Candace L. Sidner and David J. Israel. Bolt Beranek and Newman Inc. ... beliefs and wants [Cohen, 1978]) as well as a framework for reasoning about the ...
RECOGNIZING INTENDED MEANING AND SPEAKERS' PLANS
Candace L. Sidner and David J. I s r a e l B o l t Beranek and Newman I n c . 10 Houlton S t . Cambridge r MA 02138
2. INTRODUCTION
ABSTRACT Human 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 s depend upon the a b i l i t y o f t h e i r p a r t n e r s t o recognize t h e i r i n t e n t i o n s , so t h a t those p a r t n e r s may respond appropriately. I n such i n t e r a c t i o n a l the speaker can encode h i s i n t e n t i o n s t h a t the hearer a c t in a v a r i e t y o f sentence t y p e s . Instead o f t e l l i n g the hearer what to do. the speaker may Just s t a t e h i s g o a l s , and expect a response t h a t meets these goals. This paper presents a new model f o r r e c o g n i z i n g the speaker's Intended meaning in d e t e r m i n i n g a response. We show t h a t t h i s r e c o g n i t i o n makes use of the speaker's p l a n , h i s b e l i e f s about the domain and about the h e a r e r ' s relevant c a p a c i t i e s . *
One of the goals of the Knowledge Representation and N a t u r a l Language group at BBN has been to provide powerful g e n e r a l t o o l s f o r n a t u r a l language processing and to b u i l d a language understanding system f o r a d e c i s i o n maker u s i n g a graphics display. We e n v i s i o n a d e c i s i o n maker aocessing i n f o r m a t i o n from a database t h a t can be represented v i s u a l l y : he/she needs to c o l l e c t i n f o r m a t i o n from the database, add to i t , change it and d e f i n e new features. A s p e c i a l f e a t u r e of t h i s language understanding system is the assumption t h a t the human user expresses h i m / h e r s e l f n a t u r a l l y . He/she can u t t e r more than the d i r e c t imperatives and can ask questions besides the d i r e c t questions i n t y p i c a l of most c u r r e n t AI language systems.
1. STATEMENT OP THE PROBLEM
Natural communication with a partner is p o s s i b l e , however, o n l y i f the l a t t e r can reason both about the intent behind the speaker's utterances and about I t s own responses to those utterances. Such reasoning I n v o l v e s - at l e a s t (1) b r i n g i n g to bear the kinds of knowledge people have before they enter i n t o a given d i s c u s s i o n , and (2) making use of the knowledge they g a i n in the discussion.
Human 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 s depend upon the a b i l i t y o f t h e i r p a r t n e r s t o recognize t h e i r i n t e n t i o n s , so t h a t those p a r t n e r s may be capable o f responding a p p r o p r i a t e l y . For i n s t a n c e , i n the dialogue below, D1-1 S I : I want to see the drawing of the new design l a y o u t . 2 S2: OK. Here it i s . (shows sheet w i t h new design) 3 S 1 : There i s n ' t room to put in the c o l o r code charts at the bottom of the p i c t u r e . Can you move up the main l a y o u t ? 4 S2: Sure, I ' l l b r i n g back the new design in h a l f an hour.
An o r g a n i z i n g framework f o r a system t h a t can reason about the the speaker's I n t e n t i o n s has been explored in our group and is r e p o r t e d elsewhere [Brachman et a l . 1979]. In t h a t framework we experimented w i t h an Implementation of A l l e n ' s [1979] model of a plan-based approach to speech a c t recognition. This model provided us w i t h a p e r s p e c t i v e w i t h i n which to use models of speaker's b e l i e f s and wants [Cohen, 1978]) as w e l l as a framework f o r reasoning about the speaker's p l a n s . We found t h a t we needed to expand on A l l e n ' s model in two ways: ( 1 ) by r e c o g n i z i n g a r i c h e r form of p l a n s , and (2) by making e x p l i c i t the connection between the speaker's I n t e n t i o n s as s t r u c t u r e d by h i s plans and the response intended by the speaker.
the speaker's desires are encoded in a v a r i e t y of sentence t y p e s . I n s t e a d of t e l l i n g the hearer what to do, the speaker s t a t e s h i s g o a l s , and expects a response t h a t meets them, at l e a s t partway. This paper presents a model in which r e c o g n i z i n g the speaker's intended meaning plays a fundamental p a r t in determining a response.
