International Institute of Philosophy

Institut International de Philosophie

La philo sophie contemporaine
Chroniques nouvelles

par ies soins de

GUTTORM F~ISTAD
Universite d'Oslo

Tome 1
Philosophie du langage
(Editeur G. Fl¢istad)

*
Logique philosophique
(Co-editeur G.H. von Wright)

Martinus Nijhoff Publishers DordrechtiBoston/Lancaster 1986

Contemporary philosophy
A new survey

edited by
GUTTORM FI.q)ISTAD
University of Oslo

Volume 1
Philosophy of language
(Editor G. Flq,istad)

*
Philosophical logic
(Co-editor C.H. von Wright)

Martinus Nijhoff Publishers DordrechtiBoston/Lancaster 1986

Distributors:

for the United States and Canada

Kluwer Boston, Inc.

190 Old Derby Street
Hingham, MA 02043
USA

for all other countries

Kluwer Academic Publishers Group

Distribution Center
P.O. Box 322
3300 AH Dordrecht
The Netherlands

Library of Congress Catalog Card Number: 81-3972

ISBN-13: 978-90-247-3297-5 e-ISBN-13: 978-94-009-8356-4

001: 10.1007/978-94-009-8356-4

First edition 1981

Second printing 1986

Published under the auspices of the International Council of Philosophy and Humanistic Studies
and of the International Federation of Philosophical Societies, with the support of UNESCO.

Publie sous Ies auspices du Conseil International de la Philosophie et des Sciences Humaines et de
la Federation Internationale des Societes de Philopsophie, avec Ie concours de l'UNESCO.

Copyright @ 1981 by Martinus Nijhoff Publishers, Dordrechr.

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or
transmitted in any form or by any means, mechanical, photocopying, recording, or otherwise.
without the written permission of the publisher.
Martinus Nijhoff Publishers, P. O. Box 163, 3300 AD Dordrechr,
Contents/Table des matieres

G. FI¢istad, Preface VII

PART ONE/PREMIERE PARTIE

Philosophy of language/Philosophie du langage

G. FI¢istad, Introduction 3
J. Pelc, The place of the philosophy of language 11
C. Peacocke, The theory of meaning in analytical philosophy 35
J. Hintikka, Semantics: A revolt against Frege 57
J. Bouveresse, Wittgenstein et la philosophie du langage 83
N. Cocchiarella, Richard Montague and the logical analysis of
language 113
F. Kambartel and H.1. Schneider, ConstructiHg a pragmatic
foundation for semantics 155
P. Ricoeur, 'Logique hermeneutique'? 179

PART TWO/DEUXIEME PARTIE

Philosophicallogic/Logique philosophique

G.H. von Wright, Introduction 227

VI Contents

D. Prawitz, Philosophical aspects of proof theory 235

R. Barcan Marcus, Modal logic, modal semantics and their
applications 279
R. Hilpinen, Conditionals and possible worlds 299
N.D. Belnap, Jr. and J.M. Dunn, Entailment and the
disjunctive syllogism 337
R. Jeffrey, Choice, chance, and credence 367

Abbreviations used by some contributors 387

Index of names 389

Index of subjects 394

Preface

The present publication is a continuation of two earlier series of

chronicles, Philosophy in the Mid-Century (Firenze 1958/59) and
Contemporary Philosophy (Firenze 1968), edited by Raymond
KJibansky. As with the earlier series the present chronicles purport
to give a survey of significant trends in contemporary philosophi-
cal discussion. The time space covered by the present series is
(approximately) 1966-1978.
The need for such surveys has, I believe, increased rather than
decreased over the last years. The philosophical scene appears, for
various reasons, more complex than ever before. The continuing
process of specialization in most branches, the emergence of new
schools of thought, particularly in philosophical logic and the
philosophy of language, the convergence of interest (though not
necessarily of opinion) of different traditions upon certain prob-
lems, and the increasing attention being paid to the history of
philosophy in discussions of contemporary problems are the most
important contributory factors. Surveys of the present kind are a
valuable source of knowledge of this complexity and may as such
be an assistance in renewing the understanding of one's own
philosophical problems. The surveys, it is to be hoped, may also
help to strengthen the Socratic element of modem philosophy, the
dialogue or Kommu nikationsgemeinschajt.
So far, four volumes have been prepared for the new series. The
present chronicles in the Philosophy of Language and Philosophi-
cal Logic (Vol. I), are followed by chronicles in the Philosophy of
Science (Vol. II), and Philosophy of Action (Vol. III), and Phil-
osophy of Mind (Vol. IV). Each volume contains, as a rule, fifteen
chronicles each 25 pages long. However, this rule has been broken
VIII Preface

in all volumes. In some cases, as in the present volume, it turned

out to be impossible to receive chronicles in time for the date of
publication. In others the authors, if they thought it necessary in
view of their topic, were allowed to exceed the allotted number of
pages. For these reasons certain philosophical disciplines, par-
ticularly the Philosophical Logic of Volume I, are covered less
satisfactorily than others, apart from the fact that the volumes are
of unequal length.
Most of the chronicles, as to be expected, are written in English,
some in French and none in German. The German contributors
thought it necessary to write in English in order to be read. This
is, I think, a most regrettable state of affairs. It means that major
parts of the philosophical community will no longer have access to
important sources of the history of philosophy in their original
version.
The topics surveyed in the four volumes belong squarely within
the Western philosophical tradition and do little justice to philos-
ophies or ways of thinking in other cultures. This should be kept
in mind in the preparation of further volumes. The idea of trans-
cultural philosophy is still very unclear.
Bibliographical references, with a few exceptions, follow the
pattern introduced in Philosophy in the Mid-Century. The publi-
cations discussed in each chronicle are indicated by reference
numbers in square brackets, corresponding to the order in which
they appear in the text. The bibliographies themselves follow at
the end of each chronicle arranged in alphabetical order.
The bibliographies are selected by the authors themselves and
contain as a rule only those works they took to be of special
interest for the topics under discussion.
I am most grateful to a n umber of persons, who in various ways
have assisted in the preparation of the new series. My thanks are
first of all due to the former president of the Institut International
de Philosophie, Georg H. von Wright, who, in addition to writing
the introduction to Part 2, Philosophical Logic, of the present
volume, has given valuable suggestions concerning the topics to
be surveyed as well as the contributors to all volumes. The former
editor of the Chronicles, Raymond KIibansky, who has devoted
much of his eminent scholarship and his time to the editorial
work of the earlier Chronicles, provided me with much useful
 Preface IX

experience. Nils Heyerdahl, M .A., undertook most conscientiously

the unpleasant task of reviewing the methods of reference and the
bibliographies to make them comply with the given model. The
secretary of the Institute of the History of Ideas, Wenche Karin
Nesse, has with admirable patience typewritten numerous letters
in various languages and otherwise assisted in the preparations.
My thanks are also due to UNESCO for their financial support
and to Martinus Nijhoff Publishers, The Hague, for their willing-
ness to publish the Chronicles. The Secretariat of the Institut
International de Philosophie deserves special recognition for its
most helpful assistance in administrative matters. The Secretariat
has also been responsible for the contact with UNESCO and the
Publisher.
Lastly I want to thank all the scholars from various countries,
whose contributions have made it possible to complete the new
series of Chronicles.

University of Oslo, June 1980 Guttorm Fl¢istad

 PART ONE/PREMIERE PARTIE

Philosophy of language/Philosophie du langage

Introduction
GUTTORM FLq)IST AD

Language has a remarkable history in our culture. This history is

recorded differently by different disciplines. Philosophy of lan-
guage tells us one story, linguistics another, and what may be
called theology of language a third. Additionally one may well
speak of a history of ordinary or natural language itself. It changes,
due to changing internal linguistic elements on the one hand, and
external social, political, and other factors on the other.
Philosophy of language begins roughly with Plato and with
presumably the lowest possible opinion oflanguage: in the Cratyius
language is considered a mere vehicle for knowledge. It is no more
than a means for expressing and communicating knowledge. And
knowledge which is necessary is acquired independently of lan-
guage.
Linguistics describes syntactical, semantical and, fairly early in
its history, also pragmatic aspects of language.
The theological view of language is first of all known from the
Gospels. The phrase 'In the beginning was the word' accords a
creative force to language in relation to reality, which, irrespective
of different interpretations of the phrase, is almost the opposite
of the Platonic view. This theological view is much discussed later
on, particularly by medieval religious thinkers (Cusanus and
others), and several attempts are made to transfer parts of the
creative force of God's word to the human use of language.
Precisely this attempt to establish language as a creative force
(or a set of creative forces) is a salient feature of the ensuing
history of philosophy of language and linguistics. Wittgenstein's
remarks on the underlying grammar of language, Chomsky's thesis
of the mind as a system of innate and generative linguistic prin-

Contemporary philosophy. A new survey. Vol. 1, pp. 3-10.

© 1981, Martinus Nijhoff Publishers, The Hague/Boston/London.
4 G. Fl¢istad

ciples, Austin's and Searle's thesis of constitutive rules of language,

the application to language of the phenomenological view of the
constitution of the world in consciousness (Gadamer), and the
emphasis on pragmatic aspects of language in logical constructivism
(Lorenzen) are well-known examples from our own time. These
positions argue in various ways and in various degrees for a creative
or constitutive role of language in relation to reality. Language is
no longer regarded, as in the Platonic view, as a passive medium of
expression and communication, but as an important source of
reality and reality-constitution itself.
These 'active' views oflanguage may, together with an apparently
growing realization of the impossibility of necessary knowledge in
philosophy in general, also perhaps account for the central position
of the philosophy 0 f language and linguistics in modern philosophy.
The following chronicles are evidence both of the overall
tendency to establish language as a creative or constitutive force in
relation to reality, and of the eminent position of philosophy of
language and linguistics in present-day philosophy.
Jercy Pelc gives an impressive survey of all types of concerns
with language in modern science and philosophy. He defines
philosophy of language in a broad and in a narrower sense. In the
narrow sense it is distinct, e.g., from linguistics, and may be
regarded as a part of semiotics or the general theory of signs. In
the broad sense it includes other types of concerns with language
and is regarded as coextensive with semiotics in so far as semiotics
is concerned with linguistic signs. In the broad sense philosophy of
language or semiotics is linked with a number of other disciplines
such as logic, philosophy in general, and psychology.
Christopher Peacocke reviews six theories of meaning in
analytical philosophy, that is, translational semantics, model-
theoretic semantics, truth-theoretic semantics, game-theoretic
semantics, criterial and constructive semantics, and conceptual
role semantics. A few problems are picked out for discussion,
particularly Grice's earlier thesis that linguistic meaning can be
defi:1ed in terms of an agent's meaning something by an action,
and Quine's thesis of the indeterminacy of translation, both of
which have received much attention and criticism during the
period under review.
Peacocke's review of theories of meaning is a lively demonstration
 Introduction 5

of a main point in Jaakko Hintikka's article 'Semantics: A revolt

against Frege'. The various theories of meaning show how far
semantics has proceeded beyond Frege's distinction between sense
and reference. Hintikka thinks that Frege's logical system was his
most remarkable achievement. His concept of language as a
universal medium prevented him, however, from developing a
logical semantics. Hintikka then introduces his game-theoretical
semantics and the application of possible-worlds semantics to the
analysis of natural language.
The professional career of Richard Montague 'has become a
classic example of the changing and developing philosophical
views of logicians in general, in regard to the form and content of
natural language', writes Nino Cocchiarella in his chronicle,
'Richard Montague and the Logical Analysis of Language'. And
this is a development from a concern with the construction of
formal languages, in order to analyse philosophically interesting
concepts, to a concern with natural language, particularly the
context of use of concepts and expressions, or its pragmatic
aspects. Pragmatics was earlier (in Carnap and others) thought to
be the basis of descriptive semiotics and of no interest to logical
or pure semiotics. Montague develops an intensional logic with a
syntax and semantics that admits of the introduction of pragmatic
operators, and which makes his intensional logic valid for parts of
natural language. Although Montague holds that intensional logic
for natural languages in special cases could do without the dis-
tinction between intension and extension or the Fregean distinction
between sense and denotation, he incorporates this distinction in
his later attempts to develop a universal grammar for natural lan-
guages including proper translation rules.
To account for Wittgenstein's theory of language is no easy
matter, since he cannot be said to have developed his views,
particularly in the Philosophical Investigations, into anything like
a theory. Jacques Bouveresse makes the best of it by relating a
number of extra-linguistic remarks in Wittgenstein to his views on
language, for instance the sharp distinction between science and
philosophy and the therapeutic function of philosophy.
Bouveresse's idea that Wittgenstein was guided by the traditional
ideal of a complete explanation is also worth mentioning. The
basic feature of Wittgenstein's philosophy of language is the
6 G. Hr)istad

transition from the position of the Tractatus to that of the Blue

and Brown Books and Investigations. It is a transition from a
position where the distinction between language and reality,
and the logical construction of these, is al1 important, to one in
which the actual use of language determines its meaning; that is to
say, linguistic meaning is primarily a function of grammar and
contextual pragmatic elements. Language, it seems, has acquired a
position equally important, or even prior, to that of 'reality'.
Linguists and logicians usually regard questions of syntax as
fundamental to that of semantics and pragmatics. In their chron-
icle, 'Constructing a Pragmatic Foundation for Semantics', F.
Kambartel and H.1. Schneider reverse the order. They have wel1-
known ancestors in the pragmatic tradition, especially G.H. Mead
and C .S. Peirce and also Wittgenstein. The meaning of words and
sentences is a function of their use. To use language thus means to
engage in social action. In order to understand and explain such
linguistic systems of actions, the authors take into account the
logical constructivism of P. Lorenzen and W. Kamiah. The term
'constructivism' suggests a programme of explanation such that it
becomes apparent 'how one kind of linguistic act makes possible
another, which in turn makes possible a third and so on'. The
ideas of symbolic acts, of elementary sentences, and of an argu-
mentative (as opposed to a game-theoretical) interpretation of
the logical particles, are important in the logical reconstruction of
language, that is, of a 'dialogical logic'.
The pragmatist tradition is in a radical sense continued in Paul
Ricoeur's extensive contribution 'Logique hermeneutique'. This is
a very complex chronicle from the point of view of philosophy of
language: it deals not merely with problems of language, but with
a number of other topics, such as the distinction between human
and natural sciences, the explanation-understanding controversy,
problems of history and the interpretations of literary texts, the
transcendental character of parts of hermeneutic philosophy, the
intentionalist v. causalist accounts of actions, the theory of
reading, criticism of ideology, and the relation of parts of analytic
philosophy of language, expecially Wittgenstein, to Heidegger.
From the point of view of philosophy of language, Ricoeur
considerably extends the range of problems: the interpretation and
analysis of language is, particularly in the historical human sciences,
 Introduction 7

not just a matter of understanding and analysing the overt intended

meaning or surface meaning of a sentence. The historical context
of the sentence very often makes it say something different from
the surface meaning, i.e. from the points of view of the interpreter
and his own 'remote' historical situation. A similar problem arises
when the use of language is ideologically distorted. In this case the
surface meaning of a sentence or an utterance is consciously used
to conceal what is really intended. These cases are made even more
complicated, in that the use of language, according to hermeneutic
philosophy, is ultimately to be understood in terms of a dialogue,
which often has the form of a dialectic of questions and answers.
A consequence of this view is the need to work out a hermeneutics
of reading (Jauss and others), that is, the condition of under-
standing a sentence or an utterance from the point of view of the
receiver. The upshot of all this is inter alia that the pragmatic
aspects of the use of language are decisive for any proper under-
standing and analysis of language.
The other problems reviewed in Ricoeur's chronicle can be
related to problems in the philosophy of language in the following
way. Hermeneutic logic is a theory of understanding or interpre-
tation. As such it is basically a theory of the part-whole structure
of understanding, where the 'whole' is the context or the prag-
matic aspect of understanding that is prior to every understanding
of parts (or partiCUlar sentences). And this part-whole structure
pervades all human activity, as well as the sciences and philosophy
of this activity. This means that every use and analysis of language
shares certain features with other human activities and that an
analysis of one type of activity may be enriched through analysis
of other types. That is to say, for instance, that a theory of action
or of science or of history and ideology may enrich a theory of
language and vice versa. The main reason for this interplay of
explanations of different activities is obviously to be found in
pragmatics. The context of an activity usually has both linguistic
and nonlinguistic pragmatic components.
I shall point to some main problems and tendencies in the
chronicles as they appear from the point of view expressed above
on the creative (or active or constitutive) force of language.
The general tendency is no doubt to adopt a comprehensive
approach to natural language. This is of course a continuation of
8 G. H</listad

the previous view that the proper meaning of language is to be

found in its use. The analysis and understanding of what this
means extends, however, in various ways beyond the previous
ordinary philosophy of language. Intensional, constructivist, and
hermeneutic logic, each containing pragmatic or contextual
'indexes', indicate the main approaches.
It is interesting to notice that all approaches make use of the
communication situation in their account of linguistic meaning.
In intensional logic the communication situation is present in the
form of an 'index' (Montague) and more explicitly present in the
'game' in game-theoretical semantics, which according to Hintikka
may also serve as a basis for a theory of question and answer
dialogue. Logical constructivism (Lorenzen, KamIah) aims at
developing a dialogical logic. And in hermeneutics the communi-
cation situation determines primarily the use and meaning of
language in accordance with the part-whole structure of under-
standing. The communication situation is moreover regarded as
the basis for a dialectic of question and answer.
A detailed analysis is of course needed in order to decide in what
sense these striking similarities signal a convergence of approach
to natural language. But so much can surely be said at once that
the general outline of a hermeneutics of language in Heidegger and
Gadamer may certainly profit from the detailed analysis of con-
cepts and the models for natural languages developed in the other
approaches (and in intensional logic in general).
The opposite may, however, also hold true. Hermeneutics
represents in a sense, presumably, the most radical approach,
because of its origins in the phenomenological constitution-theory.
This theory states that the basic distinction within the useoflanguage
(and the philosophy of language) is a distinction between two levels
of meaning (or intensions), corresponding to the part-whole struc-
ture of understanding. This means for instance that the distinction
between language and reality, between sense and reference, or
between intension and extension are derivative. In phenomeno-
logical constitution-theory this is simply explained by saying that
objects (of whatever kind) have to be somehow understood in
order to be what they are to us, and that this understanding is
always guided by some overall purpose or strategy (or whole).
Extensions, in other words, are merely accessible as intensions,
 Introduction 9

and because they always occur within some higher order purpose
(context or situation) the intensions are always two-dimensional.
In both cases, however, our view of reality is in part determined
by language and its pragmatics.
The relations between syntax, semantics and pragmatics are
difficult to decide. The chronicles give conflicting views. Logicians
and linguists certainly have a point when they say that syntax is
the most fundamental discipline: No particular use of language
ever invents its syntax; it rather employs a given one. On the other
hand pragmatics may, historically speaking, undoubtedly be said
to have effected changes in the syntax of a given language. And
pragmatics also shows its power in syntactically incomplete
sentences. The context often gives them a perfect meaning.
Regarding the relation of a concept used in some utterance to a
hearer's linguistic expectation (a pragmatic aspect), we may at first
think that the concept acquires its meaning solely within the
utterance and its context and is then conveyed to, and eventually
reinterpreted through, the hearer's expectation. However, the
opposite probably holds true. The hearer's expectation is an
essential component of the speaker's context that helps to deter-
mine his interpretation and use of the concept. The context as a
whole thus has a significant influence on the reality presented in
the utterance and the way it is presented. In ideologically dis-
torted meanings the creative power of language is particularly
perspicuous.
Because ot difficulties such as these, it is probably true to say
that syntax, semantics and pragmatics (in logical as well as in
empirical semiotics) are mutually dependent on each other, and
that none of these disciplines has absolute priority over the
others with regard to the understanding and explanation of
natural language. The priority of one discipline is at most relative.
A cautious solution such as this is also recommended from a
different point of view. It is notoriously difficult to draw any
sharp boundaries between these disciplines and their objects,
especially between semantics and pragmatics. In their account of
logical constructivism Kambartel and Schneider hold initially
that pragmatics is the fundamental discipline, but later admit
that pragmatics comes close to (an extended) semantics. And
Hintikka thinks that text-semantics is just as viable as 'ordinary'
lOG. Fll/listad

semantics. And in hermeneutics the argument would probably be

much the same way, merely insisting on a two-level semantics, that
is, a part-whole semantics (which no doubt has some affiliations
with game-theoretical semantics in Hintikka, that is, between
functions of individuals and higher-order functions).
Finally I should like to point to the affinities that no doubt
exist between the concept of the world in hermeneutics (or
phenomenology) and the possible-worlds semantics. First, the
obvious difference is that the latter is concerned with logically
construed worlds, whereas the concern of the former is with the
ever-changing worlds in which we actually live in our under-
standing or consciousness. The worlds of hermeneutics are, as it
were, the possible worlds of logical semantics in use. The world,
primarily consisting of a system of meaning (or intensions),
becomes in hermeneutics 'experiential meaning' (to use a phrase
from Charles Taylor).
That the two approaches nevertheless have much in common is
shown by the fact that they encounter much the same problems,
for instance the problem of cross-identification of particulars.
This is both a theoretical and a practical problem. Theoretically
the task both in possible-worlds semantics and in hermeneutics
is, roughly, to state the condition (inclusive of the pragmatics or
text-semantics) that must obtain in order that cross-identification
may be possible. The conditions for an interaction of different
higher-order functions or different 'horizons' of understanding
play an important role.
Practically the problem is familiar to everyone. The cross-
identification of particulars is a main problem in all use of lan-
guage, in science and philosophy as well as in ordinary discussions
and conversations. It is decisive of success and failure in communi-
cation. It is a serious problem. To use language is after all a form
of life. To use it in the cross-identification of particulars is a form
of communal life, and philosophy of the dialogical character of
natural language is a philosophy of communal life.
The place of the philosophy of language
JERZY PELe
Warsaw University

I. THE PHILOSOPHY OF LANGUAGE

1. The philosophy of language and philosophy

What is the place of the philosophy of language among other

disciplines? The first answer that comes to mind is this: it is part
of philosophy and, as such, is one of the philosophical disciplines.
This answer sounds simple and natural: after all, is not the
philosophy of language some kind of philosophy? A moment of
reflection, however, brings doubts to mind.
The word philosophy is polysemic. For instance, Kotarbinski
([ 1] , Part V, Ch. VI) identifies six different meanings of the word.
First, the noun philosophy denotes a heterogeneous collection of
subjects traditionally taught at philosophical chairs within uni-
versities. Initially this collection consisted of metaphysics,
epistemology, formal logic, methodology of sciences, ethics,
aesthetics, history of philosophy, psychology, philosophy of
science, philosophy of history, and philosophy of religion. In the
course of time the last four components departed from philosophy
to constitute autonomous disciplines or to make part of those
disciplines with whose philosophy they were concerned. Second,
the word philosophy has been found in the names of philosophy
faculties at the universities to indicate that they have been devoted
to the so-called 'pure' sciences as different from such applied
sciences as law or medicine, and to emphasize their preoccupation
with providing education for teachers. Third, philosophy may be
contrasted with science: the former constitutes a general Weltan-
schaung, partly based on the synthesis of cognitive findings

Contemporary philosophy. A new survey. VoL 1, pp. 11-34.

© 1981, Martinus Ni;hoff Publishers, The Hague/Boston/London.
12 J. Pelc

gathered from all the various specialized disciplines and concerning

such problems as causality, determinism. free will, mind and
matter, organic and inorganic matter, psychic experiences versus
activities and their results, death and loss of self-consciousness, the
world of values and the real world, cognition versus things and
phenomena, and the universe and its creator. Fourth, philosophy
is sometimes considered as a pre-scientific stage in the develop·
ment of a given discipline, providing its foundations and elementary
theorems. Fifth, the word philusophy occurs as the first part in
the phrase the philosoph.v of such and such science and denotes
the theory of that science and the analysis and critique of the con-
cepts and methods of discipline that is the subject matter of this
philosophy. Philosophy conceived in this way constitutes a meta-
science. Sixth. metascientific character is also found in philosophy
construed as the general theory of science, i.e .. when it is not a
philosophy of this or that science, but the philosophy of science in
general. It is the science of science in this case.
As we have seen, placing the philosophy of language within
philosophy is not in itself pointing to its precise address or deter-
mining its place univocally.
It is not exclusively due. however, to the polysemic character of
the word philusophy that we find this perplexing state of affairs.
The phrase philosophy of language should also be made more
precise.
A step in this direction is perhaps the distinguishing of the
philosophy of language from linguistics, especially from theoretical
linguistics. the philosophy of linguistics, and philosophy conceived
as analysis of language, or so-called linguistic philosophy.
Thus, what is the philosophy of language'?
The editors of one of the recent comprehensive collections of
articles on this SUbject, Readings in the Philosuphy of Language,
Rosenberg and Travis. say explicitly: 'neither we nor. to our
knowledge, anyone else. know how to give such an account'
(Rosenberg and Travis 12]. p. 1). There is nothing we can do.
therefore, but resort to induction and hy enumeration of examples
look for the answer to our query.
Who is induded or at least mentioned in the philosophy of
language'? We shall limit ourselves to enumeration of a number of
past scholars and quote the names of only the most representative
 The place of the philosophy of language 13

among them: Frege, who distinguished Sinn from Bedeutung, and

the thought expressed in a sentence from the assertion of that
thought, and who characterized a connection between denotation
and truth; Husserl, who introduced the concepts of meaning
intention, syntactical category, and semantic transparency; Russell,
who developed the theory of descriptions and the conception of
logically proper names; Wittgenstein, who in Tractatus ([3] ,3.203)
declared that the object named by a name is its meaning, and in
Philosophical Investigations (Wittgenstein [4]) contended that the
meaning of an expression does not consist in a relation to any
object whatsoever but in the use of that expression, and who said
in Tractatus that a sentence is an arrangement of names reflecting
a fact and that language is a sum of sentences mapping the world,
and in Philosophical Investigations wrote that stating facts is only
one of the functions of language; Carnap, who introduced the con-
cept of intensional isomorphism and distinguished between formal
and material modes of speech; Carnap again, who along with other
logical empiricists expanded a theory in which the meaning of a
synthetic sentence is - to put it bluntly - a method of empirical
verification of that sentence or - if we are not to spare words -
an empirical synthetic sentence is equivalent to a class of elemen-
tary observational statements, registering sense data, which add up
to make this verification possible; Lesniewski, who developed a
system of hierarchy of semantic categories superposed on his
so-called ontology in order to protect his calculus from paradoxes;
Ajdukiewicz, who worked out directive theory of meaning and
categorial grammar in the scope of which the concept and method
of checking syntactic connection of sentences was devised;
C.I. Lewis, who introduced the concept of strict implication and
elaborated a theory of modal sentences; Austin, who was the
advocate of differentiation of locutionary, illocutionary, and
perlocutionary acts and/or forces, and who was the author of the
theory of performative statements, Bar-Hillel. who investigated
indexical expressions and proposed a well-known concept of
pragmatics; and Montague, who designed a formal language of
pragmatics.
This list of names, along with enumeration of problems, might
certainly be continued, going as far back as ancient Greece. Like-
wise, it is not difficult to point to contemporary philosophers who
14 J. Pelc

at least occasionally become concerned with the problems usually

included in the philosophy of language, such as: the semantic
concept of truth and the differences between object language and
metalanguages (Tarski); the concepts of synonymity, analyticity,
truth, and existence (Quine); onomatoidal expressions (Kotar-
b ifl ski) ; systematically misleading ex pressions (Ryle); and the
concept of presupposition and the connection between denotation
and speech acts (Strawson). These concepts and issues are more
numerous, e.g., an analysis of different types of terms (individual
versus general, subject terms versus predicate terms, mass-terms
versus non-mass-terms) and the ensuing problems of quantification;
furthermore, problems of modal expressions, in particular of
epistemic type; problems of meaning, denotation, and truth;
problems connected with the concepts of paraphrase, satisfaction,
logical consequence and logical tautology; problems of under-
standing utterances; the concept of semantic convention; problems
of intensional contexts; concepts of use of expressions; problems
of possible worlds and reference points; problems of the natural
and the artificial language, the formal and the informal language;
problems of the model of language, of semantic universals; the
concept of well-formed expressions and - generally - the concept
of grammaticality; and problems of distinguishing syntactics from
semantics and pragmatics.
Although incomplete, this list might serve as the basis for an
answer to the question of what the philosophy of language is
concerned with, if it were not the case that at least some of the
enumerated problems lie simultaneously in the interests of
linguistics, especially theoretical linguistics, as well as those of the
philosophy of linguistics and linguistic philosophy. As regards
those common problems, studied by each of the above-mentioned
disciplines, the issues dealt with in the philosophy of language
must tentatively be described in a highly unsatisfactory way by
saying that the last is concerned with philosophical aspects of
these problems. This circumlocution is unsatisfactory because the
word philosophical is polysemic, and additionally because such
explication would commit a didactic fallacy of ignotum per
ignotum and, in some interpretations, the logical fallacy of the
vicious circle. Finally, it is unsatisfactory because it draws a line
that does not run exactly between the philosophy of language and
 The place of the philosophy of language 15

the remaining disciplines, but rather between the numerous above-

mentioned kinds of philosophy on the one hand, and linguistics on
the other.
This introductory survey is worth keeping in mind, however, for
it points out that the philosophy of language takes language as its
subject matter and examines certain general traits it possesses, i.e.
not the traits that belong to a certain ethnic language in contrast
to some other, and that the philosophy of language is one of
philosophical disciplines, in the first, second, fifth, and, perhaps,
fourth from among the meanings of the words philosophy, and
philosophical, mentioned above.

2. The philosophy of language and linguistics

The next step toward locating the philosophy of language among

other sciences brings us in closer contact with the neighboring
fields, especially linguistics.
By definition, it studies language or languages. If its interests -
as they used to be - were now limited to such investigative tasks
as compiling dictionaries of a given language, description of its
phonetic system, declension, syntax, and writing books of
grammar or, on the practical side, to teaching languages, and, on
both theoretical and practical sides, to comparatively analyzing
different languages, we could fairly distinguish the subject and
methods of linguistics from those of the philosophy of language.
But contemporary empirical linguistics tries to clarify how it is
possible that a competent speaker of a given language can easily
understand and/or construct all of its utterances, including even
those that are new to him. In the attempt to explain this fact
the linguist develops a descriptive system that is meant to detect
regularities of a language. Such a system, containing recursive
rules, forms a substantive scientific theory (Fodor and Katz
[5 ]).
Theoretical linguistics deals with 'universal theories of grammar
and semantics, that is, with describing the principles of grammar
and meaning that underlie all possible natural languages' (Rosen-
berg and Travis [2], p. 3). Grammar is treated (Fodor and Katz
[5]) as a formal system that enables the generation of infinite
sequences consisting of lexical elements, contains in addition a
16 J. Pelc

structural description of every sentence explicating its grammatical,

i.e. syntactic, relations, and precisely characterizes the whole set
of sentences in a given language.
As we have seen, contemporary linguistics, and especially
theoretical linguistics, share in part the same subject matter with
the philosophy of language. Their interests and investigative tasks
overlap. Their methods and research techniques are frequently
similar: for instance, it becomes less and less typical in linguistics
to concentrate on the accumulation of more and more individual
data, and instead - which resembles the approach of the phil-
osophy of language in the past and at present -- the principal
attention is paid to the facts universally recognized by all com-
petent speakers (Fodor and Katz [5]), and intuition is no longer
mistrusted.
But even though the philosophy of language lies in the vicinity
of linguistics and reciprocal trespasses are not infrequent and are
justified by the common sharing of some grounds, neither disci-
pline is a tenant of the other nor can be reduced to its mere part.

3. The philosophy of language and the philosophy of linguistics

Both the philosophy of language and the philosophy of linguistics

work on the fragment of the subject matter studied by the phil-
osophy of linguistics. The latter, as one of the philosophies of
specialized sciences, comprises methodology of its discipline and
examines the conceptual apparatus and techniques employed in
linguistics. Some of these concepts (e.g., the concept of meaning)
and techniques draw also the attention of the philosophers of
language and the theoretical linguists. Moreover, the philosophy
of linguistics tries to identify the relations between the facts of
particular languages and the facts of the language conceived
generally. It also attempts to show when grammar is adequate.
These issues as I have said -- are also interesting to the phil-
osophers of language. It is even claimed sometimes that the
philosophy of language should be studied only as the philosophy
of linguistics (Fodor and Katz [5]). But surely this is not a con-
senSllS omnium. The problems of the philosophy oflanguage only
partly coincide with those of the philosophy of linguistics. but
neither discipline has entirely contained its counterpart.
 The place of the philosophy of language 17

4. The philosophy of language and linguistic philosophy

From the philosophy of language, linguistics, and the philosophy

of linguistics, we should distinguish linguistic philosophy. I t is not
one of the philosophical disciplines or doctrines, but a method
used in several of them.
This method consists in a departure from consideration of such
extra-linguistic entities as necessity, free will, or knowledge;
instead it concentrates on the analysis of corresponding words in
the natural language, in particular of the language of specialized
sciences, on the relations that obtain between the words, and on
the analysis of adopted (in a given language) classifications and
semantic discussions. This leads to the elimination of the material
mode of speech, i.e., to the removal of the so-called pseudo-
objective and at the same time quasi-syntactic sentences, and to
the substituting of them with equivalent syntactic sentences
phrased in the formal mode (Carnap [6]). Therefore, for instance,
instead of this circumstance is logically necessary, we shall have
this sentence is analytical (Ayer [7], p. 25). The former sentence,
contend the proponents of such paraphrasing, as well as the latter,
is about language, but while the former pretends that it speaks
about things and not about words, the latter openly indicates that
it is about words. It is a superstition they continue, to believe
that only linguistic philosophy deals with words while traditional
philosophy deals with nonverbal entities, the only difference
being that linguistic philosophy, by the selection of its investi-
gative techniq ues, testifies to the belief that conceptual problems
cannot be separated from the use of words, while traditional
philosophy has never admitted it openly. At the same time, how-
ever, the latter cannot but subscribe to this view and employ it in
practice, as it does in fact, whenever, for example, it examines
specialized disciplines in order to establish the meaning of terms
used in their discourse.
Also. linguistic philosophy strives to identify the meaning of
scientific terms, and it is part of its program, but the philosophy
does not stop there. It analyzes certain formulations on the
supposition that it is easier to determine what conditions make
acceptable a use of the sentence, say, John is happy, than to try
to inquire into the nature of happiness. This is why Moore, Ryle,
18 J. Pelc

Austin, and many other philosophers have brought to light ambi-

guities and equivocations in the formulations of natural language,
and why Wittgenstein in Philosophical Investigations [4] makes
an appeal to study use of expressions; in Tractatus he even says
that all philosophy is a critique of language (Wittgenstein [3],
2.4.0031). The advocates of this program show, by carrying it out
as a method in linguistic philosophy, that formalization and
mathematization are not the only weapons against metaphysical
delusions. Moreover, they point to the source of the delusions:
language, viz., a use of words that is inappropriate relative to
circumstances, inadvertance to certain important differences
lying behind superficial grammatical similarities, etc. Thus,
linguistic philosophy takes it to be its task to reveal these dif-
ferences and, by employment of alternative terms, to differentiate
between the surface structure of an expression and its deep
structure, or perhaps various deep structures that can be reached
by transformation. Carnap [6] called for the replacement of
philosophy by the logical analysis of concepts and sentences of the
languages of specialized sciences.
Linguistic philosophy, both by its method and program, has
close affinities with the philosophy of language in several respects.
First of all, the natural method of the philosophy of language, or
at least one of its methods, is that of analyzing expressions, which
- as we have seen - is so important in linguistic philosophy.
Secondly, and on the other hand, the choice of method in linguis-
tic philosophy depends on the basic conceptions about language,
and they constitute the core of the philosophy of language.
Thirdly, even if it still remains an open question, whether linguistic
philosophy is always preoccupied with the investigation of lan-
guage, namely also when its objective is to decide certain non-
linguistic philo:;ophical problems, it is unquestionable that language
frequently constitutes the subject of its analysis. In such cases,
linguistic philosophy and the philosophy of language work in the
same area. Fourthly, both disciplines often follow the same
pattern: concentrate on a simplified, schematic, and deformed
ideal language, hoping that they learn in this way how the real
languages function and what their structures are like.
The subject matter of the philosophy of language and the
linguistic philosophy is, then, fragmentarily shared, as are certain
 The place of the philosophy of language 19

investigative methods of these disciplines. And yet neither is the

philosophy of language a fragment of the linguistIc philosophy,
nor the opposite.
This is the situation, therefore: the philosophy of language,
theoretical linguistics, the philosophy of linguistics, and linguistic
philosophy are not one and the same thing; none of these disci-
plines is included in any other. even though each partially
encroaches upon the subject matter of its neighbors; some of their
methods are common; some crucial problems in one of them
cannot be neglected by some other or at least an analysis of some
problems must rely on the findings accumulated by the partners.
Trying to describe the province of the philosophy of language
we started by pointing to the country where it belongs: philos-
ophy. When it turned out that its location and topography could
not be easily determined, we made another attempt: a few out-
standing citizens of the province have been cited and the citations
said what their preoccupations were, implying indirectly the
preoccupations of the country's inhabitants. We next turned to
the neighbors and relations among them. It is time now to say
where this province lies, and where the region lies in which it will
be found. An address or geographical coordinates could add useful
information and imply even more.
]n certain cases even when a determination of place is partly
ruled by convention; for instance, whether this or that village
belongs to Europe or Asia. Location of a scientific discipline will
not be any easier. There may be diplomatic reasons that suggest
adopting one convention rather than another. Subsumption of the
philosophy of language under philosophy, or. alternatively, under
linguistics, might kindle animosity. In the former case, linguists,
and especially theoretical linguists, might feel that their rights have
been curtailed; in the latter case, the philosophers of language and
the linguistic philosophers might feel the same way. Hence a com-
promise: we may agree that the philosophy of language lies within
the boundaries of semiotics. Then by describing the territory of
semiotics we will learn some more about the place that the phil-
osophy of language occupies and, consequently, about the
discipline itself.
20 J. Pelc

II. SEMIOTICS

1. Semiotics, semiology, semantics

Carnap believed at one time that 'the task of philosophy is semi-

otical analysis,' since philosophy is not concerned with the ultimate
nature of existence, but with the semiotic structure of the
language of science, including the theoretical part of everyday
language (Carnap [8], p. 250). Semiotics is the science of signs. It
was described in a similar way by John Locke in 1690 when he
used that term to denote 'the doctrine of signs' (Locke [9], IV,
21, § 4). C.S. Peirce characterized semiotics as the 'quasi necessary,
or formal, doctrine of signs' (Peirce [10], 2.227), i.e., of 'some-
thing which stands to somebody for something in some respect
or capacity' (ibid., 2.228), or as 'the doctrine of the essential
nature and fundamental varieties of possible semiosis' (ibid.,
5.488), semiosis being for him 'an action, an influence, which
is, or involves, a cooperation of three subjects, such as a sign, its
object and its interpretant, this tri-relative influence not being in
any way resolvable into actions between pairs' (ibid., 5.484).
Ferdinand de Saussure came to use the word semio[ogie to denote
the science that would study 'la vie des signes au sein de la vie
sociale' and would try to find out 'en quoi consistent les signes,
quelles lois les regissent,' and his opinion was that 'la linguistique
n'est qu'une partie de cette science generale, les lois que decouvrira
la semiologie seront applicable a la linguistique' (Saussure [11],
p. 46 f.). Also, Charles Morris considers semiotics to be the science
concerned with all sign systems, and among them with languages,
as wen as with single signs and sign sets that do not form systems
(Morris [12], pp. 307-310 and 366).
Semantics was the third term that not long ago used to occur as
the name of the science of signs, along with semio tics and
semiology; it was in such a general sense that Husserl interpreted
the adjective semantic when he introduced, in 1913, the concept
of semantic categories (Husserl [13], II, I, pp., 294 ff., 305 ff.,
316 ff., 326 ff.); likewise, Leon Chwistek and Tadeusz Kotarbiilski
used the word semantics in the broader sense, identical or at least
cognate with that of the noun semiotics (Kotarbiilski [1] , p 3).
Today, the term semantics occurs, for example, in linguistics or
 The place of the philosophy of language 21

in the philosophy of language, in its narrower senses, each of

which differs from that of semiotics. One of such narrower con-
cepts of semantics is to be found in Henryk Hii ([ 14] , p. 40 ff.),
who took it over from Roman Jakobson: semiotics is concerned
with meaning transmitted by any medium, not by language alone;
it is thus a generalization of semantics; those properties that are
essential for meaning have been termed by him semiotic properties,
while semantic properties are those semiotic properties that are
characteristic of speech; model theory is one of the domains of
semiotics; the semiotic role of any communication-oriented
behavior is not only to convey a content, but also to express an
experience.
Another difference l.,ctween semiotics and semantics in one of
the narrow meanings of each term is mentioned by Kotarbiflski
when he writes that such words as to mean, meaning, sense,
connotation are by some called semiotic, while such as to denote,
denotation, designatum are considered as semantic; the former
show the characteristics of expressions regardless of the relations
these bear to nonlinguistic reality, whereas the latter because of
those relations, i.e., according to whether they map some section
of that reality or not: the meaning of an expression in a given
language depends merely on the intention of the speakers, whereas
the fact that, if any, things are denoted by that expression is deter-
mined not only by the speakers' intentions, but also by whether
there are objects that correspond to those intentions (Kotarbiilski
[ 1 ] ).
In the literature on the subject, however, the prevailing sense of
the word semantics is different, with the resulting different dis-
tinction between semantics and semiotics.
In 1938, Charles Morris - probably influenced in that respect
by Peirce - singled out three 'dimensions' of semiosis, i.e., of the
process in which something functions as a sign; they are the
semantic dimension, the pragmatic dimension, and the syntactic
dimension, that is, the relations a sign bears to: those objects to
which it is applied, those who interpret it, or other signs, res-
pectively. The three dimensions and the three kinds of relations
are, each of them, the subject matter of, correspondingly,
semantics, pragmatics, and syntactics; semiotics thus consists of
those three branches (Morris [12], pp. 21-24). In 1946, Morris
22 J. Pelc

suggested the following distinction: pragmatics is that branch of

semiotics which is concerned with the origin, uses, and effects of
signs - all that within the sphere of behavior in which given signs
occur, semantics is concerned with varieties of meaning, and
syntactics -- with connections among signs, but without paying
attention to the meaning of each sign involved or to the relation
it bears to that behavior in which the signs occur. Semiotics
acquired in this case a behavioristic - but not limited to human
behavior alone - tinge; Morris also emphasized that while it is
legitimate, and sometimes even convenient, to speak about one
branch of semiotics separately, it must always be borne in mind
that semiotics forms a whole (Morris [12] , pp. 301-303).

2. Subject matter of semiotics

The concept of sign is one of the pivotal concepts in semiotics,

and classification, ordering, and typology of signs are among the
primary tasks of that discipline. Different theories of sign are
current in semiotics. Usually, however, the sign is defined not only
from the syntactic point of view, i.e., by a description of its shape
and/or sound, but from the semantic and pragmatic points of view
as well, namely by a description of the relations it bears to the
object(s) it denotes or connotes, etc., and by a description of how
it expresses the various experiences of those who produce it and
how it evokes experiences in those who receive it. There are
divergent opinions on all these issues, most of them of philo-
sophical nature; for example, are signs in the category of things or
in that of events or phenomena, or are some of them in the
former, and others in the latter category, so that they cannot be
covered by any single definition? Is the relation a sign bears to
that of which it is a sign conventional in nature only, or can it also
be natural, so that so-called indexical signs, or symptoms, or
natural signs, which have no conscious producer, but only a
receiver, could be classed as signs? If something is to be treated as
a sign, is it necessary for that something to have cognitive value,
and hence to express, or at least to evoke, a person's thought that
something is so and so, or does it suffice if it expresses, or at least
evokes, a nonintellectual experience, so that we could class as
signs such human (or nonhuman) products that merely express
 The place of the philosophy of language 23

or evoke emotions only? (Kotarbiflska [15] , pp. 57-143).

Classifications of signs are based on various criteria. For instance,
a symbol, i.e., a sign that does not bear any resemblance to the
object it designates, e.g., the word horse, is opposed to an iconic
sign, i.e., an image - such as a portrait, a map, an onomatopoeic
word - that in some respect (be it topological, geometrical,
acoustic, etc.) resembles the thing to which it refers. Further,
natural signs are distinguished from conventional signs, symptoms
from signals, etc.
In semiotics, much attention is paid to the theories of meaning.
Some of them, of frequent occurrence in the philosophy of
language, confine themselves to discussing cognitive meaning only,
and in particular to what is termed referential meaning; others
take emotional meaning into account as well; the former are
semantic, and the latter, pragmatic in nature; the former are often
concerned with the meaning of the individual signs as they occur
in isolation, while others consider contextual meaning only,
because it is claimed in them that a given sign means something
solely if it occurs together with some other signs, and also in a
specified situation; this view is often combined, especially in
linguistics, with the opinion that there are no synonymous signs
as there are no identical situational contexts. The meaning itself
is interpreted in various ways: as a thought associated with the
perception of a given sign, as the organism's response to that sign
as a stimulus, or as the rule stating how that sign is to be used
under specified circumstances, or the set of properties connoted
by that sign, etc. Most of those concepts of meaning discussed in
semiotics are of importance for the philosophy of language.
The meaning of a sign is only one of its semiotic functions:
syntactic and such remaining semantic functions as designating,
denoting, connoting, truth, or satisfaction are also analyzed in
semiotics, along with pragmatic functions, such as expressing,
understanding, or acceptance of a sentence by its user. All this
is linked to such issues, well known from the philosophy of
language, as analytic-synthetic, extensional-intensional,
description-evaluation-norm, semantic antinomies, and all other
logical errors connected with the use of speech.
Finally, semiotics is concerned with the construction of the
concepts of expression, text, langue, iangage, parole, code,
24 J. Pelc

metalanguage, etc. It covers metalanguages of higher levels,

including a metalanguage of so high a level that it enables the
discussion of all other metalanguages used in semiotic research and
the relations among them. Thus an important feature of this
discipline consists in the fact that semiotic discourse takes place at
various levels and that the syntactic, the semantic, and the prag-
matic metalanguage occur even at the lowest level, and also in the
fact that methodological and metascientific - and thereby,
philosophical, in one sense of the term - reflection forms an
integral part of semiotics.
Semiotics is not the only discipline concerned with signs, and,
a fortiori, not the only discipline concerned with language. There
are sciences that are concerned with signs, too, but not solely
with them: this applies to logic, mathematics, psychology, psy-
chiatry, the sciences of literature and art, sociology, information
science, etc. All these disciplines, however, are concerned either
with signs of one kind only, or with signs of various kinds but only
from a certain single point of view; in most cases, too, signs are
only part of the subject matter - and not the principal one at
that - of the disciplines mentioned above. Semiotics, on the
contrary, is concerned with all kinds of signs from all points
of view, which links it to logic, philosophy, linguistics, psychology,
sociology, information theory, theory of art, aesthetics, symptom-
atology as a branch of medicine, genetics as the science con-
cerned with genetic codes, and with many other sciences; this fact
accounts for the interdisciplinary nature of semiotics. These
connections are noticeable not only within semiotics itself, e.g., as
connections with logic and the philosophy of language in
semantics, but also in the border areas between semiotics and
related disciplines, such as psychosemiotics, sociosemiotics, and
praxiosemiotics. Perhaps the range of interests of semiotics over-
laps with that of the philosophy of language in the broad sense of
the latter term, i.e., when the philosophy of linguistics and lin-
guistic philosophy are considered to belong to the philosophy of
language; when, however, the philosophy of language occurs in a
narrow sense of the term and is distinguished from the other two
disciplines, then its range of interests seems to be contained in
that of semiotics.
 The place of the philosophy of language 25

3. Semiotics as a method

The term semiotics happens to be used, often unconsciously, not

as the name of a certain discipline, but as the name of a certain
method. One must realize the difference between theoretical
semiotics and its applications, i.e., applied semiotics, in the sense
of results of analyses carried out in the various fields with the use
of the semiotic method.
We can point to three kinds of the applications of the method:
formalization of language, analysis of language and other sign
systems, and research on objects that primarily are not signs.
Formalization of a language consists in handling its signs
exclusively relative to their graphic shape, their so-called eidetic
meaning being disregarded, and hence in handling them within the
scope of syntactics, so that computational methods can be applied
to objects that are not numbers; the signs we are concerned with
have only what is called operational meaning, which is to say that
we merely know how to handle them in confonnity with speci-
fied syntactic rules. We thus obtain entire theories and even
disciplines as systems admitting of various interpretations, pro-
tected against intrusion of implicit rules, axioms, and pre-
suppositions (Bochenski [16], pp. 31-43). Since formalization
can be applied both to natural and artificial languages, and to
various theories and disciplines, it is an interdisciplinary factor
that integrates fields which sometimes are quite remote from one
another.
The semiotic method, which consists in treating the objects
under investigation as signs, naturally finds application in language
analysis and in the analysis of other sign systems and of sets of
signs that do not form systems, and finally also in analyzing indi-
vidual signs. This is the second kind of application of the method,
its main field being linguistics. Some researchers claim that the
subject matter of linguistics is nothing other than the semiotic
properties of natural languages and of communication by means
of those languages, while linguistics itself happens to be treated
as a branch of semiotics, whose scope overlaps that of pragmatics;
in this connection it is believed that linguistic theories of language
presuppose a semiotic theory (Lieb [17], pp. 94-119, and [18] ,
pp. 217 -249). Linguistic tenninology is based on that of semiotics,
26 J. Pelc

because - as has been claimed, e.g., by Morris ([ 12] , pp. 304-

307) - semiotics provides linguistics with its metalanguage and
criteria for classification of expressions. Some authors maintain
that linguistic facts are forms of symbolic, and hence semiotic,
behavior (Sapir [19], p. 211) and that linguistics is a branch of
the general theory of signs (Bloomfield [201 and [21]). The
expanding application of semiotic methods to the study of ethnic
languages contributed, especially after 1950, to the emergence of
modern theoretical linguistics, which to such former conceptual
distinctions as de Saussure's language, speech, and discourse, has
added new semiotic concepts, namely those of text, deep and
surface structure, presupposition, semantic universals, coherence,
competence, and performance, as well as many others. Some of
them are common property of linguistics and the philosophy of
language. As the development of theoretical linguistics goes hand
in hand with that of the philosophy of linguistics, and as the latter
and the philosophy of language have a number of interests and
problems in common, the claim that linguistics forms a part of
semiotics paves one more way for the opinion that the philosophy
of language is contained in semiotics.
Applications of the semiotic method also cover analyses of the
languages of the various disciplines. In such cases the method
under consideration consists in studying the process of construction
of the language of a given science, in analyzing the emerging set of
concepts, and in investigating the use of such concepts within the
discipline itself, i.e., interpretation of expressions and formation
of opinions. Such analyses in tum yield assessment of the
effectiver.ess of the use of the language of a given discipline,
assessment based on both formal and praxiological criteria, and
sometimes lead to the formulation of the norm of construction of
such a language and the set of its concepts. As is known, such
norms function in their most precise form as the syntactic rules of
deductive systems, especially in logic and mathematics. This is
in what the role of the semiotic method consists in construction
of these disciplines. Application of this method to the language of
disciplines that sometimes differ widely from one another brings
them closer together and may contribute to their postulated
integration or unification in one respect: semiotics would provide
a metalanguage for the description of the languages of the said
 The place of the philosophy of language 27

disciplines and for an analysis of relations among those languages.

Application of the semiotic method to the analysis of a branch
of philosophy gave rise to so-called semantic epistemology, in
which analysis of the relation between propositions, i.e., between
cognition in the logical sense, and the real facts, as is done in
ordinary epistemology, is replaced by analysis of the relation
between sentences, of which those propositions are meanings,
and the facts, so that epistemological issues are approached from
their linguistic aspect. Since language may be treated as a deductive
system, it turns out that the theory of deductive systems, i.e.
metalogic, has the same subject matter as semantic epistemology;
either can accordingly avail itself of the results obtained by the
other (Ajdukiewicz [22], pp. 155-191, and [23], pp. 140-154).
We have here one more example of sharing some grounds by
semiotics and, in this case, linguistic philosophy, which again
points to the fact that the philosophy of language, in the broad
sense of the term, has close affinities with semiotics, while the
philosophy of language, in a narrower sense of the term, is sub-
sumed under semiotics.
The third sphere of application of the semiotic method is that
of things, phenomena, and events that by their nature are not
signs, such as: buildings, costumes, human and animal behaviour,
manmade landscapes, musical compositions, paintings, sculptures,
rites, meals; none of these has been produced or invented solely or
primarily to function as a sign. The same sphere covers such
natural phenomena as the appearance of a comet, the eclipse of
the sun, an earthquake, a hurricane, a person's flushing or paling
- the last being an example discussed in La logique ou ['art de
penser (Arnauld and Nicole [24]), and thereby in a book partly
concerned with the philosophy of language - the falling of leaves,
the appearance of hoarfrost on trees, etc. In such cases the semi-
otic approach does not seem the most natural of all, but can be
chosen deliberately. Such applications of the semiotic method are
often differentiated terminologically from investigations in
theoretical semiotics, and are given a compound name, such as,
eg., the semiotics of architecture or zoosemiotics, while the funda-
mental ones are labeled just semiotics, without any further modi-
fiers. Application to any such object of the semiotic method
makes us treat that object as 'semiotically transparent,' which
28 J. Pelc

means that our attention does not focus on that object alone, but
moves toward that of which the object - in our opinion - is a
sign, and often also toward the mental experiences of the actual or
alleged producer of the sign, experiences which we believe to be
expressed by that object. In a word, such an object becomes
subject to a semiotic interpretation, in particular, to understanding.
When this occurs, we reflect on its meaning or signification: for
instance, the meaning of a rite, a symphony, the outfit of a
housing space, or a rainbow. We also try to determine the object's
role is in the process of communication.
Application of the semiotic method to human and/or animal
products and actions that are not signs by their nature, and to
natural events, makes it possible to uniform our look at the
various spheres of cognition, and in particular enables us to
analyze nonhumanistic objects in a humanistic manner, so that
they are subject to such procedures as valuation and teleological
approach. This, on the one hand, opens opportunities for inte-
grating research and making it more comprehensive by the use of
the semiotic method, which converts the objects of its analysis
into texts and other sets of signs, but, on the other hand, involves
the danger of abusing the concepts of sign, meaning, and language.
Abusing the semiotic method may sometimes result in super-
stitions and groundless interpretations of allegedly supernatural
phenomena, and in many cases ends in formulating pseudo-
scientific theories.
Whenever language is the subject matter of investigation, the
semiotic method and that employed in the philosophy of language
coincide, or, in other words, the philosophy of language makes use
of the semiotic method. What is more, the philosophy of language
may be identified with that part of semiotics which deals with
language. The close relationship between the philosophy of lan-
guage and semiotics is perhaps best visible in the case of logical
semiotics.

4. Logical semiotics

Logical syntactics together with logical semantics and logical

pragmatics combine to form what Ajdukiewicz called logical
semiotics, i.e., the logic of language (Ajdukiewicz [25] and
 The place of the philosophy of language 29

[23J, pp. 182-191, 282-294}. He interpreted logical semiotics

broadly, as the science that analyzes functions of speech, especially
the function of 'mapping of facts in speech. In this way it prepares
the set of concepts and the terminology which are indispensable
for informing about all kinds of infringements of the principle
that statements must be matter-of-fact, unambiguous, and precise
formulations; it further systematizes such infringements, and indi-
cates the ways of preventing them' (Ajdukiewicz [25 D. Logical
semiotics characterized in this way covers pure and descriptive
logical semiotics, general next to its specialized versions, formal-
ized next to nonformalized, and formal next to non formal.
The distinction pure versus descriptive comes from Camap, and
in his interpretation applied to syntactics and semantics only, but
can now be made to cover pragmatics as well. Pure syntactics
(semantics) was an analysis of systems of syntactic (semantic)
rules that consist exclusively of definitions of syntactic (semantic)
concepts relative to a given syntactic (semantic) system and of
their consequences; hence pure syntactics (semantics) is analytic
and does not include any factual elements. Pure syntactics refers
neither to semantics nor to pragmatics, and pure semantics does
not refer to pragmatics. Pure semantics is based on pure syntactics.
Descriptive syntactics (semantics) consists, according to Camap, in
an empirical study of syntactic (semantic) features of languages
that exist or whose existence in the past has been proved. Camap
also claimed that descriptive syntactics and descriptive semantics
are based on pragmatics (Carnap [6)).
As seen by Camap, the distinction between pure and descriptive
syntactics or semantics is equivalent to the distinction between
logical and extralogical syntactics and semantics. Logical syn-
tactics is that branch of the logical analysis of language, in which
the most important, fundamental concepts, such as consequence
relation, theorem, analytic sentence, pertain to the form of
expressions and to the relations that hold between those ex-
pressions solely relative to their shape. Thus the logical syntactics
of a language establishes its vocabulary, and formulates rules of
construction of sentences and transformation rules; the vocabulary
and the axioms can just be listed, the rules of construction can be
formulated without reference to the meaning of words, and the
rules of direct consequence relation can be given by listing the
30 J. Pelc

transformations that yield the direct consequence of the sentence

to be transfonned. The syntactics of the language of logic, inter-
preted in this way, was carried on, for example, by Stanislaw
Lesniewski, Jan Lukasiewicz, and Alfred Tarski in their metalogical
research, and by David Hilbert in his works in metamathematics
(cf. Ajdukiewicz (22], pp. 7-28). Likewise, a non formalized
logical syntactics of a language - including a natural language -
forms that branch of the logic of language in which reference is
made only to expressions and to relations that hold between them
relative to the external properties of those expressions, nothing
being said either about the objects to which those expressions
refer or about relations between expressions and objects or about
the mental experiences of the users of those expressions (ibid.,
pp. IS 5 -15 9). It is assumed in this connection that the syntactic
metalanguage does not include the object language, and, conse-
quently, sentences formulated in the syntactic metalanguage are
not translatable into sentences about things (Ajdukiewicz [23],
pp. 182-191). Also syntactic concepts, such as, for example, the
concept of sentence as an inscription of a specified shape, deter-
mined by the syntactic rules of construction of expressions, the
concept of consequence relation relative to certain transformation
rules, should be distinguished from semantic and pragmatic con-
cepts.
Logical semantics makes use of concepts that refer to relations
between expressions and extra-linguistic reality, such as the con-
cepts of denoting, truth, and satisfaction. As has been shown by
Tarski [26], the classical concept of truth can be defined only in
the semantic metalanguage, i.e., one having the object language as
its part. Model theory consists of the logical syntactics of a
language treated as a noninterpreted calculus and of an interpre-
tation of the language, viz., a set-theoretical approach to ontology,
describing reality as a hierarchy of sets. Thus, a semantic model of
a given deductive theory is a system of broadly interpreted objects
that satisfy the axioms of that theory and are denoted by its
primitive terms or are in the range of the variables that occur in its
axioms (Ajdukiewicz [25]).
Logical pragmatics, the third branch of logical semiotics, makes
use of concepts that refer to relations between language users and
expressions, e.g., the concept of a sentence accepted by a person
 The place of the philosophy of language 31

as a sentence he would be ready to adopt as a basis for action

(Ajdukiewicz [23], pp. 282-294) or the concept of the situation
that accompanies the use of an expression. Pragmatic meta-
language is accordingly richer than the semantic one: it includes
also names of the users of the object language and names of places
and indications of moments in which its expressions are used.
Modern formalized logical pragmatics takes into account not only
relations mentioned above, as was being done, e.g., by Carnap, but
also other elements, such as situational context or context of use;
it accordingly covers tense logic, topological logic, etc. (Montague
[27] ; Rescher, 1968). The logical pragmatics disregards, however,
all emotional speakers' attitudes toward expressions they emit
and/or receive. Thus the user of expressions is treated as one of
the factors in a given situational context, next to the circum-
stances that accompany the utterance - it being assumed that
changes in that context modify relations between the sender or
the receiver of the message and the message itself, and that, con-
versely, the modification is by assumption restricted to the
semantic sphere: the language user, affected by a change in the
context, assigns a given expression an extension that differs from
the former (Bar-Hillel [28], pp. 359-379). Contemporary for-
malized systems of logical pragmatics are based on various con-
cepts, such as that of the acceptance, by a given user at a given
time, of a given sentence (Martin [29]) or the concept of con-
text of use that consists of the speaker, the listener, the time and
place of a given utterance, the indicating gesture, and the possible
worlds. An ordered set of those elements forms the so-called
index of an expression, and each element is a coordinate of that
index. Thus, in this interpretation, a formal pragmatic represen-
tation consists of the set of indexes, the set of possible objects,
and the function that assigns interpretive properties to expressions
in a given language. This triplet is termed a possible interpretation
of a language, and in logical pragmatics plays the role that is
analogous to that of a semantic model in formal logical semantics
(Montague [27], [30], pp. 68-94; [31], pp. 373-398). The
modern formalized logical pragmatics sets itself, so far, narrower
tasks than does pragmatics as understood by Morris or
Ajdukiewicz: the latter described it as a discipline that analyses
relations between speech and thinking, the social function of
32 J. Pelc

language, its ontogenesis and phylogenesis, etc.; he also made the

acceptance of sentences dependent on observations, and inter-
preted the meaning of expressions as the mode of understanding
them, that mode being fixed by linguistic usage: he thus took
psychological and sociological factors into consideration, and,
consequently, had rather descriptive pragmatics in mind
(Ajdukiewicz [22), pp. 155-191).
To return to the distinction between pure and descriptive
semiotics: one may be at variance with Carnap's opinion that only
pure syntactics (semantics) is the logical one, and instead claim -
in agreement with Ajdukiewicz's view - that although pure semi-
otics always is the logical one, but not conversely: logical semiotics,
in the broader sense of the term, happens to be pure or descriptive.
It is descriptive when it uses logical means to analyze syntactic,
semantic, and pragmatic features of definite languages or other
sign systems or sets of signs that are not sign systems. Each branch
of descriptive logical semiotics refers, at least implicitly, to the
remaining two, because it is concerned with expressions or other
kinds of signs that have definite shapes, definite meanings or other
semantic properties, and are used by some people and/or animals.
An example of a study in descriptive logical semiotics is provided
by Ajdukiewicz's 'Conditional Statement and Material Implication'
(Ajdukiewicz [23), pp. 222-238). While logical semiotics can,
but need not be, descriptive, extralogical semiotics is always
descriptive in the sense of being an empirical investigation of
certain given sign systems and/or individual signs.
Semiotics, whether logical or extralogical, can be general if
concerned with all existing or possible languages and other sign
systems and, for example, if a generalized concept of language is
constructed for that purpose; the discipline is specialized if con-
cerned with a selected language or languages, or other sign
system(s), etc.
Semiotics is formal when its theorems are constructed solely of
logical constants and variable symbols and are thus deprived of
interpretation; it is nonformal otherwise. Finally, it is formalized
- or in Carnap's usage: pure - when it assumes the shape of a
deductive theory; it is non formalized otherwise.
The term logical semiotics may be interpreted more broadly
than the logic of language: it may be held - despite what Adju-
 The place of the philosophy of language 33

kiewicz maintained when he identified logical semiotics with the

logic of language - that the former has as its subject matter not
only language and speech, but all sign systems, and also other sets
of signs, i.e., those not forming systems, as well as individual signs
of various kinds. Likewise, the term semiotics may be given a
broader extension than was done by Ajdukiewicz, who saw in it a
theory of language (Ajdukiewicz [23J, pp. 182-191), and may be
treated as the science concerned with all kinds of signs and sets of
signs.
The above description of logical semiotics leads to the con-
clusion that the discipline, when conceived as the logic of language,
is part of the philosophy of language, and when interpreted more
broadly it partially encroaches upon the subject matter of the
philosophy of language. Semiotics as a whole, if interpreted in a
narrower way, i.e., as the science of language, includes the
philosophy of language as its part, on assumption that the latter is
concerned with language alone, and not with other kinds of signs.
Finally, semiotics sensu largo contains the philosophy of language,
even if the latter is interpreted broadly.

BI BLiOG RAPHY

Ajdukiewicz, Kazimierz [22J Ie yk i poznanie. wybOr pism z lat 1945-

1963, tom II (Language and Knowledge, Vol. II: Selected Papers
1945-1963). Warszawa: PWN, 1965.
-- [25J Pragmatic Logic. Translated by O.A. Wojtasiewicz. Synthese Library,
Vol. 62. Dordrecht: D. Reidel/Warszawa: PWN, 1974.
-- [23J The Scientific World-Perspective and Other Essays, 1931-1963.
Edited and with an Introduction by Jerzy Giedymin. Synthese Library,
Vol. 108. Dordrecht - Boston 1978: D. Reidel, 1978.
Arnould, Antoine, and Nicole, Pierre [24J La logique ou l'art de penser.
1662.
Ayer, Alfred J., Ed. [7J Editor's Introduction. In Logical Positivism. New
York: The Free Press, 1959.
Bar-Hillel, Yehoshua [28] Indexical Expressions. Mind 63 (I 954).
Bloomfield, Leonard [20] Language. New York 1933.
_. [21] Linguistic Aspects of Science. In International Encyclopedia of
Unified Science, 1,4. Chicago 1939.
Bochenski, J.M. [16] The Methods of Contemporary Thought. Translated
by Peter Caws. Dordrecht: D. Reidel, 1965.
34 J. Pelc

Camap, Rudolf [6] Logische Syntax der Sprache. 1934.

- [8] Introduction to Semantics. Cambridge, Mass., 1942.
Fodor, Jerry, and Katz, Jerrold J. [5] What's Wrong with the Philosophy of
Language. Journal ofPhilosophy 59 (1962).
Hi~, Henry [14] Logical Basis of Semiotics. In T.A. Sebeok, Ed., Perfusion of
Signs. Bloomington: Indiana University Press, 1977.
Husserl, Edmund [13} Logische Untersuchungen. Halle 1913.
Kotarbinska, Janina [15] Poj cie znaku (The Concept of Sign). Studio Logica
VI (1957).
Kotarbinski, Tadeusz [1] Spor 0 desygnat (A Controversy Concerning the
Concept of Designatum). Prace Fil%giczne XVIII (1963), 1.
- [32] Gnosiology. Oxford/Wroclaw, 1966.
Ueb, Hans-Heinrich [17] On Subdividing Semiotic. In Y. Bar-Hillel, Ed.,
Pragmatics of Natural Languages. Dordrecht: D. Reidel, 1971.
- [18] On Relating Pragmatics, LinguistiCS, and Non-Semiotic Disciplines.
In A. Kasher, Ed., Language in Focus. Dordrecht: D. Reidel, 1976.
Locke, John [9} Essay Concerning Human Understanding. Ed. A.C. Fraser.
Oxford 1894.
Martin, Richard M. [29] Toward a Systematic Pragmatics. Amsterdam
1959.
Montague, Richard [27] Pragmatics. In R. Klibansky, Ed., Contemporary
Philosophy: A Survey. Firenze: La Nuova ltalia Editrice, 1968.
- [30] Pragmatics and Intensional Logic. Synthese 22 (1970).
- [31} Universal Grammar. Theoria 36 (1970).
Morris, Charles [12] Writings on the General Theory of Signs. Approaches
to Semiotics 16. The Hague/Paris: Mouton, 1971.
Peirce, Charles Sanders [10] Collected Papers of Charles Sanders Peirce. Eds.
Ch. Hartshorne, P. Weiss, A.W. Burks. Cambridge, Mass., 1965-66.
Rescher, Nicolas [33] Topics in Philosophical Logic. Dordrecht: D. Reidel,
1968.
Rosenberg, Jay F., and Travis, Charles, Eds. [2] Readings in the Philosophy
of Language. Englewood Cliffs, N.J.: Prentice-Hall, 1971.
Sapir, Edward [19] The Status of Linguistics as a Science. Language 5 (1929).
de Saussure, Ferdinand [11] Cours de linguistique genera/e. Ed. R. Engler.
Wiesbaden 1967.
Tarski, Alfred [26] Pojc::cie prawdy w j~zykach nauk dedukcyjnych (The
Concept of Truth in the Languages of Deductive Sciences). Sprawoz-
dania TowarzystwaNaukowego Warszawskiego. Warszawa: TNW, 1933.
Wittgenstein, Ludwig [3] Logisch-philosophische Abhandlung (Tractatus
Logico-Philosophicus). Annalen der Naturphilosophie XIV (1921).
[4] Philosophische Untersuchungen (Philosophical Investigations). Eds.
G.E.M. Anscombe, R. Rhees. Oxford: Basil Blackwell, 1953.
The theory of meaning in analytical philosophy
CHRISTOPHER PEACOCKE
Oxford University

This survey will be restricted in scope not only in the respects indi-
cated in the title and by a limitation to fewer than thirty pages,
but also by the inevitable omission of discussion of the semantics
of particular expressions - proper names, demonstratives, natural
kind words, predicate and sentential operators, and so forth - the
upshot of which cannot fail to have some effect on our conception
of a semantic theory. But even within these restrictions, the field
is a vast one; and it will help to structure our discussion if we
impose a division corresponding to the following questions:
(I) What form should a theory that specifies the meaning of all
the sentences of a particular language take? I will follow the con-
vention of calling such a theory for a particular language a
'meaning theory' (MT) for that language, and will reserve the
phrase 'theory of meaning' for a theory (together with the argu-
ments for it) about the correct form for a MT.
(II) In virtue of what is one rather than another MT applicable
to a particular language in use in a community of speakers?
(III) What are the various constraints upon and infirmities of
the relations between our answers in questions (I) and (II)?
Though a theorist's motivation for answering anyone of these
questions in a particular way can hardly be independent of the
views he holds about the others, it is possible for different
theorists to agree on the answers to some (notably (I) and to
disagree on others.

Contemporary philosophy. A new survey. VoL I, pp. 35-56.

©1981, Martinus Nijhoff Publishers, The Hague/Boston/London.
36 C Peacocke

I. WHAT FORM SHOULD A MEANING THEORY TAKE?

I shall enumerate six answers of the last twelve years to this ques-
tion, before considering how they fare in respect of conditions of
adequacy derivable from not much more than the idea of a MT.
These six answers are not mutually exclusive: there are philos-
ophers and linguists who would want to combine some of them
together, each part of a total theory fulfilling a different role
demanded of a MT.
(i) Translational semantics. According to an idea inherited from
Fodor, Katz and Postal, a MT should consist of a mapping of sen-
tences of the language for which a MT is to be given into a dis-
ambiguated language, a language that perspicuously exhibits
various relations of entailment between predicates and possibly
other expressions (Katz [I], [2]). Though in generative semantics
the idea of a syntactically identified level of deep structure to
which a translational semantics is applied was abandoned, the con-
ception of interpretation as in part translation into something or
other was not: and it is only a variation on the fundamental theme
given by this idea if it is suggested that the semantic representation
use first-order logic. (For a survey, see Fodor [3].)
(ii) Model-theoretic semantics. A view developed notably by
Montague [4] and his students is that a MT for a language L must
consist of a definition of truth in an arbitrary model for each sen-
tence of L. Almost all model-theoretic semanticists agree either in
their metatheoretical commentary, or show by their actual prac-
tice, that they would in a MT for a language pick out a dis-
tinguished one of the indices relative to which models are defined,
the actual world, the present moment, or whatever; Montague [5]
wrote explicitly that this ought to be done. But beyond this point
of agreement, there are major distinctions between the kinds of
model-theory that have been advocated. (Some are perhaps not
happily called 'model theory' at all.) (a) There are those who hold
that a MT must specify the extensions of expressions not only
with respect to the actual world, but also with respect to ways
things might have been; thus David Lewis [6]. On this view, the
assignment of extensions to expressions with respect to nonactual
worlds is meant to capture the actual meaning of the expression in
the language being treated: in the case of English, with respect to
 The theory o/meaning in analytical philosophy 37

any world at all, a correct MT will assign to 'green' the set of

things that are green in that world. (b) There are model theorists
whose constructions are not so constrained in respect of the non-
actual worlds: the role the model theory is conceived as playing
does not require that they be so. Nevertheless the interpretation of
the logical constants does not on this second version vary through
the models; thus, apparently, Montague [4]. (c) There are those
who would employ model theory in a MT in such a way that it is
intended to express some notion of structural validity of infer-
ences in the language: what varies and what does not between the
models is meant to be a consequence, not a presupposition, of a
theory of this concept of structural validity; thus Evans [7 J . One
of the reasons that the label 'possible-worlds semantics' is of little
value in a survey of this kind is that it may cover not only anyone
of these three kinds of position, but also particular MTs falling
under (iii) below.
(iii) Truth-theoretic semantics. One of the major developments
of this period has been the adumbration by Davidson of the pro-
posal that a MT for a language L should consist of a finitely
axiomatized theory of truth for the sentences of L, or for the sen-
tences of some language from which these sentences can be
generated by meaning-preserving transformations [8], [9], [10].
In question here is absolute truth, not truth in a model; such rela-
tivization of the truth predicate as there is in such a theory will be
to accommodate indexicality, or, if satisfaction is regarded as a
relativization of truth, to accommodate the fact that closed sen-
tences are sometimes composed of open ones. Both (ii) and (iii)
take their inspiration from Tarski, and both have generated
research programs designed to treat particular object language
(OL) constructions with their respective conceptions of a MT: see
for example Davidson and Harman [11] , [12] .
(iv) Game-theoretic semantics. Inspired by the development of
game-theoretic techniques in the study of logical validity, Hintikka
has proposed that a MT for a language L will be a set of rules
which associate with each sentence of the language a game; the
game can be regarded as played between the asserter of the
sentence and Nature (Hintikka [13a], [13b], Saarinen [14]). At
each stage, a move is made depending on the form of the sentence,
and the move requires either the asserter or Nature to make a
38 C. Peacocke

choice; as a result of each move a sentence oflesser length is yielded

with respect to which the game is continued. The asserter wins the
game if the game ends with a true atomic sentence, and a complex
sentence can be defined as true if its asserter has a winning
strategy in the correlated game. This general conception has been
applied in some detail to natural language constructions; but
many, perhaps all, of the results on natural language sentences can
be taken over by other theories of meaning, for the game-theoretic
rules easily generate corresponding truth- and model-theoretic
axioms (Peacocke [15], Hintikka [ 16 J ).
(v) Criterial and constructive semantics. One view of the correct
form of a MT was widely employed in discussions within the
philosophy of mind in the 1950s and 1960s, and within our survey
period its bases have been made more explicit. It is regarded by its
proponents as being a statement of the views of Wittgenstein in his
later works. The view is that to specify the meaning of a sentence
is to specify what are the criteria for it (Baker [ 17] , Hacker [ 18] ,
Lycan [19]). If p's being the case is a criterion for the sentence
s, then p is said to provide noninductive evidence for s, but does
not entail the truth of s: it can be the case both that p and that
s is false. There are many criteria for a given sentence, and criteria
are defeasible; some hold that any criterion for a given sentence
can be defeated if we are clever enough in thinking up circum-
stances. The concept of a criterion has been used by some of its
advocates in attempts to answer or defuse scepticism about other
minds or the material world, but this is a claimed consequence of
the theory of criteria, and not part of the definition of the notion.
For some theorists it is undecidable sentences that particularly
stand in need of criteria for a specification of their meaning, and
then the criteria themselves must be decidable: Wittgenstein's
arguments about following a rule apply to both decidable and
undecidable notions, but in the undecidable case it is particularly
clear that there is nothing internal that guides the speaker in the
use of these notions on their classical interpretation. In any case it
is essential that the arguments that forced the inroduction of the
notion of a criterion not be applicable in turn to the criteria them-
selves. A systematic theory that is attractive to these theorists is
the intuitionistic semantics for arithmetic in terms of the theory
of constructions: here an account of the meaning of a sentence is
 The theory o/meaning in analytical philosophy 39

given in terms not of what it would be for it to be true, but in

terms of what something must be like to be a proof of the sen-
tence (Dummett [20], Kreisel [21], Wright [22]).
(vi) Conceptual-role semantics. One natural way of extending
the functionalism held at one time by Putnam is to suggest that to
specify the meaning of a sentence is to specify a thought, where
'thought' is used not in the sense of Frege, but in a sense in which
to specify a thought is to specify a role in the network of the
agent's perceptions, psychological states and actions. This
extension was proposed in outline by Harman [23], [24]. On a
formal development of the view one would expect the meaning of
a sentence s to be determined by (a) which sensory stimuli pro-
duce belief and which disbelief in s (b) from which sets of sen-
tences one infers to s, and with what probability (c) what one can
infer from s, and with what probability (d) the effects of belief
and desire expressed by s on action. In each case (a)-(d) ought
to be specified relative to every set of auxiliary hypotheses
believed (and in the case of (d), relative to desires too). A theory
dealing just with (b) and (c) along these lines was developed by
Field [25].
A TM for a language L must be such that if someone knows the
theory and knows that it meets the conditions required to be a
theory of meaning for L, then he is in a position to understand
sentences of L. Lewis and Davidson have argued forcefully that
translational semantics does not meet this condition: one can
know how to translate the sentences of one language into another
without understanding any of the sentences of the first language.
It has been said against this argument that theories such as (ii) to
(iv) as we have characteriz~d them here that do meet the con-
dition of adequacy presuppose an unexplained competence in the
metalanguage in which the theory is stated; and that if this is
allowed then the argument against (i) falls because the objection
fails to apply if we presuppose competence in the language into
which the translation takes us. (Thus Harman [23], [24].) But
this rejoinder confuses two different roles. In a theory of kind (i),
three languages are in play: there is the metalanguage in which the
theory is cast, a language which certainly must be understood if
the theory is to be understood, there is the language being tran-
slated and there is the language into which the translation maps.
40 C. Peacocke

In theories of kinds (ii)-(iv) there are two languages in question,

the metalanguage of the theory and the object language for which
the theory is a MT. The analogue of the required understanding of
the metalanguage in the cases of (ii)-{iv) if the theory is to be
understood is understanding of the metalanguage in which the
translational manual is specified. The requirement of under-
standing of the language into which the translation is given, the
requirement that must be met if the condition of adequacy is to
be met, is additional. A theory of kind (i) is adequate only if we
tacitly make use of a theory of meaning in the case of one language
where this theory is not a translational manual: but if such
theories can be given, why not give one outright for the object
language and cut out the detour? This general objection to trans-
lational semantics does not seem to rest (as Katz suggest [26]) on
a failure to distinguish sense and reference: someone can know a
translation manual from L to L' and know that it is meaning-
preserving without knowing anything about either the sense or the
reference of sentences of L.
Since theories (ii)-(iv) determine the absolute truth conditions
of sentences of the OL they treat, and since in the specification of
the truth condition that they supply a sentence that means the
same as the OL sentence is used and not mentioned, they meet the
condition of adequacy about understanding. What of theories (v)
and (vi)? In the case of criterial semantics, there are two sets of
issues. The first set concerns the form of the criterial specifications
of meaning supposed to be possible. But first a preliminary remark:
in order to set aside one irrelevant complication, I supposed in the
above description that 'is a criterion for' takes a term for a sen-
tence and a used sentence to form a sentence, rather than taking it
just as a relation between sentences (as do Baker [17] and Hacker
[18]). If the latter course were followed, MT's of kind (v) would
be open to all the objections to translational semantics. The set of
issues about the form of criteria is raised by the thought that in
many (perhaps all) cases in which a concept is given criteria, it
does not seem to be strictly a priori that these criteria are good
evidence for the application of the concept unless the criteria
embed the concept in question itself. The minor problem is then
that it is not clear that there are any decidable conditions that
embed the concept itself and which are also a priori good evidence
 The theory of meaning in analytical philosophy 41

that the concept applies. The major problem is that if a con-

dition embedding the concept is irreducible to conditions that
do not embed it (as the thought prompting these considerations
implies) then it is plausible that there will be no manifesting a
grasp of these conditions without manifesting a grasp of the
embedded concept; yet it was grasp of that concept that the
introduction of criteria was to help to explain, and not take for
granted. Proponents of type (v) TM's must either face these
questions or else show that the supposition that raised them is
incorrect.
The second set of issues is raised by the query: does provision
of a list of criteria for a sentence suffice to specify what that
sentence says? Strawson and others would deny that it does [27].
Wright replied that the objection is answerable by giving a homo-
phonic MT recursively stating the extension of 'is assertible (in
state of information i)' [28]. In the later part of our period it
became a conventional view that there was no difficulty in giving
constructively acceptable homophonic MT's (Wright [22],
McDowell [29]). Certainly it is true that there is no obstacle to
deriving all the T-sentences for a language for first-order arith-
metic in a homophonic truth theory using only intuitionistic logic
in the metalanguage: the disquotational negation axiom, for
instance, if all the connectives are read constructively, is accep-
table under this interpretation. But there are, first, difficulties of
principle in extending this to empirical discourse: we cannot say
that the negation of a sentence s is assertible in i if there is no state
i' of information that both extends i and in which s is assertible,
for in the empirical case this is not a condition we can recognize as
obtaining when it does obtain; and we have no idea how to pro-
vide a constructive version of it. A second general question con-
cerns observationality. Observationality seems the natural analogue
for empirical discourse of decidability in arithmetic (Dummett
[20]). Now the condition that is said to obtain in an observational
sentence is one which can obtain without being observed to obtain,
if observers are not suitably situated. Understanding an obser-
vational sentence involves the ability to establish it as true even
when it is not observed to be true. But such means of establishing
it may involve hypotheses concerning places and times which are
not themselves decidable. It is arguable that in empirical discourse,
42 C. Peacocke

we do not have absolute observationality, but only various rela-

tivized kinds of observationality, relativized to the type of hypoth-
eses used in canonical means of establishing it not by observation.
Conceptual-role semantics has yet to be sufficiently well
developed to permit assessment with respect to the understanding
condition of adequacy: Field's published development simply adds
to the specification of conceptual role a Tarskian truth theory that
avoids the objection. But this imposes an obligation to state how
the conceptual role and the truth theory are related, to avoid
another objection that it is implausible that meaning is determined
by an unrelated pair of notions.
A second condition of adequacy on MT's that is concerned
simply with their form is that they show how the meaning of com-
plex sentences is determined by the meaning of their parts.
Following Davidson's definition of a semantical primitive in the
period covered by the previous survey as an expression such that
the rules that suffice to determine the meaning of the sentences in
which it does not occur do not suffice to determine the meaning
of some of those in which it does, Wallace and later Davies have
considerably sharpened the condition of adequacy this formu-
lation and idea impose on a MT (Wallace [30], Davies [31]).
Roughly the restriction is that the principles used in a MT to
derive a meaning specification for an OL sentence be such that all
and only the sentences meaning specifications for which can be
derived from these principles are specification for sentences that
can be understood if the given sentence is understood. Constraints
along these lines will exclude various trivializing MT's, and also
provide a test for detailed semantical theories of particular OL
constructions. The ability to move on to new cases one has learned
the language must not be confined to previously unencountered
sentences: the ability to apply an atomic predicate to new cases is
also important. It is very far from clear that type (v) MT's accomo-
date this phenomenon: one does not really learn a vast list of
criteria in learning the sense of an atomic predicate of a
problematic kind: one learns something that unifies the criteria,
and which allows one to apply the word in new cases. If this were
not so, there would be no answer to the objection that a predicate
with multiple criteria is ambiguous. On the list view of criteria,
just altering one of the criteria for a given sentence ought to yield
 The theory of meaning in analytical philosophy 43

a new sentence with a determinate sense, and it is not obvious that

this is so: the same point could be made about the specifications
(a)-(d) in MT's of type (vi). Meaning is just what unifies all these
long lists.
It has long been taken as a goal of a MT to account for some
relations of implication amongst OL sentences. This is an area in
which theories have far outrun understanding of the conditions
they are meant to satisfy, in two respects: there is unclarity about
which implications the theory is meant to capture, and there is
unclarity about how the distinguished inferences are to be treated
in the MT. At one point Davidson suggested that any theory in
which all the T -sentences for the OL in question were provable
would necessarily validate those inferences one would want to
distinguish as valid in some favoured way [32]. Grandy [33] and
Evans [7] remarked that this will validate very few inferences, and
certainly not the adverbial modification inferences in which
Davidson was particularly interested himself. Indeed to derive the
T-sentences, only universal instantiation and intersubstitutivity of
predicates and terms provably coextensive in the theory is needed.
There was also the question of whether there was any reason that
first-order regimentations should have any special role: this is in
part the question whether there is anything distinctive of the
traditional logical constants.
It is striking that much of the further work on that last question
in this period has appealed to properties of expressions that are
truth-theoretic (Quine [34], Peacocke [35]). This suggests that
there may be some redundancy in supposing that one needs to
supply a model theory as part of a MT if one's conception of
model theory is that the interpretation of the traditional logical
constants is held fixed through the models: for the relevant
range and kind of models would be fixed by the absolute truth
theory, and it would be in only a formal sense that truth is a
special case of truth in a model. There have been attempts (Evans
[7]) to employ a more general notion of structural validity of an
inference according to which first-order inferences would not be
counted as structurally valid on the ground that they depend on
the sense of particular expressions rather than the way the sen-
tence has been built up: but a fully satisfactory criterion for this
distinction which also does succeed in counting some inferences
as structurally valid has yet to be formulated (Davies [31]).
44 C Peacocke

II. WHAT MAKES ONE RATHER THAN ANOTHER TM APPLICABLE

TO A LANGUAGE IN USE?

Grice's answer to this question was given in an earlier decade, but

was greatly refined by himself in a series of papers and by Schiffer
in a book (Grice [36], [37], Schiffer [38]). The fundamental
ideas are that linguistic meaning can be defined in terms of an
agent's meaning something by (the product of) an action ('s-mean-
ing') and that that second notion is a matter of the agent's having
a complex overt intention to produce a certain response in a
certain audience by means of the audience's recognition of his
intention. The intention is wholly overt in the sense that there is
no intention required for s-meaning that the agent does not
intend his audience to recognize. It follows that the intentions
required for s-meaning either form an infinite series or else they
are self-referential (Harman [39]).
The link between s-meaning and linguistic meaning is generally
said by those working within a Gricean framework to be provided
by the notion of convention: very roughly indeed, the idea is that
for sentence s to mean that p in community C is for it to be a con-
vention in C to utter s only if you s-mean thereby that p. The con-
cept of convention was elegantly investigated by Lewis [40]. In
fact it is not at all clear that language has to be conventional in the
sense of Lewis, and the claim that it has to be was questioned in
the later part of our period (Burge [41], Peacocke [42]). Certain-
ly any further enquiry into convention must distinguish the
following notions. There is Lewis's definition; then there is the
notion that a practice is conventional in the sense that there could
be a community (not necessarily the community in question) in
which the ends served by this practice could be served in a
different way, the laws of nature and people's goals remaining
unchanged; then there is the weaker notion still that a practice is
sustained by common expectations that others will conform to it.
There are also natural definitions of a practice being conventional
with respect to one social goal but not another. For any pair of
these notions, it is not too hard to think of examples in which one
applies and the other does not; and it seems that language is
conventional only on the weakest of them.
However the most fundamental criticisms of Grice's program
 The theory of meaning in analytical philosophy 45

have concerned not the bridge built between s-meaning and

linguistic meaning, but have been concerned with the role of s-
meaning itself. Let us look at some of these.
One such criticism, that since the Gricean account uses propo-
sitional attitudes that have the same contents as the sentences
whose meaning they are used to define the theory gives no
account of what it is to have a concept, we will defer to the dis-
cussion of modesty in Part III. More specifically it is urged that
propositional attitude sentences are to be analyzed as relations to
sentences or utterances with a certain meaning (for instance as in
Davidson [32]), so that if a Gricean reduction held there would be
the circularity that linguistic meaning is explained in terms of
propositional attitudes, and those attitudes in terms of linguistic
meaning. This objection is inconclusive not only because sentence-
based analyses of propositional attitude sentences need not
actually use 'means that' (or a synonym) in their analysis; it is also
inconclusive because there is a dilemma applicable to the supposed
account of propositional attitude sentences. Either the account
makes the OL sentence 'John can intend to shut the door regardless
of what utterances and sentences mean what' true or it does not.
If it does not, then the account is inadequate; and if it does, then
there is no circularity in combining it with Grice's idea. In neither
case is there an objection of principle to Grice's programme.
A second objection arises from the fact that one cannot ascribe
detailed intentions to a language-using agent in advance of inter-
preting much of his language (since that is such a large part of his
intentional behaviour). This point, as is emphasized by Davidson
[ 43], [44], of course shows that Gricean intentions cannot be
used as evidential base in the enterprise of radical interpretation,
and not that the Gricean reductions are not true. It is also indeed
a point that a Gricean ought to embrace, if he holds to some
degree the holism of propositional attitude psychology; linguistic
behaviour is especially rich and structured, and is also a form of
direct expression of attitudes of the agent, so any ascription of
propositional attitudes to the agent must be answerable to this
source of evidence, and will be rejected if it cannot account for
such behaviour.
A more serious source of difficulty is that the Gricean reductions
are presumably confined to the use of nonsemantic propositional
46 C Peacocke

attitudes in the definitions, that is attitudes that do not embed

'refers', 'true', 'means (says) that'. It has yet to be shown that
without such attitudes a definition of linguistic meaning can be
given. (Indeed this is one interpretation of what Searle has been
saying for some time [45].) It is particularly plausible that such
semantic attitudes would enter a proper account of disambig-
uation on a particular occasion of use of ambiguous sentences; the
utterer intends his audience to recognize that his utterance has one
truth condition rather than another. It should be noted too that
there is no immediate nonquestion-begging argument from the
claim that if two people have all their nonsemantic propositional
attitudes in common they will have the same dispositions to verbal
behaviour, so the linguistic meaning of their expressions will be
the same, hence some form of Gricean account must be correct.
The point is not just that there is not necessarily any move from
supervenience to reducibility (after all that gap may not be impor-
tant here, since if the semantic does supervene on the psychologi-
cal, it would be plausible that the semantic descriptions could be
known if the psychological ones were, and this could be so only
if there were connecting principles). It is rather that dispositions
are specified by the triggering conditions for their manifestation;
and now if the triggering condition is a semantic propositional atti-
tude, the verbal dispositions of two people with the same non-
semantic propositional attitudes need not be the same. We must
not beg the question by assuming that this semantically specified
disposition is already included in the nonsemantic dispositions.
But all this is only to criticize an invalid argument to a conclusion,
and not the conclusion itself.
A different kind of answer to question (II) was given by David-
son [43], [44]. It is of course not surpriSLrlg that one who adopts
a truth-theoretical conception of a MT should be concerned to
give a nontrivial (nondisquotational) answer to question (II) since
it is evident that more is required for an extensional truth theory
to be interpretational than its mere truth. Davidson's answer con-
sists in a suggested account of how we should identify an interpre-
tational truth theory as correct when engaged in radical interpre-
tation. The proposal very roughly stated is that we first identify
which sentences the OL speakers hold to be true; we then select as
interpretational that truth theory which assigns truth conditions
 The theory of meaning in analytical philosophy 47

to OL sentences in such a way that, subject to certain epistemo-

logical constraints, no other theory makes them hold more sen-
tences that are, by our lights, true. The crucial point is that
although it is agreed that to identify instances of holding-true we
need to know something about an agent's desires and beliefs, we
can know that someone holds a sentence to be true without
knowing its meaning on his lips or what belief he expresses by
uttering it. Any truth theory actually selected - there may be
more than one if indeterminacy holds - is of course not an
approximately true theory of the conditions under which OL
sentences are held true: it is a precise theory of the conditions
under which they are true.
As an answer to question (II) this position must be equipped
with a reply to the objection that it gives an account of how to
discover that a truth theory is interpretational, but not an account
of what it is for the theory to be interpretational, what it is that
this procedure succeeds in discovering. One way of developing a
reply is to argue that in other cases in which we draw a contrast
between methods of discovery and what is discovered we have
some conception of an alternative route to the application of the
concept in question besides the procedure under consideration.
That is not so with the concepts of truth and saying. The language
leamer's route, which certainly does not involve the application by
the learner of anything like Davidson's radical interpretation pro-
cedure, could be accommodated by this general position with an
argument that the first-language learner comes noninferentially to
apply the notion of saying in conformity with the extension of the
notion determined by a radical interpretation procedure.
At least two questions about holding true seems worthy of
further investigation. The first concerns an argument that the fact
that one can know that a man holds a sentence to be true without
knowing what it means does not by itself show that holding true
can play its intended role in radical interpretation. For it might be
the case that whenever one does know that a sentence is held true
without knowing its sense, this is because the believer is in some
ways on this particular occasion similar to the way he or others
have been when they have held an uttered sentence true; and it
might be that on these occasions one's grounds for believing that
the sentence is held true include the fact that the best total theory
48 C. Peacocke

makes the person in question believe that p and interprets this

sentence as meaning that p. Second, and more generally, does
there have to be a notion playing the role Davidson envisages for
holding true? Could we not proceed in interpretation by trying
out total combinations of truth theory and ascriptions of belief,
desire and intention, and seeing whether they in combination
satisfactorily rationalize all the agent's linguistic behaviour?
Initial and revisable hypotheses about the manifestation of holding
true may be helpful in thinking of plausible combinations to test,
but this may be true too of initial guesses about detailed attitudes
and sentence senses.
Davidson's method of moving from the evidence in terms of
holding true to a truth theory by means of the Principle of Charity
is open on some construals of it to objections from causal theories
of relational propositional attitudes (Evans [46], Grandy [47],
McGinn [48]): the subject should be more fully detailed in the
volume in this survey dealing with the philosophy of mind.

III. WHAT ARE THE VARIOUS CONSTRAINTS UPON AND

INFIRMITIES OF THE RELATIONS BETWEEN OUR ANSWERS
TO (I) AND (II)?

A full answer to this question (III) must include an adequate view

of the relations between a MT and both physics and psychology:
for vigorous statements from particular stances on these issues, see
Field [49], [50] and Fodor [51]. In order to avoid an even
greater degree of sketchiness than was displayed in the previous
sections of this article, I will concentrate on the two issues of
indeterminacy and manifestation and knowledge of semantic
properties.
Issues about manifestation and knowledge of semantic proper-
ties clearly relate to general theories of meaning rather than MT's,
since questions left open by a MT may be settled in the
surrounding general theory of meaning. A convenient starting
point for discussion is the notion of modesty introduced by
Dummett [52]: a general theory of meaning (TM) may be des-
cribed as modest if it insists that MT's for particular languages can
only state which concepts are associated with which expressions in
 The theory of meaning in analytical philosophy 49

the language it treats. Dummett sometimes says that a MT

endorsed by a modest TM is a theory that, for any concept it
associates with some object language expression, "would be
intelligible only to someone who had already grasped the con-
cept": this can make it seem that to object to modesty is to
insist on a theory that could be understood by someone who did
not already have the concepts expressed in the theory. But in fact
there are several genuine issues here.
The first issue concerns what it is to have a concept. Is it ever
correct to offer a condition for what it is for a person x to have a
concept that embeds an expression for that concept inside a
propositional attitude or indirect discourse verb that is applied to
x? One reading of modesty is that this could never be correct. It is
with respect to this issue that nonhomophonic clauses in the MT
have become associated with modesty: for a reduction of a con-
cept to others, provided that what it is to have the reducing con-
cepts is not problematic, would help to meet his demand. Since it
is also obvious that this process must stop somewhere, issues over
modesty are related too to the conditions of legitimacy of homo-
phonic clauses. In particular, are they legitimate only for decidable
notions? The second issue is what makes an axiom of a MT
correct. Must some account of this be given in ultimately non-
semantic terms or not? (On one of the views of knowledge of
language discussed below, there would be no reason to suppose
such a nonsemantic account must be available.)
A third issue which still arises if we are modest in both the pre-
vious two respects but which is not identified by Dummett is this.
Within modest theories of meaning we can distinguish what I will
label taciturn from diffident theories of meaning. A diffident
TM gives an account for each concept expressed by an object
language expression of what is an appropriate manifestation of a
belief involving that concept, an account that uses that very con-
cept but which does not presuppose that the object language
speaker has that concept. This is a middle position: for though it
offers some account of what can make it correct to attribute a
concept, it offers no reduction of the concept in question, nor
need the concept be decidable for a diffident account to be
available. Examples of diffident accounts are Quine's verdict
tables for the sentential connectives and Ramsey's account of
50 C. Peacocke

what it is to believe a universally quantified sentence in terms of a

disposition to form singular beliefs (Quine [53], Ramsey [54]).
This disposition is in the case of the Ramsey example quite
unobjectionably specified using the universal quantifier itself, and
this is what makes the account diffident. One of the attractions of
diffidence is that it offers an account for each English expression
of what makes it correct to interpret expressions in different lan-
guages of very different levels of sophistication by that English
expression. A taciturn TM on the other hand says nothing about
each particular kind of expression, and an advocate of taciturnity
will have to show why the alleged attractions of diffidence are
illusory. I f diffidence is a correct demand upon a theory of
meaning, it defines a goal for a major philosophical programme:
for it requires that for every concept we have some account of the
manifestation conditions of possession of that concept.
The question of whether speakers of a language must know
what is stated by the axioms of a MT for their language is of
course not settled by whether that theory is realist or construc-
tivist. On the one hand one might be able to answer constructivist
doubts about manifestation of recognition-transcendent beliefs
without claiming that the realistic semantical axioms are known to
speakers; while on the other hand adequate manifestation of the
correctness of a semantical axiom in a constructive theory of
meaning does not imply that the semantical axiom is known to
speakers. Nevertheless several theorists have claimed that com-
petent speakers must know, though of course only tacitly, the
axioms of a correct MT for their language. The position is held
by some followers of Chomsky (Graves et al. [55]); Chomsky'S
own position is very cautious, and he seems more concerned to
insist on the internal realization of the axioms of a correct
grammar rather than on knowledge of them by the speaker [56] .
The most defensible view that stands opposed to it is one that is
the analogue for semantics of the view suggested by Stich for
syntax [57]. This view is that competent, understanding speakers
have knowledge only of what is said by complete sentences with
which they are presented in speech, writing or thought: to under-
stand a language is to have the capacity to acq uire such knowledge
noninferentially. (As in the case of perceptual knowledge, to say
that the knowledge is noninferential is not to say that it could not
 The theory of meaning in analytical philosophy 51

be justified, but only that any justification is not in fact the route
by which the speaker arrived at it.) On this position there is not an
infinite amount of knowledge possessed by the understanding
speaker, but only a potentially infinite capacity: any requirement
that an infinite amount of knowledge in a single person have a
finite basis from which it is inferred would thus not be applicable.
This view, according to which nothing in a correct account of
understanding demands that the axioms of a MT be known by the
speakers of the language, ought not to be argued for on the ground
that all knowledge, or even only semantical knowledge, of a lan-
guage-using creature must be explicit. For there could be a
community of persons with a language that is understood by the
members of the community and who thus know on this view what
particular utterances say: yet nothing in this description of the
case requires that their language actually contain a word that
means 'says' (in the indirect sense) or 'true' or anything else that
they could use verbally to express this knowledge about particular
utterances. The view ought rather to be defended by arguments
from the very great sophistication of the concepts that may
need to be used in a theoretical description, both semantical and
syntactical, of their language. In particular, if we do attribute to a
person beliefs involving these sophisticated concepts as part of our
account of understanding a language, then we must explain why
these people do not behave on other occasions as one would
expect someone possessed of these concepts (such as sequences,
transformational cycles, and so forth) to behave. The semantic
analogue of Stich's view of syntax says that there is no such
knowledge in the ordinary competent speaker, and the expla-
nation of his not behaving in appropriate ways is simply that
he need not be possessed of the concept. Certainly it is not a
general truth that speakers know the principles that determine the
application of the concepts that are used in the expression of
truths that they do know: else we would have to say that anyone
with binocular vision was sufficiently sophisticated to solve simul-
taneous equations in visually judging distance.
My second topic in this section is indeterminacy. Quine's thesis
of the indeterminacy of translation has almost certainly received
more attention in the journals in our period than any other topic
in the theory of meaning. One way of stating the thesis is that
52 C. Peacocke

there can be two equally good translation manuals from one lan-
guage L into a second, M, which both preserve the verbal dis-
positions of all concerned with respect to the sentences in ques-
tion, and which yet map a given sentence of L onto sentences of M
that differ in truth value. (This formulation does not employ the
notion of meaning.) Quine's emphasis on his grounds for holding it
have altered in this period, as has too the sense in which he sub-
scribes to it. In 1970 he wrote that 'the real ground' for holding it
was the general undetermination of scientific theory by all
actually true (and "pegged") observation sentences [58). Yet he
also said later in our period that the only version of the under-
determination thesis that he would definitely affirm is one saying
that our theory of the world is bound to have empirically (i.e.
for Quine, observationally) equivalent alternatives such that, if we
were to discover them, we would see no way of reconciling them
by translating the predicates of one by open sentences of the other
[59). The sense, then, in which he now subscribes to indeter-
minacy insofar as it rested on the underdetermination thesis must
be correspondingly muted. This description of Quine's present
position meshes well with some remarks of Putnam [60].
Quine distinguishes an argument from above and an argument
from below for indeterminacy. The argument from above - that
is, the argument from underdetermination - is simply that since
theory can vary even though the truth value of all pegged obser-
vation sentences be fixed, the translation of a foreign scientist's
theoretical sentences can vary, preserving his dispositions to verbal
behaviour, even though the translation of his observational sen-
tences is fixed (Quine [58)). But it has been rightly objected by
Putnam [60] that Quine's notion of observationally equivalent
theories is very weak, turning as it does on agreement of the
theories of which indicative observational sentences are entailed by
the theories. A wider notion would require agreement on counter-
factuals stated in observational terms: and it is extremely unclear
that theory can vary while observationality in this sense is fixed.
(Clearly this issue depends in part on whether all counterfactuals
can be reduced to categoricals.)
Quite apart from Putnam's point, Quine's position seems to
involve an asymmetry. Though on Quine's view observation does
not determine theory, he is realistic about the physical theory he
 The theory of meaning in analytical philosophy 53

adopts: from the standpoint of his adopted theory, muons and

curved spacetime are real, independently existing entities. Why
not pick on one of the many acceptable translation manuals and,
from the standpoint of the adopted manual, make unrelativized
assertions about what the natives are saying? Quine's answer
([ 61], [62]) would be that the publicity of meaning requires that
it cannot go beyond what is behaviourally determinate. But as
Michael Friedman notes [63], this will lead to the justification
of Quine's asymmetrical position only if the language learning
that goes on in publicly observable circumstances involves no
theoretical inference, or indeed unification of those public circum-
stances by nonbehavioural notions. (In writing this I do not intend
to endorse Friedman's suggestion that one interpretation of the
indeterminacy thesis is that a MT is not weakly or strongly
reducible to physical theory. That interpretation is problematic
since apparently nothing in the notion of weak or strong
reducibility demands uniqueness in the relevant sense of the
reduced theory. If that is correct, it seems to leave open the possi-
bility that two substantively different MT's for a language both be
weakly reducible to physics.)
Arguments ffOm below for indeterminacy consist in citing
examples of unambiguous sub sentential expressions which may be
interpreted in more than one way: as Quine noted, to be evidence
for indeterminacy, it must be shown that these different possible
interpretations do indeed result in different total interpretations
of the whole sentences in which the expression occurs. The only
example in Word and Object meeting this condition which Quine
developed in detail concerned the interpretation of a word
intuitively taken as translatable by 'rabbit': he claimed that such
translations as 'undetached rabbit part' could be made to work
equally well if compensating adjustments were made in the
interpretation of other parts of the language. A powerful general
treatment of predication was developed by Evans [64] and
applied in argument against Quine's suggestion. The theory is that
taking an expression as a predicate of objects of a certain kind is
to be justified by the role of the putative predicate in contributing
to the assent 'conditions of complex sentences in which it occurs.
Natural constraints on interpretation of an expression as a predi-
cate derived from this conception can be shown to exclude the
54 C. Peacocke

unorthodox interpretations of 'rabbit'. Quine could regard these

constraints as additional canons of interpretation that do not show
there are any more facts of the matter only if he also so construed
all his own remarks about the interpretation of logical constants,
mass terms, terms of divided reference, and so forth in his own
writings. Field also responded to the argument from below by
noting that even if there were such indeterminacy, one could
retain an objective notion of truth without relativizing all semantic
notions to an arbitrarily chosen translation manual [65]; he
further separated certain genuine cases of indeterminacy of
reference that may become apparent after a scientific revolution
from the kinds of example given by Quine (Field [66]). But the
exact extent of indeterminacy, and its relation to various kinds of
semantic and psychological holism is worth further investigation.

