedures are applied; on the other hand, the status of this existence is made dubious because the very foundation of the authenticating mechanism is undermined. Ultimately, it is impossible to decide what exists and what does not exist in the fictional worlds constructed by self-voiding narratives. In the case of self-voiding narratives, the fictional world's lack of authenticity is due to a disturbance affecting the authenticating act. The roots of the disturbance are pragmatic, although its presence is manifest in the text's semantic features. However, a disruption of the fictional world's authenticity can be also achieved by a purely semantic strategy. Impossible fictional worlds, i. e. worlds which include inner contradictions, or imply contradictory states of affairs, are a

Possible Worlds and Literary Fictions

239

case in point. Let me mention O. Henry's story 'Roads of Destiny' as an example of such world-construction. Its protagonist dies three times in three different ways. Since all the conflicting versions of his demise are constructed by the authoritative narrator, they are all fully authentic. They exist in the fictional world, juxtaposed, unreconciled, unexplained. Ultimately, it is impossible to decide which event is and which is not a legitimate constituent of the fictional world. Impossible worlds are no less an 'abuse' of fiction making than selfvoiding narratives. In this case, however, the authenticity of fictional existence is denied by the logico-semantic structure of the world itself. Literature offers the means for constructing impossible worlds, but at the price of frustrating the whole enterprise: fictional existence in impossible worlds cannot be made authentic. The Leibnizian restriction is circumvented, but not cancelled. The close link between lack of authenticity and impossibility of fictional worlds is confirmed by narratives where both the pragmatic and the semantic disturbance operate. Such a double manoeuvre cancels fictional existence in Robbe-Grillet's novel La maison de rendez-vous (The House of Assignation). Robbe-Grillet's text certainly constructs an impossible world, a maze of contradictions of several different orders: a) one and the same event is introduced in several conflicting versions; b) one and the same place (Hong-Kong) is and is not the setting of the novel; c) events are ordered in contradictory temporal sequences (A precedes B, B precedes A); d) one and the same fictional entity recurs in several existential modes (as fictional 'reality' or theater performance or sculpture or painting, etc.).34 Just as O. Henry's 'Roads of Destiny' does, La maison de rendez-vous constructs several incompatible alternate plots or plot fragments (cf. Ricardou, 1973:102f.). In Robbe-Grillet's case, however, the impossible logico-semantic world structure is coupled with a pragmatic flouting of the authenticating narrative act. The act of world-constructing is tentative, unfinished, crumbling into a series of frustrated attempts. As a result, the text of the novel is a sequence of drafts, with recurring cuts, new beginnings, corrections, deletions, additions etc. La maison de rendez-vous is a self-disclosing narrative of a most radical type, an overt demonstration of fiction-making as a trial-and-error procedure.35 Robbe-Grillet's novel reconfirms the ultimate impossibility of constructing a fictionally authentic impossible world. This undertaking necessitates the ruin of the very mechanism of fiction-making. Literature, however, turns this destruc-

34

35

The 'mixing' of different modes of existence seems to be a universal feature of modern an; its explicit demonstration is the cubistic collage which incorporates real objects in paintings (cf. Hintikka, 1975: 246). This technique has been glimpsed by Sturrock in Robbe-Grillet's first novel Les Gommes; the novel dramatizes 'the conditions under which a novel comes into being, or rather tries to come into being' (Sturrock, 1969: 172). Morrissette has recovered an underlying coherent plot of La maison de rendez-vous (Morrissette, 1975:260 f.). It has to be noted, however, that this coherent plot has no privileged authenticity.

240

L Dolezel

tive process into a new achievement. Fiction-making becomes overtly what it has been covertly: a game with fictional existence. It is in such a way that literature suggests to us the sense of all quixotic attempts at squaring the circle: they are manifestations of the never-ending constructive play of the human imagination.

References Abrams, M. H. 1953. The mirror and the lamp. Romantic theory and the critical tradition. LondonOxford-New York: Oxford University Press. Adams, Robert M. 1974. Theories of actuality. Nous 8: 211-231. Austin, J. L. 1962. How to do things with words. Cambridge, Mass.: Harvard University Press. -. 1971. Performative-constantive. In: J. R. Searle, ed., The philosophy of language: 13-22. Oxford: Oxford University Press. Barthes, Roland. 1977. Image, music, text. New York: Hill and Wang. Berlin, Isaiah. 1973. Austin and the early beginnings of Oxford philosophy. In: Sir Isaiah Berlin et al., Essays onj. L. Austin: 1-16. Oxford: Clarendon Press. Birjukov, B. V. 1964. Two Soviet studies on Frege. (Ignacio Angelelli, ed. and trans.). Dordrecht: Reidel. Bradley, Raymond, Norman Swartz. 1979. Possible worlds. An introduction to logic and its philosophy. Oxford: Basil Blackwell. Christensen, Inger. 1981. The meaning ofmetafiction. Bergen-Oslo-Troms0: Universitetsforlaget. Cohen, L. Jonathan. 1980. The individuation of proper names. In: Zak van Straaten, ed., Philosophical subjects. Essays presented to P. E Strawson: 140-163. Oxford: Clarendon. Dijk, Teun A. van. 1974/75. Action, action description and narrative. New literary history 6: 273-294. Dolezel, Lubomir. 1976. Narrative semantics. PTL: A journal for descriptive poetics and theory of literature 1: 129-151. -. 1979. Extensional and imensional narrative worlds. Poetics 8: 193-211. -. 1980a. A short note on a long subject: literary style. In: Vozmi na radosf. To honourJeanne van der Eng-Liedmeier: 1-7. Amsterdam: Slavic Seminar. -. 1980b. Truth and authenticity in narrative. Poetics Today 1: 7-25. -. 1982. Mukarovsky and the idea of poetic truth. Russian Literature 12: 283-297. -. 1984. Kafka's fictional world. Canadian review of comparative literature/Revue Canadienne de litterature comparee 10: 61-83. -. 1985. Pour une typologie des mondes fictionnels. In: Herman Parret, Hans-George Ruprecht, eds., Exigences et perspectives de la semiotique. Recueil d'hommages pour Algirdas Julien Greimas: 7-23, Amsterdam-Philadelphia: Benjamins. -. forthcoming. Occidental Poetics. Chapters from a research tradition. Dummett, Michael. 1973. Frege. Philosophy of language. London: Duckworth. Eco, Umberto. 1979. The role of the reader. Bloomington, Ind.: Indiana University Press. Ejchenbaum, Boris. 1919. Kak sdelana 'Sinel' Gogolja. In: Poetika. Sbomikipo teorii poeticeskogo jazyka: 151-165. Petrograd. Evans, Gareth. 1982. The varieties of reference (John McDowell, ed.). Oxford: Clarendon Press, New York: Oxford University Press. Even-Zohar, Itamar. 1980. Constraints on realeme insertability in narrative. Poetics Today 1:65-74. van Fraasen, B. C. 1966. Singular terms, truth-value gaps, and free logic. Journal of Philosophy 63: 481-495. -. 1969. Presuppositions, supervaluation and free logic. In: K. Lambert, ed., The logical way of doing things: 67-91. New Haven-London: Yale University Press.