In this paper we w i l l describe the new theoretical model for recognizing speaker's intended meaning and the computational t o o l s we are c u r r e n t l y u s i n g to implement t h a t model. We t h i n k the model i s p a r t i c u l a r l y powerful because i t i s the basis of a system t h a t oan reason about the kinds of p l a n n i n g bugs the acknowledgement o f , and the response t o whioh are o f t e n r e f l e o t e d i n c o n v e r s a t i o n a l exchanges.
3. DEPININO INTENDED SPEAKER MEANING Our goal is to provide a computational model of the hearer's interpretation of the speaker's intended meaning. The intended meaning of an u t t e r a n c e we d e f i n e as t h a t set of < p r o p o s i t i o n a l attitude (e.g., belief, want, intend), p r o p o a i t i o n a l content> p a i r s t h a t the speaker wants to Induce in the hearer by means of the u t t e r a n c e .
*Acknowledgements: The research reported In t h i s paper was supported in p a r t by the Advanced Researoh P r o j e c t s Agency and was monitored by the Office of Naval Research under Contract No. N00014-77-C-0378.
The n o t i o n of the intended meaning of an u t t e r a n c e oan b e I l l u s t r a t e d b y c o n t r a s t i n g i t w i t h t h a t of semantic meaning. The semantic meaning of
203
a declarative utterance is the propositlonal c o n t e n t assigned to the type of the utteranoe by the semantic r u l e s of the language. For i n s t a n c e , if someone s a y s , "You're a p r i n c e , " the semantic meaning is ( r o u g h l y ) t h a t the person addressed by the speaker la tne son of a k i n g . By c o n t r a s t , the intended meaning is dependent on ( i n t e r a l i a ) the i n d i v i d u a l p s y c h o l o g i c a l s t a t e o f the speaker a t the time and place of u t t e r a n c e . The speaker may mean t h a t he t h i n k s the hearer is a r e a l l y n i c e guy and wants to t e l l him so, or he may be saying something q u i t e d i f f e r e n t . The speaker, u s i n g i r o n y , may mean t h a t the addressee is j u s t the opposite of a n i c e guy. This ( i n some ways extreme) example demonstrates t h a t speaker's intended meaning, though c o r r e l a t e d , i s not i n g e n e r a l i d e n t i c a l , w i t h semantic meaning. Comprehending the semantic meaning of the u t t e r a n c e forms the oasis for d i s c e r n i n g the intended meaning, but the l a t t e r ( t y p i c a l l y ) a l s o r e q u i r e s the use of the f o l l o w i n g kinds of b e l i e f on the p a r t o f the h e a r e r : 1. B e l i e f s about the c h a r a c t e r i s t i c s of the current s i t u a t i o n , 2. B e l i e f s goals,
about
the
speaker's
beliefs
and
3 . B e l i e f s about the c o n t e x t o f d i s c u s s i o n (the discourse context) as a s p e c i a l aspect of ( 1 ) , 4 . B e l i e f s about what conventions e x i s t between the speaker and hearer,*
for
action
the
clause
network> OK. on the pre
and Intended the system to i n c l u d e t h a t s t a t e in the d i s p l a y . If the user had been r i g h t about t h i s , the system would indeed have i n c l u d e d i t . But the u s e r ' s "No" i n d i c a t e s to the system some bug in her p l a n , a bug stemming e i t h e r from a f a u l t y model o f the domain i t s e l f o r from f a u l t y expectations about the system's model. (For s i m p l i c i t y , we assume t h a t the system is omniscient about the ATN grammar.]
