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 semantically 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 fundamental 
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 acts 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. 
744 
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. 
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 ~Lpeech act is just another name for the speech 
act type a~d does not add any more distinguishing 
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 = HdoesA. 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 
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 What is the 
direction of fit?, we will find no ready answer that is not 
circular. The direction of fit of a directive is "world-to- 
words" ~. 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 the heart of these problems 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. '~lYhe notion 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 
meaning 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 primitive.and 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 
4. Speech Acts in Communication 
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 hearer is the study 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., the 
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 that the message is 
accepted. (see Perrault \[87\] for a similar view) 
5. Critique of Habermas 
Habermas \[81\] suggests that 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 
truth conditions, moral assertions being right or wrong 
are given rightness conditions and~intention statements 
being sincere or not are given sincerity conditions. Since 
comands are legitimate or not they would, presumably, be 
given legitimacy conditions. 
Habermas' 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 market 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 am 
going 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 
theory that specifies the semantic content and pragmantic 
point. 
I will VP, 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. 
You should 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 authority 
relations exist and are accepted by all parties concerned. 
746 
But this legitimation condition does not distinguish 
Attack! from Retreat/nor from any other comand. 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 
situation 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 
semantics. The truth conditions are defined by 
presupposing the deep meaning of the sentence is already 
given. So too for all the generalized satisfaction 
conditions. The deep semantics and the pragmatic 
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. Psychology and Force 
Bring me a brick/and You 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 
situation semantics serves as a foundation for 
understanding and defining the propositional content of 
speech acts. 
A proper theory of speech act force requires a formal 
~heory of psyehology~ Such a theory of psychology must 
bc tbrmal because its variables, mental states, will be used 
\[n the ibl~ml analysis of the point. Furthermore, a formal 
psychology should sh6w the theoretical relationships 
between the mental states. Thus these mental states must 
have a su:fficient complexity to account for these 
inb~rrelatio~as and to explain the linguistic phenomena. 
Such a theory should provide the conceptual foundations 
~aeeded for the central and most important concept in 
;~peech act f;heory, in addition, it must account for the 
i'elationship between linguistic communication and social 
cooperation, 
To get a theory of meaning for speech acts we thus will 
need a theory of the intentional, informational, evaluative, 
and more generally, the cognitive states of the 
convcrsants. For a more detailed account of the general 
theory ofeommunication see Wcrner188a\]. 
7. Cognitive States 
Let TP be the set of time periods t where time 
instants ar~ordered by a relation <. Let Hist(Q) be the 
setofpossible partial histories ttt up to time t. Let Q be 
the set ofalt complete histories or worlds H. Worlds are 
series of complete situations. (See Barwise and Perry \[83\].) 
A situation s is realized 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 strategy n The cognitive or 
representational state R of a conversational 
participanL cp is described by three components R -= <I, 
S, V> . ~ is the information state of the cp. 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 evaluative 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 include the agent A's representation of B's 
representation of A's representation, Rn^^. Thus we can 
represent arbitrary levels of nesting of representations. 
Let INF, INT, and EVAL be the set of all possible 
informati.n, intentional, and evaluative states, 
~espectively. Let the representational capacity , Rep, 
be the set of all possible representational states of a cp. 
For the theoretical foundations and some of the principles 
interrelating intention, information and ability see 
Werner \[88c\]. 
8. Representational Semantics 
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 situation semantics in the style of 
l~arwise and Perry \[83\]. We call the theory of how 
representational or cognitive states are transformed by 
messages the pragmatics or representational 
~emantics. Given a language L we define a pragmatic 
operator Prag, where for each sentence a in L, Prag(a) 
is a function from Rep into Rep . Thus Prag takes a 
given sttb~'epresentational state such as I in R and 
transtblzns itinte a new substate I' = Prag(a)(I). 
9. Communication 
9.1 Syntax 
To illustrate how the representational semantics 
works we develop the pragmatic interpretation for a small 
temporal propositional language fragment we will call Lpc 
The language Lpt will include logical and temporal 
connectives: A (=and), V (--- or), -1 (=not), A¢ (= and 
then), while (= while). From atomic formulas p ,q, we 
build up complex formulas a A ~, a V ~, -7 a, a A ~ i~, 
and u while ~ in the usual way. 
9.2 Pragmatic Interpretation of Assertions 
For any formula a in L , Prag(a) : Rep ~ Rep is 
a function that distributes over the representational state 
R E Rep subject to the constraints that for all I ( INF , S 
E INT,and V E VAL,then 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. 
Prag must additionally satisfy other conditions : For 
atomic formulas 
Prag(a) : Rep ¢ Rep . Let Prag be defined for the 
formulas a and ~. 
