You are on page 1of 10



ABSTRACT. The dynamic nature of Game-Theoretical Semantics is emphasized. The role

of strategic meaning in accounting for linguistic competence is examined. The semantics of
epistemic possibility is shown to involve a dynamic ingredient. Update semantics has been
designed to capture it. The paper focuses on the interplay between logical and linguistic
competences in discourse understanding.


The invention of game-theoretic semantics by J. Hintikka in the seventies

(Hintikka 1973, Saarinen 1979) can be described as the emergence of a
new paradigm not only for semantics but also for logic and the philosophy
of mathematics.
I shall concentrate on the dynamic component of Game-theoretical
semantics (G.T.S.). The dynamic turn initiated by G.T.S. in semantics is
better understood when looked at in historical perspective. G.T.S was foreshadowed by Ch. S. Peirce in manuscripts written in the 1890s as shown
in R. Hilpinens paper On C.S.Peirces Theory of the Proposition: Peirce
as a Precursor of Game-theoretical Semantics.
According to Peirce, The utterer is essentially a defender of his own
proposition, and wishes to interpret it so that it will be defensible. The
interpreter, not being so interested, and being unable to interpret it fully
without considering to what extreme it may reach, is relatively in a hostile
attitude, and looks for the interpretation least defensible (Hilpinen 1982,
This passage clearly anticipates the idea of a game with two players
where one player wins what the other loses (De Mulder 1994, 300301).
In the case of quantifiers, according to Game-theoretical semantics, [t]he
players are Myself, who tries to find a substitution-instance of the sentence
that is true, and Nature who, like some kind of malin gnie, tries to make
the substitution instance false (De Mulder 1994, 301).
Here again, J. Hintikkas views were explicitly anticipated by Peirce
who says that the utterer may (allow) his opponent i.e., (the interpreter)
a choice as to what singular object he will take as an instance to refute
Synthese 130: 175184, 2002.
2002 Kluwer Academic Publishers. Printed in the Netherlands.



the proposition, as in Any man you please is mortal (Peirce, quoted by

Hilpinen, Hilpinen 1982, 185).
The dynamic character of Game-theoretical semantics is worth stressing. It can be traced back to the following passage. Rejecting Wittgensteins pictorial account of meaning, J. Hintikka writes: [i]n order to
understand (a given first-order sentence) F , we typically do not have time
(or memory-space) enough. Our actual understanding of first-order sentences must accordingly be based on some finite, step-by-step comparisons
between the sentence F and the world rather than the (potentially) pictorial
nature of F (Hintikka 1973, 105106, quoted by Rivenc 1998, 155).
The characteristic feature of Game-theoretical semantics does not lie,
as one might think, in the particular rules which define the success conditions of a move in the game. It rather lies in the very notion of strategy.
The novelty of the game-theoretical approach to language theory is that
language is considered a goal-directed rather than rule-governed activity
(Hintikka and Sandu, 1997, 405).
As soon as this distinction has been drawn, new and interesting questions can be raised such as How can I possibly follow in practice . . .
a strategy when there is no effective way for me to find out in general
what my next move will be? (Ibid. 375). Several options are open here.
One might provide some sort of heuristics as we do in non deterministic
proof systems. Or, as J. Hintikka and Sandu suggest, one might restrict
the strategies of Myself to recursive ones. The latter global constraint is a
very precise way of capturing the Wittgensteinian concern for the actual
playability of our language-games.
J. Hintikkas insistance on the goal directed nature of language helped
to rediscover the significance of the notion of game in the formative period
of Wittgenstein (19291931). At that time, Wittgenstein was currently
using the word game to refer to activities such as verification or falsification which are goal-directed. The semantical games studied by J. Hintikka
within the framework of his Game-theoretical semantics can be seen as a
species of language-games in a Wittgensteinian sense.
The emphasis on strategies led J. Hintikka to isolate the very useful
concept of stategic meaning. Strategic meaning is set off against abstract
meaning. Abstract meaning is quite standard: . . . it is defined by specifying in which possible worlds (a sentence S) is true and in which possible
worlds it is false (Hintikka 1987, 503). Strategic meaning is a force carried by some assertive sentences which goes beyond the abstract meaning
of S and which amounts to knowing in part or totally, what the strategy is
that enables Myself to win in G(S) or which the speaker thinks verifies S
(Hintikka, Ibid.).



