A Formal Computational Semantics and Pragmatics of Speech Acts*
Eric WERNER University of Hamburg Department of Computer Science, Project WISBER, P.O. Box 302762 Jungiusstrasse 6, 2000 Hamburg 36, West Germany and Department of Computer Science Bowdoin College, Brunswick, Maine 04011, U.S.A. *This work was in part supported by the German Ministry for Research and Technology, BMFT, andwaspartofthejointprojectWISBER. The WISBER partners include: Nixd0rf Computer AG, SCS GmbH, Siemens AG, the University of Hamburg and the University of Saarbruecken. It was also in part supported~by grants from Bowdoin College.
Abstract This paper outlines a formal computational semantics and pragmatics of the major speech act types. A theory of force is given that allows us to give a s e m a n t i c a l l y and pragmaticaly motivated taxonomy of speech acts. The relevance of the communication theory to complex distributed artificial intellince, DAI, systems is described. l Introduction In a system that engages in a dialogue with a user about some domain, like the one we are building in the WISBER project [Bergmann, Gerlaoh 87], the problem of describing the meaning of a speech act in terms of its effect on the user model and, more generally, on the system's knowledge base becomes central. The f u n d a m e n t a l problem is that, until now, there has been no general formal theory of meaning for speech acts. Previous formal semantic theories such as Montague [74] and situation semantics [Barwise and Perry 83] were limited to assertions. Cohen and Perrault [79] give only a theory of how speech actsare planned, the semantics being implicit in the add and delete lists of the operators. Appelt [85] gives a semantics for assertions based on Moore [80] that is compatible with our approach. However, there is no explicit theory of communication. As Appelt himself states, what is lacking is an explicit theory of intention. Searle and Vanderveken [85] do little more than classHy speech acts based on features that were seen as relevant to their differentiation. However, as we will argue below, those features are not motivated by any coherent theory of meaning. The crucial features that define force are left unanalyzed. None of the above theories give a formal semantics for nonassertive speech acts. As Searle also states, what is missing is formal theory of iutentions. In this paper we outline a formal semantics and pragmatics of speech acts based on an explicit formal theory of information and intention. A formal description /0f the notion ofillocutionary force is given. We take a new look at Searle's a classification of speech acts. In the process, we develop a deeper understanding of the rather vague notion of force.
We start with Searle's classification of illocutionary acts. Searle and Vanderveken's definition of force is criticized. Next Habermas is criticized. Next, referential theories of meaning are looked at critically. Then we present our theory of the semantics and pragmatics of speech acts. We then use that theory to give a theory of force. We then describe the speech a c t s in Searle's classification in terms of the semantic and pragmatic effects they have in the communicative situation. Finally we show how the semantics and pragmatics of speech acts is related to the general communcative process in the social world and show that the class!fication of speech acts really reflects the fundamental way agents use language to coordinate and make possible their social actions. 2. Searle's Classification of lllocutionary Acts What distinguishes a request from an assertion? One answer is that their force is different. But what is force? According to Searle, when humans communicate they are engaged in an activity. An utterance, according to Searle, can be broken down into two basic components, the illocutionary force F and the propositional content p. The utterance is symbolized as F(p). In order to classify the different types of force F , Searle and Vanderveken [85] attempt to reduce the force of a speech act to more primitive features. The force and the propositional content is then used to divide speech acts into six general classes. In Searle's classification of illocutionary acts [Searle 75] he distinguishes six general classes based on four dimensions. The four dimensions were the illocutionary pob~t (assertive t-, directive !, commissive C, expressive E ~declarative D, and representative declarative Dr ), the direction of fit (word-to-world $ , world-to-word t ), the psychological state (believes B, wants W, intends D, and the propositional content p. The null symbol was used to indicate, that a given dimension had no instance. The following speech act classes were distinguished: 1. Assertives: t- $ B( p ) This says that the assertive t- (the illocutionary point) has a direction of fit of word-to-world $ , the psychological state B for the sincerity condition to hold must be that the speaker belives B the propositional content expressed p. Example: The speaker states The door is open and belives that the door is open.