BIBLIOGRAPHY

Baker, G. [17] Criteria: A New Foundation for Semantics. Ratio 16 (1974):

156-89.
Burge, T. [41] On Knowledge and Convention. Philosophical Review 84
(1975): 249-55.
Chomsky, N. [56] Knowledge of Language, Minnesota Studies in the Phil·
osophy of Science VII Ed. K. Gunderson. Minneapolis: University of
Minnesota Press, 1975.
Davidson, D. [8] Truth and Meaning.Synthese 17 (1967): 304-23.
- [9] In Defense of Convention T. In H. Leblanc, Ed., Troth, Syntax and
Modality. Amsterdam: North-Holland, 1973.
- [10] Reply to Foster. In Evans and McDowell.
- and Harman, G. [11] The Logic of Grammar. California: Dickenson,
Encino and Belmont, 1975.
- and Harman, G. [12] Semantics of Natural Language. Dordrecht: Reidel,
1972.
- [32J On Saying That. In D. Davidson and J. Ilintikka, Eds., Words and
Objections. Dordrecht: Reidel, 1969.
- [43] Belief and the Basis of Meaning. Synthese 27 (1974): 204-23.
- [44] Radical Interpretation. Dialectica 27 (1973): 313-28.
Davies, M. [31] Meaning and Structure. Forthcoming in Philosophia.
Dummett, M. [20] What is a Theory of Meaning? (II). In Evans and
McDowell.
- [52] What is a Theory of Meaning? In S. Guttenplan, Ed., Mind and Lan-
guage. Oxford: Oxford University Press, 1975.
 The theory of meaning in analytical philosophy 55

Evans, G. [7] Semantic Structure and Logical Form. In Evans and McDowell.
_. [46] The Causal Theory of Names. Proc. Arist. Soc. s.v. 47 (1973): 187-
208.
- [64] Identity and Predication. Journal of Philosophy. 72 (1975): 343-63.
- and McDowell, J. [67] Truth and Meaning. Essays in Semantics. Oxford:
Oxford University Press, 1976.
Field H. [25] Logic, Meaning and Conceptual Role. Journal ofPhilosophy 74
(1977): 379-409.
- [49] Tarski's Theory of Truth. Joumal of Philosophy 69 (1972): 347-75.
- [50] Conventionalism and Instrumentalism in Semantics. Noas 9 (1975):
375-406.
[65] Quine and the Correspondence Theory. Philosophical Review 83
(1974): 200-28.
- [66] Theory Change and the Indeterminacy of Reference. Journal of
Philosophy 70 (1973): 462-81.
Fodor,J.A. [51] The Language of Thought. New York: Crowell, 1975.
Fodor, J.D. [3] Semantics: Theories of Meaning in Generative Grammar.
New York: Crowell, 1977.
Friedman, M. [63] Physicalism and the Indeterminacy of Translation. Noas
9 (1975): 353-74.
Grandy, R. [33] Some Remarks about Logical Form. Noas 8 (1974): 157-
64.
- [47] Reference, Meaning and Belief. Journal of Philosophy 70 (1973):
439-52.
Graves, C. et al. [55] Tacit Knowledge. Journal of Philosophy 70 (1973):
318-331.
Grice, H. [36] Utterer's Meaning and Intentions. Philosophical Review
78 (1969): 147-77.
- [37] Utterer's Meaning, Sentence-Meaning, and Word-Meaning. Foun-
dations of Language 4 (I 968): 1-18.
Hacker, P. [18] IInsight and Illusion. Oxford: Oxford University Press, 1972.
Harman, G. [23] Thought. Princeton: Princeton University Press, 1973.
- [24] Meaning and Semantics. In M.K. Munitz and P.K. Unger, Eds.,
Semantics and Philosophy. New York: New York University Press,
1974.
- [39] Review of Schiffer [24] . Journal of Philosophy 71 (1974): 224-9.
Hintikka, J. [13a] Quantifiers in Logic and Quantifiers in Natural Language.
In S. Korner, Ed., Philosophy of Logic. Oxford: Blackwell, 1976.
- [I3b] Quantifiers in Natural Languages: Some Logical Problems II. Lin-
guistics and Philosophy I (1977): 153-72.
- [16] Reply to Peacocke. In E. Saarinen, Ed., Game-Theoretic Semantics.
Dordrecht: Reidel, 1978.
Katz, J., and Fodor, J. [IJ The Structure of Semantic Theory. Language
xxxix (1963): 170.
- and Postal, P. [2J An Integrated Theory of Linguistic Descriptions.
Cambridge, Mass.: M.I.T. Press, 1964.
56 C. Peacocke

Katz, J. [26] Semantic Theory. New York: Harper and Row, 1972.
Kreisel, G. [21] Foundations of Intuitionistic Logic. In E. Nagel et al., Eds.,
Logic, Methodology and Philosophy of Science. Stanford 1962.
Lewis, D. [6] General Semantics. In D. Davidson and G. Harman, Eds.,
Semantics of Natural Language. Dordrecht: Reidel, 1972.
- [40] Convention. Cambridge, Mass.: Harvard University Press, 1969.
Lycan, W. [19] Noninductive Evidence: Recent Work on Wittgenstein's
"Criteria". American Philosophical Quarterly 8 (1971) 109-25.
McDowell, J. [29] Truth Conditions, Bivalence and Verification. In Evans
and McDowell.
McGinn, C. [48] Charity, Interpretation and Belief. Journal of Philosophy
74(1977): 521-35.
Montague, R. [4] Formal Philosophy. Ed. R. Thomason. New Haven: Yale
University Press, 1974.
- [5] English as a Formal Language. In Montague [3].
Peacocke, C. [15] Game-Theoretic Semantics, Quantifiers and Truth. In
E. Saarinen, Ed., Game-Theoretic Semantics. Dordrecht: Reidel, 1978.
- [35] What is a Logical Constant'? Journal of Philosophy 73 (1976): 221-
40.
- [42] Truth Defmitions and Actual Languages. In Evans and McDowell, and
in trans.
Putnam, H. [60] The Refutation of Conventionalism. In M.K. Munitz and
P.K. Unger, Eds., Semantics and Philosophy. New York: New York
University Press, 1974.
Quine, W. [34] Philosophy of Logic. Englewood-Cliffs: Prentice-Hall, 1969.
- [53] The Roots of Reference. La Salle, Ill.: Open Court, 1974.
- [58] On the Reasons for Indeterminacy of Translation. Journal of Phil-
osophy. 67 (1970): 178-83.
- [59] On Empirically Equivalent Systems of the World. Erkenntnis 9,
(1975): 313-28.
- [61] Word and Object. Cambridge, Mass.: M.LT. Press, 1960.
- [62] Olltological Relativity. Journal of Philosophy 65 (1968): 185-212.
Ramsey, F. [54J General Propositions and Causality. In The Foundations
of Mathematics and Other Essays, Ed. R. Braithwaite. London:
Routledge, 1931.
Saarinen, E. [14] Game-Theoretical Semantics. Monist 60 (1977): 406-18.
Schiffer, S. [38] Meaning. Oxford: Oxford University Press, 1972.
Searle J. [45] Speech Acts. Cambridge: Cambridge University Press, 1969.
Strawson, P. [27J Scruton and Wright on Anti-Realism Etc. Proc. Arist. Soc.
51 (1976-7): 15-21.
Stich, S. [57] What Every Speaker Knows. Philosophical Review 80 (1971):
476-96.
Wallace, J. [30J On the Frame of Reference. In D. Davidson and G. Harman,
Eds., Semantics of Natural Language.
Wright, C. [22] Truth-Conditions and Criteria. Proc. Arist. Soc. S.v. 50
(1976): 217-245.
-- [28] Strawson on Anti-Realism. Synthese 40 (1979): 283-299.
Semantics: A revolt against Frege
J AAKKO HINTIKKA
Florida State University

1. FREGE'S HEGEMONY

The most influential semanticist in the recent history of logic and

of philosophy of language did not believe in semantics and con-
sequently left the most important part of his semantical ideas
unexpressed. In the last hundred years, most of philosophical logic
and of philosophical analysis has been deeply influenced, often
virtually dominated, by the ideas of Gottlob Frege (in most cases
as propagated by Russell; cf. e.g. [85]). Yet Frege left many of his
own semantical assumptions undefended, unless the success of his
overall approach is supposed to be its defense.
These references to Frege's semantical ideas require an expla-
nation. When someone speaks in these days of Frege's semantical
views, philosophers and linguists typically think of Frege's theory
of sense and reference [I] . This is only a small and secondary part
of the total story, however. (This misleading emphasis permeates
for instance Dummett's massive book [2]). The theory of sense
and reference is introduced by Frege only to handle what he called
oblique contexts; i.e., what are nowadays often called intensional
contexts. But it is Frege's treatment of nonoblique (extensional)
contexts that differs most sharply from that of his predecessors
and has exercised the strongest influence on subsequent logic and
philosophy. Virtually all later analytical philosophers have accept-
ed Frege's conception of logical system and its ingredients, in-
cluding his views of the form of primitive (atomic) propositions.
This conception of logical system is Frege's most remarkable
achievement, as is rightly emphasized by van Heijenoort [3], [4] .
Even though it is formulated by Frege in syntactical (deductive

Contemporary philosophy. A new survey. Vol. 1, pp. 57-82.

@1981. Martinus NiihoIf Publishers, The Hague / Boston/London.
58 J. Hintikka

and axiomatic) terms, its motivation is clearly semantical. Through

his conception of logical system, Frege's tacit semantical ideas
have dominated much of twentieth-century philosophy of
language. Indeed, most of the ingredients of Frege's semantical
conceptions have been seriously challenged only in the last few
decades, in many cases only in the last few years. In this survey
article, I will consider several of the most interesting recent
developments in logical and philosophical semantics as so many
attempts to broaden the Frege framework and in some cases over-
throw it altogether. Accordingly, this paper is divided into several
different sections according to the specific Fregean assumption
that is being challenged. This results in a survey of ideas rather
than a survey of literature. In the present situation, a survey of the
former kind seems to me even more useful than one of the latter
sort.

2. LANGUAGE AS THE UNIVERSAL MEDIUM VS. LANGUAGE

AS CALCULUS

One reason why Frege's semantical ideas (other than the sense-
reference theory) have largely been overlooked by philosophers is
that Frege did not think seman tical relations could, strictly
speaking, be explicitly formulated and discussed. Hence a system-
atic theoretical study of semantics is impossible in Frege's view.
He is accordingly representative of the type of view labelled by
van Heijenoort [3] 'logic as language,' which perhaps rather
ought to be called in more general terms the view of language as
the universal medium of all discourse. According to this kind of
view we cannot escape the basic semantical relationships that
connect our language with reality. Since they are presupposed in
anything we say, they cannot be meaningfully talked about. On
the view of language as the universal medium we cannot, as we
can on the view of language as calculus, speak of the representa-
tive relationships between language and reality.
In contrast, on the view labelled by van Heijenoort 'logic as
calculus,' we can meaningfully and nontrivially discuss the links
between our own language and reality. We can even think of those
representative relationships as being varied. In other words, we can
 Semantics: A revolt against Frege 59

to some extent at least think of our language and its logic as if it

were a calculus, not in the sense that it is a meaningless formal
game as a calculus is, but in the sense that it can be reinterpreted
like a calculus. In traditional terminology, the contrast between
the two views is sometimes marked by the contrast between
mathesis universalis and lingua characterica.
This contrast is a most useful tool in trying to understand
various philosophical issues connected with recent semantics.
First, the development of all serious truth-conditional semantics
(model theory) obviously presupposes adopting the conception of
language as calculus. Semantics is the study of the representative
relations (meaning relations) between language and reality. Hence
we can practice systematical semantics only if we can meaning-
fully discuss these relationships as we cannot do on the view of
language as the universal medium. The slow historical develop-
ment of logical semantics (logical model theory) has gone hand in
hand with a gradual shift of philosophers' emphasis from mathesis
universalis to lingua characterica.
Secondly, the philosophy of language of several major thinkers
has been deeply influenced by their commitment to the idea of
language as the inescapable medium of all discourse. This is, as van
Heijenoort in effect suggests, the reason why Frege does not spell
out his own semantical vision. Indeed, the completeness of Frege's
own axiomatization of first-order logic and the very concept of
(semantical) completeness were left for much later logicians,
notably to Codel and Tarski, to prove or to formulate, respec-
tively. For logical validity is, when it is explicitly defined, a con-
cept which requires a tacit reinterpretation of the representative
relationships between language and reality: a sentence is valid if
and only if it is true on every possible reinterpretation of its non-
logical (more generally, nondesignated) concepts. Hence a com-
pleteness proof (a proof that each logically valid sentence of a
given language is provable) presupposes the idea of language as
calculus.
However, Frege is not the only nor the last major logician to
believe in the conception of language as the universal medium.
Much of Quine's philosophy of language (see Quine [5], [6], [7],
and [8]; Davidson and Hintikka [9], Orenstein [10]) cannot be
appreciated without realizing that he is another true believer. We
60 J. Hintikka

can perhaps give truth-definitions a la Tarski for restricted frag-

ments of our language, but a large-scale variation of the repre-
sentative relationships is impossible according to Quine. On such
a view semantics is likely to be futile. Small wonder, therefore,
that Quine has never practiced model theory and that the dis-
cussion of some of the main problems he has imported into
recent discussion, notably the problem of radical translation,
has remained singularly inconclusive. These problems are essentially
semantical ones, but most of the discussants have followed Quine's
lead and shunned all systematic semantical theory.
Another case in point is Wittgenstein, whose allegiance to the
idea of language as the universal medium persisted throughout his
career. (See, e.g., Wittgenstein [II], [12], [13]; Kenny [14];
Hallett [15]; Hintikka et at. [16]. With what is said here, cf.
Hintikka [17].) In the Tractatus [18] (cf. Black [19], Stenius
[20 1), the ineffability of semantics gave birth to the famous con-
trast between saying and showing. As far as Wittgenstein's late
work is concerned, his self-same semantical abstinence has prompt-
ed a major misinterpretation. In later Wittgenstein, basic seman-
tical relations (connections between language and reality) are
established and maintained by language-games. Because such
connections are inexpressible, the semantical role of language-
games is not discussed in so many words by Wittgenstein. This has
created the impression that he does not believe in such a repre-
sentative function at all. This is a completely mistaken view,
however. On the contrary, it is just the basic semantical role of
language games that makes them ineffable. Wittgenstein's apparent
emphasis on connections between different moves in language-
games rather than on representative relations from language to
reality is merely a reflection of his belief in language as the
universal medium.
The transition from the conception of language as the universal
medium to that of language as calculus is one of the first main
departures from the Fregean tradition. To some extent, it was
necessitated by the failure of another ingredient of the Fregean
configuration of ideas, viz. the failure of his idea of an axiomat-
ization of logic. Godel's incompleteness proofs showed that for
sufficiently strong languages there cannot be any complete axiom
system of the Fregean kind. In the last couple of decades, philo-
 Semantics: A revolt against Frege 61

sophical analysts of language have gradually come to realize that

explicitly semantical techniques rather than attempted axiomati-
zations of logical truths (without prior independently charac-
terized concepts of truth and validity) are the best tools for the
logical analysis of many important concepts, often concepts
imbedded in natural language.
Nevertheless, a mistake - 'the first dogma of the philosophy of
language' - often is in the offing when an attempt is made to
elicit semantical morals from the rules of logical inference which
naturally are formulated in syntactical (formal) terms and which
establish, when correctly applied, logical truth rather than truth
simpliciter. (The former is logicians' artificial notion, and is much
less basic than the plain man's concept of material truth.) Hence
several fashionable lines of thought are in need of careful scrutiny
before they can be acquitted of suspicions of too close an assimi-
lation of rules of logical inference to semantical ideas. Cases in
point are offered by Hacking [21] who tries to produce semantics
from the rules of inference of various logical systems, by Dummett
[22], [23], and by Paul Lorenzen and his followers (see [24],
[25], and Lorenz [26]). The (only partially published) program
of Dag Prawitz seems to belong to the same group of ideas (see
[27], [28]).
The development of logical semantics (model theory) is not a
recent development, even though philosophers' awareness of the
power of semantical methods largely is. Other aspects of the
revolt against Frege are of a more recent vintage, however.

3. POSSIBLE-WORLDS SEMANTICS

A somewhat more recent departure from the Fregean paradigm

is the so-called possible-worlds semantics, which grew from the
semantics of model logics developed in the late fifties. The latter
is often attributed to Saul Kripke (see Kripke [29], [30], and
[31]), who certainly did discover it independently, albeit con-
siderably later than Stig Kanger and several others. Modal logics
and hence presumably also possible-worlds semantics are dealt
with in a separate survey article. Hence I shall here restrict myself
to a few salient general points only.
62 J. Hintikka

First, it is especially natural to consider possible-worlds seman-

tics as a departure from Frege. This is seen most clearly from the
formulations of possible-worlds semantics given by Richard Mon-
tague. (See Thomason [32], Partee [33], and cf. Hintikka [34]
and Cocchiarella [35].) According to Frege, the sense of an
expression comprises more than its reference; it comprises also the
way in which the reference is given to us. If one interprets 'the
way of being given' here functionally, and if one generously
allows the arguments of the function in question to depend on the
whole world in which the prospective reference (the value of the
function) is to be located, one arrives precisely at the main idea of
possible-worlds semantics: meanings (Frege's 'senses') are
functions from possible worlds to references (extensions). The
main departures from Frege are twofold: (i) Frege firmly reifies
his senses into denizens of this world of ours instead of interworld
functions which onto logically speaking are literally neither here
nor there; and (ii) Frege's 'senses' are much more like recipes or
algorithms for calculating Montague-type meaning functions than
like these functions themselves as abstract mathematical and
logical entities.
The closeness of this fundamental feature of possible-worlds
semantics to Frege is illustrated by the fact that the right con-
dition of the substitutivity of two expressions for each other in an
intensional context, which for Frege was the identity of their
senses (the meaning functions associated with them), is now seen
to be the identity of these functions as restricted to those worlds
which we are considering in the context in question.
Secondly, there apparently is a deep difference between
different variants of possible-worlds semantics. The basic idea is a
modification of the Leibnizian idea that necessary truth in a given
world W0 means truth in every possible world. The crucial tech-
nical modification is to replace the set of all possible worlds by the
set of all worlds alternative to W0, i.e., of all worlds bearing a
certain "alternativeness relation" or "accessibility relation" to
Wo. But how rich must such a set of alternatives be, if what we are
dealing with are logical modalities'? Kripke's answer is essentially
that it can be any old set of possible worlds. This is quite
unnatural, however, for logical necessity certainly seems to require
truth in every logically possible world (with given individuals).
 Semantics: A revolt against Frege 63

This latter requirement was in effect assumed initially by Montague

(see Thomason [32]), the most forceful possible-worlds seman-
ticist, and apparently also by Kanger, the first real semanticist of
modal logic. (See Kanger [36], [37].) It is explicitly discussed by
some of Montague's former students (see Cocchiarella [38 J and
[39)). This is an important difference which, as Hintikka points
out in [40], is analogous to the difference between nonstandard
and standard models in Henkin's sense ([41 J, cf. Andrews (42))
in higher-order logics. This important distinction needs much more
attention than it has been given by philosophers. It is connected
with a number of differences between Montague and Kripke. (For
Kripke's philosophical views, see his (43), [44), and [45).)
An interesting further development is motivated by the obser-
vation that in ordinary modal logic we always move further and
further from the world W0 in which we are evaluating a sentence
from W0 to its alternatives, perhaps to their alternatives, etc.
However, in traditional modal logic we have no way of going back
to possible worlds considered earlier, even though several
phenomena in natural languages presuppose such return trips. A
few logicians, for instance Hans Kamp [46) and David Kaplan
[ 47), have proposed to deal with these phenomena by making
evaluation relative to two worlds, the initial world we start from
being preserved as a parameter all through the evaluation process.
As Esa Saarinen argues [48 J, [49), it seems better to build ways
of returning to earlier worlds into the object-language itself. It
seems that his 'backwards-looking operators' function differently
from the ways we refer anaphorically back to worlds considered
earlier in natural languages, however.
As far as the philosophical problems of possible-worlds seman-
tics are concerned, many philosophers seem to have been misled
by the metaphysically charged tenn 'possible world.' They
expect these worlds to be something quite comprehensive, not
unlike the grand Leibnizian possible worlds between which God
makes his choice. Yet logically and ontologically this compre-
hensiveness is almost a side issue. The crucial attribute of possible
worlds is not that each of them is large, but their multiplicity, that
is to say, the fact that there are several of them to be considered in
their relation to each other, as it were at the same time. 'Possible
worlds' so conceived of can be quite short histories of quite a
64 J. Hintikka

small part of the universe, for instance, different courses that an

experiment may take. It follows that the ontology of such possible
worlds is not any more outlandish than what is needed for the
applicability of probability calculus, with possible worlds playing
the role of the sample-space points of a probabilist ala Kolmogorov
[50]. Philosophers ought not overlook the fact that for statis-
ticians, too, many worlds are, to use L.J. Savage's [51] termi-
nology, 'small worlds.' As far as its metaphysical status is
concerned, possible-worlds semantics is no more suspect than
probability calculus.
By the same token, the worlds of a possible-worlds semanticist
are not any more unclear than the 'events' and 'sample-space
points' that are in the last analysis involved in any appli-
cation of probability concepts. Indeed, it seems to me that most
of the purely philosophical objections to the reality of possible
worlds are either jejune or else betray (for instance in Quine's
case) very restrictive presuppositions as to what is empirically
or ontologically acceptable.
The very real problems there are in possible-worlds semantics
concern much more specific issues. The first and foremost of
these is, as Quine has correctly pointed out, the problem of cross-
identification. It may be true that in most applications of the
possible worlds framework, cross-identification (answering ques-
tions of sameness concerning individuals in different worlds) is
unproblematic. This, however, does not mean that the question
as to how cross-identification takes place is not philosophically
interesting even when the outcome is not in doubt. There is no
consensus even as to what principles we use in our actual con-
c~ptual system to cross-identify. David Lewis [52] believes in
the similarity between individuals as the crucial criterion. It is
doubtful, however, whether such a theory is even a starter unless
structural similarities of those entire worlds are taken into account
in which the two individuals occur. Hintikka [53] has suggested
that the continuity of individuals in space and time goes a long
way towards facilitating cross-identification by helping us to trace
them to the area which is shared by the different possible worlds
we are considering and in which cross-identification is therefore
unproblematic. However, it may be that this idea should be given
the same twist as David Lewis' counterpart theory was subjected
 Semantics: A revolt against Frege 65

above: continuity considerations should be applied to the space

of different worlds rather than to particular possible worlds
considered one by one. In other words, what is unproblematic
are merely the identities of individuals under infinitesimally small
changes in the world at large. The right logical type of the task of
cross-identification is thus perhaps that of solving a set of partial
differential equations. Maybe this idea helps us to understand both
the ease with which we cross-identify in typical smooth-running
everyday situations and the breakdown propensity of our cross-
identification methods when something unexpected happens.
Perhaps what we need is a catastrophe theory of cross-identi-
fication.
Be this as it may, we seem to have fairly good qualitative grasp
of the differences between different cross-identification methods.
For instance, Hintikka (54], (55] has distinguished between
'descriptive' cross-identification relying on an ordinary public
framework and 'demonstrative' cross-identification (cross-identi-
fication 'by acquaintance') which relies on the framework of
one's first-hand epistemic relations (perception, first-hand
memory, acquaintance, etc.). He has also shown how this contrast
explains the semantic difference between the wh-constructions
(interrogative constructions) and direct-object constructions with
certain intensional verbs (perceives, sees, remembers, knows, etc.).
This is one of the most remarkable proofs of the applicability of
possible-worlds semantics to the semantics of natural languages.
It also shows that even if much of the details of cross-identi-
fication methods are unclear, a mere overall qualitative under-
standing of these methods is highly valuable in natural-language
semantics. In other words, Miller and Johnson-Laird ((56],
p. 589) are radically mistaken in thinking that the explicit logical
study of the contrast between different cross-identification
methods somehow presupposes mathematical exactness of these
cross-identifications.
Another interesting insight (or claim, if you are a skeptic) is
that, for identification by acquaintance, world lines are drawn
in the special case of lines connecting the actual world with its
alternatives by means of causal chains. One reason why this
observation is interesting is that it shows (if it is correct) at one
stroke how seriously incomplete Kripke's causal theory of naming
66 J. Hintikka

is (ef. Kripke [43], [44], and Schwartz [57]). The occasions of

"dubbing" he envisages do indeed establish a semantical link
between acquaintance-individuals and denizens of the actual
world. This does not enable us to identify individuals between
different alternative words nor does drawing descriptive world
lines as distinguished from the demonstrative ones.
Another genuine problem about possible-worlds semantics is
that it seems to commit us to logical omniscience, that is to say,
to the assumption that everyone always knows all the logical
consequences of what he or she knows, believes, all the logical
consequences of what he or she believes, etc. This problem can
be solved, however. Not only is there a natural way of distin-
guishing in syntactical terms between inferences which do and
which do not give rise to breakdowns of logical omniscience.
(See Hintikka [58] and [59], also for connections with impor-
tant earlier doctrines in the philosophy of logic and of mathe-
matics.) Recently, this syntactical solution has been given a
semantical basis by Rantala and Hintikka, as explained below
in section 5. (See their respective papers [601 and [61]).

4. ATOMIC SENTENCES

One of the crucial ingredients of the Fregean conception of a

logical system is the assumption that all atomic sentences are
formed from singular terms by means of certain given predicates
and functions. It is not always realized how restrictive this
assumption really is. The store of nonlogical primitives Frege
envisaged may be sufficient for his special purpose of analyzing
mathematical concepts and mathematical inferences. There is no
a priori reason to expect that it will suffice for other purposes,
including that of serving as a semantical framework for natural
languages. Ventures beyond the Fregean scheme have usually
taken the form of introducing special interpreted (designated)
predicates which are still of the Fregean type. These include, over
and above the membership relation studied in set theory, the part-
whole relations studies in the 'calculus of individuals' developed
by Nelson Goodman and others. (See Goodman [62]). However,
it is likely that more radical departures are needed in this direction.
 Semantics: A revolt against Frege 67

I do not see any hope of analyzing the semantics of such concepts

as mass terms (see Pelletier [63] with an extensive bibliography)
or comparatives without allowing primitive propositions of an
altogether different logical type from the ones Fregeans coun-
tenance. Since this way out has rarely been taken by logical and
linguistic semanticists, it is no wonder that both mass terms and
comparatives have prompted extensive but inconclusive dis-
cussions in the recent literature. For instance, there is little hope
of discussing the semantics of comparatives without allowing
primitive nontruth-functional propositional compounds which in
effect compare the degrees to which one and the same predicate
applies to different individuals.
An even more radical departure from the Fregean paradigm is
possible and indeed desirable here. So far we have examined
broadening Frege's conception with respect to the predicates
(Frege's 'functions') which may enter into primitive propositions.
However, doubts can likewise be entertained concerning the
adequacy of the other main ingredient of Fregean primitive propo-
sitions, the individuals (Frege's 'objects'). In possible-worlds
semantics, an individual must be recognizable in several different
possible worlds, and hence becomes very much like a function (of
a kind) special from possible worlds to their inhabitants. If our
basic individuals hence are like functions, why cannot we treat
them as such? Why do we need 'urelements' at all? Perhaps we do
not. An example of what the resulting semantics might look like is
offered by Dana Scott's interesting semantics for the X-calculus. In
it, everything is a function; there are no 'urindividuals' at all. For
a systematic treatment of Scott's ideas, with a good bibliography,
see Stoy [64). Scott's ideas are inspired by and developed so as to
apply to the semantics of computer languages. Scott and others
have also connected them in an interesting way to nonclassical
logics (see Scott [65], with further references to the literature).

5. THE DYNAMICS OF SEMANTICS

A consequence of Frege's belief in logic as language was that he

did not in any way discuss the basic semantical relations which
connect language with the part or the aspect of reality it can help
68 1. Hintikka

us to deal with. Even when those relations were later theorized

about by Tarski, Carnap, and others, they were left completely
unanalyzed. Wittgenstein's language-games can be thought of as
such links between language and the world. However, in the Witt-
gensteinian tradition, they have been given little systematic
theoretical attention.
It turns out, however, that some of the Wittgensteinian
'language-games' are also games in the precise sense of the mathe-
matical theory of games, and can hence be studied by means of
the conceptual framework provided by this theory. This obser-
vation has given rise to the game-theoretical semantics of Hintikka
and his associates. (See Saarinen [66] and the literature referred
to there.)
These semantical games are most easily understood when
thought of as attempted processes of verification of a given sen-
tence S resisted by an actively malicious opponent which can be
identified with nature. Semantical games take the players to sim-
pler and simpler sentences and eventually to atomic ones. Their
truth or falsity is determined by the mere interpretation of the
nonlogical constants of S. If the outcome is a true atomic sen-
tence, the verifier ('myself') has won; if false, the falsifier
('Nature') has won.
The truth of S can now be defined as the existence of a winning
strategy for myself in the correlated game G(S). Thus the game
rules serve the same purpose in game-theoretical semantics as the
different recursive clauses in a Tarski-type truth-definition; they
extend the concepts of truth and falsity from atomic sentences to
others. There are important differences, however. Game rules
operate from the outside in, whereas truth-definitions operate in
the opposite direction. Furthermore, game-theoretical truth-defi-
nitions are usually equivalent with the conventional ones only if
the strategy sets of the players are not appreciably restricted.
If these sets are restricted, e.g., to strategies expressible by com-
putable functions, we obtain an interpretation (both a semantical
and a pragmatic one) of certain nonclassical logics.
By expressing the existence of a winning strategy by an explicit
higher-order sentence, one can extract from game-theoretical
semantics a translation of first-order logic into higher-order logic.
The restrictions on strategy sets will then correspond to various
 Semantics: A revolt against Frege 69

nonstandard interpretations of higher-order logic in Henkin's sense.

(See Henkin [41] and cf. Andrews [42]).
Game-theoretical semantics can be viewed as codifying certain
eminently natural ways of operating with and thinking about such
logical notions as quantifiers and propositional connectives. These
ways are so natural that they are firmly entrenched in logicians'
and mathematicians' folklore, from which they sometimes surface
into explicit discussion, for instance, in connection with such con-
ceptualizations as Skolem functions and branching quantifiers
(Henkin quantifiers). (On the latter see Henkin [67], Walkoe
(68]' Enderton [69], Hintikka [70], and Barwise (71)). An
especially dramatic connection (first pointed out by Dana Scott
[72]) obtains between game-theoretical ideas and G6del's at first
sight obscurely motivated 'extension of the finitistic point of
view' [73] (cf. here also Girard [74)).
This connection with so-called functional interpretations illus-
trates also the fact that the power of game-theoretical semantics
goes way beyond quantifiers and truth-functional connectives. The
crucial idea underlying those uses of game-theoretical semantics
that are connected with functional interpretation is that of a
sUb-game. (See here Carlson and Hintikka [75]). Sub-games of
semantical games are independent parts of these games which are
played 'to the bitter end,' which is interpreted as meaning that
at least one of the players divulges his or her strategy. In the rest
of the game, players' moves depend on this strategy, wherefore
the players' overall strategies are codified by functionals (higher-
order functions) rather than by functions of individuals.
Game-theoretical semantics appears to be especially well suited
for the treatment of various natural-language phenomena,
including differences between different quantifier words, quanti-
fier ambiguities, quantifier scopes, anaphoric expressions with
quantifier phrases as their grammatical antecedents, negation,
multiple questions, and conditionals.
The crucial feature of game-theoretical semantics is its dynamic
character. This allows the use of game-theoretical ideas for pur-
poses which as such are not very closely connected with the game
concept. One such use is Rantala's notion of urn model [60].
Moves in a semantical game connected with quantifiers can be
thought of as draws from a certain "urn" called the domain of
70 J. Hintikka

individuals. By borrowing from probability theory the simple idea

of letting the population of individuals in the urn change between
successive draws, Rantala obtains a new type of nonstandard
model for first-order languages. They have already been used by
Rantala in his highly successful theory of definability [75]. Hin-
tikka [61] has likewise applied them to provide a semantical basis
for his syntactical (deductive) solution to the problem of 'logical
omniscience' mentioned earlier. It turns out that the semantical
solution to which one is led via the concept of an urn model is
the same as Hintikka had argued for earlier on intuitive grounds
and in syntactical (deductive) terms.
This solution of the problem of logical omniscience is closely
related to the philosophical uses Hintikka has made of the basic
ideas of game-theoretical semantics, partly before this theory was
developed into a full-fledged semantics. (See his [58], [77].)
Among other things, there is a close connection between the
breakdown of logical omniscience on Hintikka's construal and
Kant's views of mathematical knowledge and of the analytic-
synthetic distinction. These applications do not belong to seman-
tics any longer, however.
The theory of urn models incorporates also the theory of what
Hintikka at one time called surface models [78] . They are not real
models at all, changing or invariant, but rather summaries of what
one can find about an urn model without considering series of
draws longer than some fixed finite number n.
Game-theoretical semantics has not been presented so far in a
fully explicit formulation. (Saarinen's attempt in [48] turns game-
theoretical semantics into a variant of Tarski-type semantics.) This
makes comparisons with other approaches sometimes difficult.
However, enough work has been done on game-theoretical seman-
tics in order for us to see what some of the main differences
between it and some of the main competing approaches are.
Game-theoretical semantics and the variation in the relations
between language and the world it allows is of course not the
only way of changing the basic semantical relations which give a
language its meaning. Consider, for instance, a sentence of some
first-order language. Usually, it is said to be satisfied by (true in)
all the models (worlds) which are connected with it in the right
way by a set of recursive truth-clauses. It is quite possible,
 Semantics: A revolt against Frege 71

however, to impose further conditions on these models, inde-

pendent of the step-by-step clauses. For instance, we can require
that these models be 'special models' in logicians' sense. They
could be required to be, for instance, prime models or to be
homogeneous saturated models. This makes quite a difference to
~ur semantics. For instance, many more nonlogical axiom systems
become categorical in a power, i.e., will determine their models of
some given cardinality up to isomorphism. Nor are requirements
of this general kind unheard-of in actual axiomatic practice. The
last axiom of the best known example of an actual axiom system,
that is, of Hilbert's [79], is of this kind, imposing a special con-
dition on the models of the other axioms rather than a condition
on the elements of such models. (For this axiom, see Bemays
[80]).

6. THE FAILURE OF THE FREGE PRINCIPLE

An important role has been played in much recent semantics by

the so-called Frege Principle, a.k.a. compositionality. This prin-
ciple says that the semantical entity associated with a complex
linguistic expression is a function of the semantical entities
associated with its component expressions. In Frege, the seman-
tical entities in question can be either references or senses. Their
failure to be parallel in all contexts according to Frege is a self-
inflicted problem which need not detain us here.
Compositionality is closely related to the possibility of giving
Tarski-type recursive truth-definitions for a language. Hence
theories of semantics which aim (among other things) at such
truth-definitions are committed to compositionality. Cases in
point are the theories outlined by Donald Davidson ([ 81], [82],
cf. [83]), and Richard Montague (see Thomason [32]).
As pointed out in Hintikka [84] , compositionality presupposes
that the right direction of a semantical analysis is from inside out.
There is no general reason a priori why this should be the correct
direction of semantical processing. It fails in fact in several
different languages, both natural and formal. Some such counter-
examples to compositionality are implicit in what has been said
above. Branching quantifiers, backwards-looking operators, and
72 J. Hintikka

semantical games that can go on to infinity all provide examples

that show that compositionality in fact fails in both kinds of
languages. Its failure for certain perfectly explicit formal languages
is especially interesting in that it shows that whatever difficulties
there are in trying to uphold the Frege Principle are not due to
irregularities or other imperfections of natural languages. The
failure of compositionality is interesting in a wider perspective as
it shows that semantical theories relying on Tarski-type truth-
characterizations cannot be fully adequate. This objection hits
both Montague's and Davidson's respective approaches to seman-
tics. In contrast, in game-theoretical semantics the analysis of sen-
tences proceeds from outside in, and hence is automatically
compatible with the failure of the Frege Principle.

7. THE ALLEGED AMBIGUITY OF "IS"

The sharpest specific difference between Frege's logical notation

and the ideas of his predecessors lies in his treatment of verbs for
being. Such verbs are, according to Frege and his followers, am-
biguous in that they have to be translated into the logical notation
in at least four different ways: (i) by the identity sign '=' (the
'is' of identity); (ii) by the existential quantifier (the 'is' of exis-
tence); (iii) by predicative juxtaposition (the 'is' of predication or
the copula), and (tv) by a general implication (the 'is' of class-
inclusion). In 1914 Bertrand Russell [85] called this four-fold
distinction 'the first serious advance in real logic since the time of
the Greeks', and it has been incorporated in all the usual systems
of first-order logic (lower predicate calculus, quantification
theory). Hence everybody who has been using the notation of
first-order logic for the purposes of semantical representation is
committed to the ambiguity of 'is'. This applies to linguists and
philosophers otherwise as unlike ~ach other as George Lakoff,
Noam Chomsky, W.V. Quine, Donald Davidson, and Ludwig
Wittgenstein. Frege's distinction has thus become one of the most
widely accepted inventions of his and an integral part of most
current treatments of semantics.
Parts of the distinction have been challenged by Montague
(following ideas foreshadowed by Russell [86]), and the whole
distinction by Lesniewski (see Luschei [87]). The latter neverthe-
 Semantics: A revolt against Frege 73

less failed to build a viable semantics for his formalism.

It turns out that no Fregean distinction between different
meanings of 'is' can obtain in game-theoretical semantics, as
pointed out in Hintikka [88]. Since game-theoretical semantics is
a viable semantical theory for certain fragments of natural lan-
guages, we have a situation which is extremely interesting
theoretically. It may be called seman tical relativity. What it
amounts to is that the crucial semantical predicates of linguistic
expressions, such as ambiguity, univocity, number of readings,
etc., may depend on one's underlying framework of semantical
representation. This suggests several interesting methodological
morals.
(a) People's so-called intuitions about the basic semantical
properties and relations - ambiguity, synonymy, number of
readings, etc. - will be relative to a framework of semantical
representation which they are tacitly assuming, and can be
changed by inducing them to adopt a new framework. (Cf. Hin-
tikka [88] and [89]). Hence the task of linguistic semantics
cannot be identified with the regimentation (and/or explanation
through systematization) of our semantical intuitions as is
assumed, e.g., by Katz [90] .
(b) Even if one framework of semantical representation turns
out to be preferable to another for some given language, the
reasons need not be purely semantical. They can, e.g., turn on the
superiority of the total theory of language the former allows us to
build as compared with the latter. This total theory can for
instance include syntax (possibly also pragmatics and even phon-
etics) as well as semantics. Hence there is likely to be more inter-
dependence between syntax and semantics than there is usually
thought to be.
(c) In several approaches to language, each structurally un-
ambiguous sentence S is thought to have a unique syntactical
generation (derivation) from a unique deep structure. This is true
of Chomsky's Aspects model [91], and it is also true of the so-
called generative semantics (see George Lakoff [92] ). If the seman-
tical relativity outlined above obtains, then the deep structure of a
given sentence S cannot be identified with the semantical repre-
sentation of S. For the thesis of semantical relativity amounts to a
denial of such a unique representation.
74 1. Hintikka

Hence semantical relativity is incompatible with both the

approaches just mentioned. Even though they have been highly
influential, these approaches have in recent years been subjected
to severe criticism of several other kinds. Independently of the
developments just mentioned, the idea that semantic interpretation
operates on the deep structure has been given up among Chomsky's
closest followers. First, the rules of semantical interpretation were
allowed to take into account also the surface structure over and
above the deep structure. Much of the work of this kind was
reported by lackendoff [93J.
Recently, the pendulum has swung to the other extreme.
Chomsky has suggested (see [94 J , p. 167) that the rules of seman-
tical interpretation which determine the 'logical form' of a sen-
tence perhaps can operate exclusively on the surface form. This
presupposes that constituents of the deep structure which are lost
or moved elsewhere through the transformations (which turn the
deep structure into the surface structure) leave a 'trace' which can
be recovered from the surface form. This is in fact claimed by
Chomsky's 'trace theory' expounded in his [95] and [96].
Since the logical form postulated in Chomsky's theory is for-
mulated in terms of conventional first-order logic, it is subject to
the same objections as other uses of this logic as the only frame-
work of semantical representation.
(d) Indeed, semantical relativity is incompatible with any use of
first-order logic as one's sole canonical notation. Another variant
of such a use of first-order logic is the translational strategy of
Quine's ([6], [7], [8]) and Davidson's ([81], [82], [83]; cf.
Merrill B. Hintikka and Vermazen [97]) where the semantical
properties of natural languages are elucidated by translating or
paraphrasing their sentences into first-order notation. (Davidson
nevertheless makes it clear that he favors this strategy only
because of what he takes to be its success; see his [98].) The
problem with this strategy is that the principles of translation are
often left completely mysterious, even when we know fully well
that the translation of some given particular sentences is obvious.
Indeed, it is just the step from surface forms to the logical form
(or, rather, the reverse step) which is the focus of the efforts of
generative semanticists (see George Lakoff [92], [99]).
(e) Even more generally, semantical relativity casts a shadow
 Semantics: A revolt against Frege 75

over the concept of logical form. For one of the most important
ideas underlying this notion is that the logical form of an unam-
biguous sentence is always unique. It was introduced largely to
capture the unique true logical or semantical form of our sentence
which this varying syntactical clothing tends to hide.
(0 Among the approaches to semantics which postulate a
unique mode of semantical representation there are those postu-
lating a 'language of thought' which is thought to represent the
way language is actually processed semantically by the human
mind. A representative example of approaches of this sort is Fodor
[ 100] . It is seen that current views of this type can be considered
as latter-day versions of earlier, mostly philosophical views of
thinking as being analogous to speaking - perhaps comparable
with mind's discourse with itself without spoken words. Views of
this kind were put forward among others by Plato and Peirce, and
in the contemporary discussion by Peter Geach [10 I ]. In spite of
such illustrious ancestry, 'language of thought' theories are subject
to serious doubts. Applied to ordinary natural languages, they are
essentially translational theories, and hence open to the same
objections as these. Moreover, the proponents of the 'internal
code' idea have failed to spell out the structure of this code in
several important respects. For instance, the problem of the
ambiguity or nonambiguity of 'is' ought to be easy to solve by
reference to the internal code. Yet no such solutions are forth-
coming from the 'language of thought' theorists.

8. TEXT SEMANTICS

Frege was primarily interested in mathematical languages. Since

mathematical propositions are independent of the context in
which they are propounded, Frege was not interested in depen-
dencies between sentences which are due to the context in which
they are put forward. In current linguistic terminology, he had
little use of text semantics (which is part of text grammar) as dis-
tinguished from sentence semantics.
If Frege's shortcoming was excusable, those of his successors are
not. Even though there has been a great deal of work on text
grammar by linguists (for a survey, see Dressler [102)), the level
76 1. Hintikka

of theorizing has been much lower than in sentence syntax and

sentence semantics. The main concepts employed in text grammar,
such as 'focus', 'topic', 'theme', 'comment', etc. have not allowed
the formulation of sharp generalizations. To adapt a memorable
phrase of Patrick Suppes' for my purposes, few theorems have
been proved in text semantics.
However, the reasons for this failure may lie partly in the court
of sentence grammarians. Several important semantical phenomena
are found in essentially the same form both in sentence semantics
and in text semantics. Cases in point include anaphora, quantifier
scopes, and the phenomenon that the sub-game idea is calcu-
lated to capture. Yet many approaches to semantics do not
naturally allow for an extension of the relevant concepts from
sentence semantics to text semantics. This shortcoming is for
instance characteristic of these conceptions of semantics which tie
semantical evaluation closely to the generation process of
sentences. The theories of Katz and others in the sixties which
tried to tie semantical interpretation to a syntactical deep struc-
ture (see for instance Katz and Postal [103], Chomsky [91],
[ 104] , Katz and Fodor [105], and cf. Partee [106]) offer exam-
ples of such theories, even if these particular theories have by this
time been pretty thoroughly discredited on other grounds.
The need of specifically text semantical conceptualizations is
even greater in connection with those linguistic phenomena which
turn on differences in the epistemic background of the different
sentences uttered. The most obvious examples are offered by
questions and answers. (If a questioner and the addressee of the
question knew the same things, the question would be pointless.)
Several treatments of the semantics of questions have been
presented recently (see e.g., Aqvist [107], [108]; Harrah [109],
[110]; Katz [Ill], [90]; Belnap and Steel [112], Karttunen
[113 n. Most of them are geared to the sentential paradigm, however,
and do not yield any real analysis of the crucial question-answer
relationship. The main exception is Hintikka's theory ([ 114] and
(89] ), which is closely related to possible-worlds semantics and to
game-theoretical semantics. It can apparently serve as a basis of a
theory of question-answer dialogues; see Hintikka [115].
Conceptualizations which - like any satisfactory analysis of the
question-answer relationship -- apply to sentences uttered against
 Semantics: A revolt against Frege 77

different epistemic backgrounds offer other kinds of promises as

well. Paul Grice (see [116] , [117]) has put forward highly sugges-
tive remarks concerning the influence of conversational expec-
tations and other contextual factors on the interpretation of
people's utterances. Such influence is almost by definition beyond
the purview of sentence semantics, and has therefore often been
thought to belong to pragmatics rather than to semantics in any
reasonable sense of the latter word. There is no good reason, how-
ever, why text semantics could not tackle some of the phenomena
Grice has been interested in.

BIBLIOGRAPHY

Andrews, Peter [42] General Models and Extensionality. Journal of Symbolic

Logic 37 (1972): 395-397.
Aqvist, Lennart [107] A New Approach to the Logical Theory of Inte"og-
atives. Uppsala: Filosofiska F6reningen, 1965.
- [108] Revised Foundations for Imperative, Epistemic, and Interrogative
Logic. Theoria 37 (1971).
Barwise, Jon [71] On Branching Quantifiers in English. Journal of Philo-
sophical Logic 8 (1979): 47-80.
Belnap, Nuel B., and Steel, Thomas B. [12] The Logic of Questions and
Answers. New Haven: Yale University Press, 1976.
Bernays, Paul [80] Betrachtungen tiber das Vollstandigkeitsaxiom und
verwandte Axiomen. Mathematische Zeitschrift 63 (1955): 219-292.
Black, Max [19] Companion to Wittgenstein's Tractatus. Ithaca, N.Y.:
Cornell University Press, 1964.
Carlson, Lauri and Hintikka, laakko [75] Conditionals, Generic Quantifiers,
and Other Applications of Subgames. In F. Guenther and S.J. Schmidt,
Eds., Formal Semantics and Pragmatics for Natural Languages, pp.
1-36. Dordrecht: D. Reidel, 1979. Reprinted in Saarinen, Ed. [66],
pp. 179-214.
Chomsky, Noam [91] Aspects of the Theory of Syntax. Cambridge, Mass.:
MIT Press, 1965.
-- [104] Topics in the Theory of Generative Grammar. The Hague: Mouton,
1966.
[94] Conditions on Rules of Grammar. Linguistic Analysis 2 (1976).
Reprinted in Chomsky [95], pp. 163-210.
[95] Essays on Form and interpretation. Amsterdam: North·Holland,
1977.
- [96] On Wh-Movement. In Peter W. Culicover, Thomas Wasow and Adrian
Akmajian, Eds., Formal Syntax, pp. 70-132. Academic Press, 1977.
78 J. Htntikka

Cocchia.rella, Nino [39] Logical Atomism and Modal Logic. Philosophia

4 (I 974}: 41-66.
- [38] On the Primary and Secondary Semantics of Logical Necessity.
Journal of Ph ilosoph ical Logic 4 (1975): 13-27.
- [35] Richard Montague and the Logical Analysis of Language. In the
present volume.
Davidson, Donald [81] Theories of Meaning and Learnable Languages.
In Yehoshua Bar-Hillel, Ed., Logic, Methodology, and Philosophy of
Science: Proceedings of the i964 international Congress, pp. 383-
394. Amsterdam: North-Holland, 1965.
- [82] Truth and Meaning. Synthese 17 (1967): 304-323.
- [83] True to the Facts. Journal of Philosophy 66 (I 969): 748-764.
- [98] Action and Reaction. inquiry 13 (1970): 140-148.
- [9] and Hintikka, Jaakko, Eds., Words and Objections: Essays un the
Work of W. V. Quine. Dordrecht: D. Reidel, 1969.
Dressler, Wolfgang U. [102] Ed., Trends in Text Linguistics. Berlin: Walter
U.: Gruyter, 1978.
Dummett, Michael [2] Frege: Philosophy of Language. London: Duck-
worth,1973.
- [22] What is a Theory of Meaning? In Samuel Guttenplan, Ed., Mind
and Language, pp. 97-138. Oxford: Clarendon Press, 1975.
- [23] What is a Theory of Meaning? (II). In Gareth Evans and John
McDowell, Eds., Truth and Meaning. Oxford: Clarendon Press, 1976.
Enderton, H.B. [69] Finite Partially Ordered Quantifiers. Zeitschrift fur
mathematische Logik und Crnndlagen der Mathematik 16 (I 970):
393-397.
Fodor, Jerry A. [l00] The Language of Thought. New York: Thomas Y.
Crowell,1975.
Frege, Gottlob [1] Uber Sinn und Bedeutung. In I. Angelelli, Ed., Kleine
Schriftell. Darmstadt: Wissenschaftliche Buchgesellschaft, 1967.
Geach, Peter [101] Mental Acts. London: Routledge and Kegan Paul, 1957.
Girard, Jean Yves [74] Functional Interpretation and Kripke Models. In
Robert Butts and Jaakko Hintikka, Eds., Logic, Foundations afMathe-
matics, and Computability Theory, pp. 33-57. Dordrecht: D. Reidel,
1977.
Godel, Kurt [73] Uber eine bisher noch nicht beniitzte Erweiterung des
finiten Standpunktes. In Logica: Studia Paul Bernays dedicata, pp.
76- 83. Neuchatel: Editions Griffon, 1959.
Goodman, Nelson [62] The Structure of Appearance, 2nd ed. Indianapolis:
Bobbs-Merrill, 1966.
Grice, Paul [117] William James Lectures. Harvard University, 1967 (un-
published).
- [116] Logic and Conversation. In Donald Davidson and Gilbert Harman,
Eds., The Logic of Grammar, pp. 64-65. California: Dickenson,
Encino and Belmont, 1975.
 Semantics: A revolt against Frege 79

Hacking, Ian [21] Do-it-yourself Semantics for Classical Sequent Calculi. In

Robert Butts and laakko Hintikka, Eds., Logic, Foundations of Mathe-
matics, and Computability Theory, pp. 371-390. Dordrecht: D. Reidel,
1977.
Hallett, Garth [15] A Companion to Wittgenstein's 'Philosophical Investi-
gations.' Ithaca, N.Y.: Cornell University Press, 1977.
Harrah, David [109] Communication: A Logical Model. Cambridge, Mass.:
MIT Press, 1963.
- [110] A System for Erotetic Sentences. In Alan Ross Anderson, Ed., The
Logical Enterprise, pp. 235-245. New Haven: Yale University Press,
1975.
van Heijenoort, lean [3] Logic as Language and Logic as Calculus. Synthese
17 (1967): 324-330.
- [4] Ed., From .";fege to Godel: A Source Book in Mathematical Logic.
Cambridge, Mass.: Harvard University Press, 1967.
Henkin, Leon [41] Completeness in the Theory of Types. Journal of Sym-
bolicLogic 15 (1950): 81-91.
- [67] Some Remarks on Infmitely Long Formulas. In Infinitistic Methods.
London: Pergamon Press, 1961.
Hilbert, David [79], Grundlagen der Geometrie, 8th edition, with revisions
and supplements by P. Bemays. Stuttgart 1956.
Hintikka, laakko [77] Language-Games for Quantifiers. American Philo-
sophical Quarterly Monograph Series 2 (1968): 46-72. Reprinted in
Hintikka [58] .
- [54] Objects of Knowledge and Belief. Journal of Philosophy 67 (I970):
869-883. Reprinted in laakko Hintikka, The Intentions of Intention-
ality, pp. 43-58. Dordrecht: D. Reidel, 1975.
- [59] Knowledge, Belief, and Logical Consequence. Ajatus 32 (1970):
32-47. Reprinted in 1.M.E. Moravcsik, Ed., Logic and Philosophy for
Linguists, pp. 165-176. The Hague: Mouton, 1974.
- [55] Knowledge by Acquaintance - Individuation By Acquaintance. In
David Pears, Ed., Bertrand Russell: Modern Studies in Philosophy,
pp. 52-79. Garden City: Doubleday, 1972. Reprinted in laakko Hin-
tikka, Knowledge and the Known, pp. 212-233. Dordrecht: D. Reidel,
1974.
- [58] Logic, Language-Games, and Information. Oxford: Clarendon Press,
1973.
- 178] Surface Models. In Hugues Leblanc, Ed., Truth, Syntax, and Modality,
pp. 128-147. Amsterdam: North-Holland, 1973.
- [34] On the Proper Treatment of Quantifiers in Montague Semantics. In
SOren Stenlund, Ed., Logical Theory and Semantic Analysis, pp.
45-60. Dordrecht: D. Reidel, 1974.
- [70] Quantifiers vs. Quantification Theory_ Linguistic Inquiry 5 (1974):
153-177: Reprinted in Saarinen, Ed. [66], pp. 49-79.
- [114] Questions about Questions. In Milton K. Munitz and Peter K. Unger,
Eds., Semantics and Philosophy, pp. 103-158. New York: New York
University Press, 1974.
80 1. lIintikka

- [53] Quine on Quantifying In. In Jaakko Hintikka, Ed., The Intentions of

Intentionality, pp. 102-136. Dordrecht: D. Reidel, 1975.
- [61] Impossible Possible Worlds Vindicated. Journal of Philosophical Logic
4 (1975): 475-484.
- [89] The Semantics of Questions and the Questions of Semantics. Acta
Philosophica Fennica 28 (4). Amsterdam: North-Holland, 1976.
- [88] 'Is', Semantical Games, and Semantical Relativity. Journa!ofPhilos-
ophica! Logic 8 (I 979): 433-468.
- [115] The Logic of Information-Seeking Dialogues: A Model. In W. Becker
and W.E. EssleT, Eds., Konzepte der Dialektik, pp. 212 -- 231. Frankfurt!
Main: Vittorio Klostermann, 1981.
[84] Theories of Truth and Learnable Languages. In S. Kanger and S.
Ohman, Eds .. Philosophy and Grammar. pp. 37-57. Dordrecht: D.
Reidel, 1980.
-- [17] Language-Games. In Hintikka et al. [16], pp. 105--125. Reprinted
in Saarinen, Ed. [66] ,pp. 1-26.
- [40] Standard VS. Nonstandard Logic: Higher-Order, Modal, and First·
Order Logics. In E. Agazzi, Ed.,Modern Logic: A Survey, pp. 283-296_
Dordrecht: D. Reidel, 1981.
[16] et al., Eds., Essays on Wittgenstein In Honour ofG.H. von Wright.
Acta Philosophica Fennica 28 (1-3). Amsterdam: North-Holland, 1976.
Hintikka, Merrill B., and Vermazen, Bruce [97], Eds., Donald Davidson: A
Profile. Dordrecht: D. Reidel (forthcoming).
Kanger, Stig [37] The Morning Star Paradox. Theoria 23 (1957): 1-11.
- [36] Provability in Logic. Stockholm Studies in Philosophy, I . Stockholm
1957.
Kaplan, David [47] Bob and Carol and Ted and Alice. In Jaakko Hintikka,
Julius M.E. Moravcsik and Patrick Suppes, Eds., Approaches to Natural
Language, pp. 490-518. Dordrecht: D. Reidel, 1973.
KaTttunen, Lauri [113] Syntax and Semantics of Questions. In Henry Hii,
Ed., Questions, pp. 165-210. Dordrecht: D. Reidel, 1978.
Katz, J.J., and Postal, Paul M. 1I03] An Integrated Theory of Linguistic
Descriptions. Cambridge, Mass.: MIT Press, 1964.
- [105] and Fodor, Jerry A., Eds., The Structure of Language. Englewood
Cliffs, N.J.: Prentice-Hall, 1965.
- [Ill] The Logic of Questions. In B. van Rootselaar and J.F. Staal, Eds.,
Logic, Methodology, and Philosophy of Science III, Proceedings of the
Third Internatiollal Congress, pp. 463-493. North-Holland, Amster-
dam: 1968.
- [90J Semantic Theory. New York: Harper & Row, 1972.
Kenny, Anthony [14] Wittgenstein. Middlesex: Penguin, 1973.
Kolmogorov, A. [50] Grundhegriffe der Wahrscheinlichkeitsrechnung.
Berlin: Springer, 1933.
Kripke, Saul [29] Semantical Considerations on Modal Logic. Acta Phi/os-
ophica Fennica 16 (1963): 83-94.
 Semantics: A revolt against Frege 81

- [30] Semantical Analysis of Modal Logic I. Zeitschrift fur mathematische

Logik und Grund/agen der Mathematik 9 (1963): 67-96.
[31] Semantical Analysis of Modal Logic II. In J.W. Addison, L. Henkin
and Tarski, Eds., The Theory of Models, pp. 206-220. Amsterdam:
North .Holland, 1965.
- [44] Identity and Necessity. In Milton K. Munitz, Ed., Identity and Indi-
viduation, pp. 135-164. New York: New York University Press, 1971.
- [43] Naming and Necessity. In Donald Davidson and Gilbert Harman,
Eds., Semantics of Natural Language, pp. 253-355. Dordrecht: D.
Reidel,1972.
- [45] Speaker's Reference and Semantical Reference. Midwestern Studies in
Philosophy 2 (1977): 255-276.
Lakoff, George [92J On Generative Semantics. In Danny D. Steinberg and
Leon A. Jakobovitz, Eds., Semantics, pp. 232-296. Cambridge
University Press, 1971.
- [99] Linguistics and Natural Logic. In Donald Davidson and Gilbert
Harman, Eds., Semantics of Natural Language, pp. 545-665. Dord-
recht: D. Reidel, 1972.
Lewis, David [52] Counterpart Theory and Quantified Modal Logic. Journal
of Philosophy 65 (1968): 113-126.
Lorenz, Kuno [26] Arithmetik und Logik als Spiele. Dissertation, Kiel
1961.
Lorenzen, Paul, and Schwemmer, Oswald [24] Konstruktive Logik, Ethik
und Wissenschaftstheorie. Mannheim: Bibliographisches Institut, 1973.
- and Lorenz, Kuno [25] Eds., Dialogische Logik. Darmstadt: Wissenschaft-
Hche Buchgesellschaft, 1978.
Luschei, E.C. [87] The Logical Systems of Lesniewski. Amsterdam: North·
Holland, 1962.
Miller, George A., and Johnson-Laird, Philip N. [56] Language and Per-
ception. Cambridge Mass.: Harvard University Press, 1976.
Orenstein, Alex [10] Willard Van Orman Quine. Boston: Twayne, 1977.
Partee, Barbara Hall [106J On the Requirements that Transformations
Preserve Meaning. In C.J. Fillmore and D.T. Langendoen, Eds., Studies
in Linguistic Semantics, pp. 1-21. New York: Holt, Reinhart, and
Winston, 1971.
- [33] Ed., Montague Grammar. New York: Academic Press, 1976.
Pelletier, F.J. [63] Ed.,Mass Terms. Dordrecht: D. Reidel, 1979.
Prawitz, Dag [27] Meaning and Proofs. Theoria 43 (1977): 2-40.
- [28J Proofs and the Meaning and Completeness of the Logical Constants.
In Jaakko Hintikka, llkka Niiniluoto, and Esa Saarinen, Eds., Essays on
Mathematical and Philosophical Logic, pp. 25-40. Dordrecht: D.
Reidel, 1979.
Quine, W.V. [5J Word and Object. Cambridge, Mass.: MIT Press, 1960.
- [6] Ontological RelatiVity and Other Essays. New York: Columbia Uni-
versity Press, 1969.
82 J. Hintikka

- [8] Philosophy of Logic. Englewood Cliffs, N.J.: Prentice-Hall, 1970.

- [7] Roots of Reference. La Salle, TIL: Open Court, 1974.
Rantala, Veikko [60] Urn Models: A New Kind of Non-Standard Model for
First-Order Logic. Journal of Philosophical Logic 4 (1975): 455-474.
- [76] Aspects of Definability. Acta Philosophica Fennica 29 (2). Amster-
dam: North-Holland, 1977.
Russell, Bertrand [86] Principles of Mathematics. Cambridge University
Press, 1903.
- [85] Our Knowledge of the External World. London: George Allen and
Unwin, 1914.
Saarinen, Esa [49] Intentional Identity Interpreted. Linguistics and Philos-
ophy 2 (1978): 151-223. Reprinted in Saarinen, Ed. [66], pp. 245-
327.
- [66] Ed., Game-Theoretical Semantics. Dordrecht: D. Reidel, 1979.
- [48] Backwards-Looking Operators in Tense Logic and in Natural Lan-
guage. In Saarinen, Ed. [66], pp. 215-244.
Savage, LJ. [51] Foundations of Statistics. New York: John Wiley, 1954.
Schwartz, Stephen P. [57] Ed., Naming, Npcessity, and Natural Kinds.
Ithaca, N.Y.: Cornell University Press, 1977.
Scott, Dana [72] Unpublished notes. Stanford University, about 1968.
- [65] The Problem of Giving Precise Semantics for Formal Languages. In
Linguaggi nella societa e nella tecnica, pp. 225-254. Milano: Edizioni
di Comunitci, 1970.
Stenius, Erik [20] Wittgenstein's 'Tractatus': A Critical exposition of Its
Main Lines of Thought. Oxford: Basil Blackwell, 1960.
Stoy, Joseph [64] Denotational Semantics. Cambridge, Mass.: MIT Press,
1977.
Thomason, Richmond [32] Ed., Formal Philosophy: Selected Papers of
Richard Montague. New Haven: Yale University Press, 1974.
Walkoe, Jr. W. [68] Finite Partially Ordered Quantification. Journal of
Symbolic Logic 35 (1970): 535-550.
Wittgenstein, LudWig [II] Philosophical Investigations. Oxford: Basil Black-
well, 1953.
- [18] Tractatus Logico-Philosophicus, with a new translation by D.F. Pears
and B.F. McGuinness. London: Routledge & Kegan Paul, 1961.
- [I2] Philosophical Grammar. Oxford: Basil Blackwell, 1974.
- [13] Philosophical Remarks. Oxford: Basil Blackwell, 1975.
Wittgenstein et la philosophie du langage
JACQUES BOUVERESSE
Universite de Paris - Sorbonne

Comme Ie remarque Richardson, 'c'est la limitation imposee a

I'activite philosophique qui est la caracteristique de la philosophie
de Wittgenstein que les philosophes contemporains ont Ie plus de
difficultes a accepter' (Richardson [1], p. 42). D'un bout a
l'autre de son itineraire philosophique, Wittgenstein est reste
convaincu que les problemes philosophiques resultent d'une
incomprehension de la 'logique de notre langage' et doivent etre
resolus (ou, plus exactement, elimines) non pas par des expli-
cations scientifiques ou pseudo-scientifiques, mais par une
description correcte de cette logique. De ce point de vue, parler
d'une 'philo sophie du langage' de Wittgenstein est certainement
trompeur; car son objectif n'etait manifestement pas de con-
struire une theorie ou une philosophie du langage, mais unique-
ment de resoudre des problemes philosophiques en general, y
compris des problemes de philosophie du langage, par une analyse
du fonctionnement de nos expressions linguistiques qui ne pre-
suppose pas, mais au contraire considere a priori comme un
obstacle a surmonter, ce qu'on appelle habituellement une 'phil-
osophie du langage' ou, plus generalement, ce que nous sommes
tentes de dire sur Ie Iangage 'quand nous philosophons'.
Tout comme Lichtenberg (qu'il estimait particulierement),
Wittgenstein aurait pu dire que 'notre philosophie fausse est
incorporee dans tout Ie Iangage'. Mais sa conception de la
philo sophie est en contradiction avee ee que Lichtenberg a cerit,
par ailleurs, sur la nature de la philosophie savante comme simple
technique d'extraction ou de purification: 'La philosophie est

Contemporary philosophy. A neW!lUrvey. VoL 1, pp. 83-112.

© 1981, Martinus Nijhoff Publishers, The Hague/Boston/London.
84 J. Bouveresse

toujours un art d'analyse (Scheidekunst), quel'on tourne la chose

comme on voudra. Le paysan utilise toutes les propositions de la
philosophie la plus abstraite, mais seulement enveloppees, dis-
simuIees, liees, latentes, comme disent Ie physicien et Ie chimiste;
Ie philosophe nous donne les propositions pures.' Chez Wittgen-
stein, l'image du therapeute prend la place de celle du chimiste. Le
point de depart, qui est constitue non pas par l'usage ordinaire du
Iangage, mais par les representations que nous nous faisons et les
descriptions que nous donnons spontanement de cet usage, con-
stitue bien une matiere premiere a traiter, mais en un sens beau-
coup plus proche de celui auquel on traite une maladie. Le phil-
osophe qui cherche it comprendre Ie fonctionnement du langage ne
peut proceder comme Ie fait normalement un scientifique en
presence de son objet: il doit essayer de Ie comprendre essentielle-
ment en reaction 'contre quelque chose qui pousse it ne pas Ie
comprendre' (Wittgenstein [2], § 109). Le materiau brut auquel
il a affaire n'est pas une philosophie impure, mais quelque chose
qui, pour Wittgenstein, est de la nature d'une mythologie; et,
pour autant que les theories philosophiques savantes sont Ie
result at d 'un simple processus d 'explicitation correctrice, elles
requierent exactement Ie meme type de traitement que les
donnees qu'elles traitent.
Comme l'ecrit Wittgenstein, 'dans notre langage est deposee
toute une mythologie' ([3J, p. 142). L'origine de la mythologie et
de la magie est, comme celle de la philosophie, a chercher ala fois
dans certaines tendances fondamentales de la nature humaine et
dans certaines caracteristiques generales du langage: ' ... La magie
repose toujours sur l'idee du symbolisme et du langage' (ibid.,
p. 237). Et notre philo sophie spontanee du symbolisme et du
langage est toujours plus ou moins apparentee a la magie (voir les
remarques de Wittgenstein sur les pouvoirs 'magiques' que nous
attribuons, par exemple, a la relation occulte du nom a l'objet
qu'il designe). Wittgenstein constate que 'toutes les theories
enfantines (in fantiles) , nous les retrouvons dans la philosophie
actuelle' (ibid., p. 246). Et il remarque, a propos du processus
mythique de la substantialisation, dont it estime avoir ete lui-
meme victime dans Ie Tractatus: 'Expulsion de la mort ou meurtre
de la mort; mais, d'un autre cote, on la represente comme
squelette, comme etant elle-meme en un certain sens morte. "As
 Wittgenstein et la philosophie du /angage 85

dead as death." "Riel! n'est aussi mort que la mort; rien n'est aussi
beau que la beaute elle-meme." L'image sous laquelle on se repre-
sente la realite est que la beaute, la mort, etc., sont les substances
pures (concentrees), alors qu'elles sont presentes dans un objet
beau comme ingredient. - Et ne reconnais-je pas ici mes propres
considerations sur "objet" et "complexe"?' (ibid., p. 242).
C'est l'origine linguistique des problemes philosophiques qui
explique, aux yeux de Wittgenstein, a la fois leur profondeur,
leur difficulte particuliere, leur· nature tout a fait differente de
celie des questions scientifiques et leur persistance. 'On en tend
toujours a nouveau, a-t-it ecrit, la remarque selon laquelle la
philosophie ne fait a proprement parler aucun progres, que les
memes probIemes philosophiques qui occupaient les Grecs nous
occupent encore. Mais ceux qui disent cela ne comprennent pas la
raison pour laquelle cela ne peut pas ne pas etre ainsi. Elle est que
notre langage est reste identique a lui-me me et no us entrafne
toujours a nouveau vers les memes questions. Tant qu'il y aura un
verbe "etre", qui donne l'impression de fonctionner comme
"manger" et "boire", tant qu'il y aura des adjectifs "identique",
"vrai", "faux", "possible", tant qu'il sera question d'un flux du
temps et d'une extension de l'espace, etc., etc., les hommes se
heurteront toujours anouveau aux memes difficultes enigmatiques
et auront les yeux fixes sur quelque chose qu'aucune explication
ne semble pouvoir eliminer' (Wittgenstein [4], p. 36).
En un certain sens, Wittgenstein a pris au serieux l'idee
traditionnelle que la philosophie devrait apporter la solution
complete et definitive des problemes qu'elle traite et en a tire la
consequence (a bien des egards triviale) que cette solution ne peut
pas etre donnee sous la forme d'explications ou de theories. Pour
ceux qui sont convaincus que les questions des philosophes ne sont
pas reellement plus 'profondes' que celles des scientifiques et que
les solutions de la philosophie sont aussi partieIles, provisoires et
progressives (en depit des apparences) que celles de la science, les
considerations mHaphilosophiques de Wittgenstein et les moti-
vations essentielles de son entreprise philosophique sont tout
simplement inintelligibles. Comme on I'a remarque, il y a au moins
un point commun important entre la philo sophie de Wittgenstein
et celle de Heidegger: la conviction que les methodes et les objec-
tifs de la philosophie sont fondamentalement differents de ceux
86 J. Bouveresse

des sciences. Wittgenstein est reste jusqu'au bout un defenseur

acharne de la specificite absolue de la philosophie. Mais il est reste
egalement, sur un point essentiel, un disciple de Frege, convaincu
que c'est non seulement une tache, mais fa tache de la philo sophie ,
'de briser l'empire du verbe sur l'esprit humain, en devoilant les
illusions que l'usage linguistique fait naitre souvent de fac;:on
presque inevitable sur les relations des concepts' (Frege [5],
p. XII). S'il a renonce a la methode preconisee par Frege (la con-
stitution d'une langue formelle), il n'a jamais eu d'hesitation sur Ie
but; et tous les rapprochements qui ont ete tentes ne changent rien
au fait que sa preoccupation pour Ie langage ne peut en aucune
fac;:on etre comparee acelIe de Heidegger.
Cela serait une pietre objection que de faire remarquer qu'il n'y
a pas un langage, mais des langues et que, contrairement ace que
suggere Wittgenstein, elles ne sont pas restees les memes au cours
de l'histoire. Sur Ie premier point, la reponse est qu'un philosophe
a a resoudre des problemes qui se posent effectivement, et non des
problemes qui se posent necessairement. Wittgenstein a souligne
lui-meme que certaines questions philosophiques ne se poseraient
pas, en tout cas pas naturellement, si nous utilisions un langage
different de celui que nous utilisons en fait, et s'est adonne a
l'occasion a ce que ron pourrait appeler la 'linguistique fictive'. II
est question ici de Ia permanence des problemes a l'interieur d 'une
certaine tradition, et non de leur caractere universel ou universal-
isable. Sur Ie deuxh~me point, on peut remarquer que Wittgenstein
etait plus conscient qu'aucun autre de la contingence, de I'his-
toricite et de l'imprevisibilite de nos jeux de langage et de nos
formes de vie. De fac;:on generale, en depit de l'absence presque
complete de references aux doctrines traditionnelles dans son
oeuvre philosophique, de son mepris bien connu de l'exactitude et
de I'erudition historiques et de sa mHiance caracteristique it
regard de to ute philo sophie de l'histoire, il avait certainement un
sens historique bien plus reel et plus profond qu'on ne l'a
generalement suppose. Le langage peut evoluer de fac;:on plus ou
moins importante sous certains aspects et les hommes peuvent,
pour des raisons historiques, etre plus ou moins endins a em-
prunter les voies sans issue qu'il leur propose; mais il n'en con-
tinue pas moins a receler dans ses formes la meme mythologie
latente et les memes errements potentiels: 'Le langage tient prets
 Wittgenstein et la philosophie du /angage 87

pour tout Ie monde les memes pil~ges; I'immense reseau de fausses

routes tres practicables' (Wittgenstein [4], p. 41). Comme dit
Wittgenstein, dans la sante meme de l'entendement, nous sommes
environnes en pennanence par la folie. On peut evidemment se
perdre dans Ie Iabyrinthe du Iangage de bien des fa~ons, en suivant
les chemins les plus immemoriaux et les plus permanents, ou
au contraire les plus nouveaux et les plus circonstanciels, et meme
a la limite ne pas s'y perdre du tout, la question tout a fait class-
ique, que Wittgenstein s'est posee comme beaucoup d'autres
philosophes (cf. par exemple ibid., p. 58), etant alors d'apprecier
exactement Ie gain que represente la resolution (c'est-a-dire
l'elimination) des problemes philosophiques par rapport :i leur
simple absence. Mais il y a au moins une propriete constitutive du
langage qui ne peut reellement changer et qui constitue, en
derniere analyse, la source de toutes les difficultes philosophiques:
son enorme puissance d'uniformisation, 'Ie pouvoir qu'a Ie langage
de tout egaliser, qui s'exprime de la fa~on la plus crasse dans Ie
dictionnaire et qui fait que Ie temps a pu etre personnifie, ce qui
n 'est pas moins etrange que si nous avions des divinites des con-
stantes logiques' (ibid., p. 49), ce pouvoir qui fait de la lutte
contre Ie lang age (cf. ibid., p. 30) essentiellement un combat pour
la difterence.
Wittgenstein s'est presente lui-meme essentiellement comme un
penseur 'reproductif et reactif. II a fait remarquer que, meme dans
certains de ses essais artistiques (comme la sculpture), il n'avait pas
fait reellement oeuvre originale, mais avait plutot reagi a des
choses existantes sous la fonne d 'une tentative de clarification
(cf. ibid., p. 43). II aurait pu dire egalement que, meme en cela,
il s'etait comporte d'une maniere typiquement philosophique. Car,
meme si son oeuvre philosophique donne, selon les cas, l'impres-
sion d'apporter beaucoup plus ou beaucoup moins, il n'a jamais
con~u la philo sophie autrement que comme un 'travail sur soi-
meme' (ibid., p. 38), dont Ie but est uniquement la clarte et
l'apaisement: 'La paix dans les pensees. C'est Ie but auquel aspire
celui qui philosophe' (ibid., p. 87). Par rapport a ce but final - la
suppression des inquietudes philosophiques - ce qui se presente
dans ses ecrits sous la fonne d'une explication, d'une theorie ou
d'un resultat philosophique au sens usuel ne devrait en principe
constituer qu'un moyen plus ou moins aleatoire et personnel ou
88 J. Bouveresse

un sous-produit. De fa<;on generale, Wittgenstein n 'avait certaine-

ment aucun desir de concurrencer les scientifiques en general
(y compris les linguistes) et pas davantage de leur faciliter les
choses: il avait, au contraire, Ie sentiment que son influence sur
eux serait tout au plus dissuasive (cf. ibid., p. 118) et que la
recherche de la clarte pour elle-meme etait, du point de vue
theorique et seientifique, une entreprise non seulement impro-
ductive, mais meme, en un certain sens, parfaitement negative.
Comme il l'a eerit a propos des mathematiques: 'La clarte phil-
osophique aura sur la croissance des mathematiques la meme
influence que la lumiere du soleil sur la poussee des germes de
pommes de terre. (Dans la cave obscure, ils poussent jusqu'a des
metres de 10ngueuL)' (Wittgenstein [6] , p. 381 ).

II

II est particulierement difficile de prendre reellement au serieux,

lorsqu'il est question de la philosophie du langage de Wittgenstein,
ce qu'il a ecrit sur la nature des problemes philosophiques en
general, leur parente avec les problemes esthetiques, l'impossibilite
de les resoudre en formulant des theses ou meme des opinions, etc.
S'il y a un domaine OU il donne I'impression d'avoir abondamment
'theorise', et parfois de fa<;on contestable, c 'est bien Ie langage.
Pourtant, si I'on considere que les theories auxquelles il s'en est
pris ont, de son point de vue, la particularite de ne pas ou de ne
plus etre des theories, dans la mesure ou elles sont ou sont devenues
incapables de se percevoir eomme telles, on peut admettre qu'ef-
fectivement il n'a pas essaye de les remplacer par d'autres. Comme
ill'ecrit dans les Fiches (Zettel):
'Nous avons a present une theorie; une theorie "dynam-
ique" de la proposition, du langage, mais el1e ne nous
apparait pas comme une theorie. L'tilement caracteristique
d'une telle theorie est qu'elle considere un cas particulier,
clairement intuitif (anschaulich), et dit: "Cela montre com-
ment les choses se passent en gemhaL ce cas est Ie modele
de taus les cas." - "Naturellement! C'est ainsi que cela doit
etre", disons-nous, et nous sommes satisfaits. Nous en
sommes arrives a une forme de representation qui nous
 Wittgenstein et la philosophie du langage 89

donne I'impression d'evidence. Mais les chases se passent

comme si nous avions a present vu quelque chose qui est
sous la surface.
La tendance a generaliser Ie cas clair semble avoir en
logique sa justification stricte: on semble ici inferer avec une
entiere justification: "Si une proposition est une image,
alors toute proposition doit etre une image, car elles doivent
toutes avoir la meme nature." Car nous sommes bel et bien
dans l'illusion que Ie sublime, l'essentiel dans notre recherche
reside dans Ie fait qu'eUe apprehende une essence qui englobe
tout' (Wittgenstein [7J , § 444).

Le point crucial est done que la generalisation a partir du cas

favorable, qui paraitrait suspecte ailleurs, semble ici une operation
normale, qui engendre l'illusion de la profondeur et du definitif.
Le dogmatisme est, pour Wittgenstein, I'attitude philosophique qui
fait dependre d'un supplement d'information la reconnaissance du
fait que les choses sont effectivement ce qu'elles doivent etre:
lorsque nous aurons pousse suffisamment loin I'analyse des pro-
positions, nous constaterons qu'eUes sont reellement conformes a
ce que nous avions prevu ou, plus exactement, postuh!. Wittgen-
stein s'est reproche d'avoir assimile, dans Ie Tractatus, tous les cas
a un cas clair et d'avoir cru exhiber la veritable nature de la pro-
position, alors qu'il proposait, en fait, une simple comparaison.
Car, lorsque quelqu'un affirme que la proposition est une image,
on pourrait egalement faire remarquer que 'ce qui apparait comme
un objet de comparaison pour la proposition est un cas particulier
de notre concept general'. Ou encore: 'Lorsque j'ai compare la
proposition a une regIe graduee, j'ai, a strictement parler, utilise
une indication de longueur donnee a I'aide d'une regIe graduee
comme exemple pour toutes les propositions' ([ 6], p. 164).
Ce qui se passe en pareil cas est simplement que I'on impose aux
phenomenes un mode de description uniforme, avec Ie double
risque de la meconnaissance des faits et de la trivialite. Car 'Ie
modele doit precisement etre instaure comme tel; de telle fa90n
qu'it caracterise toute la maniere de. considerer les choses,
determine sa forme. II est, par consequent, au sammet et se trouve
etre universellement valide par Ie fait qu'il determine la forme de
l'examen des chases, et non pas par Ie fait que ce qui n'est vrai
90 J. Bouveresse

que de lui est enonce de tous les objets que 1'0n examine' ([4],
p. 35). La consequence est que 'I'on croit suivre encore et toujours
la nature, alors que 1'0n ne fait que passer Ie long de la forme a
travers laqueUe nous la considerons' ([2], § 114).
Quand nous philosophons sur Ie langage, remarque Wittgenstein,
'nous sommes dans l'illusion que ce qu'il y a de particulier, de
profond, d 'essentiel pour nous dans notre recherche reside dans Ie
fait qu'elle essaie de comprendre I'essence incomparable du
langage. C'est-a-dire, l'ordre qui existe entre les concepts de la
proposition, du mot, de l'inference, de la verite, de l'experience,
etc. Cet ordre est un super-ordre entre - pour ainsi dire - des
super-concepts. Alors qu'en fait les mots "Iangage", "experience",
"monde", s'ils ont une utilisation, doivent en avoir une qui est
d'un niveau aussi bas que celle des mots "table", "lampe".
"porte'" (ibid., § 97). Notre attitude n'est pas celie qu'un
theoricien adopte habituellement a l'egard de l'objet de ses
recherches. Nous avons l'impression que, dans Ie cas precis, il s'agit
de decouvrir non pas un ordre possible ou hypothetique, mais bel
et bien l'ordre commun au langage, a la pensee et au monde --
completement a priori (parce qu'il est 'l'ordre des possibilites qui
doivent etre communes au monde et a la pensee '), parfaitement
simple et transparent (non contamine par les impuretes et les
ambiguHes de l'experience) et omnipresent (en ce sens qu'i!
penetre et gouverne de l'interieur tous les phenomenes).
Cet ordre, qui se manifeste sur Ie mode contradictoire de
l'idealite realisee, c'est-a-dire de la presence-absence, est a la fois
'superieur' et particulierement intime et dissimule. Or Ie pathos
du sublime et celui de la profondeur constituent precisement, aux
yeux de Wittgenstein, les deux symptomes essentiels de la 'super-
stition' philosophique. Ce qui est Ie plus difficile a admettre est
que la proposition est une chose tout a fait ordinaire, et non une
entite unique en son genre et dotee par elle-meme de pouvoirs
etranges et inexplicables, que nous pouvons decrire Ie fonction-
nement du langage sans avoir a utiliser necessairement des con-
cepts d'un ordre plus eleve que ceux qu'il nous fournit et que
taus les faits importants sont, d'une certaine maniere, sur Ie
meme plan et a la surface (ce sont meme des faits si evidents que
seule leur trop grande familiarite nous empeche de les remarquer
et de no us en etonner). A la difference de la science, la philosophie
 Wittgenstein et fa philosophie du fangage 91

n'a pas la 'structure stratifiee d'une theorie' (Hacker [8], p. 119).

II en va evidemment de la philosophie du langage comme du reste
de la philosophie; et il est difficile de comprendre ce qui a pu
autoriser certains linguistes a interpreter comme une prise de
position en faveur de la linguistique 'taxinomique', par opposition
a la linguistique 'democriteenne' des remarques qui ont pour but
de caracteriser la specificite de l'approche philosophique du
Iangage par rapport a celie des sciences du langage e1Ies-memes, y
compris et en un certain sens par anticipation (dans la me sure ou
les chomsky ens lui reprochent d'avoir mis tant de temps a adopter
les procedures en usage dans toutes les autres sciences) Ia linguis-
tique. Wittgenstein veut dire que Ie langage pose au philosophe des
probh~mes profonds, en ce sens qu'ils naissent d'illusions profonde-
ment enracinees dans Ie Iangage et sont profondement revelateurs
de la nature de notre Iangage, mais non en ce sens que leur
solution pourrait dependre d 'hypotheses sur la n~alite profonde
('sous-jacente', comme dit Katz) du langage.
La contradiction apparente qui existe entre la pratique de
Wittgenstein et ses declarations sur la nature essentiellement
metaphilosophique et therapeutique de l'activite philosophique a
incite un certain nombre de commentateurs et d'interpretes a
considerer celles-ci comme denuees de pertinence reelle et plus ou
moins negligeables. Pour comprendre l'evolution des idees de
Wittgenstein sur ce point, i1 faut tenir compte du fait que, tout
comme sa critique de l'ancienne philosophie est Ie plus souvent
une critique de son ancienne philosophie, son rejet apparent de
la theorie en general est d'abord un rejet du genre de theorie
qu'il avait propose dans Ie Tractatus. A bien des egards, l'utili-
sat ion qu'il avait faite, dans cet ouvrage, de la logique comme
tt~chnique de decouverte ou d'anticipation applicable immectiate-
ment et entierement a priori au langage et a la realite, a con-
stitue par la suite, a ses yeux, Ie prototype du procecte phil-
osophique dogmatique. II ecrivait en 1913 a Russell: 'N'est-il pas
extremement remarquable de voir a quel point Ia logique est une
science grande et infiniment singuliere; je crois que ni toi ni moi
ne savions cela il y a un an et demi' ([ 9], p.4S). II en est venu par la
suite a considerer que meme la pretention plus limitee qu'avait
la logique d'exhiber Ia vraie nature du langage, du raisonnement et
de la 'realitC' mathematiques etait completement iIlusoire et a
92 J. Bouveresse

denoncer 'la funeste intrusion de la logique dans les mathematiques'

(une fonnule qui constitue peut-etre une parodie de Fregc: 'der
verderbliche Einbruch der Psychologie in die Logik', (Frege [10] ,
p. XIV». Le rejet du psychoiogisme, sous l'influence de Frege,
a ete suivi, chez Wittgenstein, d 'un rejet presque aussi categorique
du logicisme sous toutes ses formes. La logique philosophique
(au sens du Tractatus) est Ie paradigme de la theorie philosophique
qui considere l'ideal ou Ie modele non pas comme ce qu'il est .-
un point de comparaison ou un etalon de mesure (que I'on a
choisi et que I'on aurait pu choisir autrement)- , mais comme
quelque chose qui doit deja etre hi, en depit de toutes les
apparences du contraire, comme 'un prejuge auquel la realite doit
se conformer' (Wittgenstein [2], § 131). Ce qui est deconcertant
est que Wittgenstein, par reaction, a donne plus tard I'impression
de contester purement et simplement qu'il y ait interet a adopter
ce genre d'ideal ou de modele en Ie prenant simplement pour ce
qu'if est.
Dans Ie Tractatus, la logiq ue fonctionne avec Ie caractere a la
fois contraignant et fascinant d 'une mythologie; et elle possecte au
moins un element commun important avec la psychologie: Ie
recours plus ou moins ad hoc a des entites et des processus etheres
et inaccessibles sur lesquels nous savons (pour l'instant) tres peu de
chose et qui ont pour fonction, dans Ie cas precis, de concretiser
l'ideal sous la forme d'une realite occulte. Par la suite, Wittgen-
stein a associe regulierement dans sa critique Ia 'mythologie du
symbolisme' et la 'mythologie des processus mentaux', qui
reposent toutes les deux sur l'idee que les choses essentielles
pour la comprehension du phenomene linguistique se passent en
profondeur dans un milieu etrange (dans un cas, I'esprit, dans
l'autre une sorte d'univers du sens, OU deux negations sont
supposees donner une affirmation en vertu des lois logiques, a
peu pres au sens OU du carbone et de l'oxygene donnent de
l'anhydride carbonique en vertu des lois physicochimiques).
L'insistance de Wittgenstein sur les notions d'usage et de pratique
CCest la pratique qui donne aux mots leur sens') est bien moins
une tentative d'explication qu'une reaction c~ntre ce type de
representation et une incitation a voir Ies choses d 'une maniere
completement differente. Quand nous disons, par exemple, qu'il
est extraordinaire que Ie \angage puisse en quelque sorte 'alIer
 Wittgenstein et la philosophie du langage 93

jusqu'a la realite' ou que I'on puisse 'penser quelque chose qui

n'est pas', nous ne voulons pas dire qu'il s'agit du genre de
phenomene extraordinaire que Ia science finit Ie plus souvent par
expliquer; et pourtant nous ne parvenons pas a renoncer complete-
ment a l'idee d'une explication de ce genre, qui devrait etre
obtenue par la construction d'une theorie de la signification ou
de la pensee. La difficulte ne provient pas d 'un phenomene etrange
qui resiste a l'explication, mais plutot d'une chose ordinaire que
nous ne parvenons pas a apprehender comme telle.

III

On pourrait naturellement objecter a Wittgenstein que toute

theorie, qu'elle soit scientifique ou philosophique, peut etre
victime de l'illusion essentialiste (c'est un des reproches essentiels
qu'il adresse, par exemple, a la theorie freudienne du reve) et se
transformer tot ou tard en un mode de representation obsessionnel
(les exemples abondent dans l'histoire des sciences), ou encore
qu'une theorie aussi respectable que l'atomisme physique n'etait
a l'origine pas moins philosophique, speculative et dogmatique
que I'atomisme logique, et I'est restee, en outre, pendant tres long-
temps. II est difficile de ne pas juger tout a fait excessive et incon-
grue Ia severite avec laquelle Wittgenstein a juge sa premiere
philo sophie du langage, quand on Ie voit remarquer, par ailleurs,
que 'Ie merite veritable d'un Copernic ou d'un Darwin n'etait pas
la decouverte d'une theorie vraie, mais d'un nouvel aspect fecond'
([ 4] , p. 42). Apres tout, il a reconnu lui-meme que l'identification
de la proposition a une image l'avait aide a mettre en evidence un
aspect interessanl de la nature de la proposition, ou encore que
'dire que la proposition est une image fait ressortir certaines
caracteristiques dans la grammaire du mot "proposition'" ([ 6] ,
p. 163),
Meme si ce nouvel aspect ne s'est pas revele aussi fecond qu'il
l'avait espere au debut, on ne peut pas dire non plus qu'il ait ete
comph,:tement illusoire au improductif. Comme I'a souligne
notamment Kenny, Wittgenstein n'a certainement jamais considere
la 'picture-theory' comme une absurdite intrinseque, et il en a
incontestablement retenu quelque chose (probablement beaucoup
94 J. Bouveresse

moins, cependant, que ne Ie pense Kenny) dans sa deuxieme

philosophie. II n'a certainement pas non plus voulu exclure a priori
qu'elle puisse etre reinterpretee et exploitee de differentes
manieres et se reveler feconde dans une perspective et une
direction autres que celles auxquelles il songeait (par exemple,
celles qui ont ete suggerees par Hintikka). De plus, si Ie but du
philosophe est principalement de nous amener it une certaine
fa((on de voir les choses, l'utilisation de la theorie, au moins
comme 'revelateur', en plus de la description minutieuse de
faits generaiement negliges, y est apparemment aussi indis-
pensable que dans les sciences.
'Pourquoi', s'est demande Wittgenstein, 'ne dois-je pas utiliser
des expressions a l'encontre de leur usage primitif? N'est-ce pas,
par exemple, ce que fait Freud, lorsqu'il appelle meme un reve
d 'angoisse un reve de desir? Ou est la difference? Dans la fa((on
scientifique de considerer les choses, Ie nouvel usage est justifie
par une theorie. Et si cette theorie est fausse, alors Ie nouvel usage
etendu doit eire abandonne. Mais dans la philosophie ce n'est pas
sur des opinions vraies ou fausses concernant des processus
naturels que s'appuie l'usage elargi. Aucun fait ne Ie justifie, aucun
ne peut Ie soutenir' ([ 4] , p. 88). Or, d'une part, il est clair que les
'faits de langage', les seuls qui, du point de vue de Wittgenstein,
interessent directement Ie philosophe, peuvent parler jusqu 'a un
certain point pour ou contre un usage modifie d'expressions ordi-
naires dans la description que l'on veut en donner. D'autre part,
l'utilisation que fait Wittgenstein de termes cruciaux comme
'langage ordinaire', 'jeu de langage', 'grammaire', 'usage', 'para-
digme', 'critere', etc., n'est manifestement pas completement
independante d'une theorie ou, en tout cas, de certaines pre-
suppositions et options theoriques, qui sont restees, chez lui,
largement implicites et sans doute parfois inaper((ues. (Cela ne
signifie pas, il est vrai, qu'elle aura it besoin, en toute rigueur, de
la justification d 'une theorie, au sens indique plus haut).
Ce qui est etrange est de voir Wittgenstein affirmer simul-
tanement que l'element essentiel dans l'oeuvre de theoriciens
comme Darwin ou Freud a ete la production d'un arrangement
ec1airant entre des faits a premiere vue disparates, que cette
predominance de l'aspect 'esthetique' sur l'aspect causal de l'expli-
cation est egalement une caracteristique de la methode philoso-
 Wittgenstein et la philosophie du langage 95

phique et que, neanmoins, l'activite philosophique n'a rien a voir

avec la production de theories. Sans doute les remarques
apparemment tout a fait 'positivistes' sur la theorie darwinienne
on d'autres du meme genre relevent-elles bien moins de l't~pis­
temologie que de ce que I'on pourrait appeler une phenomenologie
de la persuasion et de la conviction theoriques en general. Mais,
precisement. de ce point de vue, il est particulierement difficile
d'etablir une distinction tranchee entre Ie cas des theories scien-
tifiques et celui des theories philosophiques. Certaines theories
ont tendance a s'imposer irresistiblement comme vraies, indepen-
damment de toute espece de confirmation, acause de la simplicite,
de I'unite, de la coherence et de la symetrie qu'elles conferent aux
phenomenes consideres. Dans Ie cas de la philosophie, nous nous
trouvons finalement accules a une contradiction insoluble, parce
que nous sommes dans I'incapacite de renoncer a un certain type
d'exigence: '''Ce n'est pourtant pas ainsi!" - disons-nous. "Mais
cela ne peut cependant pas ne pas etre ainsi!'" ([ 12] , § 112). Mais
y a-t-il rt!ellement une difference de nature entre la necessite ou
l'impossibilite rigides qui s'expriment dans Ie 'doit' ou Ie 'ne peut
pas' du prejuge philosophique typique et I'element prescriptif ou
prohibitif plus ou moins flexible que contiennent normalement
les theories scien tifiq ues?
Un element de reponse a ce genre de question est probablement
fourni par la remarque:
•Aussi longtemps que sous Ie nom d' "arne" on se repre-
sente une chose, un corps, qui est dans notre tete, cette
hypothese n'est pas dangereuse. Ce n'est pas dans I'imper-
fection et la grossierete de nos modeles que se situe Ie danger,
mais dans leur caractere obscur (indistinct).
Le danger commence lorsque nous remarquons que
I'ancien modele ne suffit pas et que. dans ces conditions.
nous ne Ie modifions cependant pas, mais ne faisons pour
ainsi dire que Ie sublimer' ([ 2] , p. 287).
Tout comme il est moins dangereux, a tout prendre, de dire que la
pensee est 'dans ma tete' que de dire qu'elle est 'dans mon esprit',
il est moins dangereux de considerer Ie signe propositionnel
comme une sorte de tableau que d'en appeler, pour ne pas avoir
a renoncer a un modele de ce genre, a 'une entite intermediaire
96 J. Bouveresse

pure entre Ie signe propositionnel et les faits' ([2], § 94). Ce que

Wittgenstein denonce est la tendance a faire correspondre a J'objet
de la description une sorte de double ideal ou a purifier et
sublimer l'objet lui-me me pour Ie rendre conforme ala description
que I'on en veut donner, une procedure dont il souligne qu'elle
n 'est pas reellement comparable ala simplification et l'idealisation
qui interviennent constamment dans les sciences de la nature.
L 'operation suspecte est celle qui transforme une description idea-
lisante inadequate de la realite empirique en une description
directe d 'une realite ideale. C'est ce qui se passe, pour Wittgenstein,
lorsqu'on dit, par exemple, que 'Ia logique represente une n~alite
idealisee, qu'elle ne vaut en toute rigueur que pour une langue
ideale, et d'autres chosen du meme genre' «61, p. 77), en oubliant
que ce pn!tendu 'ideal' est simplement quelque chose que nous
avons constnlit et stipule.
Les philosophes pourraient remarquer, non sans raison, que les
problemes dont ils s'occupent n'ont pas forcement Ie caractere
repetitif, obsessionnel et plus ou moins morbide que leur attribue
Wittgenstein, et que sa tendance caracteristique a concevoir la
philosophie en termes de plaies et bosses, maladies, envoutements,
crispations mentales, etc., n'est pas du tout conforme a la realite
du travail philosophique. Mais ce genre d'objection est sans grande
portee, !orsqu'on a affaire a un philosophe qui, en un certain sens,
n'a jamais pretendu etre objectif sur ce point, n'a jamais cesse de
reagir contre ses propres tentations et de regler des comptes avec
soi-meme et a admis explicitement que la raison d 'etre de ses
efforts pourrait echapper completement a des hommes qui n'ont
pas ou qui, pour des raisons historiques, n 'auront plus Ie genre de
reactions. d'inquietudes et de prejuges qu'il presuppose (cf. par
exemple [4], p. 86). Wittgenstein avait tendance a croire que
l'elimination complt~te des problemes philosophiques resulte
finalement moins de la contribution personnelle d'un individu
que d'un changement global d'attitude determine par des raisons
plus ou moins externes. et il s'est meme demande s'il ne souhaitait
pas davantage les voir disparaitre en ce sens-Ia que d 'avoir des
disciples susceptibles de continuer son oeuvre (cf. ibid .. p. 1 17).
De ce point de vue, les remarques de Hacker a propos de son
'dedain des considerations architectoniques' (op. cit .. p. 139) sont
a la fois tres pertinentes en soi et peu wittgensteiniennes. 'II n'est
 Wittgenstein et La philosophie du Langage 97

pas du tout evident, ecrit Hacker, que I'on ne puisse pas a la fois
rendre justice a I'indetermination du sens, au caractere ouvert des
concepts et a la flexibilite du langage, et egalement etre systemati-
que et englobant' (p. 140). De fait, 'bien des aspects de la
philosophie critique de Wittgenstein peuvent etre presentes de
fa~on systematique et revelent aisement Ie caractere acheve et
englobant de son oeuvre. Et, ce qui est plus important, il est loin
d'etre evident que ses dernieres conceptions sur la semantique ne
puissent pas etre representees de fa«on coherente et bel et bien
formalisees, de fayon a fournir ce que la plupart des philosophes
appelleraient une theorie englobante de la signification' (p. 141).
Mais, s'il est vrai que la philosophie n'est pas reellement une
activite theorique, it doit etre vrai egalement qu'un probleme
devenu proprement theorique n'est plus exactement un probleme
de philosophie. Chez Frege, !'idee d'une 'Iogique du vague' aurait
suscite des reactions comparables acelles qu'a pu provoquer autre-
fois I'idee d'une 'science de l'incertitude' (selon I'expression qui
a ete utilisee a propos du calcul des probabilites). Mais Iorsque
Ie flou et l'imprecision des concepts usuels ont perdu Ie caractere
philosophiquement preoccupanL voire franchement inacceptable,
qu'ils avaient pour lui (cf. Frege [10], II, § 56) et ont fini par
devenir un fait accepte et familier, en meme temps que I'objet
d 'une theorisation systematique, sous la forme d'une logique
du vague ou de quoi que ce soit de ce genre, les problemes qu'ils
posent ne ressemblent plus que superficiellement ace que Wittgen-
stein appelle un probleme philosophique. Pour lui, la philosophie
proprement dite intervient avant, au stade des resistances et des
prejuges instinctifs qui s'opposent a Ia perception des faits et
(secondairement) a la construction eventuelle d'une theorie, ou
apres coup, dans la mesure ou une theorie nouvelle et des con-
cepts nouveaux entrainent a peu pres inevitablement des
problemes de clarification nouveaux.