Prag 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'} 
Prag (a while ~)(I) =- {Ht : Ht~ I and for all to, 
t' e TP, 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'ragmatic Interpretation of Directives 
Prag 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) 
Prag (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'} 
Prag (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}. 
747 
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. Prag does this by removing all those possible plans 
of A that 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 
competence of an ideal speaker and not the actual 
performance. He may for various reasons not accept the 
message. But for him to understand the conventional 
meaning of the assertion or directive, the conversational 
participant 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). 
10. Pragmatic Operators 
The interpretation of utterances is actually more 
complicated. For a given formula a are associated several 
operators that give different kinds of information. Let fa 
= Prag( a ) the interpretation of a. To describe the act 
- of saying a we introduce another operator, acts is the 
operator giving the infomation that a was just said. ~a is 
the time operator that shifts time according to how long it 
took to say u . Here we assume for simplicity that the act 
of uttering a takes one unit of time. We could of Course 
relativize the time operator to a . Note, ~a~a ~ ~a 
Combining operators, acta~a is then the act of uttering 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, 
IB ~ acta~aIB . 
3. If B hears something was said, but does not know 
what 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 audience B 
depends on its force. Let Rll • =(IB, SB, VB). Given that 
the message is accepted, some the cases are as follows: 
1. Assertions: (Ill, SB, VB) ~ (faacta~aIB, SB, VB) 
2. Comands: (Ill,SB, VB) ~ (actaxaIB, faSB, VB) 
748 
3. Statements of Intention: 
(IB, SB, VB) ~ (acta~aIB, faSB A, VB) 
4. Statements of Value: 
(IB, SB, VB) ¢ (acta~aIB, SB, faVB ^) 
Below we will simplify matters by usually 
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 another 
operator acting on the information state ICp of the 
coversational participants cp = A, B. Since there are 
points of view, it is clear that the operator d is also a 
function of the conversational participant 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. Speech Act 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 hearer h have 
representational state Rh = (Ih, Sh , Vh ). The 
different kinds of speech acts can be differentiated by how 
they effect the cognitive state of the conversants. 
Specifically, the force of a speech act is the set of 
subrepresentations in R that are to be transformed by the 
speech act. An intentional state S forces a goal g, in 
symbols, Sn¢ giffforall 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 force a goal 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. Assertives: Example: Billopened the door. 
1.1. Ih a~ OIh 1.2. Is h a ~ OIs h 
1.3. Is Ir~ a 
Remark: Ih transforms to oI h • Assertives effect the 
informational state of the hearer. They also effect the 
hearer's representation of the speaker's beliefs• The 
sincerity condition that the speaker believe what he 
asserts is expressed by Is rl¢ a . 
2. Directives: Example: Open the door/ 
2.1. Sh a~ aSh 2.2. Ss h a~ ass h 
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. Commissives: Example: I will open the door. 
3.1. Ss a~ ass 3•2• Sh s a~ aSh s 
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 h s represents the hearer's resulting representation of 
the speaker's modified intentions. 
4. Declarations: 
Example: \[ resign, Your fired, 
4.1. Ih a~ Ulh 4.2. I s u=> UI s 
4.3. Ss u~ ass4.4. Sh a@ ash 
4.5. S institution n~ uS institution 
Remark: The both hearer and speaker update their 
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 
institutiomd effects. Namely, it effects the composite 
intentions of all those with roles involved in the 
employment relationship. As we mentioned above, the 
Prag operator is also a function of the role the cp has. 
Viewed from another perspective, the stone speech act has 
different effects on the intentions of different cp's 
according to the roles they occupy. 
5. Representative Declaratives: 
Example: I find you guilty. 
5.1. Ih 0¢ OIh 5.2. Is a~ O/s 
5.3. Ss u::~ oSs 5.4. Sh u =~ aSh 
5.5. S institution a~ aS institution 
5.6. Is II~ u 
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 • What is so interesting is that our formalism 
allows us t,~) talk about such complex social processes. It 
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. Conclusion 
We have developed the outlines of a formal theory of 
meaning (semantics and pragmatics) of speech acts. To 
accomplish this we developed a formal theory of 
intentional states. We then related language to 
information and intention states. We found that meaning 
is not identical to conditions of use, and that meaning is 
not identical to reference. The formal theory of 
information and intention made 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 subrepresentations that 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 
moaning and force of speech acts. An actual 
implementation of an algorithm for Prag and an actual 
knowledge representation scheme to describe the 
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 
global significance is the fact that our theory of 
communication fits well with yon Neumann and 
Morgenstern's economic theory \[see yon Neumann and 
Morgenstern 47\]. This is because our theories 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 - G1- 
Kongress 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 KOMMUNIKATIVEN flANI)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. 