I f , on the other hand, the system d o e s n ' t oonclude t h a t the user takes S/AUX to be among the p r e v e r b a l c o n s t i t u e n t s , and i f i t b e l i e v e s t h a t she takes t h a t to be mutually b e l i e v e d , then the system must again use i t s models of her and of her model of i t s e l f to determine what a c t i o n is intended by the user [ E . g . , should i t compress the c u r r e n t d i s p l a y to make room f o r S/AUX; should it erase the c u r r e n t d i s p l a y and b r i n g up a new one, centered on S/AUX, e t c . ? ] . This d e c i s i o n may depend on the kinds o f conventions a l l u d e d t o i n ( 4 ) above. In general, of course. people's behavior in c o n v e r s a t i o n a l s i t u a t i o n s a l s o depends on the r e l a t i v e s t a t u s o f the c o n v e r s a t i o n a l p a r t n e r s , o n what the participants think will benefit themselves, as w e l l as not harm o t h e r s , and the like. These s o c i a l c o n s i d e r a t i o n s are s i g n i f i c a n t to human i n t e r a c t i o n , but f o r the remainder of t h i s paper, w e ' l l assume t h a t the system responds in a s l a v i s h l y c o o p e r a t i v e way, t h a t l a , i t has n o i n t e r e s t beyond s e r v i n g the u s e r . There are two ways to view the i n t e n t i o n s of another agent. The f i r s t i s simply i n terms o f one's b e l i e f s about what the other person wants and believes. This is keyhole r e c o g n i t i o n (see Cohen et al [ 1 9 8 l ] ) . One person decides what he t h i n k s another intends simply by o b s e r v i n g him through a keyhole. ( E . g . I decide t h a t you are l o o k i n g f o r your u m b r e l l a , on the basis of your l o o k i n g around the room w l t n your coat o n , when I b e l i e v e you believe that i t ' s raining outside). Keyhole r e c o g n i t i o n o f a u s e r ' s wants i s c e n t r a l t o Genesereth's MACSYMA advisor [ 1 9 7 8 ] ; it a l s o forms the basis o f p l a n r e c o g n i t i o n i n both Schmidt e t a l ' a BELIEVER [19781 system and in W i l e n s k y ' s s t o r y understanding work L1978].
5 . B e l i e f s about what i s m u t u a l l y b e l i e v e d , a s between speaker and h e a r e r , w i t h r e s p e c t to (1) through ( 4 ) . A sample exchange w i l l i n d i c a t e the r o l e of these kinds of b e l i e f . In the example below, the user is i n t e r a c t i n g w i t h our system to d i s p l a y some i n f o r m a t i o n ( i n t h i s case, a p o r t i o n of an ATN network). The u s e r ' s f i r s t two u t t e r a n c e s are simple d i r e c t i m p e r a t i v e s t h a t i n d i c a t e t h a t the user wants the system to d i s p l a y a p a r t of the net and then move the focus to a subpart of the display. D2-1 U: Display network. 2 S: < d i s p l a y of 3 U: Now focus constituents. 4 S: < d i s p l a y of 5 U: No, I want
, and thought the system per el lveevrebda l t hci so nas ltsi tou.e n t sShe would then have expected
level
The intended r e c o g n i t i o n of what someone is doing, on the other hand, la relevant for communicative s i t u a t i o n s [ O r l c e 1957, A l l e n . 1 9 7 9 ] . A speaker says something to a h e a r e r , and Intends t h a t the hearer recognize the i n t e n t i o n t h a t l i e s behind the u t t e r a n c e . The speaker is a t t e m p t i n g to " g i v e the hearer a piece o f i t s mind" and I t ' s e s s e n t i a l to the success of the speaker's attempt t h a t the hearer recognize i t a s such.
verbal
subnet> OK. to see S/AUX.
What does the user mean by her t h i r d u t t e r a n c e ( u t t e r a n o e 5)? The answer depends on what she b e l i e v e s about the net o b j e c t s to which she has referred. Suppose she t h i n k s t h a t S/AUX is p a r t of the preverbal constituents. Then she is communicating t h a t the d i s p l a y la wrong and w h a t ' s wrong w i t h i t ; she Intends f o r S/AUX to be Included in the display with the other constituents. Suppose, a l t e r n a t i v e l y t h a t she t h i n k s t h a t S/AUX i s not p a r t o f the p r e v e r b a l c o n s t i t u e n t s . She i s s t i l l i n d i c a t i n g t h a t she wants to see S/AUX, but a l s o t h a t she has changed her mind about the d i s p l a y in some way and i n t e n d s S/AUX to be visible. [As t o the nature o f her change o f mind, t h i s depends on whether she has r e a l i z e d the e r r o r of her ways, or at l e a s t the mismatch between her views on S/AUX and the s y s t e m ' s . ]
4. A MODEL OF RECOGNITION OF INTENDED MEANING The h e a r e r ' s taak in r e c o g n i z i n g what the apeaker meant by an u t t e r a n o e la to be understood as f o l l o w s :
I n ohoosing among i t s a v a i l a b l e responses, the system must u t i l i z e i t s model o f the u s e r ' s b e l i e f s about the domain and i t s model of what the user takes to be m u t u a l l y b e l i e v e d between the two of them about t h a t domain. For example, the user might have thought t h a t S/AUX was one of the
1.
to produce an explanation utterance, stated in terms speaker's b e l i e f s and wants.
for of
the the
2.
to use the e x p l a n a t i o n aa a b a s i s f o r a response.