IV

Ce qui rend Ie cas de Wittgenstein tout a fait exceptionnel et en un

certain sens unique est qu'il a non seulement elabore deux
philosophies a premiere vue tres differentes, mais egalement utilise
98 J. Bouveresse

largement la premiere comme materiau de base pour une critique

des formes les plus classiques et les plus repandues de l'illusion
philosophique. 'Ie ne fais, a-t-il dit, qu'attirer l'attention de l'autre
sur ce qu'il fait veritablement, et je m'abstiens de toute affir-
mation' (Wittgenstein [11], p. 186). Un aspect essentiel de sa
reflex ion philosophique a consiste a essayer de com prendre ce
qu'il avait veritablement fait dans Ie Tractatus; et il en est venu
a traiter sa premiere philosophie exactement comme il aurait pu
traiter Ia philo sophie de queIqu'un d'autre, avec Ie meme genre
d'estime et de severite dont il faisait preuve a l'egard d'auteurs
comme Platon, saint Augustin ou Frege.
Kreisel, qui a suggere que l'idee essentielle du Tractatus avait
consiste simplement a concevoir l'analyse des propositions sur Ie
modele de l'analyse chimique des molecules, a conteste que Ie
modele ou l'image en question aient jamais pu, selon l'expression
de Wittgenstein, 'envouter' reellement qui que ce soit. Mais c'est
un fait que l'auteur du Tractatus a rejete sa premiere philosophie
du langage non pas comme quelqu'un qui abandonne une theorie
partielle, partiale, ambitieuse et irrealiste (ce qui est, a des degres
divers, Ie cas de la plupart des theories) en presence de faits
recalcitrants, mais parce qu'elle impliquait a ses yeux une con-
ception inadequate de la problematique et de la methode
philosophiques. II a utilise par la suite de nombreux autres
modeles, analogies, comparaisons, images, etc., avec la conscience
qu'un mode de description qui a pour but de nous faire voir les
faits d 'une certaine maniere est toujours susceptible, aun moment
donne, de nous empecher purement et simplement de les voir.
Lutter pour un certain 'style de pensee', par opposition a un
autre, selon la caracterisation que Wittgenstein a donnee lui-meme
de son enseignement, c'est toujours, entre aut res choses, essayer
d'imposer sa 'preference pour certaines images (Gleichnisse),; et
il a dit de lui qu'i! etait 'au fond un peintre, et souvent un tres
mauvais peintre' ([4], p. 156). C'est, bien entendu, en ce sens-Ia
plutOt qu'au sens de la coherence et de la systematicite theoriques
qu'il faut interpreter la recherche de l'ilbefsichtliche Darstellung,
qui joue un role si important dans sa deuxieme philosophie.
Or, il s'est reproche precisement d 'avoir utilise l'image centrale
du Tractatus non pas comme un 'artiste', mais comme un
physicien ou un chimiste, c'est-a-dire avec des presupposes com-
 Wittgenstein et la philosophie du langage 99

pietement errones concernant la nature de l'analyse logique, de la

signification et de la proposition: 'Ma conception dans Ie Trac tatus
logico-philosophicus etait fausse: (1) parce que je n'avais pas les
idees claires sur Ie sens de l'expression 'dans une proposition est
dissimu!e un produit logique' (et d'autres semblables), (2) parce
que moi aussi je croyais que l'analyse devait mettre au jour des
choses cacMes (comme Ie font l'analyse chimique et l'analyse
physique).' ([6], p. 210). Le 'concept errone de l'analyse logique',
qui avait ete, selon Wittgenstein, celui de Russell, de Ramsey, de
Carnap (dans I'Aujbau) et de lui-meme, entrafne comme con-
sequence que 'l'on attend une analyse logique finie des faits,
comme on attend une analyse chimique finie de combinaisons.
Une analyse par laquelle, par exemple, on trouve nSellement une
relation a 7 places, comme on trouve un element qui a effective-
ment Ie poids specifique 7' (ibid., p. 311-312).
Une etape decisive dans l'evolution des idees de Wittgenstein sur
Ie langage a consiste a se defaire du prejuge qui attribue a nos
expressions linguistiques une signification 'n~el1e' que l'analyse
logique doit decouvrir et que, par consequent, nous ne connaissons
peut-etre pas encore veritablement. 'Les philosophes, constate
Wittgenstein, parient tres souvent d'etudier, d'analyser la signi-
fication des mots. Mais n'oublions pas qu'un mot n'a pas acquis
une signification qui lui a ete donnee, pour ainsi dire, par une
puissance independante de nous, de sorte qu'il pourrait y avoir
une espece de recherche scientifique sur ce que Ie mot signifie
reellement. Un mot a la signification que quelqu'un lui a donnee'
(The Blue Book, p. 27-28). Cela signifie, bien entendu, unique-
ment que Ie langage est d'abord une creation de l'homme, et non
pas qu'il s'agit d'une creation volontaire et concertee, qu'il est
apparu et a evolue en fonction de decisions qui ont ete effec-
tivement prises. (Wittgenstein a insiste, au eontraire, speeialement
dans ses toutes dernieres remarques philosophiques, sur Ie fait
qu'it s'est developpe a partir de formes d 'action et de reaction
pre-rationnelles, et non a partir de choix plus ou moins motives).
C'est, paradoxalement, Ie modele trompeur de la demarche seien-
tifique qui est a l'origine de Ia plupart des attitudes 'meta-
physiques' que Wittgenstein a stigmatisees: Ie mepris du cas par-
ticulier et coneret, l'idee que l'usage complique et ramifie que
nous faisons de certains termes generaux doit etre unifie en
100 J. Bouveresse

profondeur par Ie sens cache, l'impression que des mots comme

'penser' ou 'comprendre' doivent designer un processus determine
sous-jacent a tous les phenomenes (superficiels) de la pensee et de
la comprehension interpretes comme de simples symptomes,
l'assimilation de la connexion entre Ie sens et l'explication de sens
a une relation du genre de celle qui existe entre un phenomene
naturel et l'explication que nous en donnons, Ie besoin d'expliquer
(au sens de reduire) avant d 'avoir decrit, et ce que Kreisel appelle
notre 'tendance gem!rale a decrire Ie langage et, en particulier,
la pratique mathematique, d ['aide de concepts dont Ie niveau
d'abstraction est plus eieve que Ie minimum dont on a besoin en
rea lite' (Kreisel [12]).
I1 est devenu habituel, depuis quelque temps, de caracteriser la
theorie de la signification que Wittgenstein a proposee dans sa
deuxieme periode comme 'constructiviste' (par opposition a la
conception 'realiste' du Tractatus) et d'insister a ce propos sur
l'influence de Brouwer. Comme Ie remarquent Baker et Hacker,
'la 'deuxieme philosophie' de Wittgenstein, comme on l'appelle,
est unifiee par son adhesion au constructivisme et sa repudiation
du realisme' (Baker [13] ,p. 270). Le realisme dont il est question
ici est la conception selon 1aquelle la signification d 'une phrase
(declarative) est donnee par ses conditions de verite, indepen-
damment de la possibilite que nous avons de savoir si eUes sont ou
non realisees: 'Comprendre une proposition veut dire savoir ce
qui est Ie cas lorsqu'elle est vraie' (Wittgenstein [14] , 4.024). Du
point de vue constructiviste, la signification d 'une phrase de ce
genre est donnee plutot par ses conditions d'assertion, par ce que
nous considerons comme une justification ou une raison qui
autorise a l'affirmer (dans Ie cas des enonces mathematiques,
une demonstration). II y a effectivement, de ce point de vue, une
concordance remarquable entre 1a philosophie des mathematiques
et la philosophie du langage de Wittgenstein. Comme Ie soulignent
Baker et Hacker, l'evolution de ses idees apartir des annees trente
a ete determinee largement par la question des differents choix
que ron peut faire, lorsqu'i} s'agit de determiner les propositions
de base de 1a reconstruction semantique et les relations de 'con-
firmation' (plus ou moins strictes) qui rattachent 1a signification
des propositions derivees a celle des propositions primitives.
Le point important est que la reponse a la question 'Qu'est-ce
 Wittgenstein et la philosophie du langage 101

qui compte comme evidence suffisante pour quoi?' doit etre

donnee par la description de ce qui compte effectivement, dans la
pratique de nos jeux de langage, comme evidence suffisante pour
quelque chose, et non de ce qui peut l<5gitimement compter
comme tel. Que la chaine des raisons et des justifications ait une
fin, n'autorise pas a supposer que nos jeux de langage s'appuient
sur un fondement insuffisant, puisque la possibilite me me d'un jeu
de langage depend de ce fait: 'Je veux dire a proprement parler
qu'un jeu de langage n'est possible que si l'on se repose sur
quelque chose. (Je n'ai pas dit "peut se reposer sur quelque
chose".)' ([ 15], § 509). La satisfaction des criteres de l'asserta-
bilite, qui font partie de la grammaire des expressions en cause,
engendre la certitude et rend Ie doute incomprehensible, mais
n'elimine pas necessairement la possibilite de 1'erreur. Le concept
central de critere foumira finalement a Wittgenstein l'argument
decisif contre Ie scepticisme: puisque la probabilite et Ie doute
presupposent I'existence de criteres et la possibilite de la certitude,
Ie scepticisme radical est tout simplement inconsistant.
La theorie semantique du deuxieme Wittgenstein peut encore
etre dite antirealiste en un sens qui se rattache plus directement a
la these (constamment reaffinnee dans les ccrits de cette period e)
de I'autonomie de la grammaire et du langage. Frege avait souligne,
dans sa critique du formalisme mathematique (Frege [10] , II,
§ § 86-137), qu'a la difference des regles du jeu d'echecs, les
regles qui gouvement l'usage des signes arithmetiques ne peuvent
pas etre arbitraires, parce qu'elles decoulent des significations des
signes (et, plus precisement, de leurs denotations: les nombres).
Le choix des regles est determine par la nature et les proprietes
des entites que les signes designent. Wittgenstein a, pour sa part,
recuse catcgoriquement l'existence d'une instance superieure et
anterieure aux regles de langage, a laquelle on pourrait en appeler
pour les justifier, les contester ou les modifier: 'On ne peut pas
penetrer derriere les regles, parce qu'il n'y a pas de derriere'
(Wittgenstein [6], p. 244).
Les regles de la negation, par exemple, ne decrivent pas une
entite qui pn!existe, eUes la constituent. C'est une erreur de
croire que 'de la nature de la negation decoulent les regles con-
cernant Ie signe de negation. De sorte que, en un certain sens, la
negation serait d'abord donnee, et ensuite les regles de la gram-
102 J. Bouveresse

maire' (ibid., p. 53). De fac;:on generale: 'La grammaire n'a a rendre

des comptes a aucune f1!alite. Les regles de grammaire determinent
d'abord la signification (la constituent), et elles ne sont, par conse-
quent, responsables envers aucune signification et dans cette
mesure arbitraires' (p. 184). Alors que Wittgenstein avait conc;u,
a l'epoque du Tractatus, la 10gique du langage comme refletant (de
fac;:on inexprimable) les proprietes essentielles de la realite, i1 lui
a attribue, au contraire, par la suite Ie caractere independant, arbi-
traire et changeant d 'une libre creation.
L'arbitraire des regles de la grammaire n'est cependant ici
qu'une determination purement negative. Wittgenstein veut dire
que, quelles que soient les explications que I'on pourrait eventuel-
Iement donner par ailleurs du fait que les regles grammaticales sont
celles que nous avons plutot que d'autres, on ne peut pas Iesjusti-
fier en disant qu'elles sont fondees dans la nature des choses:
'Nous avons un systeme des couleurs comme nous avons un systeme
des nombres. Les systemes resident-its dans notre nature ou dans la
nature des choses? Comment doit~n dire la chose? - Pas dans la
nature des nombres ou des couleurs' ([7], § 357). Meme s'il etait
vrai que, pour des raisons qui tiennent a notre nature et a
I'histoire, no us ne pouvions pas avoir d 'autres regles et d 'autres
concepts que les notres, it n 'en est pas moins depourvu de sens de
les presenter comme les seuls possibles ou les seuls corrects, si cela
veut dire les seuls qui permettent de caracteriser adequatement la
realite et egalement de les soupc;:onner de partialite ou d'injustice a
l'egard de la realite: 'Car, lorsque je dis que les regles sont arbi-
traires, je veux dire qu'elles ne sont pas determinees par la realite,
comme la description de ceUe n~alite. Et cela signifie: c'est un
non-sens de dire d'elles qu'elles concordent avec la realite; que les
regles concernant les mots "bleu", "rouge", par exemple, concor-
dent avec les faits qui ont trait aces couleurs, etc.'( [6 J , p. 246).
Et si les regles de la grammaire determinent une forme de descrip-
tion, mais ne decrivent rien, cela n'a naturellement pas non plus de
sens d'envisager qu'elles ne concordent pas avec la realite.
Sur ce point, Wittgenstein a, en un certain sens, simplement
inverse la conception qu'il avait exposee dans Ie Tractatus: 'Dans
Ie Tractatus la structure du langage ou de la pensee permettait
d 'apprehender la structure de la realite. Dans les Recherches,
la structure du langage est toujours l'objet de l'examen. En outre,
 Wittgenstein et fa philosophie du langage 103

eUe est toujours isomorphe a la structure de la realite, non pas par-

ce que Ie Iangage doit refleter Ia forme logique de I'univers, mais
parce que ce qui apparait comme la "structure de Ia realite" est
simplement l'ombre de Ia grammaire' (Hacker [8], p. 145).
Compte tenu de ce qui precede, il serait absurde d'interpreter cette
ombre projetee sur Ie monde comme une apparence trompeuse.
L'insatisfaction fondamentale que les philosophes eprouvent
frequemment, de ce point de vue, a regard de la grammaire ne
peut pas etre eliminee par une reforme de nos concepts, mais par
la description de ceux que nous avons. Envisager, comme Ie fait a
l'occasion Wittgenstein, d 'autres possibilites de conceptualisation
n'est encore qu'un moyen de parvenir a ce resultat. C'est ce qui
explique, en particulier, Ie recours a l'ethnologie reelle ou fictive:
'Lorsque nous utilisons la maniere de considerer les choses de
l'ethnologue, cela signifie-t-il que nous donnons a Ia philosophie
Ie statut d'ethnologie? Non, cela veut dire simplement que nous
prenons notre point de vue loin a I'exterieur, pour pouvoir voir les
choses de fat;on plus objective' (Wittgenstein [4], p. 76).
La position de Wittgenstein sur Ie caractere non descriptif et
Ie statut tout a fait special des propositions d'essence est restee,
sur un point important, Ia meme de puis Ie Tractatus jusqu'a la
fin. Dans De la certitude et les Remarques sur les couleurs, il a
souligne beaucoup plus qu'il ne l'avait fait auparavant Ie carac-
tere flou et fluctuant de la distinction entre Ies propositions
grammaticaies ou conceptuelles et les propositions empiriques, Ie
fait que Ie sens de certaines propositions, comme par exemple
celles de la grammaire des couleurs, oscille, pour ainsi dire, entre
I'experience et la logique, et egalement I'existence d'un grand
nombre de propositions empiriques par leur forme qui fonction-
nent neanmoins, dans notre systeme de connaissances, comme des
a priori (relatifs) et des regles ou des normes de description. Mais
it n'a jamais veritablement remis en question la legitimite de la
distinction et son importance pour Ia philosophie, qui, a la
difference de la science, s'occupe uniquement de questions con-
ceptuelles, et non pas empiriques, mais commet regulierement
I'crreur de traiter Ies problemes conceptuels comme s'il s'agissait
de probh!mes d 'experience.
II n'est jamais revenu non plus sur l'idee que les propositions
qui expriment des necessites grammaticales ou conceptuelles n'ont
104 1. Bouveresse

pas reellement de sens, parce qu'elles n'ont pas de negation signifi-

ante. Comme il I'avait ecrit a Russell: ' ... Nous comprenons une
proposition seulement si nous savons d la lois ce qui serait Ie cas
si elle etait fausse et ce qui serait Ie cas si elle etait vraie" ([91,
p. 33). Et il est reste convaincu, meme apres avoir n!pudie l'idee
d 'une semantique construite en termes de conditions de verite
qu'une proposition authentique doit creer une alternative n~elle,
qu'elle ne peut avoir de sens que si sa negation en a une et que,
par Ie fait, seules les propositions contingentes sont reellement
doUt~es de sens. Les propositions grammatica\es, qui sont des
propositions sans antithese, dans la mesure ou leur caractere
grammatical exclut que nous puissions nous representer Ie con-
traire de ce qu'elles enoncent, different beaucoup plus des
propositions dcscriptives ordinaires que ne Ie suggere I'idee
philosophique courantc qu'elles decrivent bel ct bien une
rcalite, mais une n~alite 'ideale'. (Dire que Jes propositions
mathematiques, qui constituent une espece particulierement
importante de propositions grammaticaies, traitent d 'objets
ideaux et decrivent des jaits concernant ce genre d'objets est,
aux yeux de Wittgenstein, une fa<;on de minimiser ou d 'annuler
la difference). Si 'l'essence est exprimee dans la grammaire' ([21,
§ 371), die n'y est en aucune maniere decrite. Et c'est ce qui
fait qu'on ne peut justifier Ies propositions grammaticales en
exhib:1l1t une realite dont elles sont supposees traiter. Du
Tractatus a De fa certitude, Wittgenstein est reste fidele a l'iMe
que les proprietes structurales communes au langage et a Ia
realitl' ne pe.uvent, en toute rigueur. e!re deerites dans des pro-
positions, mais seulcment montrees dans l'usage au la pratique
du langage (ef. par cxemple [15], § 50 I).

Dans les Carnets de 1914 ~ 1916 [161, Wittgenstc in avait eerit:

'La difficulte. avant rna theorie de la reproduction logique, etait
de trouver une connexion entre Ies signes sur Ie papier et un etat
de choses cxtcrieur dam, Ie monde' (27.10.14). Les deux prob-
iemes essentie.ls que la thearie du Tractatus ftait supposee
resouclre sont: I) Comment Ie langage peut-il atteindre ';1 dis-
 Wittgenstein et fa philosophie du iangage 105

tance' la realite? 2) Comment peut-il dire quelque chose sus-

ceptible d'etre vrai ou faux concernant la n~a1ite? Selon Ie Trac-
tatus, Ie contact entre Ie langage et Ie monde est assure par les
coordinations etablies entre les noms et les objets, qui sont com-
parables a des antennes par lesquelles Ie langage 'touche' en
quelque sorte la realite. Et la question de la nature de la verite,
comme relation du signe propositionnel au fait, est reglee par Ia
theorie de l'image, dont Wittgenstein dit qu'elle 'donne en pre-
mier lieu un eclaircissement sur la nature de la relation de verite'
(ibid., 20. 10. 14): la proposition est vraie ou fausse selon que les
objets dans l'etat de choses sont arranges ou non comme Ie sont
leurs noms dans Ie signe propositionnel.
Wittgenstein a completement renonce, au debut des annees
trente, a la conception selon laquelle il existe en quelque sorte
deux types de regles de langage: les unes qui mettent en relation
les signes les uns avec les autres (qui donnent au langage sa struc-
ture), et d'autres qui les mettent en relation avec des objets extra-
linguistiques (les premieres etant, d'une certaine maniere,
imposees et justifiees par Ie choix des secondes). En realite, meme
les definitions ostensives n'etablissent pas entre Ie langage et la
realite une coordination du genre de celle dont il est question dans
Ie Tractatus, elles ne font qu'ajouter des elements au ca1cul ou au
systeme du langage: 'La "liaison du langage avec la realite", par
exemple par l'intermMiaire des definitions ostensives ne rend pas
la grammaire obligatoire (ne justifie pas la grammaire). Car celle-ci
ne reste jamais qu'un calcul qui fIotte librcment dans l'espace, qui
peut etre etendu, mais non soutenu. La "liaison avec la realite" ne
fait qu'etendre Ie langage. mais ne Ie contraint a rien' ([6],
p.313-314).
Le principe selon lequel une proposition douee de sens doit
determiner completement un etat de choses, a l'existence ou la
non-existence pres, entrafnait comme consequence, dans Ie Trac-
tatus, 1a necessite de l'existence d'objets simples, qui peuvent Hre
seulement nommes, et non decrits. En rejetant Ie postulat du
carac1ere determine du sens (qui est un des aspects fondamentaux
de la theorie de l'image), Wittgenstein a rejete egalement les objets
du Tractatus et !'idee meme d'une analyse complete et univoque
de la proposition. Comme il l'a explique lui-meme dans les
Recherches, une theorie du genre de celle du Tractatus impose a
106 1. Bouveresse

priori et dogmatiquement l'existence de noms propres purs, dont

la signification est epuisee entierement par leur relation directe a
I'objet et pour la denotation desquels la question de l'existence ou
de la non-existence ne doit pas se poser. Le nom propre 'authen-
tique' doit designer quelque chose de simple et d 'indestructible;
car 'on doit pouvoir decrire l'etat dans lequel tout ce qui est
destructible est detruit. Et dans cette description il y aura des
mots; et ce qui leur correspond ne peut pas alors etre detruit, car
sans ccIa les mots n 'auraien t pas de signification' (§ 55).
Sur ce point-If! egalement, scIon une de ses propres metaphores,
Wittgenstein a effectue une rotation de cent quatre-vingts degres
par rapport a sa conception initiale: ce qui doit exister pour que
des propositions doUt~es de sens soient possibles, ce ne sont pas des
elements indestructibles de la realite, representables par des signes
simples, mais des elements de representation, qui sont des instru-
ments linguistiques, par exemple, dans Ie cas des couleurs, des
modeles qui sont utilises en liaison avec les noms (et qui font
partie, en un certain sens, du 'dictionnaire' lui-meme). De fayon
generale, 'ce que, apparemment, il doit y avoir fait partie du
langage' (§ 50). Et c'est l'intemporalite des paradigmes et des
n':gles qui rend compte de la pretendue 'etemite' des objets ou des
essences. Tous les problemes fondamentaux du Tractatus, celui du
contact entre Ie langage (ou la pensee) et la realite, celui de la con-
cordance entre la structure du langage et ceUe de la realitc, et
meme celui de la nature de la verite, doivent pouvoir etre resolus
en restan t it l'in terieur de la grammaire: 'C'est dans Ie langage
que tout se regIe' ([6J, p. 143).
Dans Ie Tractatus, la signification du signe propositionnel est
presentee comme unc relation de type projectif avec Ie fait decrit.
Les rayons de projection qui rattachent l'image propositionnelle a
la realite sont supposes annuler la difference qui pourrait subsister
entre les deux, sans Jaisser subsister aucune place pour un usage au
un mode d'emploi. Lc signe propositionnel accompagne de la
methode de projection contient deja, d'une certaine manierc,
entierement Ie fait, a i'existence pres. Mais, en realite, meme Ie
dispositif constitue par I'image ct un schema du mode de pro-
jection peut encore etre utilise de differentes fa<;ons, aussi 'cor-
rectes' les unes que les autres (cf. [2J, § 141). Le pont que la
methode de projection est censee jeter entre Ie 1angage et la
 Wittgenstein et la philosophie du langage 107

realite n'existe pas veritablement, tant que l'usage n'est pas etabli;
et seuls certains cas particuliers de ce que nous appelons, de fa~on
generale, l'usage ou l'application des signes peuvent etre decrits de
fac;on plausible et eclairante en termes de projection et de con-
cordance, au sens d'une communaute de forme ou de structure.
Selon Ie Tractatus, Ie signe propositionnel doue de sens (pris
avec la methode de projection) est Ie signe propositionnel pense.
L'intervention de la pensee, qui est pourtant eUe-meme une
image logique des faits, au meme titre que Ia proposition,
supprime la distance apparemment infranchissable qui existe
entre toute image et ce qu'elle represente en dehors d'elle-meme.
Pourtant, aussi longtemps qu'on se represente Ia pensee,
l'intention, etc., sous la forme d'une image, meme parfaitement
ressemblante, Ie probleme reste entier. Lorsque quelqu'un pense,
'les objets sur lesquels il pense ne sont assurement pas du tout
dans sa tete (aussi peu que dans sa pensee)!' ([6], p. 143). La
question cruciale de savoir comment la pensee, Ie vouloir-dire,
l'imagination, I'attente, Ie desir, etc., peuvent reellement anticiper
sur un fait determine, qui y est deja, d'une maniere ou d'une
autre, present sous une forme affaiblie, mais parfaitement uni-
voque - celie d'une sorte de double ethere qui s'effacera, Ie
moment venu, devant Ia realite tangible - est simplement
rep osee a un au tre niveau.
On pourrait reformuler ainsi Ie probleme philosophique fonda-
mental qui a domine, d'un bout a l'autre, la problematique du
signe chez Wittgenstein: qu'est-ce qui fait vivre Ie signe materiel,
qui est apparemment une chose inerte, et lui permet de dire
queIque chose? La reponse classique est que 'l'action du langage
consiste en deux parties; une partie inorganique, la manipulation
de signes, et une partie organique, que nous pouvons appeler
comprendre ces signes, leur donner un sens, les interpreter, penser'
(The Blue Boo k [17], p. 3). Mais 'tou tes les fois que nous in ter-
pretons un symbole d 'une maniere ou d 'une autre, l'interpretation
est un nouveau symbole ajoute a l'ancien' (ibid., p. 33). Le sym-
bole mental n'est en lui-meme pas plus proche et pas plus eloigne
de l'objet qu'il vise que Ie signe materiel. Et, de ce point de vue,
nous ne perdons rien a remplacer systematiquement Ie premier
par Ie deuxieme. (Comme dit Wittgenstein, ce dont nous avions
besoin est precisement Ie caractere occulte de I'objet ou du
108 1. Bouveresse

processus immateriel.) C'est ce qui explique ce qu'il a appele Ie

'behaviorisme' de sa conception et de sa demarche, son ignorance
deliberee de la distinction entre I"interieur' et I"exterieur' et son
application consequente du principe: 'Si vous avez des perplexites
concernant la nature de la pensee, de la croyance, de la connais-
sance, et d'autres choses de ce genre, substituez it la pensee
l'expression de la pensee, etc'. (ibid., p. 42). La conviction que la
philosophie, meme lorsqu'elle a affaire aux concepts psychol-
ogiques les plus typiques, n'est pas plus liee a la psychologie qu'a
n'importe queUe autre science empirique est encore un point sur
lequel il n'a jamais change d'opinion.
La plupart de nos difficultes et de nos erreurs se rattachent ici
au fait que 'nous cherchons l'usage d'un signe, mais nous Ie
cherchons comme s'il ctait un objet coexistant avec Ie signe'
(ibid., p. 5). Cette fac;on de voir est particulierement tent ante dans
Ie cas des expressions qui renvoient apparemment a des objets et
des processus internes 'prives', ou l'usage semble etre determine et
justifie uniquement par la presence immediate de quelque chose
que nous sommes les seuls a posseder et, en fin de compte, a
connaftre. L'argumentation contre la possibilite d'un langage
prive, dans les Recherches philosophiques, vise a etablir que, meme
dans ce cas, Ie jeu de langage (public) que nous jouons avec les
expressions concernees reste fondamental et premier. Les expres-
sions verbales immCdiates de l'experience interne ne sont pas
element aires, du point de vue de la theorie de la signification: elles
ont tout au plus une signification qui est derivee et dependante de
celle des propositions intersubjectivement controlables que nous
formulons sur Ie monde exterieur. Et meme dans Ie cas d 'expres-
sions comme 'avoir mal', 'penser' ou 'comprendre', I'aspect
determinant de la signification n'est pas constitue par la presence
de quelque chose que nous eprouvons ou que nous avons it un
moment donne et qui nous fait dire ce que nous disons, mais par
les criteres publics de l'usage. L'idee d'un usage prive ou d'une
justification privee de l'usage est pratiquement une contradiction
dans les termes. Cela entraine Ie rejet de l'intuition comme
pseudo-justification et 'faux-fuyant inutile' ([2], § 213). de
I'introspection comme voie royale d'acces au vouloir-dire et a la
signification et de toutes les 'absurdites' dues au fait 'que I'on
fixe son regard sur une forme verbale et oublie de se demander
 Wittgenstein et fa philosophie du fangage 109

ce qu'on fait avec elle, ou qu'on regarde fixement dans son arne
a soi pour voir si deux expressions ont la meme signification, et
d 'autres choses de ce genre' (Diamond [18] , p. 111).
Une des attaques les plus serieuses contre la philosophie du lan-
gage de Wittgenstein et sa conception de la philosophie tout court
est celle qui est provenue recemment de ce qu'on appelle quelque-
fois 'la nouvelle theorie de la reference', qui lui reproche d'etre
reste, en depit de ses reticences bien connues, foncic:hement
tributaire de la notion traditionnelle de signification. Si les repre-
sentants de cette theorie ont raison, il peut, d'une certaine maniere,
y avoir une interrogation sur ce qu'un mot signifie 'reellement' et
une question qui a trait a la signification exacte du mot 'or', par
exemple, n'est pas fondamentalement differente d'une question
scientifique sur la nature veritable de la substance en cause. Cela
oblige a reconsiderer Ie role essentiel que I'on a pris I'habitude
d'attribuer a l'analyse conceptuelle pour la solution des problemes
philosophiques et I'idee que l'essence doit etre cherchee dans des
conventions grammaticales, plutot que decouverte au fur et a
mesure que nous reconnaissons les choses pour ce qu'eUes sont
reellement. Comme I'a montre la polemique entre Putnam et
Malcolm sur la nature du reve, meme la pertinence et la superiorite
apparentes de l'analyse criteriologique dans Ie cas des termes
designant des experiences ou des processus mentaux est contestee
par la nouvelle theorie. Et ce que Putnam appelle 'la division du
travail linguistiquc' implique que la signification, si elle n'a effec-
tivement rien d'un element occulte plus ou moins prive, n'est
cependant pas non plus accessible par la simple observation de
l'usage public, dans la mesure OU les criteres d'identification
detenus par les 10cuteurs 'experts' se trouvent etre, a travers eux,
la propriete collective de la communaute linguistique, d'une
maniere telle que 'Ie fait Ie plus esoterique concernant l'eau peut
devenir une partie de la signification sociale du mot, bien qu'il
soit inconnu de presque tous les locuteurs qui acquierent Ie mot'
(,Meaning and Reference', in S.P. Schwartz (ed.), [19], p. 126).
Si cette delegation de propriete constitue reellement un aspect
essentiel du phenomene de la signification, il est difficile de
maintenir que la philosophie ne s'interesse pas a ce qui est cache
et propose quelque chose comme 'une vue "synoptique" de choses
que nous savons tous'.
110 J. Bouveresse

II n 'est pas surprenant que tou tes les objections de principe

contre la methode philosophique de Wittgenstein toument finale-
ment plus ou moins autour de la question de savoir s'il existe ou
non une frontiere nette entre les questions scientifiques et les
questions philosophiques. 11 est difficile de dire si la position de ses
adversaires sur ce point est determinee plutot par une insensibilite
relative a l'egard d'une certaine categorie de problemes ou par une
decision theorique plus ou moins consciente sur leur nature
'reelle', qui ignore deliben!ment les apparences (la maniere tout a
fait particuliere dont ils sont ressentis et l'impression qu'ils
a
donnent de resister toute espece d'explication). II est clair que,
pour Wittgenstein, la prise de conscience de I'heterogeneite totale
de Ia philosophie par rapport aux sciences est elle-meme un des
buts principaux de la recherche philosophique et la mecon-
naissance de Ia distinction une des sources principales des
difficultes du philosophe.

BIBLIOGRAPHIE

l. Ouvrages de Wittgenstein

Wittgenstein, L. [2j Philosophische Untersuchungen (Philosophical Investi-

gations). Eds. G.E.M. Anscombe and R. Rhees. Oxford: Blackwell,
1953.
- [3] Bermerkungen liber Frazers 'The Golden Bough '. Ed. R. Rhees. Syn-
these 17 (I 967).
- [4] Vermischte Bemerkungen, eine Auswahl aus dem Nachlass, heraus-
gegeben von G.H. von Wright, unter Mitarbeit von Heikki Nyman.
Frankfurt: Suhrkamp Verlag, 1977.
- [6] Philosophische Grammatik. Ed. R. Rhees. Oxford: Blackwell, 1969.
- [7] Zettel. Eds. G.E.M. Anscombe and G.H. von Wright. Oxford: Black-
well,1967.
[9] Letters to Russell, Keynes and Moore. Ed. G.H. von Wright. Oxford:
Blackwell, 1974.
- [11] Ludwig Wittgenstejn und der Wiener Kreis, shorthand notes recorded
by F. Waismann. Ed. B.F. McGuinness. Oxford: Blackwell, 1967.
- [14] Tractatus Logico-Philosophicus, German text with a new edition of
the translation by D.F. Pears and B.F. McGuinness. London: Routledge
and Kegan Paul, 1971.
- [15] Uber Gewissheit (On Certainty). Eds. G.E.M. Anscombe and G.H. von
Wright. Oxford: Blackwell, 1969.
 Wittgenstein et fa philosophie du fangage III

- [16] Notebooks 1914-1916. Eds. G.H. von Wright and G.E.M. Anscombe.
Oxford: Blackwell, 1961.
- [17] The Blue and Brown Books. Oxford: Blackwell, 1958.
- [20] Worterbuch fur Vo lksschu len. Wien: Holder-Pichler-Tempsky, 1977.
- [21] Some Remarks on Logical Form. Proceedings of the Aristotelian
Society, Supp. Vol. 9 (1929); reproduit dans Essays on Wittgenstein's
Tractatus. Eds. I.M. Copi and R.W. Beard. London: Routledge and
Kegan Paul, 1966.
- [22J Philosophische Bemerkungen. Ed. R. Rhees. Oxford: Blackwell,
1964.
- 123] Bemerkungen iiber die Farben (Remarks on Colour). Ed. G.E.M.
Anscombe. Oxford: Blackwell, 1977.

II. A utres ouvrages

Ambrose, A., and Lazerowitz, M., Eds. [24J Ludwig Wittgenstein: Philos-
ophy and Language. London: Allen and Unwin, 1972.
Anscombe, G.E.M. [25J An Introduction to Wittgenstein's Tractatus, 2nd
edn. London: Hutchinson, 1963.
Baker and Hacker [I3J Critical Notice on 'Philosophical Grammar'. Mind
85 (1976).
Black, M. [26] A Companion to Wittgenstein's Tractatus. Cambridge:
Cambridge University Press, 1964.
Bogen, J. [27J Wittgenstein's Philosophy of Language. London: Routledge
and Kegan Paul, 1972.
Diamond, C., Ed. [18] Wittgenstein's Lectures on the Foundations of Mathe-
matics. The Harvester Press, 1976.
Dietrich, R.A., [28] Sprache und Wirklichkeit in Wittgensteins Tractatus.
Tiibingen: Max Niemeyer Verlag, 1973.
Dummett, M. [29] Frege: Philosophy of Language. London: Duckworth,
1973.
Durfee, H.A., Ed. [30] Analytic Philosophy and Phenomenology. The Hague:
M. Nijhoff, 1976.
Fogelin, RJ. [31] Wittgenstein. London/Henley/Boston: Routledge and
Kegan Paul, 1976.
Frege, G. [5] Begriffsschrift. 1879.
- [10J GrundgesetzederArithmetik, I. 1893.
Griffin,1. [32] Wittgensfein's Logical Atomism. London: Oxford University
Press, 1964.
Hacker, P.M.S. [8] Insight and Illusion: Wittgenstein on Philosophy and the
Metaphysics of Experience. Oxford: The Clarendon Press, 1972.
Hallett, G. [33] Wittgenstein's Definition of Meaning as Use. New York:
Fordham University Press, 1967.
- [34] A Companion to Wittgenstein s Philosophical Investigations. Ithaca/
London: Cornell University Press, 1977.
1 12 J. Bouveresse

Hintikka, J., Ed. [35] Essays on Wittgenstein in Honour of GR. l'on Wright.
Amsterdam: North-Holland, 1976.
Hunter, J.F.M. [36] Essays ajter Wittgenstein. London: Allen and Unwin,
1973.
Kenny, A. [37] Wittgenstein. London: Allen Lane, 1973.
Klemke, E.D., Ed. [38] Essays on Wittgenstein. Urbana/Chicago/London:
University of lllinois Press, 1971.
Kreisel, G. [12] Wittgenstein's Theory and Practice of Philosophy. British
Journal for the Philosophy of Science II (I 960).
Lang, M. [39 J Wittgensteins philosophische Grammatik. Den Haag: M.
Nijhoff, 1971.
Leinfeller, E., Leinfeller, W., Berghel, H., Huebner, A., Eds. [39J Wittgen-
stein und sein Einfluss aUf die gegenwti'rtige Philosophie. Akten des
zwciten internationalen Wittgenstein Symposiums, Kirchberg aW. Wien:
Holder-Pichler-Tempsky, 1978.
Pitcher, G. [41] The Philosophy of Wittgenstein. Englewood Cliffs, N.J.:
Prentice-Hall, 1964.
-- Ed. [46] Wittgenstein: The 'Philosophical Investigations'. A Collection of
Critical Essays. New York: Anchor Books, 1966.
Rhees, R. [42 J Discussions of Wittgenstein. London: Routledge and Kegan
Paul, 1970.
Richardson, J.T.E. [1] The Grammar of Justification, An Interpretation of
Wittgenstein's Philosophy of Language. London: Sussex University
Press, 1976.
Schwartz, S.P., Ed. [19] Naming, Necessity and Natural Kinds. Ithaca, N.Y.:
Cornell University Press, 1977.
Setler, G.F. [43] Language and the World, A Methodological Synthesis with-
in the Writings of Martin Heidegger and Ludwig Wittgenstein. Atlantic
Highlands, N.J.: Humanities Press, 1974.
Stenius, E. [44] Wittgenstein's Tractatus, A Critical Exposition of the Main
Lines of Thought. Oxford: Blackwell, 1964.
Winch, P., Ed. [45] Studies in the Philosophy of Wittgenstein. London:
Routledge and Kegan Paul, 1969.
Zimmermann, J. Wittgensteins sprachphilosophische Hermeneutik. Frankfurt:
Vittorio Klostermann, 1975.
Richard Montague and the logical analysis of
language *
NINO COCCHIARELLA
Indiana University

Richard Montague was an exceptionally gifted logician who made

important contributions in every field of inquiry upon which he
wrote. His professional career was not only marked with brilliance
and insight but it has become a classic example of the changing
and developing philosophical views of logicians in general,
especially during the 1960s and 70s, in regard to the form and
content of natural language. We shall, in what follows, attempt to
characterize the general pattern of that development, at least to
the extent that it is exemplified in the articles Montague wrote
during the period in question.
The articles to which we shall especially direct our attention
are: 'Pragmatics' [1]; 'Pragmatics and Intensional Logic' ([ 2 J ;
'On the Nature of Certain Philosophical Entities' [3]; 'English as
a Formal Language' [4]; 'Universal Grammar' [5J; and 'The
Proper Treatment of Quantification in Ordinary English' [7].
Needless to say, but many of the ideas and insights developed in
these papers Montague shared with other philosophers and
logicians, some of whom were his own students at the times in
question. Montague himself is meticulous in crediting others where
credit is due, but for convenience we shall avoid duplicating such
references here.

*The author is indebted to the National Endowment for the Humanities for
support during the research and writing of this paper.

Contemporary philosophy. A new survey. Vol. 1, pp. 113-154.

©1981, Martinus Nijho!! Publishers. The Hague/Boston/London.
114 N. Cocchiarella

1. CONSTRUCTED VS. NATURAL LANGUAGES

In his earlier pre-1964 phase of philosophical development,

Montague thought of a formalization of English or any other
natural language as either impossible or extremely laborious and
not rewarding ([ 10], p. 10). There was in his view an important
difference between constructing a formal language for the purpose
of analyzing concepts that were of interest to philosophers and
investigating the behavior of the expressions in natural language
that were ordinarily used to express those concepts. He held in low
esteem the general methodology of so-called ordinary language
philosophers which was very popular during the 1950s and early
60s and thought that there was little or no philosophic interest in
attempting to analyze ordinary language itself.
At the height of his later post -1964 phase of development, how-
ever, Montague had come to completely reject 'the contention
that an important theoretical difference exists between formal and
natural languages' ([ 4], p. 188), i.e., 'between natural language
and the artificial languages of logicians' ([5], p. 222); and he set
about constructing a formal language which 'may be reasonably
regarded as a fragment of ordinary English' ([4] , p. 188). He still
rejected the methodology of ordinary language philosophy but
had now come to believe that indeed there is philosophic interest
in attempting to analyze ordinary language after all ([3], p. 186).
Nevertheless, despite this important change in view and attitude
regarding natural language, Montague continued to maintain that
ordinary language is 'an inadequate vehicle for philosophy' ([ 3],
p. 186). This is partly because ordinary language is rich in am-
biguities and needs to be disambiguated and partly because the
semantical relation of logical consequence and logical truth
cannot be characterized except with respect to a logically per-
spicuous language. However, it is also, and perhaps more especially,
because of the dubious ontological nature and status of the
entities purportedly involved in the use of ordinary language, at
least when compared to th~ entities to which they are reduced in
the logically perspicuous language Montague constructs for the
semantics of ordinary language. Indeed, here in the ontology of
that logically perspicuous language we find another, and perhaps
more fundamental, change or development in Montague's phil-
osophy of language.
 Richard Montague and the logical analysis of language 115

2. SET THEORY VS. INTENSIONAL LOGIC

In his earlier phase, Montague had maintained that philosophy

had 'as its proper theoretical framework set theory with indi-
viduals and the possible addition of empirical predicates' ([ 3 ] ,
p. 154). Philosophical analyses, on this view, were to be carried
out within definitional extensions of a set theory so supplemented;
and, in consequence, nonextensional entities such as individual
concepts, properties, relations-in-intension and propositions were
not allowed to occur in such analyses. Intensional contexts and so-
called propositional attitudes, such as knowledge and belief, were
to be analyzed instead in terms of relations between persons and
sentences or other similar expressions. Montague's own con-
tributions to such analyses (and reprinted in Formal Philosophy)
can be found in [8], [9], [7]. This last paper is especially impor-
tant for its extension of Tarski's theorem to necessity, or
necessary truth, as a property of sentences.
In his later phase, however, Montague developed a new
theoretical framework which transcends set theory and in which
individual concepts, properties, relations-in-intension, and prop-
ositions are all allowed to occur in philosophical analyses and are
found to be 'not much more mysterious than sets' ([ 3 ] , p. 152).
'Philosophy', Montague writes at the mid-point of this later phase,
'is always capable of enlarging itself; that is, by metamathematical
or model-theoretic means - means available within set theory -
one can "justify" a language or theory that transcends set theory,
and then proceed to transact a new branch of philosophy within
the new language' (ibid., p. 155). This new theoretical framework
is intensional logic.
There are two stages and two systems involved in Montague's
development of his new theoretical framework for philosophy.
The first is a second-order modal logic (which is readily extendible
to third and higher-order modal logic), and the second is an
extension of Alonzo Church's formulation of type theory in which
primitive operators for the extension-intension, or sense-
denotation, distinction have been introduced. The first, for the
most part, allows only for intuitive translations into logical
notation (as typified by the symbolizations found in elementary
logic texts) and usually requires the use of circumlocution or para-
116 N. Cocchiarella

phrase for the representation of certain intensional verbs such as

'seek', 'conceive', 'owe', etc. The second, on the other hand,
allows for a precise translation of ordinary language and for a
direct representation of intensional verbs. (ef. Gallin [11] for a
discription of both systems, together with other related systems,
and a number of proofs of results regarding all of these systems
and their interrelationships.) Before turning to these systems, how-
ever, we shall first take up Montague's development of pragmatics
which involves only first-order languages and which in fact rather
appropriately characterizes the initial phase in the development of
Montague's new orientation to the form and content of natural
language.

3. PRAGMATICS AND LOGICAL SEMIOTICS

Prior to the 1960s most formal language philosophers deliberately

avoided the whole question of how the different contexts in which
language is used affects the meaning and reference of expressions.
Logicians, with their proper concern for rigor and formal pre-
cision, tended to see the logical analysis of those features of
ordinary language that were dependent on context as either im-
possible or hopelessly complex. They preferred instead to abstract
from particular contexts of use and to provide analyses of sen-
tences and arguments of ordinary language within a timeless and
context-free setting.
Logical semiotics at that time, as opposed to descriptive semi-
otics, was in fact by definition restricted to logical (or pure)
syntax (concatenation theory) and logical (or pure) semantics
(model-theory as a branch of set-theory or as a form of higher-
order logic), where the latter amounted in effect only to a theory
of reference. Pragmatics, the third branch of the semiotical trinity,
was not only excluded from logical or pure semiotics on this
scheme, but was in fact considered to be the very basis of
descriptive semiotics (CL Camap [12]). Thus Rudolf Camap, for
example, considered pragmatics to be an empirical theory which
was to be formulated in the metalanguage of a collection of object
languages and which was in that regard to be concerned with the
relationship between the users and the signs of the object lan-
 Richard Montague and the logical analysis of language 117

guages in question. The expressions of a pragmatic theory, on this

view, including epistemic expressions for belief, doubt, etc., were
considered to be predicates belonging to the metalanguage and
were understood as such as designating psychological relations
between language users and signs of the object languages. Indeed,
Carnap clearly states even as late as 1963 that the relation of
designation in pragmatics 'is a psychological concept' (cf. Carnap
[ 13]).
Montague's development of pragmatics was not so much a
rejection of this tradition as an alternative development within
logical semiotics. What he proposed was an extension of the
semantical notion of truth (in a model or interpretation) as
defined for standard first-{)rder languages but now supplemented
with pragmatic operators, such as operators for the past and future
tenses, as well as other indexical expressions for demonstratives
and personal pronouns. Pragmatics, according to Montague, is to
follow the lead of semantics and 'concern itself also with the
notion of truth - but with respect not only to an interpretation
bilt also a context of use' ([ 1], p. 96). Thus instead of dealing
with descriptive pragmatic predicates of an empirical meta-
linguistic theory, Montague formulates a logical theory of first-
order languages supplemented with pragmatic operators and
including selected individual constants for demonstratives and
personal pronouns.

4. PRAGMATIC LANGUAGES

On Montague's approach then, a pragmatic language is a first-

order language the basic symbols of which are drawn from the
following categories: (I) logical constants, e.g., ...... , II, v, -+, -e-,
1\, V and = (with their usual readings); (2) auxiliary symbols
such as parentheses, brackets and commas; (3) individual variables;
(4) individual constants; (5) n-place predicate constants (for nEw),
including the one-place predicate constant 'E' (for 'exists'); and
(6) n-place operators, for nEw - {O}. The terms and formulas of a
pragmatic language L are built up in the usual way except that
now whenever M is an n-place operator of L and !PI' .. , , !Pn are
formulas of L, then IM'PI ... 'Pnl is also a formula of L. Thus,
I 18 N. Cocchiarella

e.g., where .~ and .F are one-place operators and I{J is a formula,

then 1.-1J~ and I.F~ are also formulas and may be read as 'It was
the case that iJ1 and lit will be the case that I{JI, respectively.
In interpreting a pragmatic language, it is of course necessary
that we take into account the possible contexts of use for that
language. We need not consider these contexts in their full com-
plexity, on the other hand, but need only consider those aspects
of such a context that are relevant to the language in question.
Montague calls such a complex of relevant aspects an index or,
following Dana Scott, a point of reference. Thus for the tense
operators each index will specify a moment of time, and for
personal pronouns and demonstratives it will also specify a
speaker, a listener, objects indicated, etc.
Besides the indices or points of reference, an interpretation for
a pragmatic language L will assign intensions (or meanings) to each
of the operators and individual and predicate constants in L. The
idea here is that an intension determines an extension of the
relevant type in each context of use, i.e., at each index of a
possible pragmatic interpretation. Intensions, or rather their set-
theoretic representations, accordingly, can be identified with
certain functions on the set of indices. Thus, e.g., the intension of
an individual constant will be a function which assigns an indi-
vidual to each index; and the intension of an n-place predicate
constant will be a function which assigns a set of n-tuples of indi-
viduals, i.e., an n-ary relation (in extension) between individuals,
to each index. Montague calls the first type of intension an indi-
vidual concept and the second an n-ary relation-in-intension
(between individuals). The intension of a one-place predicate
constant he calls a property (of individuals). Rudolf Carnap, it
should be noted, was the originator of this general type of
approach to the set-theoretic representation of intensions. Carnap,
however, as Montague notes, inappropriately restricted his indices
to models or state-descriptions.
Where Montague goes beyond Carnap, aside from his dis-
associating indices from models or state-descriptions, is in his
unified treatment of pragmatic operators - and in that regard in
his development of a general pragmatic logic. In order to under-
stand this treatment, let us note first that as an intension a
proposition determines a truth-value in each context of use, and
 Richard Montague and the logical analysis of language 119

therefore at each index of a pragmatic interpretation. In this

regard, the same proposition can be set-theoretically represented
either by the set of indices (of the interpretation) at which it is
true or by the function which assigns to each index the truth-value
the proposition determines at that index. Relative to the former
representation, the functions on indices that Montague associates
with n-place operators are those that assign to each index an n-ary
relation (in extension) between the propositions of the interpre-
tation, i.e., those that assign to each index an n-ary relation on the
set of all subsets of the set of indices of the interpretation. In
other words, since the different subsets of indices are here taken as
set-theoretic representatives of propositions, pragmatic operators
are being interpreted by Montague in effect as higher-order prop-
erties and relations-in-intension of propositions.
Formally stated, then, a possible interpretation for a pragmatic
language L is an ordered triple 0, U, F) such that (1) I and U are
the sets of indices and individuals, respectively, of the interpre-
tation; (2) F is a function defined on the set of operators and indi-
vidual and predicate constants in L; (3) for each individual con-
stant c in L, F( c) E VI; (4) for each n-plane predicate constant P
in L, F(P) E {A: A ~ un} I; and (5) for each n-place operator M in
L, F(M) is a function on I such that for each iEI, F(M) (i) ~
{J: J ~ I}n.
Where JJI is a possible interpretation for a pragmatic language L,
the fact that each nonlogical constant in L is assigned an intension
in JJI allows us to follow the recursion clauses in the definition of
the terms and formulas of L in such a way as to assign a uniq ue
intension of the appropriate type to each term and formula of L.
This then leads in a natural and obvious way to a definition of
truth for the sentences of L at an index of sf, and thereby to
definitions of logical truth and logical consequence in the sense of
general pragmatics. (We shall forgo the details of these definitions
here, however.)
Now it should be noted that particular pragmatic languages,
such as those for modal, deontic and tense logic, had of course
been studied before, both by Montague and his students as well as
by other logicians. But in each case, as Montague observes, 'truth
and satisfaction had to be defined anew,' since no 'unified treat-
ment of operators' had yet been realized ([ 2], p. 120). Indeed, it
120 N. Cocchiarella

is in fact Montague's unified treatment of the semantics of

operators which is his unique contribution to the development of
pragmatics within logical semiotics. In particular, it is only with
respect to such a unified treatment that the notions of logical
truth and logical consequence in the sense of general pragmatics
can be characterized at all.
Notwithstanding this development of pragmatics within logical
semiotics, however, philosophers are typically not interested in
general pragmatic logic but in certain specializations or disciplines,
such as the tense, modal and denotic logics already mentioned.
Such specializations or disciplines, relative to general pragmatic
logic, amount to restricting the possible interpretations to certain
classes of what might be considered the standard interpretations
for the discipline in question. Thus where K is such a class of
possible interpretations, we can restrict logical truth and logical
consequence in the sense of general pragmatic logic to the
interpretations in K. giving us instead, e.g., K-validity and K-con-
sequence. Sometimes it will even be necessary to restrict the indi-
ces as well, designating some of them as the standard points of
reference; and this will lead to further restrictions on validity and
consequence, all serving the particular discipline in question (cf.
[1] , pp. 104 f.).

5. INTENSIONAL LOGIC AS HIGHER-ORDER MODAL LOGIC

It is noteworthy that although the nonlogical constants of a prag-

matic language are all assigned intensions in a possible interpre-
tation for that language, the only entities quantificationally
referred to under such an interpretation are the individuals of the
interpretation. Pragmatic languages are after all first-order lan-
guages, and in that regard, what is indicated by the fact that prag-
matic operators can be interpreted as higher-order properties and
relations-in-intension of propositions is that pragmatic languages
amount in effect to first-order reductions of (a part of) higher-
order intensional languages, i.e., higher-order languages where the
properties and relations-in-intension between propositions that
operators in pragmatic languages represent are directly repre-
sented instead by higher-order predicate constants.
 Richard Montague and the logical analysis of language 121

Alternatively, what this also indicates is that pragmatics, as a

development within logical semiotics, might more appropriately be
viewed as contained in intensional logic where quantificational
reference to the intensional entities represented in pragmatic
interpretations is a part of the object language itself. Indeed,
according to Montague, the fact that the modalities that have
usually been associated with operators coincide in a sense with
properties and relations-in-intension of propositions 'is what
lends interest to those modalities and provides intuitive sanction
for using them to interpret operators' ([2], p. 137). And if that
is the case, then we might just as well develop an intensional logic
and proceed to do philosophy in that framework instead. This
in fact is what Montague had gone on to do next in his developing
new orientation to the philosophy of language.
Montague's first formulation of an intensional logic is really a
second-order modal logic supplemented with third-order predicate
constants. An intensional language on this formulation is one the
basic symbols of which are (I) the logical constants of pragmatic
languages; (2) parentheses and brackets; (3) individual variables
and constants; (4) n-place predicate variables, for nEw; (5) the
one-place operator 0 (read 'necessarily'); (6) the descriptive sym-
bol T (read 'the unique ... such that'); and predicate constants of
type s, for each finite sequence s of integers ~ -1.
The type of a predicate constant, as described above, indicates
the grammatical types of the argument-expressions it takes, with
-I being the type of an individual variable or constant and non-
negative integer n being the type of an n-place predicate variable.
We identify propositional variables and propositional constants
here with O-place predicate variables and predicate constants of
type s, where s is the null sequence.
The atomic formulas of an intensional language L are expressions
either of the form rr = 11--', where r, 11 are individual variables or
constants; rp = (jl, where P, Q are n-place predicate variables, for
nEw; 'p(rl , ... , rn)l, where P is an n-place predicate variable and
each ri is either an individual variable or an individual constant;
or 'R(~}, ... , ~k)l, where R is a predicate constant of type ($1' ... , Sk)
and either Sj = -1 and ~j is an individual variable or constant or
Sj ~ 0 and ~j is an sj-place predicate variable.

Given the atomic formulas, the formulas of L are then built up

122 N. Cocchiarella

in the obvious way. The descriptive symbol T, however, is allowed

to apply only to predicate variables, so that where P is an n-place
predicate variable and I.{) is a formula, "rP~ is a complex n-place
predicate expression. As used by Montague, these descriptive
phrases are all eliminable and therefore, strictly speaking, the
descriptive symbol T need not occur at all in the formulas of
intensional logic. Its utility is really one of convenience for allow-
ing certain more direct ways for representing complexly described
intensions.
Since the set-theoretic counterparts or representatives of
intensions are always functions on some set I of indices and with
values that are either drawn or constructed from a set U of indi-
viduals, it is more accurate in some contexts if we speak of these
counterparts as (I, U}-intensions. Thus an (I, U>-individual con-
cept is a function on I with values in U; and an n-ary 0, U)-
relation-in-intension is a function on I with n-ary relations (in
extension) on U, i.e., sets of n-tuples drawn from U, as values. An
(I, U>-property is a one-ary (I, U>-relation-in-intension.
An <I, U)-proposition, as already noted, can be identified either
with a subset of I, viz., the set of indices at which the proposition
is true, or with a function that assigns a truth-value to each index.
If we use I to represent the true and a to represent the false, then
an (I, U)-proposition, as a function that assigns a truth-value to
each index, amounts to the characteristic function of the set of
indices at which the proposition is true, i.e., it amounts to that
function on I with values in {a, I} such that I is assigned to all
and only those indices in the set in question. This latter identi-
fication is more convenient in our present context since we can
also identify a with both the null sequence and the empty set and
take I = {a}. Thus, since propositional variables are construed
here as a-place predicate variables, an (I, U )..proposition which is
the value of such a variable will be an (I, U>-intension which has
a or 1 as its value (extension) at each index. An <I, U)..proposition,
in other words, is a a-ary <I, U >-relation-in-intension.
A possible interpretation for an intensional language L, then, is
a triple (I, U, F> such that (1) J and U are the sets of indices and
individuals, respectively, of the interpretation; (2) F is a function
defined on the nonlogical constants of L, i.e., the single operator
o and the individual and predicate constants in L; (3) for each
 Richard Montague and the logical analysis of language 123

individual constant c in L, F(c) E VI, i.e., F(c) is an 0, V}-indi-

vid ual concept; (4) F(D) = {O, (I»: iEI}, i.e. F(D) is that property
which propositions (as subsets of I) have when and only when
they are true at every index l ; and (5) for each predicate constant
P of type (Sl, ... , Sk) in L, F(P) is a function on I such that for
JEI, F(P)(j) is a set of n-tuples (V l ' ... , Vk ) such that for all i, if
I ~ i ~ k, than either Sj = -1 and Vj E U or Sj ~ 0 and Vj is an
Sj-ary <I, V>-relation-in-intension.
As with pragmatic languages, we are able again to follow the
recursion clauses for the construction of an arbitrary formula of
an intensional language L and define, relative to a possible i~ter­
pretation d for L, the intension in d of each formula of L; and,
as before, this definition also leads in a natural way to a defi-
nition of truth (for the sentences of L) at an index of d, and
thereby to definitions of logical truth and logical consequence (in
the sense of intensional logic).
With the syntax and semantics of intensional languages charac-
terized as above, the exact sense in which pragmatics is contained
in intensional logic can now be explained as follows. For each
pragmatic language L, there is (easily constructible from L) an
intensional language L' containing (1) all the individual con-
stants of L; (2) a predicate constant P' of type {Sl' ... , sn}, where
Sj = -1, for i ~ n, for each n-place predicate constant P in L; and
(3) a predicate constant M' of type ($I, ... , sm ), where Sj = 0, for
i ~ m, for each m-place operator M in L. Similarly, for each
pragmatic interpretation 0, V, F> for L there is an intensional
interpretation 0, V, F') for L' such that (1) F'(c) = F(c) for
each individual constant c in L; (2) F'(P') = F(P) for each n-
place predicate constant P in L; and (3) for each m-place operator
M in L, F'(M') is that function on I such that for iEI F'(M')(i)
= {(cf (J d, ... , cf(Jm )}: (J I' ... , 1m) E F(M)(i)}, where cf(Jd
is the characteristic function of 1k , for k ~ m.
A translation function t which translates each formula of L
into a formula of L' can be recursively defined as follows:
t(lr = 71') = 1" = 71'; tepCrl' ... , ~n)l) = Ip'(rl' ... , rn)l; t(~IP')
= 1""t(IP)l, and similarly for", v , ~, ~>; teA Vi ~) = 1;\ Vi t(lP)l,
and similarly for Iv Vi ~l; and finally t{M 1/11 ... I/Im') = V PI
... vPm [D[PI ~t(I/II)] " ... " 0 [Pm ~> t(l/Im)] " M'(PI' ... , Pm»)·
It is now easily proved that if I{) is a sentence of Land iEI, then
124 N. Cocchiarella

I{) is true at i in the pragmatic interpretation 0, U, F) if and only if

t(l{) is true at i in the intensional interpretation <I, U, F').

6. PHILOSOPHICAL APPLICATIONS OF INTENSIONAL LOGIC

In turning to philosophical applications of his intensional logic,

Montague took up such questions as whether we need 'tolerate
such entities as pains, events, tasks and obligations' ([3], p. 148).
Sentences in which we purportedly talk about these entities 'play
a conspicuous role in philosophy, perceptual psychology, and
everyday discourse' (ibid.); and 'it therefore appears desirable to
investigate the nature of the entities in question, construct an
exact and convenient language in which to speak of them, and
analyze the pertinent notion of logical consequence' (ibid.). It
is this last task in particular which, according to Montague, is 'a
necessary preliminary to the rational treatment of certain philo-
sophical paradoxes' (ibid., p. 149).
Of course, if the entities in question can be reduced to the
properties and relations-in-intension of intensional logic, then this
last task has in fact already been accomplished; and it is not sur-
prising, therefore, to find Montague proposing just such a reduc-
tion. Thus, e.g., insofar as it is not unreasonable to regard an
(instantaneous generic) event of the sun's rising as corresponding
to the expression 'the sun rises at t', it is similarly not unreason-
able, according to Montague, to identify the (generic) event in
question with the property expressed by that expression. Instan-
taneous generic events, on this analysis, 'will form a certain class
of properties of moments of time' (ibid., p. 150); and the occur-
rence of such an event at a moment is to be explicated as the
possession by that moment of the property which is identified
with that event. Protracted generic events can of course be identi-
fied with properties of intervals, rather than of just moments, of
time.
Tasks, according to Montague, 'should be regarded as certain
two-place relations-in-intension between persons and moments'
(ibid., p. 15 I ); and therefore the performance of a task R by a
person x at a moment t is to be taken as x's bearing the relation-
in-intension R to 1. Pains, 'and more generally experiences, are of
 Richard Montague and the logical analysis of language 125

the same ontological sort as tasks: they form a certain class of

relations-in-intension between persons and moments' (ibid.).
Thus, for example, 'the experience of seeing a tree is the relation-
in-intension born by x to t just in case x sees a tree at t' (ibid.);
and therefore for x to have the experience R at t is for x to bear
the relation-in-intension R to t. Finally, obligations, according to
Montague, can also 'best be regarded as the same sort of thing as
tasks and experiences, that is, as relations-in-intension between
persons and moments' (ibid.); and therefore for a person x to
discharge or fUlfill an obligation R at a moment t is again for x to
bear the relation-in-intension R to t.
One point to be noted in these reductions is that on Montague's
view 'the basic notion of partaking of a property ... is expressed in
ordinary English by several different verbs, depending on context;
by 'perfonn' in the case of a task, 'have' in the case of an
experience, and 'occur' in the case of an event' (ibid.); and of
course by 'discharge' or 'fulfill' in the case of an obligation. It
should also be noted, however, that in none of these reductions
does Montague claim to have defined, e.g., the notion of an
experience, or of a task, etc. What he does claim is that the
property of being an experience, of being a task, etc., is to be
regarded as a property of relations-in-intension between persons
and moments.
The reduction of these 'dubious ontological categories' to one,
viz., that of relations-in-intension, led Montague to place more
emphasis on the ontological status of the indices and individuals
involved in the set-theoretic construction of relations-in-intension.
Instead of being merely possible contexts of use, indices are now
spoken of as full-fledged possible worlds and the individ uals are all
the differen t possibilia that might exist in these different possible
worlds. There is, however, some question here as to whether the
notions of a possible world and of a possible individual have an
absolute or only a relative status in the set-theoretic foundations
of Montague's new theoretical framework; and we shall return to
this question at a later point in this essay.
The important observation which should be made here in any
case, however, is that by reducing these 'd ubious ontological cate-
gories' to that of relations-in-intension, Montague has in effect
produced 'an exact language capable of naturally accommodating
126 N. Cocchil1rella

discourse about the dubious entities' as well defined 'an intuitively

satisfactory notion of logical consequence for the sentences of
that language' (ibid., p. 154); and in consequence we have at least
the necessary preliminaries for a rational treatment of a variety of
philosophical puzzles or paradoxes.
Toward the treatment of such puzzles in the present version of
intensional logic, Montague utilizes the modal operator as a means
for constructing names of specific relations-in-intension. Thus,
adopting the following abbreviatory notation:
VI'" vn I{) = dfTPAv I ... AVnD [P(v l , ... , vn ) +;"IP)
we can read'vI ... vn 0{} as 'the relation-in-intension between those
(possible) individuals VI, ... , Vn such that Ifn = 1, then Iv.J
J.
designates the property of those individ uals that satisfy I{J (as
values of the individual variable 'v.'; and for n = 0, we use '~r.pl
for Ithe proposition expressed by J.
In general, at least when the formula in question is not too com-
"2
plex, Montague identifies ry.o{) and Iv. J with the use of infini-
tives and gerunds in English. E.g., where 'Bald' is a predicate
constant of type (-1), 'vI Bald (v. ~ represents the different
English expressions, 10 be bald', 'being bald' and 'the property of
being bald'. The last expression is of course also represented by
the predicate constant 'Bald' itself so that

is taken here to be logically true; and in fact it is an instance of

the more general logical truth:
(l) AP [P = uP(u)]

It is important to note here, however, that (1) cannot be

instantiated to just any predicate expression and still result in a
logical truth. E.g., suppose we restricted ourselves to the present
moment and took experiences, tasks, etc., to be (relational)
properties which a person may or may not now possess; and
suppose that a certain person, Jones, is now having the experience
of seeing a tree, although in some other possible world the same
person, Jones, is having a completely different experience. Clearly,
there is nothing inconsistent in this situation; and the assumption
 Richard Montague and the logical analysis of language 127

that (the property of) seeing a tree is an experience and that it is

in fact the experience which Jones is now having can be plausibly
represented by:

(2) Experience (0 [u sees a tree])

(3) TP[Experience (P) 1\ P(1ones)] =0 [u sees a tree]

Moreover, the (modal) fact that Jones is not seeing a tree in some
other possible world but is having there some completely different
experience indicates that the property of having the experience
which Jones is having is not identical with the property of seeing
a tree, i.e., that

(4) O[TP[Experience (P) 1\ P(1ones)] (u)] *" O[u sees a tree]

is also true in the situation described - since otherwise the two
properties designated in (4) would have the same extension in
every possible world, which is contrary to assumption.
As Montague points out, it is fallacious to attempt to derive
a contradiction here by claiming that

(5) TP[Experience (P) 1\ P(1 ones)] = 0 [TP[Experience (P)

1\ P(1ones)] (u)]

follows from (1) by universal instantiation, and that therefore,

by (3) and Euclid's law that things equal to the same thing are
equal to one another,

O[TP[Experience (P) 1\ P(1ones)] (u)] = u[u sees a tree]

is true after all. For the fact is that the variable P occurs in (1)
within the scope of the modal operator; and in consequence (1)
can be instantiated only to those predicate expressions that
(rigidly) designate the same pro perty in every possible world,
whereas of course 1'P[ Experience (P) 1\ P(1 ones)]', by assumption,
does not designate the same property in every possible world.
Thus, while the experience which Jones is having is indeed a
property, it is nevertheless not the property of having the
experience which Jones is having.
128 N Cocchio.rella

Another example to which Montague applies his intensional

logic is in the analysis of:

(6) Jones sees a unicorn having the same height as a table actually
before him.

which according at least to some philosophers implies the exis-

tence of sense data. Here Montague takes 'sees' in its veridical
sense and in terprets its nonveridical sense as 'seems to see', where
'seems' is taken as a predicate constant of type (-1, -1, 0, as in
'x seems to y to see a unicorn', i.e., as in 'Seems (x, y, u [Vz
(Unicorn (z) 1\ Sees (u, z»])1. Montague's analysis of (6) then is
as follows:

(7) Vx [Table(x) 1\ Before (x, Jones) 1\ Seems(Jones, Jones,

U[Vy(Unicorn (y) 1\ Sees(u, y) 1\ Has-the-same-height-as
(y,x»])]

Needless to say, but no sense data are required on this analysis, so

that an argument for the existence of sense data based on (6)
must be deemed inconclusive (cf. [3], p. 171).
Circumlocutions, such as 'seems to see' for the nonveridical
sense of 'sees', are rather common in philosophical analyses and
are in general based on the assumption that opaque contexts in
ordinary language, such as that of the direct-object position in the
nonveridical sense of 'sees', are always to be analyzed by means of
a paraphrase in which a sentence or formula occurs within the
scope of a modal operator. It is the occurrence of a full sentence
or formula within the scope of a modal operator, in other words,
which on this assumption explains why, e.g., existential general-
ization or the substitutivity of identity fails for opaque contexts.
Thus, e.g., while 'x sees a unicorn' can be represented by

Vy(Unicorn(y) 1\ Sees (x, y»

when 'sees' is taken in its veridical sense, it can also be represented

by
 Richard Montague and the logical analysis of language 129

Seems (x, x, u[ vy(Unicorn(y)" Sees (u, y»] )

when 'sees' is taken in its nonveridical sense; and in the latter

analysis the formula Ivy(Unicorn(y) " Sees(u, y~ occurs within
the scope of a modal operator (which is implicit in the definition of
'u[ ... ]'), so that nothing follows about the existence ofa unicorn
that only seems to be seen.
As another example of the apparent need for circumlocution,
consider the verb 'seek', which also generates an opaque context
with respect to its direct-object position. When compared with
the verb 'find', what we notice is that while

Jones finds a unicorn; therefore, there is a unicorn.

is valid, the apparently similar argument:

Jones seeks a unicorn; therefore, there is a unicorn.

is in valid 2 ; and this amounts to a philosophical puzzle insofar as

we need to explain how of two arguments of apparently the same
logical form one can be valid and the other invalid.
Relative to the present version of intensional logic, Montague's
own approach to this example is to regard 'tries to find' as a
circumlocution for 'seeks', where the verb 'tries' is represented by
a predicate constant of type (-1, I >. The above argument is then
analyzed. by Montague as follows:

Tries (Jones, u[vy (Unicorn(y) " Finds (u, y»]); therefore,

Vy Unicorn(y).

which of course is completely different in logical form from:

Vy (Unicorn(y)" Finds (Jones, y»; therefore, Vy Unicorn(y).

Thus, upon analysis through circumlocution and paraphrase, the

two arguments are seen to have significantly different logical
forms after all, so that the validity of the one need not, and in fact
does not, carryover to the other.
Despite his use of circumlocution and paraphrase in a number
130 N. Cocchiarella

of analyses, Montague nevertheless rejects the general assumption

that opacity is always to be accounted for by the occurrence in
ordinary language of a sentence or formula within the scope of a
modal operator and that therefore sentences containing intensional
verbs, such as 'seek', 'conceive', 'worship', etc., that do not involve
such an occurrence of a sentence or formula are to be paraphrased
into sentences that do. This is because opacity, for Montague, is
but an aspect of intensionality which is present in the semantics of
a number of transitive verbs. It is an aspect, however, which
cannot be represented in the present formulation of intensional
logic without resorting to circumlocution, though, as we shall see,
it can be represented without circumlocution and paraphrase in
Montague's second and later development of intensional logic.

7. INTENSIONAL LOGIC BASED ON THE SENSE-DENOTATION

DISTINCTION

It is noteworthy that the only entities explicitly acknowledged in

Montague's first version of intensional logic, i.e., entities to which
quantificational reference can be made in the object language
itself, are possible individuals and, for each nEw, n-ary relations-
in-intension, where the latter include propositions (for n = 0) and
properties (for n = I). The apparently excluded ontological
categories of individ ual concepts and (n-ary) relations-in-extension,
it turns out, are directly reducible to relations-in-intension. E.g., as
an (I, U>-intension, we have identified an individual concept with a
function f from I into U; and therefore the same individual con-
cept can equally well be identified with a function f' from I into
singleton subsets of U, where f'(i) = {f(i)}, for each iEI. Thus, we
can identify individual concepts with those properties satisfying
the following formula (as values of the one-place predicate variable
P) (cf. [2], p. 132):

o vuAv[P(v)-B-v=u]
As n-ary relation-in-extension, on the other hand, can be identi-
fied with an n-ary relation-in-intension that has the same extension,
viz., the relation-in-extension in question, at every index. Thus, we
 Richard Montague and the logical analysis of language 131

can identify n-ary relations-in-extension with those n-ary relations-

in-intension that satisfy the following fonnula (as values of the
n-place predicate variable P) (ibid .):

Having made these reductions, one might well maintain that the
distinction between the intension and extension, or the sense and
denotation, of an expression is neither fundamental to intensional
logic - and therefore not a basic distinction for philosophical
analysis - nor necessary, as apparently Gottlob Frege thought, for
the analysis of ordinary language. One might well speak accor-
dingly, of the denotation simpliciter of individual and predicate
expressions, even though the denotations of the latter would in
this case be what others call their senses. Indeed, in [4] , Montague
explicitly endorses and carries through just such a view in his
semantics for a fragment of ordinary English. 'It is wrong to
maintain', Montague argues there, 'that an analysis of ordinary
English (or German) requires a notion of sense as well as one of
denotation' (p. 217).
In [5], on the other hand, Montague withdraws his emphasis
'on the possibility of doing without a distinction between sense and
denotation'; and he remarks that 'while such a distinction can be
avoided in special cases, it remains necessary for the general
theory, and probably provides the clearest approach even to the
special cases in question' (p. 222). And having changed his view on
this matter, Montague proceeds to fonnulate his second version of
intensional logic in which the distinction between sense and
denotation now is taken as fundamental.
The ontological framework of this new intensional logic is to be
based on a type-theoretical distinction regarding the various
entities that are to be explicitly acknowledged in the logic, i.e., the
entities to which quantificational reference is to be meaningful,
and a theory of types which is essentially an extension of Alonzo
Church's theory of simple types. To set-theoretically characterize
the types, let e, t, s be three objects, none of which is an ordered
pair; and let Type, the set of types, be inductively defined as the
smallest set such that (1) e, t E Type; (2) (a, b) E Type whenever
a, bE Type; and (s, a) E Type whenever a E Type.
132 N. Cocchiarella

Entities of type e are understood to be the possible individuals

of intensional logic, while entities of type t are the two truth-
values. An entity of type <a, b> is to be a function from entities of
type a to entities of type b; and therefore entities of type (a, t>
are of course characteristic functions of sets of entities of type a,
all of which we will hereafter speak of as being those sets them-
selves, i.e., we will not distinguish here between a set and its
characteristic function. Also, we will not distinguish here between
an n-ary relation-in-extension between entities of type a l , ... , an,
».
respectively, and the function of type (at (a2 ... <an, t) ... E.g., a
set of individuals and a binary relation-in-extension between
individuals are construed here as entities of types (e, t> and (e, (e,m,
respecti vely.
Finally, where a E Type, the sense, or intension, or an entity of
type a is an entity of type (s, a); and therefore senses, or inten-
sions, as can be seen from the definition of Type, are always
senses of an entity of some type, i.e., there are no entities of type
s simpliciter. Senses, or intensions, of sets of entities of type a,
i.e., senses of type (s, <a, t» are understood in this framework to
be the properties of entities of type a; and senses of n-ary
relations-in-extension between entities of type a l , ... , an, are
understood to be the n-ary relations-in-intension between entities
of these types. Propositicm:, on the other hand, are understood to
be the senses of type (s, t).
In regard to the syntax of intensional logic, the only primitive
syncategorematic signs, besides parentheses and brackets, that
Montague adopts are =, the identity sign; X, Church's function-
forming X-abstraction operator; and' and -, the intension-forming
and extension-forming operators. (Here we see already how
fundamental the intension-extension, or sense-denotation, dis-
tinction is on this version of intensional logic.) We also assume, for
each aE Type, the availability of denumerately many variables of
type a as well as of denumerably many constants of type a. The
set of meaningful expressions of type a, MEa, is recursively
defined (on the set of types) as follows:

(I) every variable and constant of type a is in MEa;

(2) if aEME(a ,b) and j3EMEa , then ra(j3~ E MEb ;
 Richard Montague and the logical analysis of language 133

(3) if aEMEa and u is a variable of type b, then

~u~ EM~b,a);
(4) ra
if a, (3E MEa, then = jj' E MEt;
(5) if aEMEa , then r;.~ EME<s a);
(6) if a E ME< s ,a), then r:. a::l E MEa.
'

The logical constants of the earlier version of intensional logic

can now of course be taken as defined rather than as primitive
constants. E.g., where aEType, a is the first variable of type a, p is
a variable of type t, (j is a variable of type (t, 0, and I.{), I/J are for-
mulas, i.e., I.{), I/J E MEt :

"al.{) = df XC4p = Xa[a = a]

"'I.{) =df I.{) =" pp
I.{)II I/J= df 1\(j(1/J = [J3(I.{) = (j(I/J)])
The necessity operator and our earlier formal counterpart for
infinitive constructions are definable here as follows, where I.{)
is a formula, a is a variable of arbitrary type a and u is a variable
of type e:

DI.{) = df [AI.{):;; AI\U(U:;; u))

&p = df[ -Xal.{)]

It should also be noted here that if a E type, a E MEa and

8 EM~a,t), i.e., if 8 is of the type of a set of entities of type a,
then ~(a)l expresses the 'membership' of the entity denoted by
a in the set denoted by 8. On the other hand, if 8EME(s,<a,t»
instead, i.e., if 8 is of the type of a property of entities of type a,
then we use:

to express the 'possession' by the entity denoted by a of the

property denoted by 8.
In constructing a set-theoretic semantics for the present version
of intensional logic, we again need to regard certain sets I and U as
being the set of possible worlds and the set of possible individuals,
respectively. Where a E Type, the set of possible denotations (or
134 N. Cocchiarella

values of variables) of type a as based on U and I, in symbols

Da , V, I, is defined recursively as follows:

(1) De,V,1 =V;

(2) D t v I ::;: {D, I};
. ' , D
(3) lfa, bEType,D(a,b},U,1 =(Db,v,d a,V,I;
(4) if a E type, D(s,a}, U, I = (Da, U, J)I.

Besides the set of possible worlds I, Montague wants us to also

consider a set J of contexts of use in the possible worlds in I.
The indices or points of reference now are neither possible worlds
alone nor contexts of use alone but are rather ordered pairs of
each, i.e., a reference point is a pair 0, j), where iEI and jEJ. It
should be noted, however, that Montague now takes moments,
when they are needed in the interpretation of tenses, to be com-
ponents of members of I rather than components of members of
J (cf. [5] , p. 228).
Relative to the sets V, I, J, and a type a, Montague takes the set
of possible meanings of type a, in symbols Ma , U,I, J, to be
functions from the points of reference into possible denotations
of type a, i.e., Ma , U,I, J = (Da , U,d x J. Meanings, on this inter-
pretation, are functions of two arguments: a possible world and
a context of use. Senses, or intensions, on the other hand, are
functions of only one argument, viz., a possible world (which may
include specification of a moment of time). It should be noted,
however, that only senses, and not meanings, are explicitly
acknowledged in the ontological framework of the theory, i.e.,
it is senses, and not meanings, that can be quantificationally
referred to within the logic itself. Meanings, rather than senses, are
needed to serve as the interpretations of expressions, on the other
hand, because the interpretation of a compound is always to be a
function of the interpretation of its components, including of
course indexical components such as pronouns and demon-
stratives; and functions on possible worlds alone will not always
suffice to satisfy this condition.
An interpretation for the present version of intensional logic,
accordingly, can be taken to be a 4-tuple 0, J, V, F), where
(1) I, J, V are non empty sets representing respectively, the
possible worlds, the contexts of use, and the possible individuals
 Richard Montague and the logical analysis of language 135

of the interpretation; and (2) F is a function defined on the set of

constants such that whenever a E Type and a is a constant of type
.,
a, then F(a) E Ma , U 1 J. Naturally, relative to such an interpre-
tation we can again follow (but shall avoid doing so here) the
recursion clauses in the definition of a meaningful expression of a
given type so as to characterize the meaning each such expression
is assigned under the interpretation; and of course, where a E
Type, the meaning of an expression of type a will be a member of
Ma, U, I, J.
Where I{J is a sentence, i.e., a member of MEt in which no
variable has a free occurrence, and 0, j) is a point of reference of
an interpretation as defined above, then I{J is true at G, j) (under
the interpretation) if the meaning If) has under that interpretation
has the value I at the reference point G, j). Definitions of logical
truth and logical consequence do not immediately follow, how-
ever, since it is possible that under some interpretations some of
the logical constants might not be assigned their standard or usual
meanings. We can of course restrict our considerations to the class
of those interpretations under which all the logical constants do
receive their usual meanings and define logical truth and logical
consequence with respect to this class. We can also extend this
class, however, by including interpretations under which the
logical constants receive their usual or standard extensions of all
designated or standard points of reference but not also at certain
other 'unactualizable' points of reference (cf. [5]. p. 231). Sen-
tences that are synonymous, i.e., that have the same meanings,
under all the interpretations in this latter class will then also be
logically equivalent; but then there will also be sentences that
are logically equivalent, i.e., that have the same truth-value at all
of the designated points of reference of all the interpretations in
the class, that are not synonymous under every interpretation in
the class.
This divergence between synonymy and logical equivalence,
Montague notes, allows for 'a natural treatment of belief contexts
that lacks the controversial property of always permitting inter-
change on the basis of logical equivalence' (ibid .). It should also
be noted in this regard, however, that Montague's treatment of
belief contexts in his first version of intensional logic had in fact
possessed this controversial property (cf. [2], pp. 137 ff.); and,
136 N. Cocchiarella

indeed, even here in his second version of intensional logic it is not

so much that Montague rejects his earlier analysis as that he
wishes only to point out that that approach 'has genuine alterna-
tives and is not forced upon us' ([ IS] , p. 231).

8. ON THE INTENSIONAL REALITY OF POSSIBLE WORLDS

We have already noted that the only entities to which quanti-

ficational reference can be made in Montague's first version of
intensional logic are possible individuals and, for each nEw, n-ary
relations-in-intension. And the situation is really not that much
different in the new version of intensional logic, although now
quantificational reference to individual concepts (as senses of type
<s, e» and relations-in-extension can be effected directly rather
than indirectly through relations-in-intension. The question
naturally arises then as to the status of possible worlds in the onto-
logical framework of intensional logic. That is, in what sense, if
any, does the adoption of intensional logic as a new theoretical
framework for philosophy presuppose the existence or 'being' of
possible worlds? In particular, are possible worlds among the
entities to which quantificational reference can be made within
this new theoretical framewc~k?
Now it may seem natural to maintain that possible worlds have
been assumed to exist all along in Montague's set-theoretic seman-
tics for intensional logic; and, indeed, it is true that Montague even
considers adding to the language of set theory 'two definite
singular terms, one to designate the set of possible worlds, and the
other the set of possible individuals' ([3], p. 154). But he also
goes on in the same context to reject that suggestion; and in his
definition of an interpretation 0, U, F) for an intensional lan-
guage, or of an interpretation 0, J, U, F) for intensional logic, we
are not told that the sets I and U are the set of possible worlds
and the set of possible individuals simpliciter but only that they
are nonempty sets that are to be regarded as the set of possible
worlds and possible individuals, respectively, of the interpretation
in question. And in that sense, the notions of a possible world and
of a possible individual have only a relative, and not an absolute,
status from the ontological point of view.
 Richard Montague and the logical analysis of language 137

What would render this notion absolute rather than relative in

Montague's set-theoretic semantics, or at least what would be a
necessary condition for such, would be an assumption to the
effect that if 0, V, F) and (I', V', F') are two interpretations for
an intensional language L, then I = I' and V = V'. And if Montague
had added to the language of set theory the two special singular
terms mentioned above, then he might well indeed have made just
this assumption. But in fact Montague rejects adding such singular
terms and there is no indication at all that he intends to make any
such assumption. Indeed, to the contrary, Montague explicitly
says that intensional logic transcends set theory, and that there-
fore it is not reducible to set theory even though it can be 'justi-
fied' by metamathematical or model-theoretic means available
within set theory ([3], p. 155).
The question of Montague's commitment to possible worlds,
accordingly, applies not to his set-theoretic semantics but to his
adoption of intensional logic as a new theoretical framework for
philosophy. For possible individuals, the commitment is quite
explicit: these are the entities indicated by bound individual
variables, i.e., variables of type e; and it is clear, moreover, that
whatever else a possible world is within the framework of
intensional logic, it is not an individual, since in that case relations-
in-intension would reduce to relations-in-extension. 3
Perhaps the most plausible suggestion is the identification of
possible worlds with certain 'world propositions', where the
notion of a world as a prcperty of propositions is definable as
follows, where p, q are variables of type (s, 0:

A world-proposition is true in Montague's set-theoretic semantics

in one and only one possible world (of any given interpretation)
and may therefore be taken as the intensional counterpart, i.e.,
the counterpart within the framework of intensional logic, of that
possible WO·'i.4 Vnderstood in this sense, Montague's commit-
ment to possible worlds is now quite clear; for the sentence:

DVp (World {p} A vp)

is true at every reference point of every interpretation.

138 N. Cocchiarella

9. ON QUANTIFYING OVER POSSIBLE INDIVIDUALS

Quantificational reference to individuals that do not exist (at a

given index) is sometimes objected to by philosophers who other-
wise have no qualms about quantifying over propositions and
relations-in-intension. Quantification with respect to individual
variables, these philosophers maintain, should be restricted in
content so as to apply only to the individuals that exist at the
index in question. This view, at least with respect to Montague's
overall semantic approach, is quite unwarranted and can in fact be
refuted if the claim is that direct quantification over possible
individ uals cannot be justified using only the restricted quantifiers
referring to existing individuals in modal or intensional contexts.
Consider, for example, the context of tense logic where the
restricted individual quantifiers are supposed to refer only to the
individuals existing at the time of utterance; and, for convenience,
let us use 1\ e and ve for these restricted quantifiers. Then the view
in question, as applied to this context, maintains that the only
significant way we can quantificationally refer to past or future
individuals is indirectly, i.e., where these restricted quantifiers
occur within the scope of the past and future tense operators; and,
a fortiori, on this view the only significant way we can quantifi-
cational refer to possible individuals in the context of modal or
in tensional logic is also in directly, i.e., where the restricted quanti-
fiers occur within the scope of the modal or intension-forming
operator.
Our counter-thesis is that by taking advantage of the prag-
matic vperators available within Montague's general framework
we can fully justify direct quantification over past and future
individuals in the context of tense logic and direct quantification
over merely possible individuals in the context of modal or inten-
sional logic. To show this, consider the problem of how we might
formalize the English sentence:

(8) There did exist someone who is an ancestor of everyone now

existing.

using only indirect quantification over past individuals. In this case

it is clear that the symbolization (where the variables are for con-
 Richard Montague and the logical ana,vsis of language 139

venience understood to range only over persons):

will not do, since this fonnula represents the different English
sentence:

(9) There did exist someone who was an ancestor of everyone

then existing.

Clearly, what we need here is a way, whether primitive or defined,

of directly quantifying over past as well as over presently existing
individuals. E.g., using flY and vP for such quantifiers, we can
formalize (8) above as:

VPUAev Ancestor (u, v)

And of course a completely analogous argument shows the need

for away, whether primitive or defined, of directly quantifying
over future individuals.
Now the point of our counter-thesis is not that there are exam-
ples that clearly show the need for directly quantifying over past
and future individuals in the context of tense logic, or for directly
quantifying over possible individuals in the context of modal or
intensional logic; but rather that by fully utilizing the pragmatic
orientation of Montague's semantics we can in fact actually define
the quantifiers needed for such direct quantification in terms of
the restricted quantifiers occurring within the scope of the tense,
modal or intension-fanning operators.
As it turns out, what is needed to realize this type of definition
is the pragmatic operator AI" for 'it is now the case that', as
developed by Hans Kamp, one of Montague's former students
(Kamp [14]). The introduction of this operator, Kamp has noted,
requires a procedure of double (rather than single) indexing and
a redefinition of truth at an index to truth at an index when part
of an utterance in the present context of use; and of course it also
requires specifying that index in each interpretation which is to
be regarded as the present context of use in that interpretation.
This supplementation of the relevant parameters involved in the
140 N. Cocchiarella

analysis of truth is of course completely within the spirit of the

pragmatic orientation of Montague's semantics.
Using the pragmatic now~perator in the analysis of (8), we see
now how to avoid assuming the prior introduction of the quanti-
fiers II P and v P . For in this case, the formula

fully captures all that we want to say with (8) using only indirect
quantification over past individuals. And of course what this
analysis indicates is that we can contextually define direct quanti-
fication over past individ uals in general:

=df "'.qi'''' /\ e u v~· <{J

liP U <{J
vPu <{J = df ~Veu .¥ <{J

Naturally, essentially the same pattern of definition suffices to

define direct quantification over future individuals; and assuming
that the modal operator is interpreted either as in Montague's
first version of intensional logic or is defined as above in terms of
the intension-forming operator, we can then similarly define
direct quantification over possible individuals as follows:

IIU<{J = df Dlleu ..¥<{J

Vu <{J =df oveu .Y<{J
Thus the claim that only indirect quantification over possible
individuals should be allowed in intensional logic is, as we have
said, quite unwarranted. Whether, on the other hand, there are
reasons to prefer defining II and V as above in terms of lie and ve
or for defining /\ e and Ve , as Montague (in effect) does, in terms
of II and V and a predicate for 'exists' is another question
altogether; but one which we shall not go into here. (This question
is discussed in Cocchiarella [15] .)
 Richard Montague and the logical analysis of language 141

10. UNIVERSAL GRAMMAR

While the development of intensional logic as a new theoretical

framework for philosophy marked one important shift in
Montague's philosophy of language, the development of his
universal grammar as a universal syntax and semantics applicable
to natural as well as to artificial languages has come to mark an
even more important shift not only for Montague but for many
logicians and linguists as welL Natural languages, according to
Montague, are no longer to be distinguished from the artificial
languages of logicians on the grounds that only the latter have a
mathematically precise syntax and semantics; and, indeed, in this
regard Montague's universal grammar is intended 1:0 comprehend
the syntax and semantics of both kinds of languages within a
single natural mathematically precise theory' ([5], p. 222).
It is important to note here that the mathematically precise
theory Montague has in mind for universal grammar is in effect a
definitional extension of set theory (with proper classes). This is
particularly noteworthy in that it may seem in the present context
that Montague has thrown over his new theoretical framework for
philosophy in favor 0 f the old. To the contrary, however, inten-
sional logic serves for Montague not only as an example within
universal grammar of an important artificial language but also,
and more importantly, as a semantical fra;nework in its own
right and into which extensive portions of natural language are to
be translated by means of a rigorously described theory of trans-
lation which Montague also formulates as part of universal
grammar. Such a rigorously characterized translation into
intensional logic induces, according to Montague, a more
perspicuous interpretation for the expressions of a natural
language then would otherwise be given them in purely set-
theoretic terms. The sense, moreover, in which intensional logic
transcends set theory even while being metamathematically
'justified' by set-theoretic constructions remains no less in force
in universal grammar than when these constructions were given
independently of such a single comprehensive theory.
Nevertheless, despite this transcendence, Montague does formu-
late his universal grammar in purely set-theoretical terms and
distinguishes in that grammar both a universal syntax and universal
142 N. Cocchillrella

semantics. (The term 'universal' is meant here, it should be noted,

only in the logico-mathematical sense of utmost generality). As
applied to any given language, the basic aim of the syntax is to
characterize the various syntactical categories, including especially
the category of declarative sentences, of that language, while the
basic aim of the semantics is to characterize the notions of a true
sentence (under an interpretation) and of entailment between the
sentences of that language ([5], p. 223). The two components are
intimately related by Montague in that where L is a language and
.<:1 is an interpretation for L, there exists in .wa semantical oper-
ation corresponding to each structural operation in the syntax of
L; and, furthermore, in that part of his semantics which Montague
calls the theory of reference, there will in addition be a unique
semantic category corresponding to each syntactic category of
L (though the correspondence will in general only be many-one
and not one-to-one).
Montague's theory of reference, we should note, is essentially
a model theory based upon the set of types already described for
intensional logic and involves assigning the model-theoretic
counterparts of senses or intensions as well as the more usual
extensions to expressions. Senses differ in this regard from mean-
ings in Montague's semantics in that they, but not meanings, are
'intensional entities that are sometimes denoted by expressions'
((5 J, p. 228). The types in general, moreover, whether extensional
or intensional, are in fact the semantic categories that are to be
associated with the syntactic categories of a language L; and,
moreover, it is precisely because a type is assigned to each syntactic
category of L that the more perspicuous interpretation of L which
is induced by its translation into intensional logic can be effected.
Thus, in a way the real point of the set-theoretical framework is
to serve first as a syntactical metalanguage within which the syn-
tax of any language whatsoever might be described, including the
language of intensional logic, and, secondly, to serve in that regard
as the syntactical basis of a theory of translation within or by
means of which each such language might be translated into
intensional logic and thereby be given its most perspicuous seman-
tical representation.
Montague's set-theoretical description of a language L amounts
in effect to identifying L with a system (L*, R), where L* is a
 Richard Montague and the logical analysis of language 143

(syntactically) disambiguated language (and therefore an artificial

construct) and R is a binary relation (of syntactical analysis) with
the set of proper expressions of L* as its domain and the set of
proper expressions of L as its range. R will in general by many-
one, relating several disambiguated proper expressions of L*
(as alternative analyses) with each syntactically ambiguous proper
expression of L. The disambiguated construct L* is defined by
Montague to be a system (A, F y, XI), S, c5 o>')'Er, I)E~' where
(A, F ,),>,),E r is a free algebra generated by UI) E ~ XI), with A as the
set of proper expressions of L* and each F.y a structural operation
on strings of members of A (resulting in new members of A) and
with the sets XI) being the sets of basic phrases of each of the
categories of L *, 6 being the set of indices of these syntactic
categories and 15 0 being the index of the category of the declara-
tive sentences of L*; and where S, the syntactical rules of L*, is a
set of sequences of the form (F,)" (c5 t )t<(J' 1'/>, where -yEr, (3 is the
number of places of the operation F')', c5 t E.1 for all ~ < {3, and
1'/ E .1. (A syntactical rule as described in effect stipulates that the
structural operation F')' operates on a {3-place string of proper
expressions, each of category 15 t, respectively, for all ~ < (3, and
results in a proper expression of category 1'/). The family of phrases
of the syntactic categories of L* that are generated in this way in
L* is inductively characterized as the smallest indexed family
(P6)1) E A of subsets of A such that (1) XI) S PI) , for all 15 E 6, and
(2) whenever (F,)" (c5 t >t < (1'1'/> E S and at E PI) t' for all ~ < (3, then
F')' «at)t < (1) E PTj.

11. THE TRANSLATION OF ENGLISH INTO INTENSIONAL LOGIC

We shall forgo the set-theoretical details of Montague's general

theory of translation and briefly illustrate instead the particular
translation Montague gives for the fragment of English formu-
lated in [6]. The fragment is philosophically interesting in its
own right, moreover, just for the novelties and insights it exhibits,
such as, e.g., Montague's unified treatment of quantifier phrases
and terms of English. Indeed, it is the particular type of formu-
lation Montague gives for this fragment of English as well as the
form of its translation into intensional logic that has been developed
144 N. Cocchiarella

in recent years into an active, and apparently fruitful, field of

research at the interface of linguistics and logic. 5
Montague defines the set of syntactic categories for the frag-
ment of English in question to be the smallest set K such that
(1) the two basic types t and e of intensional logic are in K, and
(2) whenever A and B are in K, then so are A/B and A/ /B.
(K. Ajdukiewicz, the originator of categorial grammar, allowed
only one compound category A/B to be generated from the more
basic categories A and B. Montague's introduction and use of two
such compounds is in itself already a novelty of his grammar, and
he notes that 'in connection with other languages it is quite
conceivable that a larger number would be required' ([6], p. 249).)
Some of the more important traditional categories that are in K
are:

IV = t/e (the category of intransitive verb phrases)

T = t/IV (the category of terms)
TV = IV/T (the category of transitive verbs)
IAV = IV/IV (the category of IV-modifying adverbs)
eN = t//e (the category of common noun phrases)

Associated with each of the above compound categories, and

several others, are the sets of words or phrases that are taken by
Montague to be the basic phrases of English belonging to these
categories and from which, by means of the syntactical rules,
the more complex phrases of these categories are to be construc-
ted. Thus, e.g., words like 'run', 'walk', 'talk' and 'rise' are stipu-
lated as being basic expressions of category IV; and adverbs like
'rapidly' and 'slowly' belong to category IA V, while adverbs
like 'necessarily' belong to category t/t. Phrases for propositional
attitudes, e.g., 'believe that', 'assert that', etc., appropriately
belong to category IV It, while phrases like 'try to', 'wish to',
etc., belong to category IV //IV, which is syntactically quite
different from the category of adverbs IAV == IV/IV.
One of the general forms of the syntactical rules by means of
which the more complex phrases of English belonging to these
categories are to be generated can be described as follows: if 0:' is
a phrase of one of these compound categories A/B and f3 is a
phrase of category B, then Fj(O:', (3) is a phrase of category A,
 Richard Montague and the logical analysis of language 145

where Fi is the structural operation associated with the syntactical

rule in question. (As a set-theoretical construction of universal
grammar the syntactical rule in question is identified with the
three-tuple (F i , <A/B, B), A). A similar rule <Fj , <A//B, B), A) is
given for the compound category eN, as defined above, and for
IV / /IV.6) Rules of this form are called rules of functional appli-
cation; and taken together with the other syntactical rules
Montague describes they yield a simultaneous inductive definition
of the sets PA of phrases of English of each syntactic category A.
In regard now to Montague's translation of this fragment of
English into in tensional logic, let us note that the translation is
to be based first upon a certain mapping of the syntactic cate-
gories of English into the seman tical types of intensional logic and
then upon the construction of a rule of translation corresponding
to each of the syntactical rules involved in the simultaneous
inductive definition of the sets PA . The translation relation con-
structed in this way will of course not be a function, given the
phenomena of semantic ambiguity in English; but by means of an
analysis relation, which Montague defines in terms of analysis trees
(cf. [4], pp. 204-207), we can speak instead of the translation of
a meaningful expression of English corresponding to any given
analysis tree for that expression. For convenience, however, we
shall ignore further mention of analysis trees here.
The form of the mapping of syntactic categories into semantic
types cannot be ignored here, however; for aside from the obvious
association of the simple syntactic categories e and t with the
simple semantic types e and t, the mapping Montague defines is
based upon two fundamental assumptions regarding the semantics
of English constructions. These are (I) that despite important
differences in their syntactical roles in English, the compound
categories AlB and AIIB are to be assigned the same semantical
type; and (2) that in those cases where English constructions are
generated by syntactical or grammatical relations associated with
compound categories, the construction is to be interpreted as a
function whose arguments are always to be senses or intensions.
In particular, the general form of the semantical rules of trans-
lation that correspond to the rules of functional application as
described above is paradigmatic for this type of interpretation:
if a E PAlB or a E PAIIB, {3 E PB , and a translates into a' and {3
146 N. Cocchiarella

translates into (3', then Fi(O', (3) and Fj(O', (3) translate into ra'C{3'Y.
The mapping which is based upon these two fundamental
assumptions is defined as follows, for arbitrary categories A and B:

fCe) =e
[(t) =t
F(A/B) = f(A//B) = «s, f(B», f(A»

12. QUANTIFIER PHRASES AS ENGLISH TERMS

Now although there are expressions of type e in intensional logic

denoting individuals, there are no such related expressions in
Montague's fragment of English. There are of course English
terms, i.e., phrases of category T, such as 'John', 'Mary', 'ninety'.
and even term variables 'he]', 'he2', etc. The category T, however,
as defined above, is not a basic but a compound category t/IV;
that is, a term phrase when combined with an intransitive verb
phrase under a certain structural operation F j results in a
(declarative) sentence-phrase, i.e., a phrase of category t (which is
either a sentence of English or like a sentence of English except
for the occurrence of a variable 'hen '). Consequently, since f(T)
= f(t/IV) = «s, f(lV», 0, and since f(lV) = f(t/e) = «s, e), t>
= the type of sets of individ ual concepts, then, instead of denoting
an individual, an English term denotes a set of properties of indi-
vidual concepts. Prima facie, this may seem somewhat anomalous,
but it really falls rather nicely into place once we notice that the
terms of English are in effect what linguists call noun phrases and
that noun phrases include quantifier phrases as well as proper
nouns and definite descriptions. The anomaly then, if any, is not
that a term denotes a set of properties, but that it denotes a set of
properties of individual concepts rather than a set of properties
of individuals. (But more of this anon.)
Proper nouns and term variables, accordingly, are not the only
term phrases in Montague's fragment. In particular, there are three
syntactical rules (F j , CN, T), for i = 0,1,2, such that if t is a
common noun phrase, where 'man', 'woman', 'fish', 'unicorn',
etc., are examples of basic CN-phrases, then F 0cn
= revery r,
F Icn (n r
= ~he f, and F 2 = 'a(n) are all phrases of category T.
 Richard Montague and the logical analysis of language 147

This is of course a significant departure from standard logical

analysis. But then it is also very much in line with the traditional
(pre-Fregean) subject-predicate analysis of English sentences; and,
not surprisingly, it allows in this last regard for a more direct
connection between the syntax and semantics of English. It also
allows, it turns out, for a more direct representation of intensional
verbs which avoids the type of circumlocution described earlier,
such as paraphrasing 'seeks' as 'tries to find'; and it provides in
that regard for a more realistic alternative to the rather common
assumption that opaque constructions in English are always to be
paraphrased into constructions involving the occurrence of a
sentence or formula within the scope of a modal operator (cf.
Partee [18]).
Consider, for example, certain individual constants 'j', 'm' and
'n' of intensional logic understood to denote the individuals
John, Mary and the number ninety. The English terms 'John',
'Mary' and 'ninety', as noted above, are not translated by Mon-
tague, into 'j', 'ro' and 'n', respectively, but into rxppcrf1 ,
'APPCmf and 'XPPCn}', where P is a variable of type (s, (<S, e),
t», i.e., of the type of properties of individual concepts. Similarly,
where ~ is a eN-phrase of English and ~' is its translation into
intensional logic, then revery r,'the r, and ra(n) f1 translate
into:

XPl\x(r'(x) ~ p {x})
XPvx(l\y(r'(y) # y = x) A P {x})
APVX(r'(x) A P{x})

respectively, where 'x' and 'y' are variables of type (s, e), i.e., of
the type of individual concepts. Here, of course, proper nouns,
definite descriptions and quantifier phrases are all translated into
expressions of intensional logic that denote sets of properties of
individual concepts.
Simple predication of a term and an IV -phase on this interpre-
tation, it should be noted, treats what the term denotes as the
function and the intension of what the IV -phrase denotes as the
argument. Thus, e.g., the English sentence 'John runs' is translated
by Montague into intensional logic as follows:
148 N. Cocchiarella

(10) APP{",j} Cruns')

where 'runs" is the translation of the English IV -phrase 'runs', and

is therefore of type f(lV) = «s, e), 0, i.e., of the type of sets of
individual concepts; and '~runs" therefore is of type (s, «s, e), t»,
i.e., of the type of properties of individual concepts. However, by
obvious principles of intensional logic (10) reduces to:

(11) runs' ej)

which is now more nearly in line with standard logical analysis.

Of course, for an extensional IV-phrase such as 'runs', the
(reduced) translation (11) is still unduly intensional. Montague
rectifies this by first defining:

8. = df AU 8Cu)

where 8EMEr(lv) and 'u' is an individual variable (of type e),

and by then giving a meaning postulate of the form:

o [8(x) ~ b. ex)]
for any 8 which translates an extensional IV or eN-phrase. (Note
that since 8 is of type {(s, e), 0, i.e., of the type of sets of indi-
vidual concepts, then 8. is of type (e, 0, i.e., of the type of sets of
individuals.) Thus, given that 'runs' is an extensional IV -phrase,
(II) reduces to:

(12) runs' .(j)

which is exactly what the standard logical analysis of 'John runs'

comes to.
Needless to say, but an entirely similar analysis applies to the
English sentence 'Every man runs', which is directly translated as:

AP !\x[man'(x) -+ P{x}] Cruns')

and which first reduces to:

1\ x [man/ex) ~ runs/ex)]
 Richard Montague and the logical analysis of language 149

but which can be reduced to:

I\. ulman' * (u) -+ runs' *(u)]

assuming that 'man' is an extensional eN-phrase.

