Scripts
Content
SCRIPTS, PLANS, AND KNOWLEDGE Roger C. Schank and Robert P . Abelson Y a l e University New Haven, Connecticut
t
USA
"Of what a stmnge nature i s b w t e d g e ! It cZings t o the mind, when i t has once seized on i t , like a
Zichen on the rock."
Frankenstein's Monster (M. Shelley, fiankenstein or the Modern Prwmetheus, 1818) Abstract We describe a t h e o r e t i c a l system intended to f a c i l i t a t e the use of knowledge i n an understanding system. The notion o f s c r i p t i s introduced to account f o r knovledge about mundane s i t u a t i o n s . A program, SAM, i s capable of using s c r i p t s to understand. The notion of plans i s introduced to account f o r general knowledge about novel situations.
-
I.
Preface
I n an attempt to provide theory where there have been mostly unrelated systems, Minsky (1974) recently described the work of Schank (1973a), Abelson (19731, Charniak (1972), and Norman (1972) as f i t t i n g into the notion of "frames." Piinsky attempted to r e l a t e t h i s work, i n what i s e s s e n t i a l l y language processing, t o areas of v i s i o n research that conform to the same notion. Minsky's frames paper has created q u i t e a s t i r i n AI and some immediate spinoff research along the l i n e s of developing frames manipulators e f i n d that ( e . g . Bobrow, 1975; Winograd, 1975). W we agree with much of what Minsky s a i d about frames and with h i s characterization of our own work. The frames idea i s so general, however, that i t does not lend i t s e l f to applications without further s p e c i a l i z a t i o n . This paper i s an attempt to develop further the l i n e s of thought s e t out i n Schank (1975a) and Abelson (1973; 1975a). The ideas presenced here can be viewed as a s p e c i a l i z a t i o n o f the frame idea. W e s h a l l r e f e r to our central constructs as " s c r i p t s . "
zation of knowledge can r e s u l t i n a r e a l understanding system i n the not too distant future. W e expect that programs based on the theory we outl i n e here and on our previous work on conceptual dependency and b e l i e f systems w i l l combine with the MARGIE system (Schank e t a l . , 1973a; Riesbeck, 1975; Rieger, 1975) to produce a working understander. We see understanding a s the f i t t i n g of new information i n t o a previously organized view of the world. We have therefore extended our work on language analysis (Schank, 1973a; piesbeck 1975) to understanding an understander. l i k e an analyzer, should be "bottom up" u n t i l i t gets enough information to make predictions and become "top down." E a r l i e r work has found various ways i n which a word in a s i n g l e sentence s e t s up expectations about what is l i k e l y to be found i n the r e s t of the sentence. A s i n g l e sentence and i t s corresponding conceptualizations s e t up expectations about what is to f o l l o w in the r e s t of a discourse o r story. These expectations characteri z e the world knowledge that bears on a given s i tuation, and i t is these expectations that we wish to explore.
-
111. S c r i p t s
A s c r i p t , as we use i t , is a structure that describes an appropriate sequence of events i n a p a r t i c u l a r context. A s c r i p t i s made up o f s l o t s and requirements about what can f i l l those s l o t s . The structure i s an interconnected whole, and what i s in one s l o t a f f e c t s what can be i n another. Scripts handle s t y l i z e d everyday s i t u a t i o n s . They a r e not subject to much change, nor do they provide the apparatus f o r handling novel situations, a s plans do (see section V). For our purposes, a s c r i p t i s a predetermined, stereotyped sequence of actions that d e f i n e a well-known situation. A s c r i p t i s , i n e f f e c t , a very boring l i t t l e story. Scripts a l l w f o r new references to o b j e c t s within them j u s t a s i f these objects had been previously mentioned; o b j e c t s within a s c r i p t may take "the" without e x p l i c i t introduction because the s c r i p t i t s e l f has a l ready i m p l i c i t l y introduced them. (This can b e found below, i n the reference to '*the waitress" i n a restaurant, f o r example.) Stor'ies can invoke s c r i p t s i n various ways. Usually a story Is a s c r i p t with one or m r e int e r e s t i n g deviations.
11.
The Problem
Researchers in n a t u r a l language understanding have f e l t f o r some time that the eventual limit on the solution of our problem w i l l b e our a b i l i t y to characterize world knowledge. Various researchers have approached world knowledge in various ways. Winograd (1972) d e a l t with the problem by severely r e s t r i c t i n g the world. This approach had the pos i t i v e e f f e c t of producing a working system and the negative e f f e c t o f producing m e that was only minimally extendable. Charniak (1972) approached the problem from the other end e n t i r e l y and has made some i n t e r e s t i n g f i r s t steps, but because h i s work is not grounded i n any representational system o r any working computational system the rest r i c t i o n o f world b o w l e d g e need not c r i t i c a l l y concern him. Our f e e l i n g is that an e f f e c t i v e characteri-
I. John went i n t o the restaurant.
4ie ordered a hamburger and a coke. He asked the waitress f o r the check and left.
11. John went to a restaurant.
Be ordered a hamburger. It was cold when the waitress brought i t . He l e f t her a very small t i p .
111. H a r r i e t went t o a birthday party.
t
The work of the second author was f a c i l i t a t e d by National Science Foundation Grant GS-35768.
151
She put on a green paper hat. Just when they s a t down to e a t the cake, a piece of p l a s t e r f e l l from the c e i l i n g Onto the t a b l e . She was lucky, because the dust didn't get a l l over her h a i r .
PTRANS s e l f to t a b l e WVE s i t d m scene 2: ordering ATRANS receive menu M R A N S read menu HBUILD decide what s e l f wants HTRANS order to waitress scene 3 : eating ATRANS receive food INGEST food scene 4: e x i t i n g receive check t i p to waitress s e l f to cashier money to cashier PTRANS. s e l f out of restaurant I n t h i s s c r i p t , the instruments f o r performing an action might vary with circumstances. For example, i n scene 2 the order might be'spoken, or w r i t t e n down with predesignated numbers f o r each item, o r even ( i n a f o r e i g n country with an unfam i l i a r language) indicated by pointing or gestures. Each act sequence uses the p r i n c i p l e of caus a l chaining (Schank, 1973b,Abelson, 1973). That i s , each action r e s u l t s i n conditions that enable the next to occur. To perform the next act i n the sequence, the previous a c t s must be completed sati s f a c t o r i l y . I f they cannot be, the hitches must b e d e a l t with. Perhaps a new action not prescribed i n the s c r i p t w i l l be generated i n order to get things moving again. This "what-if" behavior, t o b e discussed l a t e r , i s an important component of s c r i p t s . I t i s associated with many o f the deviations i n s t o r i e s such a s paragraph 11. In a t e x t , new s c r i p t information i s interpreted i n terms of i t s place i n one of the causal chains within the s c r i p t . Thus i n paragraph I the f i r s t sentence describes the f i r s t action i n scene 1 o f the restaurant s c r i p t . Sencence 2 r e f e r s to the l a s t action o f scene 2, and Sentence 3 t o the f i r s t and l a s t actions of scene 4. The f i n a l interpretation o f paragraph I contains the e n t i r e restaurant s c r i p t , with s p e c i f i c statements f i l l e d i n and missing statements (that he s a t down, for example) assumed. In paragraph 11, the f i r s t two sentences desc r i b e actions in scenes 1 and 2 . P a r t o f the t h i r d sentence i s in the s c r i p t as the f i r s t action of scene 3, but there i s a l s o the information that the hamburger i s cold. The fourth sentence ("He l e f t , her a very small tip") i s a modification of the t h i r d action of scene 4. The uwdifier, "very small," i s presumably r e l a t e d to the unexpected information about the "cold hamburger." Even a stupid processor, checking paragraph I1 against the standard restaurant s c r i p t , could come up with the low-level hypothesis that the small s i z e of the t i p must have something to do with the temperature of the hamburger, since these two Items of Information are the only deviations from the s c r i p t . They must b e r e l a t e d deviations, because i f they were unrelated the narrative would have no business ending v i t b two such unexphined features. O f course we do not vant our processor to be I n s l i g h t l y more complex examples, adestupid. quate understanding r e q u i r e s attention to the naATWS ATRANS PTRANS ATRANS
HBUILD where to s i t
I V . Harriet went to Jack's birthday party. The cake tasted awful. Harriet l e f t Jack's mother a very small t i p .
It i s Paragraph I i s an unmodified s c r i p t . d u l l . It would be even d u l l e r i f all the events i n the standard restaurant s c r i p t (see below) were included. Paragraph I 1 i s a restaurant s c r i p t with a stock v a r i a t i o n , a customer's t y p i c a l reaction when things go wrong. Paragraph I11 invokes the birthday party s c r i p t , but something wholly outside the range of normal birthday p a r t i e s occurs the p l a s t e r f a l l s from the c e i l i n g . This deviation from the s c r i p t takes over the i n i t i a t i v e i n the n a r r a t i v e u n t i l the problem i t r a i s e s i s resolved, but the birthday s c r i p t i s s t i l l a v a i l a b l e i n the indirect reference to the party hat and i n the p o s s i b i l i t y that normal party a c t i v i t i e s be resumed l a t e r i n the n a r r a t i v e . I t seems n a t u r a l f o r reference to be made to dust i n the h a i r following the p l a s t e r ' s f a l l i n g , which implies that there i s a kind of s c r i p t for f a l l i n g p l a s t e r too. (This kind of s c r i p t we c a l l a vignette (Abelson, 1 9 7 9 1 . ) NQt i c e that "the c e i l i n g " r e f e r s t o an uninteresting "room" s c r i p t , which can be used f o r references to doors and windows that may occur. Thus i t i s poss i b l e t o be i n more than one s c r i p t a t a time. Paragraph I V i l l u s t r a t e s the kind of absurdit y that a r i s e s when an action from one s c r i p t i s a r b i t r a r i l y inserted into another. That one f e e l s the absurdity i s an indication that s c r i p t s are i n inadmissable competition. I t i s conceivable that with adequate introduction the absurdity i n paragraph I V could be eliminated. With these examples, a number o f issues have been r a i s e d . Let us a t t h i s point g i v e a more extensive description o f s c r i p t s . W e have discussed previously (Schank, 1975b) how paragraphs are represented i n memory as causal chains. This work implies that, f o r a story to be understood, inferences must connect each input conceptualization t o a l l the others i n the story that r e l a t e to i t . This connection process i s f a c i l i t a t e d tremendousl y by the use of s c r i p t s . Scripts a r e extremely numerous. There i s a restaurant s c r i p t , a birthday party s c r i p t , a footb a l l game s c r i p t , a classroom s c r i p t , and so on. Each s c r i p t has players who assume r o l e s i n the action. A s c r i p t takes the point o f view of one of these p l a v e r s , and i t often changes when i t i s viewed from mother player's point o f view. The f o l l o w i n g i s a sketch of a s c r i p t f o r a restaurant from the point of view of the customer. Actions a r e s p e c i f i e d in terms of the primitive ACTS of conceptual dependency theory (Schank, 1973b.