Possible Worlds and Literary Fictions

241

Frege, Gottlob. 1892. Über Sinn und Bedeutung. Zeitschrift ßir Philosophie und philosophische Kritik 100: 25-50 (Engl. trans, in Zabeeh, KJemke, Jacobson, eds., 1974: 118-140). -. 1918/19. Der Gedanke. Eine logische Untersuchung. Beiträge zur Philosophie des deutschen Idealismus 1: 58-77 (English trans, in: E. D. Klemke, ed., Essays on Frege, 1968: 507-535. Urbana-Chicago-London: Illinois University Press). Gabriel, Gottfried. 1975. Fiktion und Wahrheit. Eine semantische Theorie der Literatur. Stuttgart-Bad Cannstatt: Holzboog. Greimas, A. J. 1966. Semantique structurale. Recherche de methode. Paris: Larousse. -. 1970. Du sens: Essais semiotiques. Paris: Seuil. Grossmann, Reinhardt. 1983. The categorial structure of the world. Bloomington: Indiana University Press. Goodman, Nelson. 1978. Ways ofworldmaking. Indianapolis-Cambridge: Hackett. Haack, Susan. 1974. Deviant logic. Some philosophical issues. Cambridge: Cambridge University Press. Heintz, John. 1979. Reference and inference in fiction. Poetics 8: 85-99. Hintikkajaakko. 1975. Theintentionsofintent onalityando ernewrnodeL·formodalü Reidel. Holenstein, Elmar. 1981. On the poetry and the plurifunctionality of language. In: Barry Smith, ed., Structure and Gestalt: Philosophy and literature in Austria-Hungary and her successor states: 1-43. Amsterdam: Benjamins. Howell, Roben. 1979. Fictional objects: How they are and how they are not. Poetics 8: 129-177. Hutcheon, Linda. 1980. Narcissistic narrative. The metaßctional paradox. Waterloo, Ont.: Wilfrid Laurier University Press. Iser, Wolfgang. 1978. The act of reading. A theory of aesthetic response. Baltimore-London: Johns Hopkins University Press. (German original, 1976). Ishiguro, Hide. 1981. Contingent truths and possible worlds. In: R. S. Woolhouse, ed., 1981:64-76. Kerbrat-Orecchioni, Catherine. 1982. Le texte litteraire: Non-reference, auto-reference, ou reference fictionnelle? Texte l: 27-49. Kripke, Saul A. 1963. Semantical considerations on modal logic. A eta Philosophica Fennica 16:83-94 (reprinted in: Zabeeh, Klemke, Jacobson, eds., 1974: 803-814). -. 1972. Naming and necessity. In D. Davidson, G. Harman, eds., Semantics of natural language 253-335. Dordrecht: Reidel. Lewis, David. 1978. Truth in fiction. American Philosophical Quarterly 15: 37-46. -. 1986. On the plurality of worlds. Oxford: Basil Blackwell. Linsky, Leonard. 1962. Reference and referents. In: C. E. Caton, ed., Philosophy and ordinary language: 74-89. Urbana: University of Illinois Press (quoted from: E. D. Klemke, ed., Essays on Bertrand Russell. 1970: 220-235. Urbana-Chicago-London: University of Illinois Press). Martinez-Bonati, Felix. 1981. Fictive discourse and the structures of literature: A phenomenological approach. Ithaca: Cornell University Press. Morrissette, Bruce. 1975. The novels of Robbe-Grillet. Ithaca-London: Cornell University Press. (French original, 1963). Parsons, Terence. 1980. Nonexistent objects. New Haven-London: Yale University Press. Pavel, Thomas G. 1975/76. Possible worlds in literary semantics. The Journal of Aesthetics and Art Criticism 34: 165-176. -. 1980. Narrative domains. Poetics Today 1: 105-114. -. 1983. Incomplete worlds, ritual emotions. Philosophy and Literature 7: 48-58. Pelc, Jerzy. 1977. Some semiotic considerations concerning intensional expressions and intentional objects. Logique et Analyse 79: 244-267. Pollard, D. E. B. 1973. Fiction and semantics. Ratio 15: 57-73. Plantinga, Alvin. 1977. Transworld identity or worldbound individuals. In: Stephen P. Schwartz, ed., Naming, necessity and natural kinds: 245-266. Ithaca, N. Y.: Cornell University Press. Ricardou, Jean. 1973. Le nouveau roman. Paris: Seuil. Ricoeur, Paul. 1983; 1985. Temps et redt. Vol. 1, 2. Paris: Seuil (English trans, of vol. 1, 1984).

242

L. Dolezel

Riffaterre, Michael. 1983. Text production. New York: Columbia University Press (French original, 1979). Russell, Bertrand. 1919. Introduction to mathematical philosophy. London: George Allen & Unwin, New York: The Macmillan Co. -. 19372. A critical exposition of the philosophy of Leibniz. London: Allen & Unwin. (1st ed., 1900). -. 1970. An inquiry into meaning and truth. Harmondsworth: Penguin Books Ltd. (first published in 1940). Ryan, Marie-Laure. 1984. Fiction as a logical, ontological and illocutionary issue. Style 18:121-139. de Saussure, Ferdinand. 1916. Cours de linguistique generale. Lausanne-Paris: Payot (quoted from: Tulio de Mauro, ed., 1972. Paris: Payot) (English trans, by Wade Baskin, London: Owen, 1960). Savan, David. 1983. Toward a refutation of semiotic idealism. Recherches semiouques/Semiotic inquiry 3: 1-8. Schirn, Matthias, ed. 1976. Studien zu Frege. Studies on Frege. Vol. 1-3. Stuttgart-Bad Cannstatt: Frommann-Holzboog. Schmidt, S. J. 1984. The fiction is that reality exists. Poetics Today 5: 253-274. Searle, John R. 1975. The logical status of fictional discourse. New Literary History. 6: 319-332. -. 1979. Expression and meaning. Studies in the theory of speech acts. Cambridge etc.: Cambridge University Press. Slinin, J. A. 1967. Teorija modalnostej v sovremennoj logike. In: P. V. Tavanec, ed., Logiceskaja semantika i modatnaja logika: 119-147. Moskva: Nauka. Sluga, Hans D. 1980. Gottlob Frege. London-Boston-Henley: Routledge & Kegan Paul. Spariosu, Mihai, ed. 1984. Mimesis in contemporary theory. Vol. 1: The literary and philosophical debate. Amsterdam-Philadelphia: Benjamins. Stalnaker, Robert C. 1976. Possible worlds. Nous 10: 65-75. -. 1984. Inquiry. Cambridge, Mass.-London: The ΜΓΓ Press. Sturrock, John. 1969. The French new novel. Claude Simon, Michel Butor, Alain Robhe-Grillet. London-New York-Toronto: Oxford University Press. Urmson, J. O. 1977. Performative utterances. In: Peter A. French et al., eds., Contemporary perspectives in the philosophy of language: 260-267. Minneapolis: University of Minnesota Press. Walton, Kendall L. 1978/79. How remote are fictional worlds from the real world? The Journal of Aesthetics and Art Criticism 37: 11-23. Wolterstorff, Nicholas. 1980. Works and worlds of art. Oxford: Clarendon Press. Woods, John H. 1974. The logic of fiction. The Hague: Mouton. Woolhouse, R. S., ed. 1981. Leibniz: metaphysics and philosophy of science. Oxford: Oxford University Press. Zabeeh, Farhang, E. D. Klemke, Arthur Jacobson, eds. 1974. Readings in semantics. UrbanaChicago-London: University of Illinois Press.

NICHOLAS WOLTERSTORFF

Discussion of Lubomir Dolezel's paper "Possible Worlds and Literary Fictions"

Fiction-making, says Prof. Dolezel, is a many-sided cultural activity; and unless those who discuss it keep this in mind, distortion and artificiality, if not downright error, will ensue. To this I would only add that fiction-making is an example of that special kind of cultural activity which is a social practice. Fiction-making is one of those long-enduring cooperative activities into which new practitioners are inducted, and which endure amidst the exploration of new possibilities and the institution of profound changes in goals and standards. Being the complex social practice that it is, fiction-making straddles many of the current demarcations of the academic disciplines. Theorists from many different disciplines find it appropriate to bring fiction into their discourse. Philosophers, for example, discuss fiction both when formulating general theories of reference and when constructing general ontologies. But, as Prof. Dolezel emphasizes, if we in the academic community wish to do more than explore fiction just in so far as it touches our own disciplines, if we wish also to understand fiction-making in its own right, then at some point the contributions from various disciplines will have to be brought together. In such bringing together we will discover that the contributions cannot be utilized in quite the form in which they are handed on to us. Revisions will be required in all directions if a unified general theory is to be constructed. In his paper, Prof. Dolezel does not attack that large project of constructing a general unified theory of fiction. He does, though, begin the important process of bringing together the recent work on fiction of philosophers and of texttheorists, allowing each to fructify the other. The core of his paper consists of staking out a position on the essence of fiction and on the ontology of fiction. Toward the end he points out the relevance of his conclusions on these matters to the work and concerns of text-theorists. Yet the center of gravity in his paper is in the ontology. My comments will mirror his emphasis. And let me say at the beginning that at a deep level my thought goes along the same lines as that of Prof. Dolezel. Central to his proposal is that we discard what he calls the One-world model frame' for fiction and see the worlds of fiction as something other than the actual world. I myself, in my reflections on fiction, have given pride of place to the fictional projection of worlds and to the worlds fictionally projected; and I have argued