2. Directives: ! ~'W(H does A) This states that the directive ! has a direction of fit of world-to-word ~, the psychololgical state must be that the speaker wants that the hearer H do the action A. Example: The speaker gives the command Open the door and wants the door to be opened. Nel~, here, as in the declaratives, we need certain role presuppositions to hold. 3. Commissives: C ~' I(S does A) The commissive C has a direction of fit of world-to~.word ~", the puychololgical state must be that the speaker S intends I to do the action A. Example: The speaker says I will open the door and intends to do it. 4. Egpressives: E 0p(P)(S/H + Property) The expressive F, has no direction of fit p, the psychological state should correspond to the emotion/attitude expressed by the speaker about the property of the spearker S or hearer H. E~ample: The speaker exclaims I like your coat and means il;. 5. D~clarations: D ~-*(p) The declaration D has a self-realizing direction of fit * (since the state described by the propositional content p is realized by the very act of saying it). There is no pyschological state necessary and, hence, no sincerity condition. However, the speaker mu.~t have a certain institutional role, e.g., In saying I resign or You're fired the speaker must have the role of employee or boss, respectively. 6. Representative Deelaratives: Dr 4~B(P) The represe~ltative declarative has a direction of fit of word-to-world $ ( since the speaker bases his declaration on the recognition of some fact p ), and a self-realizing directions of fit ~ (like declarations). There is a psychological state in the sincerity condition that the speaker must believe B the proposition expressed. Example: The umpire declaring, You i'e out or the judge, I find you guilty as charged.
the utterance. For the utterance is understood by the hearer whether the speaker wants the action or not. For example, it may well be that a general is forced to order his troops to war, after a declaration of war by the president. The general's command still has its meaning and effect irrespective of his personal wants. And, indeed, a little reflection should convince the reader that the want W does not contribute to the differentiation of the typology. So now we have reduced the force of a directive to a single parameter value the direction of fit ~. If we are so impertinent as to ask W h a t is the direction of fit?, we will find no ready answer that is not circular. The direction of fit of a directive is "world-towords" ~. In other words, the force of a directive is to change the world using words. Or, better, to influence an agent to change the world by conununicating a directive to that agent. In effect, in spite of all the symbols, the force remains essentially an unanalyzed notion. Little more is said than that the force of a directive is differenct from the force of an assertion. Nothing new there. Directives are used to change the world. Assertions are used to describe the world. True, but it is not enlightening. Hence, we have shown some of the dimensions to be redundant, vague and of questionable computational usefulness. We have also shown that the attempt to reduce the point feature to the remaining features leads to a dead end. At t h e h e a r t of t h e s e p r o b l e m s lies a more devastating pathology: We have seen that the point of the speech act is redundant for Searle's classification of speech acts. However, the point of the speech act is that part of the force that gives the central purpose of the act. Searle and Vanderveken state that the point of a speech act is a clear enough concept to be used as a primitive notion in the formal theory. '~lYhenotion of illocutionary point is the fundamental undefined primitive notion of illocutionary logic." [p.fl7, Searle and Vanderveken 85] Yet a few lines later they say the analysis of the point of speech acts requires a theory of mental states, specifically, a theory of intentionality which they say is beyond the scope of the book. Thus, the point of a speech act is a very complicated notion that requires a formal theory of psychology before that notion can be made formally rigorous. They also state that the point of a speech act is the most important component of the force [p.14, Searle and Vanderveken 85]. Since the force of a speech act is the most important element that differentiates speech act theory from traditional referential semantics, the point of the speech act is the central theoretical notion in the foundations of the theory of speech acts. Yet it is the point of the speech act that is left unanalyzed in Searle's formal account. To sum up, the features constituting the notion of force are not motivated by any coherent theory of the m e a n i n g of speech acts. As Searle and Vanderveken admit, they have no semantics for the two most central features in the definition of force, namely, the point and direction of fit of the speech act. Instead, they leave these notions p r i m i t i v e. a n d unanalyzed. That, however, amounts to leaving the notion of force an unanalyzed concept. As Searle himself states, a proper theory of force -requires a theory of intention. We outline such a theory of intention in this paper and use it to formally define the force of a speech act. 745