For extensional transitive verbs, i.e., English phrases of category
TV, the situation is analogous, though somewhat more involved
since TV = IV IT, and therefore f(TV) = «s, f(T», f(lV». That is,
what a transitive verb denotes in in tensional logic is a function
whose arguments are the intensions of terms; and for sentences
with extensional transitive verbs, this means that not only must we
reduce the intensionality of the subject term of the resulting IV-
phrase but also that of the term which is the direct object of the
transitive verb. Thus, e.g., where 'love" translates the English
TV -phrase 'loves', we translate 'Mary loves John' and 'Mary loves
every man' first into:

APP Cm) Clove'(PPCj}))

APP Cm}Clove'(Pi\x[man'(x) -+ P{x}]»

which, again by 0 bvious principles of intensional logic, reduce to 7 :

10ve'Cm, PP Cj})
10ve'Cm, Pi\x[man'(x) -+ P{x}])

which in turn, by the meaning postulate regarding the extension-

ality of 'loves's, reduce to:

love: (m, j)
I\.x[man'(x) -+ love: (m, -x)]

the second of which reduces finally to:

i\u[man: (u) -+ love: (m, u)]

The fact that transitive verbs are interpreted as functions whose

arguments are the intensions of term phrases is of course but one
example of Montague's general assumption, cited earlier, regarding
the interpretation of English constructions based upon compound
150 N. Cacchiarella

categories. It is this fact in particular, moreover, which together

with his treatment of quantifier phrases as term phrases that
enables Montague to explain and represent the opacity of transi-
tive verbs, such as 'seek' and 'conceive', without resorting to
circumlocution or paraphrase. Indeed, the fact is that this opacity
is but a feature of the intensionality that is assigned to all
transitive verbs, and it is precisely this opacity which needs to be
eliminated by meaning postulates in those contexts in which the
verb is to be given an extensional interpretation.
Consider, for example, the different treatment which 'seek'
receives in the translations of 'John seeks a unicorn' when this
is given its de dicta reading:

APPCj} Cseek' (Pvx[unicorn'(x) A P{x}]»

which reduces to:

seek'CL PVx[unicorn'(x) A P{x}])

from when it is given its de re reading:

Vx[unicorn'(x) A seek' Cj, PP{x})]

which, given the extensionality of 'unicorn', can be reduced to:

vu[unicorn: (u) A seek'Cj, PP eu})]

but which can be reduced even further in this case to:

vu[unicorn:(u) A seek: (j, u)]

Here, it is because intensionality is the rule that the opacity of

'seek' can be directly represented without resorting to circum-
locution and paraphrase; and in those cases in which the rule is
not to apply, a meaning postulate must be invoked so as to
eliminate the intensionality and consequent opacity of the verb,
and context, in question.
It is of course not only transitive verbs that are assigned this
intensionaiity, according to Montague. We already saw above
 Richard Montague and the logical analysis of language 151

that meaning postulates are needed in the case of extensional IV

and CN-phrases as well. But the fact is, however, that not all
CN and IV-phrases are extensional, and a uniform interpretation
that would render them all as such would therefore be quite
inappropriate. Consider in this regard the following clearly invalid
argument:

The temperature is ninety; the temperature rises; therefore,

ninety rises

Montague's analysis or reduced translations of the sentences in

this argument are:

VX[Ay(temperature' (y) ~y = x) 1\ l"yJ = nJ

Vx[Ay(temperature' (y) ~ y = x) 1\ rise' (x)J
rise' en)

where 'x' and 'y' range over individual concepts. Here we should
especially note that while the identity in question in the first
premise is not of two individual concepts but only of their
extensions at the present moment in the real world, the truth of
the second premise, according to Montague, depends on the fact
that 'the temperature,' loosely speaking, 'denotes' an individual
concept and not an individual and that 'rise', unlike most verbs
'depends for its applicability on the full behavior of individual
concepts, not just on their extensions with respect to the actual
world and ... [presentJ moment of time' ([6J, p. 268). Thus,
'temperature' and 'rise' are particular examples of an intensional
CN and IV -phrase, respectively; and it is precisely in order to
explain the invalidity of the above type of argument that
Montague interprets IV and CN-phrases in general as denoting sets
of individual concepts rather than sets of individuals. And, it is for
this reason of course that English terms, including quantifier
phrases, are interpreted by Montague as denoting sets of properties
of individual concepts rather than sets of properties of individuals.
152 N Cocchiarella

13. CONCLUDING REMARKS

There are many other important features of Montague's grammar

for English and of his translation of English by means of that
grammar into intensional logic that we cannot go into here. The
highly intensional nature of his semantics, for example, provides
not only a more direct analysis of the opacity of intensional verbs
but also a more direct analysis of the opacity of infinitive phrases
as well. And then there is his treatment of relative clauses and of
attributive adjectives which we have not touched upon at all.
In closing then, it will no doubt have crossed the reader's mind
that there may be some irony in the fact that Montague began his
philosophical career as an extensionalist who took set theory as
the proper theoretical framework for philosophy and as a formal
language philosopher who viewed the formalization of ordinary
language as either impossible or extremely laborious, and in any
case as certainly not philosophically rewarding. For the fact is
that Montague has made important and philosophically innovative
contributions toward a fully formalized syntax and semantics for
natural language and that the semantics in question is most
perspicuously described in an intensional logic which transcends
set theory and which constitutes in effect a new theoretical frame-
work for philosophy. If this is not a revolution, it is at least a form
of progress in the logical analysis of language.

NOTES

I. Our interpretation of 0 here as a one-place operator which is to be directly

assigned an intension under an interpretation is an inessential modification
of Montague's treatmen t.
2. Strictly speaking, 'Jones seeks a unicorn' is ambiguous between a de re
(referential) reading and a de dicta (nonreferential) reading. Such am-
biguity is typical of intensional verbs. The de re reading is unambiguously
given by 'Jones seeks a certain unicorn', from which, as premise, it does
follow that there is a unicorn. We are of course understanding the sen-
tence in the text to be given with its de dicta reading, from which it does
not follow that there is a unicorn.
3. It has been suggested that possible worlds be added as a new type of entity
altogether. E.g., let s be a basic type along with e and t, i.e., let s E Type,
and set Ds, U, I = I (eL Gallin [Ill, ~ 8.)
 Richard Montague and the logical analysis of language 153

4. From the algebraic point of view of Montague's set-theoretic semantics a

world-proposition might also be called an atomic proposition. (ef. Gallin
[II] , § II.) It should be noted here that we are assuming that all of the
logical constants involved in the above defmition have been defined in
terms of the primitives of intensional logic and that they cannot there-
fore, as so defined, be assigned 'nonstandard' meanings.
5. Cf. especially the collection of papers in Partee (ed.) [16]. Partee's essay
'Montague Grammar and Transformational Grammar' [17] is an excellent
introduction to a variety of the issues involved in combining Montague
grammar with transformational grammar.
6. Cf. the two converse structural operations F6(CX, (j) = ICX{j1 and F7(ex,{j)
= ~exl. The syntactical rules S8 and SIO of [6] apply as described above
when ex E PlY/IN and {j E PlY or when ex E PlYIN and (j E PlY. Thus S8
generates, e.g., 'try to run' as an IV-phrase, while SIO generates 'run
rapidly' as an IV-phrase .
7. Regarding the use of relational notation in intensional logic, we set

o(cx, (j) = df o({j) (ex)

where 0 E ME(b, (a, t», ex E MEa and {j E MEb.

8. Given the definition:

the meaning postulate for when 0 translates an extensional transitive

verb is:

-
where 9' ranges over properties of properties of individual concepts and
'x', 'y' range over individual concepts.

BIBLIOGRAPHY

I. Works by Richard Montague

(The first 9 articles are reprinted in Formal Philosophy, Selected Papers of
Richard Montague, edited and with an introduction by R.H. Thomason,
Yale Univ. Press, New Haven 1974. All page references will be to this volume.
The dates listed are not the dates of publication but of when Montague first
presented each paper to a philosophical aUdience.)
Montague,R. [I] Pragmatics, 1964.
- [2] Pragmatics and intensional Logic, 1967.
- [3] On the Nature of Certain Philosophical Entities, 1967.
- [4] English as a Formal Language, 1968.
154 N. Cocchiarella

- [5] Universal Grammar, 1969.

- [6] The Proper Treatment o/Quantification in Ordinary English, 1970.
- [7] Syntactical Treatments of Modality, 1963.
- and Kalish D. [8] That, 1959.
- and Kaplan D. [9] A Paradox Regained, 1960.
- and Kalish D. [10] Logic: Techniques of Fonnal Reasoning. New York:
Harco urt, Brace and World, 1964.

Works by others

Carnap, R. [12] Introduction to Semantics. Cambridge: Harvard University

Press, 1942.
- [13] Replies. In P.A. Schlipp, Ed., The Philosophy 0/ Rudolf Carnap,
p. 125. La Salle, Ill.: Open Court, 1963.
Cocchiarella, N. [15] Sortals, Natural Kinds and Re-Identification. Logique
et Ana(yse (1977): 85-120.
Gallin, D. [11] Intensional and Higher-Order Modal Logic. Amsterdam:
North-Holland, 1975.
Kamp,H. [14] Formal Properties of 'Now'. Theona 37 (1971): 227-273.
Partee, B.H., Ed. [I6] Montague Grammar. New York: Academic Press, 1976.
[17] Montague Grammar and Transformational Grammar. Linguistic
Inquiry VI (1975): 203-300.
- [18] Opacity and Scope. In M.K. Munitz and P.K. Unger, Eds.,Semantics
and Philosophy, pp. 81-101. New York: New York University Press,
1974.
Constructing a pragmatic foundation for
semantics*

F. KAMBARTEL and H.l. SCHNEIDER

Universitiit Konstanz

We will present here an approach to a philosophical theory of

language that might be called 'pragmatic in a radical sense'. It can
be characterized by putting it into the context of two schools of
thought: On the one hand there is the broad pragmatic tradition,
represented by authors such as G.H. Mead, Ch.S. Peirce, L. Witt-
genstein, J.L. Austin, and J. Searle. On the other hand there is the
school of 'constructive' philosophy, initiated by P. Lorenzen and
W. KamIah, to which the authors of this essay have some affinity.
We will first comment on some issues related to the broader
tradition; the constructive approach will be sketched later, when
some details of the theory of symbolic acts, elementary sentences,
and the logic of argumentation will be discussed (sections 4-8).

I. THE PRAGMATIC TRADITION AND THE RELATION BETWEEN

PRAGMATICS, SEMANTICS, AND SYNTAX

The pragmatic tradition, roughly characterized by the names given

above, has produced some basic and important insights which
can be summarized as follows. To understand language, one
should not look at it primarily as a means to express inner private
thoughts and ideas or to represent certain 'entities'. 1 Such an
approach would falsely presuppose that meanings are language-
independent entities. Instead, language should be looked at as a
social activity. This means that in a first step the actions con-
*The whole essay has been thoroughly discussed by both authors, we do not
however agree on every detail. So the reader may note that in a strict sense
Kambartel is responsible for sections 2,5,7, and 8, and Schneider for 1,3,4,
and 6.

Contemporary philosophy. A new survey. Vol. 1, pp. 155-178.

@1981, Martinus Nijho!! Publishers, The Hague/Boston/London.
156 F. Kambartel and HJ. Schneider

stituted by language use should be studied. This will make it

possible to understand, in a second step, certain mental acts such
as having ideas and wanting to express a certain meaning. So
'meaning' becomes a derived category as compared to 'use'; the
field of pragmatics is taken to be more basic than that of seman-
tics. This reversal of the traditional order was an important aspect
of G.H. Mead's program. 2 eh.S. Peirce must be credited for dis-
covering the speech act 3 and developing a semantic theory that comes
close to Wittgenstein's ([7], I, p. 43) later philosophy and his wel1-
known fonnula 'The meaning of a word is its use in the language'.
Basically, we think, this pragmatic understanding of language
meaning can be taken as the accepted view as far as the meanings
of words are concerned. Its counterpart, 'realistic' semantics in
the ontological sense, which takes concepts or attributes as
peculiar entities, existing independently of language, is not promi-
nent any longer (although some people, especially linguists and
logicians, still seem to be attracted by a realist way of speaking; to
this we will address ourselves later). But how can we account
for the meaning of phrases and sentences? If we accept Frege's
principle that the meaning of a sentence is constituted by two
things, namely its structure and the meanings of its constituents,
and if (following the pragmatic tradition) it is the use of a word
which makes its meaning, should it not be possible, even manda-
tory, to develop a use-theory of semantic structure? If we reject
ontological realism for word-meanings we have to do the same for
sentence-meanings. For it would be inconsistent indeed if we were
to deny an ontological realm of language-independent meanings
but were to go on to state that the semantically relevant structure
of language represents in some way the structure of certain mean-
ings, the very existence of which we have just denied.
This is the point in speaking of the approach outlined here as
one that is pragmatic 'in a radical sense': It calls for a pragmatic
understanding not only of word-meanings but also of language-
structure. In this it differs from other views, e.g., from Searle's
speech act theory. With some of Searle's ([7], p. 16) statements
we are fully in agreement, e.g., when he says 'The unit of linguistic
communication is not, as has generally been supposed, the symbol,
word or sentence, or even the token of the symbol, word or sen-
tence, but rather the production or issuance of the symbol or
word or sentence in the performance of the speech act.' But when
 Constrncting a pragmatic foundation for semantics 157

it comes to language-structure Searle adopts a Chomskyan view,4

and he fails to explain how this can be harmonized with state-
ments like the one just quoted. For Chomsky developed his syntax
independent from semantic and pragmatic considerations. S
This touches on a point that is of some importance for the view
developed here, namely the relation between syntax, semantics,
and pragmatics. In linguistics, syntax has for a long time been
taken as a starting point, and it has been taken for granted that it
can be developed without recourse to semantics. The theory of
language meaning consequently has been considered to be some-
thing additional to syntax. Semantics, like phonology, was taken
to be 'interpretative'; i.e., it was expected to relate certain 'mean-
ings' to an independently generated syntactic structure (cf. Katz
[3], p. 367). Pragmatics then was seen as just another additional
component that would take account of certain phenomena of
language use. So the use of a sentence was thought of in analogy
to the use of a natural object, e.g., a stone: the object used (the
stone) can be described independent of its possible uses.
The 'radically' pragmatic approach reverses this order. It takes
pragmatics as fundamental and construes it as a theory that will
help us to understand language as a system of complex actions. On
this view, semantics will largely coincide with pragmatics, except
that for linguists it might be useful to label as 'semantics' the field
of meaning-relations (like synonymity, homonymity, etc.).6
'Syntax' in a strict sense would then cover the field of those
aspects of language-structure that cannot be understood seman-
tically. These autonomous syntactic aspects of language must of
course be taken into consideration. 7 The pragmatic view certainly
does not deny them, but it holds that language-structure should
not be taken as a natural phenomenon that has to be described
in a detached way like, e.g., the structure of the atom. Instead,
language, as an elaborate system of actions, should in the first
place be understood; its structural aspects should be considered
under the perspective of how they make certain acts possible.

2. CRITIQUE OF TRUTH-CONDITIONAL SEMANTICS

In a radically pragmatic understanding, language structures are

reconstructed throughout as parts of a rational practice rather
158 F. Kambartel and HJ. Schneider

than as objects. Such a reconstruction cannot draw on those con-

ceptions of language and logic that at present are dominant in
formal logic and linguistics. These are based on the so-called
'semantic point of view' which, in its modern form, goes back to
A. Tarski. Here one has to take the word 'semantic' not just in
the general Greek sense, indicating only that the meaning of
language expressions is treated. Rather, the word 'semantic'
refers to a quite peculiar theory of meaning. This theory assumes
that language consists of special (structured) objects, which
become meaningful by their being related to certain other objects
in the world. Among these other objects we have to imagine such
abstract objects as functions, especially truth functions. These
meaning relations are introduced and analysed by using another
language (or level of language) which is usually called the 'meta-
language'. In the metalanguage, so we are informed, we can speak
about the object language, i.e., the language which is the 'object'
of our considerations.
As an illustration consider the following example. In logic
books written from a semantic point of view, one can find truth-
conditional definitions of the universal quantifier of the following
kind: 'Axa(x), is by definition true iff for all 'admitted' substi-
tutions xln the resultinga(n) is true. - More sophisticated versions
using the possible-worlds concept go roughly like this: A possible
world w belongs to the meaning of a universal sentence 'J\ xa(x),
iff for all x: w is an element of the meaning of a(x).
Strictly speaking, from the semantic viewpoint we characterize
with such definitions abstract objects like sets or functions and
coordinate them as meanings with certain other objects, called
language expressions or symbols. - Quite obviously this type of
explaining universal quantification has serious defects: Besides
the unjustified ontological framework it already uses expressions
like 'for all x'. Therefore it is not helpful for learning or under-
standing universal quantification, contrary to what we would
expect from logic books. This defect is not avoided, as most
logicians since Tarski think, by just taking 'Ax' as a symbol of
the object language and 'for all x' as an expression of English,
taken as a metalanguage to the language of formal logic. For if one
wants to understand the use of universal quantifiers, it would
be a petitio principii to simply use metalingual translations of
 Constructing a pragmatic foundation for semantics 159

(equivalents to) those terms of the object language that we want

to introduce. Also it would be a (more indirect) petitio principii,
if such a translation or equivalent is used to build up the meta-
lingual level. Unfortunately methodical circles of this kind are
characteristic of the semantic point of view. Not taking them
seriously means giving up the goal of understanding of what we do
when we use language, especially logical expressions and symbols.
So a semantical analysis of the Tarski-type does not provide us
with a rational understanding of meaning. Neither of the lexical or
of the categorical component of meaning. Semantics in this sense
may give us a very general description of language structures, a
description which works only on the presupposition that we have
already acquired all the relevant lexical and categorical com-
petence. Hence this sort of semantics has nothing to do with the
intensional part of language analysis - if we still want to use the
words 'Sinn' and 'in tension' in the Frege-Carnap-tradition.
Another point is worth mentioning. Treating language and the
realm of meanings as a set of objects linked by functional relations
often has dangerous consequences for the attempt to establish a
rational practice. For it takes us away from the obvious fact, that
language is a human activity which can be reasonable or irrational,
understandable or confused in a pragmatic sense. As one can wit-
ness in science and in the humanities, the semantic point of view
supports descriptive attitudes to one's own ordinary and
theoretical language, as well as the refusal to understand and
justify these languages step by step.

3. PRAGMATICS FROM A 'SEMANTIC POINT OF VIEW'

For a further clarification of the role of pragmatics in philo-

sophy of language is outlined above, it might be helpful to contrast
it with an approach that tries to incorporate the pragmatic aspects
of language into a semantic theory of the Tarski-type. The basic
idea is the following. For a semantic treatment of what are known
as 'indexical expressions' (examples are '1', 'you', 'here', 'now',
etc.), it is necessary to have among the 'input' for the semantical
rules not only the sentence under consideration (e.g., 'I am
hungry') but also a so-called 'point of reference', i.e., in our case
160 F. Kambartel and HJ. Schneider

an information about the speaker. When both are given, a semantic

rule can be applied that says something like 'the expression "I"
refers to the speaker of the utterance', and a semantic interpre-
tation of the sentence can be given in the usual way, i.e., the
sentence can be reformulated in an unambiguous and context-
independent way (cf. Montague [11]).
The idea of a semantic treatment of pragmatics tries to general-
ize from this case and makes the following consideration: Should
it not be possible to treat all pragmatic aspects of language in such
a way that additional information about speaker, hearer, situation
of utterance, etc. is given in an enriched 'point of reference', and
that, corresponding to this information, additional rules are incor-
porated into the semantic component of a grammar, so that all
possible pairs of sentence plus point of reference can be unam-
biguously related to a corresponding semantic interpretation?
We think that it is important to see that an exhaustive treat-
ment of pragmatics along these lines will not be possible, -
important, that is, for a correct appreciation of the fundamentally
pragmatic character of language as well as for an understanding
of the motives behind the approach we ourselves propose to
follow. What would be necessary to realize this program of a
semantic treatment of pragmatics? All possibly relevant features of
all possible situations of language-use would have to be listed and
categorized, so that complete descriptions of such situations could
be given. In addition, a number of rules would have to be formu-
lated that would allow us to specify exhaustively for each pair of
utterance plus situation exactly one unambiguous reformulation.
To count as a semantic treatment in the spirit of the Tarski
school, no vagueness should be allowed. At no point should an
intuitive understanding (e.g., of the motives of the speakers or
their history) enter into the process of semantic interpretation,
which must be thought of as a kind of mathematical calculation in
which argument-places of functions are filled in in order to arrive
at values of the functions at these points.
For two reasons this is an impossible task to fulfill: Firstly, we
have no metalanguage at our command that would be suited to
formulate objectively and exhaustively the rules and the features
of the 'points of reference'; and secondly, all natural languages (in
constrast to invented, small-range sign languages) show a charac-
 Constructing a pragmatic foundation for semantics 161

teristic pragmatic openness that contrasts sharply with the closed

character of language that is presupposed in the above model.
These are two philosophical points about the nature of language
that we take to be of great importance. What would it mean to
have a metalanguage of the kind required? It has to be one with
the help of which we could give an exhaustive description of any
situation of language use, without knowing anything in advance
about 'relevant' or 'irrelevant' ~spects; we just describe every single
feature of the situation. But this means that not only is a linguistic
assumption about the possibility of a certain language made, but
also an assumption about the world and our possibility to describe
it. Does it make sense to speak of anything like an absolutely
complete description of a situation? To put it objectively: Is our
world a 'Tractatus-world' in the sense described in detail by von
WrightS? We think it is not, and it does not make sense to speak
in an abstract or absolute sense ofa 'complete description'.
So if the world is no 'Tractatus-world' and if, consequently, we
have no metalanguage of Wittgenstein's 'ideal' kind, then we
would have to use our natural language for the formulation of the
'points of reference', i.e., for the description of the situations of
language-use. Any such description however must necessarily be
incomplete and it must entail a specific understanding of the situ-
ation, i.e.: it has to discriminate between its relevant and its irrele-
vant aspects. But what does 'relevant' mean? We think that here is
the point where the irreducible pragmatic dimension comes in
again: Only an understanding of the utterances as actions allows us
to select the relevant features of the situation and to choose (and
even formulate) the rules that lead to a correct disambiguation.
Thus on the level of the metalanguage exactly the same process of
intuitive, competence~uided understanding cannot be avoided
which one had hoped to eliminate by a complete functional
analysis.
Secondly, even if we had a Tractatus-world and an appropriate
'ideal' metalanguage - a hardly meaningful speculation -, we
would find it impossible to formulate for any natural language a
comprehensive system of features and rules in the sense considered
above. It is an important feature of language that its use will
always be ahead of any set of pragmatic rules. Natural languages
are historically developing systems of actions and characteristically
are not closed like, e.g., a system of traffic signals.
162 F. Kambartel and H.J. Schneider

4. CONSTRUCTIVE PHILOSOPHY AND THE CONSTRUCTION OF

LANGUAGE AS A SYSTEM OF ACTIONS

Constructive Philosophy9 has been developed in close contact

with constructive mathematics (cf. Lorenzen (14]), a certain
position in the discussion about the foundations of mathematics.
One characteristic feature is its insistence on truth and justifi-
cation where other positions speak of axioms and formal deri-
vations. The constructivist tries to reconstruct mathematics as
a system of meaningful operations with symbols, where the simple
operations (e.g. counting) make possible the more complex ones
(like addition, which in turn makes possible multiplication, and so
on). This insight, that a complex system of human actions can be
understood if we can give a reconstruction that shows how one
action makes possible another, has turned out to be relevant for
the philosophy oflanguage. Language, too, is a complex system of
human actions, and at certain points it is obvious that there is an
order in its elements that is comparable to the order among mathe-
matical operations. These are cases, where one action makes
possible another so that the competence to do the second pre-
supposes a competence to do the first. One obvious example is
that of assertions vs. questions: The action of asking a question
presupposes the competence to make an assertion. Another even
simpler example is that of the logical connectives: To use any
one of them presupposes the competence to make simple state-
ments.
So the question can be asked: Can we understand the semantic
structure of language by reconstructing it as a system of actions
in such a way that it becomes apparent how one kind of linguistic
act makes possible another, which in turn makes possible a third,
and so on. As has been pointed out above, the construction has to
proceed in such a way that no ontologicai claims, open or hidden
in an unjustified use of logical or linguistic categories, are made.
On the contrary, we think that basic categorical distinctions are to
be explained with recourse to those actions by which we make the
distinctions.
The following paragraph will offer some general considerations
about symbolic acts; then a pragmatic interpretation of the
structure of elementary sentences is discussed; and finally an
 Constructing a pragmatic foundation for semantics 163

argumentative - as opposed to game-theoretical - interpretation

of the logical particles will be given.

S. SYMBOLIC ACTS

In order to clarify the notion of symbolic acts, a few words on

the concept of an action in general will be helpful. Acts are not
natural events. As Wittgenstein ([15], I, p. 612) has put it: 'I
should not say of the movement of my ann, for example: it comes
when it comes, etc .... And this is the region in which we say
significantly that a thing doesn't simply happen to us, but that we
do it. "I don't need to wait for my ann to go up - I can raise it".'
To perfonn an act, then, means to exert an ability which we
have acquired, as a rule by learning with a teacher or just by trying
again and again by ourselves. Therefore, being able to act in a
certain way means to be able to produce more than one single
occurrence. That means that it will be necessary to have at one's
disposal a certain type of occurrence, at least under suitable
circumstances, if one wants to be independent from chance and
nature. This leads us to the distinction between act schemes and
concrete acts as actualisations of a certain act scheme 10 (W. Kam-
Iah; von Wright: generic acts/individual acts, Goldman: act types/
act tokens).
It provides no difficulty to understand the intentionality of
acts on the basis of this distinction: The relevant situation is that
someone presupposes some occurrence to be a (concrete) act. Ask-
ing for the intention then is equivalent to the question of what
the agent is doing, and that is asking for the act-scheme he is
actualizing.
There may be several intentions to one 'act occurrence': Some-
one operates a switch, thereby turns on the light and thereby gives
some information (signals something).
In everyday English (and similarly in German) the word
'intention' can be used for aims too. Our definition of intention
means that we distinguish between, e.g., the act of running which
someone is actualizing, and the psychosomatic condition he wants
to preserve or attain by regular running, which we propose to
call the aim of his performance.
164 F. Kambartel and HI Schneider

There are acts which we perform (actualize) by performing

(actualizing) other acts, like putting on the light by operating a
switch. Speech acts, as we shall see, are of this kind. We propose
to call an act a mediated by other acts if and only if a, by defi-
nition, can be actualized by actualizations of one or several of
these other acts. We may speak of mediated acts (absolutely) if
and only if an act can be performed in a mediated way only.
Let us now turn directly to those mediated acts which consti-
tute language and other human institutions. We will call them sym-
bolic acts. Symbolic acts serve the purpose of agreeing on the situ-
ation, in which we are, and of cooperating (pragmatically) by
influencing other people's actions. They, in a sense, make rational
life possible. We constitute symbolic acts by giving a special (sym-
bolic) use to other acts. This is done by agreements which restrict
(explicitly or tacitly) the actualization of these acts to certain
situations or lay down for them certain consequences in the course
of our actions. Think, e.g., of a certain phonetic act 'come' (or
rather [k/\ m]) which by agreement has to be understood as a
specific demand. A person who has uttered a demand must - by
the sort of agreements which are characteristic for demands - be
prepared to see his demand fulfilled by the addressee (no sanctions
however are implied, because demands are not necessarily
commands).
The performance of symbolic acts is built on the possibility of
following rules. However we should not confuse symbolic action
with rule-following, e.g. a rule that an act a is permitted only in
certain situations or is required in certain other situations. A rule
R(a) just concerns a regular way of doing or forbearing a. That is:
we have a different rule, if we regulate a different a' in the same
way R. Drinking tea every morning is different from drinking
coffee every morning. But in the case of a symbolic act the a
underlying the regulation R does not really matter: it is
methodically arbitrary, though there may be historical and
mnemotechnical reasons for a particular choice. Any a' about
which we make the same agreement (R) as about a, in principle
will be good for the same symbolic intention, though in fact
naturally we do not and cannot make all possible agreements.
Thus actualizing the symbolic act belonging to the rule or rule
system R means to perform one of the acts, which by agreement
 Constructing a pragmatic foundation for semantics 165

are given the same symbolic use. This illustrates the particular way
in which agreements and mere conventions are combined in sym-
bolic acts.
As is generally the case with mediating acts, we follow at least
two intentions in the performance of a symbolic act, the direct
one of the mediating act and the indirect one of the symbolic
act. We shall say that the direct intention 'carries' the symbolic
one. Phonetic acts mediating symbolic acts may be called speech
acts (in the narrow sense). In the same way we may understand
writing acts, the carrying intentions then consisting in the pro-
duction of figures (spatial forms). Speech acts and writing acts
then form the speech acts in the broad (usual) sense of the term.
- With this we have laid the foundation for a pragmatic recon-
struction of 'intensional meaning': we propose to define as
intensional the symbolic intention of speech acts in the broad
sense.
The defining agreement constitutes a symbolic act s. Naturally
we might in addition to performing s also follow rules when we
actualize s, which would then be called regulative rules.
A symbolic act s has informative meaning with reference to the
conditions which by agreement make the performance of s a
correct one, and it has performative meaning in so far as it has
constitutive pragmatic consequences. Obviously mixed cases are
possible.
But does not this approach lead us into serious methodical
problems? Are we not forced to go back to descriptions or articu-
lations 0 f the agreements constituting certain symbolic acts?
Since these descriptions already make use of language, we seem to
get into an infinite regress or into the same circle which we
criticized with respect to the semantic point of view: For it
appears that we have to use a descriptive metalanguage to learn the
acts of a pragmatically understood 'object language'.
However, a closer look at elementary speech situations shows
that in a pragmatic understanding of language there is no such
circle. That is to say, in these situations we can learn an ele-
mentary symbolic act s by exemplification and we can prag-
matically control our understanding of the agreement which is
intended. We just get acquainted with examples and counter-
examples of correct actualizations of s in an appropriate pragmatic
166 F. Kambartel and H]. Schneider

context. And in order to control and correct our understanding we

can - in elementary situations - always return to what we do,
especially what we do in nonsymbolic action. (Take for an
example the discussion of the demand 'come!' above). Therefore
in the elementary case, in order to acquire a symbolic act, we, in
principle at least, need no descriptions using other symbolic acts.

6. COMPLEX SYMBOLIC ACTS AND ELEMENTARY STATEMENTS

We turn now to a more detailed discussion of the possibility of

developing a pragmatic understanding of semantic structure, and
as an example we choose the subject-predicate distinction. The
task is the following: can we describe a way in which the use of a
complex utterance of the form 'x is P' is acquired, which satisfies
the following conditions: (1) The use of the complex utterance-
type rests on the capacity of using a simple one. (2) The step
(or steps) from the simple to the complex can be understood as an
enrichment of the possibilities to act that is (a) pragmatically
meaningful (i.e., it makes sense in a subject's system of nonverbal
and verbal competences to act) and can (b) be imagined to be
acquired step by step in a non-circular way. (3) The description
does not speak of mental acts of such a kind that is systematically
dependent on a semantic terminology. E.g., it will be excluded as a
petitio principii to speak of a mental act of referring to objects (or
actions, or whatever). On the contrary, the specific characteristics
of referring (as contrasted, e.g. to predicating) are to be explained.
Now we take it that the paragraph about symbolic acts has
made it reasonably clear how simple one-word utterances of a
certain type can be acquired along the lines described already by
Wittgenstein ([6J, I, p. 2) with his famous example of the building
site. The example is made up in such a way that the pragmatic
meaningfulness is clear from the beginning: The acquisition of
language takes place in a situation in which people engage in
actions; language in some obvious way is useful for performing
these actions. So we take it that a number of symbolic acts of the
type 'slab', 'red', 'hot' are among the repertoire of the speakers.
How can we now think of a step towards a complex symbolic act
with a SUbject-predicate structure, that satisfies the conditions
given above?
 Constructing a pragmatic foundation for semantics 167

Considering that we are looking for a pragmatically meaningful

expansion, we look for a situation in which the one-word-com-
petence reaches the limits of its usefulness and in which an
additional action of uttering a second word turns out to be prag-
matically meaningful. So we imagine that the word 'hot' has been
introduced as a simple symbolic act and has been used in a number
of ways, e.g. as a warning when hot meals are brought on the
table. Now the limit of its usefulness is reached in a situation that
has (from the point of view of the established practice) to be con-
sidered as non-standard; e.g. when not the meal but the plates are
hot. So even if the word 'hot' has been used in other (non-eating)
contexts, a misunderstanding may be anticipated (or experienced)
by the speaker. So, in order to avoid such misunderstanding, he or
she may invent as a new step that transcends the conventions
established so far the possibility of uttering the two words 'plate
hot'. Of course this utterance may not be understood; there is no
rule yet according to which it was spoken. On the other hand we
all know from travelling in countries with languages we do not
speak, how a single word can be the key to a situation which is
just one step away from being understood pragmatically: the one
word helps us to act correctly. So the invention of the complex
act might be helpful and become a first instance of a new con-
vention.
With this example in mind, we now return to the conditions
given above and ask if they are satisfied. For the first condition
the answer is positive. We have postulated that both parts of the
complex utterance ('plate' and 'hot') are known as separately
performable symbolic acts. The second condition asks for prag-
matic meaningfulness, which is satisfied quite obviously; the com-
plex act helps to avoid a dangerous incident (bumL'1g one's fingers)
in a better way than a simple one could. And such a combination
of two known acts can imagined to be learned, it seems to us, in
one step.
The third condition is satisfied in so far as our description did
not make use of a vocabulary for 'linguistic' mental acts, such as
'to refer to'. We imagined a situation in which a combination of
two words of the same kind is used to cope with a particular type
of situation. It was not necessary to speak of mental acts as con-
ditions that define or make possible certain speech acts.
168 F. Kambartel and fl.J. Schneider

If it is accepted that an extension of the competence to perform

symbolic acts is possible along the lines just sketched, how far
does this take us towards an understanding of the sUbject-predicate
structure of sentences? One point to be observed is that so far
there occurred in our reconstruction only words of one single
category ('predicators'). We have not introduced proper names,
and we did not differentiate kinds of words according to semantic
functions. Instead, we have introduced a step that gives a new
pragmatic function to a word of the old (and only) category of
predicators. In the complex utterance 'plate hot' the word 'plate'
takes over a new role, a fact that is clearly seen when we compare
its role in the complex utterance with the one it has in a one-word
sentence. Utterances of one-word sentences have direct informa-
tive or performative functions, as explained above. To say 'hot' in
an infonnative context, for example, makes sense or is meaningful
in so far as it changes the possibilities to act on the side of the
hearer; mentalistically speaking: when the speaker infonns the
hearer of something he does not know. In the structured utterance
'plate hot' this is indeed the function of the word 'hot'; but the
added element 'plate' does not function in this manner: both
partners have equal access to the symbolic act of uttering 'plate'.
Again in mentalistic terminology: both know there are plates in
front of them. The new function of the word 'plate' (the 'subject')
is to resolve an ambiguity that the utterance of 'hot' alone (the
'predicate') would have. Other acts like that of pointing could be
introduced to fulfill the same fUllction.
If this functional difference, the difference in the use of words
is properly understood, we can differentiate between the roles of
words in sentences on this basis; we now have at our disposal a
(necessary though not sufficient) condition for speaking of
subjects and predicates and of the subject-predicate structure of
sentences.
Now the important point for a pragmatic theory of language is
that on the one hand the reconstruction just sketched is more than
a purely syntactic or forrnal introduction of the vocabulary with
the help of which language structure is described. But on the other
hand it does not rely on ontological assumptions about meanings
or any other platonistic entities. We did not start from the
assumption, that there are ('in the world ') individuals and
 Constructing a pragmatic foundation for semantics 169

attributes and that for this reason we would have to expect (in
language) proper names and predicators and the functions of sub-
ject and predicate. Instead, we are proceeding from a pragmatic
understanding towards a reconstruction of these categories.
It hardly needs mentioning that the considerations just given do
not suffice for a full explication of the subject-predicate dis-
tinction. ll For one thing, more has to be said about the specific
kind of ambiguity that is exhibited by the example discussed
above, because one can think of other types of ambiguity where
an addition of a second expression will constitute another
semantic relation, e.g. attributive modification.
Secondly, the utterance-types described so far are intelligible
only in their pragmatic context, i.e. they are not independent of the
situation of their use. This independence will be reached when the
subject-expression is enriched in such a way that it becomes a
'description' in the usual sense: When more than one additional
predicator is used for disambiguation, the context may not be
needed any more.
After that, proper names can be introduced to replace these
descriptions. In this case, their difference to predicators can
clearly be understood as a functional one, which will not be the
case if they are introduced in the same way as predicators are, i.e.
by the use of examples and counter-examples. Context-independent
utterances are statements, and they are needed to proceed now to
logical connection.

7. ARGUMENTATIVE LOGIC

Elementary statements may by agreement come to carry a further

symbolic intention and thereby be transformed into basic
assertions. The need for elementary assertions arises in situations
where language competence is developed to the degree that action
depends on correctly performed elementary statements, and where
the participants do not agree on the correctness of such a per-
formance. In this case one can either give up common orientation
as provided by elementary statements, or one can try to overcome
private opinions by reaching a new level of transsubjectivity, by
argumentation.
170 F. Kambartel and fl.J. Schneider

With argumentation we mean here, quite simply, all attempts to

settle differences on the basis of previously or newly established
agreements. Thus we may go back to other elementary statements,
which are not at issue, or reproduce the original learning situation
for the words involved. In the fIrst case, e.g. if we have already
agreed upon 'P is Q' and 'Q is R', this might stop further
discussion on 'P is R'. Often it will be helpful (or even necessary)
to remember the aims and activities which originally gave rise to
the terms of a disputed elementary statement. These examples will
be sufficient to indicate what activities we can enter into for the
purpose of justifying an elementary statement by arguments.
Someone who now not only just states something, but asserts
what he is stating, must be prepared to establish by argumentation
a transsubjective agreement that his statement has been made
correctly. In this way statements characteristically come to carry a
further assertive symbolic intention, and then are called assertions.
Assertions are, in our everyday and scientific life, one of the
language institutions, whereby we can rely on others in our
orientations. They thereby make possible a division of labor in
building up a knowledge of our situation. Trivially the reliability
of assertions is undermined if people make assertions without
having the corresponding justifications at hand. Often this is done
with bad intentions, as e.g. in the case of propaganda. Less harm-
ful seems (but only seems) a practice which is widespread in the
universities, namely the practice of substituting mere assumptions
for statements. Making assumptions may be a good thing when
there is no knowledge in sight. But more often than not putting
forward assumptions serves as a means of avoiding the laborious
business of justification.
We shall now consider logically complex assertions and show
that, once we have elementary assertions, it is helpful to proceed
to more complex activities related to the logical words of our
language.
In planning common practice it is often advisable to make
action depend on more than one justified orientation. Thus it may
be that one has to do two things, justify a and justify b, to make
someone else do h. Whoever thinks himself capable of fulfilling
both these justification-tasks, claims that he can do a complex
justification-work, as one might call it. This claim is raised in
 Constructing a pragmatic foundation for semantics 171

English in the form of a complex statement, the conjunction

'a and b', in logical symbols: 'al\b'.
But are we not once again guilty of a petitio principii with this
reconstruction of logical conjunction? Did we not use the word
'and', when we put the relevant complex justification task. as
follows: 'to do two things, justify a and justify b'? - Indeed we
did. But here (in this formulation) the word 'and' is not a logical
connective, but a word which we use to form expressions for com-
plex acts, e.g. in the case of commands. And what it means to
perform an act by performing several (in our case two) other acts
we can learn by means of suitable examples. This practical learning
does not imply that we need to speak about this learning situation
and use, in doing this, logical conjunction. Thus there is no circle
of definition.
In a similar way we obtain logical adjunction 'a or (vel) b' in
symbols: 'avb'. Adjunction serves among other ends, the purpose
of planning an action which is dependent on several possibilities
(alternatives). Again, we first have to establish a pragmatic basis
for logic, that is to say learn a sort of pragmatic complexity, which
we may express as doing one of two things (acts). Knowing this
we know in particular what it means to fulfil one of the (two)
justification-tasks related to assertions a, b. And this new justifi-
cation-claim we may again attach to a new complex statement
'a or b' ('a vb') .
Logical quantifiers may be treated similarly: Asserting 'for all
x: a(x)', in symbols: ';\xa(x)' (universal quantification) means
having a procedure at hand to justify a(n) for an arbitrary sub-
stitution x/no To know a procedure does not necessarily imply
descriptions using universal quantifiers: to know how to get nails
into a concrete wall does not imply understanding sentences like
'all nails are in the wall' or 'for all nails I know how to get them
in to the wall'.
Who asserts 'there is an x, such that: a(x)', in symbols: 'Vxa(x),
(existential quantification) claims to have a substitution x/n at
hand, so that he can give a justification for the corresponding
a(n). Existential quantification in this sense is 'effective': you have
to know how to 'construct' the substitution which is necessary to
defend the existential assertion.
Obviously no logical platonism or axiomatic formalism is
172 F. Kambartel and H.J. Schneider

necessary to understand the meaning of the logical words and

symbols and to judge on logical validity. What one has to do is to
reconstruct their rational pragmatic place in our lives, i.e., one has
to understand them as part of rational action, namely in this case
of argumentation.
The purpose of dividing up justifications into parts, leads us to
logical subjunction <b, if a' ('a ~ b'). In this case the constitutive
claim is: let me have a justification of a, then I shall be able to
construct one for b. This should, for invariance reasons, include
the case that I have an independent justification of a.
As to the negator 'not' ('I '): whoever puts forward a negation
la (not a), claims to have a procedure, by which he can be sure,
that attempts to justify a, will fail. In short: the proponent of a
negation -lQ must be able to refute a. Obviously negations provide
us with an important kind of knowledge, because with them we
can e.g. give up trying to justify the negated a or reflecting on
courses of action presupposing the validity of a. - Again negation
makes use of a pragmatic distinction, namely between succeeding
in and failing at doing something (acting in a certain way).
We are now in a position to settle the question of tertium non
datur: Is a v la valid independently of the content of a, i.e. valid
by taking into account nothing else but the meaning of the logical
words v, I? Obviously not, because there are a's for which we
have neither justifications nor refutations at hand.
The contradiction principle on the other hand holds, i.e.
I (al\ ,a) is logically valid. This is so, because nobody can do both,
justify an assertion a, and at the same time show us that an
attempt to justify a must fail.
Obviously the proposed reconstructions of complex assertions
are related to the dialogical logic of P. Lorenzen and K. Lorenz. 12
The dialogical approach also claims to characterize, by schematic
rules, the logical aspect of concrete argumentation as exemplified
by Plato's dialogues. And yet the short history of dialogical logic
shows that the latter has not yet managed to fully overcome the
distortions caused by the mathematization of logic.
Already the first version in Lorenzen [13] stylizes argu-
mentation as following formal rules in a dialogical <game', analo-
gous to the moves in chess. Lorenz in [27], [29) then has worked
out this game-theoretical reconstruction of formal obligations in
 Constructing a pragmatic foundation for semantics 173

argumentation and of logical validity. He gave precise and

schematically applicable formulations not only for those steps
which depend on the logical particles (particle rules), but also for
the general frame which has to be observed when making any
particle-related moves (frame rules). For all these rules, especially
for frame rules, we can raise the question of justification, the
question, that is, whether or not they are an adequate recon-
struction of logical complexity in concrete argumentation. And
here the adequacy control can lead to a reasonable regulation only
if we already understand the rational place of logically complex
argumentation in our lives. 13
The above proposals for an argumentative use of the logical
words are meant to offer such an understanding. Classically
spoken, they play the role of a semantics in relation to the 'mono-
logical' or 'dialogical' formal systems.
Lorenzen and Lorenz on the other hand transform the ade-
quacy problem of schematical argumentation rules into a com-
pleteness problem which can be treated on the formal level alone
and thus is located again inside mathematical logic. They take the
'dialogical' games themselves as an argumentative semantics for
logical calculi, i.e. they conceive of these games as giving meaning
to the logical words. In that case (after this transformation) one
can naturally discuss whether the derivability of a logical formula
in a calculus is equivalent to the existence of a winning-strategy
in a two-person-game. But the unresolved discussion about the
justification or rationality of the dialogical rules, especially the
frame rules l4 shows that there is still an element of arbitrariness in
dialogical logic, which cannot be removed on the technical level
alone. At the beginning Lorenzen was still aware of the connection
of agonal (,eristic') or cooperative ('dialectical') argumentations
with different schematical regulations. IS Later, for unclear
reasons, the agonal version was formally favoured, and in the dis-
cussion of the frame rules technical aims like e.g. 'simplicity' or
mere plausibility considerations prevail. 16 , 17
174 F. Kambartel and H.J. Schneider

8. CONCLUSION

Let us look back on the systematic course, that is the method,

which we have followed. We may call our considerations and pro-
posals critical reconstructions of an activity, which we all know in
use, namely language. Reconstructions being constructions, let us
dwell on this point for a moment. 'Construction' in our use is a
term of the theory of action. In a (pragmatic) construction we
build up, step by step, systematically connected actions. Pragmatic
constructions are justified, in so far as they lay the basis for a
rational life. To intend justified constructions implies that, besides
the question of how we can aq uire step by step certain acts, we
have to consider the connection between our acts and our aims:
That is to say we have to judge the consequences of performing
those acts, i.e. evaluate the resulting situations. In this sense
justified pragmatic constructions include teleological con-
siderations.
In most cases of pragmatic constructions, we do not enter into
the corresponding actions for the first time. Rather we are already
involved in them in a confused and sometimes seriously distorted
way. Thus in some sense we always know what we are doing, and
in some sense we do not know; i.e. we know how to do certain
things without having a clear awareness either of the pragmatical
construction of this ability, or of its teleological place in our life.
This leads us to misunderstand our activities and sometimes there-
by to become subject to a misuse of our lives without realizing it.
Therefore in those situations it is advisable to work out clear
pragmatic and teleological constructions for the actions we are
involved in, in short: to reconstrnct what we are doing. We think
that this amounts to striving for enlightenment; and obviously
the actual use of language is confused enough to stand in need of
enlightenme nt.

NOTES

1. This is the traditional approach, taken e.g. by Locke in his Essay [1],
Vol. II, p. 3, and, until quite recently, by 1.1. Katz [2], p. 176. Katz has
later given up this view; cf. his [3], p. 38.
 Constructing a pragmatic foundation for semantics 175

2. Mead's theory is sketched in [4] , Part II.

3. This is not always noted, but Peirce is qUite explicit on this subject. Cf.
Peirce [5], 5.546 sq.
4. Searle [7], pp. 30 sq.; he is not perfectly explicit about this point, how-
ever, cf. Searle [7a] for a critique of Chomsky's approach.
5. This is true for Chomsky [8] and again, as it seems, for his later work,
like [9]. At other times he appeared to be ready to give more weight to
the semantic component of a grammar.
6. This approach is favored in Lyons [10] .
7. This has been one of the points of debate between Chomsky and the
proponents of 'Generative Semantics' (Lakoff, McCawley and others).
8. It will be helpful for the reader to consult von Wright's clear discussion
in his [12], pp. 44 sqq.
9. For a broad orientation cf. Lorenzen [16], KamIah and Lorenzen [21],
Lorenzen [15] and [18], Lorenzen and Schwemmer [20], Janich,
Kambartel and Mittelstraill [22], and Kambartel [23] _ For the context
of this essay cf. Kambartel [24], Schneider [26], Lorenz [30] and [32],
Mittelstraill [33] and [34]. For dialogical and argumentative logic see
below, notes 12 and 13.
10. This terminology ('Handlungsschema', 'Aktualisierung eines Handlungs-
schemas') has been introduced by W. KamIah: Cf. KamIah and Lorenzen
[21], II, §2. G.H. von Wright [35] uses a similar distinction between
'generic acts' and 'individ ual acts'.
II. It may be noted that among constructive philosophers the debate is
presently going on. KamIah and Lorenzen [21] have not seen any
problem in the traditional 'xEP'-analysis; Lorenz [30] and [32] is
working on an explanation for it, and Lorenzen [17] and Lorenzen and
Schwemmer [20] have proposed more complex structures for elementary
sentences. Schneider is presently working on an attempt to clarify the
issues, cf. Schneider [26a] .
12. The development of the 'dialogical' foundation of logic is now documen-
ted in the collection P. Lorenzen and K. Lorenz [19]. Cf. also Lorenzen
[16], Lorenz [31].
13. This perspective has been worked out to some degree in C.F. Gethmann
[36], [37] and Kambartel [25]. Already Lorenz in [28] took some
steps in this direction.
14. Cf. e.g. H. Lenk [38] , [39] ; W. Stegmiiller [40], but also the permanent
change in the arguments for the 'frame rules' (and correspondingly their
form) in the developing proposals of Lorenz and Lorenzen.
15. Cf. The final remarks in Lorenzen [13] .
16. Cf. e.g. Lorenzen and Schwemmer [20] p. 68, Lorenz [29] pp. 38,40
(= Lorenzen and Lorenz [19] p. 104 sq., 107).
17. These arguments concern Hintikka's proposals of a dialogical semantics
(cf. e.g. Hintikka [41], [42]) too. Only one should add that this version
of 'dialogical' games, with nature as an omniscient partner, leads us away
from the real conditions of argumentation, a fact that Hintikka himself
acknowledges in [41] p.80 seq.
176 F. Kambartel and H.J. Schneider

BIBLIOGRAPHY

Berk, U. [43] Konstruktive Argumentationstheorie. Stuttgart·Bad Cannstatt

1979.
Chomsky, N. [8] SyntacticStnictures. Den Haag 1957 .
. [9] Deep Structure, Surface Structure, and Semantic Interpretation. In
D.O. Steinberg and L.A. Jakobovits, Eds., Semantics: An Inter-
disciplinary Reader in Philosophy, Linguistics. and Psychology, pp.
183-·216. Cambridge 1971.
Gethmann, C.F. [36] Die Ausdifferenzierung der Logik aus der vorwissen-
schaftlichen Begriindungs· und Rechtfertigungspraxis. Zeitsc}zr~ft fur
allgemeine Wissenschaftstheorie 9 (forthcoming).
- [37] Protoiogik - Untersuchungen zur formalen Pragmatik von Begrun-
dungsdiskursen. Frankfurt 1979.
- [37a] , Ed., Theorie des wissenschaftlichen A rgumentierens. Frankfurt
1980.
Hintikka, J. [41] Logic, Language-Games and Information: Kantian Themes
in the Philosophy of Logic. Oxford 1973.
- [42] Quantifiers vs. Quantification Theory. DiiJiectica 27 (I973): 329-
358.
Janich, P., Kambartel, F., and MittelstraB, J. [22] Wissenschaftstheorie als
Wissenschaftskritik. Frankfurt 1974.
Kambartel, F. [23] Theone und Begrundung - Untersuchungen zum Ph i/o so-
phie- lind Wissenschaftsverstiindnis. Frankfurt 1976 .
.. [24] Symbolic Acts: Remarks on the Foundations of a Pragmatic Theory
of Language. In G. Ryle, Ed., Contemporary Aspects of Philosophy,
pp. 70-85. Oxford 1977.
- [25] Oberlegungen zum pragmatischen und zum argumentativen Funda-
ment der Logik. In K. Lorenz, Ed., Konstruktionen versus Position en,
Vol. 1, pp. 216-228. Berlin 1979.
- [25a] Pragmatische Grundlagen der Semantik. In C.F. Gethmann [37a I,
pp.95-114.
Kamiah, W., and Lorenzen, P. l21 J Logische Propildeutik- Vorschule des
verniinftigen Redens. Mannheim 1967. Pages refer to the second edition
1973.
Katz,J.J. [2] The Philosophy of Language. New York 1966.
- [3] Semantic Theory. New York 1972.
Lenk, H., [38] Kritik der logischen Konstanten. Berlin 1968.
- [39] Philosophische Logikbegriindung und rationaler Kritizismus. In
H. Lenk,Metaiogik und Sprachanalyse, pp. 88-109. Freiburg 1973.
Locke, J. fl] An Essay Concerning Human Understanding, Vol. 2. Ed. A.C.
Fraser. London/New York 1894.
LoreRz, K. [27] Arithmetik und Logik als Spiele. Dissertation Universitat
Kiel, 1961, partly repr. in Lorenzen and Lorenz [19] , pp. 17 -95.
- [28] Die Ethik der Logik. In H.G. Gadamer, Ed., Das Problem der Sprache,
pp. 81·86. Miinchen 1967.
 Constructing a pragmatic foundation for semantics 177

- [29] Dialogspiele als semantische Grundlage von Logikkalkulen. Archiv fur

mathematische Logik und Grundlagenforschung 11 (1968): 32-55,
73-100; repr. in Lorenzen and Lorenz [19] ,pp. 96-162.
- [30] Elemente der Sprachkritik. Frankfurt 1970.
- [31] Rules versus Theorems: A new Approach for Mediation between
Intuitionistic and Two-Valued Logic. Journal of Philosophical Logic 2
(1973): 352-369.
- [32] Sprachtheorie als Tei! einer Handlungstheorie - Ein Beitrag zur
Einfuhrung linguistischer Grundbegriffe. In D. Wunderlich, Ed., Wissen-
schaftstheorie der Linguistik, pp. 250-266. Kronberg 1976.
Lorenzen. P. [13] Logik und Agon. In Atti del XII Congresso Internazinnale
di Filoso[w (Venezia, 12-18 Settembre 1958), Vol. 4, 1960; pp.
187-194. Firenze, repr. in Lorenzen and Lorenz [19], pp. 1-8.
- (14] Differential und Integral - Eine konstruktive Einfuhrung in die
klassische Analysis. Frankfurt 1965.
- [15] Methodisches Denken. Frankfurt 1968.
- [16] Nomlative LOgic and Ethics. Mannheim/Zurich ]969.
- [] 7] Semantisch normierte Orthosprachen. Die wissenschaftliche
Redaktion 7, Mannheim, pp. 117-132 (1972). Reprinted in F. Kam-
bartel, and 1. Mittelstr~, Eds., Zum normativen Fundament der
Wissenschaft. Frankfurt] 973.
- [18] Konstruktive Wissenschaftstheorie, Frankfurt 1974.
- and Lorenz, K. [19] Dialogische Logik. Wissenschaftliche Buchgesellschaft
Dannstadt,1978.
- and Schwemmer, O. [20] Konstruktive Logik. Ethik und Wissenschafts-
theorie. Mannheim 1973; pages refer to the second edition 1975.
Lyons, 1. [10] Introduction to Theoretical Linguistics. Cambridge 1968.
Mead, G.H. [4] Mind. Self, and Society from the Standpoint of a Social
Behaviorist. Ed. C.W. Morris. Chicago/London, 14. impr., 1967. Mittel-
str~, 1. [33] Die Pradikation und die Wiederkehr des Gleichen. In
H-G. Gadamer, Ed., Das Problem der Sprache. Munchen, 1967.
Reprinted in the collection cited under Mittelstra~ [34], pp. 145-157.
English version in Ratio 10 (1968): 78-87 (Predication and Recurrence
of same).
- [34] Das normative Fundament der Sprache. In Mittelstr~, 1. Die
M6glichkeit von Wissenschaft, pp. 158-205. Frankfurt 1974.
Montague, R. [11] Pragmatics and Intensional Logic. Synthese 22 (1970):
68-94.
Peirce, Ch.S. [5] Collected Papers, Vols. 1-4. Eds. Ch. Hartshorne and P. Weiss.
Cambridge, Mass., 1931-35.
Schneider, H.J. [26] Pragmatik als Basis von Semantik und Syntax. Frankfurt
1975.
- [26aJ On Language Use and Language Structure. Kodikas 2 (1980): 77-85.
Searle, l.R. [7] Speech Acts: An Essay in the Philosophy of Language.
Cambridge 1969.
178 F. Kambartel and H.J. Schneider

- (7a] The Rules of the Language Game. Times Literary Supplement, Sept.
10,1976.
Stegmuller, W. [40] Remarks on the Completeness of Logical Systems
relative to the Validity-Concept of P. Lorenzen and K. Lorenz.
Notre Dame Journal of Fomzal Logic 5 (I 964): 81-112.
Wittgenstein, L. [6] Philosophische Untersuchungen - Philosophical Investi-
gations. Eds. G. Anscombe and R. Rhees. Oxford 1953.
von Wright, G.H. [12] Explanation and Understanding. London 1971.
- [35] Nomz and Action. London 1963.
'Logique hermeneutique'?
PAUL RICOEUR
Universite de Paris

Peut-on parler d'une logique hermeneutique? A vrai dire, Ie terme

meme n 'a ete employe, a rna connaissance, que par Hans Lipps dans
ses Untersuchungen zu einer hermeneutischen Logik, Frankfurt,
1938. 1 Ce fut une entreprise interrompue, d'abord en raison de la
mort prematuree de Lipps, mais surtout parce que la coupure que
represente la radicalisation de I'hermeneutique par Heidegger I'a
fait apparaitre apres coup comme trop liee a la Lebensphilosophie
d'un Georg Misch, de laquelle precisement I'hermeneutique
philosophique s'eloignait de maniere decisive. Et pourtant un
certain retour a Lipps se laisse remarquer, qui n'est pas sans
rapport avec la tendance generale de I'hermeneutique post-heideg-
gerienne a reflechir sur les conditions de possibilite de son propre
discours et par la meme a se situer par rapport aux preoccupations
logiques qui se font jour dans les autres courants philosophiques
contemporains. Mon essai sera precisement consacre, dans les
paragraphes 3 et 4, a ce double effort de la philosophie herme-
neutique de ces dernieres annees pour reflechir son propre statut
epistemologique; mais auparavant on rappellera la position philo-
sophique du probleme creee par Heidegger et developpee par
Gadamer.

I. RADlCALlSA TlON ANTI-LOGIQUE?

A premiere vue la radicalisation de I'hermeneutique par Heidegger

marque un eloignement des questions epistemologiques qui avaient
domine la periode diltheyenne. 2 Avec Dilthey il etait encore
possible de situer l'hermeneutique au plan de la discussion episte-

Contemporary philosophy. A new survey. Vol. 1. pp. 179-223.

©1981. Martinus NijhoffPublishers. The Hague/Boston/London.
180 P. Ricoeur

mologique, sans toutefois pouvoir l'y enfermer. L'enjeu etait

triple. II s'agissait d'abord de defendre l'autonomie des sciences de
]'esprit par rapport aux sciences de la nature, ensuite d 'etablir la
difference entre leur procede principal, la comprehension, et
l'explication en usage dans les sciences de la nature, -- enfin de
fonder cette difference epistemologique dans une propriete
fondamen tale de la vie spirituelle, a savoir Ie pouvoir du sujet
de se transporter dans une vie psychologique etrangere. L'her-
meneutique de Dilthey s'inscrivait done indubitablement dans Ie
debat epistemologique dans la mesure ou c'etait la scientificite
meme des sciences hermeneutiques qui etait en jeu au triple plan
de la determination de leur champ, de l'etablissement de leur
procede principal et du fondement demier de leur specificite.
Cette demiere question marquait Ie passage de l'epistemologie, au
sens precis du mot, a une interrogation transcendantale de type
kantien qui subordonnait l'enquete methodoIogique a l'investi-
gation des conditions de possibilite.
Or, c'est precisement avec ce mode d'interrogation que la
philosophie henneneutique de Heidegger paraft rompre. Et ceci
d'une triple fa<;:on egalement. Ce n'est pas d'abord a justifier la
specificite des sciences de l'esprit qU'elle s'interesse, du moins
directement. Son interrogation est ontologique: quel etre sommes-
nous, nous qui nous posons la question de l'etre? Des !'intro-
duction a Sein und Zeit, la radicalite non-epistemologique de la
question est claire. Si ron s'interroge d'abord sur l'etre que nous
sommes -- Ie Dasein, c'est parce qu'il est Ie lieu privilegie de cette
question fondamentale. C'est a titre 'preparatoire' seulement que
l'analytique du Dasein occupe l'avant-scene. La question du sens -
question hermeneutique par excellence - est celIe du sens de
l'etre. Ainsi une question d 'origine aristotelicienne supplante-t-elle
une question d'origine kantienne. Deuxieme ligne de rupture: la
question de la comprehension - impliquee dans la question du
sens de l'etre - n 'est pas elle-meme une question epistemologique,
du moins a titre primaire. Comprendre est un trait distinctif de
l'etre que nous sommes, une propriete du Dasein en tant qu'etre
dans Ie monde. L'interpretation n'est que Ie developpement de la
comprehension, dans la mesure ou comprendre quelque chose en
tant que quelque chose c'est deja l'interpreter. Cette interpretation
a son tour s'articule dans un discours qui determine et explicite les
 'Logique hermeneutique 'I 181

articulations d 'une situation et d 'une comprehension qui on t

d'abord ete liees a un niveau plus fondamental que Ie discours. La
pretention de commencer par Ie discours exprime en propositions
et done de s'etablir au sein d 'un logos apophantique, est la plus
fondamen tale mecomprehension con tre laquelle s'eleve l'herme-
neutique. Le premier milieu d'articulation, c'est l'etre-au-monde
lui-meme et c'est l'enchafnement entre situation-comprehension-
interpretation et discours qui fonde toute investigation au niveau
simplement propositionnel. Troisieme ligne de rupture: ce n'est
pas. en terme de conditions de possibilite que se pose ultimement Ie
probleme de la comprehension. Dans une perspective kantienne,
interroger sur les conditions de possibilite, c'est se rHerer it un
sujet epistemologique porteur des categories qui n~glent I'objec-
tivite de I'objet. Or ce que l'analytique du Dasein met a nu, ce
n'est pas un sujet de connaissance, mais un etre jete-projetant.
Avant la relation de connaissancc entre sujet et ohjet, il y a I'impli-
cation de la comprehension dans la structure ontologique de l'etre-
jett; et de I'anticipation. Cette modalite de !'etre est mieux
designee comme souci que comme connaissance. C'est dans Ie
souci qu'est incluse a titre de pre-comprehension ontologique la
question de I'etre dont on vient de dire qu'eUe precede l'inves-
tigation dans les sciences de l'esprit.
Cette subordination de I'epistemologie a l'ontologie atteint son
poin t decisif lorsq ue la me thodologie des sciences historiq ues se
voit supplantee par une interrogation sur l'historicite du Dasein:
avant que l'histoire ait un objet et une methode, nous sommes
historiques de part en part.
Telle est - tres grossierement csquissee -- la motivation anti-
epistemologique et anti-Iogique de l'hermeneutique radicalisee de
Heidegger dans Sein und Zeit.
Et pourtant, dans la phase post-heideggerienne qui nous
interesse ici, cette hermeneutique radicalisee ne devait pas echap-
per a une interrogation de type epistemologique, des lors que I'on
peut et que l'on doit enquerir sur la condition de possibilite propre
au discours de l'hermeneutique et Ie situer par rapport a celui des
philosophies a pretention logique et epistemologique, principale-
ment celui de la philosophie analytique contemporaine.
II faut avouer que cette reflexion de second degre n'est que
faiblemen t esq uissee dans Sein und Zeit. On en trouve neanmoins
182 P. Ricoeur

les premiers lineaments dans tous les passages ou Ie type de verite

hermeneutique est en jeu. Ainsi lit-on au paragraphe 7 que
l'hermeneutique est une espece de phenomenologie; or qui dit
phenomenologie dit logos de ce qui se montre. II faut donc situer
Ie logos hermeneutique par rapport au logos apophantique, celui
de Ia Iogique propositionnelle. Si Ie logos apophantique s'articule
a titre primaire dans la proposition qui dit quelque chose en tant
que quelque chose, quel rapport existe-t-il entre Ie 'en tant que'
apophantique et Ie 'en tant que' hermeneutique qui, comme on
l'a dit, constitue Ie moment de l'interpretation, lequel developpe et
articule Ie moment de la comprehension (Sein und Zeit paragraphe
32, 33)? Un nouveau concept de verite est ainsi suscite qui n'est
plus defini par des caracteres de la proposition, mais par la
capacite devoilante impliquee dans Ie rapport entre comprehension
et situation. Mais du meme coup la philosophie hermeneutique
eleve une pretention a Ia verite qui doit se mesurer avec Ia pre-
tention a la verite de type apophantique, c'est-a-dire propo-
sitionnel.
D'une autre far;on la philosophie hermeneutique devait s'inter-
roger sur elle-meme sur un mode epistemologique. La caracter-
isation du Dasein met en jeu un type de discours appele analytique
existentiale. Par cette expression deux choses sont impliquees.
D'une part, loin d 'etre un retour a l'ineffable et a I'irrationnel
comme on l'a quelquefois dit a tort, la philosophie hermeneutique
est une analytique, en ce qu'elle procede par distinction, deter-
mination, relation (d'ou la frequence du terme structure dans Sein
und Zeit et Ie caractere fortemen t didactiq ue de la construction de
l'ouvrage). D'autre part, cette analytique est une analytique
existentiale, en tant qu'elle articule des quasi-categories - etre-
dans-Ie-monde, situation, comprehension, etc. - qui sont au
Dasein ce que sont les categories pour les choses. Cette distinction
en tre existentiaux et categories est certes appuyee a une dis-
tinction ontologique entre plusieurs modalites d 'etre, l'etre que
nous sommes - Ie Dasein, qui seul ex-siste, et les 'choses' qui sont
soit des choses subsistantes (vorhandene), soit a portee de main,
a titre d'utensiles (zuhandene). Mais cette distinction entre modes
d'etre vient precisement au langage et au discours comme
difference categoriale. De cette fac;on, l'henneneutique ne peut se
soustraire a la question kantienne des conditions de possibilite de
son propre discours.
 'Logique hermeneutique '? 183

Elle y est reconduite d 'une troisieme maniere: aussi radicale que

soit l'interrogation henneneutique issue de Heidegger, elle ne peut
eliminer Ie fait que l'hermeneutique est nee de la problematique
des sciences de l'esprit. C'est meme de cette problematique qu'elle
est la radicalisation. Du meme coup la confrontation avec I'herme-
neutique de Dilthey est une partie integrante de sa propre entre-
prise. Sa pretention a fonder les sciences hermeneutiques demeure
la composante epistemologique irrecusable de son propre discours,
au moins au troisieme sens que nous avons attache plus haut a
l'epistemologie de Dilthey. Aussi bien de nombreuses analyses de
Sein und Zeit pointent dans cette direction sans toutefois s'y
engager completement. Ainsi, la derivation de la problematique
des sciences historiques a partir de I'ontologie de l'historicite est
esquissee dans les analyses qui montrent de queUe maniere l'his-
toricite, originairement orientee vers l'avenir, se retoume vers Ie
passe par la mediation de la 'repetition', et comment on passe du
proces de temporalisation (Zeitlichkeit), inherent a la dialectique
de la situation et de la comprehension, a l"etre-dans Ie temps'
(lnnerzeitlichkeit) qui implique la me sure et la reference a des
echeances publiques. Mais on ne va pas plus loin que cette deri-
vation des conditions les plus lointaines de la connaissance his-
torique. On ne construit pas conceptuellement les categories qui
permettraient de tenir cette derivation pour autre chose que
l'allegation d 'une anteriorite ontologique de l'historicite par
rapport a I'histoire-science. Ou, si on prefere, on ne voit pas en
quoi les conditions de possibilite de l'enquete ontologique sont
aussi conditions de possibilite de la connaissance historique
objective. II semble qu'avec Heidegger on fasse bien Ie mouvement
de remontee au fondement, mais mal la 'seconde navigation' -
pour parler comme Platon -, celIe qui ramenerait vers l'episte-
mologie des sciences humaines.
II faut en dire autant d'une autre contribution importante de
Sein und Zeit a l'epistemologie des sciences humaines. Elle con-
cerne moins Ie caractere historique de ces sciences que leur statut
textuel. L'hermeneutique ancienne n'avait cesse d'argumenter au
sujet du cercle - Ie fameux cercle hermeneutique - en vertu
duquel les anticipations de sens de l'interprete font partie inte-
grante du sens a interpreter. Pour com prendre un texte, il faut deja
l'avoir pre-compris. Et cette pre-comprehension ne peut etre
184 P. Ricoeur

eliminee, sous peine de rompre Ie pacte entre l'interpretant et

l'interprete en vertu duquel I'in terprete accede aux intentions de
sens du texte. Ou point de vue epistemologique, cette implication
de l'interprete dans la chose interpn~tee ne peut paraftre que
comme une faiblesse, une tare subjectiviste, par rapport a I'objec-
tivite que semble requerir l'ideal de scientificite. Heidegger justifie
Ie cercle hermeneutique en montrant que sa faiblesse episte-
mologique apparente derive de sa force ontologique reelle: Ie
cercle Ie plus originaire, en effet, est celui qui existe chaque fois
entre la pre-comprehension et la situation intra-mondaine a inter-
preter. Ce cerc1e n 'est pas vicieux, mais constitue la condition
positive de la connaissance la plus originaire. Cette demonstration
n 'est pas sans force; eUe fait partie de la strategie generale de
Sein und Zeit qui est de ramener la reflexion des questions de
methode a leur enracinement ontologique. Mais, on ne voit pas
comment on peut revenir de ce fondement aux difficultes propre-
ment epistemologiques ressortissant a I'intcrpn!tation des textes.
En particulier, la subordination genera Ie du cercle epistemologique
au cerde on tologiq ue ne permet pas de trancher entre diverses
manieres de se comporter par rapport a un tex te: l'ontologisation
du probleme hermeneutique implique-t-elle la liquidation complete
de sa psychologisation chez Oilthey'? En particulier, faut-il aban-
donner Ia pretention a mesurer Ie sens d'un texte a l'intention de
l'auteur? Faut-il cesser de vouloir comprendre I'auteur mieux
qu'iJ s'est compris Iui-meme'? Ou encore: faut-il abandonner
I'idee de s'egaler avec I'intention de sens du texte, de s'en rendre
contemporain (que Ie sens soit celui du texte ou celui de I'auteur)'?
Faut-il cesser de definir I'hermeneutique par la lutte contre la me-
comprehension, par I'appropriation de l'etranger? par la Iutte
con tre la distance dans l'espace et dans Ie temps? par la repro-
duction de la production originaire? Le retour au fondement,
chez Heidegger, est si radical que les questions derivees sont en
quelque sorte perdues de vue, comme si elles etaient rendues
inessentielles, voire non pertinentes, par la remontee au fonde-
ment. Et pourtant il ne depend pas de la philosophie herme-
neutique que ces questions aient ete posees dans Ie passe par
I'exegese biblique, par la philologie classique, par lajurisprudence,
et qu'eHes soient encore posees par I'hermeneutique litteraire,
dont on donnera plus loin un echantillon. Bien plus, c'est par la
 'Logique hermeneutique? 185

capacite de revenir a ces questions que se mesure la pretention de

I'hermeneutique aetre une discipline !ondamentale, au sens propre
du mot.

2. VERITE ET/OU METHODE?

La meme sorte d'interrogation est suscitee par l'oeuvre majeure de

Gadamer, Wahrheit und Methode ,3 qui constitue Ie deuxieme relai
en tre l'hermeneutiq ue des sciences humaines d 'origine diltheyenne
et la presente enquete sur la condition epistemologique de l'herme-
neutique.
A bien des egards, Ie titre de Gadamer invite a etre lu dans Ie
sens d 'une alternative: verite ou methode. C'est probablement
ainsi que Ie comprennent les lecteurs formes dans la tradition de la
philo sophie analytiq ue.
L'experience initiale d 'ou l'oeuvre procede - Ie lieu d 'ou eIle
parle - est Ie scandale que constitue, a l'echeIle de la conscience
moderne, la sorte de 'distanciation alienante' (Ver!remdung), qui
est plus qu'un sentiment ou une humeur, mais la presupposition
ontologique qui soutient la conduite objective des sciences
humaines. La methodologie de ces sciences implique ineluctable-
ment une mise a distance, laquelle a son tour presuppose la des-
truction du rapport primordial d 'appartenance - de Zugehorigkeit
- sans quoi il n'existerait pas de rapport a l'objet. Ce debat entre
distanciation alienante et experience d'appartenance est poursuivi
par Gadamer dans les trois spheres entre lesquelles se repartit,
selon lui, l'experience hermeneutique: sphere esthetique, sphere
historique, sphere langagiere. Dans la sphere esthcWque,
l'experience d 'etre saisi par la chose belle est ce qui toujours
precede et rend possible l'exercice critique du jugement dont Kant
a fait la theorie sous Ie titre du Jugement de gout. Dans la sphere
historique, c'est la conscience d 'etre porte par des traditions, qui
rend possible l'exercice d'une methodologie historique au niveau
des sciences humaines et sociales. Enfin, dans la sphere du langage,
qui d'une certaine fac;on traverse les deux precectentes, la co-
appartenance aux choses dites par les grandes voix des createurs de
discours precede et rend possible toute reduction instrumentale du
langage et toute pretention a dominer par des techniques objec-
186 P. R icoeur

tives les structures du texte de notre culture. Ainsi une seule et

meme these court a travers les trois parties de Wahrheit und
Methode.
Le debat avec les sciences de l'esprit ne constitue done pas
I'unique point d'ancrage de la philosophie hermeneutique. L'entree
dans la problt~matique par I'esthetique est au contraire irrem-
pla~able. C'est 13 que l'hermeneutique trouve dans la conscience
commune son meilleur poin t d'appui pour briser la pretention de
la conscience jugeante a s'eriger en arbitre du gout et en maitre du
sens. C'est grace a cette premiere breehe que la philosophie herme-
neutique peut operer sa seconde percee, au niveau de l'experience
historique, et ainsi renouer avec ce qui fut, chronologiquement,
son premier point de depart. AinsLl'universalite que revendiquera
l'hermeneutique dans Ie deb at qu'on evoquera plus loin, entre
hermeneutiq ue et critique des ideologies, n 'exc1ut pas, mais
requiert la multiplieite de ces ancrages concrets. II n'est toutefois
pas indifferent que Gadamer ait ecarte comme moins significative
une reflexion sur l"etre pour Ie texte' (Sein zum Texte) au il voit
Ie danger d'tme reduction de I'experience henneneutique 3 la
traduction, erigee en modele langagier du comportement humain
a l'egard du monde.
Si I'on se eoncentre main tenant sur la deuxieme partie de
Wahrheit und Methode, les trois points ou I'anti-methodologisme
de Gadamer s'avere Ie plus vif et sur lesquels ses adversaires se
sont attardes Ie plus volontiers sont la rehabilitation du prejuge,
de la tradition et de I'au torite, - la notion de Wirkungsgeschichte.
-- et celie de 'fusion des horizons'. Ce sont aussi les trois points
ou il sera possible de rattacher une lecture mains anti-methodolo-
gique de Wahrheit und Methode et donner au titre une interpre-
tation moins disjonctive.
C'est assurement avec une pointe de provocation que Gadamer
entreprenait son plaidoyer en faveur du prejuge. Le prejuge est une
categorie de l'Aufkl(irung. C'est ce dont il faut se debarrasser pour
'oser penser' - sapere au de , s'excIamait Kant! - et pour acceder a
riige 'adultc'. Mais Ie prejuge n'est si univoquement negatif que
dans une philosophie critique, dans une philosophie dujugement.
A son tour, seuic erige Ie jugement en tribunal unc philosophic
qui fait de I'objectivite, dont Ie modele est fourni par les sciences,
la mesure de la connaissance. II faut au contraire considerer avec
 'Logique henneneutique '? 187

Heidegger que nul sujet de connaissance n'accecte a un domaine

d'objet sans avoir d'abord projete sur ce domaine une pre-compre-
hension qui assure la familiarite avec celui-ci. Or cette pre-compre-
hension n'est pas entierement transparente a la reflexion. Nul
sujet transcendantal n'en a la parfaite maitrise. C'est pourquoi Ie
prejuge est simplement la projection sur Ie plan du jugement
d'une categorie hermeneutique fondamentale: la tradition. 'L'etre
homme trouve sa finitude dans Ie fait que d'abord il se trouve au
sein de traditions'. La tradition est positivement l'expression du
caractere historiquement fini de la comprehension de soi par
l'etre humain. Le troisieme terme de la trilogie: prejuge-tradition-
autorite ne fait qU'expliciter ce role de la tradition en terme
d'efficacite. Mais sa signification nous est cachee par Ie prejuge,
issu de I'Aufkliirung selon lequel l'autorite est synonyme de
domination, au sens de violence et la soumission it I'autorite
synonyme d '0 beissance aveugle.
La consequence epistemologique de ce plaidoyer est que les
sciences de I'esprit - et au premier rang l'histoire - s'edifient sur
Ie sol prealable de la transmission et de la reception des traditions.
La Forschung - inquiry - n 'echappe pas it la conscience his-
torique de ceux qui vivent et font l'histoire. C'est it partir d'une
tradition qui I'interpelle que l'histoire pose au passe des questions
'sensees'. Entre I'action de la tradition et I'investigation historique,
un pacte se noue que nulle conscience critique ne saurait delier,
sous peine de rendre la recherche elle-meme insensee.
Cette condition 'historiq ue' (geschichtlich) de I'histoire (His-
torie) s'expriJne dans Ie concept de Wirkungsgeschichte (mot a
mot: I'histoire des effets). Cette categorie ne releve plus de la
methodologie historique, mais de la conscience reflexive de la
mHhodologie. C'est la conscience d 'etres exposes it l'histoire et
a son action, d'une maniere telle qu'on ne peut objectiver cette
action sur eux, parce que cette efficace fait partie de son sens en
. tant que phenomene historique. Cette conscience de l'efficace a
une signification negative et une signification positive. Negative-
ment, elle exclut tout survol qui permettrait de maftriser du regard
I'ensemble des effets sur no us du passe; l'etre historique est ce qui
ne passe pas dans Ie savoir absolu. Ici la philo sophie de l'herme-
neutique est, a l'encontre de toute resorption hegelienne dans Ie
savoir absolu, une philosophie de la finitude. Positivement, il faut
188 P. Ricoeur

dire que Ie concept d 'histoire efficace mediatise notre rapport au

passe. A partir de cette mediation, quelque chose est significatiL
interessant, valable, memorable, bref digne d'etre relate dans une
enquete historique.
Que l'efficace historique ne signifie aucunement que nous
soyons prisonnier du passe est atteste par Ie troisieme concept
cle, celui de 'fusion des horizons'. Le concept d 'horizon complete
et corrige celui de situation. L 'horizon c'est ce que decoupe un
point de vue. Parler de fusion des horizons c'est admettre, avec
l'hermeneutique diltheyenne, qu'il est toujours possible de se
transporter dans Ie poin t de vue de I'autre. Mais ce transfert
n'est pas une enigme psychologique. S'il est possible d'entrer
dans une dialectique des points de vue, c'est paree que, dans cette
tension entre l'autre et Ie propre, une entente prealable sur la
chose meme porte la recherche d'un accord effectif. Mais a son
tour cette entente prealable ne peut etre transformee en un savoir
objectif qui abolirait l'alterite des points de vue, par mise a dis-
tance de to us les poin ts de vue. Car personne n 'aura it plus rien a
dire dans ce dialogue, personne ne s'apportant plus lui-meme au
debaL Nous n 'existons ni dans des horizons fermes, ni dans un
horizon unique, de caraetere 0 bjectif-empiriq ue ou dialectiq ue-
speculatif. L'idee de 'fusion des horizons' comme celIe d'efficace
historique' a une portee negative et une portee positive. Elle
signifie negativement Ie refus de toute inclusion abjectiviste ou
de toute Aufhebung speculative. Elle signifie positivement que
c'est a partir de la 'fusion des horizons', et de I'entente prealable
sur la chose meme qui la soutient, qu'est possible Ie transfert dans
un autre point de vue ct dans une autre culture qui est a la base de
la connaissance historiq ue.

Qu 'en est-it du rapport de fa verite d fa methode?

Disons d 'abord de fa<;on globale que Ie rapport entre verite et
methode n'est pas un rapport simple d'opposition au d'exclusion
mu tuelle. Ce serait Ie cas si la comprehension historiq ue etait
opposee sur Ie meme plan a I'explication scientifique. A cet egard,
R. Bubner 4 a raison de mettre en garde contre une interpretation
de Wahrheit und Methode qui ramenerait purement et simplement
a la diehotomie methodologique que Dilthey n'a pas reussi a
depasser. La philosophic hermeneutiq ue n 'est pas une anti-episte-
 'Logique hermeneutique '? 189

mologie, mais une reflexion sur les conditions non epistemologiques

de I'epistemologie. Les trois categories que nous avons passees en
revue marquent les conditions indepassables de constitution de
l'espace de sens dans lequel quelque chose peut valoir comme
objet historique.
A partir de cette remarque generale, il est possible de donner
de Wahrheit und Methode une interpretation plus dialectique que
la precedente du rapport entre les deux termes. Chacune des trois
categories historiques marque la place en creux d'un moment
critique approprie qui assure la mediation entre l'hermeneutique et
les sciences humaines 0 bjectives.
La rehabilitation du prejuge ne signifie pas soumission a toute
tradition, mais seulement I'impossibilite de se soustraire a la con-
dition de transmission historique. La contre-partie de I'autorite de
la tradition c'est 'Ia reconnaissance (Anerkennung) de Sl
superiorite': 'I'autorite, dit Wahrheit und Methode (p. 264), n'a
aucune relation immediate avec l'obeissance: elle repose sur la
reconnaissance'. Ce qui, en dernier ressort, a autorite, c'est la
tradition a partir de laquelle nOlls interrogeons. L'avouer n'est pas
sacrifier la raison. Car la 'preservation' (Bewahrung) d'lIn heritage
culturel ne va pas sans critiq ue, done sans un debat permanent
entre I'esprit d'innovation et l'esprit de conservation. A cet egard,
l'operation qui joue a l'egard de la tradition en general Ie role que
la verification joue a I'egard de I'hypothese scientifique, Gadamer
la n omme 'application' (Verwendung). C'est l'hermeneutique
juridique qui en a Ie mieux compris l'importance: I'application
marque l'espace de jeu dans Jequel Ie juge se distingue du iegis-
lateur. L'application re<;oit allssi ses lettres de noblesse de la
rlH~torique ancienne qui s'adresse a l'auditeur, dont eUe connaft
les passions, pour Ie convaincre et Ie persuader. Cette categorie
fondamentale atteste que l'art de comprendre n'est pas complet
sans une actualisation critique du sens dans les conditions d'une
situation culturelle nouvelle.
La categorie d 'efficace historiq ue ne va pas non plus sans une
contre-partie critique. A savoir Ie concept 'distance historique'.
La distance est d'abord un fait, la mise a distance est deja plus:
c'est un comportement methodologique. L'histoire des effets ou
de l'efficace s'exerce precisement sous la condition de la distance
au double sens d 'eloignement passif et de mise a distance active.
190 P. Ricoeur

Elle devient ainsi la proximite du lointain. L'illusion methodolo-

gique commence seulement lorsque nous nous imaginons que la
distance met fin a notre connivence avec Ie passe et cree une
situation comparable a l'objectivite dans les sciences de la nature.
Le paradoxe de l'alterite du passe tient precisement a ceci que
l'histoire efficiente, c'est l'efficace dans la distance. La distan-
ciation alienante commence lorsque Ie moment d'objectivation est
abstrait de son site concret, a savoir l'appartenance de l'historien a
l'ensemble historique dont il entreprend I'investigation.
Quant ala categorie historique de 'fusion des horizons', dont on
a souligne Ia place par rapport a toute Aujhebung speculative ou
a toute inclusion dans un savoir objectif, elle trouve aussi son
complement critique dans la structure Iangagiere developpee
dans la troisieme partie de Wahrheit und Methode. Toute compre-
hension du monde en effet y apparaft conditionnee par une
pratique iangagiere commune. Mais si cette communaute
Iangagiere se soustrait a l'objectivation du Iangage dans un systeme
de signes manipulables, eUe ne signifie aucunement que l'entente
prealable sur la chose meme, evoquee plus haut, implique un
accord effectif deja donne (ici encore R. Bubner met en garde
c~ntre la confusion de l'entente presupposee avec quelque con-
sensus effectif que ce soit). La seule 'Iogique' qui soit appropriee
a la fusion des horizons, c'est la dialectiq ue, c'est a dire, au sens
originel du terme, I'art de la question et de fa reponse. C'est sur
cette dialogique que debouche I'epistemologie de I'hermeneutiq ue.
Elle marque l'inclusion du moment critique de la question dans
l'hermeneutique de la comprehension portee par la communaute
Iangagiere.

3. LA PRETENTION DE L 'HERMENEUTIQUE A L'UNIVERSALITE MISE

EN QUESTION

Ces premieres notations sur !'inclusion d'un moment critique dans

Ie processus global de comprehension vont se preciser a Ia faveur
du debat sur la pretention de l'hermeneutique it t'universalite. La
bataille epistemologique au sujet de l'hermeneutique s'est cristal-
lise sur un point precis: qu'en est-il de sa pretention a l'univer-
salite?S
 'Logique hermeneutique '? 191

11 y a debat parce que tantot l'hermeneutique pretend reglr

toute scientificite, dans Ia mesure OU toute connaissance objective
s'enracine dans une comprehension langagil~re du monde qui Ia
precede, tantot elle paraft se limiter elle-meme au role de
Kunstlehre a l'egard d 'un groupe de sciences, celles que Ie XIXeme
siecIe finissant appelait sciences de l'esprit. Dans Ie premier cas,
l'universalite n'est pas douteuse: elle exprime Ia subordination de
toute explication a un fond prealable de comprehension; en
revanche la scientificite de cette comprehension est incertaine.
Dans Ie deuxieme cas, la scientificite de l'hermeneutique est plus
aisee it justifier dans Ie cadre d'une epistemologie des sciences
humaines, mais c'est son universalite qui devient douteuse, dans la
mesure ou l'explication tombe en dehors de la comprehension.
Deux contributions - d 'a ille urs proches l'une de l'autre -
meritent a cet egard d'etre examinees ici, celIe de K.O. Apel et
celle de 1. Habermas. Elles tendent l'une et l'autre a mettre
l'accent sur la scientificite de l'hermeneutique aux depens de son
universalite et ainsi conduisent a situer l'hermeneutique dans un
ensemble 'scientifique' plus vaste.
Schematiquement la critique par Habermas 6 peut se resumer
ainsi. La ou Gadamer emprunte au romantisme philosophique
la rehabilitation du prejuge et derive de la notion heideggerienne
de precomprehension son concept de conscience de I'efficace his-
torique, Habermas developpe un concept d'interet issu de la
critique marxiste des ideologies, lui-meme reinterprete a la lumiere
de Lukacs de Histoire et Conscience de classe et des travaux de
l'Ecole de Francfort. II est vrai que Habermas est aussi loin du
marxisme que Gadamer I'est du Romantisme allemand. Au
monisme marxiste du concept de production Habermas substitue
un pluralisme des interets dont chacun regIe, a la fac;on d'un
a priori anthropologique, un domaine scientifique. II Ie regie en
ce sens que la signification des enonces possibles relevant de cette
sphere d'objets est precteterminee et prescrite par cet interet. Ainsi
a I'interet technique ou instrumental, defini comme 'interet
cognitif au controle technique applique a des processus objectives',
correspond la sphere des enonces empirico-analytiques. A I'interet
pratique ou interet pour Ia communication inter-humaine corres-
pondent les sciences historico-hermeneutiques; la signification des
propositions produites dans ce domaine ne procede pas de la
192 P. Ricoeur

prevision possible e't de l'exploitation technique, mais de la com-

prehension du sens; cette comprehension se fait par Ie canal de
I'interpretation des messages echanges dans Ie langage ordinaire,
par Ie moyen de l'interpretation des textes transmis par la
tradition, enfin grace a l'interiorisation des normes qui
institutionalisent les roles sociaux. La discussion de ces deux
inten~ts contient l'essentiel de la critique de Marx, accuse d'avoir
confondu Ie plan technique et Ie plan pratique et d'avoir reconduit
au positivisme la critique des conditions de possibilite de l'agir
humain, issue de Kant, de Fichte et des hegeliens de gauche. Mais
c'est Ia distinction entre l'interet pour la communication et un
troisieme interet, intitule interet pour l'emancipation, qui marque
l'opposition a Gadamer. Ce troisieme in teret deplace Ie centre de
la discussion des sciences historico-hermeneutiques vers les
sciences sociales critiq ues, auxquelles appartiennen t pour I'essen tiel
la critique des ideologies et la psychanalyse. Dans ce couple la
critique des ideologies fournit Ie champ d 'application, a savoir les
formes de communication inter-humaines systematiquement dis-
tordues, la psych analyse, Ie modele d 'explication, it savoir la quasi-
objectivation des processus que leur opacite rend inaccessibles a
une simple explicitation de pre-supposWons implicites. C'est ce
detour par la quasi-observation et la quasi-explication qui marque
la limite de la comprehension hermeneutique. Celle-ci en effet
parait se borner it la clarification des formes de mecomprehension
homogenes elles-memes a la comprehension, comme dans la
tradition de SchIeiermacher et de Dilthey. Que les quasi-objec-
tivations doivent deboucher finalement sur un elargissement et
un approfondissement de la comprehension de soi, ce dernier
episode donnera lieu tout it l'heure a un assaut de pretentions
entre l'hermeneutique s'erigeant en meta-critique et la critique
s'erigeant en met·a-henneneutique. Mais avant cette confron-
tation ultime, creusons la difference.
En distinguant l'interet pour l'emancipation qui guide les
sciences sociaies critiques de !'interet pour la communication qui
regIe Ies sciences historico-hermeneutiques, Habermas fait
apparaftre Ie caractere limite de l'hermeneutique liee a un groupe
de sciences, les sciences 'traditionnelles'. Or la notion de tradition
est ambigue. Elle designe pour une part Ie fait incontestable de 1a
dependance de tout projet present a un passe transmis et compris.
 'Logique henneneutique '? 193

Mais eIle designe aussi les contenus transmis, dont certains expri-
ment des formes ideologiquement gelt~es de relations inter-
humaines. Si ces reifications du lien social ne peuvent etre trans-
formees que critiquement, c'est parce que Ie discours dans lequel
elles s'expriment marque la dependance entre, d'une part, Ie
langagc ct, d 'autre part, le couple forme par Ie travail et la domi-
nation. L'hermeneutique, ici, peut etre accusee d'idealisme
linguistique. dans la mesure ou elle ignore ce rapport de dependance
du langage a des forces sociales qui font du discours Ie siege de dis-
torsions systematiques irreductibles a des phenomenes de mecom-
prehension q u 'une comprehension plus penetrante su ffirait a
dissoudrc. L'ideologie, en effet, travaille derriere Ie dos des parten-
aires sociaux. C'est pourquoi sa dissolution requiert Ie detour de
procedures explicatives et non plus seulement comprehensives,
mettant en jeu un apparcil theorique, comme celui de la meta-
psychologic freudienne. qui ne peut deriver d 'une simple extension
dc I'art spon tane d 'in terp retation a I'oeuvre dans Ie discours
ordinaire et dans la con versa tion.
Si donc I'hermeneutique est epistemologiquement limitee -
limitee au groupe des sciences hermeneutiques - c'est parce
qu'elle est philosophiquement limitee par sa meconnaissance du
rapport entre langage et violence, suscite par Ie conflit des forces
sociaies. En ce sens Ie phenomenc ideologique, considere avec ses
variantes individuelles ou collectives, constitue unc experience
limite pour l'hermeneutique.
Le conflit entre hermeneutique et critique des ideologies
doit dre porte a un degre plus profond encore de radicalite. Pour
Habermas Ie defaut principal de l'hermeneutique de Gadamer est
d 'avoir on tologis~ l'hermeneutiq ue; il entend par lil son insistance
sur J'entente. sur l'accord, comme si Ie consensus qui nous precede
erait quelque chose de constitutif, de donne dans l'etre. Habermas
ne peut avoir que mefiance pour ce qui lui paraft etre I'hypostase
ontologique d 'une experience rare, a savoir 1'experience d 'etre
precede dans nos dialogues les plus heureux par l'entente. Mais on
ne peut canoniser cette experience et en faire Ie paradigme de
1',Jction communicative. Ce qui nous I'interdit, c'est precisement Ie
phenomene ideologique. Si J'ideologie etait seulement un obstacle
in terne a la comprehension, une mecompnShensiol1 que Ie seul
exercice de la question et de la reponse pouvait reintegrer, alors on
194 P. Ricoeur

pourrait dire que la ou il y a mecomprehension il y a entente

prealable. Des lors il appartient it une critique des ideologies de
penser en termes d'anticipation ce que l'hermeneutique pense en
termes de tradition assumee. Autrement dit, la critique des
ideologies implique que soit pose com me idee regulatrice, en avant
de nous, ce que I'hermeneutique des traditions conyoit comme
existant a I'origine de la comprehension. Cette idee regulatrice est
celie d'une communication sans bomes et sans contrainte. L'accent
kantien est ici evident; I'idee regulatrice est plusdevoiretrequ'etre,
plus anticipation que reminiscence. C'est cette idee qui donne sens
a toute critique psychanalytique ou sociologique; car il n'y a de
desymbolisation que pour un proje t de resymbolisation; et il n 'y a
de projet tel que dans la perspective revolutionnaire de la fin de la
violence. Un eschatologisme de la non-violence constitue ainsi
I'horizon philosophique demier d'une critique des ideologies. Cet
eschatologisme, proche de celui de Ernst Bloch, prend la place que
tient I'ontologie de I'entente langagiere dans une hermeneutique
des traditions.
Karl-Otto Ape!7 situe son appreciation de I'hermeneutique a
l'interieur d'un projet englobant qui serait de restituer a la Wissen-
schaftslehre ~ qui, en allemand, a toujours signifie plus que
science - son droit epistemique, sans pourtant la reduire a la 'logic
of science', au sens de la philosophie analytique. Cette restitution
ne peut se fa ire que par une reprise et un elargissement de 1a
question transcendantale kantienne, c'est-a~ire par une reflexion
sur les conditions de possibilite de ce qui, d'une maniere generale,
est re<;u comme sense (sinnvoll). La reflexion de deuxieme degre
sur I'hermeneutique sera done transcendantale, dans la mesure
ou elle porte sur la condition de possibilite du discours qu'elle
articule. Mais cette reflexion ne peut etre menee a bien que si on
ne limite pas I'investigation transcendantale a cette region du
sens que Kant a delimitee sous Ie titre de I'objectivite et qu'il a
rapportee a l'unification du divers de I'experience sous Ies cate-
gories de l'entendement par un sujet de pure connaissance. Ce
que Kant a omis, c'est qu'une unification langagiere precede
l'unification categoriale. Or cette unification langagiere met en
jeu, outre un entendement categorique, un engagement corporel,
techniquement mectiatise, et une entente de caractere inter-subjec-
tif sur les normes implicites de la communication. Cette unification
 'Logique hermeneutique '? 195

langagiere requiert donc bien plus qu'un a priori de conscience: un

a priori corporel et social; celui-ci, a son tour met en jeu les
inten~ts decrits et analyses par Habermas et qu'it faut rapporter a
une anthropologie transcendantale.
C'est dans cette perspective generale d 'une scientificite elargie
que Karl-Otto Apel developpe deux theses: la these de la comp/e-
mentarite entre ce qu'il appelle Scientistik et l'hermeneutique,
- et la these de la mediation operee par la critique des ideologies
entre les pretentions rivales de ces dernieres.
L'idee meme d'une complementarite entre Scientistik et her-
meneutique implique d'abord que }'on rejette Ie programme flt!O-
positiviste de la 'unified science' et done que l'on reprenne a
nouveau frais Ie probleme diltheyen de la distinction entre com-
prendre et expliquer. La comprehension, soutient Apel, ne se
borne pas a une operation psychologique extrinseque aux oper-
ations logiq ues; eUe ne releve pas seulement d 'une heuristique
preparatoire, comme si la science commenc;:ait seulement avec la
position d 'hypotheses verifiables ou falsifiables. La justification
de la comprehension comme dimension epistemologique propre
repose partiellement sur un argument de type diltheyen, a savoir
que dans les sciences hermeneutiques la comprehension fait partie
integrante de la constitution du sens. L 'hermeneutique trouve ici
un renfort a l'interieur du camp analytique, dans les resistances
qu'oppose l'epistemologie de la connaissance historique au modele
explicatif propose par Hempel et Popper.
Mais it est interessant de voir de queUe maniere cet argument,
issu d'une critique interne ala philosophie analytique et emprunte
a Dray, Danto, etc.,s n'est pas seulement interpole, mais veri-
tablement integre a la tradition hermeneutique d'origine
diltheyenne. C'est de Heidegger et de Gadamer que Apel rec;:oit
!'idee que toute comprehension met en jeu un rapport sense entre
un projet et une situation. C'est pourquoi la comprehension reste
Ie mode irreductible d 'in telligibilite en histoire, dans la mesure ou
une suite singuliere d'evenements tire son sens du rapport entre
les intentions de sens des agents et une situation singuliere telle
qu'elle est comprise par ceux-ci.
A son tour, ceUe integration d'un fragment de philo sophie
analytique dans l'hermeneutique se fait a la faveur d'une double
inflexion de l'hermeneutique elle-meme. D'un cote, on se detoume
196 P. Ricoeur

de la question de la difference ontico~ntoIogique, done de la

question de l'etre en tant qu'etre. Et l'on met presque exclusive-
ment l'accent sur Ie caract ere langagier de la comprehension du
monde. L'hermeneutique devient ainsi pour l'essentiel une Sprach-
hermeneutik. Celle-ci s'attache aux presuppositions implicites a
toutes les formations de sens, au premier rang desquelles les trans-
missions culturelles. Deuxieme inflexion de l'hermeneutique: Ie
caractere langagier de la comprehension du monde selon Heidegger
est rapproche de la theorie wittgensteinienne des ~eux de \angage'
consideres comme 'formes de vie'.
On y reviendra plus loin lorsqu'on considerera pour lui-meme
Ie debat entre les deux heritages de Wittgenstein et de Heidegger.
Du dernier Wittgenstein deux themes sont essentiellement retenus
en vertu de leur analogie a ceux de la Sprachhermeneutik: d'une
part l'idee qu'une articulation linguistique determine les limites de
notre monde, d 'autre part l'idee que ceUe articulation est
d'embh~e publique. A cet egard, la critique wittgensteinienne du
langage prive supprime Ie probleme diltheyen, et a certains egards
husserlien, du passage d 'une comprehension d 'abord subjective a
une comprehension inter-subjective par transfert dans les
intentions de sens d'autrui. Heidegger, Gadamer et Wittgenstein se
rencontrent ici dans une meme reaction anti-psychologiste.
Le pro bleme de la connaissance historiq ue, ainsi n~in terprete
dans les termes de la Sprachhermeneutik, elle-meme inflechie par
la reception de Wittgenstein, devient alors ceci: la comprehension
des raisons d'agir mises en oeuvre par des agents historiques est
irrCductible a une explication relevant de la Scientistik; la con-
naissance historiq ue est seulemen t la rectification - au double
sens d'une mise a l'epreuve critique et d'une complication - d'un
jeu de langage qui a deja ete joue avant que l'histoire se constitue
comme science, a savoir la pratique narrative dans l'horizon d'une
tradition culturelle.
Ce n'est toutefois pas cette reprise du probleme de la connais-
sance historique dans une perspective hermeneutique qui demontre
Ie mieux la complementarite entre la Scientistik et l'herme-
neutique. C'est au plan meme de la connaissance scientifique prise
dans son ensemble qu'il faut montrer la complementarite entre
l'explication objective et l'entente inter-subjective de caractere
pratique et iangagier. C'est de toute connaissance qu'il faut dire
 'Logique hermeneutique?' 197

qu'elle reunit les deux dimensions de la praxis: technique et

ethique. La praxis moderne la plus engagee dans la technologie
presuppose, en effet, une entente prealable sur les possibilites et
les normes d 'un etre au monde tenu pour sense. La tradition est Ie
lien et Ie milieu de cette comprehension. Si donc toute connais-
sance presuppose implicitement l'existence d 'une communaute de
communications, il revient a l'hermeneutique de thematiser cette
presupposition. C'est dans cette operation que s'atteste la comple-
mentarite entre Scientistik et hermeneutique.
Pourquoi alors recourir a une troisieme discipline pour les
concilier? C'est ici que Apel et Habermas prennent leur distance a
l'egard de Gadamer. Pour eelui-ci l'engagement existentiel dans la
tradition est a comprendre comme 'application' qui 'medie' entre
la norme du passe et la situation presente. Par l'applieation,
l'efficace de l'histoire est restituee dans sa force originelle en depit
de Ia distance temporelle. Le modele de I'interpretation est alors
a chercher dans Ie travail d 'actuaIisation de la norme par Ie juge.
Sans nier que Gadamer ait parfaitement conscience de ce que
I'hermeneutique n'a son mot a dire que lorsque la tradition a
perdu sa force, donc dans une situation de crise ou la transmission
culturelle est menacee de rupture, Apel et Habermas voient dans
I"application , davantage une limitation qu'une concretisation du
projet hermeneutique. On ne parle d'application que pour des
textes religieux dont l'autorite demeure, meme si eUe s'est
affaiblie, ou pour des textes Iitteraires 'classiques', c'est-a-dire
capables d 'une reaetualisation en toute situation cultureUe, ou
pour des textes juridiques dont la valeur normative demeure
in contestee. Or tel n'est pas Ie rapport moderne a la tradition. A
eet egard, Ie pro bk~me pose par les cultures non~uropeennes ou non-
americaines est plus clair et plus radical que Ie notre. La
reappropriation du passe ne s'y laisse plus eoncevoir comme appli-
cation, mais comme traversee d 'un doute radical qui equivaut a
une distanciation parfois douloureuse. Du meme coup on ne peut
plus mettre toute distanciation au compte de l'alienation method-
ique. Elle fait partie de la condition proprement moderne du
rapport a la tradition. En ee sens, c'est la meme distanciation qui
rend possible l'appropriation et l'abstraction methodiq ue. Apel
accorde volontiers qu'il est devenu difficile aujourd'hui de tenir
ensemble les deux theses: d'un cote en effet, il faut dire qu'il n'y a
198 P. Ricoeur

pas de point de vue neutre d'ou ron puisse regarder a distance

toutes les traditions, sous peine de retomber dans les impasses de
l'historicisme; en ce sens, la transmission de la tradition reste la
condition d'acces a n'importe quel objet culturel; a l'egard de
l'illusion methodologique l'hermeneutique fonctionne comme
revelateur de nai"vete. Mais, d'un autre c6te, il faut reconnaftre
que telle ou telle tradition ne nous parle plus et que sa reappro-
priation immediate nous est interdite. Reste alors a tirer avant age
de la distanciation elle-rneme, a pratiq uer a sa faveur une quasi-
objectivation des contenus transmis, afin d'acceder a une appro-
priation plus mediate, plus complexe, dont nous n 'avons pas
encore, dans bien des cas, la cleo
Cette quasi-objectivation est celie que pratiquent les sciences
sociales critiques dont Ie modele est la psychanalyse. Elk consiste
a traiter les formations culturelles dont Ie sens nous est devenu
etranger comme sympt6me de relations reelles, de structures
materielles appartenant a une autre dimension que Ie langage.
C'est Ie cas des effets de sens issus des rapports entre travail et
domination dans la societe industrielle capitaliste. Entre discours,
travail et domination se tissent des reseaux de relations dont
l'opacite, la non-transparence, n'est pas accidentelle mais essen-
tielle, non pas fragmentaire mais systematique. Des lors une
hermeneutique qui ne fait fond que sur la capacite d'expJiciter
l'implicite par la seule force du discours peut etre accusee
d'idealisme linguistique. Une telle hermeneutique se heurte a ce
que Apel considere comme Ie prolongement de l'histoire naturelle
dans l'histoire spirituelle. Sans cette non-transparence de principe,
les hommes pourraient s'egaler a leurs propres intentions de sens
et realiser l'ideal de mutuelle identification qui fut celui de l'her-
meneutique romantique. Or les hommes jusqu'ici n'ont pas vrai-
ment fait leur histoire; celle-ci se fait en grande partie derriere leur
dos; c'est pourquoi la reprise reflexive integrale de son sens est
impossible. Les facteurs contingents de l'existence sociale ne se
Jaissent analyser qu'au niveau d 'une explication quasi-objective
dont la meta-psychologie freudienne est Ie modele. A ce niveau,
la construction des modeles d'explication fait appel a une ter-
minologie et a une conceptualite de deuxieme degre auxquelles
les acteurs ne peuvent participer. En retour, l'hermeneutique
reprend l'avan tage pour rappeler que cette quasi-objectivation
 'Logique hermeneutique '? 199

caracteristique de la position du therapeute n'est jamais qu'une

communication partiellement suspendue qui a pour finalite la
reintegration des significations alienees dans une comprehension
de soi mediate et approfondie.
11 en resulte que Ie savoir quasi-objectif relevant du modele
therapeutique ne saurait se laisser reabsorber par I'objectivite
de la Scientistik. Si tel etait Ie cas, Ie meta-Iangage de I'expli-
cation therapeutique permettrait a celle-ci de fonder une
maitrise technique non seulement des phenomenes consideres
mais des hommes eux-memes. Cette maitrise technique servirait
seulement a etendre la domination de l'homme sur l'homme. La
seule riposte a ce danger est de subordonner la quasi-objectivite
des sciences sociales critiques a la reflexion sur soi dans Ie cadre
d'une ultime comprehension de soi.
C'est ce statut mixte des sciences sociales critiques qui fait de la
critique des ideologies la discipline mediatrice entre l'herme-
neutique et la Scientistik.