244

N. Wolterstorff

that these worlds are rarely, if ever, to be identified with the actual world. To understand fiction, and indeed, much of art in general, we must "avoid getting hung up on actuality."1 Let us begin with Prof. Dolezel's central suggestions, that the distinguishing mark of fiction is to be found in the nature of fictional worlds; and that a fictional world can be defined as a "macrostructure of compossible fictional particulars" (p. 232). In his discussion he makes clear that in saying this he wishes to be understood as affirming that fictional worlds are merely possible, not actual; and that they are indeed possible, not impossible. Consider, then, the following very brief and admittedly dull piece of fiction: "Once upon a time two boys, disobeying their parents, went walking in a woods and got lost." The world of this work seems indeed to be a possible state of affairs. But more than that: It seems to be an actual and not merely a possible state of affairs. For surely at some time two boys, disobeying their parents, did go walking in a woods and got lost. I said that the world of this work seems to be actual. Prof. Dolezel would insist that in fact it is not; for if it be granted that the worlds of works of fiction are sometimes actual, a different way of delineating fiction from non-fiction would have to be found from that which he espouses. For him, possible but non-actual is a necessary feature of fictional worlds. It is worth noting, however, that his theory as to the ontology of fictional characters, plus his theory as to the role of fictional characters in fictional worlds, yields severe difficulties for this view. As to the nature of fictional characters, his view is that they are non-existent but possible particulars, brought into being by the composition of a suitable text or narration.2. And as to the role of such entities in fictional worlds, he says that "fictional particulars are necessary and indispensable constituents of literary fiction" (p. 228). Now presumably the fictional particulars which are constituents of the fictional world of my story include the two boys. But if so, then the following question comes to mind: How can a non-existent entity created by the composition of a text be a boy who disobeys his parents and goes walking with another boy in a woods, thereupon together getting lost? Have not all disobedient woods-walking boys to be found throughout history existed at the time they were disobedient woods-walking boys? And is that not necessarily so? It would Wolterstorff, Nicholas. 1980. Works and Worlds of Art. Oxford: Oxford University Press. p. 107. There are some passages in his text which, by themselves, would lead one to think that Prof. Dolezel's theory is not that the fiction-maker brings into being, or creates, his characters; but rather that his fiction-making turns entities already in being into entities having the property of being characters in fiction. On p. 235 he says, for example, that "a special operation is needed to transmute non-actual possible into fictional entities, to assign fictional existence to possible worlds." And again, on p. 237, he says, "A non-actualized possible state of affairs becomes a fictional existent by being authenticated in a felicitously uttered literary speech act." Everything he says between these two brief passages leads me to think, however, that this is not really his view. In any case, very little in my discussion would have to be changed if I am wrong about this.

Discussion of Lubomir Dolezel's paper

245

appear, then, that if characters are indeed what Prof. Dolezel suggests they are, non-existent artist-created particulars, and if the characters of my work's world include two disobedient woods-walking boys who get lost, then the world of my work is impossible. The same thing holds, for similar reasons, for every other work's world containing fictional entities. But whatever is to be said for or against this view in its own right, it contradicts what Prof. Dolezel argues for emphatically at the end of his paper. He says that "Literature offers the means for constructing impossible worlds, but at the price of frustrating the whole enterprise: the fictional existence of impossible worlds cannot be made authentic" (p. 239). Have I perhaps allowed something to go amiss here? Prof. Dolezel's theory of fictional characters is that, though they are non-existent, yet they are possible. It might seem that I have ignored that part of his theory. For a crucial step in my argument was that it is impossible for a non-existent artist-created entity to be a boy of the sort in question. Now of course an entity which is non-existent cannot at the same time exist. But what, it may be asked, is impossible about a non-existent entity coming into existence? Part of the answer is that my argument did not depend on assuming that a non-existent entity cannot come into existence. My argument was rather that an entity when non-existent cannot have the property of being a boy, of disobeying his parents, of walking in a woods, of getting lost, etc. Such properties, I was assuming, imply existence. Accordingly, if we are to understand the situation projected in my work of fiction as that of two non-existent entities having those properties, then the world of my work is impossible. But though I did not assume that non-existent artist-created particulars could not come into existence, yet it is worth looking at the suggestion that they could. What would the coming into existence of such a being belike? If sometime in the future there existed two boys who disobeyed their parents and walked in a woods, thereupon becoming lost, would that be the coming into existence of the characters of my work? It seems absurd to say so. But if that would not do it, what would? Furthermore, as Prof. Dolezel notes, fictional particulars, if there are any, will be ontologically incomplete. Now though I shall not here argue the point, it would seem that each such entity's pattern of incompleteness will be essential to it. But if so, then we have another reason for thinking that nonexistent entities cannot become existent entities. For necessarily every existent entity is ontologically complete: every property is such that that entity either has or lacks it. And an entity which is necessarily incomplete cannot become complete. Suppose we make one more attempt to overcome the difficulties which have been rearing up. I assumed that the property, say, of walking in a woods, implied existence. And then, on the additional assumption that that is one of the properties which belongs to one of the characters in my story, I drew the conclusion that, on Prof. Dolezel's theory of characters, my story projects an impossible world. But suppose we give up that latter assumption. Perhaps the

246

N. Wolterstorff

property that my narration gives to the characters in its world is not that ordinary, existence-implying, property of walking in a woods. Perhaps it is a property which a thing can have without existing - call it simply, the property of walking-in-a-woods-without-existing. And so also for all the other properties which gave trouble. Now if we were to develop this theory we would, of course, want to explain what such a property might be. But here let us just notice that though, on this proposal, the world of my story is not impossible, it is also not merely possible. It is in fact actual. For granting that there are such entities as fictional particulars, two of those which are constituents of my story's world will actually have the property in question. Of course, the world of my story will not be actual by virtue of what transpires with boys and parents and woods in the actual world. Every fictional world, with the exception of those which assign contradictory predicates to characters, will be actual. All the fictional entities of a work's world will actually have the existencefree properties which it is created as having. Obviously I have not canvassed all the possible moves here. Nonetheless the conclusion must be, I think, that given Prof. Dolezel's theory of fictional characters, and given his theory as to the place of characters in fiction, his central insight, that fiction deals with possibilities, proves extremely elusive. It repeatedly slips away. Let us then explore briefly those two troublesome assumptions: That there is no fiction without fictional characters - or more broadly, without fictional entities; and that fictional entities are non-existent artist-created particulars. Allow me to offer a second piece of fiction, not quite so short and perhaps also not quite so dull as the first: "The day after Ronald Reagan's first inauguration as President of the United States, he invited Jimmy Carter to the White House to get his advice on how best to incorporate concern for human rights into U. S. foreign policy." This piece of fiction seems to me to be about existent entities: Reagan, Carter, the White House, Reagan's inauguration, the United States. Just as we can imagine things that don't exist, so too we can imagine the things that do exist as being somewhat different from how they are and as acting somewhat differently from how they do. And it seems to me that at least some of such imaginings yield fiction. I know, of course, that the canonical view in 20th century literary theory is that existent particulars do not enter in the way I have suggested into worlds of fiction. The modern tradition says that we must distinguish the real Reagan from Reagan-in-my-fiction, and that only the latter enters my fiction's world. Prof. Dolezel accepts that tradition. He speaks of the "sovereignty" of fictional worlds and of the "boundary" between them and the actual world. But neither he nor the tradition gives what, in my judgment, is a cogent reason for claiming that the entities of fiction cannot be existent entities. If I can speak falsely, and even lyingly, about Reagan, claiming things to be true of him which are not, why cannot I also imagine him to be other than he is ? And if I can do that, what is to stop me from fictionalizing about Reagan? Fiction does not require fictional entities.

Discussion of Lubomir Dolezel's paper

247

But even if there can &e fiction without fictional entities, most fiction is not like that. So we must move on to Prof. Dolezel's other thesis, that fictional entities are non-existent particulars created by fiction-makers. We have seen that such entities are not possible; they are impossible of existing. But to have seen that is not to have seen that there are no such entities. Here I shall forego trying to show that. I shall confine myself to making some comments about the line of thought which, so far as I can tell, leads Prof. Dolezel to hold that there are such entities. We can usefully begin with his suggestion that fictional worlds are sets of possible states of affairs. With this I agree, provided that we add "or impossible": possible or impossible states of affairs. The world of Pirandello's Six Characters is impossible. I am not persuaded that that is a reason for concluding that it is not fiction. Now when it comes to states of affairs, we can distinguish between their occurring and their not occurring. This distinction is to be distinguished from that between their existing and their not existing. An existent state of affairs may either occur or not. To see this, it helps to think of the counterpart situation for propositions - these being, in my view, identical with states of affairs, but if not that, then at least standing in one-to-one correspondence with states of affairs. Of the propositions that exist, some are true and some are false. In the same way, of the states of affairs that exist, some occur and some do not. Reagan's asking Carter for advice about human rights policy is a state of affairs that did not occur and is not occurring. Now to particulars, this distinction between occurring and not occurring does not apply. Particulars such as tables, ducks, and persons, do not occur. However, in die philosophical tradition a different distinction has sometimes been proposed for such entities, namely, between being and existence: Some philosophers have held that, within a given category, there can be non-existent as well as existent entities - non-existent persons as well as existent persons, for example. Prof. Dolezel falls within this tradition. For he holds that the characters of fiction are non-existent particulars and that there are such particulars. So far as I can tell, however, his reason for holding this is that, after rightly seeing that fiction deals mainly with possibilities, he then fails to mark the difference between the occurring or not occurring of (existent) states of affairs, and the existing or not existing of those particulars that have being. He assumes that a commitment to possible non-actual states of affairs requires a commitment to non-existent entities. He moves back and forth, without break, between speaking of states of affairs as possible and speaking of particulars as possible. But a possible state of affairs is one which is possible or occurring. A (merely) possible individual would have to be a non-existent individual which is possible of existing. Not only is the occurrence and non-occurrence of existent states of affairs to be distinguished from the existence and non-existence of beings. What is important for our purposes is that, without any difficulty whatever, one can hold that there exist non-actual states of affairs, be they possible or impossible,