3. Critique of Searle's Notion of Force Note that the actual classes of illocutionary acts that Searle distinguishes correspond exactly to the values of the illocutionary point dimension, it turns out that there is a one to one correspondence between the illocutionary point and the type of speech act. it should be evident that the point of the ~Lpeechact is just another name for the speech act type a~d does not add any more d i s t i n g u i s h i n g inibrmation for the individuation of that type from the others. Hence, one can conlude that it is the information in the remaining three dimensions ( the direction of fit, pyschological state and propositional content) that really distinguish the speech act types considered by Searle. One is still left with the nagging question: But, what is the force of an utterance? Well, if we have an utterance u == F(p) ,and, if u is a directive, then u = ! I"W(H does A)~ where the force F = I~'W and p = H d o e s A. But, if we are right about the irrelevance of the illocutionary point ! to t~e classification, then the force of a directive becomes F : ~W and the utterance n becomes u = F(p) :~ ~'W(H does A). However, one can plausibly argue that the sincerity condition W, that the speaker want the action indicated by the propositional content, is really not part of the force of
4. Speech Acts in C o m m u n i c a t i o n The necessity of social cooperation gives birth to communication. The reason that utterances have the effect of actions is because they influence the cognitive state of the conversants. It is the harmony of the cognitive states of agents that makes possible cooperative social action and forms the basis of society [Werner 88 a, b]. On our view the meaning of the speech act is best understood if we understand how the speech act is meant to influence the cognitive states of the conversants in the context of a social activity. The force of a speech act lies in its unique distribution of effect on the cognitive substates of the social agents. One objection to our view may be that the theory of how a speech act effects the h e a r e r is t h e s t u d y of perl0cutionary effect. The perlocutionary effect is subject to the idiosyncrasies of individual performance and understanding and, therefore, cannot be the meaning of the speech act. We think differently. One must make a distinction, analoguous to Chomky's for syntax, between the ideal competence of the understanding subject (i.e., t h e ability of the subject to understand the speech act) and the actual cognitive performance. The meaning of a speech act is described by how it is to effect the ideal cognitive state of the conversants, given t h a t the message is accepted. (see Perrault [87] for a similar view) 5. Critique of H a b e r m a s Habermas [81] suggests t h a t to get a theory of meaning for all sentences of a natural language, we have to generalize truth conditions to general satisfaction conditions. This would imply that assertions are given t r u t h conditions, moral assertions being right or wrong are given r i g h t n e s s c o n d i t i o n s and~intention statements being sincere or not are given sincerity conditions. Since comands are legitimate or not they would, presumably, be given l e g i t i m a c y conditions. H a b e r m a s ' hypothesis is based on a confusion between presuppositions (rightness, sincerity, legitimacy), truth conditions, and a theory of meaning. The general problem is that the rightness, sincerity and legitimacy conditions are not sufficient to specify and differentiate the meaning of one sentence from another. The general satisfaction conditions are inadequate for meeting the requirements of a semantic theory. Consider the example, I will go to the m a r k e t tomorrow. Here the intention expressed may or may not be sincere. Even if it is sincere, this sincerity in itself does not distinguish it from I a m g o i n g to go fly a kite. The content and point of the sentence is missed. The sincerety condition is independent of the meaning of the sentence. It says nothing of the semantic t h e o r y that specifies the semantic content and pragmantic point.