M R A N S ask f o r check
-
script: roles: reason:
restaurant customer, w a i t r e s s , chef, cashier to get food so a s to go up i n pleasure and down i n hunger
scene 1: entering PTRANS s e l f i n t o restaurant ATTEND eyes to where empty tables a r e
ture o f d e v i a t i o n s from the s c r i p t . A smarter processor can i n f e r from a c o l d hamburger that the INGEST i n scene 3 w i l l then v i o l a t e the pleasure goal f o r going t o a restaurant. The concept o f a v e r y small t i p can be stored with the restaurant s c r i p t as a what-if associated w i t h v i o l a t i o n s o f the pleasure goal. The general form f o r a s c r i p t , then, i s a s e t o f paths j o i n e d a t c e r t a i n c r u c i a l points that def i n e the s c r i p t . For restaurants the c r u c i a l p a r t s are the INGEST and the ATRANS of money. There are many normal ways t o move from p o i n t t o point. Ordering may b e done by M " S i n g to a w a i t e r or by s e l e c t i n g and taking what you l i k e (in a cafeteria). Likewise the ATRANS o f money may be done by going t o the cashier, o r paying the There w a i t r e s s , o r saying, "Put i t on my b i l l . " a r e a l s o paths t o take when s i t u a t i o n s don't go as planned. Paragraphs 1 1 1 and I V call up deviant paths i n the birthday party s c r i p t . All these v a r i a t i o n s i n d i c a t e that a s c r i p t i s not a simple l i s t o f events but rather a linked causal chain; a s c r i p t can branch into m u l t i p l e p o s s i b l e paths that come together a t c r u c i a l d e f i n i n g points. To know when a s c r i p t i s appropriate, s c r i p t headers a r e necessary. These headers d e f i n e the circumstances under which a . s c r i p t i s c a l l e d i n t o p l a y . The headers f o r the restaurant s c r i p t a r e concepts having t o do with hunger, restaurants, and so on i n the context of a plan o f a c t i o n f o r g e t t i n g fed. Obviously contexts must be r e s t r i c t ed t o avoid c a l l i n g up the restaurant s c r i p t f o r sentences that use the word "restaurant" as a place ("Fuel o i l was d e l i v e r e d t o the restaurant"). S c r i p t s organize new inputs i n terms o f prev i o u s l y stored knowledge. I n paragraph I, many items-that a r e p a r t of the restaurant s c r i p t a r e added t o the f i n a l i n t e r p r e t a t i o n o f the s t o r y . W e don't need t o say that a w a i t r e s s took the customer's order or that he a t e the hamburger. These Ideas a r e f i r m l y a p a r t o f the s t o r y because the restaurant s c r i p t r e q u i r e s them. I n understanding a s t o r y that c a l l s up a s c r i p t , the s c r i p t becomes p a r t of the s t o r y even when i t i s not s p e l l e d out. The answer t o the question W h o served John the hamburger?" seems obvious, because our world knowledge, as embodied i n s c r i p t s , answers i t .
it.
I n scene 2 of the restaurant s c r i p t , if what-ifs. the w a i t r e s s i g n o r e s the customer, he w i l l t r y t o catch her eye or c a l l t o her when she passes nearby. I f he can't make out the menu o r needs further information, he w i l l ask the w a i t r e s s . If she doesn't speak h i s language, he w i l l attempt her language, or make gestures, o r seek another customer t o t r a n s l a t e , o r accept her suggestion o f what t o order. I n scene 3, i f the w a i t r e s s does not b r i n g the food, he w i l l again t r y t o catch her eye. I f the food is not f i t , he w i l l send i t back. Errors have a s l i g h t l y d i f f e r e n t character from o b s t a c l e s but follow the same general r u l e s . Receiving the menu i s errorful i f the w a i t r e s s ATRANSes a p r i n t e d sheet t o the customer but i t i s yesterday's menu, or the breakfast instead o f the dinner menu. Reading the menu may y i e l d an e r r o r i f the customer g e t s the wrong idea o f what i t says say, he thinks f i l e t mignon i s a f i s h . Eere i t i s up t o the w a i t r e s s t o supply the whati f c o r r e c t i v e . Deciding what t o order may y i e l d an e r r o r i f the customer goes through the d e c i s i o n process but f o r g e t s the stored outcome. The-what; i f i s t o review the MBUILD ("Let's see now, what Giving the o r d e r t o the w a i t r e s s d i d I decide?") may b e i n e r r o r if she w r i t e s down something o t h e r than what the customer said, o r omits a p o r t i o n o f i t . The what-if i s t o repeat the order, asking the w a i t r e s s whether she i s sure she g o t a l l o f
-
I V . The SAM Program
what-Ifs
There are a t l e a s t three major ways i n which s c r i p t s can be thrown off normal course. One i s d i s t r a c t i o n , i n t e r r u p t i o n by another s c r i p t , such as the plaster's f a l l i n g from the c e i l i n g . W e w i l l not pursue here an a n a l y s i s of t h e conditions The other two and consequences of d i s t r a c t i o n . ways, o b s t a c l e and e r r o r , a r e i n t i m a t e l y connected with what-if behavior. An o b s t a c l e t o the normal seauence occurs when someone o r something prevents a normal a c t i o n from occurring or some usual enab l i n g condition for the a c t i o n i s absent. An e r r o r occurs when the a c t i o n i s completed i n an inappropriate manner, so that the normal consequences of the a c t i o n do n o t come about. I n p r i n c i p l e , e v e r y simple ACT i n a standard e ass c r i p t has p o t e n t i a l o b s t a c l e s and e r r o r s . W sume t h a t , e v e r y time an o b s t a c l e o r e r r o r occurs i n a s c r i p t t h a t i s being learned, the methods used t o remove the o b s t a c l e or redeem the e r r o r The rea r e s t o r e d w i t h the s c r i p t as what-ifs. sult of many r e p e t i t i o n s i s t h a t mst of the common what-ifs a r e attached t o the s c r i p t . Every o b s t a c l e has one or mre c h a r a c t e r i s t i c
We have created a program that uses s c r i p t s t o make i n f e r e n c e s i n domains i t knows about and have added i t t o the MARGIE program reported a t 3IJCAI. SAM ( S c r i p t A p p l i e r Mechanism) r e c e i v e s Conceptual Dependency structures as input (these a r e gotten from English by using Riesbeck's parser SAM i d e n t i f i e s the s c r i p t t o be used (1975)). (by using Riesbeck's request format) and f i l l s i n the i d e n t i f i e d s c r i p t w i t h the new inputs as they a r e received. It makes the inferences that a r e necessary t o complete a causal chain between input conceptualizations (see Schank (1973b) f o r a discussion o f causal chains). The resultant comp l e x structure can then be queried by input Engl i s h questions o r be used t o make summaries and paraphrases o f the i n i t i a l s t o r y . SAM was w r i t ten by Chris Riesbeck, Richard C u l l i n g f o r d and Wendy Lehnert. Goldmau's generation program (1975) i s used t o map the answers of paraphrases I n t o English. I t was modified f o r t h i s purpose by Gerald DeJong and W.J. Stutzman. An example of SAM's input and output i s : Input: John went t o a restaurant. The hostess seate d John. The hostess gave John a menu. The w a i t e r came t o the t a b l e . John ordered lobster. John was served quickly. John l e f t a Large t i p . John l e f t the restaurant. Questions and Answers:
Q: What d i d John e a t ? A: LOBSTER. Q: Who gave John the menu? A: THE HOSTESS. Q: Who gave John the l o b s t e r ? A: PROBABLY THE WAITER. Q: Who paid the check? A: PROBABLY JOHN.
153
A: Q: A: Q: A:
Q: What happened vhen John vent t o the t a b l e ? THE HOSTESS GAVE HIM A MENU AND JOHN SAT DOWN.
Why d i d John get a menu? SO HE COULD ORDER. Why d i d John give the w a i t e r a l a r g e t i p ?
A low-level
goal is i l l u s t r a t e d by:
John wanted t o cut h i s steak. He c a l l e d to h i s w i f e i n the kitchen.
BECAUSE HE WAS SERVED QUICKLY.
Paraphrase: JOHN DECIDED HE WAS GOING TO GO TO A B E S T W RANT. JOHN WENT TU A RESTAURANT. JOHN INDICATED TO THE HOST HE WOULD LIKE TO BE AT A TABLE. THE HOST INDICATED TO JOHN HE COULD GO TO A TABLE. JOAN WENT TO THE TABLE. THE HCST WENT TO THE TABLE. JOHN SAT DOWN I N THE CKAIR. JOHN GOT THE MENU FROM THE HOST. JOHN READ T?E MENU. THE WAITER SAW JOHN WAS AT THE TABLE. THE WAITER WENT TO THE TABLE. JOHN ORDERED LOBSTER. THE C€EF PREPARED THE LOBSTER. THE WAITER GOT THE LOBSTER FROM THE CHEF. WE WAITER SERVED JOHN THE LOBSTER. JOHN ATE THE LOBSTER. JOHN INDICATED TO TEE WAITER HE WOULD LIKE TO GET THE CHECK FROM HIM. JOHN GOT THE CHECK FROM THE WAITER. JOHN L E F I THE WAITER A LARGE TIP. JOHN PAID THE CHECK. JOHN LEFT THE RESTAURANT. Summary: JOHN WENT TO A RESTAURANT AND ATE LOBSTER. This program runs on the PDP-10 a t Yale. It currently has only a small amount of knowledge and a small vocabulary. But we f e e l encouraged that our s c r i p t theory i s workable because of the simp l i f i c a t i o n i n the inference process that has resulted from the use of s c r i p t s .