248

N. Wolterstorff

without granting that there are any non-existent particulars. Thus in principle one can preserve Prof. Dolezel's central insight, that fiction deals mainly with possibilities, without adopting his theory of characters as non-existent particulars. Indeed, we saw that his theory of characters is in severe tension with that central insight. We can hold that the maker of fiction presents to us for our consideration certain non-actual states of affairs without holding that he either creates or describes non-existent particulars. Of course, the world of a fictional work is rarely if ever a possible world. Possible worlds are just too large. If a possible state of affairs S is to be a possible world, it must be maximal; that is, for every state of affairs whatsoever, S must either include or preclude it. Few if any worlds of fiction are like that. But if one holds that there exist non-actual states of affairs and that it is mainly such states of affairs that constitute the worlds of fiction, and then goes on to hold that there are no non-existent entities, what would a theory of characters look like? In my own work I have tried to answer that question.3 Essentially what I have done is elaborate Aristotle's suggestion that characters are types not persons of a certain type but person-types. The character Hamlet is not a non-existent particular, but a type. And such types, so I hold, exist, whether or not they have examples. Where Prof. Dolezel sees the fiction-maker as creating a non-existent particular, I see her as delineating for us a certain type. We began our discussion by citing Prof. Dolezel's thesis as to the essence of fiction. We have now seen several reasons for thinking that his suggestion will not do. His thesis was that the essence of fiction is to be found in a peculiarity of the worlds of fiction: Fictional worlds are macrostructures of compossible fictional particulars. I have suggested that a fictional world need not be merely possible; it may be actual. It need not even be possible; it may be impossible. Further, it may have no fictional entities at all as constituents - nothing but real existent particulars. And the fictional entities it does have as constituents will not be particulars but types. So what then makes something fiction? My own answer is that the essence of fiction lies not in any peculiarity in the world projected but rather in the mode in which it is projected. What makes something fiction is that a world is projected in the fictional mode. The essence of fiction is to be located in the nature of the speech acts performed when telling or writing a narration. What must by all means be preserved from Prof. Dolezel's thesis, however, is the suggestion mat the worlds of works of fiction are structured - structured, indeedjin a wide variety of significantly different ways. And surely he is right in suggesting that this is one of the points where thinking of fiction as the projection of states of affairs distinct from the actual world intersects significantly with literary theory. Much of literary theory, it becomes clear, is concerned to

Wolterstorff, Nicholas. Op.cit., Part Three, Chapters VI-XI.

Discussion of Lubomir Dolezel's paper

249

distinguish significantly different structures in the worlds of fiction. And let it be added that even impossible worlds are not without significant structure. Further, even though I regard a fictional text as used to delineate types rather than to create particulars, nonetheless I agree with him that the proper interpretation of text is the means whereby the rest of us gain access to those worlds. Accordingly, what text-theorists have to tell us about the working of texts also intersects with our reflections on ontology. Let me add just one thing: The literary significance of texts is not exhausted by their projection of worlds. Texts have supra-story significance. Text-theorists have more to tell us than how worlds are projected and apprehended.

THOMAS PAVEL

Fictional Worlds and the Economy of the Imaginary One of the most attractive features of the theory of fiction developed by Lubomir Dolezel in his thought-provoking paper is the flexibility of its central notion. Possible worlds, Dolezel tells us, are unlimited in number and maximally varied; in addition they can be incomplete, semantically unhomogeneous, even erratic or impossible. Their versatility precludes essentialist definitions. Fictional worlds are not so-and-so, once and for all. But the avoidance of essentialism does not impose upon us the opposite choice, namely historicism. The flexibility of fictional worlds does not entail that they evolve homogeneously according to the laws of history. Is cultural change always consistent with historical dialectics? Does it not instead often originate in structural shifts within paradigms, as structuralist thinkers have argued? Change in fictional worlds may sometimes be attributed to external causes, such as the loss of belief in a well-organized mythology; sometimes to internal pressures, like the rise of fantastic literature during the Romantic era; sometimes just to the restlessness of taste, to the fickle moods of the Artworld. Moreover, it is far from assured that historical shifts in one location on the map of the imaginary spread with sufficient rapidity and uniformity to count as genuine transformation. Older forms coexist with, and sometimes outlive, new adventures; we all still come into contact with folktales, mythological constructs, or the classics. Opinions and tastes are divided; depending on who is speaking, the nouveau roman is by now a forgotten anomaly, or, on the contrary, the realist novel and its consistent characters have long since vanished. To be sure, one can always dismiss philistine literature and commercial criteria; but, in as far as it fosters arbitrariness and parochialism, would not such a move be self-defeating? Moreover, the suggestion that the tensions between advanced trends and more traditional imaginary worlds is attributable to the peculiar wretchedness of our time does not withstand scrutiny. Other periods nave experienced similar tensions, since the referential frameworks posited by literary fiction do not necessarily function by virtue of the ontological structure attributed to the actual world, and the ontologies of fiction enter into complex, conflictual relationships with the actual ones. The question thus arises whether it is possible to retain the notion of history of the imagination without either subscribing to a deterministic view of history, or to an essentialist conception of fiction.

Fictional Worlds and the Economy of the Imaginary

251

Fictional Landscapes Towards the end of the eighteenth century, a new cosmology was reasonably well established in the European scientific milieu, and was constantly conquering new social territories. Earlier, Fontenelle had described the pleasure and amazement with which a narrow French elite came in touch with the recent astronomy. Christian cosmology was far from dead, however. At the very end of the century, the unprecedented success of Haydn's oratorio The Creation, an enthusiastic, open celebration of the old cosmology, cannot be attributed simply to the beauty of the music. In defiance of the musical conventions of the period, The Creation relates the music to the libretto in an almost literal way. Should one think that the public was enjoying the oratorio despite its cosmology, or because of it? At least some of the contemporary admirers of Haydn must have heard of the new theories about the planets moving around the sun; they must have been peripherally aware of an innovative cosmology, while keeping their central commitments to an entirely different one, similar to a child who, at the age when it becomes more and more obvious that Santa Claus does not exist, still clings to his old beliefs, while marginally sensing that his convictions may be obsolete. This situation instantiates a remarkable property of ontological systems, namely the fact that they rarely command an unqualified loyalty. Often, objects belong to two different sets of worlds and have different properties, functions and ontological weight in each of these. The worlds in which the individual called Jesus, the pole of the Kwakiutl worship house, the grotto at Lourdes exist are clearly distinct from the worlds inhabited by the son of God, the Pole of the World, and the Holy Virgin. This is so not only for the skeptic and the materialist, but also for the believer, who often perceives the world as profane in its texture, but sanctified by the epiphany of the holy. Sacred being and objects, miraculous or prophetic grottos, holy mountains, places of worship, all these provide for the points of articulation at which two worlds meet in what could be called a series of ontological fusions. Fusions can conceivably be complete, in the sense that all the entities at the profane level play a role at the sacred level. Within the esoteric doctrines that teach universal symbolism, virtually every object belonging to the literal world has a place in the ontological framework of a secondary, symbolic world. The task of the wise is to decipher, for those objects perceived in the world of appearances, their place in the hidden, but all the more real, world. Pushed to their extreme consequences, all major religions contain projects of complete ontological fusion: does not the presence of the holy convert the entire universe by attributing to each of its parts a religious meaning? Similarly, scientific projects are often based on complete fusions: atomic physics posits an invisible level of reality coextensive with the world of everyday experience, but structurally different from it. The manifestation of the sacred does not, however, always assume a pancosmic character. Most social organizations tend to limit the expansion of the