I will V P , where VP is some verb phrase describing some action, is sincere if the speaker's intentions SA are modified to be in accord with the meaning of the sentence. Y o u s h o u l d not have done X is right if the act X is not allowed by the prevailing norms. Here again the rightness by itself is not enough to determine the meaning of the normative claim. The comand Attack! is legitimate or not depending on if the requisite roles and a u t h o r i t y relations exist and are accepted by all parties concerned.
A t t a c k ! from R e t r e a t / n o r from any other comand.
But this legitimation condition does not distinguish The legitimation claim that is raised and presupposed by giving a comand does not specify the meaning of the comand itself. There is a false analogy drawn between the general satisfaction conditions and truth conditions. While truth conditions do differentiate assertive sentences that have distinct meanings, a truth conditional semantics does not give the meaning of a sentence explicitly. They do not construct a semantical object that is meant to be the sentence's meaning. Classical non-situational semantics of basic sentences did not differentiate the meanings of sentences except indirectly through differences in the content and form of the truth conditions. However, in s i t u a t i o n semantics, where basic sentences refer to situations, we can refer to and differentiate sentence meanings directly: And that is point! The meaning of the basic sentence is already given by the deeper situation s e m a n t i c s. The t r u t h c o n d i t i o n s a r e d e f i n e d by presupposing the deep meaning of the sentence is already given. So too for all the g e n e r a l i z e d s a t i s f a c t i o n conditions. The deep s e m a n t i c s and the p r a g m a t i c interpretation of the sentence in question is presupposed in the specification/defintion of the satisfaction conditions. A crucially important conclusion follows from the above argument: It is that conditions of use of a sentence in the sense of Wittgenstein, Austin, Searle, Cohen and Perrault and Habermas, are only indirectly related to the meaning of an utterance. Meaning is not identical with the conditions of use of the sentence. It is not identical to the preconditions of the speech action. Instead, the conditions of use of a sentence presuppose the situational and pragmatic meaning of the sentence.

6. P s y c h o l o g y a n d F o r c e
B r i n g me a b r i c k / a n d Y o u brought me a brick may both refer to the same action. While their semantic reference is the same, their force is clearly distinct. The example shows that the meaning of a sentence cannot be identified with its reference. For in the example both sentences have the same reference, namely, the action of bringing the brick. The difference in meaning lies in the difference in the force, and, as we will see below, in the difference in their pragmatic interpretation.
To explain this difference in force we will give a pragmatic interpretation to the sentence. A eomand steers action by affecting the intentions of the hearer. Its point is to affect those intentions. The assertion, on the other hand, affects the information state of the hearer and its point is to give information. This difference of force is not explainable by a purely referential theory, It requires explicit or implicit reference to mental states. Thus, if we take situation semantics to be the claim that the meaning of a sentence is the external situation it refers to, then situation semantics can give no explanation of force. If it tries to, it must bring in mental states. Thereby, it is no longer a purely referential theory. However, as we will see s i t u a t i o n s e m a n t i c s s e r v e s as a f o u n d a t i o n for understanding and defining the propositional content of speech acts.
9. C o m m u n i c a t i o n
A proper theory of speech act force requires a f o r m a l ~heory o f psyehology~ Such a theory of psychology must bc tbrmal because its variables, mental states, will be used [n the ibl~ml analysis of the point. Furthermore, a formal psychology should sh6w the t h e o r e t i c a l r e l a t i o n s h i p s between the mental states. Thus these mental states must h a v e a su:fficient c o m p l e x i t y to a c c o u n t for t h e s e inb~rrelatio~as and to explain the linguistic phenomena. Such a theory should provide the conceptual foundations ~aeeded for the central and most i m p o r t a n t concept in ;~peech act f;heory, in addition, it must account for the i'elationship between linguistic communication and social cooperation, To get a theory of meaning for speech acts we thus will need a theory of the intentional, informational, evaluative, a n d m o r e g e n e r a l l y , t h e c o g n i t i v e s t a t e s of t h e convcrsants. For a more detailed account of the general theory ofeommunication see Wcrner188a]. 7. C o g n i t i v e States Let TP be the set of t i m e p e r i o d s t where t i m e i n s t a n t s ar~ordered by a relation <. Let Hist(Q) be the setofpossible p a r t i a l h i s t o r i e s ttt up to time t. Let Q be the set o f a l t complete h i s t o r i e s or w o r l d s H. Worlds are series of complete situations. (See Barwise and Perry [83].) A situation s is r e a l i z e d in H at t if s is contained in H and the domaine ofs is time period t. STRAT is the set of all possible strategies, a * is the set of possible histories consi,~tent ~vith the s t r a t e g y n The c o g n i t i v e or representational s t a t e R of a c o n v e r s a t i o n a l p a r t i c i p a n L cp is described by three components R -= <I, S , V >. ~ is the i n f o r m a t i o n state of the c p. I is ibrmalized as a set of possible partial histories. S is the intentim~a~ state of the cp. S is a set of possible strategies t.hat gaide the actions of the cp. V is the e v a l u a t i v e state of the cp. Y represents the cp's evaluation and focus on situations. The representational state RA may include the agent A's representation of B's representation, RBA. It may also i n c l u d e t h e a g e n t A ' s r e p r e s e n t a t i o n of B's representation of A's representation, Rn^^. Thus we can represent arbitrary levels of nesting of representations. Let I N F , INT, and E V A L be the set of all possible informati.n, intentional, and evaluative states, ~espectively. Let the r e p r e s e n t a t i o n a l c a p a c i t y , Rep, be the set of all possible representational states of a cp. For the theoretical foundations and some of the principles i n t e r r e l a t i n g i n t e n t i o n , i n f o r m a t i o n and a b i l i t y see W e r n e r [88c].