A plan is a s e r i e s o f actions that w i l l reali z e a goal. Often in order t o r e a l i z e one goal another must b e decided on and a plan drawn up to achieve i t . I n the f i r s t example above, a goal to a t t a i n power i s reduced to a goal to get arsenic. High-level goals a r e more i n t e r e s t i n g and v e have concentrated on them f i r s t . W e d e f i n e a "deltact" as an action o r a group o f actions that l e a d s to a desired s t a t e . Delt a c t s constitute subplans that a r e pursued because of t h e i r intended e f f e c t s . There a r e f i v e d e l t a c t s i n the present system:
AAGENCY ACONT AKNOW APROX ASOCCONT
-a
-
change i n o b l i g a t i o n t o do something f o r somebody a change i n the control of an o b j e c t a change i n vhat an a c t o r knows a change in the proximity r e l a t i o n s of o b j e c t s and a c t o r s a change in s o c i a l control over a person o r a s i t u a t i o n
v.
Plans
Plans are responsible f o r the d e l i b e r a t e behavior that people e x h i b i t . Plans describe the s e t of choices that a person has vhen he s e t s out to accomplish a goal. In l i s t e n i n g to a discourse, people use plans to make sense of seemingl y disconnected sentences. By f i n d i n g a plan, an understander can make guesses about the intentions of an action i n an unfolding story and use these guesses to make sense o f the story. Consider the f o l l o w i n g paragraph:
John knew that h i s w i f e ' s operation would
be very expensive. There v a s always Uncle Harry He reached f o r the suburban phone book.
...
How a r e we t o make sense o f such a paragraph? It makes no use of headers o r the s c r i p t s they sign a l . It would b e unreasonable to p o s i t a "paying f o r an operation" s c r i p t v i t h a l l the necessary a c t s l a i d out as in our restaurant s c r i p t . But, on the other hand, the s i t u a t i o n i s not e n t i r e l y novel, e i t h e r . The problem of understanding t h i s paragraph would not b e s i g n i f i c a n t l y d i f f e r e n t i f " w i f e ' s operation" were changed t o "son's education" or "down payment on the mortgage." There i s a general goal s t a t e in each case, r a i s i n g a l o t of money f o r a l e g i t i m a t e expense, and there is a generalized plan or group of plans that may l e a d to the goal s t a t e . Plans s t a r t with one o r more g o a l s . A highl e v e l g o a l i s i l l u s t r a t e d by the sequence:
John wanted to become king. Be vent to get Bome arsenic.
There is a l s o a s e t of lower-level d e l t a c t s & e l son, 1975a). Plans are made up of deltacts. When a collocation of d e l t a c t s i s used often enough, i t becomes a s c r i p t . A plan includes a s e t of planboxes, l i s t s of actions that w i l l y i e l d s t a t e changes and the preconditions f o r these actions, along with a s e t of questions f o r choosing the appropriate planbox. For instance, the TAKE plan has the goal o f enabling the taker to do something with an o b j e c t , whatever i s generally done with i t . T o TAKE something you must b e c l o s e to i t , so e i t h e r the obj e c t and the taker must b e i n the same location o r the taker must use a subplan APROX. Either no one e l s e must have CONTROL of the o b j e c t o r a t l e a s t there must be no bad consequences i n the taker's attempt to PTRANS the o b j e c t to himself. The TAKE plan c a l l s a PTRANS of the object i f a l l the preconditions a r e p o s i t i v e . But i f , say, someone e l s e CONTROLS the obj e c t , a plan f o r the taker's gaining CONTROL must b e c a l l e d . This subplan i s ACONT. ACONT has a Bet of planboxes attached t o i t . These planboxes define a d e l t a c t j u s t a s inferences define a primitive ACT. A planbox i s a l i s t of primitive ACTS that w i l l achieve a goal. Associated with each ACT are i t s precondltions, and a planbox checks them. A s e t o f p o s i t i v e conditions a l l o w s the desired ACT. Negative conditions c a l l up new planboxes o r d e l t a c t s that have as t h e i r g o a l the resolution o f the negative s t a t e . Preconditions f a l l into three c l a s s e s . A controlled precondition can b e f i x e d when i t i s negative by doing an ACT. A negative uncontrolled precondition cannot b e f i x e d , and another planbox must b e t r i e d . Negative mediating preconditions can b e a l t e r e d but require plans of t h e i r own t o change. Mediating preconditions usually r e f e r to the w i l l i n g n e s s o f other p a r t i e s to p a r t i c i p a t e i n plans. Further d e t a i l s on planboxes appear i n Schank ( 1 9 7 5 ~ ) . To see why an understander needs plans, cons i d e r the following sequence:
W i l l a vas hungry.
She took out the Michelin Guide.
Most readers understand that V i l l a was using the
154
Michelin Guide t o f i n d a good restaurant. But if the f i r s t sentence were subjected t o s t r a i g h t f o r ward inference (a l a R i e g e r , 1975), p r e d i c t i n g that W i l l a i s l i k e l y t o do something t o enable hers e l f t o INGEST food, the second sentence would seem t o answer t h i s p r e d i c t i o n o n l y i n the weird i n t e r p r e t a t i o n that she w i l l e a t the Michelin Guide. An understander w i l l r e j e c t t h i s i n f a v o r o f any b e t t e r path that i t can f i n d . The f i r s t sentence w i l l be analyzed f o r any g o a l that might generate a plan. "Hungry" i s l i s t e d i n the dictionary as i n d i c a t i n g the need f o r a plan t o do a ACONT o f food. One means f o r gaining c o n t r o l o f food i s a restaurant. An enablement f o r t h i s means i s going t o a restaurant, which r e q u i r e s APROX. This i n turn r e q u i r e s knowing where you are going, which may r e q u i r e AKNOW. I n the d i c t i o n a r y , a l l books a r e l i s t e d as means o f s a t i s f y i n g AKNOWs and the Michelin Guide i s l i s t e d as a book. To complete the processing o f t h i s sequence i t would, of course, be necessary t o have the information that the Michelin Guide l i s t s restaurants. Without t h i s information, the sequence might be as nonsensical as ' V i l l a was hungry. She took out Introduction t o A r t i f i c i a l Intelligence. W i t h the information that the Michelin Guide i s a source o f knowledge about restaurants, we know why the second a c t i o n was done and can pred i c t future actions. W e have transformed a seemi n g l y disconnected sequence i n t o one that provides the expectations that a r e so v i t a l t o understanding. I f the next sentence i s ' V i l l a g o t . i n her car," we w i l l know that the plan i s being e f f e c t e d . By using what we know about cars (that they a r e instruments o f PTRANS) and the s c r i p t f o r restaurants ( t h a t i s s t a r t s w i t h a PTRANS), we can make the inference that W i l l a i s on her way t o a restaurant. Some restaurant header would s t i l l be required t o i n i t i a t e the restaurant s c r i p t i n i t s f u l l glory. The procedure o f taking out the Michelin Guide when hungry, w h i l e seemingly n o v e l , could conceivably be routine for a c e r t a i n i n d i v i d u a l i n a c e r t a i n context. I f we know t h a t W i l l a i s a gourmet t o u r i s t staying i n P a r i s who enjoys going t o a d i f f e r e n t restaurant every evening, then the procedure o f looking i n the Guide might become p a r t o f h e r restaurant s c r i p t . For her there i s a scene b e f o r e scene 1 i n which she ATTENDS t o the Guide, MBUILDs a choice, and MTRhNSes a reservat i o n . A r o u t i n i z e d plan can become a s c r i p t , a t l e a s t from the planner's p o i n t o f view.
When a person reads a 300 page novel he does not (unless he i s v e r y unusual) remember a l l the conceptualizations s t a t e d i n the s t o r y i n the form of a g i a n t causal chain. Rather he remembers the g i s t of the book. Maybe 5 o r 10 pages o f summary could be extracted from him a f t e r reading the book. P r e v i o u s l y we have s a i d that Conceptual Dependency Theory w i l l account f a r memry f o r g i s t o f sentences. B u t i t cannot be s e r i o u s l y proposed that t h i s i s a l l that i s needed f o r g i s t o f long and complex s t o r i e s . Some o t h e r explanat i o n must be given. I n a recent experiment, Abelson [1975b] showed that people remember s t o r i e s b e t t e r when they a r e asked t o take some p a r t i c u l a r p o i n t of view ( o f one o f the p a r t i c i p a n t s o r o f an observer i n a p a r t i c u l a r place), and that what they remember i s contingent on which p o i n t of view they had. The r a m i f i c a t i o n s o f t h i s experiment f o r a theory o f language understanding have, t o be that when people have a c l u e o f what t o f o r g e t they do b e t t e r a t remembering. I n other words, good f o r g e t t i n g i s the key t o remembering. Likewise, if we want t o b u i l d programs that remember, we had b e s t teach them how t o f o r g e t . One method of f o r g e t t i n g i s simply not n o t i c i n g l e v e l s o f det a i l that a r e there. This can be done by t r e a t i n g the instruments f o r an a c t i o n a t a d i f f e r e n t l e v e l than the main ACTS that they explain. When looki n g a t a s t o r y a t one l e v e l o f d e t a i l we would n o t s e e the l e v e l of d e t a i l underneath i t unless spec i f i c a l l y c a l l e d upon t o do so. For example, consider the sentence, "John W e have p r e v i o u s l y repwent t o New York by bus." resented t h i s sentence by a simple ACT (PTRANS), B u t i t must b e and an instrumental a c t (PROPEL). r e a l i z e d that as w i t h any other s c r i p t , questions could be answered about t h i s sentence that were n o t s p e c i f i c a l l y i n i t . Subjects a l l seem t o agree that the answer t o "Was there a bus d r i v e r ? " i s "Yes" and t o "Did John pay money t o g e t on the bus?" i s "Probably." This seems t o i n d i c a t e t h a t the instrument o f John's PTRANS i s , i n a c t u a l i t y , the e n t i r e bus s c r i p t . Should we, as understanders, go so f a r as t o p l a c e the e n t i r e bus s c r i p t i n what we o b t a i n from understanding the above sentence? The answer seems obvious. You don't want t o do a l l that unl e s s you need to, but you want t o have quick access t o i t i n case you need t o . Consider the f o l l o w i n g s t o r y :
John wanted some cheesecake. He decided t o go t o New York. He went t o New York by bus.