We use the term " e x p l a n a t i o n " because the hearer Is t r y i n g to answer the q u e s t i o n "Why d i d the speaker say t h a t to me?* The answer to t h i s q u e s t i o n - the p r o f f e r e d e x p l a n a t i o n o f the speaker's a c t i n u t t e r i n g what he/she d i d - in t u r n produces new b e l i e f s about the speaker; theae w i l l form p a r t o f the basis of the h e a r e r ' s response. The e x p l a n a t i o n , i n g e n e r a l , w i l l have the form o f a set o f p a i r s o f p r o p o s i t l o n a l a t t i t u d e s and p r o p o s i t l o n a l contents a t t r i b u t e d by the hearer to the speaker. [ E . g . < b e l l e f , t h a t S7AUX i s p a r t o f
• T h i s , of course, is relevant only to a special c l a s s of s i t u a t i o n s ; a o l a s s which Includes the k i n d of i n t e r a c t i o n the BBN system must handle.
204
the p r e v e r b a l o o n s t i t u e n t s > . e t c . j C e r t a i n b e l i e f s play a c e n t r a l r o l e i n e x p l a i n i n g why the speaker s a i d what he d i d :
t h a t PC(S1). [ I n the above, " p " i s a schematic l e t t e r which takes verbs of proposltional attitude as s u b s t i t u e n d s ; "PC", a schematic l e t t e r which takes d e c l a r a t i v e sentences as s u b s t i t u e n d s . ] We oan apply t h i s theory d i r e c t l y t o the sample d i a l o g u e s . For example, l e t us consider a sample u t t e r a n c e from the dialogue D2, understood, however, as the i n i t i a l utterance of a discourse:
Explanatory b e l i e f s 1.
b e l i e f s about the speaker's goal and the plan t o achieve i t ,
2.
b e l i e f s about the h e a r e r ' s c a p a c i t i e s ,
3.
b e l i e f s about the h e a r e r ' s d i s p o s i t i o n s to act given information about the speaker's wants.
S I : I want to see S/AUX. I n t u i t i v e l y , we would l i k e the theory to a l l o w us to show how an audience (even a computer system) would conclude t h a t the user wants to see S/AUX, and t h a t the user wants it to b e l i e v e t h a t he/she has t h i s d e s i r e .
The problem we pose f o r ourselves is determining how to i n f e r b e l i e f s of these k i n d s , and how to use them to d i s t i n g u i s h , between Intended, and h e l p f u l , but unintended, responses. We want our system to recognize and produce the intended response whenever p o s s i b l e , and to be able to produce a h e l p f u l response when a p p r o p r i a t e .
The set of r e l e v a n t f e a t u r e s F, a t t r i b u t e s A and mappings C are (or anyway i n c l u d e ) :
To model the c o n s t r u c t i o n of the r e q u i r e d explanation, we begin w i t h G r i c e ' s theory of speaker meaning [1957, 1969]. Grice notes t h a t there are c e r t a i n kinds of evidence which are normally a v a i l a b l e to an audience on the basis of which the audience ( i s intended) to draw c e r t a i n conclusions about the speaker's intended meaning. These include the f e a t u r e s of the u t t e r a n c e , mappings between those f e a t u r e s and p r o p o s l t i o n a l attltude-proposltlonal content pairs that the audience, assumed to be a competent speaker/hearer of the language, is supposed to be able t o , and is intended to g r a s p , e t c . For example, the f e a t u r e : DECLARATIVE w i l l be mapped to the speaker's wanting the hearer to b e l i e v e the speaker believes the proposltional content o f the utterance; while i m p e r a t i v e s w i l l be mapped to the speaker's wanting the hearer to b e l i e v e the speaker wants the hearer to b r i n g about the s t a t e of a f f a i r s expressed by the p r o p o s l t i o n a l component of the u t t e r a n c e .
Intended Conclusions S1 has c e r t a i n f e a t u r e s ( c a l l them F1 . . . Fn).
2.
S 1 i s c o r r e l a t e d , i n v i r t u e o f such f e a t u r e s and the r u l e s of the language, w i t h the p a i r (p , P C ( S D ) .
3.