9.1 S y n t a x

To illustrate how the representational semantics works we develop the pragmatic interpretation for a small temporal propositional language fragment we will call Lpc The l a n g u a g e Lpt will include logical and t e m p o r a l connectives: A ( = a n d ) , V (--- or), -1 ( = n o t ) , A (= and then), while (= while). From a t o m i c f o r m u l a s p ,q, we build up c o m p l e x f o r m u l a s a A ~ , a V ~, -7 a, a A ~ i~, and u while ~ in the usual way.

9.2 P r a g m a t i c

Interpretation

of A s s e r t i o n s

For any formula a in L , Prag(a) : Rep ~ Rep is a function that distributes over the representational state R E R e p subject to the constraints that for all I ( INF , S E I N T , a n d V E V A L , t h e n Prag(a)(I) E INF, Prag(a)(S) E INT, and Prag(a)(V) E VAL , respectively. Below we assume Holds( a, H, t) is defined by induction in the usual way where for atomic formulas Holds( a, H, t) if the situation referred to by a is realized in H at t. P r a g must additionally satisfy other conditions : For atomic formulas Prag(a) : R e p R e p. Let Prag be defined for the formulas a and ~. P r a g acts on information states as follows: Prag(a A [~)(I) = Prag(a)(I) f) Prag(~)(I) Prag(a V ~)(I) -= Prag(a) (I) U Prag(~)(I) Prag(-1 a)(I) = I - Prag(a)(I) Prag (a A ~ 13)(I) -= {Ht : Ht e I and there exist times to, t' e TimePeriods, TP, where Holds( a, H, to) and Holds( ~, H, t') and to < t'} P r a g (a while ~)(I) =- {Ht : H t ~ I and for all to, t' e T P , if to contains t' then if Holds( [~, H, t') then Holds( a, H, to)} For example, the pragmatic interpretation of the sentence a = 'Jon opened the door' is arrived at as follows: n refers to the event of Jon opening the door. Prag(a) is an operator on the hearer's information state I such that Prag(a)I is the reduction of the set I to those histories where the event referred to by n occurred. The hearer A knows u if a holds in all the worlds in I. Thus, A comes to know that a as a result of receiving and interpreting the message a.

9.3 l ' r a g m a t i c

of Directives
8. R e p r e s e n t a t i o n a l S e m a n t i c s A theory of the meaning of speech acts is a theory of how the representations of the cp are updated by the co~t~nunicative process. The propositional content of the speech act ~s given by a s i t u a t i o n s e m a n t i c s in the style of l~arwise and P e r r y [83]. We call t h e t h e o r y of how representational or cognitive states are transformed by messages the pragmatics or r e p r e s e n t a t i o n a l ~ e m a n t i c s. Given a language L we define a pragmatic operator P r a g , where for each sentence a in L, Prag(a) is a function from R e p into R e p. Thus Prag takes a given sttb~'epresentational state such as I in R and transtblzns i t i n t e a new substate I' = Prag(a)(I).