252

Th. Pavel

sacred. When they succeed, the contact points between the two worlds are restricted to a well defined set of elements: the sacred spaces (temples, houses of worship, places of sacrifice), ritual objects, periods of celebration (ceremonies, feasts, festivals). The remaining space and time, the non-holy objects and activities obey only the laws and constraints of the profane world. Consequently, the world view of a given community may divide into several ontological landscapes. European society at the end of the eighteenth century was still keeping the Christian element as an essential component of its ontological territory. This territory was, however, much wider than the Christian world: even among those not primarily interested in the progress of science, the rumor would circulate that new and disturbing cosmological theories were being proposed. We might safely assume that most of Haydn's admirers never managed to study the new cosmology closely. However, an educated person could have planned to acquaint himself with the new system, the existence of which, together with a few of its general features, was common knowledge. Yet, in spite of the new territories, it was still possible to celebrate the beauty of the land that stood for centuries at the center of civilization: hence the enthusiasm with which Haydn's Creation was received. The oratorio must have been perceived as a magnificent opportunity to explore the old world, so beautiful, so close at hand, so reassuring. Conversely, one may easily think of a population which, after having lived for a long time in the shadow of an ontological fortress, starts to spread out onto a much wider surface. The steep rock, by now uninhabitable, on which the founders had built their impregnable castle, is declared a historic park and is used only as a tourist attraction. This does not prevent the castle from serving as a center and emblem for the expanding region: geocentric cosmology does not play a different role in our time. Ontological landscapes foster the plurality of worlds; freedom of choice nonetheless appears to be subject to certain constraints, since people have deep ontological commitments and a stable feeling that they live in an ontologically coherent world. If most societies seem to accomodate, or at least to authorize some diversity in the ontological landscape, there still remain means to indicate that only one of these landscapes represents the world proper, or central. Competition between neighbouring landscapes always leads to a process of ontological focalization, to a sorting out and an ordering of the worlds in place. The most conspicuous world-model may then play the role of the absolute norm, of a high court that summons neighbouring models for control and justification. In communities that adopt a central model, while still keeping other peripheral landscapes, the chosen model would serve as ultimate truth and regulating principle for the remaining versions: therefore, in case of conflict, the peripheral models have to yield. In typical European villages, for instance, popular beliefs more archaic than Christian cosmology, faith in local spirits, witchcraft, etc., coexisted with the new system for long periods. True, tolerance did not always mean freedom: accepted at the periphery, the slightest danger of

Fictional Worlds and the Economy of the Imaginary

253

expansion or conflict led to the exclusion of the archaic beliefs. If, therefore, orthodoxy imposes a dominant model, it is in order to protect a certain ontological focalization, a certain ordering of coexistent landscapes. For, observing the large number of beliefs and heresies condemned in the name of one set of convictions or another, one cannot avoid the impression that world models are in a continuous state of fermentation, change, and degradation - in a permanent movement, against which the dogmatic reinforcement of a certain order may be the best defense. Thus, rigidity of religious beliefs and more generally dogmatism of any sort could have their origin in the need for stability and for occasional normalization of the ontological landscape. Such enforcement is obtained through a few recurrent patterns of landscape organization. Models that occupy the central area of the landscape may vary between two extremes: complete fusions versus flat literal universes. A complete fusion is a double ontological structure in which every element at one level plays a role at each other level as well (the converse need not be true). A flat or literal universe is a single level construction, assumed to contain without residue all and only what there is. Both fusions and literal universes may in turn be strong or weak. A strong literal model is assumed to be the only faithful representation of all and only what there is, as happens in the case of early physicalism. A weak literal model may coexist with other weak literal versions, or even with other fusions; thus, a more tolerant philosophy of science accepts more than one literal representation of what there is; it can even accomodate, on occasion, non-literal representations of the world. At the other end of the spectrum, with strong fusions, the sacred world tends to cover the whole surface of the profane. In weak fusions, the two levels make contact with one another only selectively. The educated Christian of the end of the nineteenth century believed in a universe governed by natural laws; restricted to a few isolated points, the holy was surrounded by an entirely foreign texture. Religious philosophy solved this encirclement either by withdrawing into the existential moral field (since, according to this choice, the true place of the sacred is in the interiority of the subject), or by positing a sacred center of the world, unaffected by the progress of the profane. In both cases, in the heart of the literal we face the epiphany of a metaphor: the unseen, donor of meaning. At the margins of ontological landscapes, one finds leisure worlds, or worlds for pleasure, which often derive from older, discarded models. Each culture has its ontological ruins, its historical parks, where the members of the community relax and contemplate their ontological relics. Greek and Roman gods performed this function till late in the history of European culture. Or else, marginal models may be used as training grounds for various tasks. Thus, as Umberto Eco convincingly argued, one of the functions of fiction is to cultivate abilities such as perceptual alertness, rapid induction, construction of hypotheses, positing of possible worlds, moral sophistication, linguistic proficiency, value awareness.

254

Th. Pavel

The arrangement of ontological space strikingly resembles landscape architecture and urban planing. Strong fusions recall the use of the natural landscape by communities whose life is still closer to nature. Weak fusions are not unlike nineteenth century cities, where the inhabited space was clearly separated from green spaces, to the extent that city dwellers lived and breathed in different places. Strong literal models eliminate landscape variety, similar to the futuristic city from which the vegetable realm would have been excluded. Finally, weak literal models may be compared to our contemporary chaotic cities with their heterogeneous scattered neighbourhoods, linked only by highway networks. For, similar to urban arrangements, which render spatial coexistence of conflicting activities possible, ontological planning aims at avoiding or at least appeasing the inevitable clash between opposite world-models. It rationalizes to some extent the variety of the ontological space; as a consequence, questions such as "is proposition p true or false?" can be referred to such and such an area of the general landscape. At the beginning of our century, a proposition like "Christ is a man-god", the truth of which unquestionably stood at the center of a strong medieval fusion, remains true within the space arranged for this kind of sentence, without being true everywhere in the ontological landscape, not unlike the activity of breathing fresh air, which at some point in human history was coextensive with most human activities, but became restricted for the dwellers of the modern megalopolis to certain times and places. And similarly to the social rules of behaviour prescribing special leisure activities dedicated to the periodical consumption of fresh air, in complex ontological arrangements elaborate rules of etiquette indicate in detail which propositions should be assented to in various contexts and situations. Propositions that are obviously true in Church on Sunday are less so in different contexts, such as, say, professional meetings. Like any order, ontological planning may trigger hostile reactions: one quite serious cultural condition may DC called ontological stress. Caused by difficulties of orientation among the complexities of modern ontological arrangements, this type of stress leads to the weakening of our adjustment to ontological landscapes. Its first victim was Don Quixote, unable to distinguish actuality from fiction. Ontological distinctions having since become much more subtle and complex, users of contemporary ontological arrangements have to travel between heterogeneous if not plainly hostile landscapes to which they are expected to adjust rapidly and only for short periods of time, not unlike the modern city dwellers who cross long distances between the place where they work and that where they live and relax. But our capability for ontological adaptation cannot tolerate more than a certain amount of change, and when the threshold is reached, reactions that resemble those vis-ä-vis the pressure of modern city-life manifest themselves. Such reactions vary between nihilism and nostalgia. Nihilists interpret changes from one landscape to another as signaling the complete absence of order. To them each landscape is only the deceitful ossification of one kind of illusion. To establish oneself somewhere, to dwell in one of the world-models is in the

Fictional Worlds and the Economy of the Imaginary

255

nihilist's eyes a mortal sin. Conversely, the nostalgic is homesick for old times when ontological stability was still the rule. Nostalgjcs of various persuasions, craving for the age of innocence, the age of cathedrals, Victorian order, or Vienna at the turn of our century, are fascinated by the alleged simplicity of assent these periods required. Nihilism assumes that the simultaneous presence of several world-models cancels the credibility of each model. Their multiplicity stands to prove that all are fictitious, and any choice between these fictions is made accorcung to purely utilitarian criteria. Any version is good, provided it serves a certain purpose; marginal or obsolete landscapes have the same right to attention as central versions; conversely, the centrality of the latter is only a question of convenience, or convention, hi turn, by rejecting ontological multiplicity, nostalgics, in their own way, consider the surrounding ontologies as mere fictions, at least in comparison with the overthrown dogma. I have just employed the term 'fiction' as synonymous with error. Users of world-models, however, spontaneously make distinctions between fiction and error on the one hand, and fiction and truth on the other. They also know that world-models can serve more than one series of users. A given model may lose the assent of its users without, however, being irrevocably discarded: deposed world-versions often find secondary users. Mythologies survive this way. Nelson Goodman once suggested that we should replace the question "What is art?" by "When is art?". Likewise, one might simply DC tempted to ask "When is fiction?", and offer a pragmatic answer: fiction is when world-versions find secondary users. But an answer of this kind would not cover all cases. For if fiction were only a special use of models produced elsewhere and later abandoned as obsolete, how could one explain the observation that most societies maintain some sort of non-religious fictional activities, such as the 'laughing stories' of the Cherokees, animal stories, anecdotes, folktales, etc. To derive these from older unused or degraded myths is not always easy: on the contrary, a considerable number of folktales that are based on non-mythical material may have originated in the observation of the then-current social life. They appear to have been designed independently of any other discarded world-models. Instead of defining fiction in historical terms only, as the result of myth-decay, we should perhaps characterize it in terms of ontological landscaping and planning as well. Taking the division of the ontological space into central and peripheral models as a very general formal organization of the beliefs of a community, we may localize fiction as a peripheral region used for ludic and instructional purposes. The concrete content of the marginal ontology would thus count less than its position within the functional organization; and indeed fictional space can accomodate almost any ontological construction. This suggests that fiction is both a pragmatic and a semantic notion, since the organization of cosmological space obeys pragmatic reasons, but the structure itself is clearly semantic. Also, the choice of the particular models that fill the fictional

256

Th. Pavel

space has pragmatic undertones: do these models originate in specially constructed versions? old discarded ones? actual models used fictionally for only a short while? But the regularities of the peripheral space itself possess semantic features.