4. LA REPLIQUE DE L'HERMENEUTIQUE

La defense de I'universalite de l'hermeneutiq ue a ete conduite

a des niveaux differents par Gadamer,9 par ses disciples ou par
ceux qui, sans se reclamer de lui, s'apparentent a lui.

a) A un premier niveau encore formel, taus s'accordent pour

distinguer entre la visee universelle de I'hermeneutique et Ie
caractere limite des champs d'experience d'ou la reflexion
prend son depart. On peut bien des lors denoncer Ie caractere
limite de la problematique des sciences humaines mais. meme dans
Wahrheit und Methode, cette problematique ne couvre pas tout
i'espace parcouru par la reflexion; l'experience historique est
encadree entre I'experience de I'art et l'experience langagiere;
I'experience de I'art a meme une priorite toute speciale, dans 1a
mesure ou eUe fait paraitre c1airement l'anteriorite de la verite de
la chose sur Ie jugemen t esthetique; quant a 1'experience
langagiere, elle a certes sa limitation propre. si on la reduit a la
problematique du texte ou a celIe de la traduction, mais c'est elle
qui contient Ie principe du ctepassement de tous les domaines
limites pris pour points de depart.
200 P. Ricoeur

C'est pourquoi on peut suggerer d'autres points de depart que

ceux qu'adopte Wahrheit und Methode. Par exemple, la Rhe-
torique des Anciens fut a la philo sophie grecque ce que fut, pour la
pensee luthcrienne, puis pour Ie Romantisme allemand, l'herme-
neutique, a savoir l'entreprise de faire face it la dissolution de liens
solides avec la tradition, l'effort pour saisir quelque chose
d'evanouissant pour Ie porter a la lumiere de Ia conscience. La
Rhetorique, eUe aussi, a sa propre limite, a savoir d'abord qu'elle
concerne plutot Ie disc ours oral et son influence sur un auditoire,
alors que l'hermeneutique concerne davantage l'expression ecrite
du discours; en outre, sa pretention a la verite se borne it l'ordre
des arguments probables capables de persuader. Mais cette double
limitation n'empeche pas de retrouver }'universalite de l'herme-
neutique a partir du probleme limite de la Rhetorique. Celle-ci
a une ubiquite illimitee que l'on peut enwre verifier aujourd'hui.
La science elle-meme, en effet, n'est culturellemcnt efficace que
par Ie relais de la rhetorique qui fait appel aux ressources de com-
prehension du langage ordinaire. L'hermeneutique fait appel aux
memes ressources lorsqu'elle s'adresse a \'atopon dans notre
orientation a i'egard du monde et se propose I'appropriation
reussie de la tradition.
Que l'universalite de l'hermeneutique puisse se faire valoir a
partir de points de depart autres que ceux memes explores par
les heritiers de la problcmatique des sciences de l'esprit. deux ou
trois exemples contemporains Ie confirment.
Ainsi Hans-Robert Jauss entreprend de renouveler Ie probh~me
de l'experience esthetiq ue dans Ie cadre d 'une hermeneutique
litteraire. 1O Le caractere hermeneu tiq ue de son entreprise s'atteste
dans Ie souci de ressaisir dans son ensemble Ie rapport entre
auteur, texte et iccteur, sans se bomer a une esthetique de la pro-
duction des oeuvres. A la dimension poetiq ue de la 'production'
elle ajoute la dimension proprement esthetique de la 'reception' et
la dimension cathartique de la 'communication'. Elle renoue ainsi
avec la Rhetoriq ue de Gor.~ias et d 'Aristote et avec la Critique de
la Faculte de Juger de Kant, qui constituent les sources diverses
de ce que I'on peut appeier une esthetique de la reception. L'entre-
prise !""essortit en outre a l'hermeneutique par son effort pour
reenraciner Ie moment reflexif de l'interpretation sav~mte dans
l'experience pr.maire de la 'jouissance comprehensive'. En cela
 'Logique hermeneutique '? 201

elle procede directement des concepts d'efficace historique et

d'application de Gadamer. Elle s'en distingue neanmoins sur deux
points: la 'superiorite d'origine' et la sorte d'immutabilite que
Gadamer reconnaft aux oeuvres classiq ues, paraissent aJauss devoir
etre subordonnees a la productivite de la comprehension. Mais,
surtout, la critique que Gadamt!r adresse a l'abstraction de la
conscience esthetique au nom de la densite ontologique de
I'oeuvre et de son message de verite Ie conduit a meconnaitre Ie
concept-cle de jouissance esthetique, lequel ne peut aucunement
etre compris a partir des formes dechues de la culture esthetique
du XIXeme siecle et moins encore a partir de son exploitation par
la culture contemporaine de consommation.
Ce plaidoyer pour la ~ouissance comprehensive' n 'oppose pas
moins I'ht!rmeneutique litteraire de Jauss a l'esthetique
d' Adorno 11 qu'elle ne la distingue de celle de Gadamer. Pour
Adorno Ie plaisir esthetique est entk~rement corrompu par la cul-
ture bourgeoise qui en fait, sous forme de consommation passive,
Ie complement mensonger de l'ascetisme de la production. Seul un
ascetisme esthetique ne faisant appel qu'a Ie reflexion et non a la
jouissance pourrait, selon Adorno, riposter a I'anti-Aufkldrung de
l'industrie et de la culture. En reponse a cette 'esthetique (ie la
negativite', Jauss oppose la fonction a la fois subversive et edu-
cative de I'oeuvre d'art, en vertu precisement du caractere sociale-
ment 'indomptable' de la jouissance esthetique.
Mais la jouissance esthetiq ue ne peut etre restauree au rang
d'une experience originaire que si eile cst ramenee a ses sources par
Ie moyen d'une veritable histoire hermeneutique des affects des-
interesses. Cctte histoire conceptuelle de la jouissance, qui fait
appel a Augustin aussi bien qu'a Gorgia~ et Aristote, es! pour
I'essentiel une histoire de la fonction cathartique de l'art a travers
la rhetorique, la poetique, l'apologetique. Elle fait apparaftre la
richesse ignoree de l'experience esthetique, laquelle unit la dene-
gation du monde et sa transposition dans l'objet esthetique, la mise
a distance des roles sociaux et I'identificatioll ludique au heros
imaginaire, 1a puissance de desobeissance et la capacite d'inaugurer
de nouvelles nonnes pour l'action.
L'oeuvre de Jauss apparaft aujourd 'hui la plus propre a com bIer
une lacune importante de l'hermeneutique, a savoir I'etude des
affects desin teresses dans leur rapport it la fiction et a la poesie. La
202 P. Ricoeur

notion de jouissance esthetiq ue ajoute ainsi une dimension

nouvelle, non strictemen t langagiere, a la pre-comprehension du
monde dans laquelle s'enracinent to us les savoirs. L'entreprise con-
cerne Ie debat epistemologique en ce sens que Ie concept
d 'experience esthetique renforce la these du primat du sensus
communis sur Ie savoir conceptuel et de la communication
esthetique sur Ie consensus theorique.
La retlexion hermeneutiq ue peut surgir en d'autres regions
de l'experience humaine, des qu'est restaure un tissu concret de
relations que I 'objectivation et la specialisation scientifique
tendent a dechirer. C'est Ie cas des rapports entre politique et
i:tlzique dans I'oeuvre de Joachim Ritter: Metaphysik und Politik,
Studien zu Aristoteles und Hegel. 12 Ritter entend d'abord
recouvrer la ph~nitude originelle du concept grec d'ethos, dans sa
proximite a I'acte d'habiter et a }'habitude de la maison, par dela
toute reduction a des regles. U. s'enracinent Ie concept aristo-
telicien de nomos et Ie concept hegelien de Sittlichkeit. De la
meme maniere doit etre restitue Ie concept concret de polis, par
dela sa reduction a l'Etat, afin de redonner son amplitude Ii la
politique, sans la borner a l'exercice du pouvoir. La 'Politie'
grecque, en effet couvre l'ensemble des relations vivantes a
l'interieur de la cite. C'est de la meme maniere que Hegel subor-
donne Ie droit, au sens kantien, a !'ensemble des institutions
Cthiques. Ainsi Ie concept de praxis retrouve-t-il son amplitude en
tant que vic regll~e par les coutumes de la communaute.
Aristote offre ainsi Ie modele d'une pensee qui a rompu avec
la legitimation mythique par !es origines et par l'anciennete et
qui place dans Ie developpement des institutions en marche vers
leur maturite Ie critere concret de leur justesse. La philosophie
hegelienne du droit de son cote ne dit pas autre chose: pour eIle
aussi l'homrne trouve sa realite et sa liberte dans la vie concrete au
sein d'institutions 'ethiques'.
Cette demonstration marque la contribution de l'histoire
hermeneutique des concepts d'ethique et de politique aux tenta-
tives diverses de la philosophie hermeneutique pour redonner
vigueur a la phronesis, au sensus communis, face aux pretentions
exclusives du savoir scientifique. 13

b) A un second niveau la philosophie hermeneutique cherche a

 'Logique hermeneutique '? 203

a
n!pondre directement l'objection seion Iaquelle l'hermeneutique
est limitee par un concept etroit de mecompn!hension qui
l'enferme dans un idealisme hnguistique. Pour y parvenir eIle
s'efforce de s'elargir elle-meme en integrant au processus de la
comprehension et de l'interpretation Ie segment explicatif ou
quasi-explicatif deploye par les sciences sociales critiques.
L'argument a lui-meme deux faces. une face polemique et une
face constructive.
D'un point de vue simplement polemique, l'hermeneutique peut
faire etat de la reconnaissance par ses critiques eux-memes du
caractere incomplet de I'explication objective des formes syste-
matiques de distorsion, telle qu'elle est menee par la psychanalyse
ou par Ia critique des ideologies.. De l'aveu meme de Habermas et
de Apel, l'explication doit s'achever dans une nouvelle compre-
hension hautement mediatisee qui releve d'une hermeneutique
elargie et approfondie. Les sciences sociales critiques peuvent donc
avoir l'ambition de porter Ia comprehension au niveau scientifique
en transformant la distance alienante en alienation contr6Iee (kon-
trollierte Verfremdung). Mais cet usage critique de l'objectivatlon
ne sera garde de retomber a l'objectivation dogmatique que si Ia
critique cesse de se concevoir comme simplement antithetique a
tout rapport de caractere traditionnel au passe. Admettre que la
retlexion n'opere que quand elle demasque des fausses pretentions
et detruit Ie dogmatisme de Ia praxis quotidienne, c'est revenir
a l'opposition naIve re<;ue de l'Aufkldnmg, entre raison et prejuge.
Mais cette surelevation de la reflexion, toujours en quete de
relations substantives a dissoudre, et deliee de toute traditionalite,
mhite bien plus l'accusation d'idealisme que l'affirmation de
I'universelle mediation de la comprehension langagiere.
A cet egard Ie cas de Ia psychanalyse est singulier. D'une part,
l'hermeneutique ne peut que ratifier l'interpretation que donnent
Apel et Habermas de la meta-psychologie freudienne, a savoir
qu'elle constitue seulement une quasi-explication de phenomenes.
que ceux-ci sont a decrire comme des cas de desymbolisation
systematique et que la resymbolisation constitue la finalite du
processus entier. Mais, me me interprete ainsi, Ie modele psych-
analytique merite d 'etre critiq ue pour deux raisons. D'une part on
risque de faire porter sur la reflexion critique tout Ie poids de
l'entreprise de dissolution des fausses comprehensions resultant
204 P. Ricoeur

d'une pseudo-{;ommunication; Ie modele psychanalytique tend

alors a renforcer Ie privilege de la reflexion dans un sens idealiste.
D'autre part, Ie modele confere aux medecins un role exorbitant
d'experts soustraits au contrale de la communication sociale. A ce
stade l'immixtion de la competence psychanalytique dans Ia
correction des d ysfonctions de Ia communication devient un
facteur perturbateur dans l'echange social. Face a ces deux perils,
il importe d 'une part de rappeler que 'la pretention de I'herme-
neutique est et reste d'integrer a l'unite de I'interpretation
langagiere du monde ce qui se donne comme incomprehensible ou
non universel, et accessible aux seuls inities' (Gadamer, Replik,
291) et que 'separer la reflexion de la praxis paraft impliquer une
deviation dogmatique qui atteint egalement Ie concept de
reflexion emancipatrice' (ibid. 310). D'autre part, il faut rappeler
que la situation hermeneutique authentique procede d'un rapport
entre partenaires sociaux irreductibles a la relation entre patient et
medeein. II importe done de ne pas melanger les deux jeux de
langage de la reflexion hermeneutique et de la reflex ion psyeh-
analytique. Au surplus, la tendanee a eoncevoir tout echec du
dialogue en termes de nevrose risque de substituer a la legitime
lutte des opinions une psychiatrie sociale qui irait a l'oppose
meme du projet d 'emancipation, I'adversaire devenant un malade a
soigner et donc a soumettre au pouvoir medical. Contre quoi il
faut rappeler que la relation malade/medecin n'est saine que
si Ie malade se livre volontairement au traitement medical
en raison d'une mala die qu'il avoue et en vue d'une aide qu'iI
sollicite.
Mais I'hermeneutique, selon moi, ne preservera sa credibilite que
si elle ne se borne pas a cette polemique et depasse la reaffirm-
ation quasi-incantatoire de son universalite. Or eUe n 'attestera sa
validite dans les regions ou Ie sens n'est pas opere intention-
nellement que si eUe montre concretement comment I'explication
s'intercale entre une comprehension initiaIe, proche de celIe dont on
use dans Ies conversations ordinaires, et une comprehension ter-
minale hautement mediatisee. On a deja montre plus haut, par une
lecture moins antithetiq ue de l'oeuvre de Gadamer, que chacune
des categories maftresses de I'hermeneutique historique --
traditionalite, efficacite historique, fusion des horizons -- fait
appel a un moment critiq ue complementaire: application, distance
 'Logique hermeneutique '? 205

historique, dialectique de la question et de la rt!ponse. 11 reste a

faire la demonstration de la valeur heuristique de ce rapport dialec-
tique dans des domaines precis ou l'articulation entre comprehen-
sion et explication peut etre thematisee en tant que telle.
L'auteur du present essai14 s'est efforce d'elargir 1a breche
ouverte per cette lecture de Gadamer et d'instaurer un lien dialec-
tique etroit entre l"appartenance' qui in corp ore l'interprete a son
domaine d'investigations et la 'distanciation' qui rend possible
les procedures d 'explication et en general l'attitude critique a
l'egard de tout contenu transmis. A cet effet, il s'efforce de mettre
en parallele la situation epist~~mologiq ue qui prevaut dans des
spheres d'investigation aussi diverses que l'analyse textuelle, Ia
connaissance historique et 1a theorie de 1'action.
La correlation entre ces trois problematiques est en effet
frappante.
L'analyse textuelle, prise isolement, ramene l'hermeneutique
sur son terrain initial, l'exegese et la philologie. En ce sens eUe
marque une limitation du projet hermeneutique a 'l'etre pour Ie
texte', pour parler comme Gadamer. Mais, on l'a VU, l'herme-
neutique fait toujours valoir sa pretention a l'universalite a partir
d'une experience priviIegiee. Or I'experience textuelle est rune
d'entre elles, en raison de l'influence predominante exercee, en
France au moins, par Ie stmcturalisme. Celui-ci a cree une
situation nouvelle, qui suscite une re-evaluation complete des
rapports entre explication et comprehension. 11 est en effet apparu
de nouveaux concepts d'objectivation et d'explication, qui ne
doivent rien a un transfert de la methode des sciences de la nature
dans Ie champ des sciences humaines, comme au temps de Dilthey.
Ces nouveaux concepts sont lies au nouveau modele semiologique
qui s'est etendu successivement de la phonologie a la semantique
lexicale, puis a des entites linguistiques plus complexes que la
phrase, telles que textes litteraires, mythes, folklore, recits. Ce
modele semiologique met en oeuvre les ressources logiques d 'un
concept de structure selon lequel un ensemble de signes peut etre
segmen te en un nombre fini d 'en tites de base, defini par leurs
seules valeurs oppositives et susc:eptibles d 'engendrer, par un jeu de
combinaisons reglees, toute la variele des manifestations linguis-
tiques. En ce qui concerne l'analyse textuelle proprement dite,
l'application du modele semiologique tend aeriger Ie texte au rang
206 P. Ricoeur

d'entite autonome, regi par des codes qui fonctionnent a un niveau

comparable a celui de la grammaire profonde et qui s'expriment a
la surface du texte par divers effets de sens. La notion d'intention
de sens, con~ue comme signification visee par I'auteur et con-
jecturee par Ie lecteur, se trouve ainsi exclue du champ de
l'analyse. Rejoignant ainsi une certaine critique nietzscheenne, et
s'appuyant sur une interpretation elle-meme structuraliste de
Marx, comme celIe d'Althusser, Ie structuralisme apparaft alors au
grand public comme une contribution majeure a la 'mori: du sujet',
que diagnostique par ailleurs I"archeologie du savoir' de Michel
Foucault.
Face au deri semiologique, I'hermeneutique ne peut se borner a
affirmer que Ie texte en soi est une abstraction prelevee sur
l'operation concrete et complete qui englobe auteur, texte et
lecteur. Elle doit aussi montrer ce qui rend possible l'abstraction du
tex te et ce qui legitime Ie recours a une explication de type
semiologique. Les procedures d 'abstraction et d 'objectivation,
me semble-t-il, resultent du statut meme de l'ecriture -- ou de
I'inscription en general - par rapport it la parole vive. L'ecriture,
en effet, ne se reduit nullement a la fixation materielle du dis-
cours; celle-ci est la condition d'un phenomene beaucoup plus
fondamental, celui de l'autonomie du texte. Autonomie triple: a
l'egard de I'intention de l'auteur, it l'egard de la situation culturelle
et de tous les conditionnements sociologiques de la production
du texte, a i'egard enfin du destinataire primitif. Ce qui signifie
que Ie texte ne coi"ncide plus avec ce que I'auteur a vouiu dire;
signification verbale et signification mentale ont des destins
differents. Cette premiere modalite d'autonomie impJique deja
la possibilite que la 'chose du texte' echappe a I'horizon
intentionnel borne de son auteur et que Ie monde du texte fasse
eclater Ie monde de son auteur. Mais ce qui est vrai des conditions
psychologiques l'est aussi des conditions sociologiques, et tel qui
est pret a liquider I'auteur est moins pret a faire la meme operation
dans I'ordre sociologique; Ie propre de I'oeuvre d 'art, de l'oeuvre
litteraire. de I'oeuvre tout court est partout de transcender ses
propres conditions psycho-sociologiques de production et de
s'ouvrir ainsi a une serie illimitee de lectures, elles-memes situees
dans des contextes socio-culturels toujours differents; bref, il
appartient a l'oeuvre de se de-contextualiser, tant au point de
 'Logique hermeneutique'? 207

vue sociologique que psychologique, et de pouvoir se recontextual-

iser autrement; ce que fait I'acte de lecture. II en resulte que la
mediation du texte ne saurait etre traitee comme une extension de
la situation dialogale. Dans Ie dialogue, en effet, Ie vis-a-vis du dis-
cours est donne a l'avance par Ie colloque lui-meme; avec
I'ecriture, Ie destinataire originel est transcende. Par dela celui-ci,
I'oeuvre se cree el1e-meme une audience, virtuellement etendue
a quiconque sait lire. On peut voir dans cet affranchissement
la condition la plus fondamentale pour la reconnaissance d'une
instance critique au coeur de I'interpretation. En ce sens, la dis-
tanciation appartient a la mediation elle-meme. On ne peut donc
pas dire que Ie passage par l'explication soit destructeur de la com-
prehension inter-subjective. C'est une mediation requise par Ie dis-
cours lui-meme des qu'il est exteriorise par des marques
exterieures, inscrit dans des 'codes litteraires'.
Mais la theorie du texte n'est pas Ie seul cadre OU la dialectique
de I'explication et de la comprehension peut etre thematisee en
tant que telle. La theorie du texte n 'est, pour une anthropologie
philosophique, que I'un des 'Iieux' ou Ie present debat peut
s'enrichir.
La theorie de l'histoire est Ie second de ces lieux. La aussi
I'opposition entre positivisme et anti-positivisme demande
aujourd 'hui une approche plus dialectique des rapports entre com-
prehension et explication. Les arguments diriges par Dray, Danto,
Mink, etc., contre Ie modele hempelien rejoignent ceux de I'her-
meneutique diltheyenne. On a deja evoque cette reprise de la
querelle en langue anglaise dans la philosophie hermeneutique con-
tinentale. Mais la discussion laisse intact Ie probleme de savoir
comment l'enquete (inquiry) historique - objective et explicative
- s'articule sur la comprehension sans laquelle l'historien serait
incapable d 'attacher la moindre signification aux faits qu'il decrit
et explique. Sans doute est-ce dans I'activite narrative elle-meme
qu'il faudrait chercher l'articulation de l'explication sur la com-
prehension.
On s'arretera plus longtemps sur Ie troisieme 'lieu', celui de la
theorie de {'action. lei I'hermeneutique rencontre, de maniere
inattendue, un courant de la philosophie analytique issu de Witt-
genstein et de Austin. Le parallelisme entre les deux situations
est a cet egard particulierement instructif. Une dichotomie,
208 P. Ricoeur

semblable a celle que je combats au plan textuel, a oppose un

camp intentionaliste a un camp causaliste. Pour les premiers, Ie jeu
de langage comportant des expressions signifiantes telles que
action, intention, motif, agent, etc. est irreductible au jeu de lan-
gage con tenant les expressions de mouvement, de cause et de com-
portement. Or ce dualisme est peut-etre aussi intenable que celui
de la comprehension et de l'explication textuelle. Et cela pour
plusieurs raisons. D'abord l'experience de l'action se distribue
selon une gamme continue de cas, allant de la motivation
rationnelle a une motivation pulsionnelle OU les motifs sont des
causes au sens aristotelicien du terme. Ensuite i1 faut peut-etre dire
que nul motif n'a de valeur explicative s'il n'est pas aussi une
cause. A cet egard, une phenomenologie du monde de la vie, peu
familiere a une philosophie analytique, serait sans doute seule
habilitee, a ce stade, a fa ire entendre en quel sens Ie corps pro pre
constitue, au plan ontologique, la racine commune au regime de la
causalite et a celui de la motivation, done aussi a l'explication et a
la comprehension. Un argument plus decisif encore contre Ie
dualisme semantique et epistemologique est fourni par l'examen
des conditions sous lesquelles une action est inseree dans Ie monde.
Dans Explanation and Understanding H. von Wright 15 propose une
reformulation des conditions de l'explication, d'une part, et de
celle de Ia comprehension, d'autre part, qui permet de les com-
poser ensemble dans Ia notion d'intervention intentionnelle dans
Ie monde. La reformulation des conditions d'explication pro cede
de la theorie des systemes et d'une analyse rigoureuse de la
difference entre condition necessaire et condition suffisante. La
possibilite de l'action est introduite par la consideration des con-
ditions d'isolement d'un systeme clos soumis al'analyse precedente.
Nous apprenons a isoler un systeme en Ie mettant en mouvement.
C'est ce que nous faisons en faisant correspondre une action que
nous pouvons faire - une 'action de base' au sens de Danto -
avec l'etat initial d'un systeme. C'est ainsi qu'un agent intervient
dans Ie cours des choses. L'intervention est Ie lieu ou se croisent
la comprehension de nos pouvoirs-faire et I'explication des sys-
temes n~els. La dichotomie entre mentalisme et physicalisme parait
alors resulter d 'une meconnaissance de raction humaine dans Ie
monde. II rcsulte du modele de I'intervention que les sciences
humaines requierent un statut epistcmologique specifique mettant
 'Logique hermeneutique '? 209

en jeu une explication quasi teIeoIogique et une explication quasi

causale. Ainsi est justifie Ie statut complexe de I'explication dans
Ies sciences sociaies et en particulier en histoire. On ne peut pas ne
pas etre frappe par Ia convergence de ces conclusions avec celles de
Karl-Otto Apel concernant Ie statut des quasi-explications dans la
critique des ideologies. Aussi bien von Wright reconnait-il de son
cote, dans Ie premier chapitre de son ouvrage, Ia permanence des
'deux traditions' de l'investigation scientifique, qu'it rattache
l'une a Galilee, l'autre a Aristote. 11 n'est pas etonnant des lors que
sa propre tentative pour detenniner Ies lieux d 'in tersections
epistemologiques entre ces deux traditions rejoignent celles d'une
partie de l'henneneutique post-heideggerienne.
II resulte de cette triple analyse que comprehension et expli-
cation ne s'opposent pas comme deux methodes. A parler stricte-
ment, seule l'explication est methodique. La comprehension est
Ie moment non-methodique qui precede, accompagne et cloture
l'explication. En ce sens la comprehension enveloppe I'explication.
En retour I'explication developpe analytiquement la compre-
hension. Telle est Ia projection au plan epistemologique d'une
implication plus profonde, au plan ontologique, entre l'appar-
tenance de notre etre aux etres et a I'eire et Ia distanciation qui
rend possible to ute objectivation, toute explication et toute
critique.

c) Cette derniere remarque introduit au troisteme niveau. 11 con-

cerne Ie caractere transcendantal de la reflexion de l'henne-
neutique sur son propre discours.
Si l'hermeneutique n'est pas une anti-epistemologie, mais une
reflexion sur les conditions non epistemologiques de l'episte-
mologie (premier niveau), si meme elle implique une phase expli-
cative, un moment critique, qui confere un statut epistemologique
a la comprehension (deuxieme niveau), ne doit-eI1e pas assumer
pour elle-meme la sorte de scientificite qui, depuis I'Idealisme
allemand, est attachee aux arguments transcendantaux? C'est a
quoi invite p~!ffii d'autres R. Bubner dans un article qui s'intitule
precisement Uber die wissenschaftliche Rolle der Hermeneutik.
L'hermeneutique, en effet, est une reflexion sur Ies 'presup-
positions' de toute comprehension du monde, done sur les con-
ditions de possibilite des savoirs qui s'edifient sur cette compre-
210 P. Ricoeur

hension. En ce sens elle est un transcendantalisme, d'un genre

special toutefois, comme l'a montre Apel en opposant a priori
corporel et social aa priori de conscience.
Mais Ie rapprochement devient tout a fait egarant si l'on perd de
vue Ia difference profonde entre Ia pretention inherente au
transcendantalisme de conferer it queIque sujet epistemoIogique la
transparence a soi-meme et Ie systeme des conditions de possibilite
de tous savoirs, y compris Ie savoir sur soi-meme, --- et l'aveu,
essentiel a I'hermeneutique. de Ia radicale non-maftrise et non-
transparence a soi des conditions de tout discours. L'historicite de
ces conditions consiste precisement dans I'impossibilite de Ia
reflexion totale.
C'est en ce sens que Gadamer peut dire au debut de sa Replik:
'si l'hermeneutique est Ia Verstandigung, il est pourtant difficile de
s'entendre sur l'hermeneutique aussi Iongtemps du moins que la
discussion est gouvernee par des concepts non elucides de science.
de critique et de reflexion' (283). La these uItime de l'herme-
neutique philosophique est que cette elucidation ne peut etre
achevee. L'hermeneutique est certes plus qu'un art, eIle est une
Kunstlehre. Comme telle eUe s'apparente it la phrom?sis. qui
ctepasse toute techne. Elle enveloppe donc un savoir de type
reflexif. Mais la reflexion ne saurait devenir Ie tout de cette
activite thcorique. Elle n 'est jamais que Ie segment critique d 'une
operation totale qui ne rompt jamais ses attaches avec une compre-
hension pre-scientifique du monde porte par une communaute
langagiere (Gesprdchsgemeinschaft. 289). C'est it cette compre-
hension que s'attache a titre ultime Ie pretention a I'universalite.
dans la mesure OU aucune limite ne peut etre assignee a la
'communicabilite du comprendre'.
Cette impossibilite de la reflexion tot ale explique sans doute
pourquoi la philosophic hermeneutique est plus sensible a son
histoire q u'aucune autre philosophie. De fait, il est [rappant
que peu d'exposes de philosophie hermeneutique omettent de
commencer par une histoire de Ia question hermeneutique. 16 On
peut certes trouver it ceUe histoire un sens objectif, une teleologie
'eviden te'. C'est celIe d 'une radicalisation et d 'une universalisation
croissante; procedant d'un premier stade OU I'hermeneutique se
reduit <I un recueil de regles pratiq ues dispersees dans les trois
domaines de l'exegese biblique, de Ia philologie classique et de la
 'Logique hermeneutique '? 21 1

jurisprudence, a un deuxieme stade ou I'hermeneutique s'eleve a

une reflexion episemologique sur la comprehension en general,
enfin a un troisieme stade ou l'henneneutique se pense comme
philosophie hermeneutique et ou l'epistemologie s'enracine dans
l'ontologie. Mais cette teleologie de l'histoire de l'hermeneutique
- qui ne differe pas de celle que manifestent d'autres ecoles de
pensee dans leurs developpements - ne constitue pas I'historicite
profonde de l'hermeneutique, it savoir Ie fait qu'elle appartienne
elle-meme it une histoire de la pensee qui n 'est pas entierement
transparente it elle-meme. C'est pourquoi ceUe histoire ne se
n!duit pas it l'autobiographie d'un mouvement, mais constitue bel
et bien une interpretation hermeneutique de sa propre histoire.
Nul ne peut empecher que I'hermeneutique ne soit nee lorsque la
tradition a perdu de sa force nonnative, des l't~poque des
Humanistes et de Luther, puis au temps de l'Aufklarung et du
Romantisme, et que sa fonnulation heideggerienne et post-heideg-
gerienne soit contemporaine d'une crise plus radicale de la trans-
mission culture lIe , par conjonction en tre Ie prestige culturel de la
science et Ie soup<;on inocule au coeur de la culture moderne par
Marx, Nietzsche et Freud.
A ce premier trait de finitude, earacteristique de la 'situation'
de I'hermeneutique - et done constitutive de sa propre Be/indlich-
keit - s'ajoute un deuxieme trait de finitude, propre a son 'dis-
cours', et donc inherent a son propre Verstehen: a savoir que la
philo sophie henneneutiq ue ne peut eehapper a l'histoire de la
metaphysique. Qu'elle insiste sur Ie caractere transcendantal
de sa retlexion, elle s'inscrit dans les philosophies du sujet; qu'elle
traite de la difference entre sciences de I'esprit et sciences de la
nature, elle se situe dans I'heritage de la philosophie hegelienne de
I'esprit objectif; qu'elle tente de rattacher les productions de
l'esprit aux expressions de la vie, eUe s'engage dans une syn these
peut-dre impossible entre philosophie de I'esprit et philosophie de
la vie; qu'elle rapporte au Dasein la comprehension historique du
monde, elle reste peut-etre encore prisonniere d 'une anthropologie
philosophique qui cumule et resume sans les abolir les trois
heritages de Kant. de Hegel et du Romantisme. La 'Kehre' de
Heidegger n'aurait pas de sens sans cet aveu. Quant a la phil-
osophie heideggerienne posterieure a la 'Kehre' et a ses tentatives
successives pour repenser I'ontologie, puis pour transcender celle-ci
212 P. Ricoeur

au benefice d'un penser proche et distant du poetiser, el1e atteste

qu'on n'en a jamais fini avec la tradition speculative de la
philosophie occidentale, et que sa 'destruction' ne peut Hre
qu'une maniere de penser plus originairement ce que cette
tradition a pense.
La est la raison profonde du debat hermeneutique avec sa
propre histoire. Ce debat n'est pas une preface chronologique a
une tache intemporelle. C'est une lutte, au sein de sa conscience
de finitude, pour expliciter ses presuppositions implicites, son
propre non~it et elle sait, a cet egard, qu'une explication integrale
est impossible. Cette finitude de l'interpretation qui affecte Ie
discours de l'hermeneutique sur elle-meme est ce qui empeche
finalement ce discours de s'assimiler a la philo sophie transcen-
dantale.

5. LA CONFRONTATION ENTRE HERMENEUTIQUE ET PHILOSOPHIE

ANALYTIQUE

Nous nous sommes assigne deux taches dans cette chronique:

reflechir sur les conditions et les presuppositions du propre dis-
cours de l'hermeneutique, ~ situer ce discours par rapport acelui
des philosophies pour lesquelles les preoccupations logiques et
epistemologiques sont determinantes. Nous avons acheve la
premiere tache. Quant ala seconde, eUe a ete largement amorcee a
l'occasion de la discussion sur les limites de l'hermeneutique. En
particulier on a commence d'evoquer l'analogie presumee entre
hermeneutique du langage, au sens de Heidegger et de Gadamer, et
'jeux de langage' au sens du dernier Wittgenstein. Le moment est
venu de reprendre Ie deb at pour lul-meme, dans la mesure ou il
est devenu un des enjeux les plus irnportants de la periode post-
heideggerienne.
Que les deux courants de pensee se rencontrent sur la question
du langage est aujourd'hui un truisme. Mais cette rencontre ne
suffit pas a creer une convergence. La divergence semble bien
plutot surgir au lieu de la plus grande convergence, meme si,
ulterieurement, l'approfondissement de cette divergence suggere
des formes plus sub tiles d'interaction.
Karl-Otto Apel,17 qui a ouvert ce debat, propose de prendre
 'Logique hermeneutique '? 213

pour point de depart l'analogie superficielle entre Ia fonction

assignee de part et d'autre au Iangage dans la comprehension du
monde. A partir de hi il institue deux niveaux de confrontation.
Au niveau de Carnap et du Wittgenstein du Tractatus, d'une part,
au niveau des Investigations Philosophiques, d'autre part.
Au premier niveau, on peut resumer toutes les oppositions dans
un rapport inverse entre sens et comprehension. Du cote de l'her-
meneutique, de Luther a Schleiermacher, Ie sens, peut-on dire,
n'est pas douteux; c'est Ie comprendre qui fait probleme. Ou Ie
credit fait au sens demeure, meme Iorsque ceIui-ci a perdu sa
force normative. Ainsi, avec Ddthey, Ie sens est l'expression de
Ia vie infinie. Puis, avec Heidegger, Ie fondement du sens n 'est
plus dans Ia vie, mais dans Ie Dasein qui a toujours deja compris
son insertion dans Ie monde. Avec Gadamer, Ia certitude du sens
precede encore I'in terrogation sur Ie com prendre: Ia beaute de
l'oeuvre d'art m'a deja saisi avant que je la juge, la tradition m'a
deja porte avant que je la mette a distance, Ie Iangage m'a deja
enseigne, avant que je Ie maitrise comme systeme de signes dis-
ponibies. De toutes ces manieres I'appartenance au sens precede
to ute logique du langage. C'est pourquoi l'hermeneutique est
finalement une lutte contre la mecomprehension de ce qui a
toujours deja ete compris, soit que la me-comprehension procede
des confusions vehiculees par la metaphysique (confusion entre
I'etant que nous sommes et les etants subsist ants ou manipulables,
confusion de I'etre en tant qu'etre avec un etant supreme dans
l'onto-theologie; meconnaissance de la difference ontologique
entre etre et etant), soit que la mecomprehension procede de
l'objectivation et de I'alienation methodoIogique.
Pour la philosophie analytique, ce n 'est pas la comprehension
qui est en question, mais la presupposition du sens. La question
est celle du critere de difference entre ce qui peut etre tellU a
priori pour sense ou denue de sens. La seule maniere d 'en decider,
c'est d'etablir les criteres de sens a priori, avant to ute consider-
ation de contenu. lei, d'ailleurs, commencent les divergences a
l'in terieur de la philo sophie analytiq ue: les uns placen t Ie critere
dans la forme logique du langage, les autres dans la verifiabilite
empirique des propositions, d'autres enfin dans une fonne ou
I'autre d'efficacite pratique ou de valeur operationnelle. Mais sous
ces trois formes la 'critique du sens' conduit a la conclusion que la
214 P. Ricoeur

soi-disant pre-comprehension invoquee par la philosophie herme-

neutique ne s'enonce que dans des propositions qui s'averent, non
pas fausse s, mais denuees de sens. Si elles semblent sensees, c'est
parce qu'elles presentent, pour une grammaire superficielle, une
structure semb lable a celIe des propositions porteuses de sens.
A cet egard, les attaques contre l'hermeneutique issues des
partisans du premier critt~re de sens - par la fonne logique du
langage - sont les plus significatives. Ce sont precisement celles
de Carnap et du Tractatus. Leur objection peut se formuler
brievemen t dans les termes suivants: l'hermeneutique ne peut
formuler ses enonces sur Ie sens de i'etrc sans traiter i'etre comme
un predicat, donc sans rctomber dans la confusion deja denoncee
par Kant dans sa critique des preuves classiq ues de I'existence de
Dieu. Face a cette critique, la seule ressource de l'hermeneutique
est de riposter que les criteres de sens q u 'on lui applique com-
portent des presuppositions inavouees qui relevent a leur tour
d'une hermeneutique. Selon celle-ci, les criteres 'logiques' de sens
exprimen tune pensee qui a d 'avance limite tout etre susceptible
d'etre rencontre a des choses subsistantes ou disponibles. Autre-
ment dit la Sprachlogik et la Grammatik restent dans les limites
d'une ontologi des choses subsistantes ou disponibles. Or c'est
une des taches de l'hermeneutique de preserver la difference entre
Ie disponible et ce en vue de quoi il y a du disponible en general.
Et c'est pour preserver cette difference que l'hermeneutique
fait appel a une dimension du langage qui ne passe pas dans
I'analyse proportionnelle et qui est pourtant Ie lieu ou s'opere
l'entente prealable sur Ie sens du disponible et du non-disponible.
La projection de cette dimension de langage sur Ie plan gram-
matical et logique n'en epuise ni l'intention, ni I'efficace. C'est
pourquoi une discussion sur I'usage predicatif du mot etre n'epuise
aucunement Ie 'sens' pre-propositionnel des enonces de l'herme-
neutique, selon la distinction deja evoquee entre Iogique apophan-
tique et logique hermeneutique.
La confrontation au niveau du Tractatus paraft done se
resoudre dans une mutuelle fin de non-recevoir. D'autrcs auteurs,
ncanmoins, ant ten te de discerner certaines convergences entre
Heidegger et Ie Tractatus. Ainsi Jorg Zimmermann,18 eIargissant
une breche ouverte par Fahrenbach, tient pour henneneutique la
question de la condition de possibilite de l'homologie de structure
 'Logique herrneneutique?' 215

postulee par Ie Tractatus entre Ie sens propositionnel et les etats

de chose mondains. A cet egard la these que Ie sens se 'montre',
sans que puisse etre 'dit' ce qui rend possible la fonction de
Darstellung des enonces, definit la cadre hermeneutique du Trac-
tatus. Le these que 'la proposition montre son sens' constitue la
reponse la plus generale de Wittgenstein a la question herme-
neutique des conditions de la comprehension du sens. II faut en
dire autant de son affirmation que 'Ie langage prend soin de lui-
meme'. (o.c. p. 7, 23-4, 50). A egalement une portee herme-
neutique la reflexion sur les lirnites possibles du 'sens', donc sur
la frontiere entre sinnvoll et unsinnig et son implication
'solipsiste', selon laquelle Ie sujet constitue la limite du monde.
Releve enfin de l'hermeneutique Ia these negative selon laquelle
la philosophie ne peut se constituer en un meta-Iangage capable
d"expliquer' son rapport au langage-objet, ainsi que la these posi-
tive qui en est la contre-partie,. selon laquelle Ie rapport reflexif
de la philosophie au langage ne s'expriroe pas dans des propo-
sitions sinnvollen, mais dans des Erlduterungen, qui sont unsinnig,
puisqu'elles font voir sans dire . II faut encore ajouter que Ie fait
que les reflexions de Wittgenstein sur Ie rapport entre dire,
montrer et se taire aient conjugue leur influence avec celles de
Hofmannsthal, de Rilke et de Kafka, constitue un temoignage
indirect sur Ie caractere implicitement hermeneutique des theses
du Tractatus.
Ces notations de Zimmermann ne contredisent pas formellement
celles de Karl-Otto Apel. S'il y a une hermeneutique de Wittgen-
stein, c'est precisement une autre hermeneutique. La question
posee sera precisement de savoir si l'on peut parler d'herme-
neutique en dehors de la tradition 'hermeneutique'.
Le rapport plus apparent entre l'hermeneutique de Heidegger et
les Investigations Philosophiques de Wittgenstein permet de serrer
de plus pres Ie probh~me.
C'est en effet a ce niveau que Karl-Otto Apel voit I'analogie la
plus etroite et la divergence la plus profonde entre les deux
penseurs. L'analogie n 'est pas douteuse: Ie lien etabli par Wittgen-
stein entre jeux de langage et formes de vie n 'evoque-t-il pas Ie lien
decele par Heidegger entre comprehension et souci? L'idee de
Wittgenstein que toute connaissance du monde est articulee au
plan Iinguistique n'a-t-eUe pas un echo dans la notion de dimension
216 P. Ricoeur

iangagiere d u monde chez Gadamer? La critique du langage

prive dans les Investigations Philosophiques ne se prolonge-t-elle
pas dans I'idee de I'entente prealable, de la transmission culturelle,
commune a Gadamer et a ses successeurs, Habermas et Apel com-
pris? Et la theorie des actes de discours ne rejoint-elle pas les
anticipations de Ia logique hermeneutique de Hans Lipps? La
tentative de rapprochement paraft encore encouragee par Ie fait
que les deux ecoles ont pour adversaire commun la metaphysique
c1assique, qu'il s'agisse de la ch0se en soL du sujet absolu, du
so\ipsisme. Pour !'une ce sont des pseudo-problemes que demasque
la grammaire profonde, pour I'autre ce sont les formes les mieux
retrancllees de la me-comprehension propre a la 'metaphysiq ue'
de la philosophie occidentale. Mais un abime se creuse precisement
entre une critique qui dissout Ie probleme comme une illusion
metaphorique ct une pensee qui voit dans la tradition speculative
d~ l'Occident des fonnes dogmatiques alienees qu'il est possible de
reconduire it une fonne plus originaire. ('ette difference profonde
d'appn'ciation elu passe a pour consequence une appreciation tres
differente de la tache meme de la philosophie. Si \'011 a pu avoir
quelque difficu!te a definir I'hermeneutique en terme de reflexion
transcendantale, cette definition n'est pas probh~matique mais tout
simplement impossible a partir des Investigations Philosophiques.
Aux yeux de la philosophic henneneutiq ue, la theorie des jeux
de langage est incapable de reflechir sur ses propres conditions de
possibilite. Si Ie discours est toujours pris dans un jell de langage.
comment la reflexion sur cette condition Iangagiere est-elle
possible? Si la comprehension de I'etre s'identifie a un jeu de lan-
gage. dans quel jeu de langage Ie dit-{)n? Si I'on ne veut pas reciuire
l'activite philo~ophique a une convention verbale, il faut bien
erraciner Ie jeu de langage philosophique dans une comprehension
de I'etre au moncle prealable a tout jeu de langage. ("est la que
!'appreciati()n divergente de la tradition philosophique developpe
toutes scs consequences: si tout langage speculatif vehicule seule-
ment des pseudo-problcmes qu'il faut, non resoudre, mais
dissoudre, n'en est-il pas de me me de la philosophie qui Ie dit?
Karl-Otto Ape\ n'exclut pas qu'une nouvelle convergence puisse
na ftre de cctte ultime divergence. Mais il est encore trop t6t pour
Ie dire. Tout au plus peut-{)n faire la double suggestion suivante;
tout<: tentative de la philosophic analytique pour echapper au pur
 'Logique hermeneutique '? 217

conventionalisme, ou pour rendre compte de la reflexion fonda-

mentale du langage sur ses propres determinations et ses propres
institutions, pointe en direction d'une hermeneutique du langage.
Inversement, tout effort de la philosophie hermeneutique pour
definir les conditions d 'une inscription propositionneUe de sa
logique hermeneutique paraft devoir ramener vers une retlexion
sur Ie rapport, aussi bien createur que critique, entre Ie jeu de
langage de la philosophie et les autres jeux de langage insti-
tutionalises. L'enjeu de ce debat c'est Ie rapport de la philosophie
avec sa propre histoire. La philosophie du langage marque-t-eUe
une rupture sans precedent avec I'ensemble de la tradition, it
savoir une rupture telle que cesse meme d'etre compris ce avec
quoi elle rompt? Ou bien, la 'destruction de la metaphysique' est-
elle Ie chemin d'une reappropriation mediatisee de la tradition de
la philosophie speculative, qui permettrait au dialogue historique
de l'humanite avec eUe-meme de se poursuivre? C'est sur cette
question que s'acheve la double confrontation de Karl-Otto
Apel entre Sinnkritik et Sprachhermeneutik.
C'est une autre issue que Zimmermann entrevoit: l'herme-
neutique des jeux de langage lui paraft d'autant plus feconde
qu'elle ne doit rien a la tradition romantique, ni a l'entreprise
philosophique de Heidegger, Lipps et Gadamer, mais procede
exclusivement d'une reflexion critique sur ses propres prejuges
qui rejoignent ceux de la pensee rationaliste-empiriste. C'est it ce
prix que cette hermeneutique apporte un complement et un cor-
rectif a I'hermeneutique. Le caractere hermeneutique de la
reflexion de Wittgenstein reside pour l'essentiel dans la desig-
nation du langage naturel comme base et horizon de toute compre-
hension de sens. On peut, de cette fayon, parler ce cercle herme-
neutique chez Wittgenstein. 11 etait present, des Ie Tractatus, dans
la these de l'isomorphisme entre proposition et etat de chose, ainsi
que dans son eorollaire l'impossibilite de philosopher du dehors
sur Ie langage. Mais, avec les Investigations Philosophiques, I'arti-
eulation de la pre-comprehension inc1use dans Ie langage ne con-
siste plus dans Ie 'laisser-voir" d'une structure apriorique d'une
nature onto-logique, mais dans la reconnaissance des pre-
suppositions [actuelles du parler dans Ie contexte des formes de
vie. L'auteur, il est vrai, observe que Ie rapprochement possible
entre Wittgenstein et Heidegger-Gadamer ne vaut que dans la
218 P. Ricoeur

mesure ou l'on peut ecarter une interpretation pragmatiste, voire

behavioriste, des Investigations Philosophiques. Or, Ie recours ala
'grammaire profonde' ne s'oppose pas moins aux criteres 'externes'
du comportement qu'aux criteres 'internes' du psychisme prive.
Le behaviorisme des lors n'est que l'envers du mentalisme (224-
237), L'henneneutique des Investigations Philosophiques consiste
bien plutot dans Ia contingence de l'ordre grammatical, en tant
quil constitue l'a priori factuel qui precede tout rapport au monde.
Zimmennann est certes d 'accord avec Apel pour denoncer un con-
flit profond entre Ia pretention it l'universalite de Ia grammaire
philosophique qui exclut toute reflexion du dehors, et sa pn~ten­
tion critiq ue a detruire un jeu determine de langage, celui de la
'metaphysique' (251). Mais les reserves importantes que Zimmer-
mann fait a I'hermeneutique de Wittgenstein importent moins que
Ies consequences qu'il tire pour I'hermeneutique elle-meme de la
simple existence d 'une henneneutique qui s'est constituce hors du
champ de la reflexion sur les conditions de possibilite des sciences
de I'esprit.
A cet egard, les importantes convergences evoquees plus haut
doivent servir a dessiner Ie cadre d 'une henneneutique plus large
dans laquclle I'henneneutique des sciences de l'esprit recevrait,
comme on l'a dit, un complement et surtout un correctif. Le com-
plement apparait plus net si I'on ne borne pas, comme Karl-Otto
ApeI, a Ia comparaison entre Sein und Zeit et les Im!estigations
philosophiques. La conception de I'autonomie du langage, selon
Untenvegs zur Sprache, fait paraitre sous un jour nouveau I'affir-
mation de Wittgenstein que Ie langage 'prend soin de lui-meme'.
Une parente plus secrete encore se decouvrirait si l'on faisait
entrer en ligne de compte les ecrits de Wittgenstein sur l'ethique,
l'esthetique et la religion et tout ce qui, chez lui, se laisse inscrire
SOllS Ie titre de ce qui se montre, voire de cc qui se tait et qui,
peut.etre, parle autrement. 19 Mais Ie correctif importe peut-etre
plus que Ie complement. L'interet de I'hermeneutique de Wittgen-
stein c'est qu'elle n'a pas son lieu originel dans la problematique
des sciences de l'esprit et ne prend pas pour axe la transmission
historiquc. Pour eIle, l'historicite est seconde, pour ne pas dire
suspecte par principe, par rapport ala facticite des jeux de langage
quotidiens. Le moindre effet de l'inclusion de Wittgenstein dans Ie
champ hermeneu tiq ue, c 'est de contester Ie privilege de I'his-
 'Logique hermeneutique'? 219

toricite des mediations langagil!res. - On peut seulement se

demander si Ie divorce n'est pas sans remede entre une philosophie
pour laquelle quelque chose continue de nous parler dans la
tradition speculative de l'Occident, et une philosophie qui met
l'accent principal sur la pratique: actuelle du discours dans un con-
texte de quotidiennete. On voit ainsi se dessiner, a la faveur de
cette discussion, une certaine ramification du concept meme
d'hermeneutique, ou son identih~ risque de se perdre.
Est-ce d'abord une Kunstlehre de la comprehension langagiere
pour laquelle I'in terpretation des textes reste Ie modele privilegie?
Ou bien Ie langage lui-meme est-il secondaire par rapport a une
'experience comprehensive' qui peut etre non-verbale, comme
dans Ie cas de la jouissance esthetiq ue?
Si l'on choisit la premiere branche de I'alternative , faut-il dire
que la tache de la comprehension est de 'repenser' ce qui a etc
une fois 'pense', de 'reproduire' ce qui a ete une fois 'produit'?
Ou bien la comprehension est-eile 'productrice' de sens dans la
nouveaute des contextes de culture et de vie? Dans la premiere
hypothese, l'experience de base reste la mediation historique,
c'est-a-dire finalement la tradition qui s'adresse a nous et nous
donne a penser avant que nous soyons en etat de la reflechir. Dans
la deuxieme hypothese, la mediation historique est seconde par
rapport a la contextualisation nouvelle du discours dans les
situations inectites de la vie quotidienne.
Mais la premiere hypothese se divise a son tour en deux
branches: a l'egard de la transmission culturelle, l'accent est-il
sur I 'auto rite de la tradition, porteuse de sens et de verite possible?
Ou bien sur l'instance critique qui transforme tout heritage rec;:u en
une 'circonstance'?
A cette alternative s'ajoute la suivante: l'hermeneutique est-elle
universelle, pour la raison que: toute comprehension est de sa
competence? Ou bien est~lle limitee, s'il est vrai que toutes les
structures de I'existence tant sociales qu'individueiles ne sont
pas intentionelJes? Dans la premiere hypothese, elle peut admettre
des mediations innombrables, s'etendre et s'approfondir, sans
cesser d'etre interpretation. Dans la deuxieme hypothese, la com-
prehension est un mode limite qui laisse hors de lui l'explication
et appelle la mediation d 'une tierce discipline, dialectique ou
critique.
220 P. Ricoeur

La meme alternative se redouble dans la suivante: si toute com-

prehension est historiquement mediatisee, faut-il dire que Ia
reflexion de l'henneneutiq ue sur elle-meme l'est aussi? Ou bien
faut-il dire que I'on ne comprend la relativite contextuelle de
l'interpretation qu'a Ia lumiere de I'idee regulatrice de discours
'rationnel' ou de 'communication sans borne et sans entrave'?
Dans Ie premier cas, I'henneneutique avoue sa propre finitude, au
coeur meme de sa pretention universelle, autrement dit son
caractere 'epochal', OU se reflete sa dependance a l'egard des
'epoques' de I'etre et du penser. Dans Ie deuxieme cas, I 'her-
meneutique se transcende elle-meme dans la reflexion et se
n.~inscrit dans la tradition de la philosophie transcendantale. Dans
Ie premier cas, l'henneneutique de l'hermeneutique reste fidele a
sa these de base, mais exclut toute scientificite. Dans Ie deuxieme
cas, elle plaide pour un concept de scientificite distinct du concept
(au moins anglo-saxon) de science, mais elle renie sa propre these
de la priorite de la pre-comprehension sur la reflexion.

NOTES ET BIBLIOGRAPHIE

1. Lipps, H., Untersuchungen zu einer hermeneutischen Logik, Frankfurt

1938, 1959 2 • Extrait in Seminar: Philosophische Hermeneutik, ed.
H.-G. Gadamer et G. Boehm, Frankfurt 1976, 286--316. Otto F.
Bollnow precise ses rapports a la 'philosophie de la vie' de G. Misch dans
'zum Begriff der hermeneutischen Logik', in: Hermeneutische
Philosophie, ed. O. Poggeler, Munich 1972,100-122.
2, Outre les dernieres publications d'inedits par Heidegger lui-meme et Ie
debut de la publication de la Gesamtausgabe, Frankfurt/Main (1975- )
(4 vol. deja publics en 1978), on notera les travaux consacres a I'inter-
pnhation de sa philosophie: O. P6ggeler, ed., Heidegger. Perspektiven
zur Deutung seines Werkes, Cologne/Berlin 1969. Du meme: Herme-
neutische Philosophie (Textes de Dilthey, Heidegger, Gadamer, Ritter,
Apel, Habermas, Ricoeur,O.Becker,BoIlnow),Munich 1972, -Tugend·
hat, E. Der Wahrheitsbegriff bei Husserl und Heidegger, Berlin 1967.
--H.-G, Gadamer, 'Martin Heidegger und die Marburger Theologie', in:
Heidegger, Perspektiven .. , -- Kockelmans, J J., ed., On Heidegger and
Language, Evanston 1972.
3. Gadamer, H.-G. Wahrheit un..1 Methode. Tiibingen 1960,1965 2 ,1967 3 .
On tiendra compte des importantes 'Introductions' de la 2e et de la 3 e
ed. Du meme: Kleine Schriften, I, II, Till:Jingen 1967
4, Bubner, R. 'Uber die wissenschaftstheoretische Rolle der Hermeneutik',
in: Bubner. R., Dialektik und Wissenschaft, Frankfurt 1973,89-111.
 'Logique Izermeneutique '? 221

Seule une interpretation transcendentaliste de l'hermeneutique, peut,

selon l'auteur, repondre a la vive attaque de H. Albert, Traktat uher
kritische Vernunft, Tiibingen 1968. Du meme: 'Transzendentale Her-
meneutik?', in: Simon-Schafer und Zimmerli, ed., Wissenschaftstheorie der
Geisteswissenschaften, Hamburg 1975, 56- 70.
5. Le dossier de ce deb at a ete rassemble par Habermas, Dieter Henrich et
Jacob Taubes, sous Ie titre: Hermeneutik und Ideologiekritik. Theorie-
Diskussion, Frankfurt 1971. Cf., en outre, Betti, E., Allgemeine Aus-
legungslehre als Methodik der Geisteswissenschaften, Tiibingen 1967
(Teoria generale della Interpretazkme, Milan 1955).
6. Habermas, J. 'Zur Logik der Sozialwissenschaften', Philosophische
Rundschau, Beiheft 5, Tiibingen 1967. Extrait in Hermeneutik und
Ideologiekritik, 'Zu Gadamers "Wahrheit und Methode"', 45-56.
- Der Universalitatsanspruch der Hermeneutik, in: Bubner, R., Cramer,
K_, Wiehl, R., ed_, Hermeneutik und Dialektik, Festschrift fUr H.-G.
Gadamer, Tiibingen 1970, I, 73-1 04;repris in Hermeneutik und Ideo logie-
kritik, 121)-159. Du meme: Erkenntnis und Interesse, Frankfurt; Theorie
und Praxis, Frankfurt; Technik und Wissenschaft als 'Ideologie', Frankfurt
1968; 'Vorbereitende Bemerkungen zu einer Theorie der kommunika-
tiven Kompetenz', in: Habermas, 1., and Luhmann, N., Theorie der
Gesellschaft oder Sozia/technologie.?, Frankfurt 1971, 101-142.
7. Apel. K.O., 'Scientistik, Hermeneutik, Ideologiekritik', Wiener Jahrbuch
fiir Philosophie, Bd. I, 1968, 15-45. Repris in: Hermeneutik und
Jdeologiekritik, 7-44, et in: Transformation der Philosophie, Frankfurt
1973, 1. II, 96-127. - Cf. Kisiel, Th., 'Zu einer Hermencutik natur-
wissenschaftlicher Entdeckung', in: Zeitschrift fiir allgemeine Wissen-
schaft5theorie, Bd. II, 195 sg. - Radnitzky, G., Contemporary Schools of
Metascience, Goteborg 1968.
8. Dray, W., Laws and Explanation in History, Oxford 1957; Danto. A.,
A nalytical Philosophy of History, Cambridge 1965.
9. Gadamer, H.-G., 'Rhetorik, Hermeneutik und Ideologiekritik', in: Herme-
neutik und Ideologiekritik, 57-82, repris in: Kleine Schriften II, 113-
130. 'Replik', ibid., 283-317. - Pour une critique de Gadamer, cf.
Pannenbcrg, W. 'Hermeneutik und Universalgeschichte', in: Zeitschrift
fiir Theologie und Kirche, 1963, 90-121; Hirsch, E.D., Validitv in
Interpretation, New Haven/London 1967. - Comparez avec la philo-
sophie du langage de Lohmann. - Sur Ie rapport general de l'herme-
neutique aux sciences, Seminar: die Hermeneutik und die Wissen-
scha/ten, ed. Gadamer, H.-G. et Boehm, G., Frankfurt 1978.
10. J!luss, H.R., Kleine Apologie der iisthetischen Erfahrnng, Konstanz 1972:
Asthetische Erfahrung und literarische Hermeneutik, Munich 1977.
- Cf. egalement Iser, W., Der implizite Leser, Munich 1972; Der Akt
des Lesens, Munich 1976. -- Stierle, K. Text als Handlung, Munich
1975. - Lotmann, J.M., Die Struktur literarischer Texte, Munich 1972.
- M~kai6vsky, J., Kapitel aus der Poetik ,Frankfurt 1970. Kapitel aus
der Asthetik, Frankfurt 1970. - Kosellek, K.R. und Stempel, W.-D.,
Poetik und Hermeneutik, Arbeitsergebnisse einer Forschungsgruppe,
222 P. Ricoeur

t. I---VIII, Munich (1970-1977). Coseriu, E., Thesen tiber Sprache und

Oichtung', in: Stempel, W.O., ed., Textlinguistik, Munich 1971. Blumen-
berg, H., Der Prozess der theoretischen Neugierde, Frankfurt 1973.
11. Adorno, Th.W.,A·sthetische Theone, Frankfurt 1970.
12. Ritter, J., Metaphysik und PolWk. Studien zu Anstoteles und Hegel,
Frankfurt 1969, 1972 2 . Extrait: 'Politik und Ethik in der praktischen
Philosophie des Aristoteies', in: O. Poggeler, ed., Hermeneutische Phi-
losophie, 153-176. - Cf. O. Poggeler, 'Die ethisch-poiitische Dimension
der hermeneutischen Philosophie', in: G.G. Grau. ed., Probleme der
Ethik, Freiburg/Munich, 45-81. Ou meme: Philosophie und Politik bei
Heidegger, Freiburg/Munich 1972.
13. Sur les implications de i'hermeneutique dans Ie domaine de I'exegese
biblique et de la theologie chretienne, cf. Esbroeck, M. van, Herme-
neutique, structuralisme et exegese, Paris 1968; Lorenzmeier, B., Exegese
und Hermeneutik. Eine vergleichende Darstellung der The%gie Rudolf
Bultmanns, Herbert Brauns und Gerhard Ebelings, Hamburg 1968.
Loretz, 0., und Strolz, W., ed., Die hermeneutische Frage in der Theo-
logie, Freiburg 1968. Stachel, G., Die neue Hermeneutik. Ein Uberblick,
Munich 1968. Fruchon, P., Existence humaine et Revelation. Essais
d 'Hermeneutique, Paris 1976. Biser, E., Theologische Sprachtheone
und Hermeneutik, Munich 1970. Gerhard Ebeling. Wort und Glaube, t.
II. TUbingen 1969.
Sur Ie renouvellement de la metaphysique par l'hermeneutique, cf.
Coreth, E., Grundfragen der Hermeneutik, Freiburg 1969; Seebohm,
ThM., Zur Kritik der henneneutischen Vernunft, Bonn 1971; Pareyson,
L., VeritiL e interpretazione, Milan 1971; Theunissen, M., Hegets Lehre
vom absoluten Geist als theologisch-politischer Traktat, Berlin 1970.
14. Ricoeur, P., 'Qu'est-ce qu'un texte? Expliquer et comprendre', in:
Hermeneutik und Dialektik, II, 181-200; De I1nterpretation_ Essai
sur Freud, Paris 1965; Ie con/lit des Interpretations, Paris 1969; La
metaphore vive, Paris 1977. - On rapprochera ces ecrits de ceux de
quelques auteurs franyais qui ne se reclament pas de la tradition
herm~neutique; Levinas, E., Totalite et Infini, La Haye, 1961; Au dela de
l'essence au autrement qu 'etre, Paris 1976. Maldiney, H., A ftres de fa
langue et demeures de fa pensee, Lausanne 1975. Henry, M., L 'Essence
de la Manifestation, Paris 1963; Marx, 2 vol., Paris 1976. Oans la meme
direction, en langue anglaise, Taylor. Ch. Explanation of Behaviour, New
York 1964.
IS. Von Wright, G.H. Explanation and Understanding. Ithaca, N.Y.: Cornell
University, 1971.
16. Henrichs, N_, Bibliographie der Hermeneutik und ihrer Anwendungs-
bereiche seit Schleiermacher, DUsseldorf 1968. -- Le fameux travail de
Dilthey - Ie Preisschrift de 1860 -- Leben Schleiennachers (publie
seulement en 1966 par Redeker, M. (Gesammelte Schriften t. XIV, 2)
etait deja une histoire de I'hermeneutique. De meme I'abrege de 1900,
Die Entstehung der Hermeneutik (G.S., t. V, 317-338). Gadamer,
 'Logique hermeneutique'? 223

H.-G., Artikel 'Henneneutik', in: Historisches Worterbuch der

Philosophie, tid. Ritter. J., Bale. Stuttgart 1974; du meme: 'Henneneutik',
in: Contemporary Philosophy: A Survey, ~d. Klibansky, R., Florence
1969, 360-372. - P6ggeler, ()., 'Einfiihrung', in: Hermeneutische
Philosophie, 7-71. Munich 1972. - On trouvera dans Poggeler, 0., et
Boehm, G., Seminar: Philosophische Hermeneutik, Frankfurt 1976, une
serie d'extraits d'auteurs appartenant ala 'prehistoire de l'hermeneutique
romantique', a l"henncneutique romantique', a 'Dilthey et son ecole',
enfin a quelques contemporains dont H. Lipps. Cf. egalement Palmer,
R.-E., Hermeneutics: Interpretation Theory in Schleiermacher, Dilthey,
Heidegger, and Gadamer, Evanston 1969; Coseriu, E., Die Geschichte der
Sprachphi/osophie von der Antike bis zur Gegenwart. Eine Ubersicht.
I, Stuttgart 1969; II, Tiibingen 1972.
17. Apel, K.O., 'Wittgenstein und Heidegger. Die Frage nach dem Sinn von
Sein und der Sinnlosigkeitsverdacht gegen aller Metaphysik' (1967);
'Heideggers philosophische Radikalisierung der 'Hermeneutik' und die
Frage nach dem 'Sinnkriterium' der Sprache' (1968); 'Wittgenstein und
das Problem des hermeneutischen Verstehens' (1965), in: Transfor·
mation der Philosophie, Frankfurt 1973, t. r, 225-335.
18. Fahrenbach, H., 'Die logisch-herrneneutische Problemstellung in Wittgen-
steins "Tractatus" " in: Hermeneutik und Dialektik, Tiibingen 1970, t. II,
25-54; 'Positionen und Probleme gegenwartiger Philo sophie , Tei! II:
Philosophie der Sprache', in: Theologische Rundschau, 1970,277-306;
1971, 125-144,221-243. - Zimmerman, Jorg, Wittgensteins sprach·
philosophische Hermeneutik, Frankfurt 1975.
19. Janik, A. et Toulmin, St., Wittgenstein's Vienna, New York 1973; Brand,
G., Die grund/egenden Texte von Ludwig Wittgenstein, Frankfurt 1975;
Bouveresse, J., La Pel/see silencieuse, Paris 1977.
 PART TWOjDEUXIEME PARTIE

Philosophical logic/Logique philosophique

Introduction
G.H. von WRIGHT
University of Helsinki

1. Logic is traditionally a part of philosophy and should in my

opinion continue to be thus regarded. After half a millennium of
relative stagnation, pace Leibniz, it awoke to fresh life after the
middle of the nineteenth century. There can be little doubt that,
viewing things in a broader historical perspective, the revival of
logic is the most striking new development in philosophy in the
last one hundred years.
As a consequence, logic has not only vastly expanded in breadth
but also branched off in direction which are of lesser relevance to
philosophy. One sometimes refers to this with the terms philo-
sophical and mathematical logic. They should, however, be used
with caution" "'Philosophic logic" - There is no such thing',
Wittgenstein wrote in an often quoted letter to C.K. Ogden.
Something similar could perhaps be said about the term 'mathe-
matical logic'. But it is also a fact that much of what today goes
under the name 'logic' belongs to mathematics proper and should
not count as philosophy. The borderline may be difficult to draw.
Still the division is there. With this in mind the Editors of the
present volume of the Chroniques have used the title Philosophical
Logic to adumbrate the area within which the survey articles, here
published, fall.
2. Philosophical logic is sometimes defined as an application of
logic to the analysis of concepts and conceptual structures in
which philosophers have traditionally taken an interest. This defi-
nition fits perhaps best the logical study of modal concepts and
its several 'off-shoots' into related conceptual domains. (See
below.) Pursuits in this spirit sometimes shed interesting light on
old problems of philosophy. More often perhaps they give rise to

Contemporary philosophy. A new survey. Vol. 1, pp. 227-233.

© 1981, Martinus Nijho[[ Publishers. The Hague/Boston/London.
228 C.H. von Wright

new problems and steer the interests of philosophers in new

directions. The characterization of formal logic as a tool and of
philosophical logic as an application of formal methods to phil-
osophy is therefore too narrow and easily misleading. In my
opinion philosophical logic is better described as a form of phil-
osophy, characteristic of contemporary thought but with affinities
to a climate of opinion which also prevailed in philosophy during
some periods in Ancient and Medieval times.
3. The discipline, or better: trend, which I here call philo-
sophical logic, can be assigned a birthdate. This is the year 1879
when Frege published Begriffsschrift. Frege's work was not by any
means the first major contribution to the revival of logic, however.
More than thirty years earlier (1847) Boole had published his
Mathematical Analysis of Logic. But the tendencies of the two
authors pointed in opposite directions. Boole was interested in
what was mathematical about logic. His aim was to establish 'that
the ultimate laws of logic are mathematical in their form'. Frege's
concern was with what is logical about mathematics. He wanted to
show that truth and proof in mathematics 'sich allein auf die
Gesetze griindet, auf denen aIle Erkenntnis beruht'.
As a simplification one could therefore say that Boole was the
founding father of 'mathematical', Frege of (modern) 'philo-
sophical' logic.
Frege's efforts renect an interest in foundations and a quest for
increased rigor which is characteristic of almost all the major
innovators in mathematics of the nineteenth century. Two great
mathematicians after Frege with whom the same philosophic urge
took a somewhat different turn were Hilbert and Brouwer.
Developments in logic during the fIrst three decades of this
century mainly followed the currents of the three mainstreams of
research into the foundations of mathematics, viz. the logicism of
Frege and Russell, the formalism of Hilbert, and the intuitionism
of Brouwer.
With time the philosophic drive behind the three mainstreams
either faded or was modified, the mathematical and 'technical'
aspects associated with them assuming greater prominence. - The
program of the 'logicists' to reduce mathematics to logic gradually
merged with axiomatic set-theory. A major work from a later date
in the philosophic tradition inaugurated by Frege and Russell is
 Introduction 229

Quine's Set Theory and Its Logic (1963). At about the time of its
publication axiomatic set-theory entered a new phase of lively
development stimulated by the sensational proof by Paul Cohen
of the independence of Cantor's Continuum Hypothesis.
Hilbert's grand idea was to exhibit the whole of classical mathe-
matics in the form of an edifice of ax iomati zed and formalized
calculi. His program called for the creation of a meta-mathematics
in which the consistency and other 'perfection properties' of the
mathematical calculi (including formalized logic) could be estab-
lished. Under the impact of Godel's discoveries in the early
1930s the philosophical program of the formalist school became
in part discredited - but from its modifications grew a major part
of what now counts as 'mathernaticallogic' proper: formal seman-
tics and model theory, theory of computability and recursive
functions, theory of automata (mathematical machines). In the
course of these developments, influences from the seHheoretic
trend which grew out of logicism and from the constructivist
trend inherent in intuitionism have to a large extent become
intertwined with the proof-theoretic aspirations of Hilbertian
formalism. - In the paper by Dag Prawitz in this volume the
reader will fmd an evaluation of the philosophic relevance of
contemporary proof-theory.
It is probably right to say that the ideas underlying intuitionism,
more than the foundation programs of Frege and Russell and of
Hilbert, continue to have a seminal influence on philosophical
logic. From the beginning these ideas represented a counter-trend
both to logicism and formalism. (Poincare attacked Russell and
Brouwer Hilbert.) Wittgenstein's criticism in his later writings of
every program for a 'foundation' of mathematics may be regarded
as a continuation of a tradition inaugurated by intuitionism and
semi-intuitionism early in this century. The constructivism of
intuitionist mathematics also has affinities with the thoughts of
Wittgenstein.
In recent years ideas stemming partly from Brouwer and partly
from Wittgenstein have been elaborated for purposes of theories of
truth and linguistic meaning. The most influential representative
of this new trend in philosophical logic and the philosophy of
language has been Michael Dummett.
4. I would regard the revival of modal logic as the philosophically
230 G.H. von Wright

most important new development in logic in the mid-century and

after. Early precursors of this revival were c.I. Lewis and Jan
l.ukasiewicz. With Lewis modal logic emerged as a by-product of
his efforts to construct a propositional logic which better than the
one of Frege and Russell captured the notion of 'implication'.
With Lukasiewicz the theory of modality was linked with ideas of
a many-valued or polyvalent logic. Thus both authors conceived of
modal logic as an alternative to 'classical' truth-logic. The label
'non-classical' for modal logic, still sometimes attached to it, is
unfortunate, however. It makes the subject appear in a distorted
perspective. A correction was inlplicit already in a short paper by
Godel 2 from the early 1930s. In it modal propositional logic is a
super-structure based on the Russellian propositional logic. This
conception has now become standard.
Further important developments took place in the 1940s with
the creation of a modal predicate calculus, and in the 1950s with
the analogical extensions of modal logic to the study of cllfono-
logical, deontic, doxastic, and epistemic discourse. The first line of
development became a target of forceful criticism by Quine. This
in turn stimulated a more thorough scrutiny of the subject's philo-
sophical foundations also by those who 'defended' modal logic
against its critics. The second line has brought within the orbit of
logical study a great number of concepts and problems familiar
from traditional epistemology, ethics, and metaphysics. It has
therefore done much to enhance the importance of logic to
philosophy in general.
Since the days of Lewis several systems of ('alethic') modal
logic were known to exist. Could one of them claim to be 'the
true' logic of the modal notions? Intuition seemed an unsafe guide
to answering this question. A semantics for modal logic(s) was
needed. It was partly the criticism by Quine which stimulated
efforts to develop it. Its breakthrough dates from the years round
1960 and the pioneers were Hintikka, Kanger, and Kripke. The
new field which it opened up has become known as possible
worlds semantics. This has turned out to provide a singularly
useful framework for dealing, not only with pro blems in traditional
modal lOgiC, but also with other forms of 'intensional' discourse,
induding the notorious difficulties presented to logic by counter-
factual (and other) conditionals. (The reader is referred to the
 Introduction 231

article by Hilpinen in the present volume.) At the philosophical

level possible worlds semantics has challenged reflexion about the
nature of necessity and the other modalities and about indi-
viduation, identity, naming, and reference. Much of this work has
drawn inspiration from ideas first put forward by Kripke.
Semantical methods in modal logic and their philosophical
implications are discussed below in the article by Ruth Barcan-
Marcus, herself a major name in the modern history of the subject.
A line of development, related to modal logic and yet distinct
from it, is represented by relevance logic. It was founded by Alan
R. Anderson and Nuel Belnap in collaboration, and may be said
to continue the early efforts of C.1. Lewis (see above) to capture
by formal means the meaning of 'implication' (entailment).
Themes of thought stemming from intuitionism are also further
developed within relevance logic - as evidenced by the survey
article by Belnap and Dunn in this volume.
5. Particularly in the Anglo-Saxon tradition it has been cus-
tomary to divide logic into deductive and inductive. Inductive
logic is closely allied to the study of scientific method and to the
philosophy of the natural and 'moral' sciences. A prominent place
withm it was traditionally held by the logic and philosophy of
probability.
Much work in inductive logic has been inspired by Carnap's
conception of probability as the degree of confIrmation which
some 'evidence' confers on a "hypothesis'. Confirmation theory
in the spirit of Carnap became soon linked with decision theory
and personalist or subjective probability theory. These latter
branches are in turn connected with the logical study of pref-
erence and value (utility). Some main developments in this rather
heterogeneous field of research are described in the article by
Jeffrey on choice, chance, and credence.
Other formalist trends within the philosophy of science are
farther removed from 'inductive logic' in the traditional sense.
Mention should here be made of two such trends which have
acquired prominence .in recent years. One is the formal approach,
inaugurated by Sneed and StegmUller, to describing theory
ch,mge. 3 This approach takes its point of departure in the debate
originally stirred by Thomas Kuhn's influential work on 'scientific
revolutions'. The second trend that should here by mentioned is
232 C.R. von Wright

the (re-awakened) interest in the logica! foundations of quantum

physics. One of the issues in this area is whether many-valued or
polyvalent logics can be used for purposes of clarifying the logical
structure of the micro-world.
6. Rapid developments and profound changes in a subject
usually at first provoke a rejection of the subject's past to be
succeeded later on by a renewed interest in it. This is eminently
true of modern logic.
The logic of Aristotle and the Schoolmen, long considered
obsolete and then superseded by modern developments, has
turned out to be, not only a goldmine for historical research, but
also a source of inspiration for creative work in philosophical
logic. This can he seen, for example, [rom the numerous writings
in the last twenty years or so about Aristotle's 'problem of the
sea-battle' (the futura contingentia of the scholastics), about the
Master Argument of Diodorus, and about the idea known as the
Principle of Plenitude. A memorable pioneer in this field was
Arthur N. Prior (d. 1969).
The revival of modal logic in particular has stimulated research
in the history of logic. It is but natural that the creation of a
modal semantics of 'possible worlds' should also have led to a new
interest in the logical work of the philosopher, Leibniz, who was
the first to use the term.
Finally, mention ough t to be made of the lively interest which
the logical doctrines of Frege, the principal founding father of
modem philosophical logic (see above), continues to attract.
Frege's works have been reprinted and his posthumous writings
and correspondence published. Thorough examinations of Frege's
philosophy have been undertaken by M. Dummett and others. The
strong interest in Frege must also be seen against the background
of recent developments in linguistic theory - including attempts
to apply formal methods to the study of natural languages. A
pioneer of importance in this study was Richard Montague
(d. 1970).
 Introduction 233

NOTES

1. Cf. the comments on the terminological situation by Prawitz in his essay in

the present volume, p 235.
2. 'Eine Interpretation des intuitionistischen AussagenkalkUls', Ergebnisse
eines mathematischen Kolloquiums, 4, 1933.
3. C'f. the section 'Theory Change' in the Proceedings of the fifth Inter-
national Congress of Logic, Methodology and Philosophy of Science,
Vol. 4, Reidel, Dordrecht 1977.

SELECTED BIBLIOGRAPHY

Thi~ list includes only a few selectt~d works which are either of value as
surveys or have been of seminal importance to the recent development of the
subject. Some landmarks are also listed from branches of philosophical logic
which are not covered by the survey essays in the present vvlume.

Cohen, P.J., Set Theory and the Continuum Hypothesis. New York 1966.
Dummett, M., Frege, Philosophy of Language. London 1973.
- Elements of Intuitionism. Oxford 1977.
Frege, G., Nachgelassene Schriften und wissenschaftlicher Briefwechsel, 1-2.
Hrsg. von Hans Hermes, Friedrich Kambartel und Friedrich Kaulbach.
Hamburg 1969-1976.
Hintikka, J., Time and Necessity. Oxford 1973.
Lewis, D., Counterfactuals. Oxford 1973.
Mostowski, A., Thirty Years of Foundational Studies. Oxford 1966. Acta
Philosophica Fennica, fasc. XVII.
Plantinga, A., The Nature of Necessity. Oxford 1974.
Prior, A.N., Past, Present and Future. Oxford 1967.
Quine, W. van Orman, Set Theory and its Logic. Cambridge, Mass., 1963.
Wang, Hao, From Mathematics to Philosophy. London 1974.
Wittgenstein, L., Bemerkungen iiber die Grundlagen der Mathematik. Frank-
furt 1974. (Remarks on the Foundations of Mathematics, revised and
enlarged edition, Oxford 1978.)
Modal and Many-Valued Logics. Acta Philosophica Fennica, fasc. XVI.
Helsinki 1963. Contains several papers which have been of seminal
importance to later developments - by Hintikka, Kripke, Montague, and
others.
Logic and Philosophy. Proceedings of the Colloquium of Institut Inter-
national de Philosophie in Dusseldorf 27 August - 1 September 1978.
The Hague 1980. Contributions by E. Agazzi, R. Barcan-Marcus, M.
Dummett, R. Hilpinen, R.M. Chisholm, D. Prawitz, W. Stegmiiller,
P. Suppes. Supplements the picture of recent work in philosophical
logic given by the survey articles in the present volume.
Philosophical aspects of proof theory
DAG PRAWITZ
University of Stockholm

I. INTRODUCTION. DIFFERENT DIRECTIONS OF THE SUBJECT

1. The subject and its name

The term proof theory, in Gennan Beweistheorie, was introduced

by Hilbert [1] in 1922 as the name of a subject which he had
already described in a lecture [2] in 1917 as 'an important, new
field of research' where 'we must make the concept of specific
mathematical proof itself object of investigation, just as also the
astronomer pays attention to his place of observation, the
physicist must care about the theory of his instrument, and the
philo sopher criticizes reason itsdf'.
Hilbert had great expectations of this new discipline. In [2]
he mentions a number of questions that belong to it: 'the problem
of the solvability of every mathematical question, the problem of
checking the result of a mathematical investigation, further the
question of a criterion for the simplicity of a mathematical proof,
the question about the relation between matter and form (lnhalt-
lichkeit und Formalismus) in mathematics and logic, and finally,
the problem of the decidability of a mathematical question by a
finite number of operations.' He even thought he had found a
solution of the continuum hypothesis by the help of his proof
theory [3 J. But first of all, proof theory was meant to be the
essential ingredient in Hilbert's program. The aim of this program
was to obtain a finitary foundation of mathematics, and the way
in which this was to be done has been variously formulated as
showing that the use of transfinite principles leads to correct
finitary results or that all transfinite elements can be eliminated in

Contemporary philosophy. A new survey. Vol. 1, pp. 235-277.

©1981. Martinus NiihoffPublishers. Tile Hague/Boston/London.
236 D. Prawitz

principle from proofs of finitary results or, more simply, as

demonstrating the consistency of mathematics, which problems
are all eq uivalent under certain conditions.
In spite of Hilbert's greater ambitions, to which also Bernays
[4] has recently called attention, proof theory has to a large
extent been dominated by Hilbert's program and has even come
to be identified with work on it. Therefore, with growing interest
in other problems concerning proofs, there has seemed to be a
need to distinguish between a narrower and a broader use of the
term proof theory. Prawitz [5, 6] suggests the name reductive
proof theory for a study of proofs aimed at a reduction of trans-
finite or in general abstract principles to more elementary ones
along the lines suggested by Hilbert and the name general proof
Theory for a broader investigation when proofs are taken as objects
of study for their own intrinsic interest and not for reductive aims.
This terminology is also followed e.g. in the recent monograph on
proof theory by Cellucci [7].
Proof theory taken in a very broad sense is certainly not a
creation of Hilbert. The subject can fairly be said to have started
with Aristotle's works in logic. Indeed, logic as it was cultivated
until quite recently was to a great extent identical with proof
theory: it studied arguments, distinguished between valid and
invalid forms of arguments, and tried to give a systematic presen-
tation of the valid ones. In short, its general aim was to give a sys-
tematic account of correct, strict reasoning or, in other words, of
what constitutes a proof.
In Hilbert's vision of the subject, however, proof theory was to
be distinguished from the older tradition of logic in especially two
ways: firstly, instead of proofs in a more absolute sense, he
wanted to study formal proofs, i.e. derivations in given formal sys-
tems, and, secondly, he thought of the subject as a strictly mathe-
matical one. The first feature was made possible by Frege's
sharpening of the axiomatic ideal by the creation of logistic or
formal systems. The second feature was essentially Hilbert's own
idea. Frege and his immediate successors such as Russell and
Whitehead may be said to have studied proofs in the sense of
having isolated different rules of inference and having investigated
by what principles various mathematical proofs can be carried out,
but the proofs themselves did not then occur as objects of a
mathematical theory.
 ,
Philosophical aspects o/proof theory 237

Proof theory in Hilbert's sense thus makes its appearance at

about the same time as model theory begins in the shape of
Skolem's investigations and Tarski's semantics where also formal
languages are objects of mathematical study. But in contrast to
proof theory, model theory tries to study a notion of truth and
logical consequence independently of the question of how we
come to know that a sentence is true or a logical consequence of
given premisses - it is true that most presentations of model
theory contain proof-theoretical elements, but pure model theory
may be presented without mentioning a deductive apparatus for
the language studied (as most consistently done by Kreisel and
Krivine [8]).
If proof theory is to pay attention to the epistemological
nature of proofs, i.e. to their raison d'etre as the means of getting
to know truths and logical consequences, then, however, it is
highly doubtful whether the two special features which charac-
terize proof theory in Hilbert's sense are appropriate. Firstly,
we know from G6del's first incompleteness proof that the notion
of proof cannot in general be characterized in terms of formal
systems: given any formal notion of proof, there is e.g. a universal
numerical proposition which has no proof of the given kind but
which nevertheless is not only true but can also be proved to be
true:'by Godel's method. Secondly, one would expect that a study
of proofs in their capacity as means of getting to know truths
must take into account also the meanings of the sentences
involved, and that it thus has to merge with a meaning theory as
argued by e.g. Prawitz [9, 10]. It may be only a terminological
question whether such a theory is called mathematical or not,
but in any case, it has to attend to concepts quite different from
the ones Hilbert had in mind.
It is also to be noted (as in Prawitz [10]) that the meaning
theory in question must be quite different from Tarskian seman-
tics, and that there is no direct route from the notion of logical
consequence to the notion of proof: a proof may correctly be
said to consist of a number of valid inferences, but the condition
that the conclusion is a logical consequence of the premisses is
clearly only a necessary and not a sufficient condition for the
appearance of an inference in a proof; nobody counts just some
axioms immediately followed by a non-trivial theorem as a proof
238 D. Prawitz

even if the theorem happens to follow logically from the axioms.

Obviously, some kind of evidence is also required of the inferences
in the proof; one must in some way take seriously that the con-
clusion is seen to be true.
In conclusion, three kinds of proof theory emerge: (1) reductive
proof theory directed towards questions of consistency or, more
generally, towards reductions of abstract principles to more
elementary ones, (2) a more embracing, mathematical proof
theory as envisaged by Hilbert (to be distinguished from his
program), and (3) a more general proof theory not restricted by
the features Hilbert had in mind but attending to the general
epistemological nature of proofs and questions of validity of argu-
ments.
Kreisel has in some papers, e.g. [11, 12J , suggested the name
theory of proofs for a study that concentrates on such things as
the structural complexity of proofs but disregards questions about
what makes a proof a proof, or in other words, what makes an
argument valid, which has been the focus of papers such as Prawitz
[10, 13J. What Kreisel has in mind here seems to be proof theory in
the sense of (2) except that it is to be explicitly separated from the
aims in (1) and (3). Kreisel [14] argues that there are reasonably
adequate data for a 'natural history' of proofs (comparable to
early mineralogy or biology) but that there is no hope at present
for a rewarding, systematic science of proofs (by which he means
a theoretical science that is rewarding in the same way as "success-
ful" natural sciences like crystallography or genetics, to continue
his examples above). We may here see a fourth suggested direction
of proof theory if we want, viz. in the form of case studies of
striking phenomena of proofs, exemplified by Kreisel [14 J (but in
the terminology of that paper, also proof theory in the sense of
(2) and (3) is to be classified as 'natural history').
There is thus no complete agreement in the literature about
what direction the subject should or could take nor about the
question what it should be called. As to the last question, the idea
to use 'proof theory' and 'theory of proofs' as two non-synony-
mous terms denoting different aims in the study of proofs does
not strike one as very suggestive (and Kreisel need not be under-
stood as proposing that but only as wanting to change the name of
the subject to avoid associations with directions that according
 Philosophical aspects of proof theory 239

to him are better disregarded). One may ask whether it is proper

to widen the term proof theory to cover also studies of kind (3),
whose aims more or less coincide with that of traditional logic as
noted above. But in view of the fact that modern logic embraces
also other branches not concerned with proofs, it seems reasonable
to use proof theory as a broad name of that part of modern logic
which is concerned just with proofs and add the qualification
'general' when the need arises to distinguish it from proof theory
in a more restricted sense.
To the conflicting views about what directions proof theory
should take (or could take at present), I shall return in several later
sections.

2. The philosophical relevance of the subject

The term mathematical logic was formerly used to denote logic

treated by the mathematical method, i.e. using symbolic aids and
proceeding ded uctively , see e.g. Church [15] . It has later also been
used to refer to logic that is of special interest for mathematics,
and today it often seems to stand for mathematical research into
problems of only distantly logical origin. This shift in the meaning
of the term from 'logic treated by the mathematical method' to
'mathematics originating from logical problems' is parallel to a
certain general, current tendency in (what is called) logical
research. The idea of a 'philosophical logic' may be understood as
a reaction against this trend in the development oflogic. The term
sometimes seems to be used in a general sense for logic of philo-
sophical interest, but, more often, it seems to stand for the logic
of concepts that are especially used by philosophers, such as
various modalities, or are consid ered to be the special business of
philosophers to discuss, such as the notion of preference.
This terminological situation seems to me unfortunate. It is not
necessarily the logic of concepts occurring in philosophical con-
texts that are of the greatest philosophical interest. But, moreover,
the terminology obscures the fact that logic (i.e. what is properly
so called) as well as the closely related foundations of the deduc-
tive sciences, regardless of whether these areas are studied with the
help of mathematical techniques or not, are essentially philo-
240 D. Prawitz

sophka! disciplines aimed at the understanding of certain aspects

of reasoning and related problems.
A discussion of the philosophical relevance of proof theory may
illustrate the general situation. As is the case of today's mathe-
matical logic, many parts of proof theory are not of any special
philosophical interest. But, at the same time, proof theory con-
tains many themes of an epistemological and meaning-theoretical
nature. Therefore, in spite of its rather technical character, much
of proof theory seems to touch more central themes of philosophy
than many other branches of so-called mathematical or philo-
sophical logic. This is true both for general proof theory with its
stress on evidence and on how we get to know truths and for
reductive proof theory with its attempt at a foundation of mathe-
matics.
In the rest of the discussion in this paper of philosophical
aspects of proof theory with special attention to the development
in the last decade, I shall especially deal with Hilbert's program
and reductive proof theory (section II). Although the formulation
of Hilbert's program belongs to the twenties, the proper way of
understanding and evaluating it is something that has still been
much discussed in the last decade, and technical works as well as
monographs in this field have also been appearing very recently.
The greater space given to reductive proof theory is not owing
to the belief that this part of proof theory is philosophically the
most interesting one - on the contrary, my conclusion is the
opposite -- but merely to the fact that my discussion of reductive
proof theory turned out to require most of the available space. In
the next subsections, I shall therefore restrict myself to a quick
sketch of some elements of the early history of the subject (1.3)
and to a brief presentation of different directions in general proof
theory (1.4). It should be stressed again that this paper is in no
way intended to survey recent developments in proof theory but
to present and discuss some philosophical aspects of the subject.
Furthermore, since proof theory was not treated in the two pre-
ceding series of philosophical chronicles, it did not always seem
advisable to respect the time limit of the last decade.
 Philosophical aspects of proof theory 241

3. Early history of the subject

Frege's foundation of modern logic consisted firstly in the dis-

covery of a way of analyzing logical forms (where, as stressed by
Dummett [16], his way of analy zing universal quantification is
especially crucial) and secondly in a related codification of logical
inferences. f'rege is the first to analyze a significant enough area of
proofs, and his work obviously constitutes also the foundation of
modern proof theory in the wide sense of the term. Consecutive
work by Russell, Zermelo and Hilbert, further developed and
modified Frege's analysis. While they all analyze reasoning that
partly dates very far back, partly emanates from Cantor, Heyting
obtained, by a surprising slight formal modification of this work, a
codification of some of the reasoning involved in the revised, so-
called intuitionistic mathematics proposed by Brouwer.
Also most logical work in the twenties and thirties, when meta-
theoretical investigations began into the systems of inference rules
that resulted from the analysis of Frege and his successors, was of
proof-theoretical concern. This, of course, is especially true of
Godel's completeness and incompleteness results about the scope
and limitation of these systems.
Most of this work including that by Godel does not directly
deal with proofs, however, but rather with the derived notion of
provability (in a formal system). The same is true for certain
formulations of Hilbert's program, but, as will be discussed in
section II when I return to this part of proof theory, the study of
proofs played a part in the contemplated execution of the pro-
gram.
The fIrst ones to deal in an essential way with proofs as an
object of study are Gentzen and - to a lesser extent - Herbrand.
They are noteworthy both for their contributions to Hilbert's
program, where their ideas have allowed a leap forward, and for
their interest in the structure of proofs, i.e. not only the forms of
the various local inferences of which the proofs are made up but
also certain overall structures to which the various inferences may
give rise. In addition, Gentzen [17] analyzed anew proofs in first
order languages, which resulted in a new set of primitive inference
rules that was significant in several ways. In particular, his choice
of primitive inference rules can be seen as dictated by what it is
242 D. Prawitz

in the end that makes an inference evident. This analysis forms the
basis of Gentzen's other results and has been a main stepping-stone
in present-day proof theory.
Gentzen's main idea is to break down the inferences into two
kinds of steps, called introductions and eliminations, where each
step involves only one logical constant. Each rule is here such that
one cannot imagine other more basic rules in terms of which the
rule would be derivable. The rule for the introduction of sentences
of a given form states a condition for inferring sentences of that
form. The rule for the elimination of sentences of a given form
states what direct conclusions can be drawn from premisses of
that form. Furthermore, the rule for elimination is the inverse of
the corresponding introduction rule in the sense that a proof of
the conclusion of an elimination is, roughly speaking, already
available if the premiss of the elimination is inferred by an intro-
duction. This fact has later been called the inversion principle.
As a typical example, an introduction of an implication A ~ B
has the form

B
A~B

where in other words the condition for inferring A ~ B is a proof

of B from hypotheses A, which are said to be discharged or bound
by this inference. An elimination of an implication A ~ B has the
form

which is the inverse of the corresponding introduction in the sense

that if the condition for inferring A ~ B stated by the introduction
rule is satisfied, i.e. there is a proof of B from the hypothesis
A, then a proof of B is aiready available by replacing every hy-
pothesis A by the given proof of A.
 Philosophical aspects o/proo/theory 243

A collection of introduction and elimination rules is known as a

system 0/ natural deduction. On the basis of such sy~tems,
Gentzen formulated another somewhat more artificial kind of
systems, called calculi of sequents, and for these systems he
proved the so-called Hauptsatz according to which each proof can
be put in such a form that nothing is introduced in the proof
which is not also contained in the end result - the proof is in this
way direct, containing no auxiliaIY elements.
It was shown by Prawitz [18 J and, independently, to some
extent by Raggio [19J that already the proofs in the systems of
natural deduction can be put in a certain normal form and that
this normal form theorem, which is equivalent to the Hauptsatz,
can be seen as directly depending on the inversion principle. In
fact, the inversion principle gives rise to certain reductions appli-
cable when a sentence is obtained as the conclusion of an intro-
duction and is used as the premiss of an elimination. The
reduction removes such a round-about.
As a typical example, we may consider a proof of the form
shown to the left below. In accordance with the explanation of
the inversion principle above, it reduces to the proof shown to
the right

A
A
B
A A-~ B B
B

where the conclusion B is obtained without the detour of first

introducing A -+ B and then applying implication elimination.
The normal form is arrived at by successively carrying out all
such reductions, and its existence is proved by showing that a
series of these reductions terminate. A normal proof can be seen
as consisting of two parts, one upper, analytical part that proceeds
from hypotheses by successive eliminations and one lower, syn-
thetic part, where the obtained results are used to infer con-
clusions by successive introductions. A normal proof proceeds in
this way directly from the hypotheses to the conclusion of the
244 D. Prawitz

proof without introducing any sentences in the proof that are not
used in building up the conclusion of the proof.

4. Directions in general proof theory

4.1. The results mentioned above, Gentzen's Hauptsatz and the

normal form theorem, were first obtained for first order logic. In
recent years, they have been extended to a great number of other
systems. Such an extension involves essentially two steps, whose
exact formulation depends on wheth~!r it is formulated for a cal-
culus of sequents or a system of natural deduction. In the latter
case, they consist of, fin:tly, an analysis of proofs in the area in
question in such a way that they are broken Jown into intro-
ductions and eliminations with accompanying reductions, and,
secondly, a proof of the fact that each proof can be put in normal
form, i.e. in a form where all possible reductions have been carried
out. Some of these extensions depend essentially on the analysis
of the proofs involved, i.e. the right formulation of the intro-
duction and elimination rules, while the proof of the normal form
theorem raises no new difficulties; in other extensions, the latter
has been the main obstacle.
In Prawitz [18], where results in the natural deduction formu-
lation were obtained for first order classical, intuitionistic, and
minimal logic, extensions were made to some systems of modal
logic (S4 and S5 in classical and intuitionistic variations),
to certain non-standard systt'ms (like a relevance logic and a
system similar to Fitch's set theory), and to ramified second order
logic.
An exttnsion to an infinite system of arithmetic of natural
numbers (in the form of a system similar to a calculus of sequents)
had already been obtained in 1951 by Schutte [20]. Later, Tait
[21] and Martin-LOf [22] developed classical and intuitionistic
systems, respectively, for infinite propositional logic.
Among other various extensions essentially depending on a
conceptual analysis of the proofs, I want to mention one by
Stenlund [23] to an interesting theory of descriptions.
For unramified (impredicative) second order logic, which has
attracted special attention since it allows a formulation of classi-
cal mathematical analysis, and, more generally, for higher order
 Philosophical aspects o/proo/ theory 245

logic (or simple type theory), the first step of the extension is
straightforward (in analogy with first order), but the proof of the
normal fonn theorem (for a long time known as Takeuti's con-
jecture) turned out to be more difficult. It was solved by Tait
[24], Takahashi [25, 26], and Prawitz [27, 28] by utilizing a
certain correspondence between the c:llculi of sequents for classi-
cal logic and Tarskian truth definitions (see further section 4.2
below). Extensions to intuitionistic logic of second and higher
order were obtained by Prawitz [29] and Osswald [30], respec-
tively. However, all these results for impredicative systems showed
only the existence of a normal form for the proofs in question and
did not establish the termination of any particular series of reduc-
tions. This situation drew attention to the difference between such
a normal form theorem and a normalization theorem that estab-
lishes the termination of some reduction procedure; for the predi-
cative systems, this latter and stronger result had usually been
implicit in the proofs.
For first order Peano arithmetic, Gentzen [31] had in fact
(although he did not formulate it in these terms) obtained a
normalization theorem for proofs of numerical equations. An
extension to all proofs in the system was obtained by lervell
(32]. More generally, Martin-Lor [33] analyzed intuitionh;tic
proofs with ind uctive and iterated inductive definitions into
introduction and elimination inferences and proved a corre-
sponding normalization theorem. This result is of special interest
for two reasons: it extends the idea of introductions and elimi-
nations to apply not only to logical constants but also to atomic
predicates, and it allows a constructive development of a con-
siderable part of mathematics. The normalization theorem for this
system depended among other things on two essential discoveries,
namely (1) of the correspondence between proofs and certain
functionals, which was to be expected in view of certain
intuitionistic theones about proofs (see section 4.3 below), and
(2) of the resulting possibility to carryover to proofs a certain
notion of computability introduced earlier by Tait [34] for a
system of functionals under the name convertibility. This allowed
Martin-Lof to formulate a fairly general method of proving
normalizability via proofs of computability.
A considerable extension in a quite different direction but also
246 D. Prawitz

depending on (1) above is made by Martin-Lof {35] who formu-

lates an intuitionistic theory of types which is supposed to allow
the formalization of existing, constructive mathematics. Intro-
duction and elimination rules are here formulated for con-
structions in general, of which proofs are only special cases.
The correspondence between proofs and functionals made it
also possible to strengthen the normalization theorem for many
systems by showing that all reduction sequences (starting from a
given proof) terminate in a unique normal form either by carrying
over similar results from systems of functionals or by modifying
Martin-Lors version of computability. Results of this kind, called
strong normalization theorems, are surveyed in Prawitz [5] . The
translation between systems of natural deduction and calculi of
sequents established by Prawitz [18] to obtain the equivalence
between the normal form theorem and the Hauptsatz was
improved by Zucker [36] and later by Pottinger {37] to obtain
something similar to the strong normalization theorem also for the
calculi of sequents.
Furthermore, Girard [38] was able to extend the computa-
bility notion to second order systems of functionals, which
enabled Girard [38,39], Martin-Lo[ [40,41], Prawitz [51, and
Troelstra [42, 43] to obtain various normalization and strong
normalization theorems for second and higher order intuitionistic
logic. These results can also be extended to certain classical sys-
tems, see Statman [44] and Prawitz [45] . Much before, however,
Takeuti [46] had obtained something similar to a normalization
theorem for a subsystem of classical second order arithmetic
with quite different methods in his work on Hilbert's program,
see further section 11.5.