Strategic meaning is not a unified phenomenon. It covers background

information, conversational expectations and the principle of charity. One
might be tempted to question the claim that strategic meaning is meaning
at at all and argue that it should rather he conceived as an expression of
rationality involved in the use of language. J. Hintikka is aware of this
possible objection, but he has an answer.
Let us consider the sentence: Tom accepted the job Dick offered him,
but Harry refused it. If it is given a pronoun-of-laziness reading, that
pronoun is a mere syntactical placeholder for the job Dick offered him
where him automatically refers to Harry. This is clearly the intended
meaning. If it is not given a pronoun-of-laziness reading, that pronoun
refers to the job accepted by Tom. This however is nonsense. As J. Hintikka
observes, Harry could not have refused a job Tom had already accepted.
Hence the second interpretation is discarded by appealing to common
Why do we need strategic meaning over and above a principle of rationality? Why do we refrain from analyzing the phenomenon in terms
of enthymemes? The reply is straightforward. There are similar sentences
for which syntactic clues play a crucial role. Consider the sentence: John
carries his house key in his pocket, but Joan usuallly carries it in her purse.
As J. Hintikka observes the pronoun-of-laziness reading here is excluded
by the gender of his.
A game is a sequence of moves performed in accordance with rules
and articulated by a strategy. Hence it has a dynamic structure. GameTheoretical Semantics is well equipped to account for discourse phenomena which examplify a dynamic character such as discourse anaphora. By appealing to ordering principles over and above binding rules,
Game-Theoretical Semantics offers an elegant account of the donkeysentences.Instead of regimenting the sentence If Peter owns a donkey,
he beats it into x ((Donkey (x) Own (Peter, x)) Beat (Peter,
x)) which distorts the surface form, J. Hintikka deconstructs the notion
of scope and distinguishes the priority scope [ ] from the binding scope
( ). The sentence of natural language is rendered by x [(Donkey (x)
Own (Peter, x))] Beat (Peter, x). As logical rules are governed by
priority scopes and not by binding scopes, the above sentence is logically
equivalent with x[(Donkey (x) Own (Peter, x)) Beat (Peter, x)]
as it should be, even though it retains the surface form of the initial sentence. Observe that the last occurrence of x is not bound in the last two
On J. Hintikkas account, the apparent transformation of an existential
quantifier into a universal one in donkey sentences is easy to explain. It



can be seen as an instance of the well-known rule of passage (Quine 1982,

142147) which says that whenever the scope of an existential quantifier
comprises only the antecedent of a conditional it can be made to cover the
entire conditional by changing it into a universal one . . . (Hintikka 1997,
Beside being dynamic, the structure of Game-theoretical semantics is
also global. It obeys the context dependence principle of meaning which
says that the meaning of an expression can depend on the meaning of a
larger expression of which it is a part. If we put aside its philosophical
dimension (Hintikka 1979, 21, Sinaceur 1998, 137, Rivenc 1998, 167),
Game-theoretical semantics is an alternative to the kind of semantics developed by Montague for natural languages. The latter is built on the
model of Tarkis semantics for formal languages. It rests on the principle
of compositionality which says the meaning of a compound expression is a
function of the meanings of its constituent parts and of their syntactic combination. The principle , often called Freges principle, is forshadowed in
the following passage of Wittgensteins Tractatus: (A proposition) is understood by anyone who understands its constituents (Wittgenstein 1922,
paragraph 4.024).
For all its apparent clarity, the principle of compositionality raises
problems. It depends upon the notion of constituent which is not as
straightforward as it seems. As W. Hodges observes, [I]f a natural unit
of meaning in a sentence seems to depend on scattered pieces of the sentences, there is a strong temptation to say that these pieces form a deep
level constituent . . . (Hodges 1997, 540). The definition of constituent is
adjusted in order to rescue compositionality.
J. Hintikka holds that the principles of compositionality and the principle of context-dependence are incompatible: [i]f the meaning of a
complex expression depends only on the meaning of its parts, it can never
depend in its context on a still more comprehensive expression (Hintikka
and Sandu 1997, 370). At this stage, the question arises whether the incompatibility between compositionality and context-dependence cannot be
removed by switching from a static to a dynamic conception of meaning.
I shall address this question in the next section.