A Functionalist View of Fiction Though they strive to include history as an important component, the above suggestions are not historicist in the narrow sense of the term. They do not imply that history's course is in any way determined by material or cultural causes. Instead, the model intimates a less deterministic picture, one that resembles in some respects a strategic game, chess for instance, wherein the strategic configuration at each stage of the game depends simultaneously on the rules of the game, the configuration in force at earlier stages, and the particular strategic aims pursued, with more on less talent, by the participants. Similarly, the economy of the imaginary must obey various general constraints, in all probability more flexible than those followed by strategic games. These might include cognitive and developmental restrictions, such as the compositional constraint that requires non-empirical beings to be made up mostly of empirical components, or the ordering constraint according to which the development of the imagination follows irreversible - but not deterministic sequences: complex forms presuppose simpler ones; drama develops from ritual while the converse is not true (pace Arthaud); when myths become fiction they do so massively, through conversion of entire Pantheons, while stories grow into myths one by one and in an unobtrusive fashion, etc. The economy of the world-models in use at a certain time depends on earlier stages, as centuries of research about such dependence have proven beyond doubt. But, again as in a strategic game, the economy of the imaginary also rests on systems of purposes and expectations that, within the historical horizon provided by the already existent solutions, help delineate the new stage. Admittedly, the notions of socio-cultural purpose and expectation are far from being transparent; but perhaps it is enough to mention them in order to call attention to the teleological aspect of cultural enterprises, and challenge our reluctance to handle final causes. And if indeed entire cultures are too complex to have definable holistic purposes, at least fictional economy can be characterized as a functional system. It should be noted that a functionalist view of fiction does not entail the existence of explicit, unique and constant purposes, nor does it suggest that there are stable means of achieving such purposes. Fiction pursues families of goals with uneven and ever-changing determination, switcning emphasis from one aim to another, abandoning projects before completion, succeeding in secondary areas, just when it fails in more important endeavours, obtaining by luck what had been denied to labor. Production of fiction bears the mark of instability; this may be attributed to the multiplicity of purposes, the multiplicity of

Fictional Worlds and the Economy of the Imaginary

257

structural means of achieving them, the lack of stable dependencies between structure and aims, or to all these factors combined. Among the teleologic aspects of fictional activity, I would include the referential purposes of fiction. Reference in fiction rests on two fundamental principles which, while shared by fiction and other activities, have for a long time constituted the privileged core of the fictional order: the principle of distance and the principle of relevance. Creation of distance could well be assumed to be the most general aim of imaginary activity: the journey epitomizes the basic operations of the imagination, be it realized as dreams, ritual trance, poetic rapture, imaginary worlds, or just the confrontation of the unusual and tne memorable. Scandal, the unheard of, the unbearable tensions of everyday social and personal life, are expelled from the intimacy of the collective experience and set up at a distance, clearly visible, their virulence exorcized by exposure to the public eye, by the safety net of exemplary distance. Sagas and epic poems offer the paradigms for the literary framing of real characters and happenings. But this semantic pattern can be extended to all fictions. Symbolic distance is meant to heal wounds carved with equal strength by unbearable splendour and monstrosity in the social tissue. But the cure cannot work unless it is somehow shown to pertain to actuality. Symbolic distance must be complemented by a principle of relevance. Hence the Proposition Theory of artistic meaning, which, though open to criticism in its more punctilious versions, captures an important, ineuminable intuition, namely that literary artefacts often are not projected into fictional distance just to be neutrally beheld, but that they vividly bear upon the beholder's world. Their relevance does not have to be conceived according to a single model, be it that of logical entailment, conversational implicature, judicial decision or ideological presupposition. These models have a subordinate status, insecurely linked to the encompassing, emergent role of relevance. Depending on various decisions and conventions, some may be overemphasized, others absent. Medieval exempla are constructed as generalizations: since Λ; who did mischief was so horribly punished, all x's who do mischief shall share his fate. The venerable pattern, already noticeable in the most archaic folktales, valiantly survived through virtually any vicissitude of artistic evolution. It is not only Phaedra's misplaced love and its mythological punishment that impress us; The Scarlet Letter, Anna Karenina, the narrator's tragic infatuations in Remembrance of Things Past, or E. Sabato's Tunnelpress the same point. And once the generalisation is reached, a new conclusion gives these fictions vividness and appeal: since this can happen to anyone, it also can happen to me. But the relevance of fiction does not limit itself to logical conclusions and moral generalizations. Fictional texts often interest their public by just offering information on less familiar periods, regions, cultures, trades, kinds of behaviour. We do not read Moby Dick only to witness the hybris and tragic end of Captain Ahab, but also to learn about life on a whaling ship in the first half of the nineteenth century. With inexhaustible patience, the author strives to enlighten

258

Th. Pavel

his public on whaling matters, and some of us are caught up in the endless documentary passages, just as some of Balzac's readers enjoy the meticulousness of his descriptions. Most often, documentation is itself determined by the need for conspicuousness and scandal; this causes the unending flow of fictional texts on wars, holocausts, exceptional tragedy in the life of ruling classes or outcasts. Change in taste threatens entrenched solutions and favours diversity of procedures. Critics of contemporary culture have not failed to notice the trend towards a drastic reduction of fictional distance that brings fictional worlds as close as possible to the beholder. Frames and conventional borders seem to vanish, and the purpose becomes achievement of immediacy. The principle of relevance has been challenged too: many modern fictional texts playfully prevent the audience from reaching swift conclusions, moral or otherwise. Yet, since distance cannot be altogether abolished, the gap between us and the hyperrealist statues in contemporary parks will never be bridged; likewise Henry James' or Kafka's ambiguities do not signal the end of relevance; at most they may indicate a transformation in the kind of relevance aimed at. Just as the tonal simplicity of Vivaldi's compositions is gradually replaced by more sophisticated procedures, culminating in the tonally equivocal systems of Wagner, Richard Strauss, and Mahler, simple inferential practices give way to richer literary systems, and inferential relevance (Lear's pride leads him to destruction) gradually leads to diffuse relevance (Kafka's Castle as an undetermined parable of the human condition). Fictional worlds are the main repositories of structural features employed for referential purposes. They are, in most cases, related to the worlds of common sense, and bear the weight of ontological and epistemic assumptions, but they also reflect the technical sophistication of the author and his milieu, and the different purposes that the construction is meant to achieve. Cosmologies, we saw, vary between salient, multi-level organizations, and flat, literal ones. The more luxuriant worlds are closer kin to tne wealth of early mythologies, while later fictional worlds bear a notable mark of austerity. In historical societies we are perhaps entitled to see fiction as an institution destined to take over the place of myths and fairy tales. In this sense, did not literary fiction begin as a truth-project, meant to record memorable facts and to bring some credibility to the wÜderness of myths? Yet the opposite tendency, rooted in fairy tales, has not failed to take its revenge: the fantasy-project deploys itself periodically, apparently meaning to break up the compactness of flat worlds and to bring back the kaleidoscopic opulence of less constrained imagination. Romance versus epos and tragedy, fantastic literature versus realism, these were and still are cyclical fights between the two projects, with varying chances of temporary success according to the religious context (since naive religiosity tends to favour romance), to the social milieu (with idle groups preferring fantasy-projects and compulsive ones inclining towards truth), to the level of saturation with the opposing project. Thus, Quixote could have appealed to a period overfed with