4.2. Although Gen tzen [17] arrived at his calculi of seq uents by
reflecting upon the systems of natural deduction and, in the case
of classical logic, generalizing the deducibility relation by allowing
several sentences taken disjunctively as conclusions (which to his
surprise, as Gentzen notes in his paper [31], removed certain
problems connected with classical negation), there is an equally
interesting aspect of the calculus of sequents for classical logic,
namely the close correspondence between its rules and the clauses
of the Tarskian truth definition, which was perhaps first explicitly
 Philosophical aspects of proof theory 247

noted by Beth [47] in terms of his semantic tableaux. More

precisely, the introduction and elimination rules in the systems of
natural deduction become formally rules for introducing sentences
in the succedent and antecedent, respectively, of a sequent but can
also be read as the conditions for the truth and falsity, respec-
tively, of sentences appearing as members of an iterated dis-
junction. In this way, Gentzen's Hauptsatz may be looked upon as
a generalization of the fact that the conditions for the truth and
falsity of a sentence cannot be satisfied simultaneously.
Furthermore, systematic applications of the rules of the cal-
culus of sequents backwards, so to say, yield because of this
correspondence all possible countermodels to the sequent that one
started with, if such countermodels exist. When they do not, i.e.
when the procedure of constructing countermodels breaks down
at a finite stage, this will manifest itself in a demonstration of the
fact that no assignment of truth values that refutes the sequent
is consistent in the sense of not assigning both truth and falsity
to the same sentence. Because of a certain property of duality,
this demonstration constitutes at the same time a proof of the
sequent in question. As a result, we get an immediate proof of
Godel's completeness theorem for first order predicate logic, and
as a further immediate corollary, we obtain Gentzen's Hauptsatz.
Proofs of the completeness theorem have been given in this way
by Beth [47] ,Hintikka [48] , Kanger [49] , and Schutte [50] .
This aspect of Gentzen's work is discussed in more detail by
Prawitz [51] and is further generalized by Kreisel, Mints, and
Simpson [52]; for later comments on this generalization, see
Kreisel [53]. It was extended to second and higher order logic to
get Gentzen's Hauptsatz for those cases by the proofs of Tait,
Takahashi, and Prawitz mentioned above, which are further
discussed and reworked by Girard [54]; in this connection, see
also Pappinghaus [55] .

4.3. While the calculus of sequents for classical logic lends itself
to establishing a link between proof theory and classical semantics
(model theory), the system of natural deduction is especially
suitable for certain discussions of the semantics of intuitionistic
logic.
Intuitionistically, the logical constants are often interpreted in
248 D. Prawitz

terms of proofs or constructions as perhaps first explicitly done by

Kreisel [56]. For instance, the implication A ~ B is taken to assert
the existence of a construction c, which applied to a proof of A
yields a proof of B, together with a proof of the fact that c has
this property. In other words, proofs are not only thought of as
demonstrating the truth of the sentences, but the sentences are un-
derstood as speaking about proofs. There is then an obvious common
ground between the semantics of intuitionistic logic and general
proof theory. A correct intuitionistic interpretation of sentences
of a certain form must, it seems, at the same time state what con-
stitutes a proof of a sentence of that form. Kreisel's work in [56]
has been elaborated by among others himself in [57]; see also
Troelstra [58].
A similar but slightly different approach had already been
vaguely suggested by Gentzen [17] . He looked upon the condition
stated by an introduction rule for inferring sentences of a certain
form as giving the meaning of the logical constant in question. For
instance, an implication A ~ B would be understood as expressing
the existence of a proof of B from the hypothesis A. The corre-
sponding elimination rule, on the other hand, got its justification
from this meaning since, as we saw in section 3, given that the
condition for inferring the premisses of an elimination by intro-
duction is satisfied, a proof of the conclusion is already available,
or more precisely, is obtained by a reduction.
The normalization theorem can be seen as a sharpening of this
idea; it shows that the whole system of elimination rules is justi-
fied in the sense that any proof in the system of a sentence (not
depending on hypotheses) is transformed by a series of reductions
to a non-reducible proof, which necessarily has to end in an intro-
duction, i.e. the canonical way of arguing according to this point
of view. But Gentzen's general idea has been further elaborated in
two slightly different ways. One way is to note a similarity
between natural deduction and an extended A-calculus for defining
functionals. For instance, a derivation of B from an hypothesis A
may be looked upon as an open term t(x) in which x is a variable
ranging over proofs of the hypothesis A and which yields a proof
t(p) of B whenever a proof p of A is substituted for the variable x.
The proof of A ~ B obtained by implication introduction applied
to the derivation of B from A may then be identified with the
 Philosophical aspects of pro of theory 249

function Xxt(x) which applied to a proof p of A yields the

proof t(p) of B. To say as Gentzen does that implication intro-
duction determines the meaning of implications may then be
understood as saying that the meaning in question is determined
by the requirement that a proof of A ~ B is to consist of a
function that applied to a proof of A gives a proof of B and that
can be written in the form Xxt(x), which corresponds to the first
half of Kreisel's interpretation of implication.
On the other hand, a proof obtained by applying implication
elimination to a proof q of A ~ B and a proof p of A may be
identified with the result of applying the function q to p. If q is
written in the form AXt(X), the proof may then be written
{Axt(x)}(p), which is identical with t(p). In the terminology of X-
calculus, the latter term t(p) is said to come from {Xxt(x)}(p) by
X-conversion. Implication introduction is in this way seen to
correspond to X-abstraction, implication elimination to function
application, and the reduction associated with implication to
X-conversion.
The formal analogy between implicational logic and X-calculus
seems first to have been noted by Curry and was more systemati-
cally developed and extended to full predicate logic by Howard
[59] and Prawitz [5, 60]. Howard's formulation, the corre-
spondence between propositions and types of functionals is
formulated as an isomorphism (in Prawitz' formulation as an
homomorphism). The correspondence has been utilized especially
by Martin-Lor as mentioned above (section 4.1).
This way of looking at proofs also leads to the identification of
two proofs when one can be obtained by a reduction of the other
as discussed by Martin-Lof [61], Kreisel [62], and Prawitz [5] .
Feferman [63] has objected to this idea on the ground that a
proof PI of a sentence A may reduce to a proof P2 but may con-
tain much more information than P2; on its way of proving A, PI
may establish the truth also of many other propositions not estab-
lished by P2' This cannot be denied, but it must be observed that
on the view presented here, a proof is not looked upon as a
collection of sentences but as the result of applying certain oper-
ations to obtain a certain end result; when understood in this way, it
seems difficult to deny that e.g. the proof expressed by {Xxt(x)}(p)
is identical to the one expressed by t(p). The main question
250 D. Prawitz

must thus be to what extent the identification of proofs and con-

structions such as those obtained in the A-calculus is relevant; in
this connection, see especially Martin-Lof [35]. However, these
questions about the identity of proofs cannot yet be considered
settled in the literature.
The other direction in which Gentzen's idea about justifications
of inferences has been elaborated is as a suggestion for a criterion
for the correctness of a proof or, rather (since if a proof is rightly
called a proof, it has to be correct), for the validity of an argu-
ment. Here, the starting point is thus a notion of argument, which
is defined as a collection of sentences structured in a certain way.
An attempt to work out such an idea has been made by Prawitz
[ 13, 64]. It used a slight modification of Martin-LOf's notion of
computability mentioned above, which was now worked out as a
notion of validity of an argument. Roughly speaking, an argument
is defined as valid if it either ends with an introduction whose
premisses are proved by valid arguments or can be transformed to
such a form by reductions. In other words, the introductions are
considered as self-justifying, as constituting the canonical ways
of arguing for the sentence forms in question, while the elimi-
nations are justified by the very reductions that generate the
normal form of proofs. The validity of a set of inference rules as
defined by Prawitz [13 J is a general sufficient condition for the
normalizability of the proofs obtained by applications of these
rules. The question of the completeness of the logical constants
from the point of view of such a semantics is discussed by
Prawitz [ 10] and Zucker and Tragesser [65] .
However, it is to be noted that this notion of valid inference
suffers from the same weakness as discussed in section 1: to be
permissible in a proof, it is not enough that an inference happens
in fact to be justifiable, it must also be justified, i.e. it must be
seen to be correct.
Dummett [66, 67] has broadened the perspective above by
suggesting that the meaning of any expression is determined by
two features, on the one hand the condition for correctly uttering
the expression, and on the other, the direct consequences of
uttering it, or what the uttering of the expression commits the
speaker to; which features have to be in harmony with each other.
In the case of affirmative sentences of the usual logical form, these
 Philosophical aspects ofproof theory 251

two features are given by Gentzen's introduction and elimination

rules, and the required harmony is expressed by the inversion prin-
ciple (mentioned in section 3). In this way, following general ideas
of Wittgenstein, Dummett wants to replace the truth condition of
a sentence as central to its meaning by the condition for correctly
asserting it. For this latter condition to be satisfied, the mere
truth of the sentence is of course not sufficient - the speaker
must also be in possession of a certain knowledge that warrants
the assertion. In a similar way, the question of the correctness of
an inference or of an argument may be related to the speaker's
knowledge as discussed by Prawitz [9, 10].
It should be noted that none of the approaches discussed in
this subsection is especially concerned with proofs in formal sys-
tems or proofs restricted to the means expressible in a formal lan-
guage. The main question that has been put here is what properties
proofs of a sentence of a given form are to have - in this respect,
the notion of proof is tied to questions of linguistic forms, but for
well-known reasons, the proofs or sentences in a given language
cannot in general be restricted to that language.
So far we lack a general proof theory with a definition of an
absolute notion of provability that is generally agreed upon com-
parable to e.g. recursion theory with its definition of algorithm or
model theory with its definition of logical consequence. Kreisel
[12, 14] doubts that such a subject is now ripe for study or
doubts the fruitfulness of such studies. Godel [68] in a paper
reprinted in the series of chronicles preceding this one (Contem-
porary Philosophy f) where he discusses provability instead of
proofs, expresses the opposite conviction. Stressing that his incom-
pleteness theorem excludes the possibility of identifying prova-
bility in a calculus with the (absolute) notion of provability, he
notes that this negative result only rules out certain ways of
defining provability but does not make such a definition
impossible. Similar remarks have been made by Myhill [69J,
stressing the need of a general notion of proof. Some of the works
referred to above constitute an attempt towards such a general
notion of proof.

4.4. Since the discussions in this area have not yet resulted in any
definite or generally accepted viewpoints, one cannot expect a
252 D. Prawi(z

survey like the present one to do full justice to all the participants,
i.e. I want to make the reservation that further developments will
soon show that the accents should have been put in different
places or that the whole picture should have been painted differ-
ently.
In conclusion, I also want to mention that other surveys of the
topics discussed here have been made in papers by Kreisel [62],
Prawitz [5], and Troelstra [43] in the last decade, that a mono-
graph by Cellucci [7] dealing with certain parts of the subject has
recently appeared, and that Gentzen's collected works have been
available in an English translation by Szabo [70] for some time
- for an extensive review of that volume, see Kreisel [71].

II. HILBERT'S PROGRAM. APPRAISAL. RECENT DEVELOPMENTS

I. The goal of the program

Hilbert often expressed himself as if the goal of his program was

simply to establish the consistency of mathematical theories. This
problem was stated by Hilbert [72] in 1900, was repeated in a
number of articles in the twenties and is the basic question from
which the exposition in the standard work Grundlagen der Mathe-
matik by Hilbert and Bernays [73] starts.
In view of Hilbert's emphasis on the consistency question, one
may think that his position was that of strict formalism by which
I mean the idea that mathematics may be understood as a number
of games where formulas are derived according to prescribed
formal rules, where no meaning is to be attached to the formulas
(although, outside of mathematics, some of them may turn out to
be useful) and where the only requirement to be put on the game
is consistency. It should be clear, however, that Hilbert took a
fonnalist position only towards sentences that contained reference
to infinite totalities, which sentences he called ideal, while he
considered the other sentences that did not make such reference,
the so-called real sentences, completely meaningful. It should also
be clear that strict formalism is not really coherent if one attaches
importance to the consistency problem and considers it to be a
mathematical problem, because then there are at least some mean-
 Philosophical aspects of proof theory 253

ingful mathematical sentences, viz. the ones expressing the consis-

tency of certain formal systems. The question cannot be whether
these sentences are provable in one or other formal system but
whether they are true. As soon as this is realized, it is clear that
there must be at least one realm of mathematics, viz. the one in
which the proof of consistency is carried out, which cannot be
viewed formalistically but must be understood as having a direct
content; as is well-known, for Hilbert this realm was what he
called finitary reasoning.
Now, if one takes this position of moderate formalism, where
a part of mathematics is considered as directly meaningful, con-
stituting what mathematics is directly about, while another part,
viz. the one containing reference to actual infinity, is taken as an
ideal or formal part, added to smooth and round off the real
part, then the important question must be whether proofs that
use ideal elements in order to prove results in the real part always
yield correct results. To establish this is just the goal of Hilbert's
program. In some of Hilbert's publications, in particular in [3,
74], the program is formulated more or less in this way, but the
one-sided emphasis on the consistency problem in many other
publications has sometimes come to overshadow the proper for-
mulation, to which in particular Kreisel has repeatedly called
attention, among others in some recent publications [ 11, 75] .
Although Hilbert's program as here formulated is equivalent
to the consistency problem under certain conditions (see below),
its proper formulation is important when it comes to evaluating
its significance. Naively one may think that it is always of interest
to establish the consistency of a mathematical theory. But as we
have already seen, from the point of view of strict formalism, the
question of consistency, let alone the question of how consistency
can be convincingly proved, cannot even be put in mathematics.
The consistency problem becomes significant in mathematics only
when some part of mathematics is accepted as meaningful (with a
definite content) and correct, while another part is put in more
doubt. But then, as said above, the important question must be
not only whether a mathematical theory is consistent but whether
we can be sure that sentences belonging to the accepted part, i.e.
real sentences to use Hilbert's terminology, which are proved using
elements from the doubted part of the theory, are always true.
254 D. Prawitz

Starting from the consistency problem we thus arrive at the more

mature Hilbert's program. But, as already remarked above, in his
last publication [73] (together with Bernays) on this subject, the
program is still formulated in terms of consistency.
In conclusion, we may summarize Hilbert's general program as
containing the following two steps

1) a separation of an immediately accepted part of mathematics,

called the real part, which was identified by Hilbert with what
he called finitary mathematics, from another not immediately
accepted part, called the ideal part, which according to Hilbert
typically contains references to the transfinite, and
2) a demonstration carried out in the real part of mathematics of
the fact (if a fact) that every provable real sentence is true, i.e.
that every sentence belonging to the real part which is proved
by possible use of the ideal part is nevertheless true (according
to the standards of the real part).

The program requires of course that we are able to describe the

mathematics spoken about in some suitable way. A first necessary
step is thus to represent the mathematics in question, say some
more or less inclusive part of mathematical practice, in the form
of a theory of some kind.
The formulation of such a theory can be more or less articu-
lated. To present it as a formalized system, i.e. to give the language
and the principles of the theory in such a way that it can be
decided whether a given expression belongs to the theory or not
and whether something is a proof in the theory or not, is to be
viewed as a way of articulating the theory and not to reflect a
formalistic attitude, where sentences and proofs are considered as
just formal objects. But (as will be seen in section 2 below)
Hilbert's ideas about how the program was to be carried out
req uired formalization.

2. Execution of the program. General method

The theory to which Hilbert wanted to apply his program was in

particular arithmetic for real numbers by which he meant a theory
sufficient for developing usual mathematical analysis. Formal
 Philosophical aspects o/proo/ theory 255

systems for such a theory were developed in the Hilbert school by

modifying Whitehead's and Russell's system in Principia Mathe-
matica, in particular by taking the notion of natural number as a
primitive and adding well-known axioms for this notion. Such a
system was believed to cover or represent most of mathematical
practice; hence, a successful realization of Hilbert's program for
this system would give a foundation of mathematics in general.
To carry out stage I) of the program as formulated in the pre-
ceding section, we have to make a separation both with respect to
what is immediately accepted as meaningful and with respect to
what is immediately accepted as evident, the fIrst being a linguistic
or semantic division of sentences into real and ideal ones and the
second being a division of principles, i.e. both axioms and rules of
inference, into those that can be used as evident in a real proof
and those that cannot be so used. According to Hilbert, the real
sentences embrace firstly all sentences expressing decidable
propositions about (finite) configurations. Thus sentences expres-
sing that a certain sequence of formulas is a proof, in a certain
formal system, of a certain formula belong typically here, and since
natural numbers are understood by Hilbert as series of strokes,
sentences like 2 + 3 = 5 or 10 10 . 10 10 = 10 20 also belong here.
Since truth-functional compositions of decidable propositions
are again decidable, the application of the corresponding sen-
tential connectives to decidable sentences gives rise to new real
sentences. The crucial question is then what kind of sentences
universal and existential quantifications over, for example, all
natural numbers give rise to.
Such quantifications when understood as infinite conjunctions
and disjunctions, respectively, are of course prime examples of
transfinite propositions, but on the other hand already the con-
sistency statement asserts something about all possible proofs of
the system in question. Hilbert's solution of this dilemma consists
in considering another 'finitary' interpretation of the universal
quantification over e.g. natural numbers, namely (as he says in
[3]) to understand a corresponding assertion as 'a hypothetical
judgement which asserts something for the case when a numerical
example is given'. Similarly, Gtmtzen [76] explains the construc-
tive meaning of the sentence 'all natural numbers have the
property P' as saying: 'Regardless of how far we progress in the
256 D. Prawitz

formation of new numbers, the property P continues to hold for

these new numbers'.
Although a similar finitary interpretation of an existential
judgement 'there exists a natural number with the property P' as
attesting that 'we have found a natural number with the property
P' is possible as Gentzen [76] remarks, Hilbert refers the existen-
tial sentences to the ideal ones, and hence, he does the same for
sentences that are negations of universal ones or have the form of
an implication with a universal sentence as antecedent. Since a sen-
tence V Xl VX2 .. . Vx n A(X 1 ,X 2 , ... , Xn) with A quantifier-free can
always be replaced by an equivalent sentence V xA'(x), where A'
is again quantifier-free, we can say in summary that the real sen-
tences comprise the decidable ones and the ones of the form
VxA(x) where each instance A(t) is decidable; the rest are con-
sidered to be ideal.
Furthermore, all proofs that can be carried out in arithmetic
using only real sentences are accepted as immediately evident as
long as universal sentences do not appear as assumptions in the
proof. Thus, in particular universal introductions and proof by
induction, where the induction hypothesis is not universal, are
accepted. (Since we never essentially operate on a universal
formula in a real proof, the universal quantifier is dispensable in
the real part, and universal sentences are usually replaced by free-
variable formulas.)
The demonstration in step 2) of the program that proofs of
real sentences, even when using ideal elements, always lead to true
sentences, could in principle be carried out in many ways. For
instance, one could hope for some special interpretation of the
ideal part under which all principles assumed in the ideal part
become true (like the interpretation of complex numbers as pairs
of real numbers). But, since the real part embraces only finitary
reasoning and the ideal part contains transfinite concepts, such
an interpretation seemed improbable, and a distinctive feature of
Hilbert's program is not only its goal but also the method by
which one hoped to reach the goal, namely by an investigation of
the proofs of the theory in question. Hence the need of a proof
theory. Furthermore, this method where proofs were to be objects
of a finitary proof theory required a formalization of the investi-
gated theory.
 Philosophical aspects of proof theory 257

Given a formalization of a theory T of arithmetic, we note that

the result to be established for T can be expressed by the real
sentence:

(1) For each proof pinT and for each real sentence A in T:
if p is a proof of A in T, then A is true.

The sentence (I) is of course not a real sentence ofT, but it is not
required that the demonstration of the truth of the provable real
sentences of a theory T is carried out in the real part of T itself.
By introducing a function g that assigns Gadel-numbers to the
objects of T and by defining a relation Pr and a property Tr to
hold for Gadel-numbers when the corresponding objects of T
satisfy the relation 'proof of and 'true', respectively, sentence (I)
can be written

(I') For each proof pinT and for each real sentence A in T:
Pr(g(p), g(A)) ~ Tr(g(A».

(I') is of course still not a sentence of T. But it now implies for

each real sentence An of T with Gadel-number n:

(2) Vx(Pr(x, ii)~ An} (where ii is the numeral denoting n)

which indeed is a real sentence of T, when Pr is defined directly

by primitive recursion (so as to coincide with Pr as defined above).
Kreisel [75] takes (2), which is called the local reflection principle
by Kreisel and Levy [77] , to be the statement that is to be demon-
strated in Hilbert's program, but it seems more appropriate to note
it as a consequence in the real part of T of the sentence (1'),
called the global reflection principle by Kreisel and Levy [77] ; (1)
or (1') is what really has to be established.
As is well-known, the Hilbert school never succeeded in proving
(I) (or (2» finitarily for a theory T containing usual arithmetic
(for real numbers). The only result obtained was the one by
Gentzen for a first order theory of natural numbers. I shall report
on some further recent progress for subsystems of analysis in
section [5 ] below.
As regards the formulation of the result to be proved in
258 D. Prawitz

Hilbert's program, we should finally note that when the real sen-
tences are delimited in Hilbert's way, the consistency problem
happens to be equivalent to Hilbert's program. Trivially, (l)
implies consistency. Conversely, assume that T is consistent and
that a real sentence A has been proved in T. If A is a quantifier-
free sentence and 4ence decidable, then A must be true because
otherwise I A would be true and the completeness of T with
respect to decidable sentences would imply that also IA was
provable, contrary to the consistency of T. If A is of the form
VxB(x), then B(n) would also be provable for each numeral n,
and hence, by what has just been said, B(n) must be true for each
n. But this isjust what is required for the truth ofVxB(x).
This equivalence is in effect noted by Hilbert, e.g. in [74], but
was not enough focalized. With Kreisel, who has especially stressed
the equivalence, e.g. recently in [11], one must say that the
reason for this is insufficient discussion of the significance of
Hilbert's project, which was usually just stated as the task of
proving consistency. In particular, it is important to note that
Hilbert's way of drawing the demarcation line between the real
and ideal sentences is crucial for the eq uivalence. As noted above,
existential sentences 3xA(x), where A(x) is quantifier-free, can as
well as the universal ones be interpreted constructively and hence
counted as real sentences. But then the equivalence between the
consistency problem and Hilbert's program fails: nothing prevents
a sentence 3xA(x) to be provable in a consistent theory T
although A(n) is false for each numeral n. And of course this is
fatal regardless of whether 3xA(x) is read in a finitary or in a
transfinite way; in either case, consistency is no guarantee that
the theory does not contain provable sentences which are simply
false.
We may also note that the terminology 'finitary' and 'trans-
finite' is not very felicitous since the finitary sentences contain
sentences formed by universal quantification over all natural
numbers. Thus, the infinite enters into both the finitary and trans-
finite sentences but in the first case as potential infinity and in the
second case as actual infinity as explained above. What this differ-
ence amounts to is not too obvious; it is one of the main issues
that has to be clarified or perhaps rather reformulated in the
dispute between classical and intuitionistic logic. However, I shall
here follow this terminology of the Hilbert-school.
 Philosophical aspects ofproof theory 259

3. Motives for the program

Hilbert's program is meant as a contribution to the foundations of

mathematics, but one can be interested in the program from a
foundational point of view for various reasons.

3.1. Moderate formalism taken in a very literal or extreme way

would deny the ideal part of mathematics any meaning and would
consider it just as a convenient formal machinery for producing
results about the real part. Such a position makes a successful
realization of Hilbert's program very necessary since pending this,
there are no reasons, except purely empirical ones, to assume the
machinery to work in the intended way. But the natural mathe-
matical reaction towards such a machinery would be, I think, to
ask why it works, to try to understand it, and probably nobody
in the Hilbert school understood (moderate) formalism in a literal
way. Gentzen certainly did not; indeed, as already remarked, in
[76] he discusses the existential sentences of the ideal part and
gives them both a finitary and a transfinite interpretation.

3.2. The position of Gentzen is rather that the ideal sentences have
some kind of meaning, which agrees with the usual way of reason-
ing about them, but that this meaning is somewhat shaky; the
ideal sentences are not yet entirely understood, and therefore one
has to give them a finitary foundation, making sure that they do
not give rise to any inconsistencies. Gentzen [76] says e.g.:
'Indeed, it seems not entirely unreasonable to me to suppose that
contradictions might possibly be concealed in classical analysis."
Although something would certainly be gained by a consistency
proof when one maintains an attitude like Gentzen's, the obvious
weakness with this position is that consistency or, eq uivalentIy,
truth of provable real sentences, is far too little. Since also the
so-called ideal sentences are granted a meaning, even if somewhat
shaky, one must demand of a foundation that it also shows the
results about ideal sentences to be in agreement with this meaning.
And as we have already seen, Hilbert's program does not rule out
that a sentence3xA(x) is provable although A(n) is false for all n.
260 D. Prawitz

3.3. Hilbert's various statements are more ambiguous or inconsis-

tent than Gentzen's. On one side, he declares in [3] that 'we must
realize that the infinite in the sense of an infinite totality, where
we still find it used in deductive methods, is something fictitious',
which seems to show that he takes moderate formalism literally.
On the other hand, unlike Gentzen he never expressed any doubts
about the consistency of classical analysis and, as Kreisel [781
points out, never spoke about investigating whether analysis is
consistent but only about establishing the consistency; indeed
Hilbert [I] asserts that a complete and justified certainty about
inferences prevails in analysis, and he accuses a critic like Weyl of
seeing ghosts.
Therefore, there are good reasons to think that Hilbert saw his
program not as removing any doubt about the correctness of trans-
finite reasoning that he had himself but as a continuation of an axio-
matic tradition, which wants to build mathematics on a base as
small and simple as possible (and which in addition should serve to
convert the sceptics). Hilbert is perhaps best understood in the
same way as Frege, who did not doubt ordinary mathematical
reasoning but nevertheless wanted a rigorous foundation of mathe-
matics and especially a reduction of mathematics to logic, except
that Hilbert wanted a reduction not to logic but to what he con-
sidered to be the most proper base, namely finitary mathematics.
What Hilbert here means by a reduction is indicated in the lecture
[3 J where, speaking at the 100th anniversary of Weierstrass, he
wants to continue Weierstrass' work of eliminating the infinitely
large and infinitely small from analysis by showing how also
references to infinite totalities in universal and existential quanti-
fication can be replaced by finite procedures which yield exactly
the same results.
When Hilbert's motives are understood in this way, the interest
is shifted from the question of the truth of the provable real sen-
tences, which is not in doubt, to showing that any proof of a real
sentence carried out in the ideal part by use of transfinite reason-
ing can be replaced by a proof of the same sentence in the real
part using only finitary reasoning. When this is shown by finitary
means, there is a sense in which we reduce the transfinite to the
finite, not by an interpretation, but by showing how the use of
transfinite notions can be eliminated when we are only interested
in results about the finitary.
 Philosophical aspects o/proo/ theory 261

Although the interest of the program no longer resides in

demonstrating the truth of the provable real sentences in itself,
such a demonstration if finitary provides just the reduction
wanted as pointed out e.g. by Kreisel [11] (and, in fact, before by
Hilbert [74]).
In other words, if PT is a fmitary proof theory in which the
sentence (1) of section 2 above is demonstrated and PT contains
the real part of T, there is a transformation r such that

(a) for every proof pinT of a real sentence A, r(p) is a proof in

PTof A.

To see that this is so, we note that a proof in PT of (1) gives a

finitary defined function I{) such that

(b) if a is a proof in PT of the fact that p is a proof of the real

sentence A in T, then I{)(Q') is a proof in PT of the fact that A
is true.

Furthermore, PT can be assumed to be complete enough to con-

tain uniform proofs about what objects in T are proofs, i.e. there
is a function 1/1 defined in PT such that

(c) if P is a proof of A in T, then I/I(p) is a proof in PT of the fact

that p is a proofof A in T.

Furthermore, any proof in PT of the truth of a real sentence A in

T can trivially be transformed to a proof of A in PT when it is
assumed that PT contains the real part of T; i.e. there is a
finitarily defined function X such that

(d) if a is a proof in PT of the truth of the real sentence A in T,

then X(o:) is a proof in PT of A.

If we set rep) = X[I{)(I/I(p»], then (b) - (d) gives the desired

result (a).
262 D. Prawitz

4. The impact of G6del 's incompleteness theorems

Of Godel's two incompleteness theorems, the second one has

often been thought to affect Hilbert's program seriously. Accord-
ing to it, the consistency of a theory T that satisfies certain con-
ditions usually satisfied by most codifications of arithmetic such
as containing a sufficient part of first order arithmetic for natural
numbers, cannot be proved by using only principles of T. It may
be true that Hilbert thought that the consistency of arithmetic,
and hence, the sentence (2) of section 2 for any real sentence A of
T could be proved in the real part of T itself, but this is just the
possibility excluded by Godel's second incompleteness theorem.
However, as already noted, the leading idea of Hilbert's program
as formulated above only requires that the demonstration of the
truth of the provable real sentences of T (or, equivalently, of the
consistency of T) is carried out in real mathematics, not neces-
sarily in the real part of T itself. As long as there exists finitary
mathematics that is not codified in a given theory T, the possi-
bility of carrying out the program for T thus remains in spite of
Godel's result.
Nevertheless, Godel's second incompleteness theorem is of
course an important result about the form that the execution of
Hilbert's program must take. For recent discussions of this
theorem, see Kreisel and Takeuti [79] and Smorynski [80] .
It should be noted however that in several ways Godel's first
incompleteness theorem affects Hilbert's program just as much or
more than the second one. The theorem (in the form given by
Rosser) constructs for every consistent theory T that contains
a sufficient part of first order arithmetic a real sentence V xG(x)
such that

(i) For each numeral n, I-G(n); but neither I-VxG(x) nor 1--,VxG(x)

Two ways in which this result affects Hilbert's program are to be

noted. Firstly, after the discovery of this theorem, it is not only an
abstract possibility that there should exist a consistent theory with
a provable false sentence of the form 3x-A(x) where A(x) is
quantifier-free (as discussed at the end of section 2 above). Given
(i) above, we need only add 3x-,G(x) as a new axiom to the ones
 Philosophical aspects of proof theory 263

of T; the new theory T' is consistent (since otherwise, classically,

VxG(x) is provable in T) but 3x-,G(x) is false given that all the
provable G(n) are true. The fatal effect for Hilbert's program of
the existence of such so-called w-inconsistent theories was
remarked upon in a short note by Godel [81] and has been
stressed by Kreisel e.g. in [II] .
Secondly, it is remarked by Kreisel [75] that the choice of
codification, i.e. the formulation of the theory T of arithmetic
now becomes a crucial problem since it will necessarily be incom-
plete. It may be said that the whole idea of securing transfinite
reasoning or reducing it to finitary reasoning (the two alternatives
of section 3) now becomes less convincing when this cannot be
done once and for all; it may seem that one needs insights into the
transfinite in order to know how to extend a given codification.
And there will always remain transfinite principles that are
intuitively correct (from a transfinite point of view) but have not
been secured or reduced. One could try to defend the program by
noting that the studied theories are complete in the empirical
sense of codifying the actual, existing mathematical practice.
Kreisel's point is then that mathematical practice, if that is what
one is interested in, can be codified by other systems that use less
strong transfinite principles. Hence, in any case, the discovery of
incompleteness must lead to greater attention to the choice of
codifications.

5. Progress in the program

No one in the Hilbert-school succeeded in carrying out anything

like the Hilbert program for a system that embraces classical
analysis. A result in this direction for a subsystem of analysis, so-
called TIl-analysis, was obtained by Takeuti [46] in 1967. His work
builds on ideas in Gentzen's consistency proof for first order arith-
metic 0 btained in the thirties.
Gentzen published two versions of his proof, the first in 1936
[76] and the second in 1938 [31], both using transfinite
induction up to the ordinal number denoted Eo. An earlier version
of Gentzen's consistency proof became publicly known because
of a paper by Bernays [82] and was recently published in the
name of Gentzen [83]. It was submitted by Gentzen in 1935 but
264 D. Prawitz

was withdrawn after criticism directed against the means of

Gentzen's proof which were considered to be too strong. For a
discussion and a criticism of the criticism, see Bemays [82] and
Kreisel [84] .
Takeuti's proof pursues the lines of Gentzen's second published
proof, which may be described in summary as consisting of the
following steps:

I) Certain ordinals, i.e. a certain system of notations which may

intuitively be understood as denoting the ordinal numbers of
a certain initial segment, viz those up to Eo are introduced and
a decidable ordering relation < between them is defined;
2) it is shown that any descending sequence of the ordinals is
finite, or rather, a demonstration is given of the principle of
transfinite induction for < up to Eo applied to decidable sen-
tences A(x), which can be formulated

(TO from the premiss 'if A(~) for all ordinals ~ < (3, then A({3)'
infer the conclusion 'A( -y) for all ordinals -y' up to € 0;

3) a reduction procedure for proofs in first order arithmetic is

introduced of the same kind as those discussed in section 1.3
with the property that an irreducible proof (i.e. one to which
no further reduction is applicable) of a decidable sentence
belongs to the real part of the theory, and it is shown by
finitary means for a certain assignment of ordinals < € 0 to
proofs of decidable sentences that the assigned ordinal is
lowered by a reduction, which by (TI) implies that every
proof of a decidable sentence in fIrst order arithmetic can be
reduced to a proof in the real part of arithmetic.

The third step thus shows how to transform proofs in a theory

T of decidable sentences to proofs in the real part of T. As we
noted in (a) of section 3, success in carrying out Hilbert's program
for a theory T implies a transformation for proofs in T of all real
sentences in T, i.e. not only the decidable ones but also those of
the form VxA(x), to real proofs, but the latter proofs do not
necessarily belong to the real part of T; in fact, it can be shown
that for certain V xA(x) provable in first order arithmetic there can
 P~ilosophical aspects o/proo/ theory 265

be no such proof in the real part of T. But the transformation

obtained in step 3) is obviously sufficient for Hilbert's program
since it shows that every instance A(n) of a provable real sentence
YxA(x) is true from which follows that alsoYxA(x) is true.
The only doubts that one can have from a finitary point of view
about these three steps concern step 2), which is the only point
where the finitary part of the object theory in question is tran-
scended (as Godel's incompleteness result requires). To avoid
confusions, it should be noted that so-called transfinite induction
is used in 2) only for decidable sentence while the object theory in
question contains ordinary induction used on any sentence,
decidable or not.
Takeuti follows the same pattern I) - 3) but uses a higher
ordinal and considers a stronger theory, viz second order arith-
metic where the comprehension axiom

3XYx(xEX ~A(x»

is restricted to formulas A(x) that are n~ in-the-wider-sense, which

means that A can be transformed in a simple way to an equivalent
formula of the form VYB(y) where B(Y) is first order. This system
is sufficient to codify an essential part of classical analysis but is
weaker than the theory that Hilbert had in mind which had no
restrictions on the comprehension axiom.
Takeuti obtains his result by using ordinal notation for a much
larger initial segment of the ordinals, and possible doubts about
the finitary character of the demonstration in step 2) of the
principle (TI) are accordingly strengthened. Two recent mono-
graphs, one by Takeuti [85] and one by SchUtte [86], give
detailed expositions of Takeuti's consistency proof.

6. The significance of the program. Recent discussions

As we have seen, Hilbert's own motives for his program were

rather ambiguous. Certain of his statements seem to indicate that
he takes moderate formalism literally while others express a com-
plete trust in all the methods used in classical analysis, in par-
ticular the use of the excluded third. A more representative
position for most work on Hilbert's program seems to be the one
266 D. Prawitz

taken by Gentzen, who ascribes different kinds of meaning also to

the ideal sentences but takes an actual transfinite interpretation
and the corresponding forms of inferences to be more doubtful
and therefore to require a consistency proof.
The most prominent, recent exponent of such a view is Takeuti.
In his book [85], he discusses the standpoint of the 'infinite
mind', who is supposed to be able to examine infinitely many
objects one by one, to form arbitrary subsets of a given domain, to
let these subsets form a new domain of objects and so on. He con-
siders classical mathematics in the form of second order arithmetic
or the theory of finite types to be quite clear from such a stand-
point but not set theory, which even from this standpoint contains
problematic parts.
But regardless of how clear and transparent this world of the
infmite mind may appear, it is still an imaginary world for us, and
we need to be reassured of its existence in one way or another,
Takeuti says. He admits that 'mathematicians have an extremely
good intuition about the world of natural numbers as conceived
by an infinite mind' and that therefore the consistency question
is not particularly important there. But he insists that 'in con-
trast, we can conceive of the world of sets only through our
imagination and our mathematical experience' and that 'con-
sequently, the problem of consistency of the comprehension
axiom is a serious and important foundational question'. 'Con-
sistency' is here used as 'a sort of foundational watchword',
whose frrst implication is that no contradiction can be derived;
Takeuti adds that this implication is the most important assurance
about the imaginary world of the infinite mind, but that we would
sometimes like to know more.
The most outspoken recent critic of this position (which I have
called moderate formalism) is Kreisel, e.g. in [ 11, 75, 77, and 84] .
Some of his points are the following ones:

(a) Long, tedious proofs that use elementary means are often less
reliable (because of the great risk of error) than short proofs
that use more abstract means and make us understand why
the result holds; hence, the whole idea of increasing the
reliability of abstract principles by a combinatorically com-
plex consistency proof is misconceived.
 Philosophical aspects of proof theory 267

(b) As a matter of fact, Gentzen's consistency proof has not

increased our confidence in natural number theory. But if any-
thing like the doctrine of moderate formalism is valid, a con-
sistency proof should increase our confidence in the theory by
100%. Hence, empirical experience refutes the doctrine.
(c) Since consistency does not exclude the possibility that an exis-
tential sentence 3xA(x) is provable while all its instances
A(n) are false, consistency does not ensure the existence of
the particular concepts for which the axioms were intended.
(d) The ideas and principles of classical mathematics, including
set theory understood in terms of Zermelo's cumulative
hierarchy, are 100% reliable and hence a proof of their con-
sistency cannot increase our confidence in them.

Of these arguments, (a) and (b) seem to me rather weak. The prem-
iss of the argument (a) about the need of abstract notions is an
observation that most people in the Hilbert-school would certainly
accept, at least partly, as far as the risk of errors in long combina-
torial proofs is concerned (see e.g. Bernays [87]) - after all,
they wanted to save classical mathematics just because of the ease
with which we obtain short and elegant proofs when using abstract
notions. But of course, the premiss is not a carte blanche for the
use of any abstract principles, and if one really has doubts about
some of them, the use of trusted principles in one consistency
proof can be carefully checked once and for all, even if very long,
and this would then certainly increase our confidence.
As for the argument (b), Takeuti [88] explicitly asserts that his
confidence in natural number theory is increased by Gentzen's
consistency proof. But most people probably agree with Tarski's
[89J often quoted statement in a discussion of Gentzen's use of
Eo-induction: '1 cannot say, however, that the consistency of
arithmetic is now ... more evident than by epsilon .. .'. However,
the reasons why this proof has not increased our confidence in
the theory are clear: firstly, as already pointed out, almost nobody
takes moderate formalism literally, but at most certain classical
concepts are doubted, and then in particular the impredicative
ones of classical analysis; secondly, the doubts there can be about
the use of tertium non datur in first order natural number theory
are expelled by the double negation (or equivalent) interpretations
268 D. Prawitz

of classical first order number theory into intuitionistic first order

number theory (due to Kolmogorov, Code] and Gentzen) if one
accepts the latter theory. Hence, Kreisel's argument (b) is hardly
relevant against most proof theorists who seek a consistency proof
for impredicative notions; at most, it is an open question what
reactions such a proof would meet.
As an (empirical) argument against those who take moderate
formalism literally, however, Kreisel's argument (induding the
talk about 100%) is most relevant, and the important conclusion
to draw is the one already mentioned above (section 3.2): for
those who doubt transfinite principles, it cannot be enough to
establish con,!j~tency or the truth of the real sentences. A reason-
able foundation of mathematics cannot treat the transfinite part
of mathematics as an instrument, a black box, that happens to
give correct results; the weakness of such an instrumentalistic
position, i.e. the position of moderate formalism taken literally,
is obvious since the foundational task must be to explain why the
instrument works, i.e. to understand it. Of course, we do attach
some meaning to the transfinite notions, which guides our formu-
lation of transfinite principles, and in case this meaning is not
sufficiently clear, the task must be to explicate it or extract its
mathematically fruitful ingredients. In short, to make Hilbert's
program at all credible, one must require that it yields an inter-
pretation of also the ideal sentences. This is a point stressed e.g.
by Kreisel [78) (see also the parenthetical remark in point (3)
below).
The same is true if the motive for the program is not any
doubts about the transfinite part but a demand for a constructive
foundation as discussed in section 3.3. A demonstration of con-
sistency, of the truth of the provable real sentences, or of the
eliminability of ideal elements from proofs of real sentences
cannot be expected to yield in itself a reasonable interpretation
of the ideal sentences. And as explained in more detail in earlier
sections, Kreisel's argument (c), already adduced by Godel [81], is
a counterexample to such an expectation.
The need of something more than a consistency proof is also
admitted by Takeuti [85) as already mentioned and in effect by
most proof-theorists. Hilbert, Herbrand, Centzen, and Takeuti
have all tried to find certain interpretations of the ideal sentences
 Philosophical aspects ofproof theory 269

that occur in proofs, namely, interpretations which depend on the

proofs in question.
The crucial foundational problem is thus not the truth of the
provable real sentences or the eliminability of ideal elements from
proofs of real sentences but the pro blem of the meaning of the so-
called ideal sentences. The significance of Hilbert's program must
therefore be discussed in such a broader perspective when the pro-
gram is understood or modified so as to include this question of
interpretation. Three or four main lines may be distinguished in
this discussion.

(l a) In want of a complete trust in the usual understanding of the

classical conceptions, one may seek a constructive interpre-
tation in terms of which all principles of classical mathe-
matics become valid, in other words, a constructive reduction
of the entire classical mathematics. This is e.g. Gentzen's and
Takeuti's position.
(1 b) Without any distrust in classical mathematics, one may seek a
constructive foundation of it as the fulfilment of the axio-
matic tradition - this may be Hilbert's position (sec. 3.3).
(2) Doubting the classical principles, one may seek a constructive
foundation of mathematics, not by an interpretation of all
classical concepts, but by developing mathematics construc-
tively as far as possible at the possible expense of sacrificing
certain classical principles. This is the line fIrst taken by
Brouwer.
(3) A third line is to maintain that classical mathematics can be
made or is already intelligible in its own non-constructive
terms and therefore is completely reliable. This line is repre-
sented by Kreisel (argument (d) above), who refers to the
cumulative hierarchy when it comes to interpreting set-
theory and who seems to think that no more explicit mean-
ing-theory is needed or that the subject is not ripe for such a
foundational study. In a less negative or pessimistic version of
this line, one would try to analyze the meaning of the classi-
cal notions further in a non-reductive way. (However, speak-
ing about Kreisel it should be remarked that although he
argues that there is no need in the sense of (I a) or (I b) for a
constructive interpretation, he has nevertheless already in
270 D. Prawi!z

[90J made a contribution to the project of finding such an

interpretation, viz. the so-called no-counterexample-interpre-
tation.)

Kreisel's attitude towards these matters has influenced several

logicians, including e.g. Bernays [91 J. But others like Feferman
[92 J maintain with Takeuti that we do not really know what set-
theoretical statements mean in the same way that we know what
arithmetical statements mean; we have a complete mental picture
of the totality N of natural numbers but not of the supposed
totality of the subsets of N. If this view about meaning in mathe-
matics is granted, there is of course a question about consistency
(although there can be different pragmatic considerations concern-
ing the likelihood of an inconsistency).
This controversy cannot be decided in an easy way, and it can
hardly be said that line (3) has been supported by a careful analy-
sis. The greater confidence in set theory today compared to the
beginning of the century is rather 'based partly on experience
partly on habit and simply not thinking about the subject' as
Takeuti [88 J says. Kreisel's attitude seems to be that nothing
more can be done in that direction, but otherwise one may regard
line (3) as constituting an open problem, another program for the
foundation of mathematics.
However, the doubts that one may have about line (3) do not
make line (la) much more reasonable. In a discussion of that
position, one must consider in more detail what interpretations
it offers. For nt-analysis Takeuti [85 J pays especial attention to
what he calls modulations of formulas which are obtained by taking
the closure of certain instances of quantified formulas, the details
of which depend on the proof in question. (E.g., V XA(X) may be
replaced by VX 1 VX 2 ••. VXn VX 1 VX2 ... VxmA(T), where T is a
second order term or formula such that ACT) has been inferred
from V XA(X).) As a consequence of a consistency proof for
(second order) arithmetic that proceeds by proving the Hauptsatz,
Takeuti notes the following corollary: if p is a proof of a sequent
r -+ ~ in (full) second order arithmetic, then there is a so-called
modulation r' -+ ~' of r -+ ~ determined by p such that r' -+~'
is provable in predicative second order arithmetic and logically
implies r -+ ~.
 Philosophical aspects of proof theory 271

The reductive interpretations of classical theories have not been

convincing, however. For instance, in Takeuti's [85] interpre-
tation by modulations, two provable sequents A ~ Band B ~ e
may be interpreted by their stronger and predicatively provable
modulations A' ~ B' and B" ~ e", but from this we cannot con-
clude that A' ~ crt is provable predicatively (although B' ~ B"
holds, it is in general not provable predicatively), and hence, the
role of ,~, is drastically changed by this interpretation.
It cannot be excluded that other more convincing interpre-
tations may be found along these lines. But if a foundation of
mathematics is to be based on certain elementary principles such
as finitary, intuitionistic or predicative ones, the natural procedure
seems to be to develop mathematics with the help of such
principles and to see how far one gets. Feferman [93], Friedman
[94] and Martin-Lof [35] have recently developed intuitionistic
theories along this line (2). (Friedman sometimes calls his work a
contribution to Hilbert's program but from the point of view dis-
cussed here it is clear that it is rather a contribution to Brouwer's
program.) As Feferman [92] puts it, it is of interest to isolate that
part of mathematics that can be developed predicatively; and as is
wen-known, most classical analysis can be developed in this way.
Takeuti [85] rejects what he calls the quasi-foundational ten-
dency to replace the results of mathematics by weaker construc-
tive theorems. Although he does not make very clear what is
meant by the derogatory term 'quasi-foundation', it probably only
denotes the attempts to replace theorems one by one by weaker
constructive theorems and not the endeavors towards a construc-
tive building of mathematics.
The position (I b) stays aloof from discussions about the reli-
ability of classical mathematics but suffers of course from the same
weakness as (la) when it comes to provide a constructive interpre-
tation.
In conclusion, it must be said that there is a strange tension in
foundational works along the lines of Hilbert's program. On one
side, one highly appreciates and wants to justify the use of all
classical notions. On the other side, one requires the justification
to be constructive, and to obtain such a justification, one tends to
consider the classical part of mathematics as a mere instrument.
When this untenable, instrumentalistic or formalistic position is
272 D. Prawitz

given up, and the so-called ideal sentences are admitted to have a
meaning, one tries to give a constructive interpretation of all the
classical notions but rejects what seems to be the more natural
alternative of developing mathematics constructively as far as
possible. Such a position is not very convincing, and it is hardly
surprising that it has not been possible to arrive at a foundation
of mathematics in this way.
This does not mean of course that work on the program cannot
be relevant for other aims. The eliminability of transfinite reason-
ing from proofs of finitary results - to the extent that this can be
demonstrated - is in itself of interest. In general, work along the
lines of Hilbert's program has given us insight in axiomatic ques-
tions about the relation between certain systems; for a survey of
works in proof theory from this aspect, see Kreisel [75].
Indirectly, such results may contribute to a better understanding
of certain foundational questions.
But it seems fair to say that the philosophically most interesting
problems of proof theory now concern certain general questions
about proofs about which there is no general agreement today.
The main philosophical significance of Hilbert's program may be
the discussions that it has given rise to and the resulting doubts
that a satisfying foundation of mathematics can be obtained in
that direction.

Acknowledgement

I am greatly indebted to Carlo Cellucci, Georg Kreisel, Per Martin-

Lof, and Goran Sundholm for detailed and valuable comments on
earlier drafts of this paper and to Neil Tennant for checking my
English.

BIBLIOGRAPHY

Bernays, P. [4] Mathematics as a Domain of Theoretical Science and of

Mental Experience. In H.E. Rose and J.C. Sheperdson, Eds., Logic
Colloquium '73, pp. 1-4. Amsterdam 1975.
- [82] On the Original Gentzen Consistency Proof for Number Theory. In
J. Myhill et al., Eds., Intuitionism and Proof Theory, pp. 409-417.
Amsterdam 1970.
 Philosophical aspects of proof theory 273

Bernays, P. [87] Hilberts Untersuchungen uberdie Grundlagen der Arithmetik.

In D. Hilbert, Gesammelte Abhandlungen, pp. 196-216. Berlin 1935.
- [91] Nachwort. lahresbericht der deutschen Mathematiker Vereinigung
81 (1978): 22--24.
Beth, E.W. [47] Semantic entailment and formal derivability. Mededelingen
der Koninklijke Nederlandse Akademie van Wetenschappen, afd. /etter-
kunde, New Series 18. Amsterdam 1955.
Cellucci, C. [7] Teoria della dimostrazione. Torino 1978.
Church, A. [15] Introduction to Mathematical Logic. Princeton 1956.
Dummett, M. [16] Frege's Philosophy of Language. Worcester 1973.
[66] The Philosophical Basis of Intuitionistic Logic. In H.D. Rose and
J .C. Shepherdson, Eds., Logic colloquium '73, pp. 5--40. Amsterdam
1975. Also reprinted in M. Dummett, Truth and other Enigmas.
Worcester 1978.
- [67] Elements of Intuitionism. Oxford 1977.
Feferman, S. [63] Review of Prawitz l5] (Ideas and Results in Proof
Theory). The Journal ofSymbolic Logic 40 (1977): 232-234.
- [92] Review of Takeuti [85] (Proof Theory). Bulletin of the American
Mathematical Society 83 (1977): 351-361.
- [93] Theories of Finite Type Related to Mathematical Practice. In J.
Barwise, Ed., Handbook of Mathematical Logic, pp. 913-971. Amster-
dam 1977.
Friedman, H. [94] Set Theoretic Foundations for Constructive Analysis and
the Hilbert Program. Mimeographed, State University of New York at
Buffalo, 1975.
Gentzen, G. [17] Untersuchungen tiber das logische Schlie~en. Mathe-
matische Zeitschrift 39 (1934): 176-210 and 405-431. English
translation in Szabo [70]. French translation by J. Ladriere and R.
Feys, Presses Universitaires de France 1955.
- [31] Neue Fassung des Widerspruchsfreiheitsbeweises fUr die reine Zahlen-
theorie. Forschung zur Logik und zur Grundlegung der exakten Wissen-
schaften, New Series, No.4: 19-44. Leipzig 1938. English translation
in Szabo [70] .
- [76] Die Widerspruchsfreiheit der reinen Zahlentheorie. Mathematische
Annalen 112 (1936): 493-565. English translation in Szabo [70].
- [83] Der erste Widerspruchsfreiheitsbeweis fur die klassische Zahlentheorie.
Archiv fiir mathematische Loglk und Grundlagenjorschung 16 (1974):
97 --112. English translation in Szabo [70] .
Girard, J.Y. [38] Une extension de l'interpretation de Godel a l'analyse,
et son application a \'elimination des coupures dans \'analyse et la
theorie des types. In J.E. Fenstad, Ed., Proceedings of the Second
Scandinavian Logic Symposium, pp. 63--92. Amsterdam 1971.
- [39] Interpretation Fonctionnelle et Elimination des Coupures de l'Arith-
mhique d'ordre superieur. These, Universite Paris VII, 1972.
- [54] Three-valued logic and cut-elimination: The actual meaning of
Takeuti's conjecture. Dissertationes Mathematicae IJ~. Warzawa 1976.
274 D. Prawitz

G6del, K. [68] Remarks before the Princeton Bicentennial Conference on

Problems in Mathematics. In M. Davis, Ed., The Undecidable: Basic
Papers on Undecidable Propositions, Unsolvable Problems, and Com-
putable Functions. New York: Hewlett. Reprinted in R. Klibansky,
Ed., Contemporary Philosophy, Vol. I. Firenza 1968.
- [81] Contribution to: Diskussion zur Grundlegung der Mathematik.
Erkenntnis 2 (1931): 135-155.
Hilbert, D. [l] Neubegriindung der Mathematik. Erste Mitteilung. Abhand-
lungen aus dem mathematischen Seminar der Hamburgischen Universi-
tiit I (1922): 157-177. Reprinted in D. Hilbert, GesammelreAbhand-
lungen. Berlin 1935.
- [2]Axiomatisches Denken. Mathematische Annalen 78 (1918): 405-415.
Reprinted in D. Hilbert, Gesammelte Abhandlungen. Berlin 1935.
- [3] Uber das Unendliche. Mathematische Annalen 95 (1926): 161-190.
Reprinted in D. Hilbert, Grundlagen der Geometrie, 7th ed. Leipzig
1930; English translation in P. Benacerraf and H. Putnam, Eds., Phi·
losophy of Mathematics: Selected Readings. Englewood Cliffs 1964;
and in J. van Heijenoort, From Frege to Gode!. Cambridge, Mass.
1967.
- [72] Mathematische Probleme. Archiv der Mathematik und Physik, 3rd
series, I (1901): 44-63,213-237. Reprinted in D. Hilbert, Gesam-
melte Abhandlungen. Berlin 1935.
- [74] Die Grundlagen der Mathematik. Abhandlungen aus dem mathe-
matischen Seminar der Hamburgischen Universittit 6 (1927): 65 -85.
Reprinted in D. Hilbert, Grundlagen der Geometrie, 7th ed., Leipzig
1930. English translation in J. van Heijenoort, From Frege to Godel.
Cambridge, Mass. 1967.
Hilbert, D., and Bernays, P. [73] Grundlagen der Mathematik I. Berlin 1934.
Hintikka, J. [48] Form and Content in Quantification Theory: Two Papers in
Symbolic Logic. Acta Philosophica Fennica 8 (Helsinki 1955): 7-55.
Howard, W. [59] The Fonnulae-As-Types Notion of Construction. Un·
published manuscript 1969.
Jervell, H.R. [32] A Normal-Form in First Order Arithmetic. In J.E. Fenstad,
Ed., Proceedings of the Second Scandinavian Logic Symposium,
pp. 93-108. Amsterdam 1971.
Kanger, S. [49] Provability in Logic. Stockholm 1957.
Kreisel, G. [11] What Have We Learned from Hilbert's Second Problem? In
F. Browder, Ed., Proceedings of the Symposia in Pure Mathematics
28 (Providence 1976): 93-130.
- [12] Some Facts from the Theory of Proofs and Some Fictions from
General Proof Theory. In J. Hintikka et aI., Eds., Essays on Mathemati·
cal and Philosophical Logic, pp. 3-23. Dordrecht 1979.
- [14] On the Kind of Data Needed for a Theory of Proofs. In J. Barwise et
a!., Eds., Logic Colloquium 76, pp. 111-128. Amsterdam 1977.
- [53] Observations on a Recent Generalization of Completeness Theorems
due to Schutte. In J. Diller and G.H. Muller, Eds.,ProofTheorySym-
 Philosophical aspects of proof theory 275

posium Kie11974: Lecture Notes in Mathematics 500: 164--181. Berlin

1975.
- [56] Foundations of Intuitionistic Logic. In E. Nagel et aI., Eds., Logic,
Methodology, and Philosophy of Science, pp. 198-210. Stanford
1962.
- [57] Mathematical Logic. In T.L. Saaty, Ed., Lectures on Modern Mathe-
matics, Vol. III, pp. 95-195. New York 1965.
- [62] A Survey of Proof Theory II. In J.E. Fenstad, Ed., Proceedings of the
Second Scandinavian Logic Symposium, pp. 109-170. Amsterdam
1971.
- [71] Review of the Collected Papers of Gerhard Gen tzen. The Journal of
Philosophy 68 (1971): 238-265.
- [75] A Survey of Proof Theory. The Journal of Symbolic Logic 33 (1968):
321-388.
- [78] Hilbert's Programme. Dialectica 12 (1958): 346-372. Reprinted in
P. Benacerraf and H. Putnam, Eds., Philosphy of Mathematics: Select-
ed Readings. Englewood Cliffs 1964.
- [84 J Wie die Beweistheorie zu ihren Ordinalzahlen kam und kommt.
Jahresbericht der deutschen Mathematiker Vereinigung 78 (1976):
177-223.
- [90] On the Interpretation of Non Finitist Proofs - Part I. The Journal of
Symbolic Logic 16(1951): 241-267.
Kreisel, G., and Krivine, J .. L. [8J Elements of Mathematical Logic. Amster-
dam 1971.
Kreisel, G., and Levy, A. [77] Reflection Principles and their Use for Estab·
lishing the Complexity of Axiomatic Systems. Zeitschrift fiir mathe-
matische Logik und Grundlagen der Mathematik 14 (1968): 97-142.
Kreisel, G., Mints, G.E., and Simpson, S.G. [52] The Use of Abstract
Language in Elementary Mathematics: Some Pedagogic Examples. In
R. Parikh, Ed., Logic Colloquium: Lecture Notes in Mathematics 453:
pp. 38-131. Berlin 1975.
Kreisel, G., and Takeuti, G. [79] Formally Self-Referential Propositions for
Cut-Free Classical Analysis and Related Systems. Dissertationes Mathe-
maticae 118 (1974).
Martin-LOf, P. [22] Infmite Terms and a System of Natural Deduction. Com-
positioMathematica 24 (1972): 93-103.
- [33] Hauptsatz for the Intuitionistic Theory of Iterated Inductive Defi-
nitions. In J.E. Fenstad, Ed., Proceedings of the Second Scandinavian
Logic Symposium, pp. 179-216. Amsterdam 1971.
- [35J An Intuitionistic Theory of Types: Predicative Part. In H.E. Rose and
J.C. Shepherdson, Eds., Logic Colloquium 73, pp. 73-118. Amster-
dam 1975.
- [40] Hauptsatz for the Theory of Species. In J.E. Fenstad, Ed., Proceed·
ings of the Second Scandinavian Logic Symposium, pp. 217--233.
Amsterdam 1971 .
276 D. Prawitz

- [41] Hauptsatz for Intuitionistic Simple Type Theory. In P. Suppes et al.,

Eds., Logic, Methodology, and Philosophy of Science IV, pp.279-
290. Amsterdam 1973.
- [61] About Models for Intuitionistic Type Theories and the Notion of
Defmitional Equality. In S. Kanger, Ed., Proceedings of the Third
Scandinavian Logic Symposium, pp. 81-109. Amsterdam 1975.
Myhill, J. [69] Some Remarks on the Notion of .Proof. The Journal of
Philosophy 57 (1960): 461-471.
Osswald, H. [30] Vollstandigkeit und Schnittelimination in der intuitioni-
stischen Typenlogik. Manuscripta Mathematica 6 (1972): 17--3 L
Pappinghaus, H. [551 Existentiell schwache Vollstiindigkeit der Logik,
zweite Stufe und Schnittelimination. Dissertation, Heidelberg 1977.
Pottinger, G. [37] Normalization as a Homomorphic Image of Cut-Elimi-
nation. Annals of Mathematical Logic 12 (I977): 323-357.
Prawitz, D. [5] Ideas and Results in Proof Theory. In J .E. Fenstad, Ed.,
Proceedings of the Second Scandinavian Logic Symposium, pp. 235-
307. Amsterdam 1971.
- [6] The Philosophical Position of Proof Theory. In R. Olson and A. Paul,
Eds., Contemporary Philosophy in Scandinavia, pp. 123·-134. Balti-
more 1972.
- [9] Meaning and Proofs: On the Conflict between Classical and Intuition-
istic Logic. Theoria 48 (1977): 2-40.
- [10] Proofs and the Meaning and Completeness of the Logical Constants.
In J. Hintikka et al., Eds., Essays on Mathematical and Philosophical
Logic, pp.25-40.Dordrecht 1979.
- [13] Towards a Foundation of a General Proof Theory. In P. Suppes et
al., Eds., Logic, Methodology, and Philosophy of Science IV, pp. 225-
250. Amsterdam 1973.
- [18] Natural Deduction. A Proof-Theoretical Study. Uppsala 1965.
- [27] Completeness and Hauptsatz for Second Order Logic. Theoria 33
(I967): 246-258.
- [28] Hauptsatz for Higher Order Logic. The Journal of Symbolic Logic
33 (1968): 452--457.
-- [29] Some Results for In tuitionistic Logic with Second Order Quanti-
fiers. In J. Myhill et aI., Eds., Intuitionism and Proof Theory, pp.
259-269. Amsterdam 1970.
- [45] Validity and Normalizability of Proofs in First and Second Order
Classical and Intuitionistic Logic. To appear in the proceedings of the
Italian Congresso Nazionale di Logica, Montecatini 1-5 October 1979.
- [51] Comments on Gentzen-Type Procedures and the Classical Notion of
Truth. In J. Diller and G.H. Muller Eds., Proof Theory Symposium
Kiel 1974: Lecture Notes in Mathematics 500: 290-319. Berlin 1975.
- (60] Constructive Semantics. In Proceedings of the First Scandinavian
Logic Symposium. Uppsala 1968.
- [64] On the Idea of a General Proof Theo!),. Synthese 27 (I974): 63-77.
Raggio, A. [19] Gentzen's Hauptsatz for the Systems NI and NK. Logique
et Analyse 30 (1965): 91-100.
 Philosophical aspects of proof theory 277

SchUtte, K. [20] Beweistheoretische Erfassung der unendlichen Induktion in

der Zahlentheorie.MathematischeAnnalen 122 (1951): 369-389.
- [50] Ein System des verkniipfenden Schlie~ens. Archiv fur mathematische
Logik und Gmndillgenforschung 2 (1956): 55-67.
- [86] Proof Theory. Berlin 1977.
Smorynski, C. [80] The Incompleteness Theorems. In J. Barwise, Ed.,
Handbook of Mathematical Logic, pp. 821-866. Amsterdam 1977.
Statman, R. [44] Structural Complexity of Proofs. Dissertation, Stanford
1974.
Stenlund, S. [23] Descriptions in Intuitionistic Logic. In S. Kanger, Ed.,
Proceedings of the Third Scandinavian Logic Symposium. Amsterdam
1975.
Szabo, M.E., Ed. [70] The Collected Papers of Gerhard Gentzen. Amsterdam
1969.
Tait, W. [21] Normal Derivability in Classical Logic. In J. Barwise, Ed., The
Syntax and Semantics of lnjinitary Languages: Lecture Notes in
Mathematics 72: 204-236. Berlin 1968.
- [24] A Nonconstructive Proof of Gentzen's Hauptsatz for Second Order
Predicate Logic. Bulletin of the American Mathematical Society 72
(1966): 980-983.
- [34] Intentional Interpretation of Functions of Finite Type I. The Journal
of Symbolic Logic 32 (1967): 198-212.
Takahashi, M. [25] A Proof of Cut··Elimination Theorem in Simple Type
Theory. The Journal of the Mathematical Society of Japan 19 (1967):
399-410.
- [26] A System of Simple Type Theory of Gentzen Style with Inference
on Extensionality, and the Cut·Elimination in it. Commentarii Mathe-
maUci Universitatis Sancti Pauli 18 (1970): 127-147.
Takeuti, G. [46] Consistency Proofs of Subsystems of Analysis. Annals of
Mathematics 86 (1967): 299-348.
- [85] Proof Theory. Amsterdam 1975.
- [88] Consistency Proof and Ordinals. In J. Diller and G.H. MUller, Eds.,
Proof Theory Symposium Kiel 1974: Lecture Notes in Mathematics
500: 365 -369. Berlin 1975.
Tarski, A. [89] Contribution to Theorie de la preuve formalisee: Discussion.
Revue lnternationale de Philosophie 8 (1954): 15-21.
Troelstra, A. [42] Notes on Intuitionistic Second Order Arithmetic. In
A.R.D. Mathias and H. Rodgers, Eds., Cambridge Summer School in
Mathematical Logic, pp. 171-205. Berlin 1973.
- [43] Metamathematical Investigation of Intuitionistic Arithmetic and
Analysis. Lecture Notes in Mathematics 344. Berlin 1973.
- [58] Principles of Intuitionism. Lecture Notes in Mathematics 95. Berlin
1969.
Zucker, J.1. [36 J Cut-Elimination and Normalization. Annals of Mathe·
matical Logic 7 (1974): 1-156.
Zucker J.I., and Tragesser, R.S. [65] The Adequacy Problem for Inferential
Logic. Journal of Philosophical Logic 7 (1978): 501-516.
Modal logic, modal semantics and their
applications
RUTH BARCAN MARCUS
Yale UniJJersity

1. Modal logic and modal semantics are primarily concerned with

the notions of necessity, possibility and related concepts. Their
development may in retrospect be seen to have proceeded in
several stages. c.1. Lewis L90] on intuitive considerations pro-
posed formal systems of modal propositional logic (MPL) extend-
ing standard propositional logic (PL) to include an operator '0' for
possibility and defining the necessity operator as '-0-', later abbre-
viated as '0'. Alternative notations in common use are 'M' and 'N'.
Central to c.1. Lewis' intuitions was the Leibnizian notion of
possible worlds where 'Op' and 'Dp' may be viewed respectively as
saying that the proposition p holds in some possible world and
that the proposition p holds in all possible worlds.
Also informing Lewis' intuitions was an alternative interpretation
of '0' and '0' as those predicates of sentences corresponding to
consistency and provability. On those readings 'O(p:)q)', ab breviated
as 'p-3q' was to be understood as 'q is deducible from p.' In sub-
sequent developments of modal logic the possible worlds interpre-
tations are more clearly distinguished from the provability inter-
pretations. Although major interest has focussed on the former,
there have been interesting results in which the modal operators
are taken as sentential predicates rather than operators. See § 13
below.
During its first propositional logic stage and in the absence of a
clear semantics and definition of validity for MPL, various alterna-
tive bases were proposed and debated on intuitive grounds. There
was particular disagreement about interpreting iterated and nested
model operators.
As refined in the subsequent literature it was seen that the

Contemporary philosophy. A new survey. Vol. 1, pp. 279-298.

© 1981, Martinus Nijhof! Publishers, The Hague/Boston/London.
280 R. Barcan Marcus

Lewis systems which were supported by some coherent set of

intuitions about the modalities, were extensions of PL which con-
tained in addition

1.1 (a) O(p:J q) :J (Op :J Oq) (Distribution)

(b) Op:Jp (Reflexitivity)
and the rule
(c) If~Athen~OA

among the debated principles were

1.2 (a) Op:JOOp (fransitivity)

(b) p:JOOp (Symmetry)
(c) Op:JOOp

which distinguished S4, the Brouwersche system and S5 respec-

tively. (The designations in parentheses characterize properties of
the relationship between worlds which must hold in possible-world
semantics if those principles are to be held valid. See §3 below).

'2. The second stage in the evolution of modal logic came with the
extension of MPL to include standard quantification theory (MQL)
and identity (MQL=) as in Barcan (Marcus) [7, 8] and Carnap
[19,20] and higher orders (MQ2 L=) as in Barcan (Marcus) [8].
Theses which held in those original extensions were

2.1 (a) O(3x)q,:J(3x)Oq,

and
(b) (3x)Oq,:JO(3x)q,

The Marcus [8] version of MQL= also contained the controversial

theorem

2.2 (x)(y)(x=y:JD(x=y»

Such extensions generated further debates about interpretation.

Indeed the plausibility, coherence and usefulness of quantified
 Modal logic, modal semantics and their applications 281

modal logic has persistently been questioned. See for example

W.V. Quine [125, 126]. Among the controversies around which a
large literature evolved are

2.3 Essentialism. The interplay of quantifiers and modal

operators commits interpretations of MQL to the Il'eaning-
fulness of essentialist sentences; a commitment which is
held to be invidious. An essentialist assertion is one which
seems to sort properties into those which a thing has
necessarily and those which it does not have necessarily as for
example in de re readings of '(3.x)OFx' and '(3.x)(GxO-Gx)'.
2.4 Identity and substitutiJlity. If identity is introduced into
MPQ in a classical way 2.2 is a theorem. This would seem to
deny that there are contingent identities. But what of 'the
morning star = the evening star'?
If the principle of substitutivity of names of identicals is
preserved then interpretations of MQL= would seem to
generate problems common to substitution in opaque con-
texts. For example since the evening star and the morning
star are identical one can derive the apparent falsehood
'0 (the evening star = the morning star)'.
2.5 Proliferation of objects. Given 2.1-2.4, interpretations of
MQL and MQL= would seem to require intensional objects;
individual concepts as well as actual individuals; attributes as
well as sets and so on, which, contra Occam, creates a meta-
physical morass.

Despite the persistent criticism, interest in various syntactic

constructions of formal systems of modal logic did not abate and
still continues. As noted above, a primary motivation was to give
a more systematic account of some intuitive conceptions of logical
or more generally, metaphysical necessity and possibility. The
categorization 'metaphysical necessity' was introduced since it is
not supposed that all necessary truths are also logical truths; con-
sider for example mathematical propositions. It was also seen that
such systems could be adapted to alternative interpretations of tht
operators. Some of those readings ranged over more restricted
notions of possibility and necessity as in causal, physical or
282 R. Barcan Marcus

temporal modalities. Parallel formal systems could be constructed

alone on analogous lines, or combined in appropriate
axiomatizations. A system of causal or temporal modalities might
for example be embedded in a system of metaphysical modalities.
See for example Prior [117-120J, Burks [17,181.
In addition to the application of modal logic to the several con-
cepts of possibility and necessity it was also observed that the
"logic" of some epistemic concepts such as 'it is known that' 'it is
believed that' and deontic concepts such as 'it ought to be the case
that' shared syntactic features and inference patterns with the
more familiar operators and were included under a more general
categorization of modal logic. Von Wright [144 J for example
proposed four modal categories: alethic, deontic, epistemic and
existential. Analogous systematizations for such alternative
readings were constructed, and with respect to such systemati-
zations puzzles about identity, substitutivity and essentialism
reemerged. It remained for the subsequent development of modal
semantics to articulate the connections between various readings
of the operators.

3. In its third stage modal logic was provided with a very general
model-theoretic semantics within which validity for sentences of
ML, MQL, and MQL= could be defined. The central ideas were
elaborated hy Kanger [72, 73 J, Kripke [80-82J and Hintikka
[63--66] .
Kripke's formulation for the semantics of MPL (here somewhat
revised) begins with a model structure (G ,K,R) where K is a set
of worlds, whi~h has the actual world G as a member, R is a
relation between worlds. A model on a model structure is a
function if; which first assigns a truth value to each atomic sen-
tence at each world. True in a world can be inductively defined
by extending 1>. The clauses for truth functional connectives are
straightforward extensions of standard semantics relativised to
worlds. For the modal operator, 1>(DA ,II) = T just in case 1>(A ,H')
= T for every H' such that HRH'. A sentence is valid just in
case it is true at every world in every model on every model
structure.
The semantic clarifies questions about iterated modalities.
Validity of l.l(b), 1.2(a) and 1.2(b) requires that R be reflexive,
 Modal logic , modal semantics and their applications 283

transitive, and symmetric, respectively. A variety of formal sys-

tems of modal propositional logic constructed over the years may
in terms of the semantics be seen as placing additional constraints
on R. For a partial survey of such systems see Feys [40], Prior
[ 117), Segerberg, [131), Hughes and Cresswell [69], Lemmon
[39), Zeman [153] .
The semantics of MQL expands the model structure to
(G,K,R,U) where U is a universe of individuals. A function 1/1
assigns to each world H a non-empty subset of U; the domain of
H. Such domains mayor may not intersect. In addition to assign-
ing truth values to sentence letters, the function ¢(pn ,H) assigns
to each n'adic predicate n ~ I a set of n'tuples at H where H
ranges over worlds, and the n'tuples are ordered elements of U.
Kripke adopts a convention whereby the elements which make up
the n'tuples assigned to predicates at a world need not be in the
domain of that world. For atomic predicates ¢(pn Xl .... Xn ,H) =T
just in case the elements assigned to Xl . ... Xn make up an ordered
n'tuple which is a member of ¢(pn H). The induction clause for
quantification says that ¢«X)A(XJlI, ... Yn),H)=T just in case
¢(A(XJ1I, ... Yn), H) =T for every assignment to X of an element
of the domain of H and for every assignment to Yi of an element
of U. Quantification is world bound.
Just as debates about iterated modalities were clarified by
the semantics for MPL, so debates about mingling quantifiers and
modal operators were clarified by the semantics for MQL. Theses
like 2.1 (a) and 2.1 (b) were seen to require inclusion relations
amo ng domains of worlds. If each world is assigned the same
domain then 2.1 (a), (b) are both valid. On such an assignment
possibilia, objects outside the domain of the actual world would
be excluded.
It remained to provide a semantics for MQL=, which requires
the clause that ¢(x=y),H)=T just in case x andy are assigned the
same object. Since for Kripke variables are assigned the same
objects across worlds it follows that if ¢«x=y),H) =T then
¢«x=y),H')=T for all H' where HRH', thus validating 2.2; the
principle of necessity of identity which preserves substitutivity of
identity in modal contexts.
There are of course alternative semantical bases for identity in
MQL. One may want to distinguish "contingent" identities from
284 R. Barcan Marcus

non-contingent ones, and allow divergence and convergence of

objects across worlds. Assignments to singular terms may then be
world bound in their reference. See for example Hintikka [66].
Necessity of identity might, with multiple reference, still be pre-
served by requiring that if x and yare assigned the same object in
H then they are assigned the same object in all H' where HRH',
although the object assigned to both x and y may vary across
worlds. Alternatives multiply. In D. Lewis [91, 93] domains of
worlds are disjoint and absolute identity is world bound. However,
he introduces a quasi-identity relation across worlds which holds
between an object in a world and its 'counterpart' in an alternative
world.
Semantics for MQL= which accommodates multiple reference will
of course require modification of standard principles like Leibniz'
law.

4. Summary of recent developments. Not all of the recent work in

modal logic followed directly in the wake of semantical accounts.
Syntactic constructions of modal systems continued to be ex-
plored, their formal properties examined and intuitive readings of
the operators proposed. This was especially true in deontic logic
and tense logic. (See Von Wright [146-148], Prior [118-121],
Rescher and Urquhart [128], Hilpinen [62].) More prominent
were the variety of interesting ways in which the semantical
accounts were employed. They were used to clarify old and
persistent controversies about identity, substitution and essential-
ism in modal contexts. They helped to illuminate more general
metaphysical and epistemological issues surrounding identity,
essences, natural kinds, and counterfactuals. They formed a basis
for new theories of meaning and reference. They suggested impor-
tant ways of viewing the relationship between epistemological
concepts like a-prioricity, metaphysical concepts like necessity,
and linguistic concepts like analyticity. The theories also received
application as a semantical base for the "logic" of concepts such as
temporality, causal necessity, obligation, as well as those charac-
teristics as propositional attitudes. Those who were diffident
about modal logic were stimulated to explore alternative frame-
works which could do the work of modal theories. There was also
a revived interest in syntactical interpretations of modal operators.
 Modal logic, modal semantics and their applications 285

As was inevitable, modal semantics generated its own puzzles.

What is a possible world, what is a possible individual - and how
are such individuals to be tracked across worlds. There was also
the question, despite the apparent successes, whether modal
languages, on some assumptions about adequacy, can be provided
with an adequate truth theory. Only a few of these topics will be
touched on in the present survey.

5. Essentialism. Interpretations of MQL permit the formation of

'essentialist' sentences such as 'Socrates is necessarily a man' or
'Someone is necessarily a man.' Such assertions on the narrow-
scope (de re) reading of the modal operator seem to say of a thing
that is has a property in a certain way or mode, i.e. essentially or
necessarily. It has therefore been claimed that modal logic is
"committed to essentialism" (See 2.3 above). In what sense com-
mitted? Granted such sentences are well-formed, is every model
of a modal system committed to the truth or more strongly, the
validity of essentialist sentences? Marcus [101], consistent with
Aristotle's claims about essential properties, proposed that an
essential property is (1) one which some objects have and other
objects fail to have and (2) objects which have them, have them
necessarily. Vacuous properties such as being human or not human,
and "referential" properties such as being identical to Socrates, are
thereby excluded. Parsons [ 112"1 showed that on such a character-
ization, there are interpretations of modal systems, maximal
models, which are consistent with the falsehood of all essentialist
sentences. Stalnaker [135] noted that the basis for the exclusion
of vacuous and "referential" properties was intuitive and, given the
extensional nature of the Kripke semantics, was not built into the
structure. To that end, Stalnaker [135] appended to the semantics
a theory of properties (qualities) as regions in a quality space in
terms of which the trivial necessary properties can be explicitly
excluded.
In answer to criticisms therefore, an anti-essentialist with res-
pect to properties need not reject modal logic. Indeed it provides
languages in which such a rejection may be clearly framed. It also
provides languages within which essentialist truths can be framed.
Ordinary as well as scientific languages with their natural-kind
terms would seem to require it.
286 R. Barcan Marcus

An additional question about essentialist sentences is raised,

independently of older debates, by the semantics of MPQ=.
'Socrates is necessarily human' means in the modal framework
that Socrates has the property of being human in all worlds in
which he exists. This, it is claimed, supposes that Socrates can be
tracked across worlds, although some of his properties may vary
across worlds. See for example Chisholm [22], Quine [125 -126]
and Kaplan [76]. One response, proposed by Kripke [183-184],
is that the claim is misleading. It confuses metaphysical issues
with epistemological ones. 'Socrates', it is argued, refers to Socrates
in whatever world he may be found. Tracking Socrates in the sense
of coming to know that an object is Socrates may require criteria
of identification; some uniq uely identifying description whether it
be in this alternative worlds. But Kripke [183 -184 J points out
support of the claim that counterfactuals like 'Socrates might not
have been executed' are nonnally understood as saying that in
alternative circumstances (worlds) Socrates, not some surrogate,
was not executed. The view has ramifications for theories of
reference. It suggests that proper names have a special role; they
refer directly and without mediation of identifying descriptions.

6. Identity and substitutivity. LIn Kripke's semantics there are no

contingent identities; it is not possible for a thing to be other than
itself. Some account must therefore be given of sentences which
on the surface appear to be contingent identities. The theory of
descriptions provides one analysis that can be given for sentences
like 'the evening star = the morning star' which contain descriptive
phrases. That analysis will yield a sentence which expresses a con-
tingent truth. As for sentences like 'Hesperus = Phosphorus' they
are not taken to be contingent identities. To view them as such is
to confuse how we come to know the truth-status of a proposition
from the truth-status it has. It can be seen that backing for such
claims requires a new theory of reference which radically dis-
tinguishes the role of proper names from singular descriptions.
See §8 below.
Frameworks which accommodate contingent identities as in
Hintikka [63-66] must be backed by some theory of divided
reference of singular terms over worlds. On the assumption that
there are contingent identities, substitutivity will fail in modal
 Modal logic, modal semantics and their applications 287

contexts. In semantical theories which accommodate such identi-

ties, there must be constraint on substitutivity.

7. Proliferation of objects. If there are contingent identities then

there are substitution failures in modal contexts. Analogously
there will be failure of substitutivity of coextensive predicates in
modal contexts. On a Fregean view, modal discourse is therefore
indirect discourse and not about the ordinary referents of terms
but about intensional objects such as senses. Since the senses of
'Hesperus' 'Phosphorus' 'the evening star' 'the morning star' differ,
substitutivity is proscribed. See Camap [20], Church [23],
Montague [107], Bressan [14], Kaplan [76]. A ground for rejec-
ting such a version of a Fregean view is that it does violence to the
actualities of modal discourse. In 'Necessarily, Socrates is human'
or 'Socrates is necessarily human' it is Socrates who is the object
of reference. Nor, as we have seen, will there be a substitution
failure in modal contexts for proper names alternative to
'Socrates'. Singular descriptions can be analyzed within the modal
framework without the addition of intensional entities over which
quantifiers range, such as individual concepts. Similarly, the
semantics proscribes substitutivity of coextensive predicates unless
they are necessarily coextensive. The modal framework does of
course accommodate some attenuated entities such as possibilia
but they are not required. Domains of worlds could be restricted
to those which are coextensive. Possibilia are not directly identi-
fiable with Fregean intensions. The latter may more properly be
understood as those functions which assigns at each world, the
extensions of terms at that world.
The point to be made is that contra Quine [125, 126] putative
failures of substitutivity are not a ground for bloating ontology.
There may of course be quite L'1dependent reasons for counten-
ancing intensional entities. Leibniz' principle of plenitude stands
in opposition to Occam's razor.

8. A new theory of reference. What emerged from the discussions

about interpretations of modal logic is a new theory of reference
and claims about logic and language which are not merely of
interest to philosophers of language but purport to illuminate
fundamental issues in epistemology and metaphysics.
288 R. Barcan Marcus

In [98] Marcus suggested that ordinary proper names 'tag'

objects and function differently from singular descriptions. No
theory was given or available to account for the difference. In
[83-84] Kripke proposed a theory of how it is that proper names
might link up with objects independent of mediating descriptions
or of speakers encounters with the object named. Proper names
are words in a language, and languages with their vocabularies are
social institutions which evolve over time. At some juncture or
junctures in that evolution some objects are given (dubbed) one
or more names. Those objects are the semantic referents of their
names, independent of any set of properties used to pick out the
object or any set of properties a speaker may subsequently have in
mind in using these names. An object's name is passed along
speaker to speaker in a chain of communication. In using a proper
name, the semantic referent is the object which initiated the chain.
A grammatically proper name is not a genuine proper name if
there is no chain which takes it back to the object initially named.
The theory has alternatively been called the 'causal theory of
names' and the 'historical theory of names.' Proper names are in a
class of designators called 'rigid' by Kripke [83-84]; such desig-
nators pick out the same object across possible worlds. But a
proper name is to be distinguished from rigid singular descriptions.
Singular descriptions have propositional content beyond the
object named. One might say in accordance with an early view of
Russell that the propositional content of a proper name is the
object itself. See Kaplan [77] .
The theory has been enlarged by Kripke [83-84] and Putnam
[ 123 -124] to include the claim that certain common nouns work
like proper names. A sample of a substance or a species may be
dubbed for example as being 'water' or 'human' without a
mediating definition of 'water' or 'human'. Such names name
substances, or kinds, or species. In the modal framework they are
essential properties; properties which if a thing has it, it has it in
all possible worlds. Elaboration and criticism of the theory has
generated a large literature. See for example Donnellan [34-35] ,
Evans [38], Peacocke [113], Plantinga [114], and Wiggins
[151].
 Modal logic, modal semantics and their applications 289

9. Metaphysical and epistemological ramifications. As noted

above, debates about interpreting modal languages issued in novel
views about meaning and reference with consequences for episte-
mology and metaphysics as well as philosophy of language. They
suggested that distinctions be made between concepts which had
traditionally been conflated. 'A-priori,' 'a posteriori,' 'conceivable,'
are, it is argued, most coherently viewed as epistemological
notions. 'Necessary,' 'contingent,' 'possible,' as metaphysical
notions. On that understanding, 'Necessarily, Hesperus is Phos-
phorus' could be true yet its falsehood conceivable (although not
possible), and its truth known a posteriori. 'Necessarily Hesperus is
Hesperus' is true as well as a priori. Similarly, for general terms:
'Necessarily water is H 2 0' is true yet its falsehood conceivable and
its truth known a posteriori. 'Necessarily water is water' is true
and also a-priori. The analysis is one step along the way to answer-
ing Frege's question as to how it is that 'a=a' is less informative
than 'a=b'.
These views can be further refined and extended. A possible
world may be seen as only one of the many features of a con-
text on which the truth of a sentence depends. The truth of sen-
tences with demonstratives and tenses will also depend on
additional features of the context of utterance such as the time,
place, speaker and the like. In {77] Kaplan points out that a sen-
tence like 'I am here now' is on any occasion of utterance true, yet
its propositional content may vary with context and it does not
express a necessary truth. Kaplan calls such sentences 'analytic',
which is close to Kripke's characterization 'a priori'.
More complete discussion of similarly generalized theories of
semantics and pragmatics will take us too far afield from the
present survey. See Montague [107], D. Lewis [93], Stalnaker
[ 134 J , and Cresswell [28] .

10. Applications. As noted above, independent of the employment

of semantical methods, systems of modal logic were adapted to a
variety of interpretations of the modalities. Some of those were
variant uses of 'possibility' and 'necessity' as in causal, physical
and temporal interpretations which could be adapted with little
modification of the underlying axiomatic structures. Other inter-
pretations such as deontic readings or epistemic reading, flowed
290 R. Barcan Marcus

from the observation that with some modification of one or

another of the familiar systems, they could be adapted in various
ways to widely different interpretations. (See §4.) The adequacy
of such systems was generally tested against their intuitive accepta-
bility when translated into ordinary language. How well for
example did they fit "moral facts" or "epistemic facts"?
The development of model theoretic semantics for modal logic
provided a new perspicuous approach to the variant interpre-
tations of the modalities. '0' was defined in such a way that 'OA'
holds in a world H in case it holds in all worlds possible relative to
H. Similarly 'A is causally necessary' holds in a world H just in
case A holds in all possible worlds relative to H where the laws of
nature obtained; A is known, believed, perceived by an individual a
just in case it holds in all worlds consistent with a's knowledge,
beliefs and perceptions respectively. A is obligatory if it holds in
all worlds where only right actions are performed. Such interpre-
tations required appropriate constraints. For example in deontic
logic the accessibility relation is not reflexive at a deontically im-
perfect world. In epistemic logic there may be constraints on substi-
tutivity of logical equivalences at a world where for example the
epistemic operators are interpreted to fit such facts as the possible
failure of an epistemic subject to know (or believe) the logical
consequences of what he (knows) or believes. Alternatively,
knowers (believers) must have complete knowledge of logical
consequences. There are further alternatives. See Hintikka [62,
65, 66]. Hansson [58]. Also the introduction by F¢l1esdal and
Hilpinen in [61] .

11. Alternative frameworks. One of the more influential proposals

for an alternative to modal logic is that of D. Lewis [91-93] who
sees modal logic and semantics as derivative from a more general,
purely extensional theory. He takes a realistic view of worlds: the
actual world is one among many. Each world is stocked with its
own domain of individuals, disjoint from that of other worlds.
There is no identity of objects across worlds but they may have
counterparts across worlds. The counterpart relation, unlike the
world bound identity relation, is neither symmetric, nor transi-
tive, nor one"'->ne. Philosophically, Lewis sees the theory as best
capturing Leibniz' view of individ uals. It also has greater generality
 Modallog;c, modal semantics and their applications 291

than modal logic in that sentences of quantified modal logic can

be translated in to counterpart theory, but not conversely. Further-
more it preserves strong extensionality. Modal operators may be
viewed as quantifiers over worlds. Of course the extensionalism is
preserved at a large metaphysical price. Under counterpart theory,
1.2(a), 1.2(b), 2.1 (a), 2.2 are not valid. 2.1 (b) is valid.
It is generally acknowledged that a motivation for modal logic
is counterfactual discourse. Lewis has in [93-94] attempted to
apply a generalization of counterpart theory to give such an
account. Worlds may be ordered with respect to similarity in
some appropriate set of respects. A counterfactual 'P [fr Q' which
has a false antecedent, may be taken as true in a world if it is true
in a closest similar world where 'P' is true. The ordering of worlds
with respect to similarity suggests analogous epistemic and deontic
interpretations. It can for example be used as a basis for a theory
of degrees of belief. For an alternative modal theory of counter-
factuals see Stalnaker [ 134] .

12. New problems or old problems in new guise. Talk of alterna-

tive states of affairs is a part of ordinary as well as scientific dis-
course. It figures crucially in probabilistic reasoning and decision
making. Modal semantics has elaborated on such discourse. But
how is a possible world to be construed'? Are they useful inventions
helpful in talking about the way the world might have been but
not to be mistaken for something analogous to a distant object?
See Kripke [83-84], Hintikka [63-66]. Are they objects on a
par with others independent of anything we might know or con-
ceive or say about them? See Lewis [91,93].
Is a world to be viewed as what corresponds to a consistent or
more strongly, maximally consistent set of sentences? Similar
questions may be raised with respect to non-actual possibilia.
Are those who, unlike D. Lewis, deny that they are realists
about possible worlds correct in their denial? Quantification over
worlds is after all carried out in the metalanguage of modal lan-
guages. One response in support of the anti-realist claim is that the
ontology of a theory is given by the values of variables within the
theory. Within modal languages construed in the manner of
Hintikka [63] and Kripke [81] quantification is over individuals
not worlds.
292 R. Barcan Marcus

13. Syntactical interpretations. Montague [104] argued that what

appears to be a natural interpretation of the modal operators as
predicates of sentences corresponding to 'provable' and
'consistent' is not feasible. He showed that on such a syntactical
treatment of modality the standard system of modal logic in-
cluding the weak system l.l(a)-(c) could not be so interpreted.
Boolos [12] has revived an interest in such interpretations in his
very extensive explorations of a still weaker system of modal logic
with modal principles l.l(a), 1.1(c) and 'D(DP:::)P) :::)DP' where
the propositional variables range over names of sentences. Also see
Solovay [133] .

14. Conclusion. This brief survey has emphasized semantical

methods in modal logic and their philosophical ramifications. Dis-
cussion of the many interesting results about formal features of
the variety of modal axiomatic systems has been neglected. The
bibliography however, includes a listing of many of those papers.

SELECTIVE BIBLIOGRAPHY··

Adams, R.M. [1] Theories of Actuality. Nous 8 (1974): 211-231.

Anderson, A. [2] The Formal Analysis of Normative Systems. In N. Rescher,
Ed., The Logic of Decision and Action. Pittsburgh 1967.
- [31 Some Nasty Problems in the Formal Logic of Ethics. Nous I (1967):
345-360.
Anderson, A., and Belnap, N. [4] Entailment: The Logic of Relevance and
Necessity. Princeton 1975.
Aqvist, L. [5] Interpretations of Deontic Logic. Mind 73 (1964): 246-253.
- [6J Modal Logic with Subjunctive Conditionals. Journal of Philosophical
Logic 2 (1973): 1-76.
Barcan, R. (also see Marcus) [7] A Functional Calculus of First Order Based
on Strict Implication. Journal of Symbolic Logic 11 (1946): 1-16.
- (also see Marcus) [8 J The Identity of Individuals in a Strict Functional
Calculus of Second Order. Journal of Symbolic Logic 12 (1947):
12-15.

*. With the exception of items mentioned in the text, the bibliography is

coni1ned to representative works published after 1965. Collections of essays
whether those of an individual author or essays by different authors will be
included where they contain items of interest not mentioned separately.
Collections will be marked by an asterisk' *'.
 Modal logic, modal semantics and their applications 293

Belnap, N. [9] See Anderson and Belnap [4] .

Bennett, J. [10] Counterfactuals and Possible Worlds. Canadian Journal of
Philosophy 4 (1974/5): 381-402.
Blackburn, S. [11] Meaning, Reference and Necessity * . Cambridge 1975.
Boolos, G. [12] The Unprovability of Consistency. Cambridge 1979.
Bowen, K. [13] Normal Modal Model Theory. Journal of Philosophical
Logic 4 (1975): 97-131.
Bressan,A. [14] A GeneralInterpretedModalCalculus. New Haven 1972.
Bull, R.A. [15] On Modal Logics with Propositional Quantifiers. Journal of
Symbolic Logic 34 (1969).
Burge, T. [16] Reference and Proper Names. Journal of Philosophy 70
(1973): 200-209.
Burks, A.W. [17] The Logic of Causal Propositions. Mind 60 (1951): 363-
382.
- [18] Choice Chance and Reason. Chicago 1977.
Carnap, R. [19] Modalities and Quantification. Journal of Symbolic Logic
11 (1946): 33-64.
- [20] Meaning and Necessity, 2nd edition. University of Chicago Press,
1956.
Castaneda, H.C. [21] Acts, the Logic of Obligation and Deontic Calculi.
Critica 1 (1967): 77-95.
Chisholm, R. [22] Identity Through Possible Worlds. Nous 1 (1967): 1-8.
Church, A. [23] A Formulation of the Logic of Sense and Denotation.
In P. Henle et al., Eds., Structure Method and Meaning*. New York
1951.
Cohen, L.1. [24] The Diversity ofMeaning, 2nd edition. Methuen, 1966.
Copi, I., and Gould, J. [25] Contemporary Readings in Logical Theory.
Macmillan, 1967.
Cresswell, M.J. [26] See Hughes and Cresswell [69], [70].
- [27] Classical Intensional Logics. Theoria 36 (1970).
CressweIl,M.1. [28] Logics and Languages. Methuen, 1973.
- [29] Frames and Models in Modal logic. In J.N. Crossley, Ed., Algebra
and Logic*. Berlin/Heidelberg/New York 1975.
Davidson, D., and Harman, G., Eds. [30] Semantics of Natural Languages*.
Reidel,1972.
- and Harman, G., Eds. [31] The Logic ofGrammar*. Dickenson, 1975.
Davis, J.W., Hackney, D.1., and Wilson, WK. [32] Philosophical Logic"'.
Reidel, 1969.
Dishkant, H. [33] Imbedding of the Quantum Logic in the Modal System of
Brouwer. Journal of Symbolic Logic 42 (1977): 321-328.
Donnellan, K.S. [34] Reference and Defmite Descriptions. Philosophical
Review 75 (1966): 284-304.
- [35] Proper Names and Identifying Description. Synthese 21 (I 970):
356-379.
Dunn, J. [36] A Truth Value Semantics for Modal Logic. In Leblanc [87].
294 R. Barcan Marcus

Eberle, R. [37] A Logic of Believing, Knowing and Inferring. Synthese 26

(1973): No.5.
Evans, G. [38] The Causal Theory of Names. Aristotelian Society, Supple.
mentary Volume XLVII (1973): 187--208.
- and McDowell, J.H. [39] Essays in Semantics. Clarendon, 1975.
Feys, R. [40] Modal Logics. Ed. J. Dopp. Paris: Gauthier-Villars, 1965.
Fine, K. [41] Propositional Quantifiers in Modal Logic. Theoria 35 (1970):
336-46.
- [42] Some Connections Between Elementary and Modal Logic. Proceed-
ings of the Third Scandanavian Logic Symposium Uppsala 1973. North-
Holland,1975.
- and Prior, A. [43] Worlds Times and Selves. Duckworth, 1977.
- [44] Properties, Propositions, and Sets. Journal of Philosophical Logic
6 (1977): 135-191.
Fitch, F. [45] A Complete and Consistent Modal Set Theory. Journal of
Symbolic Logic 32 (1967): 93-103.
- [46] A Correlation Between Modal Reduction: Principles and Properties
of Relations. Journal of Philowphical Logic 2 (1973): 97 -10 1.
Fitting, M. [47] An Embedding of Classical Logic in S4. Journal of Sym-
bolic Logic 35 (1970): 529-34.
F¢llesdal, D. [48] Quine on Modality. Synthese 19 (1968/69): 147-157.
- and Hilpinen, R. [49] Deontic Logic: An Introduction. In Hilpinen [61].
French, P., Uehling, T., and Wettstein, H. [50] Contemporary Perspectives
in the Philosophy of Language *. University of Minnesota Press, 1978.
Gabbay, D. [51] Investigations in Modal and Tense Logics with Applications
to Problems in Philosophy and Linguistics. Dordrecht 1976.
Gallin, D. [52J Intensional and Higher Order Modal Logic with Applications
to Montague Semantics. Math. Studies 19. North-Holland, 1975.
Gardies, J-1. [53] La Logique du Temps. Paris: Collection SUP, 1975.
Geach, P.T. [54] Logic Matters*. Blackwell, 1972.
Goldblatt, R. [55] Metamathematics of Modal Logic. Reports on Mathe-
matical Logic 6 (1976): 41-47 and 7 (1976): 21-52.
Haack, S. [56] Del'iant Logic. Cambridge University Press, 1975.
- [57] Philosophy of Logics. Cambridge 1978.
Hansson, B. [58] An Analysis of Some Deontic Logics. Nous 4 (I970):
373-398. Reprinted in Hilpinen [61].
Harman, G. [59] Is Modal Logic, Logic? Philosophia 2 (1972): 75-84.
Harman, G., and Davidson, D. [60] See Davidson and Harman [30] , [31].
Hilpinen, R., Ed. [61] Deontic Logic: Introductory and Systematic
Readings*. Reidel, 1971.
Hintikka, K.1.J. [62] Know/edge and Belief Cornell University Press, 1962.
- [63] The Modes of Modality. Acta Philosophica Fennica 16 (1963):
65--82. Reprinted in Linsky [96] .
- [64] Individuals, Possible Worlds and Epistemic Logic. Nous 1 (1967):
33-62.
- [65] Semantics for Propositional Attitudes. In J.W. Davis et a1., Ed.,
Philosophical Logic. Reidel, 1969.
 Modal logic, modal semantics and their applications 295

- 166] Models for Modalities *. Reidel, 1970.

- 167] Moravczik, J., and Suppes, P., Eds. Approaches to Natural Languages:
Proceedings of the 1970 Stanford Workshop on Grammar and Seman-
tics *. Reidel, 1973.
Hockney, D., Harper, W., and Freed, B. [68] Contemporary Research in
Philosophical Logic and Linguistic Semantics*. Reidel, 1975.
Hughes, G., and Cresswell, M.J. [69] An Introduction to Modal Logic.
Methuen, 1968.
- and Cresswell, MJ. [70] Omnitemporal Logic and Converging Time.
Theoria41 (1975): 11-34.
Hughes, G. [71] Ornnitemporal Logic and Modal Time. Reports on Mathe·
matical Logic 8 (1977): 41-61.
Kanger, S. [72] Provability in Logic. Stockholm Studies in Philosophy 1.
Stockholm,1957.
- [73] On the Characterization of Modalities. Theoria 23 (1957): 152-55.
- [74] New Foundations for Ethical Theory. Privately printed, Stockholm
1957. Reprinted in Hilpinen [61].
- 175] Ed., Proceedings of the Third Scandinavian Logic Symposium· 82.
Oxford/New York: North-Holland, 1975.
Kaplan, D. [76] Quantifying In. Synthese 19 (1968/69): 178-214.
- [77] Dthat* and On the Logic of Demonstratives. In P. French et aI.,
Ed., Contemporary Perspectives in the Philosophy of Language. Uni-
versity of Minnesota Press, 1978.
Klibansky, R. [78] Contemporary Philosophy I: Logic and Foundations of
Mathematics*. La Nuova Italla, 1968.
Korner, S., Ed. [79] Philosophy of LOgic*. Blackwell, 1976.
Kripke, S.A. [80] Seman tical Analyses of Modal Logic I. Zeitschrift fur
mathematische Logik und Grundlagen der Mathematik 9 (1963): 67-
96.
- [81] Semantical Considerations on Modal Logic. Acta Philosophica
Fennica 16 (1963). Reprinted in Unsky [96].
- [82] Semantical Analysis of Modal Logic II. In J. Addison, L. Henkin and
A. Tarski, Eds., The Theory ofModels. Amsterdam, 1965.
- [83] Identity and Necessity. In M. Munitz, Ed.,1dentity and Individuation.
New York: New York University Press, 1971. In Munitz [108].
- [84] Naming and Necessity. In D. Davidson and G. Harman, Eds., Seman-
tics ofNatural Languages. Reidel, 1972.
- [85] Speakers Reference and Semantic Reference. In P. French et aI., Eds.,
Contemporary Perspectives in Philosophy of Language. University of
Minnesota Press, 1978.
Lambert, K., Ed., [86] Philosophical Problems in LOgic*. Reidel, 1970.
Leblanc, H., Ed. [87J Truth, Syntax and Modality*. North-Holland, 1973.
Lemmon, E. [88 J Algebraic Semantics for Modal Logics. Journal of Symbolic
Logic 31 (1966): 46-65,191-218.
- [89J An Introduction to Modal LOgic. in collaboration with D. Scott and
edited by K. Segerberg. Oxford 1977.
296 R. Barcan Marcus

Lewis, c.I., and Langford, C.H. [90] Symbolic Logic, 1st edition, New York,
1932, 2nd edition, Dover, 1959.
Lewis, D. [91] Counterpart Theory and Quantified Modal Logic. Journal of
Philosophy 65 (1968): 113~126.
~ [92] General Semantics.Synthese 22 (1970/71): 18-67.
-- [93] Counterfactuals. Blackwells, 1973.
- [94] Counterfactuals and Comparative Possibility. Journal of Philosophical
Logic 4 (1973): 418-446.
Linsky, L. [95] Reference, Essentialism and Modality. Journal of Philosophy
66 (1969): 687-700. Reprinted in Linsky [96].
_. [96] Reference and Modality*. Oxford University Press, 1971.
Makinson, D. [97] On Some Completeness Theorems in Modal Logic. Zeit-
schrift fiir mathematische Logik und Grundlagen der Mathematik 12
(1966): 379~84.
Marcus, R. Barcan (also see Barcan) [98] Modalities and Intensional Lan-
guages. Synthese 13 (1961): 303~322. Reprinted Wartofsky [149],
and Copi and Gould [25].
- (also see Barcan) [99] Essentialism in Modal Logic.Nous 1 (1967).
~ (also see R. Barcan) [100] Modal Logic. In Klibansky [78],1968.
- (also see R. Barcan) [101] Essential Attribution. Journal of Philosophy
68 (1971).
- [102] Dispensing with Possibilia. In Proceedings of the American Philo-
sophical Association of 1976, 1977. Newark, Delaware.
Meyer, R.K. [103] Coherence in Modal Logics. Logique et Analyse 14
(1971): 658-68.
Montague, R. [104] Syntactical Treatments of Modality. Acta Philosophica
Femzica 16 (1963): 153--168.
~ [105] Pragmatics. In Klibansky [78],1968.
~ [106] Pragmatics and Intensional Logic. Synthese 22 (1970/71): 68~94.
Reprinted in Davidson and Harman [30] .
~ [107 J Forrnal Phiiosophy*. Ed. R.H. Thomason [139] . Yale, 1974.
Munitz, M.R., Ed. [108] identity and lndil'iduation*. New York University
Press, 1971.
~ Ed. [109] LogicandOntoiogy. New York University Press, 1973.
Olson, R., and Paul, A. [110] Contemporary Philosophy in Scandinavia"'.
Baltimore/London) 1972.
Parsons, T. [Ill] Grades of Essentialism in Quantified Modal Logic. Nous 1
(1967): 181-19l.
~ [112] Essentialism in Quantified Modal Logic. Philosophical Review
78 (1969): 35-52. Reprinted in Linsky [96].
Peacocke, C. [113] Proper Names, Reference and Rigid Designation. In
S. Blackburn, Ed., Meaning Reference and Necessity_ Cambridge 1975.
Plantinga, A. [114] The Nature of Necessity. Clarendon) 1974.
Poilock, J.L. [115] Basic Modal Logic. Journal of Symbolic Logic 32 (I967):
355 --365.
 Modal Logic, modal semantics and their applications 297

- [116] Logical Validity in Modal Logic. Monist 51 (1967): 128-135.