On the bus he met a n i c e o l d lady who he t a l k e d t o about the p r i c e s i n the supermark e t . When he l e f t the bus he thanked the d r i v e r f o r the r i d e and found the subway t o On the subvay he was reading go t o Lindy's. the ads when suddenly he was robbed. He wasn't hurt though and he got o f f the t r a i n and entered Lindy's and had h i s cheesecake. When the check came, he said he couldn't pay . and was t o l d he would have t o wash dishes. L a t e r he went back t o New Haven. I d e a l l y , our representation o f t h i s s t o r y should account f o r the f a c t that hearers o f t h i s s t o r y i n v a r i a b l y f o r g e t the sentences about t a l k i n g t o the o l d lady and the bus d r i v e r , but always remember the mugging, i t s consequence of dishwashing, and the main g o a l o f going t o New York t o e a t cheesecake.
V I . Conclusion and Prognosis
It i s c l e a r that i n order t o understand one neras muwleage. Knowledge i s a p o t e n t i a l l y unw i e l d y thing, so what we must do is determine the types o f knowledge that t h e r e a r e and f i n d out how t o apply them. The SAM system i s a f i r s t s t e p a t adding structured knowledge t o the MARGIE system e are currently building (Schank e t a l , 1973). W up our knowledge base by adding more s c r i p t s t o SAM. I n a d d i t i o n we a r e adding a plan component PAM. These two programs should b r i n g us up t o the l e v e l o f understanding simple s t o r i e s about a l a r g e range of known domains. B u t what about complex s t o r i e s ? Is the kind o f understanding that humans e x h i b i t on real s t o r i e s l i k e l y t o resemble the mechanisms t o b e found i n SAM?
-
W e propose t o represent s t o r i e s t h e r e f o r e i n the f o l l c w i n g way: There w i l l b e a causal chain connecting the main events o f the s t o r y . (Here the PTRANS t o the restaurant, the INGEST, and the PTRANS back home.) Underneath each o f these main events w i l l be the instrumental s c r i p t t h a t underl i e s each o f them. (The bus s c r i p t , the subway s c r i p t , and the restaurant s c r i p t . ) These s c r i p t s w i l l be "forgotten" t o be reconstructed l a t e r , with the exception that any event t h a t occurred w i t h i n them that vas not predicated by them w i l l e m e m be placed on a "weird l i s t " t o be s p e c i a l l y r bered. The f i n a l representation of a s t o r y w i l l cons i s t o f the events connected d i r e c t l y t o the g o a l s and plans t o r e a l i z e those g o a l s made by the part i c i p a n t s . These g o a l s w i l l be t i e d t o the events that a c t u a l l y occurred and t o the weird events and t h e i r consequences. Thus four l i s t s r e p l a c e our o r i g i n a l (and growing) causal chain. An event l i s t (with s c r i p t events l e f t out); a g o a l l i s t ; a plan l i s t ; and a weird l i s t . What these l i s t s do is h e l p us f o r g e t . And o f course f o r g e t t i n g helps us remember. There are two ways i n which t h i s occurs: by omission and by prototyping. Events which enter none of the four l i s t s (such as the conversation v i t h the o l d lady), a r e dropped e n t i r e l y . (More p r e c i s e l y , they a r e retained only u n t i l the constructed f i n a l representation i s t r a n s f e r r e d from working memory t o long-term memory. Anything not in t h i s f i n a l representation i s l o s t . ) Also, the event l i s t and plan l i s t a r e condensed by using p o i n t e r s t o prototypes. The d e t a i l s a r e thus "normalized" ( B a r t l e t t , 1932) ; what i s remembered i s that a normal plan f o r s a t i s f y i n g such-and-such g o a l was used, including normal enactments of appropriate s c r i p t s . The function o f the weird l i s t i s t o mark the i n t e r e s t i n g departures from these normalities. What we a r e saying then i s that one o f the major issues i n A r t i f i c i a l I n t e l l i g e n c e research much be the c r e a t i o n o f the theory o f f o r g e t t i n g . I t simply i s not p o s s i b l e t o assume that people do, o r that machines should, remember everything they encounter. I n l i s t e n i n g t o a speaker, readi n g a book, o r engaging i n a conversation, people could n o t possibly remember everything they a r e t o l d verbatim. I n attempting t o g e t the g i s t o f a sequence, they must employ what we c a l l f o r g e t t i n g h e u r i s t i c s . As part o f these f o r g e t t i n g heurist i c s , a r e h e u r i s t i c s that search out items o f major importance. The s e l e c t i o n o f these major e don't r e a l l y items i s the key t o f o r g e t t i n g . W wish t o a s s e r t that people couldn't p o s s i b l y remember everything they hear. Rather we wish t o f i n d a procedure that w i l l l e t us see only the maj o r items, y e t a l s o f i n d , with some d i f f i c u l t y , the thoughts o r statements that u n d e r l i e them, and the ideas that u n d e r l i e those, and so on. Thus, the key t o understanding must be, i n order t o f a c i l i t a t e search among what has been understood, an o r g a n i z a t i o n o f the new information, i n such a fashion as t o seem to f o r g e t the unimportant m a t e r i a l and t o h i g h l i g h t the important material. F o r g e t t i n g h e u r i s t i c s must do t h i s f o r us. So the f i r s t task b e f o r e us i s t o e s t a b l i s h what the most s i g n i f i c a n t items i n a t e x t are l i k e l y t o be, and then t o e s t a b l i s h the h e u r i s t i c s which w i l l e x t r a c t and remember e x a c t l y those items.
References Abelson, R.P. (1973). The structure of b e l i e f systems. I n R.C. Schank and K.M. Colby (eds.), Computer m d e l s of thought and language. San Francisco: Freeman. Abelson, R.P. (1975a). Concepts f o r representing mundane r e a l i t y i n plans. I n D. Bobrow and A. C o l l i n s (eds.), Representation and understanding: Studies i n c o g n i t i v e science. New York: Academic.
Does a Story Understander Abelson, R.P. (1975b). need a p o i n t o f v i e w ? I n R.C. Schank and B. Nash-Webber (eds.), Using Knowledge t o Understand, i n Proceedings of the Conference on T h e o r e t i c a l Issues i n Natural Language Processing.
B a r t l e t t , F. (1932). s i t y Press. Remembering. Oxford Univer-
Dimensions o f representation. Bobrow, D. (1975). I n D. Bobrow and A. C o l l i n s (eds.), Representat i o n and understanding: Studies i n Cognitive-Science. New York: Academic. Towards a model o f c h i l Charniak, E. (1972). dren's s t o r y comprehension. A I TR-266, Mass. I n s t i t u t e o f Technology, Cambridge, Mass. Goldman (1975). Conceptual generation. I n R. Schank, Ed., Conceptual Information Processing. North Holland Publishing, Amsterdam. Uinsky, M. (1974). Frame-systems. AI. Memo. Mass. I n s t i t u t e o f Technology, Cambridge, Mass. Norman, D. (1972). Memory, knowledge, and the answering o f questions. Center f o r Human I n f o r mation Processing Memo CHIP-25. Univ. o f Cali f o r n i a a t San Diego. Rieger, C. (1975). Conceptual memory. I n R. Schank, ed., Conceptual Information Processinq. North Holland Publishing, Amsterdam. Riesbeck, C. (1975). Conceptual analysis. I n R. Schank, ed., Conceptual Information Processing. North Holland Publishing, Amsterdam. Schank, R. (1973a). I d e n t i f i c a t i o n o f conceptuali z a t i o n s underlying natural language. I n Schank and Colby (eds.), Computer Models o f Thought and Language. W.H. Freeman Press. Schank, R. (1973b). Causality and reasoning. Technical Report #1. I s t i t u t o per g l i s t u d i Semantic1 e C o g n i t i v i . Castagnola, Switzerland. Schank, R. (1975a). The Role o f Memory in Language Processing. To appear i n C. Cofer and R. Atkinson (eds.), The Nature o f Human Memory. W.H. Freeman Press. Schank, R. (197%). The structure o f episodes i n memory. I n D. Bobrow and T. Collins (eds.), Representation and understanding: Studies i n c o g n i t i v e science. New York: Academic.
Schank e t a l . (1973). R. Schank, N. Goldman, c . R i e g e r , and C. Riesbeck. MARGIE: Memory Analy s i s Response Generation and I n f e r e n c e on Engl i s h . Proceedings of the 3IJCAI.
I
I
Schank, R.
+ .
I n R. Schank and B. Nash-Webber (eds.), Proceedings o f the Conference on T h e o r e t i c a l Issues i n L a t u r a l Language Processing.
(1975~). Using Knowledge to Under-
Winograd, T. (1972). Understanding Natural Language. Academic Press. Winograd, T. (1975). Frame Representations and the Declarative/Procedural Controversy. I n D. Bobrow and A. C o l l i n s (eds.), Representation and Understanding: Studies i n C o g n i t i v e Science. New York: Academic.