I will now briefly examine a recent contribution to semantics of natural language which stands hetween Montague Semantics and Game-theoretical
semantics. The dynamic predicate calculus spelled out by J. Groenendijk
and M. Stokhof (Groenendijk and Stokhof 1991) can be seen as an at-



tempt to reconcile compositionality with context-dependence. Moreover

the notion of context and the notion of interpretation get interconnected:
[c]ontext and interpretation are interdependent: interpretation depends on
the context but also changes the context (Groenendijk and Stokhof 1998,
31). The invention of the dynamic predicate calculus, and for that matter H.
Kamps discourse representation theory, (Kamp 1994) are major advances
in the theory of meaning. To appreciate the novelty of their approach, we
have to remember how the role of the context was understood in the early
Twenty four years ago, R. Stalnaker recognized an interaction between
content and context (Stalnaker 1974, 212). However he stuck to the definition of meaning that has prevailed at least since Wittgenstein: the meaning
of a declarative sentence consists of its truth-conditions. The definition
has been formulated in these terms by Wittgenstein in the Tractatus: To
understand a proposition means to know what is the case if it is true
(Wittgenstein 1922, paragraph 4.024). Sir Peter Strawson is still more
explicit when he writes: to know the meaning of a [statement-making]
sentence . . . is to know under what conditions someone who used it would
be making a true statement (Strawson 1952, 211).
The distinction between truth-value and truth-conditions which underlies the above definition of meaning is at the root of one of the major
philosophical discoveries of the twentieth century philosophy, i.e., possible
worlds semantics. The point has been made by M. J. Cresswell: [k]nowing
the meaning of (a declarative sentence) . . . is simply having the ability to
distinguish between worlds in which it is true and worlds in which it is
false. The idea leads directly to what is called possible worlds semantics
(Cresswell 1978, 12). We have seen above that J. Hintikkas account of
abstract meaning as opposed to strategic meaning agrees with Cresswells
statement. In contrast, the notion of strategic meaning involves the notion
of strategy. And the latter is defined for the actual world.
There is much to say in favour of the truth-conditional account of
the meaning of assertive sentences. It highlights the narrow connection
between truth-conditions and truth-value and at the same time the difference between the two concepts. The difference is blatant: I can know the
truth-conditions of the declarative sentence there is life on Mars without
knowing its truth-value.
For all its virtues, the classical definition has its drawbacks. It does
not suffice to account for the meaning of assertive sentences just as a
purely extensional semantics cannot do justice to the distinction between
informative identity statements such as a = b and non informative identity statements such as a = a. The following pair of sentences due to B.



Partee brings out the shortcomings of the definition of meaning in terms of

Consider the following sentences:

I dropped ten marbles and found all of them, except for one. It
is probably under the sofa.


I dropped ten marbles and found only nine of them. It is probably under the sofa.

The two underlined sentences convey the same information about the
actual situation. They have the same truth-conditions. Yet the second, but
not the first, produces an unacceptable discourse. Admittedly discourse (2)
can be seen as an enthymeme whose missing premisse can be retrieved in
the light of some principle of rationality. This line of explanation, however,
fails to account for the grammatical ill-formedness of discourse (2).
The source of the trouble has been diagnosed in this way by J. Groenendijk and M. Stokhof. For being able to interpret the pronoun it, one
must introduce an appropriate discourse referent that creates the interpretation context. The opening sentence of discourse (1) supplies such a
discourse referent. It introduces a first discourse referent for the group of
the ten marbles which have fallen and another one for the marble which
was not found. In the case of discourse (2), the opening sentence introduces
a first discourse referent for the original group of ten marbles which have
fallen and another one for the nine which were found but no discourse
referent is supplied for the missing marble. This is the reason why the
pronoun it cannot be provided with an interpretation (Groenendijk and
Stokhof 1998, 33).
The occurrence of discourse referent Except for one does not change
our information about the world but it changes our information about
the discourse itself. It is somehow a self-referential item of the text. As
discourse develops, both information about the situation and information
about the text are updated.
The first kind of information can be described as a set of possible
situations, the set of possible situations that the agent cannot distinguish
from the actual situations (Groeneveld 1995, 11). The second kind of
information consists of the text items available at that stage. Updating
information over the world amounts to throwing out possible situations,
i.e., eliminating alternatives. Updating information over the discourse
consists of adding or withdrawing text-items.
This account, however, is still too rough. As Groenendijk, Stokhof and
Veltman observe Discourse in itself is not just a list of sentences. It has a