Prior, A. [117] Formal Logic, 2nd edition. Oxford University Press, 1962.
- [118] Past Present and Future. Oxford University Pre~s, 1967.
- [119] Papers on Time and Tense*.Oxford 1968.
- [120] Objects of Thought. Eds. P. Geach and A. Kenney. Oxford Uni-
versity Press, 1971.
- and Fine, K. [121] Worlds, Times and SeiJ.'es. Duckworth, 1977.
Putnam, H. [122] Mathematics without Foundations. Journal of Philosophy
64 (1967): 5-22.
- [123] Is Semantics Possible? Metaphilosophy I (1970): 50-63.
- [124] Meaning and Reference. Journal of Philosophy 70 (1973): 187-
201.
Quine, W.Y. [125] From a Logical Point of View, revised. Harvard Press,
1961.
- [126] The Ways ofParadox and other Essays. Columbia, 1966.
Rescher, N. [127] A Theory ofPossibility. Blackwell, 1975.
- and Urquhart, A. [128] Temporal Logic. Springer, 1971.
Schumm, G. [129] Solutions to Four Modal Problems of Sobocinski. Notre
Dame Journal of Formal Logic 13 (1971): 335 -340.
Schutte, K. [130] Vollsttindige Systeme modaler und intuitionistischer
Logik. Berlin/Heidelberg/New Yark 1968.
Segerberg, K. [131] An Essay in Classical Modal Logic. Uppsala 1971.
Sobocinski, B. [132] Certain Extensions of Modal System S4. Notre Dame
Journal of Formal Logic 11,3 (1970): 347-368.
-, A New Class of Modal Systems. Ibid. (I 971): 371-377.
Solovay, R. [133] Provability Interpretations of Modal Logic. Israe/Journal
ofMathematics 25 (1976): 287-304.
Stalnaker, R. [134] Theory of Conditionals. In N. Rescher, Ed., Studies in
Logical Theory. American Philosophical Quarterly Monograph. Black-
well, 1968.
-- [135] Anti-essentialism. In P. French et aI., Eds.,MidwestStudiesinPhil-
osophy IV, Studies in Metaphysics, 1979.
- and Thomason, R.B. [136] A Semantic Analysis of Conditional Logic.
Theoria 36 (1970): 23-42.
Stenlund, S. [137] Logical Theory and Semantic Analysis. Reidel, 1974.
Thomason, R.H., and Stalnaker, R. [138] Modality and Reference. Nous 2
(1968): 359-372.
- [139] Modal Logic and Metaphysics. In K. Lambert, Ed., The Logical
Way of Doing Things. Yale, 1969.
- [140] Semantic Analysis of Tense Logic. Journal of Symbolic Logic
37 (1972): 150-159.
- [I 41]. Ed. Fomlal Philosophy: Selected Papers of Richard Montague.
Yale, 1974.
Thomason, S. (142] Reduction of Tense Logic to Modal Logic I. Journal
of Symbolic Logic 39 (1974): 549-5l.
298 R. Barcan Marcus

Van Bentham, J. [143] Modal Reduction Principles. Journal of Symbolic

Logic 41 (1976): 301-312.
von Wright,G.H. [144] An Essay in Modal Logic. North-Holland, 1951.
- [145] Norm and Action. Routledge and Kegan Paul, 1963.
- [146] An Essay in Deontic Logic and the General Theory of Action.
Acta Philosophica Fennica 21. North-Holland, 1968.
- [147] The Logic of Practical Discourse. In Klibansky [78], 1968.
- [148] On the Logic and Ontology of Norms. In J.W. Davis et a1. [32],
1969.
Wartofsky, M. [149] Boston Studies in Philosophy of Science. Reidel,
1962.
Weaver, G. [ISO] Logical Consequence in Modal Logic. In Leblanc [87].
Wiggins, D. [151] Identity, Designation, Essentialism and Physicalism.
Philosophia 5 (1975): 1-30.
Woolhouse, R.S. [152] Tensed Modalities. Journal of Philosophical Logic
2 (1973): 393-415.
Zeman, J. [153] Modal Logic: The LewisModal Systems. Oxford 1973.
Conditionals and possible worlds*
RISTO HILPINEN
University of Turku

Since its inception in the late 1950s, the possible worlds semantics
of modal logic has been applied with considerable success to
several areas of philosophical logic, for example, to the logic of
knowledge and belief (Hintikka [1-4]); the logic of perception
(Hintikka [5], [4], ch.4,Thomason [6], Niiniluoto [7], Bacon [8],
Smith [9]); deontic logic and the logic of imperatives (Hintikka
[ 10, II], Kanger [12], Chellas [13, 14]); the logic of action
(Kanger [15], Aqvist [16], Porn [17, 18], Lehrer [19]); the
logic of conditionals (Stalnaker [20J, Stalnaker and Thomason
[21], Lewis [22-24], Aqvist [25], Chellas [26]); and the logic
of fiction (Lewis [27]). These applications have illuminated
various problems and puzzles of philosophical logic and analytical
philosophy, but they have also generated new conceptual problems.
Not surprisingly, questions concerning the ontology of possible
worlds have been among the main sources of controversy and
confusion in this area. David Lewis ([23], ch. 4), Alvin Plantinga
[28,29] and Robert Stalnaker [30,31] have tried to clarify the
nature of possible worlds, but their efforts have been criticized
by Brian Ellis [32], Susan Haack ([33] ,[34], ch. 1O),lohnMackie
([35], pp. 88-92) and Lawrence Powers [36] (among others),
who have expressed various misgivings about possible worlds
semantics as a tool of philosophical analysis. The present paper
surveys recent work in the theory of conditionals, where possible
worlds semantics has engendered both logical illumination and
philosophical controversy. Throughout the paper I shall concen-
• Submitted after the introduction was written (Ed. note).

Contemporary philosophy. A new survey. Vol. 1, pp. 299-335.

©1981, Martinus Niihoff Publishers, The Hague/Boston/London.
300 R. Hilpinen

trate on conceptual and philosophical issues, and give relatively

little attention to the purely formal and technical developments
in the semantics of conditionals.

II

The conditional statements of ordinary discourse cannot always -

or even usually - be regarded as material (truth-functional) con-
ditionals. Almost eighty years ago Charles S. Pierce ([37], para.
4.546) observed that if we take the following sentence to be a
material conditional,

(I) There is some married woman who will commit suicide in

case her husband fails in business,

it turns out to be equivalent to

(2) There is a married woman who will commit suicide if every

married man fails in business.

According to Peirce, the equivalence of these two propositions is

the 'absurd result of admitting no reality but existence'. Peirce's
solution to this problem was to regard (l) and (2) as strict (neces-
sary) conditionals; thus he took (1) to mean that

every possible course of events would either be one in which

the husband would not fail or one in which the wife would
commit suicide.

In possible worlds semantics strict conditionals are interpreted in

the way suggested by Peirce: a strict conditional 'If A, then B' is
regarded as true in a given situation or 'possible world' u if and
only if the material conditional 'A ::J B' is true at every world v
which is possible relative to u or accessible from u. Thus a strict
conditional may be expressed by 'O(A ::J B)" where th\ necessity
operator '0' represents (universal) quantification over possible
worlds.
However, this interpretation is not applicable to all conditional
sentences, e.g., to subjunctive and counterfactual conditionals.
 Conditionals and possible worlds 301

Certain inference patterns which are valid for strict conditionals (as
well as material conditionals) fail for subjunctive conditionals.
Perhaps the most obvious example of such an inference is the rule
of strengthening the antecedent:

(3) D(A :J C)
D(A &B:J C)

is a valid rule of inference of modal logic, but (for example)

(4) If this match were struck, it would light

does not entail

(5) If this match were struck and no oxygen were present, it

would light;

thus the following schema should not be valid in the logic of

subjunctive conditionals:

(6) (If A, then C):J (If A &B, then C).

Subjunctive and counterfactual conditionals also fail to satisfy

the principle of transitivity and the principle of contraposition.
For example,

(7) If the leaders of the superpowers were not afraid of a nuclear

holocaust, the world would be destroyed

and

(8) If no country had nuclear weapons, the leaders of the super-

powers would not be afraid of a nuclear holocaust

are acceptable counterfactllals, but

(9) If no country had nuclear weapons, the world would be

destroyed
302 R. Hilpinen

is obviously false. The following sentences constitute a counter-

example to the principle of contraposition:

(10) If Bertie were to see the cow-creamer, he would not steal it.

(11) If Bertie were to steal the cow-creamer, he would not have

seen it.

I f Bertie is an honest man, (10) is true, but (11) is nevertheless

(presumably) false.

III

These logical peculiarities of subjunctive conditionals are explained

by the theory of conditionals formulated by Robert Stalnaker in
[20]. Stalnaker's point of departure is Frank Ramsey's ([38],
p. 347n.) suggestion that

if two people are arguing 'If p will q?' and are both in doubt
as to p, they are adding p hypothetically to their stock of
knowledge and arguing on this basis about q.

This interpretation of conditional statements is not directly

applicable to cases in which the antecedent of the conditional
is known or believed to be false. In such situations the antecedent
cannot be simply added to a person's system of beliefs: in order
to preserve consistency, the belief system must also be revised
by changing or deleting those beliefs which conflict with the
antecedent. Thus Stalnaker describes the evaluation of conditionals
as follows ([20], p. 102):

(12) First, add the antecedent (hypothetically) to your stock of

beliefs; second, make whatever adjustments are necessary to
maintain consistency (without modifying the hypothetical
belief in the antecedent); and finally, consider whether the
consequent is then true.
 Conditionals and possible worlds 303

Belief systems can be regarded as systems of propositions. (In

Brian Ellis's theory of conditionals [39, 40, 32] the expression
'belief system' has a special, technical meaning, which differs from
its present use.) If any belief system containing a proposition A
is termed an A-system, a system obtained from S by adding A
to it and making whatever revisions are necessary for preserving
consistency (but no other revisions) can be said to be an A-system
minimally different from S or the minimal A-revision of S, and de-
noted by 'SA'. (If A belongs to S, SA = S.) If A is inconsistent with
S, that is, if A is a 'belief-contravening' supposition, there are usually
several different ways of making the system consistent with A.
Nicholas Rescher ([ 41], ch. 2) has called this phenomenon 'the
contextual ambiguity of belief-contravening assumptions'. (It has
also been discussed by Nelson Goodman in his classic paper [42] .)
To ensure the uniqueness of the system SA, it is necessary to
assume that some way of revising S is preferred to all others, or,
if there are several equally acceptable ways of revising the system
S, SA should include only beliefs which survive all acceptable ways
of making the adjustments required by A. (Cf. Gardenfors [43],
p. 403, n. 9; for a discussion of a similar 'rule of ties' in theories
or rational belief, see Levi [44], pp. 84-85, and Hilpinen [45],
pp. 158, 167-168.) The preferability of various revisions of S
depends on pragmatic considerations which can by by-passed in
the logic of conditionals. (Nicholas Rescher has discussed in [46]
and [47] various pragmatic principles which might be used to
detennine the preference ordering among belief systems.)
The evaluation procedure described by (12) can now be formu-
lated as the following condition concerning belief systems:

( 13) A conditional 'If A, then B' belongs to a belief system S if

and only if B belongs to the minimal A -revision of S.

The truth-conditions of conditional statements should explain

and justify the procedure employed in judging their acceptability.
According to Stalnaker, the concept of possible world is 'the
ontological analogue of a stock of hypothetical beliefs' ([20],
p. 102), and he proposes to explain the evaluation procedure
described by (12) in tenns of the following condition:
304 R. Hilpinen

(14) Consider a possible world in which A is true, and which

otherwise differs minimally from the actual world. 'If A,
then B' is true (false) just in case B is true (false) in that
possible world.

Thus Stalnaker assumes that the concept of minimal revision of a

belief system can be explicated in terms of the concept of minimal
difference between possible worlds.

IV

In addition to the standard apparatus of possible worlds semantics

(a" set U of possible worlds, an accessibility relation R defined on
U, and an interpretation function I I which assigns to every state-
ment A a set of possible worlds, IA I, called the proposition ex-
pressed by A), the formal apparatus of Stalnaker's theory contains
a selection function s which assigns to each possible world u and
statement A a world seA, u) at which A is true (briet1y, an A-world)
and which is otherwise as similar to u as the truth of A permits
(s(A, u) is the A-world minimally different from u). Such a world
may also be called the closest A-neighbourofu. Stalnaker expresses
the conditional connective by'>' and defines the truth of a con-
ditional statement as follows:

(15) 'A > B' is true at u if and only if B is true at s(A, u).
(I follow the customary practice of speaking of the truth of a
statement at a world, even though the expression 'true of a world'
would be logically more correct.) If A is an impossible statement,
that is, false at every world accessible from u, S(A, u) cannot be
a possible world. (Here the concepts of possibility and truth are
always thought of as being relative to some world u even when
this is not explicitly mentioned.) In this case Stalnaker takes
s(A, u) to be the absur-l world A at which all statements (including
contradictions) are tru\;. A is not possible with respect to any
other world, and no other world is possible with respect to A- The
sole purpose of A is to permit the evaluation of conditionals with
impossible antecedents: according to Stalnaker's theory, all such
conditionals are true.
 Conditionals and possible worlds 305

The function s is subject to certain formal constraints which

are necessary for the interpretation of seA, u) as the closest A-
neighbour of u:

(16) (i) A is true at seA, u) (for any statement A and world u).
(ii) s(A, u) = A if and only if A is true at no world accessible
from u.
(iii) If A is true at u, seA. u)::;: u.
(iv) If B is true at seA, u) and A is true at s(B, u), then
s(A, u) = s(B, u).

According to conditions (iii) and (iv), the function s(u,-) deter-

mines a total ordering of all selected worlds with respect to each
'reference world' u (in which the reference world precedes all
others in the order).
These conditions and Stalnaker's truth-definition for 'A > B'
show how the principle of transitivity and the principle of contra-
position can fail for conditionals. If B holds (is true) at the closest
A-neighbour of u but A does not hold at its closest B-neighbour
(that is, if the closest B-neighbour of u is closer to it than its
closest A-neighbour), and C holds at the closest B-neighbour of
u but not at its closest A-neighbour, both 'A > B' and 'B > C' are
true at u but 'A > C' is false. And if the closest A -neighbour of
u is closer to it than its closest ""lB-neighbour, 'A > B' is true at
u, but if A is also true at the closest ""lB-neighbour of u, '1B > -lA'
is false; consequently the principle of contraposition fails to hold
for Stalnaker conditionals. (However, Stalnaker conditionals
satisfy e.g. the rules of modus ponens and modus (ollens, which
should hold for any reasonable concept of conditional.)

Stalnaker's remark that 'a possible world is the ontological analogue

of a stock of hypothetical beliefs' is inaccurate in two respects.
First, it is misleading to speak here of an 'ontological' analogue
since possible worlds can be regarded as convenient theoretical
fictions without any ontological import. (ef. van Fraassen [48],
pp. 266-270. As was indicated at the beginning of this paper, the
306 R. Hilpinen

ontology of possible worlds is a controversial topic, even though

their semantical role is reasonably well understood.) Since possible
worlds correspond to (and can indeed be regarded as) complete
sets of propositions, a stock of hypothetical beliefs is not analogous
to a single possible world, but to a set of possible worlds, viz. the
set of all worlds which satisfy or are compatible with the beliefs in
question. Thus Stalnaker's condition (13) can be formulated in
terms of possible worlds as follows:

( 17) 'If A, then B' belongs to a belief system S if and only if

B is true at every possible world compatible with the
minimal A -revision of S.

According to Stalnaker, each world consistent with the minimal

A -revision of a person's belief system S can be thought of as being
the closest A-neighbour of some world compatible with S; thus
condition (17) can be formulated in terms of the concept of
minimal difference between possible worlds as follows:

(18) 'If A, then B' belongs to S if and only if B is true at every

world which is the closest A -neighbour of some world
compatible with S.

(In (18), the universal quantifier is supposed to have wide scope:

the right-hand side of (18) means that for every world w, if there
exists a world v such that v is compatible with Sand w is the
closest A -neighbour of v, then B is true at w.)
Often neither 'If A, then B' nor 'If A, then ,B' belongs to a
person's belief system; in such a case the person can Uustifiably)
say that if A were the case, then B might be the case and ,B might
also be the case. According to Stalnaker's semantics, this 'might'
is always an expression of ignorance and due to the incompleteness
of a person's belief system, Either B or ,B is true at the closest A-
neighbour of any world u, and consequently 'A > B' or 'A > ,B'
holds at any world. (If a belief system is completed, its minimal
A -revision will either be a B-system of a ,B-system.) Thus the
principle of conditional excluded middle,

(19) (A > B) V (A > ,B),

 Conditionals and possible worlds 307

is valid in Stalnaker's semantics. However, it is clear that this

principle - or the assumption that every world has a unique
closest A -neighbour - is not required by (17) or (18). From the
standpoint of the acceptance conditions of conditionals, principle
(18) is superfluous: (17) and (18) are also consistent with the
assumption that a world may have several closest A -neighbours
(or that several A-worlds may be minimally different from a given
world u). This assumption can be represented in possible worlds
semantics by letting s(A, u) be a set of possible worlds, not a
single world. A semantical theory of this type has been developed
by David Lewis [22-24]. The selection functions of Lewis's
semantics are set selection functions, not world selection functions.
Lewis expresses the conditional connective by '~' and defines
the truth of a conditional in terms of a selection function in the
following way (to distinguish Lewis's function from that used by
Stalnaker, I denote it by 'r):

(20) 'A ~ B' is true at u if and only if B is true at every world

inf(A, u).

If the proposition expressed by B (the set of B-worlds) is denoted

by 'IBI', the right-hand side of (20) can be expressed in the form
'f(A, u) ~ IBI'. Like Stalnaker's s-function, the f-function is sub-
ject to certain formal constraints which are necessary for the
interpretation of f(A, u) as the set of the closest (preferred) A-
neighbours of u, and which generate a reasonable logic of counter-
factual conditionals (Lewis [23] , p. 58):

(21) (i) If A is true at u,f(A, u) = {u}.

(ii) ftA, u) is included in IA I.
(iii) If IA I is included in IB I and ftA, u) is nonempty,ftB, u)
is also nonempty.
(iv) If IA I is included in lEi and IA I overlaps (has a non-
empty intersection with) feB, u), then f(A, u) is the
intersection of IA I and ftB, u).

The 'absurd world' A is not needed in Lewis's semantics: if A is

impossible, f(A, u) is the empty set. Lewis ([23], p. 58) notes that
analysis (20) has been suggested by John Vickers, and further
308 R. Hilpinen

studied by Peter Woodruff in Notes on Conditional Logic (Univer-

sity of California, Irvine, 1969). A similar analysis is given in
Brian Chellas [26] .
In Lewis's semantics the principle of conditional excluded
middle is not valid: the falsity of 'A D+ B' means that B is false
at some selected A -world, but not necessarily at all of them. This
may be expressed by a might-conditional: a might-conditional
'If A were the case, B might be the case', symbolized by Lewis
'A ~ B', is defined as follows:

(22) 'A ~ B' is true at u if and only if B is true at some world

in teA, u).

Unlike Stalnaker's definition (IS), Lewis's analysis of conditionals

makes it possible to distinguish the truth-conditions of might-
conditionals from those of would-conditionals. According to
(21)-(22),

(23) I(A D-+ B) == (A ~ IB),

whereas Stalnaker's analysis entails

(24) I(A > B) => (A > IB).

(24) can be strengthened into an equivalence if A is possible, but
if A is impossible, both 'A > B' and 'A > IB' are true. Above
the principle of conditional excluded middle was criticized on
the grounds that it is not necessary for the explanation of the
acceptance conditions of conditionals. The assumption that for
any two sentences A and B, either 'If A were the case, B would be
the case' or 'If A were the case, B would not be the case' must
be true, seems therefore gratuitous, and might even be regarded
as an unfounded metaphysical postulate. This assumption also
seems intuitively implausible in situations in which there are several
equally plausible (and equally possible) ways of satisfying the
antecedent of a given conditional, only some of which satisfy
the consequent. For example, if I miss a bus which I tried to
catch, I need not believe either
 Conditionals and possible worlds 309

(25) If I had run, I would have caught the bus

or

(26) If (i.e., even if) I had run, I would not have caught the bus,

and this is not just a matter of ignorance: neither (25) nor (26)
need be true in the situation since my catching the bus would
have been dependent on how fast I would have run, and this is
left undetermined by the antecedent. In such a situation it would
be true to say only:

(27) If I had run, I might have caught the bus.

This consequence of Lewis's semantics is in agreement with the

fact that sometimes the only reasonable reply to the question
'Would B be the case if A were the case?' is That depends' - and
if the antecedent A is an assumption about purely hypothetical
circumstances, this reply is not merely an expression of ignorance.
(ef. Quine's ([49], p. 15) famous Bizet-Verdi example: If Bizet
and Verdi had been compatriots, would Bizet have been Italian?
or Verdi French?) Nevertheless many philo~ophers have regarded
(19) as an obvi'}us truth; for example, in 150] Richmond Thomason
says that 'this principle seems t() be so clearly valid a feature of
conditional reasoning that it is surprising that some logicians
have chosen to deny it' (p. 408). Similar views have been expressed
by Ramsey ([38], p. 247n.) and Goodman [42].
Bas van Fraassen [5 I] has discussed the relationship between
Lewis's and Stalnaker's theories and shown that from the stand-
point of Lewis's theory, Stalnaker's interpretation of conditionals
is analogous to a hidden variable theory of quantum mechanics:
Stalnaker-type selection functions and the truth-value assignments
based upon them can be regarded as hidden variables added to
Lewis's theory, and on the other hand, Lewis-type interpretations
of conditionals can he defined as families of Stalnaker interpre-
tations by means of van Fraassen's technique of supervaluations.
(For hidden variable theories, see van Fraassen [51,52] ; Greechie
and Gudder [53], pp. 159-162; for the concept of super-
valuation, see van Fraassen [54], eh. V, sections 3 and 4.)
310 R. Hilpinen

VI

In [22-24] Lewis discusses alternative ways of formulating the

semantics of conditionals. Instead of defining the truth-conditions
of conditionals in terms of a selection function which is supposed
to reflect the overall similarity among possible worlds, they can
be specified directly in terms of the concept of similarity as
follows ([24], pp. 424-425):

(28) 'A D-+ B' is true at u if and only if

(i) A is true at no world accessible from u, or
(ii) some A-world at which B is true is closer (more similar)
to u than any A -world at which B is false.

In case (i), the conditional is said to be vacuously true. The con-

cept of similarity used in (28) can be represented in possible
worlds semantics in various ways, e.g., in terms of a three-place
relation on the set of possible worlds (v is more similar to u than
w is), or by assigning to each world u a system of spheres of
similarity around u (a set of nested subsets of U) which represent
different degrees of similarity to u (Lewis [23], chs. 1-2).
If there are, for each possible world u and statement A ,A-worlds
closest (most similar) to u, definition (20) is equivalent to (28) if
we let f(A, u) be the set of all such worlds. However, definition
(28) does not require the existence of such a subset for every A
and u: according to (28), a conditional 'A D-+ B' may be non-
vacuously true at u even if u has no closest A -neighbours, but the
A -worlds form an infinite sequence of closer and closer worlds.
Lewis calls the assumption that any world u has a set of closest
A -neighbours the 'Limit Assumption', and he regards it as a virtue
of definition (28) that it does not require the truth of the limit
assumption. His rejection of the limit assumption is motivated by
the fact that certain plausible ways of determining the comparative
similarity of worlds do not satisfy it. In [23], p. 20, he discusses
the following example: Let a be a line just under an inch long, and
let A be the counterfactual supposition that a is more than an
inch long. If we assume that the similarity of various A -worlds
to the actual world depends solely upon the length of a, there is
clearly no A-world (or set of A-worlds) closest of the actual
 Conditionals and possible worlds 311

world: for any E (E > 0), if v is a world in which a is I + E inches

long, there are A -worlds in which a is shorter than 1 + E inches
(say, 1 + EI2 inches). However, as John Pollock [55, 56] has
pointed out, the rejection of the limit assumption leads to para-
doxical results. For example, in the case described above, Lewis's
analysis (28) implies that every conditional of the form

(29) If a were more than an inch long, it might be + E inches

long

is false (for any E, however small), and consequently

(30) If a were more than an inch long, it would not be 1 +E

inches long

is true for any E (E > 0). Since

(31) If a were more than an inch long, it would be more than an
inch long

is obviously true, the counterfactual consequents of supposition

A (Le., the set of all sentences B such that 'A G+ B' is true) form
an unsatisfiable set in spite of the fact that A itself is perfectly
consistent (Hertzberger [57]). Such counterfactual inconsistencies
cannot arise if the limit assumption holds, and the limit assumption
might therefore be justified by the following condition of countcr-
factual consistency (Hertzberger [57] , p. 84):

(32) If A is possible, it has a satisfiable set of counterfactual

consequents.

Lewis discusses in [58] (p. 6) a deontic analogue of counterfactual

inconsistency and its relationship to the limit assumption, but
does not regard a consistency condition analogous to (32) as a
general condition of adequacy for deontic logic.
Some philosophers have argued on the basis of the counter-
factual inconsistencies generated by Lewis's theory that the
evaluation of counterfactuals cannot be based on the ordinary,
intuitive concept of similarity or resemblance. According to
312 R. Hilpinen

John Pollock, the evaluation of counterfactuals is based on the

notion of 'minimal change made to the actual world to accommo-
date the counterfactual supposition', but the minimality of change
need not be correlated with the comparative similarity of worlds
(in the ordinary sense of 'similarity') :'Two different minimal
changes can result in worlds that differ in their similarity to the
real world' (Pollock [55], p. 473).
The limit assumption can also be defended on pragmatic grounds.
Hans Kamp ([ 59], p. 71) has characterized the pragmatics of
conditionals in the following way:

Conditionals have the primary function to inform the hearer

that in any of the plausible worlds in which the antecedent
holds the consequent holds as well. They offer, as it were,
for consideration a class of worlds - the plausible worlds in
which the antecedent holds - and assert that each of these
satisfies the consequent.

This characterization clearly presupposes the limit assumption,

and is consistent with definition (20) - 'the plausible worlds'
mentioned by Kamp are those selected by the I-function. On the
other hand, on Lewis's analysis (28) conditional statements need
not refer to any particular set of 'plausible worlds'; thus Kamp's
pragmatic characterization of conditionals is compatible with (28)
only in the &pecial case in which the limit assumption holds.

VII

Stalnaker's and Lewis's theories of conditionals have found

interesting applications in other areas of philosophical logic. For
example, the application of Lewis-type semantics to deontic
logic has clarified various problems associated with the concept
of conditional obligation (derived obligation, commitment)
(Hansson [60], Lewis [58]. Bengt Hansson developed his theory
of conditional obligation independently of Lewis's semantics of
counterfactuals.) In these applications the concept of conditional
obligation is interpreted in tenns of a relation of deontic pre-
ferability or betterness (among possible worlds): a state of affairs
 Conditionals and possible worlds 313

A is regarded as obligatory under circumstances B if and only if

A is true at the 'best' B-worlds. (This approach to deontic logic
has been anticipated by Lennart Aqvist [61].) Brian Chellas [62]
and Peter Mott [63] have presented especially perspicuous analyses
of conditional obligation in which the ideas of obligation and
conditionality are separated from each other, and conditional
obligation is defined in terms of the standard (monadic) obligation
operator and a Lewis-type conditional connective. David Lewis
[64] has applied his theory of conditionals to the analysis of
causation and causal dependence and to the semantics of fictional
discourse [27], and in [65] Risto Hilpinen has adapted Lewis's
semantics of counterfactuals to the analysis of Karl Popper's
concept of verisimilitude (truthlikeness). Recent research in
quantum logic has shown that conditional propositions can be
formulated in quantum logic by means of a connective which
is essentially similar to the Stalnaker conditional: the quantum-
logical conditional can be defined in terms of a selection function
satisfying conditions analogous to (l6i-iv) (van Fraassen [52],
Ha rdegree [66]).

VIII

The possible worlds semantics of conditionals has been criticized

in three ways. First, some philosophers have argued that the
concept of possible world is philosophically and onto logically
unintelligible and that Lewis's and Stalnaker's theories rest on
a shaky philosophical foundation (Ellis [32], Haack [33] , Mackie
[35]). This criticism does not really touch possible-worlds seman-
tics as a semantical theory (as was suggested above, possible worlds
can be interpreted in various ontologically innocuous ways), but
is directed against David Lewis's realistic interpretation of possible
worlds. Lewis's intuitive characterization of possible worlds is not
entirely unambiguous: he describes possible worlds as 'ways things
could have been' ([23], p. 84), but says also that '1 and my
surroundings' constitute the actual world, and other possible
worlds are entities of the same type (p. 86). Lewis's critics have
claimed that these characterizations are mutually incompatible
(Stalnaker [31], pp. 66-68; Nute [671): the latter characterization
314 R. Hilpinen

seems to involve an 'ultra-realistic' conception of possible worlds

(worlds as concrete particulars), while the former suggests only
'moderate' realism (worlds as complex states of affairs or states of
the universe; cf. Stalnaker [31], p. 70; Plantinga [29]).
The second type of criticism concerns the concept of similarity
in Lewis's semantics (Bennett [68], Fine [69], Richards [70],
Tichy [71, 72], Jackson [73], Slote [74], Bowie [75]). (This
topic was already discussed at some length in section VI in con-
nection with the limit assumption.) For example, it has been
suggested that an explanation of the truth-conditions in terms
of similarity is bound to be circular, since the similarity of worlds
depends on agreement in propositions, including conditional
(counterfactual) propositions (Fine [69], p. 455). If the com-
parative similarity of possible worlds is judged solely in terms of
particular (,atomic') facts, the evaluation of conditionals on the
basis of similarity is apt to yield completely unacceptable results
(Bennett [68], p. 395; Fine [69], p. 452; Tichy [71,72]; Bowie
[75], p. 484-485). As was already pointed out in section VI, the
concept of similarity underlying subjunctive conditionals cannot
be based on 'uneducated', intuitive judgments of similarity, but
is a special concept, dependent on the type of conditional in
question. In [76] (pp. 466-467) Lewis remarks:

The thing to do is not to start by deciding, once and for all,

what we think about the similarity of worlds, so that we can
afterwards use these decisions to test [analysis (28)] ...
Rather we must use what we know about the truth and
falsity of counterfactuals to see if we can find some sort of
similarity relation - not necessarily the first that springs to
mind - that combines with [analysis (28)] to yield the
proper truth conditions.

Similarity is a vague concept, and the vagueness can be resolved

in different ways by using different criteria of similarity. Analy-
sis (28) daives its explanatory power from the formal character-
ization of the concept of similarity and from the specification of
the criteria of similarity used in the evaluation of various types
of conditionals. The logic of conditionals depends on the formal
properties of the similarity relation (or on the formal properties
 Conditionals and possible worlds 315

of the selection function, if the truth-conditions of conditionals

are formulated in terms of a selection function). In [76] David
Lewis has discussed in detail the criteria of similarity associated
with what he calls the 'standard resolution' of the vagueness of
counterfactuals (p. 457). For example, the truth-conditions of
causal conditionals are based on the 'standard resolution'.
Many of Lewis's critics have defended a consequence theory
of counterfactual conditionals (also called the cotenability theory
or metalinguistic theory) (cf. e.g. Bennett [68]).
The following formulation of the consequence theory is due to
Nelson Goodman [42] and Barry Loewer [77] :

(33) 'If A, then B is a true counterfactual if and only if (i)

there is a finite set of true statements D such that A is
cotenable with D, and A and D together with some laws
imply B, and (ii) there is no set of true statements H such
that A is cotenable with H and A and H together with
some laws imply ,B.

The cotenability of A and D means that A and (all members of)

D can be true together. Logical consistency does not imply coten-
ability, and this creates a problem for the consequence theorist
since the cotenability of A and D requires that no counterfactual
'If A were the case, then D would not be true', where D belongs
to D, is true. (Nelson Goodman [42] concluded from this that
'we can never explain a counterfactual except in terms of others,
so that the problem of counterfactuals must remain unsolved'
(p. 16).) In [23] Lewis has analysed the concept of cotenability
by means of possible worlds semantics (pp. 57,69-71), and argued
that the consequence theory is not incompatible with analysis
(28), but a special case of the possible worlds account. This view
has been supported by Loewer [77], who has also replied to
Jonathan Bennett's [68] objections to Lewis's analysis to coten-
ability. According to Loewer, Goodman's truth-conditions for
counterfactuals are compatible with Lewis's analysis, and a logic
of conditionals based on a Goodman-type analysis is a subsystem
of Lewis's theory.
One obvious difference between Lewis's analysis and the cus-
tomary formulations of the consequence theory is that according
316 R. Hilpinen

to the latter, all counterfactuals presuppose general laws, but

Lewis's theory does not contain such a requirement. However,
this feature of consequence theories may be a weakness: some
philosophers have argued that causal dependence among individual
events (which is a special case of counterfactual dependence) does
not presuppose general (causal) laws (Anscombe [78], Dretske
and Snyder [79 D. Barry Loewer [77] (pp. 106-107) has suggested
that general laws are epistemically rather than ontologically (or
semantically) necessary for counterfactuals: counterfactuals can
be justified or refuted in terms of known general laws, but the
truth of a counterfactual statement need not depend on under-
lying general laws.
Lewis's semantics has also been criticized on logical grounds: it
has been argued that Lewis's theory legitimizes inference patterns
which are in fact not valid, but fails to justify certain logically
cogent forms of inference. Most of this criticism has concerned
two rules, viz.

(34) A,B
A D-?B

and

(35) (A V B) Q-l- C.
AQ-l-C

(34) is valid in Lewis's semantics, but (35) is not. Jonathan Bennett

[68] and John Bigelow [80] have argued that (34) should not be
valid. Their rejection of (34) is based in the view that the truth of
a subjunctive conditional requires some connection between the
antecedent and the consequent. This view seems mistaken, how-
ever: subjunctive conditionals (especially 'even if' -conditionals;
cf. examples (10) ar.d (26) above) can also be used to express a
lack of connection between the antecedent and the consequent;
thus John Pollock ([81, [56], ch. II) has argued that (34) holds
for 'simple subjunctives' and fails only for 'necessitation con-
ditionals'. (Donald Nute [82] has also suggested that there are
different types of counterfactual conditionals, subject to different
logical principles.) If A and B are known to be true, the utterance
 Conditionals and possible worlds 317

of 'If A were the case, B would be the case' would be odd and
unidiomatic, but it should not therefore be regarded as false:
Lewis notes that 'oddity is not falsity; not everything true is a
good thing to say' ([23], p. 28; cf. also Pollock [56], p. 39). In
Lewis's semantics the validity of (34) follows from the assumption
that no other world is as similar to u as u itself is. If the semantics
of conditionals is formulated in terms of selection functions, this
assumption is equivalent to (21.i). If (21.i) is weakened to

(21.i*) If A is true at u, u belongs tof(A,u)

(Le., no world is more similar to u than u itself is), (34) is no

longer valid (Lewis [23], p. 29).
The controversy concerning rule (35) is based on the fact that
some seemingly valid inferences involving subjunctive conditionals
appear to rely on it. For example,

(36) If a door or a window were open, we would get some fresh

air

seems to entail both

(37) If a door were open, we would get some fresh air

and

(38) If a window were open, we would get some fresh air,

and this suggests that

(39) (If A VB, then C) ::> (If A, then C)

(the principle of elimination of disjunctive antecedents or simpli-

fication of dinjunctive antecedents) should be a valid principle of
the logic of (subjunctive) conditionals. However, if we assume that
the logic of conditionals is a classical intensional logic (cf. Lemmon
and Scott [83], p. 18) which satisfies the principle of in tensionality ,
318 R. Hilpinen

(40) If A is logically equivalent to Band C is logically equivalent

to D, 'If A, then C' is logically equivalent to 'If B, then D',

(39) entails the objectionable principle (6) (the principle of

strengthening the antecedent): according to (40), 'If A, then
C' is equivalent to 'If (A & B) V (A & IB), then C', and by (39)
this implies 'If (A & B), then C'.
Stalnaker's and Lewis's theories of conditionals are classical
logics, and neither (6) nor (39) is valid in them. For such theories
conditionals with disjunctive antecedents present a problem: they
cannot explain in any straightforward way why (37) and (38)
seem to follow from (36). Brian Ellis, Frank Jackson and Robert
Pargetter [84] have presented the problem of disjunctive ante-
cedents as an 'objection to the possible worlds semantics for
counterfactual logics', but in fact the problem is not specific to
possible worlds semantics, but concerns any theory of conditionals
based on classical logic (i.e., any theory in which logically equi-
valent expressions are interchangeable). It arises also in Brian
Ellis's 'epistemic' theory of conditionals ([32], pp. 68-74). The
problem of disjunctive antecedents has been discussed in several
recent publications, e.g., in Fine [69], Hill and Creary [85],
Ellis, Jackson and Pargetter [84], Lewis [86], Nute [87, 67,
88], Loewer [89], McKay and van Inwagen (90], and in Humber-
stone [91].
Logicians have tried to solve the problem of disjunctive ante-
cedents in two ways. First, it has been suggested that the word
'or' occurring in examples similar to (36) does not correspond to
the logician's 'V', but functions as a wide-scope conjunction.
(Lewis [86], p. 360; Ellis [32], pp. 72-74). However, this is an
unsatisfactory ad hoc solution. It is evident that (36) is equivalent
to a wide-scope conjunction, but this is precisely the problem:
normally the word 'or' functions as a disjunction, not as a con-
junction, and it is not clear how 'or' can function as a wide-scope
conjunction in (36) if the elimination principle (39) is not valid.
Donald Nute [87,67,88] has attempted another way out of this
difficulty: he has argued that we should accept the elimination
principle and reject the principle of intensionality; this would
block the inference of (6) from (39) and enable us to avoid other
untoward consequences of (39) as well. Nute's approach has been
 Conditionals and possible worlds 319

criticized by Loewer [89] and by McKay and van [nwagen [90J.

Perhaps the most effective objection to Nute's solution is McKay
and van Inwagen's observation that there are among the counter-
factuals of ordinary discourse clear counter-examples to the
elimination principle. For example, it seems possible and perhaps
even reasonable to regard the following counterfactual as true:

(41) If Spain had fought in World War II on the Axis side or on

the Allied side, she would have fought on the Axis side.

According to the elimination principle this would imply

(42) If Spain had fought in World War II on the Allied side,

she would have fought on the Axis side,

which is absurd. This suggests that counterfactuals with disjunctive

antecedents are semantically ambiguous: sometimes the disjunction
can be eliminated in accordance with principle (39), but in other
cases this is not possible. (In [88] N ute reports a similar example
given by Michael Dunn, and concludes from Dunn's example
that in some cases the SDA-rule (simplification of disjunctive
antecedents) is inappropriate, and that 'there is no unique logic
of counterfactual conditionals' (p. 324).)

IX

At this juncture it is interesting to observe that the problem of

disjunctive antecedents is not an isolated puzzle: an essentially
similar problem arises in deontic logic. According to the 'stan-
dard system' of deontic logic (cf. F¢>llesdal and Hilpinen [92],
pp. 13-19), a proposition or a state of affairs A is permissible for a
person X if and only if it is compatible with his obligations, that
is, if A is true at some possible world at which X fulfills all his
obligations. Such worlds may be termed deontically perfect worlds
(as far as X's obligations are concerned). Thus the permissibility of
A means, according to the standard system, that A is true at some
deontically perfect world. (In the present context 'possible worlds'
should not be interpreted literally as worlds or even as states of
320 R. Hilpinen

the world, but rather as outcomes of some action situation or as

possible action sequences of the agent X; cf. Hilpinen [93).) If
A is true at some deontically perfect world, so is A VB, but not
conversely; thus

(43) PA ~ peA V B),

where 'P' is the permissibility operator, is a logical truth of the

standard system, but its converse is not. This feature of the
standard system conflicts with our intuitions concerning the
logical strength of disjunctive permissions. Often a disjunctive
permission (or what appears to be a disjunctive permission) seems
logically stronger than a permission involving just one of the
disjuncts; for example,

(44) You may buy oranges or bananas

seems to permit more than

(45) You may buy oranges,

not the other way round. This suggests that instead of (43) we
should regard its converse

(46) peA V B) ~ PA

as a logical truth of deontic logic. This conflict between the

standard system and our logical intuitions concerning disjunctive
permissions has been called the problem of free choice permission
(if a disjunctive permission implies the permissibility of both
disjuncts, the agent is free to choose either of them; cf. von Wright
[94), p. 22); it may also be termed the problem of disjunctive
pennission. Georg Henrik von Wright [94, 95] has tried to solve
the problem of disjunctive permIssion by developing a logic of
permissibility in which (46) is a logical truth: he calls the concept
of permission satisfying (46) a strong permission and distinguishes
it from the 'weak' permission concept of the standard system.
However, this attempt to solve the problem of disjunctive per-
mission creates new problems which are analogous to those arising
 Conditionals and possible worlds 321

from the acceptance of (39) as a logical truth of the logic of con-

ditionals: (46) and the deontic version of the principle of inten-
sionality.

(47) If A IS logically equivalent to B, PA is logically equivalent

toPB,

entail

(48) PA => peA & B),

which is strongly counterintuitive for any reasonable concept of

permissibility. For example, it is obvious that a permission to buy
an orange does not imply a permission to buy an orange and buy
a banana. The analogy between the problem of disjunctive ante-
cedents and the problem of disjunctive permission has been pointed
out by Kit Fine [69] and by Barry Loewer [89] (among others).
Hans Kamp [59] has observed that if a permission sentence
'PA' is used performatively, that is, for the purpose of giving a
permission to someone and not merely for reporting that some-
thing is permissible for an agent, its function is to lift a prohibition
against A and make permissible certain states of affairs (satisfying
A) which were prohibited before. A conjunction of two permission
sentences, 'PA & PB' lifts the prohibition against A and the pro-
hibition against B, and thus makes permissible the union of the
states of affairs which would have been made separately permissible
by the conjucts. A pern1ission of this type can be naturally expressed
by a disjunctive permission sentence without departing from the
rule that the connective 'or' always stands for set theoretic union
(Kamp [59], p. 65). However, as Kamp points out, disjunctive
permissions do not always function in this way: in some cases they
lift the prohibition against the disjunction as a whole without
making both disjuncts permissible. Kamp ([59], p. 67) gives the
following example of such a situation:

(49) Consider a king telling his vassal: 'You may pillage city X
or city Y. But first take counsel with my secretary.' The
second part of this statement makes !t clear that the vassal
should not infer from the first part that he may make his
own choice of city.
322 R. Hilpinen

Thus disjunctive permissions exhibit the same ambiguity as the

antecedents of conditionals. In many cases the disjunction can be
eliminated (that is, the word 'or' seems to function as a wide-scope
conjunction), but there are also cases in which this is not possible.
Kamp ([ 59] , pp. 69-70) explains this ambiguity as follows:

Permission sentences contain occurrences of a certain one-

place [sentential] operator which are often concealed in the
surface structure; I represent this operator by the letter 'F'.
I call F the 'focus-operator' as it is its function to 'move'
the sub-formula which stands within its scope into focus and
thereby subject it (as a whole) to the particular semantic-
pragmatic operation which is associated with the mode, or
pragmatic function, of the speech act in question.

This explanation suggests that the difference between (44) and the
permission sentence occurring in (49) can be made explicit by
expressing (44) in the form

(50) P'(FA V FB)

and the latter as

(51) P'F(C V D).

It should be observed that the P'-operator in (50) and (51) is not

the permission operator of the standard system of deontic logic.
(50) is a free choice permission: it entails

(52) P'FA,

but 'P'FC' does not follow from (51). A permIssIon sentence

'P' FA' does not render permissible all states of affairs (possible
worlds) satisfying A, but only some of them: a permission to do
A does not mean that all possible ways of doing A are permissible.
Kamp's focus-operator F can be regarded as a selection operator
which selects the class of possible worlds made permissible by a
permission sentence. If the set of possible worlds satisfying A is
again denoted by 'IAI', we can say that (50) makes permissible
 Conditionals and possible worlds 323

the union of the classes of possible worlds selected by F from

IA I and IB I, respectively, whereas (5 1) renders permissible
the class of possible worlds selected from IC V D I. In general,
'F(A V BY does not determine the same class of worlds as
'FA V FB'. (In some cases it is unclear which possible worlds are
made permissible by a given permission sentence (cf. Lewis [96]).
This indeterminacy is analogous to the contextual ambiguity of
belief-contravening suppositions discussed by Rescher [41]. Here
I assume that the indeterminacy is resolved by the selection
function is some way or other.) The problem of disjunctive per-
mission is also discussed by Hans Kamp in [97] .

x
If the truth-conditions of conditional statements are expressed
by means of a selection function, it is easy to see how Kamp's
explanation of the ambiguity of disjunctive permissions can be
adapted to the antecedents of conditionals. If (36) is equivalent
to the conjunction of (37) and (38), its truth-conditions can
(according to definition (20)) be given in the form

(53) [(A, u) ~ ICI andf(B, u) ~ ICI

(with appropriate interpretations of A, B and C). (52) is equivalent

to

(54) f(A, u) U feB, u) ~ ICI,

butnotequ~a~ntto

(55) f(A V B, u) ~ ICI.

According to (21.i-iv), (54) implies (55), but not conversely. In

ordinary discourse both (54) and (55) can be expressed by a
conditional with a disjunctive antecedent, and in both cases the
world 'or' has its ordinary meaning; it stands for set theoretic
union. In a formal language in which conditionals are formulated
by means of a single two-place connective, (54) can be expressed
324 R. Hilpinen

only in the conjunctive form corresponding to (53). However, if

the language of conditionals contains a special selection operator
(or a 'focus operator'), both (54) and (55) can be expressed by
conditionals with disjunctive antecedents, and distinguished from
each other in the way suggested by Kamp, i.e., by the scope of
the disjunction.
In natural languages conditionals are usually expressed by
means of the words 'if' and 'then'. Traditionally they have been
treated by logicians as a single connective, but the preceding dis-
cussion suggests that their semantical roles might be separated.
Following Kamp, we might say that the function of the word 'if'
in a conditional statement is to offer or select for consideration
a certain class of worlds, viz. the plausible worlds at which the
antecedent holds, and the statement following the word 'then'
is an assertion about the worlds introduced by the if-clause.
According to this interpretation of conditional statements, the
word 'if' is a sign of a selection operator or focus-operator, and
conditionals with disjunctive antecedents can thus be given two
alternative readings, viz.

(56) If (A or B were the case), then C would be the case

and

(57) (If A were the case) or (if B were the case), then C would
be the case,

which are not equivalent: the elimination principle is applicable

to (57), but not to (56). The following observation concerning the
use of conditionals in ordinary discourse supports this analysis:
If a conditional with a disjunctive antecedent, for example,
sentence (36),

If a door or window were open, we would get some fresh air,

is understood in such a way that the disjunction can be eliminated

from the antecedent, it can be paraphrased as

(58) If a door or if a window were open, we would get some

fresh air,
 Conditionals and possible worlds 325

but a sentence to which the elimination principle is not applicable

cannot be paraphrased in the same way without changing its
truth-value. For example, even though McKay and van Inwagen's
sentence (41) might be considered acceptable,

(59) If Spain had fought in World War II on the Axis side or if

Spain had fought on the Allied side, she would have fought
on the Axis side

is definitely false. (This 'or if'-test of the applicability of the

elimination principle has been suggested in correspondence by
Peter van Iwagen.)

XI

According to (57) and (58), an expression of the form 'If A were

the case' is a meaningful part of a conditional sentence. Such an
expression cannot be used for the purpose of stating or asserting
anything, but it is capable of functioning as an antecedent (or as
a meaningful part of the antecedent) of a conditional statement.
(It is immaterial here whether the word 'if' is regarded as a part
of the antecedent or as a sign indicating that the clause following
it is being used as antecedent (or as a part of the antecedent) of
a conditional statement.) Expression of this type may be termed
assumptions or suppositions, and expressions which can be used
for stating something can be called simply statements. To see how
the truth-conditions expressed by (55) and (54) can be assigned to
(56) and (57), respectively, let us consider a simple formal language
Le (the language of conditionals) in which the structure of con-
ditional statements is analysed in the way suggested above. The
vocabulary of Le consists of sentence letters (Po, PI> ... ), truth-
functional connectives I , & V, and two intensional connectives,
If and then. The sentences (well-defined formulas) of LC are
either assumptions or statements, and are defined as follows:

(60) (i) Sentence letters Po, PI>' .. are statements.

(ii) If A and B are statements, lA, A & B and A V Bare
statements.
326 R. Hilpinen

(iii) If A is a statement, 'If A' is an assumption.

(iv) If A and B are assumptions, A & B and A V Bare
assumptions.
(v) If A is an assumption and B is a statement, 'A, then B'
is a (conditional) statement.
(vi) An expression is a statement or an assumption only if
it is one by virtue of (i)-(v).

These rules are similar to those governing the construction of con-

ditional statements in natural languages. For example, according
to (60-iv) two assumptions can be joined together into a complex
(conjunctive or disjunctive) assumption, but an assumption cannot
be negated, i.e., it cannot contain a negation sign as the main
connective, even though a negative statement can be transformed
into a negative assumption by prefixing it with 'If'. According to
(60.v), a conditional statement is well-formed only if its ante-
cedent is a simple or complex assumption (and hence contains
at least one occurrence of 'If').
An interpretation of Le assigns to each sentence (statement or
assumption a subset of U of possible worlds, called the prop-
osition expressed by A. An assumption 'If A' is interpreted in
terms of a selection function in the way suggested by Kamp: the
proposition expressed by 'If A' consists of the plausible or selected
worlds satisfying the statement A. Since the plausibility of an
A-world depends on the reference-world or the context in which
the assumption is considered or entertained, the proposition
expressed by an assumption is also relative to the reference world.
Thus the proposition expressed by 'If A' in U is defined by

(61) IIf A lu =f(A, u),

where the selection function f is subject to conditions (21.i-iv).

The propositions expressed by conjunctive and disjunctive assump-
tions are defined in the usual way:

(62) IA & Blu = IA lu n IBlu.

(63) IA V Blu = IA lu u IBlu.

 Conditionals and possible worlds 327

The proposition expressed by a conditional statement 'A, then

B' (a conditional proposition) can be defined as follows:

(64) lA, then BI = the set of all worlds v such that IAlv is in-
cluded in IB I.

If a proposition P is included in Q, we say that P implies Q.

According to (64), a conditional proposition consists of all
possible worlds in which the antecedent would express a prop-
osition which implies the proposition expressed by the con-
sequent. Unlike an assumption, a statement expresses the same
proposition regardless of the context of utterance (or reference
world): the selection functionf(which brings about the dependence
of assumptions on the reference world) is not needed for the inter-
pretation of atomic statements (sentence letters) and their truth-
functional combinations, and (64) shows that the proposition
expressed by a conditional statement is also independent of the con-
text-world. Hence the subscript 'u' can be deleted from 'IA lu' if A is
a statement, and it is possible to speak of worlds at which a state-
ment A is true (,A-worlds') without regard to the reference world.
Assumptions, on the other hand, require a two-dimensional
concept of truth (or satisfaction). For this reason the first argu-
ment of the selection function must always be a statement; other-
wise it would not express a determinate proposition from which
the class of plausible worlds could be selected. (This restriction
is consonant with rule (60.iii), according to which 'If' can be
attached only to statements. Alternatively, we might take the first
argument off to be a proposition, not a statement; cf. Lewis [23],
p.60.)
According to (64), the truth of a conditional statement 'If A,
then B' is defined as follows:

(65) 'If A, then B' is true at u if and only if f(A, u) is included

in IBI.

Thus the present analysis gives simple conditional statements

exactly the same truth-conditions as analysis (20). In Le, it is
possible to distinguish two types of conditionals with disjunctive
antecedents, viz.
328 R. Hilpinen

(66) 1f(A V B), then C

and

(67) (If A) V (1.f B), then C,

which can be regarded as the formal counterparts of (56) and (57).

According to (64), the truth-conditions of (66) and (67) are
expressed by (55) and (54), respectively; consequently

(68) «(If A) V (If B), then C) ::> (If A , then C)

is a logical truth, but the elimination principle does not hold for
(66) (or for (56)): if the most plausible (Le., selected) B-worlds are
more plausible than the most plausible A-worlds, the former
worlds are also the most plausible worlds satisfying A VB; in this
case f(A V B, u) = f(B, u), and f(A, u) is not included inf(A VB, u).
McKay and van Inwagen's example (41) (with B = 'Spain fought
on the Axis side') illustrates such a case. The semantics outlined
a bove shows how the word 'or' can function as a wide-scope
conjunction in the antecedent of a conditional: if the connective
'or' has a wide scope in the antecedent, as in (57), the conditional
in question implies the conjunction of two simple conditionals.
In natural languages the word 'or' occurring in the antecedent
of a cond itional can nomlally be regarded as having wide scope
(even if the word 'if' has been deleted from the second disjunct);
but occasionally (as in McKay and van Inwagen's example) it has
narrow scope, and then the elimination principle fails. The narrow
scope interpretation can be ruled out by expressing the con-
ditional in form (57).
The preceding analysis of conditionals shows why the principle
of intensionality may appear problematic (ef. Nute [87,67,88]).
If the disjunction in the antecedent of

(69) If the window were open and the door were closed, or the
window were open and the door were open, we would get
some fresh air

is regarded as havine wide scope, (69) entails

 Conditionals and possible worlds 329

(70) If the window were open and the door were closed, we
would get some fresh air,

even though (70) is not entailed by

(71) If the window were open, we would get some fresh air.

However, this is not a counterexample to the principle of in ten-

sionality, since the wide-scope reading of the antecedent of (69)
is not logically equivalent (does not express the same proposition
as) the antecedent of (71).
According to the analysis outline above, conditionals with
conjunctive antecedents are also semantically ambiguous: a
conjunction in the antecedent of a conditional can have narrow
scope or wide scope and according to (61)-(63).

(72) If(A & B), then C

is not logically equivalent to

(73) (If A & If B), then C.

The principle of strengthening the antecedent fails to hold only

for (72): (73) is true if and only if

(74) f(A, u) n f(B, u) ~ ICI,

and is thus entailed by 'If A, then B'. The conditionals of natural

languages do not exhibit this ambiguity. For example,

(75) If this match were struck and if no oxygen were present, it

would light

appears to state exactly the same thing as (5), and neither (5) nor
(75) is entailed by sentence (4).
These observations suggest that a conjunction in the antecedent
of a conditional should always be regarded as having narrow scope,
or should be interpreted in such a way that it is equivalent to a
narrow-scope conjunction. This feature of conjunctive antecedents
330 R. Hilpinen

may be explained as follows: Condition (21.iv) implies that if

t(A, u) n t(B, u) is not empty, t(A & B, u) =tCA, u) n t(B, u); in
other words, if teA, u) n feB, u) =1= t(A & B, u), the intersection of
teA, u) and feB, u) is empty, and (72) is vacuously true. Thus a
conditional with a conjunctive antecedent is nonvacuously true
(and hence significantly assertable) only if the conjunction has
narrow scope or if the antecedent is equivalent to an assumption
involving a narrow-scope conjunction. Apparently the selection
function in terms of which conditionals are evaluated is subject
to various pragmatic pressures, and is usually adjusted to the
linguistic context in such a way that the entire utterance (i.e.,
conditional statement) is semantically and pragmatically significant;
consequently a conjunction in the antecedent of a conditional is
always interpreted in such a way that it is equivalent to a narrow-
scope conjunction.
The analysis of conditionals discussed above resembles that
presented by Lennart Aqvist in [25] . Aqvist represents subjunctive
conditionals in the form

(76) 0(* A J B),

where ,*, is a selection operator. I.L. Humberstone [91] has

pointed out that Aqvist's analysis permits a distinction between
two types of conditionals with disjunctive antecedents, viz.

(77) 0«* A V * B) J C)

and

(78) 0(* (A V B) J C),

of which only the former satisfies the elimination principle. How-

ever, as (76)-(78) show, Aqvist does not distinguish assumptions
from statements: his *-operator transforms statements into new
statements, and has no obvious counterpart in the conditional
sentences of natural languages. In addition, in Aqvist's semantics
the two-dimensional concept of truth required by the selection
operator is carried over to all sentences containing '*', unless
they are transformed back into 'ordinary' sentences (requiring
 Conditionals and possible worlds 331

the familiar one-world conception of truth) by a special operator

which again lacks a counterpart in natural-language conditionals
(cf. Lewis [23J, pp. 62-64; Aqvist [25], Appendix). The analysis
of conditionals in terms of the connectives 'If' and 'then' is a
simplification of Aqvist's analysis, and it represents conditionals
in a form which reflects their formulation in natural languages.

Ac knowledgement

I wish to thank I.L. Humberstone, Petervanlnwagen, Steven Kuhn,

Pekka Lahti, Donald Nute, Alke Stroup and Timothy Stroup for
their comments on the topics discussed (or mentioned) in this
paper, and Florida State University for support of research.
The topics discussed in section IX-XI are treated more
thoroughly in the author's paper 'Disjunctive Permissions and
Conditionals with Disjunctive Antecedents', in Proceedings of the
Second Soviet-Finnish Logic Conference (Acta Philosophica
Fennica, 1981).

BIBLIOGRAPHY

Anscombe, G.E.M. [78J Causality and Determination: An Inaugural Lecture.

Cambridge 1971.
Bacon, J. [8] The Logical Fonn of Perception Sentences.Synthese41 (1979):
271-308.
Bennett, J. [68] Counterfactuals and Possible Worlds, Canadian Journal of
Philosophy 4 (1974): 381-402.
Bigelow, J.C. [80] If-then Meets the Possible Worlds. Philosophia 6 (1976):
215-235.
Bowie, G.L. [75] The Similarity Approach to Counterfactuals: Some Prob-
lems. Nous 13 (1979): 477-498.
Chellas, B.F. [13] The Logical Form of Imperatives. Stanford 1969.
- [14J Imperatives. Theoria37 (1971.): 114-129.
- [62] Conditional Obligation. In S. Stenlund, Ed., Logical Theory and
Semantic Analysis, pp. 23-33. Dordrecht 1974.
- [26] Basic Conditional Logic. Journal of Philosophical Logic 4 (1975):
133-153.
Dretske, F.I., and Snyder, A. [79] Causal Irregularity. Philosophy of Science
39 (1972): 69-71.
Ellis, B. [39] Epistemic Foundations of Logic. Journal of Philosophical
Logic 5 (1976): 187-204.
332 R. Hilpinen

- [40] A Unified Theory of Conditionals. Journal of Philosophical Logic 7

(1978): 107-124.
- [32] Rational Belief Systems. Oxford 1979.
Ellis, B., Jackson, F., and Pargetter, R. [84] An Objection to Possible World
Semantics for Counterfactuals. Journal of Philosophical Logic 6 (1977):
355--357.
Fine, K. [69] Critical Notice of D. Lewis' Counterfactuals. Mind 84 (1975):
451-458.
Fraassen, B. van [54] Formal Semantics and Logic. New York 1971.
- [52] Semantic Analysis of Quantum Logic. In C.A. Hooker, Ed., Contem-
porary Research in the Foundations and Philosophy ofQuantum Theory,
pp. 80-113. Dordrecht 1973.
- [51] Hidden Variables in Conditional Logic. Theoria 35 (1974): 176-190.
- [48} Probabilities of Conditionals. In W.L. Harper and C.A. Hooker, Eds.,
Foundations ofProbability Theory, Statistical Inference, and Statistical
Theories of Science. Vol. I: Foundations and Philosophy of Epistemic
Applications of Probability Theory, pp. 261-300. Dordrecht 1976.
F¢ll1esdal, D., and Hilpinen, R. [92] Deontic Logic: An Introduction. In
R. Hilpinen, Ed., Deontic Logic: Introductory and Systematic Readings
pp. 1-35. Dordrecht 1971.
Goodman, N. [42} The Problem of Counterfactual Conditionals. In N. Good-
man, Ed., Fact, Fiction and Forecast, 2nd ed., pp. 3-27. Indianapolis
1965; reprinted from Journal of Philosophy 44 (1947): 113-128.
Greechie, RJ., and Gudder, S.P. [53} Quantum Logics. In C.A. Hooker, Ed.,
Contemporary Research in the Foundations of Quantum Theory,
pp. 143-173. Dordrecht 1973.
Gardenfors, P. [43} Conditionals and Changes in Belief. In I. Niiniluoto and
R. Tuomela, Eds., The Logic and Epistemology of Scientific Change,
(Acta Philosophica Fennica 30, Nos. 3-4), Amsterdam 1979.
Haack, S. [33} Lewis's Ontological Slum. Review of Metaphysics 30 (1976-
77): 415-429.
- [34] Philosophy of Logics. Cambridge 1978.
Hansson, B. [60] An Analysis of Some Deontic Logics, Nous 3 (1969):
373-398.
Hardegree, G.M. [66} The Conditional in Quantum Logic. Synthese 29
(1974): 63-80.
Hertzberger, H.G. [57] Counterfactuals and Consistency. Journal of Philos-
ophy 76, 83-88 (1979).
Hill, C.S., and Greary, L.G. [85} Review of Counterfactuals. Philosophy of
Science 42 (1975): 351-354.
Hilpinen, R. [45] Decision-theoretic Approaches to Rules of Acceptance. In
R.E. Olson and A. Paul, Eds., Contemporary Philosophy in Scandinavia,
pp. 147-168. Baltimore 1972.
- [65] Approximate Truth and Truthlikeness. In M. Przele~ki, K. Szaniawski
and R. Wojcicki, Eds., Fonnal Methods in the Methodology ofEmpirical
Sciences, pp. 19-42. Dordrecht/Wroclaw 1976.
 Conditionals and possible worlds 333

- [93J Deontic Logic and the Semantics of Possible Worlds. In A.G. Conte,
R. Hilpinen and G.H. von Wright, Eds., Deontische Logik und Semantik,
pp. 82-88. Wiesbaden 1977.
Hintikka, 1. [IOJ Quantifiers in Deontic Logic. Societas Scientiarum Fennica,
Commentationes Humanarum Litterarum, 23 (4). Helsinki 1957.
- [1 J Knowledge and Belief' An Introduction to the Logic of the Two
Notions. Ithaca 1962.
- [2J Individuals, Possible Worlds and Epistemic Logic. Nous 1 (1967):
32-62.
- [5] On the Logic of Perception. In N.S. Care and R.H. Grimm, Eds.,
Perception and Personal Identity, pp. 140-173. Cleveland 1969. Re-
printed in 1. Hintikka, Models for Modalities, pp. 151-183. Dordrecht
1969.
- [3] Semantics for Propositional Attitudes. In J.W. Davis, D.l. Hockney
and W.K. Wilson, Eds., Philosophical Logic, pp. 21-45. Dordrecht 1969.
Reprinted in Models for Modalities, pp. 87-111. Dordrecht 1969.
- [11] Some Main Problems of Deontic Logic. In R. Hilpinen, Ed., Deontic
Logic: lntroductory and Systematic Readings, pp. 59-104. Dordrecht
1971.
- [4] The Intensions of Intentionality and Other New Models for Modalities.
Dordrecht 1975.
Humberstone, I.L. [91] Two Merits of the Circumstantial Operator Language
for Conditional Logics. Australasian Journal of Philosophy 56 (1978):
21--24.
lackson, F. [73] A Causal Theory of Counterfactuals. Australasian Journal 0/
Philosophy 55 (1977): 3-21.
Kamp, H. [59] Free Choice Permission. Aristotelian Society Proceedings N.s.
74 (1973-74): 57-74.
-- [97] Semantics versus Pragmatics. In F. Guenther and S.1. Schmidt, Eds.,
Formal Semantics and Pragmatics/or Natural Languages, pp. 255-287.
Dordrecht 1979.
Kanger, S. [12] New Foundations for Ethical Theory. In R. Hilpinen, Ed.,
Deontic Logic: Introductory and Systematic Readings, pp. 36-58.
Dordrecht 1971.
- [15] Law and Logic. Theoria 38 (1972): 105-132.
Lehrer, K. [19] 'Can' in Theory and Practice: A Possible Worlds Analysis. In
M. Brand, and D. Walton, Eds., Action Theory, pp. 241-270. Dordrecht
1976.
Lemmon, E.J., and Scott, D. [83] An Introduction to Moda! Logic. Ed.
K.Segerberg, American Philosophical Quarterly Monograph No. 11.
Oxford 1977.
Levi, I. [44] Gambling with Truth. New York 1967.
Lewis, D. [22] Completeness and Decidability of Three Logics of Counter-
factual Conditionals. Theoria 37 (1971): 74-85.
- (23] Counter/actuals. Oxford 1973.
- (241 Counterfactuals and Comparative Possibility. Journal of Philosophical
Logic 2 (1973): 418--446.
334 R. Hilpinen

- [64] Causation. Journal of Philosophy 70 (1973): 556-567.

- [58] Semantic Analyses for Dyadic Deontic Logic. In S. Stenlund, Ed.,
Logical Theory and Semantic Analysis, pp. 23-33. Dordrecht 1974.
- [86] Possible World Semantics for Counterfactual Logics: A Rejoinder.
Journal of Philosophical Logic 6 (1977): 359-363.
- [27] Truth in Fiction. American Philosophical Quarterly 15 (1978):
37-46.
- [76] Counterfactual Dependence and Time's Arrow. Nous 13 (l979):
455-476.
- [96] A Problem about Permission. In E. Saarinen et al., Eds., Essays in
Honour ofJaakko Hintikka, pp. 163-175. Dordrecht 1979.
Loewer, B. [89] Counterfactuals with Disjunctive Antecedents. Journal of
Philosophy 73 (1976): 531-537.
- [77] Cotenability and Counterfactual Logics. Journal of Philosophical
Logic 8 (1979): 99-115.
McKay, T., and van Inwagen, P. [90] Counterfactuals with Disjunctive
Antecedents. Philosophical Studies 31 (I977): 353-356.
Mackie, J.L. [35] Truth, Probability and Paradox. Oxford 1973.
Mott, P. [63] On Chisholm's Paradox. Journal ofPh ilosoph ical Logic 2 (1973):
197-211.
Niiniluoto, 1. [7] Knowing That One Sees. In E. Saarinen, Ed., Essays in
Honour of JaakkoHintikka, pp. 249-282. Dordrecht 1979.
Nute, D. [82] Counterfactuals. Notre Dame Journal of Formal Logic 16
(1975): 476-482.
- [87] Counterfactuals and the Similarity of Worids. Journal of Philosophy
72 (1975): 773-778.
- [67] David Lewis and the Analysis of Counterfactuals. Nous 10 (1976):
355-361.
- [88] Simplification and Substitution of Counterfactual Antecedents.
Philosophia 7 (1978): 317-326.
Peirce, C.S. [37] Collected Papers, Vol. 4. Eds. C. Hartshorne and P. Weiss.
Cambridge 1933.
Plantinga, A. [28] The Nature ofNecessity. Oxford 1974.
- [29] Actualism and Possible Words, Theoria 42 (1976): 139-160.
Pollock, J. [81] Four Kinds of Conditionals. American Philosophical Quarterly
12 (1975): 51-59.
- [55] The 'Possible Worlds' Analysis of Co unterfactuals. Philosophical
Studies 29 (1976): 469-476.
- (56] Subjunctive Reasoning. Dordrecht 1976.
Powers, L. [36] Comments on R. Stalnaker's paper 'Propositions'. In Issues
in the Philosophy of Language, pp. 93-103. A.F. MacKay and D.
Merrill, Eds., New Haven 1976.
Porn, I. [17) Some Basic Concepts of Action. In S. Stenlund, Ed., Logical
Theory and Semantic Analysis, pp. 93-102. Dordrecht 1974.
- [18] Action Theory and Social Science. Dordrecht 1977.
Quine, W.V.O. [49] Methods of Logic, revised edition. New York 1959.
 Conditionals and possible worlds 335

Ramsey, F.P. [38] General Propositions and Causality. In R.B. Braithwaite,

Ed., The Foundations of Mathematics, pp. 237-255. london 1931.
Rescher, N. [41] Hypothetical Reasoning. Amsterdam 1963.
- [46] The Coherence Theory of Truth. Oxford 1973.
- [47] Plausible Reasoning. Assen 1976.
Richards, T. (70] The Worlds of David Lewis. Australasian Journal of Philos-
ophy 53 (1975): 105-118.
Slote, M. [74] Time in Counterfactuals. Philosophical Review 87 (1978):
3-27.
Smith, D.W. [9] The Case of Exploding Perception. Synthese 41 (1979):
239-269.
Stalnaker, R. [20] A Theory of Conditionals. In Studies in Logical Theory,
pp. 98-112. Ed. N. Rescher, American Philosophical Quarterly Mono-
graph No.2. Oxford 1968.
- [30] Propositions. In A.F. McKay and D. Merrill, Ed.,Issues in the Philos-
ophy of Language, pp. 79-91. New Haven 1976.
- [31J Possible Worlds. Noas 10(1976): 65-76.
Stalnaker, R., and Thomason, R. [21] A Semantic Analysis of Conditional
Logic. Theoria 36 (1970): 23-42.
Thomason, R. [50] A Fitch-Style Formulation of Conditional Logic. Logique
et Analyse 13 (1970): 397-412.
- [6] Perception and Individuation. In M. Munitz, Ed., Logic and Ontology,
pp. 261-285. New York 1973.
Tichy, P. [71] A Counterexample to the Stalnaker-Lewis Analysis of
Counterfactuals. Philosophical Studies 29 (1976): 271-273.
- [72] A New Theory of Subjunctive Conditionals. Synthese 37 (1978):
433-457.
Wright, G.H. von [94] An Essay in Deontic Logic and the General Theory
of Action. Acta Philosophica Fennica 21. Amsterdam 1968.
- [95J Deontic Logic and the Theory of Conditions. In Deontic Logic:
Introductory and Systematic Readings, pp. 159-171. Dordrecht 1971.
Aqvist, L. [61 J Deontic Logic Based on a Logic of 'Better'. In Proceedings of
a Colloquium on Modal and Many-valued Logics, pp. 285-290. (Acta
Philosophica Fennica 16). Helsinki 1963.
- [25J Modal Logic with Subjunctive Conditionals and Dispositional Predi-
cates. Journal of Philosophical Logic 2 (1973): 1-76.
- [16] A New Approach to the Logical Theory of Actions and Causality. In
S. Stenlund, Ed., Logical Theory and Semantic Analysis, pp. 73-92.
Dordrecht 1974.
Entailment and the disjunctive syllogism
NUEL D. BELNAP, Jr. and J. MICHAEL DUNN*
University of Pittsburgh

1. INTRODUCTION

'Entailment' we use for a connection between statements corres-

ponding to there being a valid argument from one to the other,
and we invoke the witness of two thousand years to sanction our
putting on this connection the requirements of relevance and
necessity. More locally, we rely on the tradition depending from
Ackermann [I], and summarized in Anderson and Belnap [2]
(1975, henceforth, AB75).
Entailment can come grammatically as either a connective or a
predicate (AB75, appendix). As a connective, we can read A -+B as
'that A relevantly and necessarily implies that B'; AB75 motivates
a certain calculus E as exactly matched to this concept, deploying,
as catching the required concept of relevance, the use of the ante-
cedent A in getting to the consequent B. In this paper we are, in
the main, going to ignore entailment when clothed in the garb of
a connective, concentrating instead of its relational sense; never-
theless, for reference below, we mention here that AB75 also
motivates a calculus R of relevant implication containing a connec-
tive exactly matched to the simpler concept 'that A relevantly
implies that B' (without commitment to necessity).

1.1 Tautological entailment. We turn now and for the rest of this
paper to entailment as a relation between sentences; furthermore,
we address ourselves to the special case when the necessity

·We thank R.K. Meyer for discussions on these topics across the years and
oceans, and S. Kripke for more recent conversations across tea tables.

Contemporary philosophy. A new survey. Vol. 1, pp. 337-366.

© 1981, Martinus Nijho!! Publishers, The Hague/Boston/London.
338 N.D. Belnap, Jr. and M. Dunn

involved is logical necessity. So let A, B, etc., be truth-functional

formulas in, say, &, V, and --, with other truth-functions being
defined in the standard way. AB75 motivated and then proposed
as adequate to the relational idea of entailment the concept of
tautological entailment, which may conveniently be described as
follows. To test whether A -"" B is a tautological entailment, first
put A in a disjunctive normal form A' and B in a conjunctive
normal form B', using only de Morgan's, Double Negation, and
Distribution principles. Then for each disjunct A * of A' and each
conjunct B* of B', ask if some conjunct of A * is exactly the same
as some disjunct of B*, and count A -"" B a tautological entailment
just in case the answer is an invariable yes'.
The reader can easily check that we have correctly sorted the
following few examples:

Tautological entailments Not tautological entailments

(l)p&~p-""p (4)p&-p-""q
(2)p-""pvq (5) p -""qv-q
(3) ~p&(pvq)-""qV(p&"'p) (6) p&(-pvq)-""q
(7) -p&(pvq)-""q
(8) (p & - p ) Vq -"" q
(9) p-""p &(qv-q)

1.2 The disjunctive syllogism. Amongst those on the 'bad side' it

has been (7) perhaps which has received most attention. It repre-
sents, of course, the disjunctive syllogism (henceforth, the d. s.),
the argument pattern, from A v Band -A to infer B. The rejec-
tion of the d. s., we must say frankly, has been taken as noxious
by many logicians, both relevant and irrelevant, critical of the
AB75 enterprise.! On the one hand this reaction does not surprise
us; the d. s. was, after all, one of the Stoic's 'Five Indemonstrables.'
Still, on the other hand, we do take some small solace in the fact
that it was the Fifth-recalling the tradition of notoriety, starting
with Euclid's Elements, regarding fifth postulates. Also we suppose
that it is better to deny an Indemonstrable than a Demonstrable.
Still, we can imagine a reader who does not find these con-
siderations conclusive. For this reader's sake, who doubtless shares
with the critics at least surprise at the relevant failure of the d. s.,
we are going to devote the lion's share of our considerations to
the conflict between relevance on one side and the d. s. on the other.
 Entailment and the disjunctive syllogism 339

We pause to portend confusion. The logics with which we deal

all postulate double negation, so that (7) - the d. s. - and (6) -
modus ponens for material 'implication' - come to the same
thing. We shall therefore on occasion take the liberty of con-
sciously confusing the two, using 'the d.s.' as the name for both.

1.3 Relevance logic and relevantism. In the course of these con-

siderations we should like to address ourselves to a number of
questions. In order to do so in a convenient way, we presume to
impose on the reader some Terminology. There are in the first
place those who believe that it is worth paying attention to the
concept of a relevant connection between statements, and worth
using formal techniques in an effort to get clear on the idea of
relevance; such a person we label a relevance logician in the wide
sense. Within this group there are those who adopt the position
that the concept of tautological entailment as just described
represents a stable and accurate and interesting analysis of relevant
implication; these are relevance logicians in the narrow sense, but
for purposes of this paper, and with all due respect to our wide-
sense colleagues, we are going to drop 'narrow' and thus reserve
relevance logician for members of this group. (Contrary to the
relevance logician is the irrelevant logician, the fellow who thinks
there is nothing whatsoever, beyond freshman confusion, to the
topic of relevance in logic; but we shall have very little to say
about this not-so-rara avis.)
There is a further distinction to be made for which we lay the
groundwork by reference to intuitionism. We all know that there
are those who believe intuitionists have got hold of an interesting
idea and are prepared to try to throw formal light on this idea
without themselves being intuitionists, e.g., Kripke [3]. In con-
trast, there are those who (like Brouwer) take intuitionistic
principles as the very standard of reasoning and by so much
reject classical two-valued logic. A relevantist we define to be like
Brouwer - he rejects classical logic and instead adopts tautological
entailmenthood as the proper standard of correct inference. (By so
much we can be seen to be giving a narrow sense to 'relevantist' in
this paper.) The contrary of the relevantist is the classicalist, who
subscribes to two-valued logic as the organon of inference.
Evidently one can be a relevance logician without being a
340 N.D. Belnap, Jr. and M. Dunn

relevantist; and indeed, in contrast to the case of intuitionism, it is

not clear that there are any full-blooded relevantists, though there
are certainly lots of relevance logicians.
This Terminology permits us to describe what we are up to as
follows: we propose to defend the claim of the relevance logician,
and to investigate the claim of the relevantist. That is, we shall be
defending the view of the relevance logician that indeed tautologi-
cal entailment is a stable and interesting concept of relevant
logical implication; but we are not in this paper going to defend
relevantism. Instead, we shall be investigating the nature, co-
herence, and ramifications of the relevantist claims that tauto-
logical entailmenthood represents the correct norm of inference,
and that the irrelevant inferences of the classicalist, with all their
prewmed subtlety, must in the end be labeled as little better than
exhibitions of brute cunning (AB75, § 25.1).

1.4 Our plan. In more detail, we plan to proceed as follows. In

§ 2 we begin to speak of relevantism, and in particular we discuss
the mysterious concept known as 'Boolean negation' and what it
means, or doesn't, to a relevantist. So far we shall have stayed at
the level of propositional logic; in § 3 we briefly discuss the
extension of relevance concepts to richer domains: individual
quantifiers, higher-<>rder quantifiers, and, importantly, arithmetic.
In this section, too, we introduce the topic of the 'admissibility'
of the d. s. as a metalinguistic property of logistic systems, and
indicate what facts relevance logicians have established about this
matter. In the next section, § 4, we shall probe more deeply into
the language and logic used by the relevance logicians, especially
as regards the admissibility of the d. s., asking to what extent
relevance logicians could, if they wished, count themselves as
relevantists. Finally, in § 5, we essay a discussion of what it would
really be like to be a reJevantist, centering our discussion on the
dual topics of the admissibility of, and the use of, the d. s.

2. BOOLEAN NEGA nON

There are some things one sometimes wishes had never been
invented or discovered: e.g., nuclear energy, irrational numbers,
 Entailment and the disjunctive syllogism 341

plastic, cigarettes, mouthwash, and Boolean negation. The reader

may possibly have not yet heard of this last named threat, and
indeed it is our present purpose to inform and caution him regard-
ing it.

2.1. Background. First we give some background. Meyer and

Routley [4] discovered that it was possible to add to relevance
logics a new negation -', called Boolean negation, which contrasts
with the usual negation -- present in the relevance logics, called
de Morgan negation. Boolean negation, unlike de Morgan negation,
has such surprising properties as that A & 1 A entails B, and that
1.11 &(AVB) entails B. Meyer [51 uses Boolean negation to give
elegant axiomatizations of the relevance logics similar in style to
the well-known Godel-Lemmon axiomatizations of the Lewis
modal logics, in which one starts with a base containing all classi-
cal tautologies (these being expressed with classical negation
interpreted as Boolean and not de Morgan negation).
The interesting thing about the Meyer axiomatizations, besides
their style, is that they seem to demonstrate that in the full
systems of relevance logic (implication as a connective) de Morgan
and Boolean negations can live side by side in a peaceful co-
existence. Indeed, Meyer shows that his axiomatizations are 'con-
servative extensions' of the usual ones in that no new theorems
arise in the standard vocabulary of relevance logic (which, of
course, excludes Boolean negation).
We should stress, though, that new such derivable rules do
arise in the Meyer axiomatizations, and so they are not 'con-
servative extensions' in an extended sense. In particular one can
derive the terrible ex falsum quod libet in rule form with de
Morgan negation: A & -A ~B. This makes problematic any straight-
forward use ofaxiomatizations of relevance logic involving
Boolean negation for the purpose of developing interesting
(potentially) inconsistent theories, at least if one wants those
theories to be dosed under derivability. Indeed, any theory that
contained the logical theorems ('regular' theories - see AB75)
would be so closed (the other theories are merely closed under
entailment - there might be some merit in looking at these in
the context of the Meyer axiomatizations). Closely connected
with the above is the fact, recently noted by Meyer (corre-
342 N.D. Belnap, Jr. and M. Dunn

spondence), that if propositional quantifiers are allowed, then

Boolean negation forces the extension to be nonconservative even
in the unextended sense: 3 p [p & (p &A & - A ~ q)] witnesses the
trouble, choosing p, of course, as --, (A &- A).
The idea of Boolean negation first arose in the context of the
Routley-Meyer semantics for relevance logic, but in the setting in
which we have placed ourselves in this paper (entailment as a
relation), it is easier to discuss it in the context of a four-valued
semantics developed in Dunn [6] and Belnap [7] (cf.Belnap [8];
also Meyer [9] , which explicitly introduced Boolean negation into
this context).
The problem in Belnap [7] was to devise a good logic for com-
puters (mechanical question-answering systems) to use when there
is real risk that the data-base from which answers to questions are
to be inferred may be inconsistent. Earlier work of Dunn (pub-
lished in 1976) suggested that in such a situation sentences be
considered to have not two but four values, which we here follow
Belnap [7] in calling T, F, None, Both, representing the cases in
which the reasoner has been told about a certain sentence that it is
true but hasn't been told that it is false; has been told that it is
false, but not that it is true; has not been told anything; and has
been both told that the sentence is true and that the sentence is
false.
The four values form a 'logicallattice' when they are ordered as
follows:

T
Both VNone
F

An upward-directed path from a to b is to be thought of as a's

implying b. Operations are defined on these four values so that
a&b == the greatest lower bound of a, b; aVb == the least upper
bound of a, b; and -T == F, -F == T, "'Both == Both, "'None
==None.
These operations give rise to a four-valued logic that is easily
seen to be the same as tautological entailmenthood. It is possible to
motivate their definition in different ways (cf. Belnap [7] and
 Entailment and the disjunctive syllogism 343

Dunn [6]), but we will not go through all that here again. We do,
however, want to focus attention on the definition of "'. Pre-
sumably we do not have to argue for the definitions of - T and
-F; and given what we mean by the four values, we feel the
definitions of -B and -N are almost as obvious. Thus, if the com-
puter (or anyone else for that matter) has been told that A is both
true and false, then there is certainly a point to saying that it has
been told that -A is also both true and false (true because A is
said to be false, false because A is said to be true). And there is a
similar point to saying that if the computer has been told nothing
about the truth value of A, then it has been told nothing about the
truth value of -A either.
Following Meyer, let us call the operation '" de Morgan ne-
gation. Now the problem of this section is that there is another
operation on the four values with some claim to be considered as
negation, again following Meyer, Boolean negation" defined so
it behaves the same as the de Morgan negation""" on T and F, but
so -lBoth = None and ,None = Both. One can imagine motivating
that ,Both = None by saying something like this: if a sentence A
is marked as both true and false, then ,A cannot be marked as
true since in order for this to be. the case it would have to be that
A is not marked as true (but it is). And similarly ,A cannot be
marked as false since then A would have to be not marked as false
(but it is). So ,A must be marked as None. And the reader can go
through similar moves for himself in order to motivate that ,None
=Both.
In a nutshell, the difference between - and, would seem to be
that - is a kind of 'internal' negation, whereas, is a kind of
'external' negation; -A might be read as 'A is false,' whereas ,A
should be read as 'it is not the case that A is true.'
At least two questions force themselves upon us. (1) Which is
the real negation? (2) Even supposing that Boolean negation is not
the real negation, why should we not have it in our logic anyway?
(1) of course drags along with it a subsidiary question about just
what kind of question it is anyway. Is it a question of meta-
physics, linguistics, of logic? We shall not presume to put these
questions to rest here, but we do wish to address ourselves to some
ancillary questions they raise.
344 N.D. Belnap. Jr. and M. Dunn

2.2 A dilemma. The orthodox relevantist account of the inter-

action of negation with entailment goes something like this. In
logic we find the ordinary truth functional connectives, in par-
ticular conjunction, disjunction, and negation, and what classical
logic tells us about them is partly true and partly false. The true
part is what it tells us about which sentences involving only the
truth functions (and quantifiers) are logically true. The false part
is what it tells us about which such sentences entail one another.
In particular AB75 complains bitterly about ex falsum quod libet
(A and not-A entails B) and the d. s. They presume to be talking
about the same truth-functional connectives (and quantifiers) as
does the classical logician, but urge that a different, tighter
relationship be taken as entailment.
Now once the existence of Boolean negation is noticed by the
classicalist, he can well reply that the relevantists are doing more
than merely insisting that he should use 'entailment' in a non-
classical sense - they are also insisting that he use 'negation' in a
non-classical sense. Classical negation is Boolean negation, and ex
falsum quod libet and the d. s. hold for Boolean negation, as the
reader can easily check for himself, using the logical lattice above.
And the classicalist could go to point out that even if some rele-
vantist should succeed by some clever argument in showing that
de Morgan negation was after all the 'real' negation, perhaps even
the same after all as 'classical negation,' Boolean 'negation' would
still remain. And on the face of it, that A & I A entails B would
seem to be as objectionable upon AB75 style intuitions about
relevance as would A & ~ A entails B.
Now it seems to us that the relevantist has a reply to the classi-
calist, and we shall attempt to sketch it here. But we want to
admit at the outset that Boolean negation is a complicated topic,
and we may just be confused.
The gist of our reply for the relevantist is that he does not have
to, indeed should not, recognize the legitimacy of Boolean
negation.
Let us look closely at the distinction between de Morgan and
Boolean negation, using a construct of Dunn [6] . There, in effect,
Both was interpreted as the set it, f}, None as the empty set <p,
T as {t}, and F as {f}. For any sentence A , let IA I be the value of
A. Then de Morgan and Boolean negations may be neatly com-
 Entailment and the disjunctive syllogism 345

pared and contrasted in their truth (and falsity) conditions as

follows:

de Morgan: (t-) tEl-AI ~ fE IAI

(f-) fE I-AI ~ t E IAI

Boolean: (t ,) t E I,AI ~ not (t E IAI)

(h) fE I,AI<f-+ not (fE IAI)

These clauses certainly seem to support the kind of distinction

between de Morgan and Boolean negations suggested earlier in this
section, where de Morgan negation is 'internal,' the Boolean
'external.' They do anyway until one stops to ask what kind of
metalinguistic 'not' it is that occurs in the Boolean clauses above.
This is a profound question, and upon it we construct a
dilemma along the following lines. If the 'not' is a de Morgan
negation (as surely it would be for the relevantist), then given
plausible semantical principles, the 'internal/external' distinction
collapses and ,A co-entails -A, and so we have only one kind of
negation after all. On the other hand, if the 'not' is Boolean, then
the relevantist can simply and consistently claim not to under-
stand it. Thus recognizing only one kind of 'not' (the de Morgan
one) is at least a stable position. But we shall go on to argue that
not only can the relevantist take such a position, further he
should, given motivations of concern for reasoning in situations
of possibly inconsistent information.

2.3 Horn 1. That was the bare bones of the dilemma; we now
begin to flesh it out. Taking the first horn then, let us suppose that
the metalinguistic 'not' is de Morgan. Let us symbolize it by -,
letting context determine whether - is meta-linguistic or object-
linguistic. Surprisingly, we can show, given plausible semantical
assumptions,

(*) fE IAI ~ -(t E IAI).

We say 'surprisingly' since (*) seems to be a rejection of the

four-valued semantics; (*) seems to say that a sentence A is
assigned precisely one of the two truth values (never Both or
None).
346 N.D. Belnap, Jr. and M. Dunn

What are the plausible assumptions about the semantics of the

metalanguage? They are: (1) the metalanguage too should be given
a four-valued interpretation (where '{) is a metalinguistic expression,
we shall let IIIPII be the value of IP); (2) the meta linguistic '" should
be evaluated as in the clause for de Morgan negation above
(replacing I I by II II, etc.); and (3) the metalinguistic sentences
tElA I and f E IA I should be evaluated as follows:

II t E IAIII = IAI; II fE IA III = I-AI.

It is not our point here to defend all of these assumptions as the
only ones that could have been made about the semantics of the
metalanguage; we merely wish to point out that these are all
plausible assumptions for a relevantist to make. Assumption (2),
of course, merely represents our choice to explore one horn of the
dilemma we are constructing. Assumptions (I) and (2) have
definite Tarskian-Davidsonian overtones about them, and reflect
some sort of decision to treat the object language and the meta-
language in very similar ways. As well, (3) has Ramseyan under-
tones about a redundancy or disappearance theory of truth (and
falsity).
We now go about deriving (*). To make our presentation
concise, we adopt yet one more use for the symbol "'. We already
have ~ as a connective of the object language and ~ as a connec-
tive of the metalanguage. We now want as well ~ as an operation
upon the four values as explained at the beginning of this section
(~T = F, "'F = T, ~Both = Both, "'None = None), with these four
values now thought of as sets of the usual two truth values t, L
of course. Then

(i) II f E IA III = I~ AI (Semantical assumption (3))

(ii) I~AI = ~IA I (Easy to check)
(iii) ~IAI = ~II tElA III (Semantical assumption (3), substitu-
tion of identicals)
(iv) ~II t E IAIII = 1I~(t E IAI) II (Semantical assumption (2),
and the checking at (ii))
(v) II f E IAIII = II ~(t E IAI) II ((i)-(iv), transitivity of =)
 Entailment and the disjunctive syllogism 347

Then (*) is an immediate consequence of (v), since (v) says the

left and right hand sides of the biconditional (*) evaluate the same.
Now that we have (*), its dual form

(**) t E IAI ~ -(fE IAI)

follows by contraposition and double negation (both relevantly

valid principles).
Finally, t E I-A I ~ t E /.AI follows directly from (t-) and
(t.) (interpreting the 'not' in (t-l) as -, since we are on horn one),
using (*). And similarly f E I-A I ~ f E I.AI follows from (f-)
and (h), using (**). SO /-AI = I.AI, and-A and.A co-entail
one another, and so are fundamentally indistinguishable, as
promised.
We find these considerations more than a little perplexing, and
we shall return to reflect and moralize upon them a bit later. But
first let us turn to exploring the second horn of the dilemma. The
point there is simply that if the 'not' in the clauses (tI), ([.) is
Boolean, then the relevantist can claim not to understand it,
recognizing, as he does, only one negation. But the further ques-
tion is, of course, should he understand it? We think not, for
reasons we shall now develop.

2.4 Horn 2. Let us put ourselves in the existential situation of the

computer of §2.3 (or other reasoner, perhaps ourselves) having
to make inferences in an environment of possibly inconsistent
information. It seems to us that such a reasoner has no conceivable
use for Boolean negation.
Let us suppose that this reasoner's sentences are being marked
in the four-valued way. Then '~A makes perfect sense, both as
input and as output (input occurs when the reasoner is told things
- by a programmer, informant, nature, whatever - and output
occurs when the reasoner is asked things). Thus, e.g., if the
reasoner receives -A marked t as input, then A is to be marked f;
and the reasoner can output -A marked t if A is already marked
as f. This is just the practical content of the clause (t-), and (f-)
has similar practical content about -A being marked as f if, and
only if, A is marked as t.
But what possible practical content can be ascribed to the
348 N.D. Belnap, Jr. and M. Dunn

clauses (t-,) and ([-,)? We can say that (t-,) instructs that -,A
should be marked as t if, and only if, A is not marked as t. But
what does this mean, practically speaking, from an input-output
point of view?
On the input side, does it mean that the reasoner receiving-,A
marked t as input, should 'unmark' A as t (erase any marking of
A as t)? Then what is the reasoner supposed to do when it receives
A &-,A marked t as input? Both mark and unmark A as t (or mark
A as t and then erase)? Trying to both mark and unmark A as t
seems to invite psychotic breakdown (and the parenthetical
alternative of marking and erasing is not without problems: it
would seem that A&-,A would differ from -,A&A, and there are
deeper troubles, to which we shall advert after we discuss the rule
of -,A as output).
Let us think about the conditions under which the reasoner can
produce -, A marked t as output. The reasoner would first have to
verify that it has not been told A, either explicitly or implicitly.
This last is most important, and we have not stressed it previously.
If the reasoner can ultimately deduce A (output A marked t), then
we would not want the reasoner to report out -,A (output -, A
marked t) purely on the basis of his not yet having gotten around
to the appropriate deduction of A. Indeed, -,A is a claim that such
a deduction does not exist. Thus -, would be an 'ineffective' con-
nective in the technical sense, since it is well-known (Church's
Theorem) that there is in general no mechanical procedure for
determining whether such deductions of A exist (at least if quanti-
fiers are present, which we suppose they are in any interesting
case).
Reflecting upon what has just been said reveals new problems
for -,A on the input side as well. Taking it as an instruction to
'erase A' is not really a viable alternative, at least in the absence of
some formal (mechanical?) model of how a reasoner 'takes things
back'. On being told -,A, we do not merely want the reasoner to
'keep quiet about A', even though all its information points to
A's being true. It is still then implicitly told A. What we would
want, we guess, is for the reasoner to correct its information so
that A is no longer deducible. But how is it to do this? We need a
theory of theory correction, and there seems to us to be no such
fully developed theory on the market. Without such a theory -,A
 Entailment and the disjunctive syllogism 349

cannot be viewed as an explicit instruction - it is at best a pious

hope. (This point relates to one which could be put more tech-
nically in the language of Belnap [8]: Boolean negation is not
'ampliative.' It is also true that it is not 'continuous' in the sense
of that paper and thus would be ruled out by what is there called
'Scott's thesis.')
Let us back away from the horrid detail of the problems of
treating I A as input/output to a question-answering machine,
and summarize our feelings in a somewhat metaphorical way.
Classical negation (ordinary two-valued negation) is a connective
fit only for God. It is an ontological negation that can be used as
an epistemic negation only by the omniscient. De Morgan negation
is the appropriate epistemic negation for the poor finite reasoner,
be it machine or human. Four-valued Boolean negation is the
temptation that must be resisted, promising as it does to combine
the ontological and epistemic. Thus IA is supposed to mean that
A really is not marked t.
Boolean negation seems to us to be best understood as an
attempt to push down into a four-valued object language the two-
valued negation of the classical metalanguage. Relevance logicians
have so far invariably used a classical metalanguage, a practice that
might be excused by the relevantist as preaching to the heathen in
his own language. But the true relevantist should for himself use a
relevant metalanguage, with the only negation being de Morgan
negation. And then, as we have seen in exploring the first horn of
our dilemma, it would appear that the truth (and falsity) con-
ditions for Boolean negation can no longer be stated so as to dis-
tinguish it from de Morgan negation.

2.5 A puzzle. The reader may still feel a sense of puzzlement

about all this (we do). How is it in the end that what started as a
four-valued semantics set forth in a classical metalanguage ends up
a two-valued semantics when reinterpreted in a relevant meta-
language (cf. (*) above)?
One point that can be made is that one would not have expec-
ted that a four-valued semantics would have been needed with a
relevant metalanguage. If all of the connectives in sight are 'rele-
vant,' then on Tarskian-Davidsonian intuitions there could not be
too much wrong with the 'homophonic'
350 N.D. Belnap, Jr. and M. Dunn

(#) (-A) is t +-+ -(A is t).

On the other hand, one would have definitely expected that some
additional apparatus beyond the usual two-valued approach was
necessary to do a semantics for relevance logic in a classical meta-
language.
But these considerations do not by themselves dispel the puzzle.
Although one need not give a four-valued semantics in a relevant
metalanguage, couldn't one? The answer seems to be no, given at
least the semantical assumptions we adopted in exploring the first
hom of our dilemma. But why is it no? It seems to us that the
answer is something like the following.
There is no way in a relevant metalanguage (without Boolean
negation) to express that a sentence A takes on, say, just the
value t (and not f as well). The most that one can say is that A is
at least t (t E IAI). The metalinguistic sentence

(tEIAI)&-(fE IAI

does not do the trick it might be thought to do, since (at least
upon our plausible semantical assumptions) it asserts that A is at
least true and it is not the case that A is at least false. But this last
conjunct does not mean that A is really not at least false; all it
means in the end, given our analysis of -, is that A is at least true,
the same as the first conjunct.
Similarly, there is no way to say that A takes on just the value
f - all that one can say is that A is at least f. The two values of the
relevant metalanguage are, as it were, 'at least t' and 'at least f,'
whereas the usual two values of a classical metalanguage are, as it
were, )ust l' and 'just f'
The surprising result (*) above now should look much less sur-
prising. The shock of it was that it appeared to say that A took on
precisely one of the values t, f - that A was at least false if, and
only if, A was not at least true. But now we see that since the 'not'
in question is de Morgan negation, the right-hand side does not
mean really that A is not at least true. All it means is that it is at
least false that A is at least true, which reduces (on our semantical
principles) to the left-hand side - A is at least false. Got it?
 Entailment and the disjunctive syllogism 351

3. QUANTIFIERS AND ARITHMETIC IN A RELEVANT SETTING

The chief topic of this section is to indicate, with all brevity, the
ways in which relevance logic has been enriched by quantifiers,
and, with almost equal brevity, how arithmetic has been enriched
by relevance logic. We offer these indications partly for their
intrinsic interest and partly as preparation for some further dis-
cussions below of relevance logic, relevantism, and the d. s. But
first a preliminary definition.

3.1 Admissibility. We say a rule is admissible in a given logistic

system if whenever its premisses are theorems, so is its conclusion.
From early on, relevance logicians have been interested in the
admissibility of the d. s. for various relevance logics. It needs
emphasising that the question is entirely non-trivial, for not only is
this rule not represented in relevance logics as an entailment
between premisses and conclusion; further, it is not even present
as a derivable rule. That is, in these systems there is no way to
start with""' A and A VB and then add enough axioms to permit a
derivation of B from those hypotheses (see AB75, §25.l). Well,
perhaps the claim of universal nontriviality is not entirely accurate;
for example, it was early and easily known that E and R (see § 1.1)
were closed under the d. s. at least for formulas not containing any
arrows, just because it was easy to show that E and R contained all
tautologies. But aside from such limited results, the general ques-
tion of the admissibility of the d. s. has always been hard. The
first substantial result was that of Meyer and Dunn (see AB75,
§ 25.2) which established that in fact the d. s. is admissible in E, R,
and other relevance logics as well. This result was particularly
important for E, as it was required to establish that E was equiva-
lent to the system of Ackermann [I], which AB75 credits as
providing the initial impetus for the entire enterprise of relevance
logic (narrow sense).

3.2 Entailment as a relation between quantified statements.

Anderson and Belnap [10] showed that the relational ideas of
tautological entailmenthood survive when first order quantifiers
are added. Further, it is easy to see that additional extensions to
the higher orders are natural and easy; though such an extension is
352 N.D. Belnap, Jr. and M. Dunn

nowhere in print (as such; cf. Meyer's work - mentioned in §3.3),

it is clear that there is just one way to carry it out, taking as a
guide that universal quantification shall be in some sense a general-
ization of conjunction. (Only one way, that is, except for possible
limitations on the principle of universal specification at the higher
orders. It may be that a relevance logician would wish to put on
this schema restrictions of certain kinds deriving from the same
intuitions as led initially to relevance logic itself; see Belnap [II] .)

3.3 Adding quantifiers to E and R. Anderson [12] added indivi-

dual quantifiers to E, and Belnap [13] added them to R. These
and other matters not yet published are to be reported in volume
II of Entailment, including the addition of propositional quan-
tifiers to the relevance logics. The work of Meyer, Dunn and
Leblanc [14] established that the d. s. is admissible also in the
presence of individual quantifiers, and Meyer [15] showed how
an argument could similarly be adduced for the admissibility of
the d. s. in higher order extensions of various relevance logics.
(With reference to the possible limitations of universal speci-
fication mentioned in the parenthetical remark of § 3.2, Meyer
showed that the admissibility of the d. s. continues to hold no
matter what limitations are set.)
Though (probably) not germane to our concern with the d. s.,
we nevertheless pause to mention a principal open problem for
relevance logicians. Fact: Rout1ey and Meyer [16] showed us how
to provide a sort of Kripke-like semantics for propositional
relevance logics, and offered completeness proofs. Their semantic
ideas extend to quantifiers in a natural way; but whether or not
completeness holds is irritatingly open.