P r a g acts on the intentional substates as follows: Prag(a A 13)(S) = Prag(a)(S) N Prag(13)(S) Prag(u V ~)(S) -- Prag(o)(S) U Prag(~)(S) Prag(-7 a ) (S) = S - Prag(u)(S) P r a g (a h=~ l}) (S) -- {n : for all H e n * and there exist times to, t' ~ TP where Holds( u, H, to) and Holds( 13, H, t') and to < t'} P r a g (o while 13)(S) = { n : for all H e n*, exists t, t' e TP such that Holds( a, H, t) and Holds( a, H, t') and t' contains t}.
For example, if a: = 'Open the door' , a refers to the situation of the addressee A opening the door. Prag(a) operates on A's intentional state SA such that A opens the door. P r a g does this by removing all those possible plans of A t h a t do not force a. Viewed constructively, a plan is incrementally built up by the Prag algorithm. And those are the plans n that have some world H e n * where the situation referred to by a is not realized in H. The result is that the agent performs the directive in parallel to other goals he may have. Again, we are talking about the ideal pragmatic competence. Note that Prag describes the pragmatic c o m p e t e n c e of an ideal s p e a k e r and not the a c t u a l performance. He may for various reasons not accept the message. But for him to u n d e r s t a n d the conventional m e a n i n g of the assertion or directive, the conversational p a r t i c i p a n t must know what the effect of the message is supposed to be if he were to accept it, Thus, a Participant will not just have an actual informational and intentional state I and S but also hypothetical representational states HI and HS that are used to compute the pragmatic effect of a given message. If the participant then accepts the message , HI or HS will become a part of the actual representational state R = (I, S, V).
3. S t a t e m e n t s of I n t e n t i o n :
(IB,SB, VB) ~ (acta~aIB, faSB VB) A,
4. S t a t e m e n t s of Value: (IB, SB, VB) (acta~aIB, SB, faVB^) B e l o w we w i l l s i m p l i f y m a t t e r s by u s u a l l y suppressing the action operator acts as well as the time operator ~a where the above cases indicate how the a more complete description would look like. Note too, that the discourse situation d can be viewed as yet a n o t h e r operator acting on the information state ICp of the coversational participants cp = A , B. Since there are points of v i e w , it is clear that the operator d is also a function of the c o n v e r s a t i o n a l p a r t i c i p a n t and more generally of the role of the cp. Similarly, as we will see, the pargmatic operator is also a function of the role of the cp in question.

11. S p e e c h A c t Theory We now give a semantic, pragmatic description of some of the speech acts in Searle's taxonomy. First some needed definitions. Given our formal pragmatics Prag for an utterance < s, a , h > , where s is the speaker, a is the sentence expressed and h is the hearer in the discourse situation d, let the speaker s have representational state Rs = (Is , Ss , Vs ) and the h e a r e r h have representational state Rh = ( I h , Sh , Vh ). The different kinds of speech acts can be differentiated by how t h e y effect the c o g n i t i v e state of t h e c o n v e r s a n t s. Specifically, the f o r c e o f a s p e e c h a c t is the set of subrepresentations in R that are to be transformed by the speech act. An intentional state S f o r c e s a g o a l g, in symbols, S n g i f f f o r a l l n e S, n forcesg, i.e.,ifffor all H ~ n*, g is realized in H. By definition the composite S + S' of two intentional states S, S' together f o r c e a g o a l g, in symbols, S + S' I1~ g ifffor all n e S , nO e S', and for all H ~ n* Cl no*, g is realized in H. Below we will use the shorthand notation of o1 for Prag(a)I = fa[ 1. A s s e r t i v e s : 1.1. Ih a ~ OIh Example: Billopened the door. 1.2. Is h a ~ OIsh
10. P r a g m a t i c O p e r a t o r s The interpretation of u t t e r a n c e s is actually more complicated. For a given formula a are associated several operators t h a t give different kinds of information. Let fa = Prag( a ) the i n t e r p r e t a t i o n of a. To describe the a c t o f s a y i n g a we introduce another operator, acts is the operator giving the infomation that a was just said. ~a is the t i m e o p e r a t o r that shifts time according to how long it took to say u. Here we assume for simplicity that the act of u t t e r i n g a takes one unit of time. We could of Course relativize the time operator to a. Note, ~a~a ~ ~a C o m b i n i n g operators, acta~a is then the a c t o f u t t e r i n g a Note, acta~a = ~aactQ. f, acta~a is the act of uttering and interpreting a. We have several cases possible when A asserts the informative a to the audience B. 1. If B hears, understands and accepts a then IB =I, f, acta~aIB. 2. If B hears, but does not understand or accept a , then IB acta~aIB. Note, if a is a directive then fa acts on SB and not on Ill. Yet it is known what was said. So here too,
1.3. Is Ir~ a Remark: Ih transforms to oI h Assertives effect the informational state of the hearer. They also effect the h e a r e r ' s representation of the s p e a k e r ' s beliefs The sincerity condition t h a t the s p e a k e r b e l i e v e w h a t he asserts is expressed by Is r l a. 2. D i r e c t i v e s : 2.1. Sh a~ Example: Open the door/ aSh 2.2. Ss h a ~ ass h