The speaker intends the audience to b e l i e v e t h a t the speaker p's t h a t PC(S1).
4.
By s i n c e r i t y (see below), does p t h a t PC(S1),
5.
the
speaker
intended t h a t
the
F1 = S1 is in d e c l a r a t i v e mood
o
F2 = S1 was u t t e r e d i n t e n t i o n a l l y by U
o
F3 = S1 was i n t e n t i o n a l l y d i r e c t e d at S
o
A1 = s is a computer system graphics d i s p l a y , and U knows t h i s
o
A2 = S b e l i e v e s U Is s i n c e r e
o
C1 s F1 maps to U's wanting the intended audience to b e l i e v e t h a t U b e l i e v e s t h a t U wants to see S/AUX.
Our proposed guaranteeing and apply cl the intended
Somewhat more f o r m a l l y : an audience, f o r the u t t e r a n c e of a c e r t a i n sentence S1, who la believed by the speaker to have c e r t a i n a t t r i b u t e s A, is expected to be able to recognize c e r t a i n f e a t u r e s of the u t t e r a n c e and to be able to draw from those f e a t u r e s c e r t a i n conclusions about what the speaker intended in u t t e r i n g S1 in t h a t c o n t e x t . [One such audience a t t r i b u t e , of course. Is competence in the language of S 1 : o t h e r s are both more i n t e r e s t i n g and more s i t u a t i o n - s p e c i f i c . ] These conclusions are ( o r a t l e a s t i n o l u d e ) :
1.
o
with a
system w i l l make d e f a u l t assumptions F2, F3, A1 and A2, recognize t h a t F 1 , to S1. The system can then u t i l i z e conclusions and i n f e r d i r e c t l y t h a t :
1.
U Intended (S to reoognlze) t h a t S1 is c o r r e l a t e d w i t h U's wanting S to b e l i e v e t h a t U b e l i e v e s t h a t U wants to see S/AUX ( d e r i v e d from Intended conclusion 3 and C1).
2.
By s i n c e r i t y : U b e l i e v e s t h a t U wants to see S/AUX.
3.
By r e l i a b i l i t y : U wants to see S/AUX.*).
T h i s , o f course, i s what, o n i n t u i t i v e grounds, wanted the system to conclude.
•Simply s t a t e d these r u l e s , f o r the case of b e l i e f , a r e : S i n c e r i t y ! I f x wants y t o b e l i e v e t h a t x b e l i e v e s t h a t q, then x b e l i e v e s t h a t q. R e l i a b i l i t y : If y believes that x believes that q and t h a t x is r e l i a b l y informed about q, then y w i l l b e l i e v e t h a t q . The basis f o r these r u l e s i s the i n t u i t i o n t h a t the speaker i s s i n c e r e about h i s b e l i e f s , and t h a t what he b e l i e v e s he b e l i e v e s r e l i a b l y , at least f o r c e r t a i n subjeot matters, such as h i s own present s t a t e of mind.
speaker
the hearer p*
• A c t u a l l y the hearer may be intended to have a different proposltional attitude p' toward a related proposition. For s i m p l i c i t y , w e ' l l assume these are the same. 205
206
plan. To account f o r the bug, the system can reason in e i t h e r of two ways*. On the one hand, If it now has reason to b e l i e v e t h a t the user b e l i e v e s t h a t the p r e v e r b a l c o n s t i t u e n t s I n c l u d e S/AUX, I t w i l l conclude t h a t t h e r e is a bug in the u s e r ' s p l a n . T h i s is a p r i v a t e ( n o t mutual) b e l i e f ; but it prevents the system from responding in the way i n t e n d e d by the user ( t o d i s p l a y S/AUX) f o r not enough o f the u s e r ' s i n t e n t i o n s are c l e a r t o decide how to do the d i s p l a y ( e . g . , to i n c l u d e S/AUX or to show It a l o n e ) . Henoe the system w i l l respond by i n d i c a t i n g what the bug la and by a s k i n g about the p a r t i c u l a r mode o f d i s p l a y d e s i r e d . On the o t h e r hand, if the system b e l i e v e s t h a t the user b e l i e v e s the p r e v e r b a l c o n s t l t u t e n t s are ( m u t u a l l y b e l i e v e d to be) d i s j o i n t from S/AUX, then the system w i l l oonolude t h a t the user has scrapped h i s c u r r e n t p l a n , and t h a t t h i s c o n c l u s i o n I s one the system I s Intended t o deduce. I n t h i s case, d i s p l a y i n g S/AUX Is the Intended response, but the system must