157
t
USA
"Of what a stmnge nature i s b w t e d g e ! It cZings t o the mind, when i t has once seized on i t , like a
Zichen on the rock."
Frankenstein's Monster (M. Shelley, fiankenstein or the Modern Prwmetheus, 1818) Abstract We describe a t h e o r e t i c a l system intended to f a c i l i t a t e the use of knowledge i n an understanding system. The notion o f s c r i p t i s introduced to account f o r knovledge about mundane s i t u a t i o n s . A program, SAM, i s capable of using s c r i p t s to understand. The notion of plans i s introduced to account f o r general knowledge about novel situations.
-
I.
Preface
I n an attempt to provide theory where there have been mostly unrelated systems, Minsky (1974) recently described the work of Schank (1973a), Abelson (19731, Charniak (1972), and Norman (1972) as f i t t i n g into the notion of "frames." Piinsky attempted to r e l a t e t h i s work, i n what i s e s s e n t i a l l y language processing, t o areas of v i s i o n research that conform to the same notion. Minsky's frames paper has created q u i t e a s t i r i n AI and some immediate spinoff research along the l i n e s of developing frames manipulators e f i n d that ( e . g . Bobrow, 1975; Winograd, 1975). W we agree with much of what Minsky s a i d about frames and with h i s characterization of our own work. The frames idea i s so general, however, that i t does not lend i t s e l f to applications without further s p e c i a l i z a t i o n . This paper i s an attempt to develop further the l i n e s of thought s e t out i n Schank (1975a) and Abelson (1973; 1975a). The ideas presenced here can be viewed as a s p e c i a l i z a t i o n o f the frame idea. W e s h a l l r e f e r to our central constructs as " s c r i p t s . "
zation of knowledge can r e s u l t i n a r e a l understanding system i n the not too distant future. W e expect that programs based on the theory we outl i n e here and on our previous work on conceptual dependency and b e l i e f systems w i l l combine with the MARGIE system (Schank e t a l . , 1973a; Riesbeck, 1975; Rieger, 1975) to produce a working understander. We see understanding a s the f i t t i n g of new information i n t o a previously organized view of the world. We have therefore extended our work on language analysis (Schank, 1973a; piesbeck 1975) to understanding an understander. l i k e an analyzer, should be "bottom up" u n t i l i t gets enough information to make predictions and become "top down." E a r l i e r work has found various ways i n which a word in a s i n g l e sentence s e t s up expectations about what is l i k e l y to be found i n the r e s t of the sentence. A s i n g l e sentence and i t s corresponding conceptualizations s e t up expectations about what is to f o l l o w in the r e s t of a discourse o r story. These expectations characteri z e the world knowledge that bears on a given s i tuation, and i t is these expectations that we wish to explore.
-
111. S c r i p t s
A s c r i p t , as we use i t , is a structure that describes an appropriate sequence of events i n a p a r t i c u l a r context. A s c r i p t i s made up o f s l o t s and requirements about what can f i l l those s l o t s . The structure i s an interconnected whole, and what i s in one s l o t a f f e c t s what can be i n another. Scripts handle s t y l i z e d everyday s i t u a t i o n s . They a r e not subject to much change, nor do they provide the apparatus f o r handling novel situations, a s plans do (see section V). For our purposes, a s c r i p t i s a predetermined, stereotyped sequence of actions that d e f i n e a well-known situation. A s c r i p t i s , i n e f f e c t , a very boring l i t t l e story. Scripts a l l w f o r new references to o b j e c t s within them j u s t a s i f these objects had been previously mentioned; o b j e c t s within a s c r i p t may take "the" without e x p l i c i t introduction because the s c r i p t i t s e l f has a l ready i m p l i c i t l y introduced them. (This can b e found below, i n the reference to '*the waitress" i n a restaurant, f o r example.) Stor'ies can invoke s c r i p t s i n various ways. Usually a story Is a s c r i p t with one or m r e int e r e s t i n g deviations.
11.
The Problem
Researchers in n a t u r a l language understanding have f e l t f o r some time that the eventual limit on the solution of our problem w i l l b e our a b i l i t y to characterize world knowledge. Various researchers have approached world knowledge in various ways. Winograd (1972) d e a l t with the problem by severely r e s t r i c t i n g the world. This approach had the pos i t i v e e f f e c t of producing a working system and the negative e f f e c t o f producing m e that was only minimally extendable. Charniak (1972) approached the problem from the other end e n t i r e l y and has made some i n t e r e s t i n g f i r s t steps, but because h i s work is not grounded i n any representational system o r any working computational system the rest r i c t i o n o f world b o w l e d g e need not c r i t i c a l l y concern him. Our f e e l i n g is that an e f f e c t i v e characteri-
I. John went i n t o the restaurant.
4ie ordered a hamburger and a coke. He asked the waitress f o r the check and left.
11. John went to a restaurant.
Be ordered a hamburger. It was cold when the waitress brought i t . He l e f t her a very small t i p .
111. H a r r i e t went t o a birthday party.
t
The work of the second author was f a c i l i t a t e d by National Science Foundation Grant GS-35768.
151
She put on a green paper hat. Just when they s a t down to e a t the cake, a piece of p l a s t e r f e l l from the c e i l i n g Onto the t a b l e . She was lucky, because the dust didn't get a l l over her h a i r .
PTRANS s e l f to t a b l e WVE s i t d m scene 2: ordering ATRANS receive menu M R A N S read menu HBUILD decide what s e l f wants HTRANS order to waitress scene 3 : eating ATRANS receive food INGEST food scene 4: e x i t i n g receive check t i p to waitress s e l f to cashier money to cashier PTRANS. s e l f out of restaurant I n t h i s s c r i p t , the instruments f o r performing an action might vary with circumstances. For example, i n scene 2 the order might be'spoken, or w r i t t e n down with predesignated numbers f o r each item, o r even ( i n a f o r e i g n country with an unfam i l i a r language) indicated by pointing or gestures. Each act sequence uses the p r i n c i p l e of caus a l chaining (Schank, 1973b,Abelson, 1973). That i s , each action r e s u l t s i n conditions that enable the next to occur. To perform the next act i n the sequence, the previous a c t s must be completed sati s f a c t o r i l y . I f they cannot be, the hitches must b e d e a l t with. Perhaps a new action not prescribed i n the s c r i p t w i l l be generated i n order to get things moving again. This "what-if" behavior, t o b e discussed l a t e r , i s an important component of s c r i p t s . I t i s associated with many o f the deviations i n s t o r i e s such a s paragraph 11. In a t e x t , new s c r i p t information i s interpreted i n terms of i t s place i n one of the causal chains within the s c r i p t . Thus i n paragraph I the f i r s t sentence describes the f i r s t action i n scene 1 o f the restaurant s c r i p t . Sencence 2 r e f e r s to the l a s t action o f scene 2, and Sentence 3 t o the f i r s t and l a s t actions of scene 4. The f i n a l interpretation o f paragraph I contains the e n t i r e restaurant s c r i p t , with s p e c i f i c statements f i l l e d i n and missing statements (that he s a t down, for example) assumed. In paragraph 11, the f i r s t two sentences desc r i b e actions in scenes 1 and 2 . P a r t o f the t h i r d sentence i s in the s c r i p t as the f i r s t action of scene 3, but there i s a l s o the information that the hamburger i s cold. The fourth sentence ("He l e f t , her a very small tip") i s a modification of the t h i r d action of scene 4. The uwdifier, "very small," i s presumably r e l a t e d to the unexpected information about the "cold hamburger." Even a stupid processor, checking paragraph I1 against the standard restaurant s c r i p t , could come up with the low-level hypothesis that the small s i z e of the t i p must have something to do with the temperature of the hamburger, since these two Items of Information are the only deviations from the s c r i p t . They must b e r e l a t e d deviations, because i f they were unrelated the narrative would have no business ending v i t b two such unexphined features. O f course we do not vant our processor to be I n s l i g h t l y more complex examples, adestupid. quate understanding r e q u i r e s attention to the naATWS ATRANS PTRANS ATRANS
HBUILD where to s i t
I V . Harriet went to Jack's birthday party. The cake tasted awful. Harriet l e f t Jack's mother a very small t i p .
It i s Paragraph I i s an unmodified s c r i p t . d u l l . It would be even d u l l e r i f all the events i n the standard restaurant s c r i p t (see below) were included. Paragraph I 1 i s a restaurant s c r i p t with a stock v a r i a t i o n , a customer's t y p i c a l reaction when things go wrong. Paragraph I11 invokes the birthday party s c r i p t , but something wholly outside the range of normal birthday p a r t i e s occurs the p l a s t e r f a l l s from the c e i l i n g . This deviation from the s c r i p t takes over the i n i t i a t i v e i n the n a r r a t i v e u n t i l the problem i t r a i s e s i s resolved, but the birthday s c r i p t i s s t i l l a v a i l a b l e i n the indirect reference to the party hat and i n the p o s s i b i l i t y that normal party a c t i v i t i e s be resumed l a t e r i n the n a r r a t i v e . I t seems n a t u r a l f o r reference to be made to dust i n the h a i r following the p l a s t e r ' s f a l l i n g , which implies that there i s a kind of s c r i p t for f a l l i n g p l a s t e r too. (This kind of s c r i p t we c a l l a vignette (Abelson, 1 9 7 9 1 . ) NQt i c e that "the c e i l i n g " r e f e r s t o an uninteresting "room" s c r i p t , which can be used f o r references to doors and windows that may occur. Thus i t i s poss i b l e t o be i n more than one s c r i p t a t a time. Paragraph I V i l l u s t r a t e s the kind of absurdit y that a r i s e s when an action from one s c r i p t i s a r b i t r a r i l y inserted into another. That one f e e l s the absurdity i s an indication that s c r i p t s are i n inadmissable competition. I t i s conceivable that with adequate introduction the absurdity i n paragraph I V could be eliminated. With these examples, a number o f issues have been r a i s e d . Let us a t t h i s point g i v e a more extensive description o f s c r i p t s . W e have discussed previously (Schank, 1975b) how paragraphs are represented i n memory as causal chains. This work implies that, f o r a story to be understood, inferences must connect each input conceptualization t o a l l the others i n the story that r e l a t e to i t . This connection process i s f a c i l i t a t e d tremendousl y by the use of s c r i p t s . Scripts a r e extremely numerous. There i s a restaurant s c r i p t , a birthday party s c r i p t , a footb a l l game s c r i p t , a classroom s c r i p t , and so on. Each s c r i p t has players who assume r o l e s i n the action. A s c r i p t takes the point o f view of one of these p l a v e r s , and i t often changes when i t i s viewed from mother player's point o f view. The f o l l o w i n g i s a sketch of a s c r i p t f o r a restaurant from the point of view of the customer. Actions a r e s p e c i f i e d in terms of the primitive ACTS of conceptual dependency theory (Schank, 1973b.