Fictional Worlds and the Economy of die Imaginary

259

fantasy, while The Master and Margaret exploded in a context in which realism was the sternly enforced obligation. Variation in ontological wealth is supplemented by epistemic principles. Again in opposition to early myths, fictional production brought along a technique of editing and rationalizing the sequence of events that resulted in well-balanced stories or plays, optimally focused on small groups of states of affairs. Compared with what we know about primitive myths, Greek myths included, everyday narratives, anecdotes, and folktales display a remarkable compositional unity and a maximal concentrating of interest toward the narrative topic. But narrative discipline and compositional unity never had the power to impose themselves decisively upon fictional production. Just as the ontological severity of the truth project has periodically been mitigated by the intervention of fantasy worlds, the compositional severity of well-focused narration is periodically contradicted by dispersion, faulty arrangement, incoherence, all eminently visible in primitive myths, but reemerging in Greek novels, in chivalric novels, in the menippean and carnivalesque traditions, in many Baroque and eighteenth century novels, and again in our century in the modernist and post-modernist context. Yet, in the end, all this restlessness appears quite well-tempered, for, seen from afar, every period, no matter how diverse and sophisticated its functional arrangements, seems to have at its command a certain number of indispensible elements. In virtually every place and time we find a more or less complete thematic set, covering the main human concerns, social or existential. Birth, love, death, success and failure, authority and its loss, revolution and war, production and distribution of goods, social status and morality, the sacred and the profane, comic themes of inadequacy and isolation, compensatory fantasies, and so much more, are always present, from early myths and folktales to contemporary literature. Changes of taste, shifts of interest, seem to affect the inventory only marginally. Since we need an alien space in which to deploy the energy of the imagination, there have always been and always will be distant fictional worlds; but we may also use close fictional worlds for mimetic purposes, in order to gather relevant information, or just for the pleasure of recognition. Moreover, we find incoherent fictional worlds everywhere, worlds destined to provide a space of dizziness and playful transgression. Distance, relevance, and dizziness, mixed in various proportions, give each period its peculiar flavour, but virtually never is any of these components missing. Cultural arrangements may well attempt to stabilize these recurrent fictional components into durable consensual patterns; their perennial failure to arrest kaleidoscopic multiplicity allows the history of fiction to unfold. Yet this inevitable displacement incessantly brings forth the familiar elements, and history itself is but the ever surprising sway of colours radiated by the same few shiny splinters.

SAMUEL R. LEVIN

Duality and Deviance: Two Semantic Modes Discussion of Thomas Pavel's paper "Fictional Worlds and the Economy of the Imaginary"

A fictional world can be viewed as a projection from the actual world onto a sequence of sentences called a novel or poem. In saying this it is understood that what we mean by 'actual world' is neither a fixed nor a simple notion - how complex a notion is made clear by Pavel in his rich and illuminating discussion of the historical and intellectual currents which run through and serve to define the background against which an author's conception of the actual world - his metaphysics - is formed. Besides the cosmological aspect, there is comprised also by 'actual world' the various religious, social, political, and cultural institutions that compete for ascendancy at any given time, and as Pavel indicates, an author can form, from the varied and frequently disparate elements presented by that world, one of several metaphysical outlooks. The type and degree of fictionality projected in his novels or poems will then be a function of the author's imagination, with that outlook serving as a parameter. Granted that fictionality is a function of imagination, our immediate concern is rather with fiction, i. e. texts. In any case talk of an author's imagination reduces ultimately to examination of his language. This suggests that a useful way to complement Pavel's comprehensive discussion of the background of fictions, with its highlighting of Ontological landscapes' and the 'plurality of worlds' that those landscapes make available, would be consideration of the ways in which an author's fictions implement and exploit that plurality. In particular, I will try to show that to the various types of ontological landscape presented by Pavel the semantic systems manifested in fictional texts display an interestingly analogous character. Semantic Duality By 'semantic system' I mean the set of lexical items in a language and their associated meanings. Obviously, the semantics that a language as a whole comprises and makes available will be adequate to describe the sum of ontologies represented in the various landscapes that obtain at any given period. At the same time, however, individual authors appropriate for their purposes only

Duality and Deviance: Two Semantic Modes

261

portions of that system, and it thus becomes possible to compare with Pavel's various ontological situations the different semantic subsystems deployed in particular fictional worlds. Thus, correlated with a flat literal ontology there is the pedestrian semantics of a novel by Theodor Dreiser, a semantics so unproblematic that understanding depends on grasping only the basic, most ordinary senses of words. The authors of such monoplanar works know a limited amount of experience extremely well, and their imaginative effort consists almost entirely of conveying a sense of verisimilitude in the description of that aspect. Since that view of experience is ordinary and generally accessible, the possible world described in these 'realistic' novels is, we might say, extremely 'possible'. Opposite to the flat and literal in Pavel's classification is what he calls ontological fusion. Necessary for fusion is not simply the collateral existence of two ontologies, but their coincidence at salient points in such a way that the same element plays a relative role in each landscape. Most religious systems, Pavel suggests, may be approached in this way - as projecting their elements in a point for point fashion onto elements of the secular world. This means that each such element has dual signification, its meaning being in a characteristic sense context-dependent. Thus, in the world defined by historical process, Jesus is a carpenter of Nazareth, in the world of Christian theology he is the son of God. When the articulation between the two ontological levels is total the result, in Pavel's nomenclature, is strong fusion. We have a semantic counterpart of this ontological model in the fictional genre of allegory. Pilgrim's Progress, for example, describes the adventures of a certain Christian who seeks the way to salvation and who, in the course of his quest, passes through and is tested at such places as Vanity Fair and the Slough of Despond. Like Christian, who is both a representative Englishman in the story and symbolic also of the earnest and troubled Christian, Vanity Fair and the Slough of Despond at one and the same time represent both physical locations and trials of the spirit. As is typical in allegories, the names in Pilgrim 's Progress evince a systematic ambiguity. As such, they represent semantic nodes in the story, points at which reference forks and whereby the two ontologies are fused. If we reel on reading the allegory that the dual reference projected by the names is of such a high charge that it spreads through and permeates everything else in the story, then we have the equivalent on the semantic side of strong fusion. As a novel whose semantics reflects a weak ontological fusion one might instance One Hundred Years of Solitude by Garcia Marquez. In this novel there occur a variety of preternatural events: flowers rain from the sky, bodies are levitated, insomnia spreads like a plague, a deluge lasts for almost five years after which there is a ten-year dry spell, dead men make reappearances, and the like. All these events have it in common that they violate or infringe upon physical laws. So that, according to any metaphysics defined by conditions in the actual world, the fictional landscape depicted in Garcia Marquez' novel is 'impossible'. The novel does not, however, represent a consistently 'impossible' world. For it

262

S. R. Levin

is not as though all human beings, when they die, ascend to heaven, nor that all dead men reappear as revenants, nor that it always rains tiny yellow flowers. Nature in One Hundred Years of Solitude does not abdicate her dominion over the world; she merely takes an occasional holiday. One character in the novel remarks this fact in noticing 'a certain confusion in nature', another in observing that 'time ... stumbled and had accidents'. In general, however, what the novel reflects is not a world totally and systematically different from the world of our everyday experience; what it reflects is essentially our world, with an occasional aberration. It is at these points of 'aberration' that elements from another ontology impinge upon the essentially flat landscape that serves as the backdrop for One Hundred Years of Solitude. We have here then, in the semantics of the novel, the counterpart of a partial or weak fusion. Semantic Deviance In the semantic systems that we have discussed to this point the feature that interested us has been duality of reference. Thus, correlated with flat ontological landscapes has been a semantics that displayed no such duality, whereas in relation to fused landscapes duality has been either thoroughgoing, in the case of strong fusions, or sporadic, in fusions that are weak. I will now consider a different type of semantic feature, that of anomaly or deviance, and consider the ways in which it correlates with ontological landscapes. Moreover, because semantic deviance is a characteristic feature of metaphor, and since metaphor is associated more with poetry than with prose, I will in this section shift from the fictional landscape of novels to those of poems. As we shall see, it turns out that the approach that one takes to the semantic analysis of deviant sentences in poems is critically related to the question of ontological landscape. In Eliot's poem 'Love Song of J. Alfred Prufrock' there occurs the line 'The yellow fog that rubs its back upon the window-panes'. Now the reason that semantic deviance is of interest in a discussion involving ontological landscapes lies in the fact that a corollary of such deviance is the description or assertion of counterfactual states of affairs. In our example, it is contrary to fact that fog has a back, much less that it rubs it upon anything. In the standard interpretive approach to this line one proceeds on the assumption that the expression is not to be taken at face value, that it is, instead, to be interpreted 'metaphorically'. Such a reading of the line might then be that the fog, in its characteristic spatial disposition, simply encroaches upon everything in its environment, in the present instance, upon the window-panes. On this interpretation, what the sentence is held implicitly to describe, i. e. once its semantics is properly construed, is a state of affairs that is factually commonplace - flat; it is merely the language used to describe this state of affairs that is out of the ordinary. Now although the option does not usually come in for discussion, it is possible, theoretically, to invert this procedure; that is, we may elect to take the language