IB ~ acta~aIB.

3. I f B hears something was said, but does not know w h a t was said, then IB ~aIB 4. More odd is the case where B does not hear a but gets the information of what was said subliminally and knows time passed. Then Ill ~ fQ~aIll. The full interpretation of a for the a u d i e n c e B depends on its force. Let Rll =(IB, SB, VB). Given that the message is accepted, some the cases are as follows: 1. A s s e r t i o n s : (Ill, SB, VB) ~ (faacta~aIB, SB, VB) 2. C o m a n d s : (Ill,SB, VB) ~ (actaxaIB, faSB, VB)

2.3 Icp ~ acta~aIcp Remark: The comand updates the hearer's intentions to aS h where h does the action a. aS sh describes the speaker's representation of the heater's new intentions. 3. C o m m i s s i v e s : Example: I will open the door. 3.1. Ss a ~ a s s 32 Sh s a ~ aShs Remark: The speaker commits himself to following those strategies that insure the propositional content of a , i.e., all the worlds in each. * realize the action referred to by a. aS hs represents the hearer's resulting representation of the speaker's modified intentions.
4. D e c l a r a t i o n s : Example: [ resign, Your fired, 4.1. 4.3. 4.5. Ih a ~ Ulh 4.2. I s u=> UIs
information, intentional, and evaluative states requires making significantly more detailed system design decisions. Within a wider context, we have aimed at providing a general theoretical framework for designing systems with a communicative competence using natural language. Our theory fits well with planning models in robotics. It also fits well with discourse and speech act theories. Of more g l o b a l s i g n i f i c a n c e is t h e fact t h a t o u r t h e o r y of c o m m u n i c a t i o n f i t s w e l l w i t h yon N e u m a n n a n d Morgenstern's economic theory [see yon N e u m a n n and M o r g e n s t e r n 47]. T h i s is b e c a u s e o u r t h e o r i e s of communication and intention allow us to define complex social roles and social structures [Werner 88a, 88b]. This will hopefully help to clarify the complex relationship between language and society. References Appelt, D. E., PLANNING ENGLISII SENTENCES; Cambridge University Press, New York, 1985. Barwise, J., and Perry, J., SITUATIONS AND ATTITUDES, Bradford Books/MIT Press, 1983. H. Bergmann/M. Gerlach Semantisch-pragmatische Verarbeitung yon.~,uBerungen im natiirliehsprachtichen Beratungssystem WISBER, in: W. Brauer, W. Wahlster (Eds.): Wissensbasierte Systeme - G1Kongress 1987. Springer Verlag, Berlin 1987, pp. 318-327 Cohen, P. R., and Perrault, C. R., "Elements of a Plan-Ba~ed Theory of Speech Acts," COGNITIVE SCIENCE, 3, pp. 177 - 212, 1979. Grosz, B. J., "The Structures of Discourse Structure," Techn. Note 369, Artificial Intelligence Center, SRI International, Menlo Park, California, 1985. Habermas, J., TIIEORIE DES KOMMUNIKATIVENflANI)ELNS, Frankfurt-a.M., 1981. Moore, R. C., "Reasoning About Knowledge and Action", Tech. Note 191, Artificial Intelligence Center, SRI International, Melno Park, California, 1980. Montague, R., "The Proper Treatment of Quantification in Ordinary English", In Thomason, R., (ed.), FORMAl, PtIILOSOPIIY: Selelcted Papers of Richard Montague, NewHaven: Yale University Press, pp. 247-270, 1974. Perrault, C.R., "An Application of Default Logic to Speech Act Theory",Report No. CSLI-87-90, CSLI, SRI International, Menlo Park, California, 1987. Searle, J. R., "A Taxonomy of Illocutionary Acts", in K. Gundm'son (cd.), LANGUAGE, MIND, AND KNOWLEDGE, MINNESOTA STUDIES IN TIIE PIIILOSOPHY OF SCIENCE, Vol. 7, Minneapolis: University of Minnesota Press, 1975. Searle, J. R., and Vanderveken, FOUNDATIONS OF ILLOCUTIONARY LOGIC Cambridge University Press, London, 1985. Von Neumann, J. and Morgenstern, O., THE THEORY OF GAMES AND ECONOMIC BEHAVIOR , Princeton University Press, Princeton, NJ, 1947. Wernor, E., "Toward a Theory of Communication and Cooperation for Multiagent Planning", TttEORETICAL ASPECTS OF REASONING ABOUT KNOWLEDGE: Proceedings of the Second Conl'erence,Morgan Kaufman Publishers, 1988a. Werner, E., "Socializing Robots: A Theory of Communication and Cooperation for Distributed Artificial Intelligence", to be presented at 1988 Distributed Artificial Intelligence Workshop, Lake Arrowhead, California, 19988b. Werner, E., "Intention, Information and Ability: Outline of a Unified Theory", unpublished manuscript, 1988c. 7z~9