M R A N S ask f o r check
-
script: roles: reason:
restaurant customer, w a i t r e s s , chef, cashier to get food so a s to go up i n pleasure and down i n hunger
scene 1: entering PTRANS s e l f i n t o restaurant ATTEND eyes to where empty tables a r e
ture o f d e v i a t i o n s from the s c r i p t . A smarter processor can i n f e r from a c o l d hamburger that the INGEST i n scene 3 w i l l then v i o l a t e the pleasure goal f o r going t o a restaurant. The concept o f a v e r y small t i p can be stored with the restaurant s c r i p t as a what-if associated w i t h v i o l a t i o n s o f the pleasure goal. The general form f o r a s c r i p t , then, i s a s e t o f paths j o i n e d a t c e r t a i n c r u c i a l points that def i n e the s c r i p t . For restaurants the c r u c i a l p a r t s are the INGEST and the ATRANS of money. There are many normal ways t o move from p o i n t t o point. Ordering may b e done by M " S i n g to a w a i t e r or by s e l e c t i n g and taking what you l i k e (in a cafeteria). Likewise the ATRANS o f money may be done by going t o the cashier, o r paying the There w a i t r e s s , o r saying, "Put i t on my b i l l . " a r e a l s o paths t o take when s i t u a t i o n s don't go as planned. Paragraphs 1 1 1 and I V call up deviant paths i n the birthday party s c r i p t . All these v a r i a t i o n s i n d i c a t e that a s c r i p t i s not a simple l i s t o f events but rather a linked causal chain; a s c r i p t can branch into m u l t i p l e p o s s i b l e paths that come together a t c r u c i a l d e f i n i n g points. To know when a s c r i p t i s appropriate, s c r i p t headers a r e necessary. These headers d e f i n e the circumstances under which a . s c r i p t i s c a l l e d i n t o p l a y . The headers f o r the restaurant s c r i p t a r e concepts having t o do with hunger, restaurants, and so on i n the context of a plan o f a c t i o n f o r g e t t i n g fed. Obviously contexts must be r e s t r i c t ed t o avoid c a l l i n g up the restaurant s c r i p t f o r sentences that use the word "restaurant" as a place ("Fuel o i l was d e l i v e r e d t o the restaurant"). S c r i p t s organize new inputs i n terms o f prev i o u s l y stored knowledge. I n paragraph I, many items-that a r e p a r t of the restaurant s c r i p t a r e added t o the f i n a l i n t e r p r e t a t i o n o f the s t o r y . W e don't need t o say that a w a i t r e s s took the customer's order or that he a t e the hamburger. These Ideas a r e f i r m l y a p a r t o f the s t o r y because the restaurant s c r i p t r e q u i r e s them. I n understanding a s t o r y that c a l l s up a s c r i p t , the s c r i p t becomes p a r t of the s t o r y even when i t i s not s p e l l e d out. The answer t o the question W h o served John the hamburger?" seems obvious, because our world knowledge, as embodied i n s c r i p t s , answers i t .
it.
I n scene 2 of the restaurant s c r i p t , if what-ifs. the w a i t r e s s i g n o r e s the customer, he w i l l t r y t o catch her eye or c a l l t o her when she passes nearby. I f he can't make out the menu o r needs further information, he w i l l ask the w a i t r e s s . If she doesn't speak h i s language, he w i l l attempt her language, or make gestures, o r seek another customer t o t r a n s l a t e , o r accept her suggestion o f what t o order. I n scene 3, i f the w a i t r e s s does not b r i n g the food, he w i l l again t r y t o catch her eye. I f the food is not f i t , he w i l l send i t back. Errors have a s l i g h t l y d i f f e r e n t character from o b s t a c l e s but follow the same general r u l e s . Receiving the menu i s errorful i f the w a i t r e s s ATRANSes a p r i n t e d sheet t o the customer but i t i s yesterday's menu, or the breakfast instead o f the dinner menu. Reading the menu may y i e l d an e r r o r i f the customer g e t s the wrong idea o f what i t says say, he thinks f i l e t mignon i s a f i s h . Eere i t i s up t o the w a i t r e s s t o supply the whati f c o r r e c t i v e . Deciding what t o order may y i e l d an e r r o r i f the customer goes through the d e c i s i o n process but f o r g e t s the stored outcome. The-what; i f i s t o review the MBUILD ("Let's see now, what Giving the o r d e r t o the w a i t r e s s d i d I decide?") may b e i n e r r o r if she w r i t e s down something o t h e r than what the customer said, o r omits a p o r t i o n o f i t . The what-if i s t o repeat the order, asking the w a i t r e s s whether she i s sure she g o t a l l o f
-
I V . The SAM Program
what-Ifs
There are a t l e a s t three major ways i n which s c r i p t s can be thrown off normal course. One i s d i s t r a c t i o n , i n t e r r u p t i o n by another s c r i p t , such as the plaster's f a l l i n g from the c e i l i n g . W e w i l l not pursue here an a n a l y s i s of t h e conditions The other two and consequences of d i s t r a c t i o n . ways, o b s t a c l e and e r r o r , a r e i n t i m a t e l y connected with what-if behavior. An o b s t a c l e t o the normal seauence occurs when someone o r something prevents a normal a c t i o n from occurring or some usual enab l i n g condition for the a c t i o n i s absent. An e r r o r occurs when the a c t i o n i s completed i n an inappropriate manner, so that the normal consequences of the a c t i o n do n o t come about. I n p r i n c i p l e , e v e r y simple ACT i n a standard e ass c r i p t has p o t e n t i a l o b s t a c l e s and e r r o r s . W sume t h a t , e v e r y time an o b s t a c l e o r e r r o r occurs i n a s c r i p t t h a t i s being learned, the methods used t o remove the o b s t a c l e or redeem the e r r o r The rea r e s t o r e d w i t h the s c r i p t as what-ifs. sult of many r e p e t i t i o n s i s t h a t mst of the common what-ifs a r e attached t o the s c r i p t . Every o b s t a c l e has one or mre c h a r a c t e r i s t i c
We have created a program that uses s c r i p t s t o make i n f e r e n c e s i n domains i t knows about and have added i t t o the MARGIE program reported a t 3IJCAI. SAM ( S c r i p t A p p l i e r Mechanism) r e c e i v e s Conceptual Dependency structures as input (these a r e gotten from English by using Riesbeck's parser SAM i d e n t i f i e s the s c r i p t t o be used (1975)). (by using Riesbeck's request format) and f i l l s i n the i d e n t i f i e d s c r i p t w i t h the new inputs as they a r e received. It makes the inferences that a r e necessary t o complete a causal chain between input conceptualizations (see Schank (1973b) f o r a discussion o f causal chains). The resultant comp l e x structure can then be queried by input Engl i s h questions o r be used t o make summaries and paraphrases o f the i n i t i a l s t o r y . SAM was w r i t ten by Chris Riesbeck, Richard C u l l i n g f o r d and Wendy Lehnert. Goldmau's generation program (1975) i s used t o map the answers of paraphrases I n t o English. I t was modified f o r t h i s purpose by Gerald DeJong and W.J. Stutzman. An example of SAM's input and output i s : Input: John went t o a restaurant. The hostess seate d John. The hostess gave John a menu. The w a i t e r came t o the t a b l e . John ordered lobster. John was served quickly. John l e f t a Large t i p . John l e f t the restaurant. Questions and Answers:
Q: What d i d John e a t ? A: LOBSTER. Q: Who gave John the menu? A: THE HOSTESS. Q: Who gave John the l o b s t e r ? A: PROBABLY THE WAITER. Q: Who paid the check? A: PROBABLY JOHN.
153
A: Q: A: Q: A:
Q: What happened vhen John vent t o the t a b l e ? THE HOSTESS GAVE HIM A MENU AND JOHN SAT DOWN.
Why d i d John get a menu? SO HE COULD ORDER. Why d i d John give the w a i t e r a l a r g e t i p ?
A low-level
goal is i l l u s t r a t e d by:
John wanted t o cut h i s steak. He c a l l e d to h i s w i f e i n the kitchen.
BECAUSE HE WAS SERVED QUICKLY.
Paraphrase: JOHN DECIDED HE WAS GOING TO GO TO A B E S T W RANT. JOHN WENT TU A RESTAURANT. JOHN INDICATED TO THE HOST HE WOULD LIKE TO BE AT A TABLE. THE HOST INDICATED TO JOHN HE COULD GO TO A TABLE. JOAN WENT TO THE TABLE. THE HCST WENT TO THE TABLE. JOHN SAT DOWN I N THE CKAIR. JOHN GOT THE MENU FROM THE HOST. JOHN READ T?E MENU. THE WAITER SAW JOHN WAS AT THE TABLE. THE WAITER WENT TO THE TABLE. JOHN ORDERED LOBSTER. THE C€EF PREPARED THE LOBSTER. THE WAITER GOT THE LOBSTER FROM THE CHEF. WE WAITER SERVED JOHN THE LOBSTER. JOHN ATE THE LOBSTER. JOHN INDICATED TO TEE WAITER HE WOULD LIKE TO GET THE CHECK FROM HIM. JOHN GOT THE CHECK FROM THE WAITER. JOHN L E F I THE WAITER A LARGE TIP. JOHN PAID THE CHECK. JOHN LEFT THE RESTAURANT. Summary: JOHN WENT TO A RESTAURANT AND ATE LOBSTER. This program runs on the PDP-10 a t Yale. It currently has only a small amount of knowledge and a small vocabulary. But we f e e l encouraged that our s c r i p t theory i s workable because of the simp l i f i c a t i o n i n the inference process that has resulted from the use of s c r i p t s .