Duality and Deviance: Two Semantic Modes

263

literally (and seriously, i. e. assume that what it asserts is neither 'patently' false nor nonsensical). If we do this, then in order to arrive at an interpretation we must modify our conception of reality: we must try, that is, to conceive of the fog as animate, as having a corporeal consistency, with a front and a back, and as being able to intend rubbing its back on a window-pane. This latter approach to the interpretation of metaphoric expressions has interesting implications for any discussion of literature carried out in the context of possible worlds analysis. For its inclusion makes it possible to think of a general (chiastic) pattern of relations holding between the approach that we take to the language of the deviant expression and the type of world that we project from it. If we take the language as metaphoric and proceed to interpret it, the result is an accommodation of its meaning to actual world truth conditions; on this approach what the metaphoric expression describes, we might say, is a 'literal' world. On the other hand, if we take the language of the expression literally, then it is the world it describes, exceeding as it does anything in the range of our actual-world experience, that is 'metaphorical'. This is the 'chiastic' pattern alluded to above: metaphoric meaning/literal world, literal meaning/metaphoric world - with semantic deviance as the hinge. From the preceding we see that deviant expressions incorporate a kind of semantic ambiguity. They have one significance taken metaphorically, another taken literally. Earlier we saw that semantic duality correlates with literary symbolism (including at the limit allegory). We see now that there is a comparable correlation between semantic ambiguity and metaphor. These two semantic characteristics - duality and ambiguity - which on the surface appear to be quite similar, in fact represent quite different semantic properties and give rise to very different effects. In the first place, dual meanings function simultaneously, ambiguous meanings function alternatively: dual meanings are both/and; ambiguous meanings are either/or. If we abstract for the moment from Pavel's taxonomy of ontological landscapes and consider instead the landscapes projected in (and from) fictions, then a dual semantics projects a fused fictional landscape, whereas an ambiguous semantics projects two totally opposed and disparate such landscapes. The correlation of duality with literary symbolism is no more than what we should expect; that the type of ambiguity I have described pertains to metaphor and that it correlates inversely with types of possible world is perhaps not so obvious. The pointing out of tnis special type of ambiguity where deviant sentences are concerned should not be seen as the mere working out of a logical pattern. Election of one or the other interpretation of the ambiguous structure can have significant implications in determining the proper evaluation of 'poetic' texts. These implications can be brought out by an examination of Giambattista Vice's reflections on the language of early man. The aspect of his account that is relevant to our present interest stems from his recognition that for early man the relation between language and reality was fundamentally different from what that relation consists in for us. For us (as for those generations that succeeded

264

S. R. Levin

early man) what we think the world is like consists essentially with its actual nature. This consistency developed as man came more and more to know the properties and principles of physical reality. Our language reflects the historical process whereby fact and our understanding have been brought into conformity. The mistake of those philosophers and scholars who speculated on the origin of language lay, according to Vico, in their unexamined assumption that early man was equipped with the same capacities for understanding his world as were men of subsequent eras. All such accounts were therefore rational accounts, presupposing that early man possessed an understanding of the physical world much like the one possessed by us. This for Vico is a fundamental mistake. The 'master key' to his Science, 'which cost [him] the persistent research of almost all [his] literary life', lay in the discovery that the early gentile peoples 'were poets who spoke in poetic characters' and this 'by a demonstrated necessity of nature'. The demonstration proceeds from Vico's explanation of early man's metaphysical outlook, an outlook which sees nature as animate and abstract properties as embodied in personifications ('class concepts'). From this outlook there necessarily follows a poetry in which the language is, from our point of view, semantically deviant. To the first poets, however, this same language was entirely straightforward and was intended quite literally. If anything was 'metaphorical' it was their conception of the world, not their language. Properly to understand their 'poetry', therefore, one must take their deviant expressions literally. The pivotal role played by semantically deviant expressions can be demonstrated another way, this time as viewed in connection with Pavel's discussion of the referential function of fiction. Pavel points out that fictional reference rests on two fundamental principles, those of distance and relevance. The worlds of fiction, for all their being imaginary, must still be conceivable. Hence referential distance must be tempered by relevance. Depending on the relative weights assigned to these two principles, fictional worlds wifl tend in the direction of either fantasy or realism. Pavel's discussion considers all these factors against a general background of historical development and aesthetic purpose. These same factors, however, are applicable to the specific case of a deviant expression. Thus, when taken literally the referential function of such an expression is attenuated, the world it projects distant, 'fantastic'; conversely, if the expression is construed metaphorically, its referential function becomes more authentic, its projected world relevant, more 'realistic'. These general and theoretically interesting consequences ensue from the taking of deviance to range over worlds and not merely words. The observations made to this point about semantic deviance have been essentially synchronic in nature. The alternation of 'fantastic' with 'real' fictional landscapes derives from the ambiguous status of such expressions, where both readings are prima facie feasible, one or the other being selected on the basis of interpretive goals and strategies. It is possible, however, to approach the ambiguity of deviant expressions also from the perspective of diachrony. Of interest

Duality and Deviance: Two Semantic Modes

265

on this approach is not the synchronic alternation of a metaphoric and a literal reading, of the sort discussed earlier, but the diachronic replacement of a metaphoric by a literal reading. Metaphors fade and eventually die. Thus the same expression which was at one time deviant and susceptible of a metaphoric interpretation (its literal interpretation being left out of account in this connection), becomes in the course of time semantically naturalized or domesticated, whereupon its meaning assumes a literal application. An expression like 'lame excuse', for example, was at one time semantically deviant. (Disregarding for present purposes the 'possible worlds' interpretation of the phrase) it originally constituted a fresh metaphor and required construal. Today it is a faded, if not quite a dead, metaphor. As such, the meanings of lame' and 'excuse' have been modified to incorporate senses which make the collocation semantically compatible. Pertinent to the interpretation of deviant expressions we may therefore say is not only a semantics defined over possible worlds but also one defined over possible languages. Postscript A book that I brought along to the Conference was The Europeans by Henry James. In it a brother and sister, Felix and Eugenia Young, come to America to 'seek their fortune'. Unlike the status of her brother, which is that of an extremely attractive and eligible bachelor, Eugenia's situation is complicated by a (morganatic) marriage to the younger prince of a German principality, whose older brother, the Reigning Prince, wishes to dissolve the marriage. Eugenia is not averse to the dissolution (for which her agreement is needed), but she hopes first to improve her situation and has made the trip to America with that end in view. The potential 'improvement' is represented in the person of Robert Acton, an American of impeccable background and substantial fortune. Despite the great attraction that she holds for him, however, a proposal is not forthcoming and Eugenia, frustrated and disappointed, decides to return to Germany. After a visite d'adieux to Acton's dying mother, Eugenia leaves the old lady's bedroom and, as she descends, reflects on the scene that lies before her: The broad staircase made a great bend, and in the angle was a high window, looking westward, with a deep bench, covered with a row of flowering plants in curious old pots of blue China-ware. The yellow afternoon light came in through the flowers and flickered a little on the white wainscots. Eugenia paused a moment; the house was perfectly still, save for the ticking, somewhere, of a great clock. The lower hall stretched away at the foot of the stairs, half covered over with a large Oriental rug. Eugenia lingered a little, noticing a great many things. 'Comme c'est bienf she said to herself; such a large, solid, irreproachable basis of existence the place seemed to her to indicate. And then she reflected that Mrs Acton was soon to withdraw from it. The reflexion accompanied her the rest of the way downstairs, where she paused again, making more observations. The hall was extremely broad, and on either side of the front door was a wide, deeply-set window, which threw the shadows of everything back into the house. There were high-backed chairs along the wall and big Eastern vases upon tables, and, on either side, a large cabinet with a glass front and little curiosities within, dimly gleaming. The doors were

266

S. R. Levin

open - into the darkened parlour, the library, the dining-room. All these rooms seemed empty. Eugenia passed along, and stopped a moment on the threshold of each. 'Comme c'est bien? she murmured again; she had thought of just such a house as this when she decided to come to America.

'Comme c'est bien': the phrase recalls, for someone who has read the Njalssaga, the scene where Gunnar, outlawed and sentenced to exile, turns for a last look at his homestead, Hlictarendi, and says, Tpgr er hlidin, svä at mer hefit aldri jam f