s t i l l ask how t o d i s p l a y I t , s i n c e I t I s not c l e a r whether the user Intended I t t o b e d i s p l a y e d alone or w i t h the subnet. A p e r s o n . In such c i r c u m s t a n c e s , would probably conclude t h a t S/AUX should be d i s p l a y e d a l o n e , because s/he oould deduce t h a t I n g e n e r a l I f a p l a n I s scrapped, e f f e c t s o f I t s p a r t i a l r e a l i z a t i o n are n o longer desired. However, t h i s h e u r i s t i c may be too general for systems which s t i l l have l i m i t e d r e a s o n i n g c a p a c i t i e s , and hence we have chosen not t o I n c l u d e such r u l e s . This example demonstrates two aspects of our system: I t s use o f p l a n - a t t r i b u t i o n s , I n f e r r e d I n the course o f I n t e r p r e t i n g the u s e r ' s I n t e n t i o n s , t o recognize bugs I n p l a n s , and i t s use o f p r i v a t e a s w e l l a s mutual b e l i e f s t o determine i t s response when what the user i n t e n d s Is u n c l e a r due to a "buggy" p l a n . The next example Is a v a r i a t i o n of D 1 . The task here is to I n t e r a c t w i t h a user when a g r a p h i c s display is available for representing information about a database. The database Is represented in KL-OWE [Brachman et a l , 1 9 7 9 ] , and the d i s p l a y c o n s i s t s of a g r a p h i c r e p r e s e n t a t i o n of XL-ONE concepts. D3-1 U: I want to see the g e n e r i c concept named employee. 2 S: OK. ( d i s p l a y s concept In m i d screen) 3 U: I c a n ' t f i t a new i n d i v i d u a l concept below i t . 4 Can you move it up? 5 S: Sure. (moves up the g e n e r i c concept) The system's problem In responding to "Can you move i t up?" l a t o determine whether the user meant h i s u t t e r a n c e d i r e c t l y as a q u e s t i o n about the system's a b i l i t i e s , or i n t e n d e d r a t h e r to be taken to be d i r e c t i n g the system to move the conoept under discussion. I t s d e c i s i o n depends o n i n f e r r i n g the s p e a k e r ' s p l a n and, i n p a r t i c u l a r , o n what i t b e l i e v e s the u s e r ' s model o f i t s own c a p a c i t i e s t o be. T h i s example i l l u s t r a t e s a f e a t u r e o f n a t u r a l i n t e r c h a n g e s : a user may have a plan in mind, and c a r r y out a part of i t , without considering ossible undesired s i d e e f f e c t ; when one o c c u r s , t may be recognized and e l i m i n a t e d . In D3 the user i s c a r r y i n g out the plan o f accessing the concept f o r employee so t h a t a/he can add a new employee to the database. S/he wants the system to
p
d i s p l a y the employee concept, but has not foreseen t h a t i t s d i s p l a y l o o a t l o n might b e i n f e l i c i t o u s . A f t e r the i n a p p r o p r i a t e n e s s i s d i s c o v e r e d , the user i n d l o a t e a the d i f f i c u l t y and expects i t t o b e corrected. J u s t how the bug in D3 is corrected depends on whether the user a l r e a d y b e l i e v e s the system can move t h i n g s up and intends the system to d o s o , o r has t o f i n d t h i s out f i r s t . From the system's p o i n t o f v i e w , the d e c i s i o n about what the user means may cause it to respond d i f f e r e n t l y i n v a r i o u s cases. Suppose the system thlnjcs the user b e l i e v e s t h a t the system can move up concepts on the s c r e e n . Then when the user i n d l o a t e a t h a t h i s p l a n has a f l a w ( D 3 - 3 ) , the system must conclude that the user's p l a n is blocked by the l a c k of space f o r a new concept. When the q u e s t i o n about moving the employee concept i s r a i s e d , the system w i l l oonolude t h a t the user i n t e n d s to t e l l the system to perform the move by a s k i n g about a p r e c o n d i t i o n of the a c t i o n s/he wants, a precondition which consists i n the system's having a c a p a c i t y I t i s m u t u a l l y b e l i e v e d to possess. The system is i n t e n d e d to move up the conoept, not s i m p l y