A plan is a s e r i e s o f actions that w i l l reali z e a goal. Often in order t o r e a l i z e one goal another must b e decided on and a plan drawn up to achieve i t . I n the f i r s t example above, a goal to a t t a i n power i s reduced to a goal to get arsenic. High-level goals a r e more i n t e r e s t i n g and v e have concentrated on them f i r s t . W e d e f i n e a "deltact" as an action o r a group o f actions that l e a d s to a desired s t a t e . Delt a c t s constitute subplans that a r e pursued because of t h e i r intended e f f e c t s . There a r e f i v e d e l t a c t s i n the present system:
AAGENCY ACONT AKNOW APROX ASOCCONT
-a
-
change i n o b l i g a t i o n t o do something f o r somebody a change i n the control of an o b j e c t a change i n vhat an a c t o r knows a change in the proximity r e l a t i o n s of o b j e c t s and a c t o r s a change in s o c i a l control over a person o r a s i t u a t i o n
v.
Plans
Plans are responsible f o r the d e l i b e r a t e behavior that people e x h i b i t . Plans describe the s e t of choices that a person has vhen he s e t s out to accomplish a goal. In l i s t e n i n g to a discourse, people use plans to make sense of seemingl y disconnected sentences. By f i n d i n g a plan, an understander can make guesses about the intentions of an action i n an unfolding story and use these guesses to make sense o f the story. Consider the f o l l o w i n g paragraph:
John knew that h i s w i f e ' s operation would
be very expensive. There v a s always Uncle Harry He reached f o r the suburban phone book.
...
How a r e we t o make sense o f such a paragraph? It makes no use of headers o r the s c r i p t s they sign a l . It would b e unreasonable to p o s i t a "paying f o r an operation" s c r i p t v i t h a l l the necessary a c t s l a i d out as in our restaurant s c r i p t . But, on the other hand, the s i t u a t i o n i s not e n t i r e l y novel, e i t h e r . The problem of understanding t h i s paragraph would not b e s i g n i f i c a n t l y d i f f e r e n t i f " w i f e ' s operation" were changed t o "son's education" or "down payment on the mortgage." There i s a general goal s t a t e in each case, r a i s i n g a l o t of money f o r a l e g i t i m a t e expense, and there is a generalized plan or group of plans that may l e a d to the goal s t a t e . Plans s t a r t with one o r more g o a l s . A highl e v e l g o a l i s i l l u s t r a t e d by the sequence:
John wanted to become king. Be vent to get Bome arsenic.
There is a l s o a s e t of lower-level d e l t a c t s & e l son, 1975a). Plans are made up of deltacts. When a collocation of d e l t a c t s i s used often enough, i t becomes a s c r i p t . A plan includes a s e t of planboxes, l i s t s of actions that w i l l y i e l d s t a t e changes and the preconditions f o r these actions, along with a s e t of questions f o r choosing the appropriate planbox. For instance, the TAKE plan has the goal o f enabling the taker to do something with an o b j e c t , whatever i s generally done with i t . T o TAKE something you must b e c l o s e to i t , so e i t h e r the obj e c t and the taker must b e i n the same location o r the taker must use a subplan APROX. Either no one e l s e must have CONTROL of the o b j e c t o r a t l e a s t there must be no bad consequences i n the taker's attempt to PTRANS the o b j e c t to himself. The TAKE plan c a l l s a PTRANS of the object i f a l l the preconditions a r e p o s i t i v e . But i f , say, someone e l s e CONTROLS the obj e c t , a plan f o r the taker's gaining CONTROL must b e c a l l e d . This subplan i s ACONT. ACONT has a Bet of planboxes attached t o i t . These planboxes define a d e l t a c t j u s t a s inferences define a primitive ACT. A planbox i s a l i s t of primitive ACTS that w i l l achieve a goal. Associated with each ACT are i t s precondltions, and a planbox checks them. A s e t o f p o s i t i v e conditions a l l o w s the desired ACT. Negative conditions c a l l up new planboxes o r d e l t a c t s that have as t h e i r g o a l the resolution o f the negative s t a t e . Preconditions f a l l into three c l a s s e s . A controlled precondition can b e f i x e d when i t i s negative by doing an ACT. A negative uncontrolled precondition cannot b e f i x e d , and another planbox must b e t r i e d . Negative mediating preconditions can b e a l t e r e d but require plans of t h e i r own t o change. Mediating preconditions usually r e f e r to the w i l l i n g n e s s o f other p a r t i e s to p a r t i c i p a t e i n plans. Further d e t a i l s on planboxes appear i n Schank ( 1 9 7 5 ~ ) . To see why an understander needs plans, cons i d e r the following sequence:
W i l l a vas hungry.
She took out the Michelin Guide.
Most readers understand that V i l l a was using the
154
Michelin Guide t o f i n d a good restaurant. But if the f i r s t sentence were subjected t o s t r a i g h t f o r ward inference (a l a R i e g e r , 1975), p r e d i c t i n g that W i l l a i s l i k e l y t o do something t o enable hers e l f t o INGEST food, the second sentence would seem t o answer t h i s p r e d i c t i o n o n l y i n the weird i n t e r p r e t a t i o n that she w i l l e a t the Michelin Guide. An understander w i l l r e j e c t t h i s i n f a v o r o f any b e t t e r path that i t can f i n d . The f i r s t sentence w i l l be analyzed f o r any g o a l that might generate a plan. "Hungry" i s l i s t e d i n the dictionary as i n d i c a t i n g the need f o r a plan t o do a ACONT o f food. One means f o r gaining c o n t r o l o f food i s a restaurant. An enablement f o r t h i s means i s going t o a restaurant, which r e q u i r e s APROX. This i n turn r e q u i r e s knowing where you are going, which may r e q u i r e AKNOW. I n the d i c t i o n a r y , a l l books a r e l i s t e d as means o f s a t i s f y i n g AKNOWs and the Michelin Guide i s l i s t e d as a book. To complete the processing o f t h i s sequence i t would, of course, be necessary t o have the information that the Michelin Guide l i s t s restaurants. Without t h i s information, the sequence might be as nonsensical as ' V i l l a was hungry. She took out Introduction t o A r t i f i c i a l Intelligence. W i t h the information that the Michelin Guide i s a source o f knowledge about restaurants, we know why the second a c t i o n was done and can pred i c t future actions. W e have transformed a seemi n g l y disconnected sequence i n t o one that provides the expectations that a r e so v i t a l t o understanding. I f the next sentence i s ' V i l l a g o t . i n her car," we w i l l know that the plan i s being e f f e c t e d . By using what we know about cars (that they a r e instruments o f PTRANS) and the s c r i p t f o r restaurants ( t h a t i s s t a r t s w i t h a PTRANS), we can make the inference that W i l l a i s on her way t o a restaurant. Some restaurant header would s t i l l be required t o i n i t i a t e the restaurant s c r i p t i n i t s f u l l glory. The procedure o f taking out the Michelin Guide when hungry, w h i l e seemingly n o v e l , could conceivably be routine for a c e r t a i n i n d i v i d u a l i n a c e r t a i n context. I f we know t h a t W i l l a i s a gourmet t o u r i s t staying i n P a r i s who enjoys going t o a d i f f e r e n t restaurant every evening, then the procedure o f looking i n the Guide might become p a r t o f h e r restaurant s c r i p t . For her there i s a scene b e f o r e scene 1 i n which she ATTENDS t o the Guide, MBUILDs a choice, and MTRhNSes a reservat i o n . A r o u t i n i z e d plan can become a s c r i p t , a t l e a s t from the planner's p o i n t o f view.
When a person reads a 300 page novel he does not (unless he i s v e r y unusual) remember a l l the conceptualizations s t a t e d i n the s t o r y i n the form of a g i a n t causal chain. Rather he remembers the g i s t of the book. Maybe 5 o r 10 pages o f summary could be extracted from him a f t e r reading the book. P r e v i o u s l y we have s a i d that Conceptual Dependency Theory w i l l account f a r memry f o r g i s t o f sentences. B u t i t cannot be s e r i o u s l y proposed that t h i s i s a l l that i s needed f o r g i s t o f long and complex s t o r i e s . Some o t h e r explanat i o n must be given. I n a recent experiment, Abelson [1975b] showed that people remember s t o r i e s b e t t e r when they a r e asked t o take some p a r t i c u l a r p o i n t of view ( o f one o f the p a r t i c i p a n t s o r o f an observer i n a p a r t i c u l a r place), and that what they remember i s contingent on which p o i n t of view they had. The r a m i f i c a t i o n s o f t h i s experiment f o r a theory o f language understanding have, t o be that when people have a c l u e o f what t o f o r g e t they do b e t t e r a t remembering. I n other words, good f o r g e t t i n g i s the key t o remembering. Likewise, if we want t o b u i l d programs that remember, we had b e s t teach them how t o f o r g e t . One method of f o r g e t t i n g i s simply not n o t i c i n g l e v e l s o f det a i l that a r e there. This can be done by t r e a t i n g the instruments f o r an a c t i o n a t a d i f f e r e n t l e v e l than the main ACTS that they explain. When looki n g a t a s t o r y a t one l e v e l o f d e t a i l we would n o t s e e the l e v e l of d e t a i l underneath i t unless spec i f i c a l l y c a l l e d upon t o do so. For example, consider the sentence, "John W e have p r e v i o u s l y repwent t o New York by bus." resented t h i s sentence by a simple ACT (PTRANS), B u t i t must b e and an instrumental a c t (PROPEL). r e a l i z e d that as w i t h any other s c r i p t , questions could be answered about t h i s sentence that were n o t s p e c i f i c a l l y i n i t . Subjects a l l seem t o agree that the answer t o "Was there a bus d r i v e r ? " i s "Yes" and t o "Did John pay money t o g e t on the bus?" i s "Probably." This seems t o i n d i c a t e t h a t the instrument o f John's PTRANS i s , i n a c t u a l i t y , the e n t i r e bus s c r i p t . Should we, as understanders, go so f a r as t o p l a c e the e n t i r e bus s c r i p t i n what we o b t a i n from understanding the above sentence? The answer seems obvious. You don't want t o do a l l that unl e s s you need to, but you want t o have quick access t o i t i n case you need t o . Consider the f o l l o w i n g s t o r y :
John wanted some cheesecake. He decided t o go t o New York. He went t o New York by bus.