3.4 Relevant arithmetic. The idea of relevant arithmetic is to

formalize the intuition that some pairs of arithmetical state-
ments may be not just materially or even modally connected,
but even relevantly connected, while others are not. For example,
the reader should be able to locate a point of view from which
(the impossibility) I = 2 can be seen to relevantly imply (the
impossibility) I =3, while the converse fails. Without further
pursuit, we simply report our opinion that Meyer [17] (there is
 Entailment and the disjunctive syllogism 353

also a lot of unpublished material) has found a stable formal

account of arithmetic relevance, an account which should be of
considerable interest even to the classicalist Gust as a classicalist
can be interested in whether or not a result is intuitionistically
available). In his work Meyer considers in effect four arithmetical
theories. p# and p## ('P' for Peano) are cast in purely extensional
terms, involving in their (common) language extensional connec-
tives, quantifiers in terpreted as ranging over the in tegers, identity,
zero, successor, addition, and multiplication. p# is the usual
Pea no system, including axioms for addition and multiplication, as
found say in Kleene [18]. P##, with the same axioms, uses in
place of standard induction instead 'infinite induction' (to infer
(x)A (x) from the infinitely many premisses A (n», which makes
p## coincident with (extensional) arithmetic truth. R# and R##
are the relevant theories of arithmetic. Their (common) language
includes th~ arrow of relevant implication, which figures in their
postulates in a most significant way. R# uses a form of induction
involving delicate use of the arrow, while R## uses infinite
induction.
Meyer has produced interesting proofs of the absolute consis-
tency of R# and R##, but those are not our interest here. Instead
we are concerned with questions on the side of completeness.
Because of infinite induction, it is easy to see that R## is 'exten-
sionally complete' in the sense of containing p##; and the usual
Codel point of view suffices to establish that that R# is not in this
sense extensionally complete. The question arises, though, as to
whether R# is as extensionally complete as p# - i.e., does R#
contain p#? It is easy to prove the axioms of P#, but of course
since R# is a relevant theory, the d. s. (or, equivalently detach-
ment for the material conditional) is not available as a primitive
rule, so that one cannot conclude with any ease to the provability
of all the theorems of p#. And efforts to establish the admissi-
bility of the d. s. for R# have so far failed, although Meyer has a
proof that it holds for R ##. In lieu of the d. s. for R # , Meyer has
proved each and everyone of the theorems of p# which are
proved in Kleene [18], so that anyhow it is known that R# is
'Kleene-complete: In the course of this effort Meyer found some
of Kleene's proofs to be relevantly invalid and so not available in
R#; in these cases he was able to find substitute proofs which were
354 N.D. Belnap. Jr. and M. Dunn

indeed relevantly valid. We want to note that although the con-

clusions of these proofs were purely extensional, the proofs
themselves typically take detours through the arrow of relevant
implication.

4. RELEVANT ARGUMENTS FOR THE ADMISSIBILITY OF THE d. s.

Relevance logicians have used fonnal or formalizable arguments in

order to establish various facts about relevance logics; can such
facts be established by using only arguments which the relevantist
takes as valid? This question, which relevance logicians have dis-
cussed amongst themselves for a number of years, without, let it
be said, making much headway, has again been posed in conver-
sation in most strenuous tenns by Kripke in the following guise:
one of the principal results concerning relevance logics is that for
the central cases, the d. s. is an admissible rule. For example,
whenever -A and AVB are theorems of the system E, (which
includes the arrow as an object sign), so is B. Kripke asks: is there
a proof of this fact which does not itself use the disjunctive syllo-
gism in the course of the proof?
The question is a complicated one; we cannot hope to answer
it here, but only to explain a bit just why it is so complicated, and
to clarify it some.
There are several choices which complicate the question,
choices we take up in turn.

4.1 Readings. There is a question as to whether the various natural

language expressions occuring in the various arguments and con-
clusions in the work on the admissibility of the d. s. shall be given
(a) wholly extensional or (b) partly relevant readings - the latter
in the sense that the relevant arrow shall be used in translating at
least some of the English conditionals which occur (explicitly or
implicitly). There appear to us to be four important possibilities.
I. The conclusion of the argument and furthermore all the
language in the argument itself shall be given a purely extensional
reading. The conclusion, on this reading, is that either . . . .A is not
an E-theorem, or AVB is not an E-theorem, or B is an E-theorem;
this we call ex tensional admissibility. And we have added as part
 Entailment and the disjunctive syllogism 355

of this option that the entire proof be given an extensional gloss.

Under this option it seems to us straightforward that none of any-
one's arguments for the d. s. are relevantly valid. But this doesn't
have much to do with the fact that it is the admissibility of the
d. s. which is being proved, as the following example shows. Let
all natural language constructions be given their 'usual' extensional
reading. Now suppose someone argues 'If A then B; but A; so B.'
He is quite evidently employing the d. s. (since the first premiss is
a material 'implication') and hence is not arguing relevantly. He
may be (under this construal) a relevance logician, but he is no
relevan tist.
We take it that the previous investigations of Meyer into R#
indicate that we have not uncovered some surprising scandal;
typically arguments for purely extensional theorems of R# need
to take a detour through relevant connectives. If one had tried to
argue for, say, the Kleene-theorems using only extensional ideas,
one would have been forced into using relevantly bad arguments.
So much for this option.
2. The conclusion of the argument shall be given a relevant
reading. On this option, the admissibility of the d. s. is to be taken
as the statement that the joint E-theoremhood of .....A and AVE
relevantly implies the E-theoremhood of B; this we call relevant
admissibility. We know of no mathematical proof one way or the
other as to whether the d. s. is relevantly admissible in E, but
several relevance logicians have opinions. Meyer (correspondence),
for example, takes it that the d. s. is not relevantly admissible.
And after stating the third alternative, we give a reason why we
tend to agree, noting at this point that since on this option the
conclusion of the argument is relevant, it is certain that any proof
would also require relevant connectives.
3. Suppose we combine (1) and (2) in the following way. Like
(I), the conclusion of the argument shall be purely extensional:
either -A is not an E-theorem or AVB is not an E-theorem or B
is an E-theorem (extensional admissibility). But like (2), the proof
shall be allowed to involve relevant connectives. Thus we have the
typical case for R # -extensional conclusions requiring relevant
proofs. Obviously this option is a stand-in for a host of options,
since there are so many choices possible in giving readings to
various parts of the argument; there mayor may not be a relevant
356 ND. Belnap, Jr. and M. Dunn

reading of some extant proof of the d. s. which renders it entirely

acceptable to the relevantist. To this extent, we can only
commend the enterprise suggested by this option as interesting,
but still, we do have a guess.
Everyone of the extant proofs (Meyer and Dunn [ 19] , Routley
and Meyer [20], Meyer [21], AB75, §25.3) involves certain
pleasant 'theories'. One finds (never mind how) a theory T with
the E-theorems ""A and AVB in it and the non-E-theorem B (for
reductio) not in it. 2 The theory is sufficiently pleasant so that
"'A's being in it implies A is not, and that A VB's being in it implies
that either A or B is in. Now it might appear that the extant
proofs proceed by the d. s. to conclude that B is in - a contra-
diction.
If appearance is reality, then there are consequences for both
options (2) and (3). With respect to (2), the classicalist will have
to reject that relevant (but not extensional) admissibility has been
established by the extant proofs. For though the d. s. is classically
acceptable, even the classicalist knows that it cannot suffice to
establish a relevant connection. With respect to (3), if appearance
is reality, then the relevantist will have to reject that (even)
extensional admissibility has been relevantly established, for he
forbids all use of the d. s.
But is appearance reality? One matter up for grabs is whether
the 'or' in 'either A or B is in' is extensional or intensional ('A's
not being in relevantly implies B's being in'). If the 'or' were to be
intensionaL then clearly this part of the argument would serve as
no bar to a fully relevant proof of full relevant admissibility; but
after substantial (if not wholly conclusive) analysis of the extant
arguments, we report to the reader our view that in fact this 'or'
cannot be taken intensionally. But this does not settle the matter,
for even with an extensional 'or' the argument can be just slightly
restructured to avoid use of the d. s., and indeed it was originally
so structured in the extant proofs mentioned above. 3 It is, as we
noted, a reductio, for which any absurdity will do. And it is easy
to see that although relevant moves will not produce 'B is in and
not in' without the d. s., it is straightforward to get 'either B is in
and not in, or A is in and not in' - which is enough. But at this
point there is a divergence between options (2) and (3), laid bare
by asking, enough for what? To make the point, let us suppose
 Entailment and the disjunctive syllogism 357

(without doxastic commitment) that all the other parts of the

extant arguments are relevantly acceptable. Then this reductio
argument would in fact suffice to establish in a relevant way the
extensional form of admissibility, but not the relevant form.
Upshot: we think the extant proofs do not give a proof of relevant
admissibility; but they might turn out to give a relevant proof of
ex tensional admissibility.
4. The fourth option is to conclude that the expressive power of
the languages so far in vestigated by relevance logicians is not
enough to handle their arguments in a relevant way, but that
instead additional features must be added.
One possibility for translating the extant arguments lies in
Boolean negation (§ 2). By a classical-relevantist let us mean one
who takes as an organon the logics with both Boolean and de
Morgan negation (see § 2 above) . Now in contexts in which neither
arrow nor de Morgan negation appears, i.e., contexts involving
only Boolean negation and positive extensional connectives,
including quantifiers, it is trivial that there is no difference
between the classicalist and the classical-relevantist. It follows that
if we take a proof of the admissibility of the d. s. and translate
everything, conclusion and argument alike, extensionally, but with
(only) Boolean negation, the argument is bound to be classical-
relevantly valid since classically valid. So that in the presence of
Boolean negation and the foregoing policy, the whole question of
the provability of the admissibility of the d. s. in a relevant way is
trivialized and cannot be sensibly asked. We must rule out Boolean
negation to make the question interesting, even for the c1assicalist.
(The relevantist, as we suggested, in §2, doesn't recognize Boolean
'negation' anyhow.)
A last possibility is to introduce some logical apparatus which is
both relevant and of use. We don't have any firm candidates for
this role; still with all hesitation, we mention the possible relevant
usefulness of some forms of restricted quantification, which
should be generalizations of conjunction and disjunction. (Neither
(x)( - Ax VEx) nor (x )(Ax ~ Ex) is such a generalization, and simi-
larly for existential quantification). This seems to us an important
line to pursue; but we cannot here take the space to follow up
this suggestion, beyond indicating what the d. s. itself might come
to on this reading: consider the space of all formulas B such that
358 N.D. Belnap, Jr. and M. Dunn

for some A, -A and AV Bare E-theorems; within this restricted

space, everything is an E-theorem.

4.2 'E'quilJalent' forms. Let the d. s. be stated materially, in accor-

dance with option (3) of §4.1. Still not all is settled, for what it
means will depend on what is meant by 'E-theoremhood.' In a way
perhaps this is noncontroversial, but why it is worth mentioning is
that there may be differing accounts of E-theoremhood which can
be shown to be provably materially 'equivalent'. Now suppose we
have a relevantly valid proof of the admissibility of the d. s. under
one of these accounts of E-theoremhood. In the absence of the
d. s. itself, it is clear that we cannot use that proof as the front end
of a proof of the admissibility of the d. s. for a materially 'equiva-
lent' account of E-theoremhood.
Perhaps the most important example of this phenomenon is due
to having both a syntax and a semantics for E, and to having a
proof that (in the usual terminology) E-theoremhood on the syn-
tactic side is 'equivalent' to E-validity on the semantic side
(Routley and Meyer [22], [20]). It could therefore be that at
some point in time someone will provide a relevant proof of the
d. s. for E-theoremhood, but not a proof of the d. s. for E-validity,
even given the proven 'equivalence' of E-validity and E-theorem-
hood. Another important example would emerge if someone were
to produce a Gentzen system for E - perhaps especially interest-
ing because one might thereby be led to a constructive proof of
the admissibility of the d. s., and from there perhaps even to a
relevant proof. See AB75 for partial Gentzen systems.
In what follows we decide to concentrate on the presently
available syntactic version of E-theoremhood. And we note that,
in the usual Godel way, we can represent E-theoremhood in the
vocabulary of p# and hence in R#. So we can find, in the purely
extensional language of arithmetic, a sentence which can be read
as a formulation of the d. s. - let's call this arithmetic sentence
DS#. (Note: the relevantist might already object to some of these
classical moves; we ignore this possibility, but not because it's not
a real one.) We are thereby led to the following surprisingly
definite question: is DS# provable in R#? We just don't have
much information about this. DS# may be like some of Kleene's
arithmetic theorems, which are provable in R# even though
 Entailment and the disjunctil'e syllogism 359

Kleene's own arguments for them are relevantly invalid. But the
analogy can't really be close in any straightforward sense, because
the extant proofs not only involve steps raising the question of
relevance; they also all involve second order considerations -
quantification over sets of sentences (theories). By so much the
arguments cannot be merely 'relevantized' to become available in
Rlit (which is first-order). We return to this point. The second
alternative is that DS# may be unprovable in R#. In the latter
case, DS# may even be unprovable in p# because, perhaps, of the
second-order considerations mentioned above; but that would not
be of as much interest as if it turned out that DS# were provable
in p# but not in R#.5 Of course, as we said, we don't know
whether or not p# is wholly included in R#; but if it is not, it
would somehow be striking if the arithmetic statement of the d. s.
itself were a witness to the fact. We must, however, record the
sober guess that DS# is not provable even in P#, since (as we said
above) the extant proofs of the admissibility of the d. s. involve
second order considerations. But who knows, the picture might
change again for second order R #: it could be that the extant
arguments can be carried out in second order R # , or it could even
be that DS# is provable in second order p# without being so in
R# .5

4.3. The foregoing makes little mention of the difference between

a classicalist and a relevantist, but our next point relies on this
distinction. Consider R## and p## instead of R# and p#. Of
course any classicalist will believe that DS# is provable in R#,
since he will believe that it's true, and hence in P#, and that
infinite induction gets p## inside R##. So he could come to
believe that DS# is relevantly provable with infinite induction (in
R##), but not with only finite induction (in R#). On the other
hand, the true relevantist could not use this argument if the lemma
on which it depends is that p## is only materially 'contained' in
R##, for the conclusion that DS# is inside R## would then
involve the d. s. (Furthermore, the relevantist would want to have
a look at the proof that p## is, even materially, inside R ##; but
that is just another level of the same kind of complexity.)
Let us put to one side, now, the question of how one has
arrived at the admissibility of the d. s., and just suppose, now, that
360 N.D. Belnap, Jr. and M. Dunn

one believes that never -A and A VB provable in E while B not. An

interesting phenomenon arises: if one is a classicalist, being given a
provable -A and a provable AV B, one will not hesitate to infer the
provability of B. But of course the relevantist cannot do this, for
to so infer would be precisely to employ the d. s.! Which brings us
to our next (and last) topic.

S. THE PHENOMENOLOGY OF RELEV ANTISM

What we are after in this section might be called 'the phenomen-

ology of relevantism,' at least in the vicinity of the d. s. We do not
at this point want to argue for relevantism, but only to indicate
some features of the relevantist situation as best we can.
To be a little concrete, we ask you to imagine a relevantist who
accepts a proof of the 'extensional admissibility' of the d. s. as
described at the end of the last section, and who has also got
proofs in E of some -A and AVB. He knows, as a true relevant-
ist, that he cannot infer the E-provability of B. So what does he
say to himself?

5.1 I'm alright, Jack. One might think as follows. The point of
relevantism is taking seriously the threat of contradiction. But
there is in this vicinity (that of fairly low level mathematics) no
real such threat. So here it's O.K. to use the d. s. and conclude
the theoremhood of B. That sounds O.K., but is it? After all, we
suppose that 'here there is no threat of contradiction' is to be con-
strued as an added premiss. But a little thought shows that no such
added premiss should permit the relevantist to use the d. s., for a
very simple reason: as we said, avoidance of the d. s. was bound up
with the l:hreat of contradiction, and one thing that is clear is
that adding premisses cannot possible reduce that threat.lf in fact
the body of information from which one is inferring is contra-
dictory, then it surely doesn't help to add as an extra premiss that
it is not. That way lies madness.
We take the opportunity to point out a disanalogy between the
relevantist situation and the intuitionist situation. The intuitionist
while overhearing with dismay the meanderings of some classicalist
can always say: 'Poor fellow! He actually thinks he is reasoning.
 Entailment and the disjunctive syllogism 361

Still, there is some sense that can be made of his musings. What he
seems to be doing is assuming (without warrant) a bunch of
excluded middles. So I can charitably interpret him as con-
structing an enthymematic argument which can be made
(intuitionistically) correct by adding the appropriate excluded
middles as premisses.' The relevantist, as we have seen, cannot
make an analogous charitable interpretation.

5.2 The relevantist/deductivist parallel. One is reminded of

induction: if induction is risky to begin with, it does no good to
add as an extra premiss that here after all it is all right to use it. We
all know about the circles we wind up in if we start out in that
direction. So in this, and the next couple of subsections, we want
to consider a parallel between the relevantist and the deductivist.
The deductivist, as the reader will recall, is that hero of elementary
logic texts who says that the only correct arguments are deduc-
tively valid arguments. We suppose the deductivist must have his
moments when he wonders just what it is that seduces those other
poor fools (and himself as well on occasion) to use inductive prin-
ciples - is it just blind irrationalism, or can some justification be
given? Both the deductivist and the relevantist have set themselves
severe standards of personal conduct with regards their reasoning,
and both should feel the need to explain, or at least apologize for,
any lapses from these standards.

5.3 The leap of faith. One tack that has been taken with respect
to induction is just to count it as involving some judgment (not
to be construed as an extra - useless - premiss) that in the par-
ticular case induction is appropriate, and then the making of a
leap to the conclusion, a leap known to be risky, a leap based
perhaps partly on faith as well asjudgment.
Perhaps the relevantist could or should take a precisely parallel
tack, given his proof of the extensional admissibility of the d. s.,
and given the E-theoremhood of -A and A VB. He cannot and
must not count the inference to the E-theoremhood of B as 'good
logic,' but perhaps he could judge that nevertheless it would be
appropriate to leap to the E-theoremhood of B anyhow, even
though (because of the threat of contradiction) he knows the leap
to be risky and hence based partly on faith.
361 N.D. Belnap, Jr. and M. Dunn

5.4 The toe in the water. There is another option, which like the
foregoing pictures the relevantist as at least tempted to jump to
the E-theoremhood of B from the E-theoremhood of --A and
AVB, but with a great deal of hesitancy. This option involves just
a little technicality. Observe first that although the inference
(6) (of § I), from """p and (PVq) to q, is relevantly abhorrent, there
is nothing wrong with the inference (3) (also of § 1), from ""P and
(PVq) to qV(p&""p) - i.e., to 'q, unless there is something awfully
wrong in our information about p.'
This suggests the technical maneuver of introducing a propo-
sitional constant f, to be interpreted (in this context - see AB75,
§ 27.1.2) as the generalized disjunction of all contradictions (given
propositional quantifiers, f could be defined as 3p(P&""p». Then
it is easy to see that

(3') ""p&(PVq) -1- qVf

is relevantly acceptable - from ,....'P and pVq to infer: q unless

we've got ourselves a contradiction.
Given f, there is then available to the relevantist a sort of copy-
cat procedure: whenever the classicalist infers q, the relevantist
infers qVf instead. (It turns out, as is also easy to see, that ad-
ditional uses of the d. s. do not require somehow more and more
f's - one f will do the trick for the whole argument.) The relevant-
ist thus can give a charitable interpretation of the classicalist's
'reasoning' that is du:.l to the intutionist's interpretation. The
relevantist can say of the classicalist: 'Poor incautious fellow! He
concludes that q outright, when what he should conclude is
merely that q unless his information is all screwed up.'
Such a relevantist might go further and come to see himself as
employing a special speech act, that of 'disjoining/,sotto voce; so
that every time the relevantist was heard to assert 'B is an E-
thorem' (say), he would be understood as having added 'or else
there's real trouble.' One could certainly do so without collapsing
the meanings of q and qVf, just as one can presume that a claim to
warranted belief is part of the speech act of assertion without
being part of the meaning of what is asserted. (The OED might let
us call our new speech act of diffident assertion, 'diffidation.')
One here too is reminded of induction, because there are people
 Entailment and the disjunctive syllogism 363

of basically deductivist persuasion who urge that one should never

conclude by inductive argument something like 'All crows are
black' - all that one is really entitled to conclude is something
like 'probably all crows are black' (or maybe, 'it is probable,
relative to my data, that all crows are black').
But it seems to us in the end that the ploy of diffident assertion
is a pretty shabby one for the true relevantist to use, since on his
own account the inference from qV/ to q is invalid (it is precisely
equivalent to the d. s., as it turns out) - so that the practice of
such a speech act would conceal what are for him real differences.
Furthermore, in the actual case the relevantist generally has more
information than a barren disjoined /; he knows, if he has done
his homework, which contradiction is at issue. Of course this point
will not interest the c1assicalist, who cannot tell the difference
between contradictions. But the relevantist can, so that for him
using/, whether suppressed or not, is to lose information.
We conclude that this option has demerits without compen-
sating advantages, except, perhaps for the profoundly dubious
values of dissembling. It is not so easy to tell apart a classicalist
and the sotto voce relevantist we have described. For this reason,
someone might want to claim to be really a relevantist after all,
even though having no qualms about using the d. s. Such a
person's defense of the apparent inconsistency in such a policy
might be that he is not after all using the d. s., but only its rele-
vantly valid cousin with disjoined t, coupled with the practice of
systematically suppressing such disjunctions.
We think such a position is probably coherent, but we certainly
do not admire it. It smacks of waffling, wavering, backsliding, and
similar characteristics not to be encouraged.

5.5 The true rele van tist. The final option which we describe is the
one to which we have become more and more attracted as this
paper has emerged: the true relevantist should not at all even be
tempted to use the d. s. After all, the temptation presumably
comes from continuing to take pVq as some kind of logical link
between p and q - perhaps a weak one, but still more than
nothing. Perhaps the true relevantist should just stop before he
starts, declaring pVq to be no link at all.
Apply this to the case at hand. The relevantist so described
364 N.D. Belnap, Jr. and M. Dunn

might be interested in 'extensional admissibility" But not at all

because of hoping to be able to use it as some kind of major
premiss once he has got an E-provable . . . . A and an E-provable
A VB. We grant it is hard then to see why he would be interested -
that is the point of the might.
And we are then led to our last thought. We do think 'admis-
sibility' has some kind of an 'if-then' in it. So if what we have
heretofore called 'extensional admissibility' does not, then what is
needed is a new theorem - not just a relevant proof of an old one.
Perhaps it would be a new theorem using restricted quantification
as described in §4.1 (4). But however that comes out, this true
relevantist has to look around the logical landscape and say, not
just that he has not seen a relevant proof of the admissibility of
the d. s., but that he has not even seen a bad proof of it: Of
course such a claim would outrage a c1assicalist; but Our Hero
should not let that bother him.
In any event, we applaud the true relevantist as described
here. In contrast to the old-fashioned logical empiricists and the
new-fashioned nominalists and such, the true relevantist is truly
toughminded, with nary a soft spot in his head. His brow wrinkles,
his jaw juts, and he will never, never use the d. s.

NOTES

I. For example Geach, in a recent review in the Times Literary Supplement

of a book in philosophical logic dealing almost entirely with other matters,
chastises his author for 'bad taste and judgment' in briefly suggesting a
use for AB75-style entailment; Geach's grounds are that it counts the dis-
j uncti ve syllogism as in valid!
2. Act ually the extant proofs work from the assumption that -B is in T. We
have stated a weaker assumption since it is clearer that it can have a rele-
vant proof (thOUgh Meyer has written us he thinks the stronger assum-
ption can as well, and we have no reason to doubt him).
3. Modulo the small difference mentioned in note 2. We credit Meyer for
having corrected a gaffe in our fIrst draft regarding the form of the extant
proofs.
4. In a communication received after completing this paper, Meyer pointed
out that DS# is a 'secondary unequation,' and so, by results recorded in his
unpublished work, cannot be in p# without being in R #.
5. Meyer's results about secondary unequations mentioned in the previous
note do not, as far as we know, extend in any immediate way to second
order R#.
 Entailment and the disjunctive syllogism 365

BIBLIOGRAPHY

Ackerman, W. [1] Begrtindung einer strengen Implikation. The Journal of

Symbolic Logic 21 (1956): 113-128.
Andersen, A.R. [12] Completeness Theorems for the Systems E of Entailment
and EQ of Entailment with Quantification. Technical Report, No.6. Con·
tract No. SAR/Nonr-609 (16), Office of Naval Research, New Haven,
1959. Reprinted in Zeitschrift fur mathematische Logik und Grundlagen
der Mathematik 6 (1960): 201-216.
and Belnap, N.D., lr. [10] First Degree Entailments. Technical Report,
No. 10. Contract No. SAR/Nonr-609 (16), Office of Naval Research,
New Haven. Reprinted in Mathematischen Annalen 149 (1963): 302-
319.
- and Belnap, N.D., lr. [2] Entailment: The logic of relevance and necessity,
Volume 1. Princeton University Press, 1975.
Belnap, N.D., Jr. [7] A Useful Four-valued Logic. In 1.M. Dunn and G. Epstein.
Eds., Modern Uses of Multiple-valued Logic. Dordrecht: D. Reidel, 1977.
- [8] How a Computer Should Think. In G. Ryle, Ed., Contemporary
Aspects of Philosophy, pp. 30--55. Oriel Press; 1977.
-- [11] Modal and Relevance Logics: 1977. In E. Agazzi, Ed., Modern Logic
- a Survey. Dordrecht: D. Reidel, 1981.
- [13] Intensional Models for First Degree Formulas. The Joumal of Sym-
bolic Logic 32 (1967): 1-22.
- [23] The Fonnalization of Entailment. University doctoral dissertation,
Yale 1959.
- [24] Tonk, Plonk and Plink. Analysis 22 (1962): 130-134.
Dunn, J.M. [6] Intuitive Semantics for First-<legree Entailments and Coupled
Trees. Philosophical Studies 29 (1976): 149-168.
Kleene, S.C. [18] Introduction to Metamathematics. New York: Van Nos-
trand,1952.
Kripke, S.A. [3] Semantical Analysis of Intuitionistic Logic I. In 1.N. Cros-
sley and M.A.E. Dummett, Eds., Fonnal Systems and Recursive
Functions, pp. 92-129. Amsterdam: North-Holland, 1965.
Meyer, R.K. [5] New Axiomatics for Relevant Logics, I. Journal of Philo-
sophical Logic 3 (1974): 53-86.
- [17] Relevant Arithmetic. Bulletin of the Section of Logic (Polish
Academy of Sciences, Institute of Philosophy and Sociology, Wroctaw),
5(1976): 133-137.
- [IS] Ackermann, Takeuti, und Schnitt: For Higher-order Relevant Logics.
Bulletin of the Section of Logic (Polish Academy of Sciences, Institute
of Philosophy and Sociology, Wroctaw), 5 (1976): 138-144.
- [21] Coherence Revisited. Forthcoming.
- [9] A Boolean-valued Semantics for R. Forthcoming.
- and Dunn, 1.M. [19] E, R, and 'Y. The Journal of Symbolic Logic 34
(1969): 460-474.
366 N.D. Belnap, Jr. and M. Dunn

- Dunn, 1.M., and Leblanc, H. (14] Completeness of Relevant Quantification

Theories. Notre Dame Journal of Fonnal Logic 15 (1974): 97-121.
"- and Routley, R. [4] Classical Relevant Logics. Studio Logica 32 (1973):
51-66.
Routley, R., and Meyer, R.K. [16] The Semantics of Entailment. In H.
Leblanc, Ed., Truth, Syntax and Modality. Proceedings of the Temple
University Conference on Alternative Semantics, Amsterdam: North-
Holland. 1973.
- and Meyer, R.K. [22) The Semantics of Entailment III. Journal of Philo-
sophical Logic I (1972): 192 -208.
- and Meyer, R.K. [20] The Semantics of Entailment IV: E, n' and n The
Journal oj'Symbolic Logic, forthcoming.
Choice, chance, and credence
RICHARD JEFFREY
Princeton University

This is an examination of certain lines of thought on the role of

probability in theoretical and practical reasoning, with no sugges-
tion that precisely the items noticed here are the most important.

1. RELIABLE COMPOUNDS OF UNRELIABLE INDICATORS

To set the scene for some of the subsequent action, let us consider
how (in especially ~lmple cases) truth functional compounds
of unreliable observation reports E j can be more reliable than
anyone is, separately. (This bears on Mary Hesse's [1] 'self-
correcting observation language', which she discusses further in
[2],pp.125-131.)
Suppose that we have n independent reports ±E 1 , ••• , ±En,
each of which is a 'yes' or 'no' answer to the same question,
'Is it the case that H?' (H might say that the accused struck the
plaintiff, or that there was a brown precipitate from a certain
solution, etc.) Suppose, too, that each report has the same prob-
ability a of being positive when H is true, and has the same,
smaller probability (3 of being positive when H is false. Example:
O! = 80% and (3 = 51 %. For any truth functional compound of
these reports, e.g., (E 1 • E 2 )v(E 3 • E 4 ), we have a probability
polynomial hex) which gives the probability that the compound
is true when x is the common probability of truth of the Ej, e.g.
for the present compound, hex) = 2X2 - X4, which maps the unit
interval onto itself with fixed points hex) = x at x = 0, x = .62, and
x = 1: see Figure 1 (adapted from Moore and Shannon [3], p. 194,
Figure 3). This mapping x 4 hex) moves points in the unit interval

Contemporary philosophy. A new survey. Vol. 1, pp. 367-386.

© 1981, Martinus Nijhoff Publishers, The Hague/Boston/London.
368 R. Jeffrey

away from the middle fixed point, toward the other two, in such
a way as to make the truth value of the compound be a more
reliable indicator of the truth value of H than is the truth value
of any of the Ej, for with H true the probability that the com-
pound is true is h(cx) = 87% instead of 0: = 80%, and with H false
the probability that the compound is false is l-h(~) = 55% instead
of 1-~ = 49%.
In general, reliability is increased when (as in Figure I) we have
h(o:) > 0: and hem < ~ so that the compound is more likely than
any of the E j to have the same truth value as H, be it truth or
falsity. Moore and Shannon [3] show that this sort of reliability
enhancement is obtainable to any degree short of the ideal step
function simply by using enough indicators E j in appropriate
ways, and they show that on the whole, the Utst compounds are
what they call 'quorum functions' of the Ej, i.e., truth functions
of n arguments where for some m between 0 and n, the compound
is true/false if more/fewer than m of its arguments are true.
(See their theorem 2, being watchful for absent-minded inter-
changes of greater and less in the proof on pp. 203-204.) Quorum
functions are optimal in another sense, too: they correspond to
likelihood ratio tests, viz., here, Bayes strategies (see e.g. Chernoff
and Moses [4], appendices Ell and E l2 .)
Hesse ([2], p. 125) accepts the view of c.1. Lewis ([5], Ch. X
and [6], p. 186) that we cannot even have probable knowledge

Figure 1. Probability function for (E,.E,)v(E,.E.)

Figure 2. h(x) for a quorum function with large n.

 Choice, chance, and credence 369

of empirical matters in the absence of certainty about observation

(reports), but she differs from Lewis in identifying the certain
reports not as 'terminating judgements' about sense data, but
as reports ±Ej about uncertain reports H, viz., ±Ej = 'Report
number i asserts/denies that H' where H is an ordinary sort of
observation report.
It is noteworthy that in cases where the E j would ordinarily be
viewed as indicators of H, a and {3 are well enough separated so
that their precise values are unimportant if we have enough inde-
pendent reports to work with: see Figure 2. A quorum function
with large n and middling m then serves to transform (near-)
certainties about what has been reported into near-certainty about
the matter reported upon. But let us see just how that happens.
Where Moore and Shannon interpret a and {3 objectivistically,
in terms of a physical (or objective) probability function Po Hesse
interprets them epistemically, in terms of a rational credence
function cr. We can represent the situation by thinking of p as
a mapping of possible stages s in the development of the world
into probability functions Ps' An unknown member r of the set
of all such possible stages is currently real, and so the real prob-
ability function Pr is the value that p assumes at r. The assumption
that the real chance PrEj of Ej (= the real propensity for Ej to be
true) depends only upon whether H is true or false is represented
by the fact that PsEj is ex at stages s at which H is true, and is {3
where H is false. (Then H is an S-determinate statement, i.e., it
has a definite truth value at each stage, independently of how the
future development of the world may go.) The uncertainty about
whether PrEj is a or {3 is thus reduced to an uncertainty about the
location of r within S. (The statement Ei need not be S-determin-
ate.) Then the Moore-Shannon assumption is this:

ani if H is true at s . .
0.1) Ps(Eil ....• Ej )={ ., (11 < ... < 1m)
m {3m if H IS false at s

On the other hand, Hesse would interpret a and (3 in terms of

conditional independence relative to rational credence:

0.2)
370 R. Jeffrey

Now let E be a quorum function of Ej, ... ,En with an h of

the general sort shown in Figure 2. Then cr( -E/H) = 1 - h(a) ;: : : 0,
so that cr( - E' H) ;: : : 0, and similarly cr(E/---H) = h(IJ) ;:.: : 0, so that
cr{E'-H) ;:.: : 0. Then credence in the symmetric difference (-E'H)
v(E'-H) is effectively 0, i.e., credence in the biconditional (E'H)
v(~E'~H) is effectively I, i.e., as close as makes not odds, E and H
are thought to be interchangeable salva veritate. Thus in simple
cases, where (1.2) holds, quorum functions of unreliable obser-
vation statements E j form highly reliable observation statements
concerning the original question, viz., 'H or ~d?'.
Where H is the proposition that the accused struck the plaintiff,
one might think that the objective chance PrH of H must be 1 or
depending on whether H is true or false at stage s, even though one
°
does not know that truth value. In such a case one might have
nonextreme credence (or subjective or personal probability)
crH in H, e.g., one might be perfectly clear that the two sides of
a bet on H would be equally attractive at odds of (say) 3:2, so that
crH = 3(5. Here Cr is one's real credence. We need not consider
what sense can be made of Cs for arbitrary stages s.

2. CHANCE AND CREDENCE

Statements about the past and present are S-determinate, i.e., they
have definite truth values at each stage s of each possible world W,
regardless of how W may evolve after stage s. On the other hand,
future contingent statements need to be evaluated at stages of
worlds: at the present stage of the actual world it may be false
that there will be a female President of the United States in the
twentieth century, but if that statement is genuinely contingent,
there will he a different possible world, agreeing with the real one
up to the present stage, but differing thereafter so that in the
present stage of that world, there is a female twentieth century
President of the United States. Then future contingencies are {r}-
indeterminate: such a statement may have one truth value at
stage r of world W, but have the other truth value at the same
stage, r, of a different world, X. In a third category are the eternal
statements, i.e., statements whose truth values depend on worlds
but not stages: E is eternal iff the truth value of E at stage s of
 Choice, chance, and credence 371

world W is the same as its truth value at stage t of world W, for

any possible world Wand any stages of sand t of W.
Let us now use 'E', 'H', etc. to denote propositions instead of
statements. Propositions will be understood to be sets of pairs
(s, W) where s is a stage and W is a world of which s is a stage.
(Stages are cumulative: think of s as the complete history of W,
up to a certain point.) Now H will be the proposition consisting
of those pairs (s, W) at which the statement formerly denoted by
'If is true.
To round the account out, we take S to be the (nonempty)
field of a forest relation L ('begins'), i.e., an irreflexive, transitive
relation that is nonconvergent in the sense that if each of two
distinct stages begins a third, then one of the two must begin the
other: if s =1= t and sLu and tLu then sLt or tLs. (The forest may
consist of a single tree, i.e., the relation L may be rooted, in the
sense that if neither of two stages begins the other, some (third)
stage must begin them both.) A possible world W will be a subset
of S that is linearly ordered by L (Le., L holds in one direction or
the other between any two distinct members of W), and contains
all initial stages of its members (i.e.), SEW if sLt€W). But not every
such W need count as a possible world. Those that do, comprise
the set 1f'". Then the propositions are the members of some field
of subsets of the set n = {(s,W):S€W€1f'"}, and it is propositions
that will be the arguments of credence Cr and chance Ps. Here s
may be any member of S but r is fixed. Notionally, r is the current
stage of whatever member R of if"" it is, that corresponds (time-
lessly) to reality. To say that H is true at stage s of world W is to
say that (s,W)€H. To say that H is (simply) true is to say that
(r,R)€H. But here 'r' and 'R' serve as extralogical constants.
We use '[ = PsE] , etc. as shorthand for the S-determinate propo-
sition that the chance of E equals (etc.) what it is at stage s:

(2.1) [=PsE] = {(t,W):t€W & ptE = PsE}

[~PsEJ = {(t,W):tEW & ptE ~ PsE}

We suppose that these are propositions whenever E is. (If PsE

should be undefined for a particular proposition E and stage s,
then [=PsEJ = [~PsEJ = ¢.) Now if PsE is defined for all s, it
would seem that crE should also be defined, and be related to
p as follows:
372 R. Jeffrey

Where PsE assumes only countably many distinct values as s ranges

over S (the discrete case), the cr-mean of PsE will be the sum over
the partitioning {[=PsE] :S€S}, of the products PsEcr[=PsEl, e.g.
in the Moore-Shannon case the cr-mean of PsE j is acrH + (kr - H.
In general, the right-hand side of (2.2) is given by the abstract
Lebesgue integral fsE'S PsE dcr(s), but we shall generally discuss the
discrete case, leaving it to the interested reader to reformulate
matters in the general case.
Given that cr satisfied the basic probability axioms (Kolmogorov
[71 I, § I ) and is defined on the propositions [=PsE] , it follows (in
the discrete case) that crE is the cr-mean of the random variable
clEf [=PsE]). To get from that to (2.2) is sufficient to have

This strikes me as a fair copy into present notation of what Hacking

([ 8], p. 135) described as follows (and called, or miscalled, 'the
frequency principle'):

One principle about support and chance seems so universally

to be accepted that it is hardlY ever stated. It is not so much
a principle as a convention to which everybody is firmly
wedded. Roughly, if all we know are chances, then the
known chances measure degrees of support. That is, if all
we know is that the chance of E on trials of kind K is p,
then our knowledge supports to degree p the proposition
that E will occur on some designated trial of kind K.

By 'support' Hacking means rational credence. I interpret Hacking's

'K' as my's', so that PsE as in (2.3) is Hacking's chance oj E on
trials of kind K. Mellor ([9], p. 67) seems to dispute this sort of
interpretation of Hacking; but see Hacking ([ 10], § 4.) Braithwaite
[11] states and defends essentially the same principle. Miller [12]
offers a booby trapped version of it, from which he deduces the
shattering conclusion cr = c p in one step. (In effect, Miller sets
s = r in (2.3) and argues that as [=PrEl surely contains r, we must
have cr[=PrE] = 1. See Jeffrey [13] for fuller discussion.)
 Choice, chance, and credence 373

David Lewis (private communication) points out that (2.3)

becomes equivalent to (2.2) in the presence of the condition
that chances aren't chancy, i.e., in the discrete case, (2.4) ptl=PsE]
= 1 or 0 depending on whether or not t € [=PsE] .

Proof We have already seen that (2.3) implies (2.2). Now observe
that by (2.2), crCE-[=PsE]) is the cr-mean ofpt(E-[=PsE]), which
(by (2.4) and the probability axioms for cr ) assumes the value
ptE for t in [=PsE] and assumes the value 0 for all other t. Then
cr(EI [=PsE]) = cr(E - [=PsE] )1 Cr [=PsE] = PsEc r [=PsE] ICr [=PsE] =
PsE, Q.E.D.

Fraasen [14], D. Lewis [15 Land Skyrms [16] treat the re-
lationship between chance and credence much as I have reported
here (but the present representation of propositions as sets of pairs
is drawn from Burgess [17 J with small modifications). The idea
(2.2) that credence ought to be the cr-mean of chance figured
basically in Camap's ([ 18], §41D) early formulation of his ideas
on induction, but Camap was then a frequentist about chance.
Good ([ 19J, p. 41) characterized chances as (subjective) prob-
abilities that 'are independent of any further experiments' and
freely applied (2.2): see also Good [20]. Braithwaite [21] inter-
prets de Finetti's ([ 22] , III) theorem on exchangeability via (2.2),
i.e. (from de Finetti's point of view) backwards. The 'mixed
Bayesianism' of Jeffrey [23] stems from these sources, but with
a different nonfrequentist 'objectivism' about chance. The possi-
bility of reducing chance to credence together with nonprobabil-
istic objective factors has been studied by Jeffrey ([ 24 ], § 12.7),
Mellor [9], Skyrms [25], and David Lewis [15 J. The view that
chances are propensities to produce appropriate relative frequencies
was Pierce's ([ 26], § 644) fallback position from frequentism -- a
position to which many have rallied since its adoption by Popper
[27], [28J. For a useful survey of the following 15 years of work
along such lines, see Kyburg [29].
374 R. Jeffrey

3. FREQUENCY, CHANCE, AND PURE SUBJECTIVISM

Such efforts to connect chance with credence are quite alien to

the uncompromising subjectivism of Bruno de Finetti, on whose
view 'objective chance' can only refer to the truth values (1 or 0)
of propositions, i.e., to something quite different from the magni-
tude discussed above. But of course de Finetti recognizes the
objective existence of truth frequencies in finite sequences E =
(E b • . . ,En) of propositions, and he is at pains to show how
credence concerning such frequencies is connected by the prob-
ability calculus with credence concerning elements of E. (He
postulates no principle that, like (2.3) goes beyond what can be
deduced from the basic probability axioms for cr.) The connection
between frequency and credence is this (paraphrased from de
Finetti [22], II):

The arithmetical mean of one's credences in the members of

E must equal the cr-mean of the truth frequency fs,wE.

Here fs,wE is l/n times the number of Ei that contain (s,W).

Analogously to (2.1), we abbreviate

(3.1) [=fs,wE] = {(t,X):t€X & ft,xE = fs,wE}

In analogy to (2.2) we have

That follows from the probability axioms. So does this:

(3.3) cr(Ed[=fs,wE]) = fs,wE, if cr(Ed[=fs,wE]) =

... = cr(En /[ =fs;w E]

It is a consequence of (3.2) or of (3.3) that de Finetti ([30] , p. 114)

applies in case

there are n events, m of which are known to have occurred

or to be going to occur, but we are ignorant of which, and it
is judged equally probable that anyone of the events should
 Choice, chance, and credence 375

occur. The only possible evaluation of the probability in this

case leads to the value ... mIn.

Here de Finetti seems to be citing this principle:

(3.4) If crE J = ... = crEn and cr[=fs,wE] =1 then crE j = fs,w E.

Similarly we might take Hacking ([ 8 J, p. 135 as above) to have

been citing not (2.3) but the following weaker consequence of
(2.2):

The frequency principles (3.2) -. (3.4) are provable on the assump-

tion that credence Cr obeys the basic probability axioms, but even
the weakest of the chance principles, (3.5), is a substantive assump-
tion about the relationship between chance and credence - an
assumption that may rule out certain credence functions as
unacceptable, even though they satisfy the basic probability
axioms.
Thus, suppose someone is sure that the chance of ace on the
i'th roll of a certain die is 1/6: cr[=PsEd =1 where PsE j = 1/6. Then
by (3.5) he cannot take himself to have additional information
that leads him to be sure that the ace will turn up after all, for
crE j must be 1/6, not I. Similarly, suppose someone is sure that
the chance of ace on each of the first two rolls of the dice is 1/6.
Then by (3.5) he must have credence 1/6 in each of El and E 2 , so
that by (3.4) he cannot have full credence in any of the three
possible frequencies of ace in the first two tosses, for if he had. he
would have to have the same number (0 or Y2 or 1, and not 1/6) as
his credence in E j •
We can also use the chance principles together with the assump-
tion that they hold for any credence function Cr that one might
reasonably have, to discover properties of chance. In particular,
let us define E is a proposition now knowable (i.e., knowable at
stage r) as follows:

(3.6) crE = I and cr[=PrE] "*

0 for some credence function Cr
that someone might reasonably have now.
376 R. Jeffrey

Using this definition, we can prove that

(3.7) PrE = 1 if E is a proposition now knowable.

Proof By (2.3), cr(E/[=PrE]) = PrE, whence by Bayes' theorem

'*
cr([=PrE]/E) = PrEcr[=PrE1/crE and, since crE = 1, cr[=PrE] =
PrEcr[=PrE] or, since cr[=PrE] 0, PrE = 1. Q.E.D.

By putting -E for E in (3.7) and defining

(3.8) V(s W E) = { 1 ~f (s,W)eE (seW)

" 0 If not

(so that V(s,W,E) is the truth value of E at stage s of world W) we

have

(3.9) PrE = V(r,W,E) if one of E, -E is knowable at reW.

(The knowable one, if either, will be the true one.) De Finetti's

thoroughly subjectivistic position is that this tells only part of
the truth, viz., that where E is {r}- determinate, PsE = V(r,W,E)
whenever reW: the whole truth is gotten by the further obser-
vation that where E is not {r}- determinate, PrE is undefined
(but one might define PrWE = V(r,W,E).)
The contrast between de Finetti's view and the eclectic view
sketched in § 2 is highlighted by the following two examples:
see de Finetti ([31], par. 11.4.4). From the eclectic point of
view these examples illustrate the fact that subjective probabilistic
independence is compatible with objective dependence (in the
P6lya example) and with objective independence (in the Bayes-
Laplace example).

A P6lya urn process. (Feller [32], p. 109. Kuipers [33], p. 48;

Johnson and Kotz [34], par. 4.2). Initially an urn contains one
black ball (color b) and one ball of another color (a), and balls are
drawn successively with duuble replacement, i.e., after each trial
the ball drawn is returned to the urn together with one more ball
of its color (and the contents are thoroughly stirred, etc.). Then
after a particular sequence of trials, n(s) in number, of which n(s,i)
 Choice, chance, and credence 377

have resulted in i-balls (i = a,b), the ratio of i-balls to all balls in

the urn (= the chance of drawing an i-ball next) is

(3.10) Ps(i next) = 12 : nd~~~) Ps (this stage is not final)

and the same equation will hold with 'Ps' replaced by 'cs' for one
who knows that the actual outcomes so far have been as in sand
that he is watching a P61ya process with double replacement.

A Bayes-Laplace situation. Here balls are drawn with single replace-

ment from an urn of fixed, unknown composition, for which
initially the (constant) chance of i's is thought equally likely to lie
in either of any two equal subintervals of the unit interval. Then
initially (at stage e), credence in the proposition that a particular
sequence s will be forthcoming is

(3.11) ce(s eventually)

= ce([=n(s)] eventually) fJ x n(s,a)(1_x)n{s,b)dx

where [=n(s)] is short for {(t,W):tEW & net) = n(s)}. Then Ce is

Camap's [1945] function c*, and at stage s credence ought to
be

. )
(3 . 12) Cs( 1 next = 21+n(s,i)
+ n(s) Cs (t
h'
IS stage
.
IS
f' 1)
not ma

for one who knows that the actual outcomes so far have been as
in s and that balls are being drawn with single replacement, and
for whom initially, credence about chance was uniform, so that
(3.11) holds.
In these examples we may take S to be the set of all 'words' on
the 'alphabet' A = {a,b}, i.e., S is the set of all finite sequences
of a's and b's, including the empty sequence, e, of length O. The
'product' st of two words is their concatenation, and e is the
identity element, i.e., se = es = s for all sinS. Define ~t as holding
iff su = t for some u in S (with sLt if u =1= e). Define W(s);:: {t: thS}
and set 11"= {- W(S):SES}.
Define the product of sets of words: BC = {st:sEB and t€C}.
378 R. Jeffrey

Then for sets of words, ¢ is the zero element (B¢ = ¢B = ¢) and

{e} is the identity element (B{e} = {e}B = B). Defining BO = {e}
and Bn+ I = BnB for all B<;:'S, An is seen to be the set of all words
of length n, and A* = A OUA1 UA 2 U ... = S. For further infor-
mation about the algebraic structure of A * , see papers in the Moore
[35] collection, e.g. Medvedev's, and §6 of Rabin and Scott's.
Let us now translate the arguments of the probability functions
in (3.10) and (3.11) from English into our sets-of-pairs notation.
In (3.11), 's eventually' denotes the eternal proposition {(t,W):
Sf W}, a nd similarly '[ =n( s) eventually], goes over as '{( t ,W) :
[=n(s)] . W =f ¢}' where again, 't' is a vacuous place-holder. In
(3.10) we have some latitude because (as David Lewis has pointed
out to me) we want p to satisfy the condition

(3.13) PsE = Ps([s] oE), where [s] = Df {(S,W):SfW}.

Thus, it makes no difference whether we take 'i next' in (3.10) to

denote the eternal proposition {(t,W):SifW} or the noneternal
proposition {(t,W):tiEW} or even {(s,W):siEW}, for the conjunc-
tion of anyone of these three with [sl is the third. The case is
similar for 'this stage is not final", of which the eternal form would
be {( t ,W): saEW or sbEW}. Observe that the idealized credence
function Cs (=Ps) of (3.12) happens to satisfy (3.13).
The main point of the P61ya/Bayes-Laplace comparison is that
rational credences in all propositions concerning outcomes are
precisel) the same in the two examples. From the eclectic point
of view, this shows that we need to speak of chance as well as
credence, to give a full account of either example; but from
de Finetti's point of view the same fact shows that we leave
nothing out, when we take no account of chance. (If we expand
the domain of cr " to include propositons about the composition
of the urn it is reasonable to expect differences between the two
examples to appear in subjective probabilities of outcomes con-
ditionally upon hypotheses about composition, but as compo-
sitions are not chances, de Finetti can count this as support for
his pure SUbjectivism.)
 Choice, chance, and credence 379

4. CHOICE AND CHANCE

Do runners have better hearts than nonrunners because running

is good for hearts, or because good hearts conduce to running?
The answer makes all the difference to the question, 'Would it
be reasonable for me to take up running, given that I don't like
to run but would like to have a healthy heart?'
In Chapter 1 of Jeffrey [24] an account of decision-making
is offered that seems to slight the difference just noted, between
therapeutic and diagnostic values of acts, and so seems to fail
where actions are symptoms, not causes, of states of affairs that
the agent would promote or prevent. Not all such cases can be
dismissed as lacking in theoretical or practical importance, even
though philosophical discussion of them in the wake of Nozick's
[36] essay on Newcomb's Problem has focussed gleefully on
examples so bizarre as to suggest that in this possible world, the
set of decision problems in which imputed direction of causality
matters is empty. Then it is of some interest to find real examples
of such problems, i.e., cases where a real deliberating agent does
give strong credence to the possibility that his actions are symp-
toms, not causes, of states he would promote or prevent, and
where although he may be making a factual error in taking
that view, the view cannot be ruled out as simply irrational for
that agent at that time. Here is such a case, I think.

Smoking and cancer: Three hypotheses. R.A. Fisher [37] argued

that the undoubted correlation that had then been demonstrated
between cigarette smoking and lung cancer admitted of three
explanations, among which the medical establishment had rashly
opted for the first: (l) Smoking causes cancer. (2) A precancerous
condition in the lungs causes smoking. (3) Smoking and lung
cancer are effects of a common cause, viz., a certain genetic
makeup. Fisher urged that the evidence available in 1958 made
(3) a lively alternative to (1). Coupled with his ([ 37], p. 7) reference
to 'the mild and soothing weed', this might lead one to wonder
whether Fisher was a smoker who thought that while abstinence
might be of prognostic interest, it was probably useless either as
prophylaxis or as therapy.
Let us suppose that for Fisher, smoking dominated abstinence,
380 R. Jeffrey

i.e., suppose that he preferred smoking with cancer to abstinence

with cancer, and preferred smoking without cancer to abstinence
without cancer. We do not thereby suppose that he preferred smo-
king to abstinence, according to Jeffrey [24]. Rather, the situation
could be as illustrated in Figure 3, where the numbers are desir-
abilities (= utilities) and where abstinence (AI) is preferred to
smoking (A 2) in spite of the dominance because the conditional
probability of cancer (C I ) on smoking is greater by 0.2 than that
of cancer on abstinence: with C2 = -C I and A2 = -AI we have

(4.1) deAl) = d(At-Ct)c(CdA t ) + d(A I -C 2)c(C 2/A I ) = 9

d(A 2) =d(A2 -CI)c(Cd A2) + d(A 2-C 2 )c(C 2 1A2) = 8

Here are the credences c(C l lAd = 0.1 and c(C l IA 2 ) = 0.03 are
uniquely determined by equations (4.1), given the fanciful d-values
of Figure 3. (For brevity, the subscript 'r's have been dropped.)
The prima facie difficulty posed here for my [24] account of
decision-making is that the c-values may be as above when Fisher
thinks that abstinence has no prophylactic or therapeutic value at

11 A 2 • C2

10 A 1 ·C 2

9 A I (abstain)
8 A2 (smoke)

7
IOc(C 2 )+1 A2
6
IOc(C 2 ) A,
5

Figure 3. A non-Fisherian desirability ;;cale.

Figure 4. Desirabilities consistent with Fisher's (1958) preferences.

 Choice, chance, and credence 381

all, in which case I do agree with Nozick and most others that
dominance should imply practical preferability, i.e., here, ration-
ality of smoking despite the lower d-value it gets in (4.1). As
Gibbard and Harper [38] say, the case in which c(C I /A I ) < c(C I /
A2 ) he cause (as Fisher thought) abstinence is a sign offreedom from
the unalterable genetic makeup in the presence of which the
chance of cancer is high, is a case in which Fisher's preference for
AI over A2 as a news item (Jeffrey [24] §S.7) or as an item in
his biography is compatible with his preference for A2 over Al
as a course of action.
I call this a 'prima facie' difficulty because in a plausible elabor-
ation of my conjecture about Fisher's beliefs it is not the smoking
itself but the desire to smoke that he took to shed direct light on
his genotype. Then if Fisher knew (D) that smoking dominated
abstinence in his preferences, and if he saw the further information
that AI (or A 2) is true as shedding no further light on his chance
of getting cancer, we have c(D) = I and so d(Do X) = d(X) and
c(Cj/AjOD) = c(Cj/D = c(Cj). Thus the equations

for the conditional expected desirabilities d(Ajo D) reduce to

(4.2) deAl) = deAl °CI)c(Cd + deAl °C 2)C(C 2 ) = lOc(C 2)

d(A 2) = d(A 2 °C I )C(C 1 ) + d(A 2 "C 2)c(C 2) = lOc(C 2)+1

in view of which A2 has the higher d-value regardless of the

value of c(C 2): see Figure 4. Then in this example the d-scale is
adequately sensitive to the agent's view that in the presence of D
(dominance of smoking over abstinence), choice and enactment
of A I or of A2 has no efficacy regarding the conditions Cj.
In a less simplistic account of his views, Fisher is less than
certain of the falsity of explanation (I), and the foregoing reso-
lution of the old problem appears as part of the resolution of the
new one. If explanation (I) is accepted then equations (4.1) are
applicable, while of (2) or (3) is accepted, equations (4.2) are
applicable. Then if 'H' ranges over {H I ,H 2} where H I says that
(I) is true and H2 says that one of the other two is true, and if
'A' and 'C' range over {A t ,A2} and {C 1,C 2}, the conditional
expected utility of act A will be
382 R. Jeffrey

(4.3) d(A) = LH,cd(HoCoA)c(HoC/A) = LHc(H/A)Lcd(HoCoA)

c(C/HoA)

Here the equation has been written so as to apply to the discrete

case generally, as well as to the problem at hand.
We can push matters further by setting H = [s] = {(s,W):S€W} =
the proposition that we are now at stage s, and then working with
relativised credence and desirability, c(X/[s)) and d(Xo[s)). In
particular, let us explore the consequences of the assumption that

(4.4) PsX = c(X/[s]) (Chance = Relativised Credence)

On this assumption, once we set H = [s] at the right of (4.3), we

have c(C/HoA) = c(C![s] °A) = c(CoA/[s]c(A/[s]) = Ps(C·A)/PsA =
Ps(C/A), and similarly c(H/A) = c([s]/A) = c(A/[s])c[s]!c(A) =
PsAc[s1/c(A), so that (4.3) yields

(4.5) d(A) = c(~) ~c[S]PsA ~d([sloCOA)Ps(C!A)

Note that if (4.4) is to serve as a subjectivistic reduction of the
concept of chance, it must be used to guide definition of c(X/ [s))
where c[s] = o.
In the simplest version of the Fisher problem we were able to
get the satisfactory result (4.2) by conditioning on D. Perhaps
such devices will continue to resolve prima facie difficulties in
more complex cases. But perhaps not. Independently of that,
some may follow Spohn [39], [40] in seeing problems about
credence in acts (Jeffrey [41], [24], Example 5.3) as fatal to
the hypothesis that the d-ranking is a rational preference ranking
of acts. And even where the d-ranking is acceptable (at least
locally), it may be that the other accounts of decision-making will
shed a better light on the matter. Recently, a number of such
alternatives have been floated, with a fair degree of convergence
emerging among some that had seemed quite different.
Following a suggestion of Robert Stalnaker's, Gibbard and
Harper [38] have proposed revising (4.1) by replacing conditional
credences in ('counterfactual') conditionals: in place of (4.1) they
have
 Choice, chance. and credence 383

(4.6) e(A) = Icd(A·C)c(A~)

(Here I use the simple arrow for the counterfactual conditional
connective where Gibbard and Harper follow D. Lewis [42) in
using a box followed by an arrow.) It is essential that the counter-
factual conditional in (4.6) be given a nonbacktracking rea 1ing,
e.g., 'Al~2' is meant to be a statement about what would happen
if Fisher (with his actual genetic makeup) were to quit smoking,
not about what would happen if Fisher had been born the sort of
person who does not smoke. If Fisher takes abstinence to have no
prophylactic or therapeutic value at all, then c(A~) = c(C) in
(4.6), so that the dominant act has the highere-value: e(A 2 » e(A 1 )
instead of d(A 2 ) < d(A I ) in (4.1).
It is clear that Gibbard and Harper get the right answer, but it
is less clear that they get it for the right reason: one that applies as
well to more complicated cases and goes to the heart of the matter.
They themselves point to certain features of their analysis as
simplifying assumptions that they hope to do without, in time.
Example: the assumption that the counterfactual conditional
satisfies the Stalnaker [43] condition, viz., equivalence of -(A-+B)
with A-+-B when A is not impossible.
But there seems to be a deeper-lying difficulty: instead of
dealing directly with the independence of the chance of C from
the act A, the Gibbard-Harper analysis used a consequence of that
independence, Le., the equality of c(A~) with c( -A~). Direct
reference to the propensity might make the analysis more broadly
applicable, e.g., to complex cases like the hitherto unmentioned
one in which Fisher takes there to be both prohylactic and merely
symptomatic value in abstinence, e.g., because he thinks that one's
chance of getting cancer is determined partly by one's genotype
(which also affects the chance that D is true) and partly (whatever
one's genotype) by one's smoking habits. Here the propensity for
C is thought to be partly dependent on A, but perhaps not so
much as to make (4.1) give the right result.
Developments of and alternatives to the Gibbard-Harper pro-
posal, that preference goes bye, not d, are in the offing: see
Cartwright [44], Levi [45], D. Lewis [46], Skyrms [25] and
[16], Sobel [47], Toda [48], and, no doubt, other work that I
have overlooked or simply not encountered. The next few years
should be lively ones for the philosophy of choice and chance.
384 R. Jeffrey

BIBLIOGRAPHY

Braithwaite, R.B. [11] Why Is It Reasonable to Base a Betting Rate upon an

Estimate of Chance? In Yehoshua Bar-Hillel, Ed., Logic, Methodology
and Philosophy ofScience. Amsterdam: North-Holland, 1965.
- [21] On Unknown Probabilities. In Komer [49],1957.
Burgess, John [17] The Unreal Future. Theoria 45 (1979).
- [49] Logic and Time . .!. Symbolic Logic 44 (1979).
Camap, Rudolf [18] Logical Foundations ofProbability. Chicago: University
of Chicago Press, 1950, 1962.
- [SO] On Inductive Logic. Philosophy of Science 12 (1945).
Cartwright, Nancy [44] Causal Laws and Effective Strategies. Forthcoming.
Chernoff, Hermand, and Moses, Lincoln [4] Elementary Decision Theory.
New York: Wiley, 1959.
Feller, William [32] An Introduction to Probability Theory and its Appli-
catiuns, Volum~ 1, 2:1d edition. New York: Wiley, 1957.
Finetti, Bruno de [22J and [30] La prevision: ses lois logiques, ses sources
subjectives. Annates de I'Institut Henri Poincare 7 (1937); translated in
Henry Kyburg and Howard SmokIer, Ed., Studies in Inductive Prob-
ability, New York: Wiley, 1964.
- [31] Theory of Probability , Volume 2, New York: Wiley, 1975.
Fisher, Ronald A. [37] Smoking. London: Oliver and Boyd, 1959.
Fraasen, Bas van [14] Tenses, with Conditionals and Chance. Forthcoming.
Gibbard, Alan, and Harper, William [38] Counterfactuals and Two Kinds of
Expected Utility. In C.A. Hooker, J.1. Leach and E.F. McClennen, Ed.,
Foundations and Applications of Decision Theory, Volume I. Oor-
drecht: Reidel, 1978.
Good, 1.1. [19] Probability and the Weighing of Evidence. London: Charles
Griffin, 1950.
- [20] The Estimation of Probabilities. Cambridge, Mass.: M.I.T. Press, 1965.
Hacking, Ian [8] Logic of Statistical Inference. Cambridge University Press,
1965.
- [10] Propensities, Statistics and Inductive Logic. In Suppes et al. [51] ,Ed.,
1973.
Hesse, Mary [1] A Self-correcting Observation Language. In B. van Rootselaar
and J.F. Stahl, Eds., Logic, Methodology and Philosophy of Science,
Volume 3. Amsterdam: North-Holland, 1968.
-- [2] The Structure of Scientific Inference. Berkeley and Los Angeles:
University of California Press, 1974.
Jeffrey, R.C. [13] Review of Miller et aI., J. Symbolic Logic 35 (1970):
124-127.
- [23] Mises redux. In Robert E. Butts and Jaakko Hintikka, Eds., Basic
Problems in Methodology and Linguistics. Oordrecht: Reidel, 1977.
- [24] The Logic of Decision. New York: McGraw-Hill, 1965.
- [41] A Note on the Kinematics of Preference. Erkenntnis 11 (1977).
Johnson, Norman L., and Kotz, Samuel 1341 Urn Models and their Appli-
cation. New York: Wiley, 1977.
 Choice, chance, and credence 385

Kolmogorov, A.H. [7] Grundbegri//e der Wahrscheinlichkeitsrechnung.

Ergebnisse der Mathematik (I933); translation (1950), New York:
Chelsea.
Komer, Stephen, Ed. [49] Observation and Interpretation. London: Butter-
worth,1957.
Kuipers, Theo A.F. [33] Studies in Inductive Probability and Rational
Expectation. Dordrecht: Reidel, 1978.
Kyburg, Henry E., Jr. [29] Propensities and Probabilities. British Journal/or
the Philosophy o/Science 25 (1974): 358-375.
Levi, Isaac [45] Newcomb's Many Problems. Theory and Decision 6 (1975).
Lewis, C.I. [5] Mind and the World Order. New York: Dover, 1956. Republi-
cation of 1929 first edition.
- [6] An Analysis 0/ Knowledge and Valuation. La Salle, Ill,: Open Court,
1946.
Lewis, David K. [15] A Subjectivist's Guide to Objective Chance. In Richard
Jeffrey, Ed., Studies in Inductive Logic and Probability, Volume 2,
Berkeley and Los Angeles: University of California Press, 1979.
- [42] Counter/actuals. Cambridge, Mass.: Harvard University Press, 1973.
- [46] Causal Decision Theory. Lecture at University of Pittsburgh, November
1978.
Mellor, D.H. [9] TheMattero/Chance. Cambridge University Press, 1971.
Miller, David [12] A Paradox of Infonnation. British Journal/or the Philos-
ophy o/Science 17 (1966).
Moore, Edward F., and Shannon, C.E. [3] Reliable Circuits Using Less
Reliable Relays. Journal of the Franklin Institute 262 (I956).
- [35] Sequential Machines. Reading, Mass.: Addison-Wesley, 1964.
Nozick, Robert [36] Newcomb's Problem and Two Principles of Choice. In
Nicholas Rescher, Ed., Essays in Honor 0/ Carl G. Hempel. Dordrecht:
Reidel, 1969.
Peirce, C.S. [26] Collected Papers. Charles Hartshorne and Paul Weiss, Ed.
Cambridge, Mass.: Harvard University Press, 1931.
Popper, Karl R. [27] The Propensity Interpretation of the Calculus ofProb-
ability, and the Quantum Theory. In Komer [49], Ed., 1957.
- [28] The Propensity Interpretation of Probability. British Journal for the
Philosophy o/Science 10 (1959).
Skyrms, Brian [16] Causal Necessity. New Haven: Yale University Press,
1979.
- [25] Statistical Laws and Personal Probabilities, PSA 1978, Volume 2.
Philosophy of Science Association, 1979.
Sobel, Jordan H. [47] Chance, Choice and Action. Working paper, revised.
Scarborough College, University of Toronto, 1978.
Spohn, Wolfgang [39] Where Luce and Krantz Do Really Generalize Savage's
Decision Model, Erkenntnis 11 (I 977).
- [40] Gn.md/agen der Entscheidungstheorie. Kronberg: Scriptor. 1978.
Stalnaker, Robert C. [43] A Theory of Conditionals. In Nicholas Rescher,
Ed., Studies in Logical Theory. Oxford: Blackwell, 1968.
386 R. Jeffrey

Suppes, Patrick, Henkin, Leon, Joja, A., and Moisil, C.R., Ed. [51] Logic,
Methodology, and Philosophy of Science, Volume 4, Amsterdam:
North-Holland, 1973.
roda, Mananao [48] Causality, Conditional Probability and Control. In
Ahmet Ayka~ and Carlo Brumat, Ed.,New Developments in the Appli-
cations of Bayesian Methods. Amsterdam: North-Holland, 1977.
Abbreviations used by some contributors

MT = meaning theory
TM = theory of meaning
OL = object language
IV = the category of intransitive verb phrases
T = the category of terms
TV = the category of transitive verbs
IAV = the category of IV-modifying adverbs
eN = the category of common name phrases
MPL = modal propositional logic
PL = standard propositiona11ogic
MQL = quantified modal logic
I ndex of names

Ackermann, W. 337, 351 Boole, G. 228, 340-351,357

Adorno, Th.W. 201 Boolos, G. 292
Ajdukiewicz, K, 13,27-33,144 Bouveresse, J. 5
Althusser, L. 206 Bowie, G.L. 314
Anderson, A.R. 231, 337, 351, 352 Braithwaite, R.B. 372,373
Andrews, P. 63 Bressan, A. 287
Anscombe, G.E.M. 316 Brouwer, L.E. 100, 228,229,241,
Apel, K.O. 191, 194, 195, 197,198, 269,271,280,339
203,209,210,212,215-218 Bubner, R. 188, 190,209
Aqvist, L. 76, 299,313,330,331 Burge, T. 44
Aristotle, 200-202, 209,232,236,285 Burgess, J. 373
Arnould, A. 27 Burks, AW. 282
Augustin, 201
Austin, 1.L. 4,13,18,155,267
Ayer, A.J. 17 Cantor, G. 229, 241
Carlson, L. 69
Carnap, R. 5,13,17,18,20,29,31,
Bacon, 1.B. 299,357 32, 68, 99, 116-118, 159, 213,
Baker,G.38,40, 100 214,231,287,373,377
Barcan, R. vide Marcus Cartwright, N. 383
Bar-Hillel, Y. 13,31 Cellucci,C.236,2S2
Barwise, J. 69 Chellas, B.F. 299, 308, 313
Bayes, T.368,373,376,377,378 Chernoff, H. 368
Belnap, N.D., Jr. 76-231 Chisholm, R. 286
Bennett, 1. 314-316 Chomsky, N. 3,50, 72-74, 76, 157
Bemays, P. 71, 236,252,254, 263, Church,A. 115,131,239,287,348
264,267,270 Chwistek, L. 20
Beth, EW. 247 Cocchiarella, N. 5,62,63
Bigelow, J.C. 316 Cohen, P.l. 229
Black, M. 60 Creary, L.G. 318
Bloch, E. 194 Cressweill, M.J. 283, 289
Bloomfield, L. 26 Curry, H.B. 249
Bochenski, J.M. 25 Cusanus, N. 3
39C Index of names

Danto, A. 195,207,208 Gadamer, H.G. 4, 8, 179, 185, 186,

Darwin, C. 93, 94 189, 191-193, 195-197, 199,
Davidson, D, 37,39,42,43,45-48, 201, 204, 205, 210, 212, 213,
59,71,72,74,346,349 216,217
Davis, M. 42, 43 Galilei, 209
Diamond, C. 109 Gallin,D.116
Dilthey, W. 179,180,183,184,188, Gardenfors, P. 303
192,205,213 Geach, p.r. 75,338 n.
Donnellan, K.S. 288 Gentzen, G. 241-249,251,252,255,
Dray, W. 195,288 259,260,263,264,266-269,358
Dressler, W.U. 75 Gibbard, A. 381-393
Dretske, F.I. 316 Girard, l.Y. 69, 246,247
Dummett, M.A.E. 39, 48, 49, 57, 61, Gbdel, K. 59, 60,69,229,230,237,
229,232,241,250,251 241, 247, 251, 257, 262, 263,
Dunn, I.M. 231,319 265,268,341,353,358
Goldman, L. 163
Good, U. 373
Ellis, B. 299,303,313,318 Goodman,N.66,303,309,315
Enderton, H.B. 69 Grandy, R. 43,48
Euclid, 127,338 Graves, C. 50
Evans,G.37,43,48,53,288 Greechie, RJ. 309
Grice, H.P. 4,44-46, 77
Gudder, S.P. 309
Fahrenbach, H. 214
Feferman, J. 249,270,271
Feller, W.376 Haack, S. 299, 313
Feys, R. 283 Habermas, J. 191-193, 195, 197,
Fichte,J.G.192 203,216
Field, H. 39,42,48,54 Hacker, P.M.S. 38, 40, 96,97, 100,
Fine, K. 314,321 103
Finetti, B. de, 373--376, 379 Hacking, I. 61,372,375
Fisher, R.A. 379-383 Hallett, G. 60
Fitch, F. 244 Hansson, B. 290,312
Fodol,J.A.48, 75, 76 Hardegree, G.M. 313
Fodor, lD. 15, 16,36 Harman,G.H.37,39,44
F~llesdal, D. 290,319 Harper, W. 381-383
Foucault, M. 206 Harrah, D. 76
Fraassen, B. van, 305,309,313,373 Hegel,G.F.W, 202, 211
Frege, G. 5, 13,39,57-62,66,67, Heidegger, M, 6, 8, 85,86, 179-181,
75, 86, 92, 97, 101, 131, 156, 183, 184, 187, 195, 196, 211-
159, 228, 229, 232, 236, 241, 215,218
260,287 Heijenoort, J. van, 57, 58
Freud, S. 94,211 Hempel, C.G. 195
Friedman, H, 271 Henkin, L. 63, 69
Friedman, M. 53 Herbrand, J. 241,268
 Index of names 391

Hertzberger, H. G. 311 Kenny, A.1. 60,93,94

Hesse, M. 367, 368 Kleene, S.c. 353, 355,358
Heyting, A. 241 Kolmogorov, A.H. 64, 268, 372
Hilbert, D. 30, 71, 228, 229, 235, Kotarbiilska, J. 23
236, 240, 241, 246, 252~269, Kotarbiilski, T. 11, 12, 14,20,21,
271,272 22
Hill, C.S. 318 Katz, S. 376
Hilpinen, R. 23 1, 284, 290 Kreisel, G. 39, 98, lOO, 237, 238.
Hintikka, J. 5, 8-10,37,38,94,230, 247-249, 251-253, 257, 258,
247,299 260-264,266,269,270,272
Hintikka, J.J.J. 282, 284,286,290, Kripke, S.A. 61 ~66, 230, 231, 282,
291 283,285,286,288,291,227 n.,
Hintikka, M.B. 74 339,352,354
Hii, H. 21 Krivine, J.L. 237
Hofmannsthal, H. von, 215 Kuhn,l.231
Howard, W.A. 249 Kuipers, T.A.F. 376
Hughes, G. 283 Kyburg, H.E., Jf. 373
Humberstone, I.L. 318, 330
Husserl, E. 13,20
Lakoff, G. 72, 74
Laplace, 376~378
Inwagen, P. van, 318, 319,325,328 Lebesgue, H. 372
Leblanc, II. 352
Lehrer, K. 299
Jackendorff, R. 74 Leibniz, G.W. von, 62, 63,227,232,
Jackson, F. 314, 318 279,284,287,290
Jakobson, R. 21 Lemmon, E.J. 283, 317, 341
Jauss, H.R. 7, 200, 201 . Lesniewski, S. 13,30, 72
Jeffrey, R. 231 Levi, 1. 303, 383
Jervell, H.R. 245 Levy, A. 257
Johnson, N.L. 376 Lewis, C.1. 13, 230, 231, 279, 280,
Johnson·Laird, P.N. 65 368,369
Lewis, OX. 36,44,64, 65,284,289-
291, 299, 307-318, 323, 327,
Kafka, F. 215 331,341,373,378,383
KambarteI, F. 6, 9 Lewis, L. 39
KamIah, W. 6,8,155,163 Lichtenberg, G. 83
Kamp, H. 63, 139, 312, 321-324, Lieb, H. 179,216,217
326 Locke,J.20
Kanger, S. 6] ,63,230,247,282,299 Loewer, B. 3 15 , 3 16, 3 18, 319, 321
Kant, 1. 70,185,192,194,200,211, Lorenz,K.61, 172, 173
214 Lorenien, P. 4, 6,8,61,155,172,
Kaplan, D. 63,286·-289 173
Karttunen, L. 76 Lukacs, G. 19
Katz, J.1. 15, 16,36,40,73,76,91, Lukasiewicz, 1. 30, 230
157 Luschei, E.C. 72
392 Index of names

Luther, M. 211,213 Pappinghaus, H. 247

Lycan, W. 38 Pargetter, R. 318
Parsons, T. 285
Partee,B.H.62,76,147
Mackie, J.L. 299, 313 Peacocke,C.4,288
Malcolm, N. 109 Peano, G. 245, 353
Marcus, R. Barcan, 231,280 Peirce, C .S. 6,20, 75, 155, 156,300,
Martin, R.M. 31 373
Martin-Lor, P.244-246,249,250,271 Pelc, J. 4
Marx, K. 192,206,211 Pelletier, F.J. 67
McGinn, C. 48 Plantinga, A. 288, 299, 314
McKay, T.318,319,325,328 Plato, 3,4, 75,172,183
Mead,G.H.~, 155, 156 Poincare, H. 229
Medvedev, LA. 378 Pollock,J.311,JI2,316,317
Mellor, D.H. 372, 373 P6lya, G. 376,378
Meyer, R.K. 337 n., 341--343, 351- Popper.K.R.195,313,373
353,356,358 Porn, 1. 299
Miller, D. 372 Postal, P.M. 36, 76
Miller, G.A. 65 Pottinger, G. 246
Mink, L.O. 207 Powers, L. 299
Mints, G.E. 247 Prawitz, D. 61 , 229
Misch, G. 179 Prior, A.N. 232, 282-284
Montague, R. 5,8, 13,31,36,37, Putnam, H. 39, 52.109,288
62, 63, 71, 72, 113-154, 160,
232,287,289,292
Moore, E.F. 17,367,369,372,378 Quine, W.V.O. 4, 14, 43, 49-54,
Morgan, Ch. de, 338, 341,343--346, 59, 60, 64, 72, 74, 229, 230,
349,350.357 281,286,287,309
Morris, C. 20, 21,22,26,31
Moses, L. 368
Mott, P. 313 Rabin, M.O. 378
Myhill, 1. 251 Raggio, A. 243
Ramsey, F.P.49,50,99,302,309,346
Rantala, V. 66, 69
Newcomb, W. 379 Rescher, N. 31,284,303,323
Nicole, P. 27 Richards, T. 314
Nietzsche, F. 211 Richardson, 1.T.E. 83
Niiniluoto, I. 299 Ricoeur, P. 6, 7
Nozick, R. 379,381 Rilke, R.M. 215
Nute,D.JIJ,316,JI8,319,328 Ritter, J. 202
Rosenberg, J.F. 12, 15
Rosser, J .B. 262
Occam, W. of, 281, 287 Routhly, R. 341,342,352,358
Ogden, C.K. 227 Russell, B. 13,57,72,91,99,104,
Orenstein, A. 59 228,229,236,241,255,288
Osswald, H. 245 Ryle, G. 14, 17
 Index of names 393

Saarinen, E. 37,63,68,69 Tarski, A. 14, 30, 37, 42, 59, 68,

Sapir, E. 26 70-72, 115,158-160,237,245,
Saussure, F. de, 20, 26 246,267,346,349
Savage, LJ. 64 Taylor, C. 10
Schiffer, E. 44 Thomason, R.H. 62, 63, 71, 299,
Schleiermacher, F. 192, 213 309
Schneider, H.J. 6, 9 Tichy, P. 314
Schutte, K. 244, 247, 265 Toda, M. 383
Schwartz, S.P. 66,109 Tragesser, R. 250
Scott, D. 67,69,118,317,349,378 Travis, C. 12, 15
Searle, J. 4, 46, 155 f. Troelstra, A. 246, 248, 252
Segerberg, K. 283
Shannon, C.E. 367-369,372
Simpson, S.G. 247 Urquhart, A.J.F. 284
Sko)em, T. 69, 237
Skyrms,B.373,383
Slote, M. 314 Vermazen, B. 74
Smith, DW. 299 Vickers, J. 307
Smorynski, C. 262 von Wright, G.H. 161, 163, 208,
Sneed, J. 231 209,282,284,320
Snyder, A. 316
Sobel, J .H. 383
Solovay, R. 292
Walkoe, W., Jf. 69
Spohn, W. 382
Wallace, J. 42
Stalnaker, R.C. 285, 289,291,299,
Weierstrass, K. 260
302-309, 312-314, 318, 372, Weyl, H. 260
383 Whitehead, A.N. 236,255
Statman, R. 246 Wiggins, D. 288
Steele, T .B. 76 Wittgenstein, L. 3, 5, 6, 13, 18,38,
Stegmilller, W. 231 60, 68, 72, 83-110, 155. 156,
Stenius, E. 60 161, 163, 166, 196, 207, 212,
Stenlund, S. 244 213,215,217,218
Stich, S. 50 f. Woodruff, P. 308
Stoy, J. 67 Wright, C. 39,41
Strawson, P. 14,41
Suppel'>, P. 76
Szabo, M.E. 252
Zeman, J. 283
Tait, W. 244, 245, 247 Zermeto, E. 241,267
Takahashi, M. 245, 247 Zimmermann,J. 214,215, 217, 218
Takeuti, G. 245, 246, 262-271 Zucker, J.I. 246, 250
I ndex of subjects

Actions Analyticity, 284

intentionalistic vs causalistic, 6
logic of, 299
system of, 163
theory of, 7, 207
utterances as, 155, 161
Acts
complex symbolic, 166 ff.
generic vs individual, 163
intentionality of, 163
mediated, 164 f.
mental, 166 f.
- schemes vs actualisations, 163
speech, 156 f., 164, 167,362
symbolic, 6, 163 ff.
- types vs tokens, 163
writing, 165
Adjunction, logical, 171
Admissibility, 340, 351 ff., 358 ff.
extensional, 354 ff., 360 f., 364
Aesthetics, 201
ascetisme, 201, 219
conscience, 201
- experience, 200 f.
jouissance, 201 f.
- of reception, 200
Algorithm, 251
Ambiguity, 114, 143, 145, 168 f.,
303,322 f., 329
._. of 'is', 72 ff.
Analytical
philosophy, 35 ff.
vs synthetic truth, 70
Argumentation, 169 ff., 236
agonal vs cooperative, 173
enthymematic, 361
eristic vs dialectical, 173
reductio, 357
relevant, 354 ff.
validity of, 250
Arithmetic, 263 ff., 340, 351 ff.
- of natural numbers, 244, 262
- of real numbers, 254 f., 257
Peano -, 245
relevant, 352 ff.
theory of, 257
Assertions, 169 f., 285, 325
diffident, 362 f.
essen tialist, 281
Assumptions, 352 f.
Axioms
axiomatic set-theory, 228 f.
axiomatization of logic, 60, 341
categorical, 71
comprehension, 265 f.
disquotational negation, 41
mathematical, 255, 260
- of meaning theory, 49 ff.
probability, 372 ff.
semantical, 50
Bayes
strategies, 368
theorem, 376
 Index of subjects 395

Behaviorism, 218 logic of, 299

Boolean negation, 228, 340 ff., 357 material, 300 ff.
might-, 308
pragmatics of, 312
Calculus, 337 strict, 300 ff.
axiomatized,229 subjunctive, 300 ff.
formalized, 229 would-,308
logical, 173 Confirmation theory, 231
lower predicate, 72, 230 Conjunctions, 255
mathematical, 229 generalizations of, 357
- of sequents, 243 ff. logical, 171
probabili ty , 374 wide-scope, 318,322,328
Cantor's Continuum Hypothesis, 229 Connectives, 337, 341
Catastrophe theory, 65 homophonic, 349 f.
Causal theory, 65 f. ineffective, 348
causality, 208 intensional,325
causation, 313 logical, 162
Chance, 367 ff. truth-functional, 325, 344
objective, 374 Connotation, 21
Change, theory, 231 vide Meaning; Sense and Reference
Choice, 367 ff. Consequence
Church's logical, 119 f., 123 ff., 135,251,
theorem, 348 374 f.
theory, 115, 131 theory of, 315 f.
Circumlocutions, 128 f., 147, 150 Consistency, 252 ff., 259 f., 262,
Clarification, 87 265 ff., 279, 302 f., 306, 315,
Classicalist, 339,344 f., 353, 356 ff., 353
363 f. counterfactual, 311 ff.
Completeness, 353 Constants
Compositionality, 71 f. individual, 117 ff., 147
Computability, 245 f., 250 logical, 117 ff., 245, 247, 250
Concatenation theory, 116 non logical, 119 ff.
Concepts predicate, 120 ff.
conceptualisation, 103 Constitu tion theory, phenomeno-
deontic, 282 logical,8
general, 89 Constructions, 174, 246
individual, 118, 130,281 constructivism, 4, 6,8 f., 229
modal, 227 vide Intuitionism
open character of, 97 Context, 75
psychological, 108, 117 decontextualize,206
relations of, 86 intensional, 57,115
super, 90 oblique, 57
Conditionals, 299 ff. opaque, 128 fL, 281
causal, 315 pragmatic, 169
counterfactual, 300 ff., 382 f recontextualize, 206 f.
-- excluded middle principle, 306 ff. Contradiction, 304, 360, 362 f.
- principle, 172
396 Index a/subjects

Contraposition, principle of, 301 f., non-epistemological conditions of,

305 189,209
Convention, 45 f. - of humanistic sciences, 183
Convertibility, 245 - subordination to ontology, 181,
Copula, 72 184
Cotenabili ty theory, 315 vs historical consciousness, 195
Credence, 367 ff. vs ontology, 183 f.
rational, 372, 378 Equations
Cross-identification, 10,64 f. differential, 65
numerical, 245
Equivalence, 358
Decision theory, 231
between the consistency problem
Deduction, 36 I and Hilbert's program, 258
system of natural, 243 f., 246
logical, 135,290
-- theory, 32
Eschatology, 194
Deductivist, 361 Essentialism, 281 f., 285 f.
Definability, theory of, 70 Ethics
Denotation, 5, 13, 21, 30 -- and politics, relation between,
vide Meaning; Sense and Reference 202
sense, 115, 130 ff. - institutions, 202
Descriptions Ethnology, 103
theory of, 244 Euclid's law, 127
uniform, 89 Events, generic, 124
Designatum, 21 Exchangeability, 373
vide Denotation; Meaning; Sense Expectations, 9
and Reference
conversational, 77
Differential equations, partial, 65
Explanation, 203
Diffidation, 362 f.
and understanding, 205, 207 ff.
Diodorus, Master Argument of, 232
complex status of, 209
Disappearance theory of truth, 346
conditions of, 208
Disjunctions, generalizations of, 357
quasi-teleological and quasi-causal,
Disjunctive antecedents, problem of,
318 ff. 209
Distribution, 280 therapeutic, 199
- principles, 338 Expressions
Dogmatism, 89 indexical, 159 f.
truth-theoretic, 43
Extension, 62
Eliminations, 242 ff. vide Meaning; Sense and Reference
- of disjunctive antecedents, - al context, 57
3 J7 ff.. 328 - al theory, 290 f.
Energy, nuclear, 340 - al truth-theory, 46
Entailment, 337 ff. - ality, 36 f.
tautology, 337 f. conservative,341 f.
Epistemology, 180,183 non extensional entities, 115
epistemological questions, 179 f. vs intension, 5, 8,115,118,127,
epistemological status, 208 130 f., 149

Fiction
logic of, 299
theoretical, 305
Fitch's set-theory, 244
Formalism, 228 f.
formalization of language, 25
moderate, 253, 259 f., 265 ff.
strict, 252
Forms, equivalent, 358 f.
Formula
atomic, 121 f.
modulations of, 270
Frege Principle, 71 f.
Frequency, 374 ff.
- principle, 372, 374 f.
Functions, 10, 67, 118, 122, l32,
134, 261,287,368 ff.
vide Recursion
credence, 375
functionals, 245 f., 248 f.
higher -order, 69
intf'fpretation, 69, 304
probability, 378
quomm, 368 ff.
selection, 304, 307, 310, 313, 315,
317,323,326 f., 330
Skolem,69
translation, 123
truth, 158,282,300,338,344,
367 ff.
Game theory, 172 f.
game-theoretic semantics, 4 f., 8,
37,68 ff., 76
language-games, 60, 68
sub-game, 69, 76
Gentzen system, 358
Global reflection principle, 257
Grammar, 102 ff., 106
philosophical, 218
rules of, 102
uni versal, 141
Hauptsatz, 243, 246 f., 270
Hermeneutics, 8 ff., 179 ff.
aesthetical sphere of, 185 f.
Index a!subjects 397
and analytical philosophy, 212
ancient, 183
anti-epistemological and anti-
logical motivation of, 181
categories of historical, 204
complementarity between - and
Scientistics, 195 f.
epistemological status of, 179, 185,
209
hermeneutical circle, 183 f., 217
hermeneutical reflection, transcen-
dental character of, 211
hermeneutical vs apophantical
logos, 182,214
historical-hermeneutical vs critical
social sciences, 192
historical sphere of, 185
history of, 210 ff.
juridical, 189
limitations of, 193, 203,212
literary, 184, 200 f.
logical,6
- of language, 196, 212, 217
- of reading, 7
ontology of, 184,193
radicalisation of, 179, 181, 183
romantic, 198
sphere oflanguage, 185
universality of, 190 f., 199 L
204 f., 210, 219 f.
vs criticism of ideology, 186,
193
vs rhetoric, 200
Historical
consciousness, 183, 188, 196,205
d tance,189
l,xperience, 186, 199
inquiry, 207
meditation, 219
Historicity, 183, 218
History, 181,187
natural into spiritual, 198
theory of, 207
Holism, psychological, 54
Homomorphism, 249
Identification of proofs, 249 f.
398 Index of subjects

Identity, 280 ff. Inversion principle, 242 f., 251

contingent, 286 f. Isomorphism, 249
Imperatives, logic of, 299 ~ntensiona1. 13
Implication, 230 f., 327, 341 f.
general,72
material,355 Judgements, terminating, 369
Indemonstrables, the Stoic's Five, 338 Juxtaposition, predicative, 72
Index, 8, 118 f., 122 f., 125, 139 f.
- of an expression, 31
Indicators, unreliable, 367 ff. Kleene complete, 353
Individuals Knowledge
vide Obiects logic of - and belief, 299
actual,281 non inferential, 50 f.
property of, 118, 125
Induction, 361,373
infinite, 353, 359 Language, 212 f., 214 f., 217,219
transfinite, 263,265 vide Metalanguage
Inference, 339 f., 362 f. and reality, 58 f., 105 f.
rules of, 236, 241 ff., 255, 260, arithmetical, 41 f.
266,301,316 f. as universal medium vs as calculus,
Intension, 119, 134 58 ff.
- al isomorphism, 13 autonomy of, 218
- ality, principle of, 317 f., 321, experience of, 199
328 f. flexibility of, 91
levels of, 8 formal, 86
Intention. 21,44 f., 163 f. functioning of, 84
- a1 objects, 281, 287 Gespriichsgemeinschaft, 210
- alism vs causalism, 6 incomparable essence of, 90
symbolic, 164, 169 f. instrumental reduction of, 185
Interpretation, 207 mathematical 75
- functionals, 69,304 nature of, 9 I
- of intensional language, 136 f. pragmatic, 117 ff.
- of literary texts, 6 private, 108
- of non classical logics, 68 f. rules of, 101,105
provability, 279 -- structure, 102, 156 f.
theory of, 7 theology of, 3, 88
Introductions and eliminations, 241 ff. Language games, 60, 68, 101, 108,
Intuition, 16,230 196,204,208,212,216 f.
semantical, 73,279 and ways of life, 215
Intuitionism, 228 f., 231, 339 f., contingency of, 86
360 ff. hermeneutics of, 217
Intuitionist, 339, 360 historicity of, 86
Intuitionistic logic of, 213
- mathematics, 241,268 practice of, 101
- semantics, 38 f. Lattice, logical, 342
- theory, 245 f., 271 Lebensphilosophie, 179
 Index ofsubjects 399

Lebesgue integral, 372 relevance, 231,244,339 ff.

Leibniz principle of plenitude, 287 tense, 31, 119 f., 138 f., 284
Umit Assumption, 310 ff., 314 topological,31
Unguistic Logicism, 92
- articulation, 196
- behaviour, 45 f.
- idealism, 193, 198,203
Linguistics, 3,12,15 ff., 91,157 Marxism, 191
fictitious, 86 criticism of, 192
taxinomical17s democritian, 91 Mass terms, 67
Logic Mathematical
vide Action, - of; Conditionals, - analyses, 244, 254 ff.
- of; Fiction, - of; Impera·· - axioms, 255, 260
tives, - of; Language games, - calculi, 229
- of; Modal, - of; Perception, consistence of - theories, 252 f.
- of; Knowledge and Belief, - formalism, 101
-of .-- languages, 75
and Mathematics, 66, 91 f., 173, - logic, 173,227 f., 239 f.
227 f., 235, 239, 260 - operations, 162
argumentative, 169 ff. - theory of games, 68,252
as calculus, 58 ff. Mathematics, 88, 100,245
as technique, 91 vide Logic and Mathematics; Meta-
axiomatization of, 60 f. mathematics and Logic con-
deductive vs inductive, 231 structive, 162
deontic, 119 f., 284,290,299, finitary vs transfinite, 254
311 ff., 319 ff. 260 ff.
dialogical, 6, 8,172 f. fOl'ndations of, 162,229,235 ff.,
epistemic, 290 240,252,255,259 ff., 268 ff.
first-order, 72, 74, 244 ideals vs real, 254 ff., 260 ff.
hermeneutic,7 f., 179,216 intuitionistic, 241
higher-order, 63, 68,116,120 ff., Mathesis universalis, 59
244 f. Meaning, 4 f., 7 f., 13, 32, 38 f.,
implicational, 249 99 f., 102, 106, 108 ff., 155 f.,
infmite propositional, 244 159, 213, 215, 229, 249 ff.,
intensional - vs set-theory, 115 f., 259
120 ff., 143 ff., 317 f. vide Denotation; Sense and Ref-
intuition~stic, 41,244 ff., 258 erence
logique du vague, 97 criteria of, 213 f.
mathematization of, 172 criticism of, 213
minimal,244 eidetic, 25
non classical, 67, 230 ex perimental, 10
~- of science, 194 informative, 165
polyvalent, 230, 232 limits of, 215
propositional, 279, 340 - of ideal sentences, 269
quantum,3l3 operational,25
ramified second order, 244 f. performative, 165
400 Index ofsllbjects

presupposition of, 213 Necessity, 115, 231, 279, 281 ff.,

- relations, 59 337
surface, 7 logical, 62 f., 337
verbal vs mental, 206 Negation, 172
Meaning theory, 7, 23,35 ff., 48 f., am pliative , 349
97,100,108,158,237 Boolean, 340 ff., 357
constructive, 50, 100 continuous, 349
diffident,49 f. de Morgan, 341 , 343 ff., 349 f.,
homophonic, 41 f.,49 357
modest, 48 f. epistemic,349
taciturn, 49 f. external, 343, 345
Mechanics, quantum, 309 internal, 343,345
Metalanguage, 14, 24, 25, 30 f., ontological,349
39 ff., 116 f., 142, 158 ff., 165, Newcomb's Problem, 379
291,346,349 f. Normalization theorem, 245 f., 248
Metalinguistics, 340, 345 f. Notation, logical, 115
theory of, 315 Numbers
Metamathematics, 115, 137,229 complex, 256
Mind, infinite, 266 Godel -,257
Modal irrational, 340
- concepts, 227,282 ff. natural, 255 ff., 266 ff.
~~ logic, 63, 115, 119 f., 229 f., real,256
279 ff., 299 ff. Numerical equation~, 245
- operator, 126 ff., 147,279 ff.
- sentences, 13
Modality, 62, 280 ff. Objects, 67,158
causal, 281 Obligation, conditional, 312 f.
physical, 281 Occam's razor, 287
temporaL 282 Omniscience, 349
Model logical, 66, 70
Aspects, 73 Ontology, 13,30,64,114,125,130,
maximal, 285 136,156,168 f., 181. 211,287,
standard vs nonstandard, 63, 70 291,299,305 f.,313,316
surface, 70 vs ontics, 196
- theoretic semantics, 36 f. Operators
theory, 30, 43,59 ff., 116, backward-looking, 71
142,237,247 epistemic, 290
urn, 69 f. focus, 322, 324
Modulations, 270 f. modal, 126 ff., 147, 279 ff.
de Morgan's Negation, 338, 341, monadic ohligation, 313
343 fL, 349 f., 357 necessity, 300
Mythology, 84 n-place, 117 ff.
mythical legitimation, 202 one-place, 322
-- of mental processes, 92 permissibility, 320
- of symbolism, 92 pragmatic,S, 117 ff.
Names, causal theory of, 288 selection, 322, 324
 Index ofsubjects 401

sentential,322 conditional, 326 f.

Ordinals, 264 f.
Paradox, philosophical, 124, 126
Peano system, 353
Perception, logic of, 299
Permissibility, 319 ff.
disjunctive, 320
strong vs weak, 320
Philosophy, 11 ff.,97, 108, 110
Phonetics, 73
vide Speech acts phonology, 157
Phrases
vide Propositions; Sentences;
Statements
basic, 144, 146
noun, 146 f.
quantifier, 146 ff.
Physics, quantum, 232, 309
Plenitude, principle of, 232
P61ya urn process, 376 f.
Positivism vs anti-positivism, 207
Possibility, 279, 281. 304
Pragmatics,5,9,13,22,28 ff., 115 ff.
and logical semiotics, 116 f.
Predicate, 117, 279,337
atomic, 245, 283
predicate systems, 245
Preference and value, 231, 239,303,
312 f., 381 f.
Prejuge, 186,217
rehabilitation of, 189, 191
l'S reason, 203
Probability, 64, 70,231,241,291,
363,367 ff.
subjective, 370,373
Process, P61ya urn, 376 f.
Proliferation of objects, 281
Proof theory, 235 ff.
Property, theory of, 285, 288
Propositions, 88 ff., 92, 99, 103 f.,
213,215,326,371
Ilide Phrases; Sentences; State-
ments
analysis of, 89, 98,105
atomic, 57
descriptive, 104
dynamic theory of, 88
eternal,378
grammatical vs em pirical, I 03 f.
mathematical, 75, 104,281
meaningful, 105 f.
primitive, 57, 67,100
system of, 303 ff.
world,137
Provability, 251,259,271,279,292,
353,357 ff.
Psychoanalysis, 198, 203 f.
Psychologism, 92
Quantification, 351 ff.
branching, 71
existential, 72,171,255,260,
357
Henkin-,69
higher-order, 340
individual, 340,352
modal-theory, 380 ff.
over possible individuals, 138 ff.
propositional, 342, 352
universal, 158,171,241,255,258,
260,300,306,352
Questions and answers, 7 f., 76 f.,
347 ff.
Reading, theory of, 6
Recursion
primitive, 257
recursive truth-clauses, 70 f., 119,
123,135
- theory, 251
Reductions, 243, 245 f., 248 f., 260
Redundancy theory of truth, 346
Reference
vide Denotation; Meaning; Sense
and Reference
multiple, 284
points of, 118 ff., 134 ff., 159 ff.
quantificational, 130, 136
theory of, 109, 142, 286 ff.
402 Index of subjects

Reflexitivity, 280, 282, 290 realistic, 156

Relations, 62 f text vs sentence, 75 f
accessibility, 290, 304 translational, 4, 36, 74,141 f.
basic semantical, 67 f., 70 truth-conditional, 157 ff.
binary, 143 truth-theoretic, 4, 37
meaning, 157 universal, 141 L
Relativity, semantical, 73 f. Semiology, 20 ff.
Relevance, 337 Semiotics, 20 ff.
- logic and relevantism, 339 fL logical vs descriptive, 116 f.
phenomenology of relevantism, Sense and Reference, S, 8, 13,40,
360 fL 57, 62, 71, 116, 142, 145, 156,
Relevantist, 339 L, 344 f., 347, 349, 159,231,284 ff.
354 ff., 361 ff. vide Denotation; Meaning
Revolutions, scientific, 231 Sense data, 13,128,369
Rhetoric, 200 Sentences
Rules vide Phrases; Propositions; S tate-
admissible, 351 ments
derivable, 351 affirmative, 250 f.
dialogical, 173 analytical 17, 289
grammar, 102 arithmetic, 358 f.
inference, 236,241 ff., 255,260, atomic, 66 ff., 282
266,301,316 f. conditional, 300 f.
introduction and elimination, 244, decidable, 255 f., 264 f.
246,248 declarative, 142 f.
language, 101, 105 existential, 256, 258 L, 267
-- of ties, 303 highcr-{)rder,68
particle vs frame, 173 ideall's real, 252,256 ff., 259 f.,
regulative, 165 266
observation, 52
particular, 7
Satisfaction, 30,119,327 permiSSion, 321 ff.
Scepticism, 10 1 pseudo-objective, 17
Scott's thesis, 349 quasi-syn tactic, 17
Semantics, 9, 20 ff., 36 ff., 97, 104, universal, 256
153 ff. Sequences, reduction, 246
conceptual-role, 4,39 ff. Sequents, calculi of, 243 f., 246 f.
critical and constructive, 4,38,40 Set theory, 66,136 L, 141,152,
game-theoretic, 4,5,8,37 ff., 266,269 f.
68 ff., 72 f, 76 axiomatic, 228 f.
generative, 73 f. Fitch's, 244
intuitionistic, 38 L -- selection functions, 307
modal, 279 ff. vs intensional logic, 115 f.
model-theoretic, 4, 37 Sign
possible worlds, 5,10,61 ff., 76, comprehension of, 107
230 fL, 279 ff., 299 ff. identity, 72,132
pure vs descriptive, 29, 32,116 f. propositional, 106 f.
 Index of subjects 403

syncategorematic, 132 Theory change, 231

-- systems, 32 L Trace theory, 74
theory of, 4,20 fL Tradition, 187, 192 L, 213
use of, 108 assum ed, 194
Similarity, concept of, 310 L, 314 authority of, 189.219
Simplification of disjunctive ante- existential engagement in, 197
cedents, 317, 319 transmission of, 198
Sinn und Bedeutung, 13 Transformation, 18,29 f., 74, 261
ride Sense and Reference meaning-preserving, 37
Speech-act theory, 156 f., 164 f. Transitivity, 280, 283
Statements, 357 f. principle of, 301,305
vide Assertions; Propositions; Sen- Translation, 60,141,143 fL, 246
tences - function, 123
arithmetical, 352, 359 indeterminacy of, 4, 52 ff.
atomic, 327 intuitive, 115
conditional, 302 ff.. 327 translational semantics, 39 f., 74
elementary, 13, 166 fL, 169 L Truth, 13, 23, 30, 185, 229, 237,
eternal, 370 f. 253
observation, 13,370 conditions, 40 f., 46 fL. 59, 68,
performative, 13 71 f., 119. 123, 135 fL, 158,
Strategies, 8, 69 247, 251, 254, 289, 303 IT.,
Structuralism, 205 f. 314 f., 323, 325,327 f., 349
Structure, part-whole, 7 L contingent. 289
Sub-game, 69, 76 defmitions, 245 ff., 260
Subjectivism, 374 fL disappearance truth theory, 346
subject ps predicate, 166, 168 L extensional truth theory, 46 f.
Subjunction, logical, 172 - frequency, 374
Substitutivity of identity, 128 f., - functions, 158,282,300,338,
281 ff. 367
Supervaluation, 309 homophonic truth theory, 4 I f.
Supposition, 325 iogical,61,114,119f.,123,126,
Syllogism, disj unctive, 338 fL 135,281,320 f., 328, 344
Symmetry, 280, 283 necessary, 62, 1 15,281,289
Syntax. 5 f.. 9. 50 f., 73, 76. 116, recursive truth-causes, 70 f.
123,132.141 f., 147, ISS ff. redundance truth theory, 346
simpliciter, 61
-- values, 118 f., 122, 132,135,
Teleology, 174 282 f., 342 f., 346, 368, 370,
tertium non datta, 172,267 374
Theorem Type theory, 1 15,131 f., 245, 257
Bayes's. 376
Church's, 348
elementary, 12 Underdetermination, 52 f.
extensional,355 Understanding, 184, 189, 191, 195.
Kleene, 355, 358 208f.,219f.
logical, 341 - and care, 215
404 Index ofsubjects

- and sense, 213 279 ff.

- as epistemological dimension,
195
dialectic of, 207
epistemological status of, 209
henneneutical, 192
initial vs tenninal, 204
inter-subjective, 207
moment of, 182
pre-, 187,202,214,217,220
problem of, 181
theory of, 7
the taste of, 219
Uniqueness, 303
Urelements. 67
Validity
logical, 173
- of an argument, 250
- of modal proposition logic,
Verification, 13
Verisimilitude, 313
Weltanschauung. 11 f.
Worlds
absurd, 304, 307 f.
deontically perfect, 319
micro, 232
plausible, 312,324,326 ff.
possible, 10,61 fr., 76,125 ff.,
134, 136 f., 158,230 ff.,
279 ff., 299 ff., 371
reference, 326 f.
selected, 326
small,64
Tractatus-, 161
Zennelo's cumulative hierarchy, 267