to answer the q u e s t i o n . A d i f f e r e n t s c e n a r i o is as f o l l o w s . Suppose the system t h i n k s the user i s uaware t h a t the system i s oapable of moving up the conoept. Then when the user I n d i c a t e s t h a t h i s plan has a f l a w and asks about moving the employee concept, the system w i l l conclude t h a t the user I n t e n d s t o f i n d out whether i t has t h a t a b i l i t y , a s p a r t o f f i n d i n g a means o f r e s o l v i n g the bloolc. I n t h i s case, i f the system moves the concept, that is a b i t of h e l p f u l b e h a v i o r , one n o t Intended to be recognized as Intended by the u s e r . Ve w i l l o u t l i n e in some d e t a i l how our system reasons in such c o n t e x t s by showing what plans are deduced, what r u l e s are needed and how the reasoning proceeds in the case of D3-3 and 4. The r e l e v a n t user p l a n i s : Add-Data < a c r e e n - l o o a t i o n > Consider-aapect Put User at < a o r e e n - l o c a t i o n > The Add-Data p l a n s t a t e s t h a t to add d a t a , a user must c o n s i d e r some aspect of a network p a r t ( n e t p l e c e ) and then put some d a t a at a screen location. Even a f t e r r e c o g n i z i n g t h a t the user wants some data d i s p l a y e d a f t e r D 3 - 1 , the system can not deduoe t h a t add-data i s the u s e r ' s p l a n . Since t h e r e are many ways to oonsider some aspect o f a net (ask f o r a d i s p l a y , t h i n k about i t , ask t o be Informed about i t s c o n t e n t s ) , as w e l l as many o t h e r plans f o r which d i s p l a y i n g a n e t p i e c e is a f i r s t s t e p , the user cannot Be understood to have i n t e n d e d the system to recognize t h a t h i s p l a n was to Add-Data . A l l the system can oonolude is t h a t the user wants the employee concept d i s p l a y e d , and i t responds a c c o r d i n g l y . I n r e a s o n i n g about D3-3, " I c a n ' t f i t a new i n d i v i d u a l oonoept below i t , " the system concludes that among the speaker's intentions mutually presumed to be recognized is t h a t the user produoed a d e c l a r a t i v e u t t e r a n o e w i t h the p r o p o s i t i o n a l c o n t e n t t h a t the user cannot f i t a new i n d i v i d u a l conoept ( a b b r i e v a t e e2) below the g e n e r i c concept (abbreviated e1): BELIEF1; (Say User System ( D e c l a r a t i v e (Not (Can User ( P i t User e2 (below e 1 ) ) ) ) ) ) Prom t h i s , the system ooncludes t h a t the user wants the system t o b e l i e v e t h a t the user b e l i e v e s t h a t i t c a n ' t f i t e 2 below e 1 , and t h a t the user i n f a c t b e l i e v e s t h a t he c a n ' t . The system then i n f e r s the embedded p r o p o s i t i o n [ ( N o t (Can . . . ) ) ] , and t h a t the user i n t e n d e d t h a t t h a t p r o p o s i t i o n be m u t u a l l y believed. Using a ( d e f a u l t t y p e ) r u l e t o the e f f e c t t h a t whenever a user says t h a t i t c a n ' t
• A c t u a l l y t h e r e is a t h i r d case—the system was wrong about what " p r e v e r b a l c o n s t i t u e n t s " r e f e r s t o . As mentioned above, we Ignore t h i s case.
207
bring about a certain a t a t a - o f - a f f a i r a or p a r f o r m a cartain action, tha User la t a i l i n g tha ayataa that It wanta that s t a t a - o f - a f f a i r s brought about, tha ayataa concludaa that It la intandad to baliava that tha uaar wanta it to baliava
•baaad on rulaa about ha1ping out whan ona haa an approprlata capacity. Aa with othar hauriatica, it ■ay ba wiaa to aonltor oarafully what tha ayataa doas to ba halpful alnoa aoaa halpful actiona, if not In faot daalrad by tha apaakar can ba aaally undooa, but otnara hava n o n - t r i v i a l aida~affacta. • • I c » and " r o l a with valua" ara KL-OWE taraa
ballafa about tha ayataa'a oapaoltlaa affaot tha ayataa*a dataralnatlon of tha uaar'a intantiona, but alao that tha f u l l axplanation of aaon uttaranoa daapana tha ayataa*a undaratandlng of tha uaar*a goala and aubaaquant uttaranoaa.