more complex structure (Groenendijk et al., 1996, 27). Saliency considerations matter. Consider the following example: A man came to see the
doctor. The man said . . . . The definite description The man might refer
to the doctor. It is however more natural to take it as referring to the individual introduced by the indefinite description A man. The crucial point
here is the descriptive contents shared by The man and A man. This
makes a man more salient qua object corresponding with the description
(the man). . . . than . . . the doctor . . . (Ibid.).
G. Sandu argues that Game-theoretic semantics , as opposed to Dynamic predicate calculus, can treat sentences such as If every man is given
a gun, then some man will fire it. By Sandus lights, game-theoretical
semantics alone can explain the functional dependency between the men
who are given a gun and those who fire it (Sandu 1997, 167). His argument against dynamic predicate calculus loses some of its bite, however,
if we bring saliency considerations to bear on the matter. The reason why
we can rewrite the sentence as If every man is given a gun, then some of
them will fire it lies in the shared descriptive content depicted by the the
two occurrences of men in the initial sentence.
Consider however the following sentence: If each soldier is given a
rifle or a submachine gun, some soldier will have to be taught to use it.
The pronoun it refers either to a rifle or to a submachine gun depending
whether the soldier referred to has been given a rifle or a submachine
gun. Sandu shows that Game-theoretical semantics can account for the
dependency between soldiers and their weapons in that intricate sentence.
The question arises whether an appeal to the logical skills involved
in restoring missing premisses in enthymemes would not provide an alternative explanation. Take the following sentence which is similar to the
previous one but a little simpler: John is a very absent-minded man. He
always leaves something behind. Sometimes his hat, sometimes his umbrella. He finds it when he returns home in the evening. The hearer might
pin down the referent of it by making a constructive dilemma: either
John leaves his hat behind or John leaves his umbrella behind. If he has left
his hat behind, he finds it in the evening when he returns home. If he has
left his umbrella behind he finds it in the evening when he returns home.
Hence, john finds his hat or John find his umbrella in the evening when he
returns home. Of course this comment is not meant to settle the issue. A
large scale empirical study would be needed to compare the explanatory
power of the rival accounts of complex anaphora.
J. Hintikka supplies an example of strategic meaning which involves
but does not reduce to an enthymeme (Hintikka 1987, 507). Imagine that
the addressee knows that I am not a Dutchman. I can utter the sentence



The witness is lying, or I am a Dutchman with the force of The witness

is lying.
Why? Simply because the addressee will take my disjunctive sentence
as the premiss of the following inference: The witness is lying, or I am a
Dutchman, I am not a Dutchman. Therefore the witness is lying. This is
not the whole story however. There are also syntactic clues operating here
as the non commutativity of the disjunction shows.
Whatever the the results of such an inquiry might be, the new semantics
invented by Groenendijk and Stokhof should be credited with a major innovation. We owe them a new account of meaning. The dynamic character
of that account has been vividly expressed by F. Veltman: [t]he slogan
You know the meaning of a sentence if you know the conditions under
which it is true is replaced by this one: You know the meaning of a
sentence if you know the change it brings about in the information state
of anyone who accepts the news conveyed by it (Veltman 1996, 221).
At first blush one might be reluctant to adopt the new definition of
meaning. It seems too broad. There is, however, a remedy. As J. van Benthem observes, if we suppose that cognitive states are ordered by some
pattern of inclusion by informational content, we can define, among
other things, the notion of minimal updating -up (P ) which is formally
rendered by xy x y P y z(x z y P z) (van Benthem 1994, 118). Taking advantage of this new operator, we can define the
meaning of an assertive sentence as the minimal change in informatonal
content the sentence triggers.
Updates can be conceived as relations between a cognitive state s, the
state of the hearer before hearing the assertive sentence and the state t,
i.e., the state of the hearer after hearing that sentence. On this account,
sentences are mapped onto relations between information states, not onto
sets of possible worlds.
This is a significant change indeed. Relational algebra is richer than
Boolean algebra. A fine-grained account of the relationship between the
premisses of a given argument becomes available. The new modelling
at our disposal can account for the role of the sequential order of the
statements in a discourse.
Updates can also be captured by functions; a sentence can be seen as
a function which takes an information state I as its argument and an
updated information state I  as value: []I = I  . Not much hinges on
this choice. As J. van Benthem observes, [t]here is no conflict between
relational and functional approaches. Functions are deterministic total relations. And conversely, every binary relations R on (a set of states) S



induces a function R # from pow(S) to pow(S) . . . (van Benthem 1996,



In this survey paper devoted to recent developments of semantics and

logic, I gave a short presentation of Game-theoretical semantics and Dynamic predicate calculus. Although these two epoch-making contributions
to semantics are sometimes seen as rivals, I am inclined to think that they
complete each other in so far as the practice of natural language rests upon
logical as well as linguistic competence.
Dynamic predicate calculus handles changing anaphoric bindings. Update semantics deals with information flow and epistemic changes. The
treatment of complex linguistic phenomena requires a combination of
these two formalisms. The examination of the recent proposals made for an
integration of Dynamic predicate calculus with Update semantics would
take us too far away. The interested reader is invited to have a look at
Coreference and Modality (Groenedijk et al. 1996).