Ss u ~ ass4.4.

S h a@ a s h
S institution n ~ uS institution
R e m a r k : The both h e a r e r and s p e a k e r update t h e i r information states to UIh and OIs, respectively, where they know the resulting state brought on by the declaration. Furthermo:ce, a declaration such as "you're fired" has specific intentional consequences such as no longer being paid. uS im~titution indicates that the declaration also has i n s t i t u t i o m d effects. Namely, it effects the composite i n t e n t i o n s of all t h o s e w i t h roles i n v o l v e d in the employment relationship. As we mentioned above, the P r a g operator is also a function of the r o l e the cp has. Viewed from another perspective, the stone speech act has d i f f e r e n t effects on the i n t e n t i o n s of d i f f e r e n t cp's according to the roles they occupy. 5. R e p r e s e n t a t i v e D e c l a r a t i v e s : Example: I find you guilty. 5.1. 5.3. 5.5. 5.6. Ih 0 OIh 5.2. 5.4. Is a ~ O/s
Ss u::~ oSs S institution a ~ Is II~ u

Sh u =~ a S h

aS institution
Remark: The representative declarative differs from the declaration in that the former must be based on certain facts obtaining. Is I1~ a expresses this condition. Again we see how social roles in an institution are affected by a declaration. The judge's declaration of guilt and sentencing has very specific intentional consequences for the police and parole board, etc. These complex intentions are packed into the composite institutional role structure aSinstitution W h a t is so interesting is that our formalism allows us t,~) talk about such complex social processes. I t takes a small step toward a better understanding of the relationship between linguistic communieatibn and social structure, lit is this property of our theory that makes it a promising candidate for the design of the complex systems being contemplated in distributed artificial intelligence.
12. C o n c l u s i o n We have developed the outlines of a formal theory of m e a n i n g (semantics and pragmatics) of speech acts. To a c c o m p l i s h t h i s we d e v e l o p e d a f o r m a l t h e o r y of i n t e n t i o n a l s t a t e s. We t h e n r e l a t e d l a n g u a g e to information and intention states. We found that meaning is not identical to conditions of use, and that meaning is n o t i d e n t i c a l to r e f e r e n c e. The f o r m a l t h e o r y of information and intention m a d e it possible for us to construct a rigorous theory of the force of illocutionary acts. The illocutionary force is defined in terms of the specific s u b r e p r e s e n t a t i o n s t h a t the speech act is to modify. The subrepresentations are only sketched. But the point ef the approach is quite clear. The cognitive states of the conversational participants, for example, system and user, play a dominant role in the theory of m o a n i n g a n d f o r c e of s p e e c h a c t s. An a c t u a l implementation of an algorithm for Prag and an actual knowledge representation scheme to describe the