On the bus he met a n i c e o l d lady who he t a l k e d t o about the p r i c e s i n the supermark e t . When he l e f t the bus he thanked the d r i v e r f o r the r i d e and found the subway t o On the subvay he was reading go t o Lindy's. the ads when suddenly he was robbed. He wasn't hurt though and he got o f f the t r a i n and entered Lindy's and had h i s cheesecake. When the check came, he said he couldn't pay . and was t o l d he would have t o wash dishes. L a t e r he went back t o New Haven. I d e a l l y , our representation o f t h i s s t o r y should account f o r the f a c t that hearers o f t h i s s t o r y i n v a r i a b l y f o r g e t the sentences about t a l k i n g t o the o l d lady and the bus d r i v e r , but always remember the mugging, i t s consequence of dishwashing, and the main g o a l o f going t o New York t o e a t cheesecake.
V I . Conclusion and Prognosis
It i s c l e a r that i n order t o understand one neras muwleage. Knowledge i s a p o t e n t i a l l y unw i e l d y thing, so what we must do is determine the types o f knowledge that t h e r e a r e and f i n d out how t o apply them. The SAM system i s a f i r s t s t e p a t adding structured knowledge t o the MARGIE system e are currently building (Schank e t a l , 1973). W up our knowledge base by adding more s c r i p t s t o SAM. I n a d d i t i o n we a r e adding a plan component PAM. These two programs should b r i n g us up t o the l e v e l o f understanding simple s t o r i e s about a l a r g e range of known domains. B u t what about complex s t o r i e s ? Is the kind o f understanding that humans e x h i b i t on real s t o r i e s l i k e l y t o resemble the mechanisms t o b e found i n SAM?
-
W e propose t o represent s t o r i e s t h e r e f o r e i n the f o l l c w i n g way: There w i l l b e a causal chain connecting the main events o f the s t o r y . (Here the PTRANS t o the restaurant, the INGEST, and the PTRANS back home.) Underneath each o f these main events w i l l be the instrumental s c r i p t t h a t underl i e s each o f them. (The bus s c r i p t , the subway s c r i p t , and the restaurant s c r i p t . ) These s c r i p t s w i l l be "forgotten" t o be reconstructed l a t e r , with the exception that any event t h a t occurred w i t h i n them that vas not predicated by them w i l l e m e m be placed on a "weird l i s t " t o be s p e c i a l l y r bered. The f i n a l representation of a s t o r y w i l l cons i s t o f the events connected d i r e c t l y t o the g o a l s and plans t o r e a l i z e those g o a l s made by the part i c i p a n t s . These g o a l s w i l l be t i e d t o the events that a c t u a l l y occurred and t o the weird events and t h e i r consequences. Thus four l i s t s r e p l a c e our o r i g i n a l (and growing) causal chain. An event l i s t (with s c r i p t events l e f t out); a g o a l l i s t ; a plan l i s t ; and a weird l i s t . What these l i s t s do is h e l p us f o r g e t . And o f course f o r g e t t i n g helps us remember. There are two ways i n which t h i s occurs: by omission and by prototyping. Events which enter none of the four l i s t s (such as the conversation v i t h the o l d lady), a r e dropped e n t i r e l y . (More p r e c i s e l y , they a r e retained only u n t i l the constructed f i n a l representation i s t r a n s f e r r e d from working memory t o long-term memory. Anything not in t h i s f i n a l representation i s l o s t . ) Also, the event l i s t and plan l i s t a r e condensed by using p o i n t e r s t o prototypes. The d e t a i l s a r e thus "normalized" ( B a r t l e t t , 1932) ; what i s remembered i s that a normal plan f o r s a t i s f y i n g such-and-such g o a l was used, including normal enactments of appropriate s c r i p t s . The function o f the weird l i s t i s t o mark the i n t e r e s t i n g departures from these normalities. What we a r e saying then i s that one o f the major issues i n A r t i f i c i a l I n t e l l i g e n c e research much be the c r e a t i o n o f the theory o f f o r g e t t i n g . I t simply i s not p o s s i b l e t o assume that people do, o r that machines should, remember everything they encounter. I n l i s t e n i n g t o a speaker, readi n g a book, o r engaging i n a conversation, people could n o t possibly remember everything they a r e t o l d verbatim. I n attempting t o g e t the g i s t o f a sequence, they must employ what we c a l l f o r g e t t i n g h e u r i s t i c s . As part o f these f o r g e t t i n g heurist i c s , a r e h e u r i s t i c s that search out items o f major importance. The s e l e c t i o n o f these major e don't r e a l l y items i s the key t o f o r g e t t i n g . W wish t o a s s e r t that people couldn't p o s s i b l y remember everything they hear. Rather we wish t o f i n d a procedure that w i l l l e t us see only the maj o r items, y e t a l s o f i n d , with some d i f f i c u l t y , the thoughts o r statements that u n d e r l i e them, and the ideas that u n d e r l i e those, and so on. Thus, the key t o understanding must be, i n order t o f a c i l i t a t e search among what has been understood, an o r g a n i z a t i o n o f the new information, i n such a fashion as t o seem to f o r g e t the unimportant m a t e r i a l and t o h i g h l i g h t the important material. F o r g e t t i n g h e u r i s t i c s must do t h i s f o r us. So the f i r s t task b e f o r e us i s t o e s t a b l i s h what the most s i g n i f i c a n t items i n a t e x t are l i k e l y t o be, and then t o e s t a b l i s h the h e u r i s t i c s which w i l l e x t r a c t and remember e x a c t l y those items.
References Abelson, R.P. (1973). The structure of b e l i e f systems. I n R.C. Schank and K.M. Colby (eds.), Computer m d e l s of thought and language. San Francisco: Freeman. Abelson, R.P. (1975a). Concepts f o r representing mundane r e a l i t y i n plans. I n D. Bobrow and A. C o l l i n s (eds.), Representation and understanding: Studies i n c o g n i t i v e science. New York: Academic.
Does a Story Understander Abelson, R.P. (1975b). need a p o i n t o f v i e w ? I n R.C. Schank and B. Nash-Webber (eds.), Using Knowledge t o Understand, i n Proceedings of the Conference on T h e o r e t i c a l Issues i n Natural Language Processing.
B a r t l e t t , F. (1932). s i t y Press. Remembering. Oxford Univer-
Dimensions o f representation. Bobrow, D. (1975). I n D. Bobrow and A. C o l l i n s (eds.), Representat i o n and understanding: Studies i n Cognitive-Science. New York: Academic. Towards a model o f c h i l Charniak, E. (1972). dren's s t o r y comprehension. A I TR-266, Mass. I n s t i t u t e o f Technology, Cambridge, Mass. Goldman (1975). Conceptual generation. I n R. Schank, Ed., Conceptual Information Processing. North Holland Publishing, Amsterdam. Uinsky, M. (1974). Frame-systems. AI. Memo. Mass. I n s t i t u t e o f Technology, Cambridge, Mass. Norman, D. (1972). Memory, knowledge, and the answering o f questions. Center f o r Human I n f o r mation Processing Memo CHIP-25. Univ. o f Cali f o r n i a a t San Diego. Rieger, C. (1975). Conceptual memory. I n R. Schank, ed., Conceptual Information Processinq. North Holland Publishing, Amsterdam. Riesbeck, C. (1975). Conceptual analysis. I n R. Schank, ed., Conceptual Information Processing. North Holland Publishing, Amsterdam. Schank, R. (1973a). I d e n t i f i c a t i o n o f conceptuali z a t i o n s underlying natural language. I n Schank and Colby (eds.), Computer Models o f Thought and Language. W.H. Freeman Press. Schank, R. (1973b). Causality and reasoning. Technical Report #1. I s t i t u t o per g l i s t u d i Semantic1 e C o g n i t i v i . Castagnola, Switzerland. Schank, R. (1975a). The Role o f Memory in Language Processing. To appear i n C. Cofer and R. Atkinson (eds.), The Nature o f Human Memory. W.H. Freeman Press. Schank, R. (197%). The structure o f episodes i n memory. I n D. Bobrow and T. Collins (eds.), Representation and understanding: Studies i n c o g n i t i v e science. New York: Academic.
Schank e t a l . (1973). R. Schank, N. Goldman, c . R i e g e r , and C. Riesbeck. MARGIE: Memory Analy s i s Response Generation and I n f e r e n c e on Engl i s h . Proceedings of the 3IJCAI.
I
I
Schank, R.
+ .
I n R. Schank and B. Nash-Webber (eds.), Proceedings o f the Conference on T h e o r e t i c a l Issues i n L a t u r a l Language Processing.
(1975~). Using Knowledge to Under-
Winograd, T. (1972). Understanding Natural Language. Academic Press. Winograd, T. (1975). Frame Representations and the Declarative/Procedural Controversy. I n D. Bobrow and A. C o l l i n s (eds.), Representation and Understanding: Studies i n C o g n i t i v e Science. New York: Academic.
157