I thank Dr. P. Dekker, Prof. B. Dreben, Prof. J. van Eijck, Prof. P. Gribomont, Prof. S. Haack, Prof. H. Kamp, Prof. D. Fllesdal, Prof. J Hintikka
and Prof. T. Imamichi for their comments on a former version of this
paper which was presentend at a meeting of the International Institute of


van Benthem, J.: 1994, General Dynamic Logic, in D. M. Gabbay (ed.), What is a Logical
System? Clarendon Press, Oxford, pp. 107139.
van Benthem, J.: 1996, Exploring Logical Dynamics, CSLI & FoLLI, Stanford.
Cresswell, M. J.: 1978, Semantic Competence, F. Guenthner and M. Guenthner-Reutter
(eds), Meaning and Translation, Duckworth, London, pp. 927.
De Mulder, W.: 1994, Game-Theoretical Semantics, in J. Verscheren, J.-O. stman, and
J. Blommaert (eds), The Handbook of Pragmatics Benjamins, Amsterdam, pp. 300307.
Gochet, P., and Gribomont, P.: 1994, Logique. Mthodes formelles pour ltude des
programmes, Paris, Herms.
Groenendijk, J. and Stokhof, M.: 1991, Dynamic Predicate Logic, Linguistisc and
Philosophy 14, 39100.



Groenendijk, J. and Stokhof, M.: 1998, Betekenis in beweging, Algemeen Nederlands

Tijdschrift voor wijsbegeerte 90, 2653.
Groenendijk, J., Stokhof, M., and Veltman, F.: 1996, Coreference and Modality, in Sh.
Lappin (ed.), Lappin Handbook of Contemporary Semantic Theory, Blackwell, Oxford,
pp. 179213.
Groenendijk, J., Stokhof, M., and Veltman, F.: 1996, Changez le contexte!, Langages
123, 829.
Groeneveld, W.: 1995, Logical Investigation into Dynamic Semantics, Ph.D. thesis, ILLC,
Hilpinen, R.: 1982, On C. S. Peirces Theory of the Proposition: Peirce as a Precursor of
Game-Theoretical Semantics, The Monist 62, 182189
Hintikka, J.: 1973, Logic, Longuage-Games and Information, Clarendon Press, Oxford.
Hintikka, J.: 1987, Language Understanding and Strategic Meaning, Synthese 73, 497
Hintikka, J.: 1997, No Scope For Scope?, Linguistics and Philosophy 20, 515544.
Hintikka, J. and Sandu, G.: 1997, Game-Theoretical Semantics, J. van Benthem and A.
ter Meulen (eds), Handbook of Logic and language, Elsevier, Amsterdam and The MIT
Press, Cambridge MA, pp. 361410.
Hodges, W.: 1997, Compositional Semantics for a Language of Imperfect Information,
Logic Journal of the IGPL 5, 539563
Hughes, G. E. and Cresswell, M. J.: 1996, A New Introduction to Modal Logic, Routledge,
Kamp, H.: 1994, Discourse Representation Theory, J. Verschueren, J. stmann, and J.
Blommaert (eds), The Handbook of Pragmatics, Benjamins, Amsterdam, pp. 253261.
Rivenc, F.: 1998, Quantifier nest pas jouer, in J. Hintikka (ed.), Questions de logique et
de phnomnologie, E. Rigal, Paris, Vrin, pp. 149169.
Sandu, G.: 1997, On the Theory of Anaphora: Dynamic Predicate Logic vs. Gametheoretical Semantics, Linguistics and Philosophy 20, 147174.
Saarinen, E. (ed.): 1979, Game-Theoretical Semantics, Essays on Semantics by Hintikka,
Carlson, Peacocke, Rantala, and Saarinen, Dordrecht, Reidel.
Sinaceur, H.: 1998, Du Modle la stratgie, in J. Hintikka (ed.), Questions de logique
et de phnomnologie, par E. Rigal, Paris, Vrin, pp. 127138.
Stalnaker, R.: 1974, Pragmatic Presuppositions, in M. Munitz and P. K. Unger (eds),
Semantics and Philosophy, University Press, New York, NY, pp. 197213.
Strawson, P. F.: 1952, Introduction to Logical Theory, Methuen, London.
Veltman, F.: 1996, Defaults in Update Semantics, Journal of Philosophical Logic 25,
Wittgenstein, L.: 1961 [1922], Tractatus logico-philosophicus, new translation by D. F.
Pears and B. F. McGuinness, London, Routledge and Kegan Paul.
Paul Gochet
University of Liege
Section de Philosophie
7 Place du XX Aout
4000 Liege