PAM - A Program That Infers Intentions
Proceedings of the Fifth International Joint Conference on Artificial Intelligence. Natural Language. Pages 15--1.
© 1977 International Joint Conferences on Artificial Intelligence
Abstract
- Schank, R. C. and Ableson, R. P. (1977). ~'?_'!_l_s. •. P_l._ '!_'!S_!. ~<::_~!:.P_t:_~ <!.!!.<! U_l!c:!_~~t_a_~d-~:_ f §l. ~n_qu_i:_ry i,_n~~ l!!!_m_'!_l! ~o~l_~d.&~ ~t.~l!_C_t:_l!_r.~s..!.
- Erlbaum Press, N. J. Wilensky, R. (1976). !!_s_i_n_g_ _ I'_l,_'!_n~ .!:.~ Understand ~a.t:.l!.t"2.l. l::~'!&.'!a.Jl~• In ~r-~t:_~~c!!_n_g_~ ~{ -~tl~-~9:!~ .!!.O.l!S_t~l! ' _7_6_.__ ACM, New York. 'latllrc.l Lanp;llap:e-1: 1-Ji 1 ensky
References were extracted automatically from the PDF and may contain errors.
Full text extracted automatically from the PDF; figures, tables and equations may differ. The PDF is the version of record.
PAM - A Program That Infers Intentions* Robert \Ulensky Yale University
New Haven, Ct. 06520
An intelligent story understanding program requires a great deal of knowledge about human intentionality. For example, consider the fol lowing story:
- (l) John needed money.
- lie got a gun
- and walked into a liquor store.
- John told the owner he wanted his money.
- The 0\mer gave John the money
- and John left.
To understand this story : we must realize that John's actions were motivated by a plan to obtain money by threatening someone. Since this inter pretation is not explicit in the story, it must be inferred by the understander.
PAM (Plan Applier Mechanism) is a program that understands stories by analyzing the inten tions of the story's characters, and relating their · intentions to their actions. PA11 reads in sentences in English, produces Conceptual De pendency (CD) meaning representations for them, and connects together these CD's to form a story representation. The story representation con tains, in addition to the actual events of the story, a Knowledge Structure Interpretation (KSI), which explains why these events occurred.
For example, PM! can understand the story c~posed of sentences 1-6 above, and produces the following KSI (simplified for expository purposes):
EPISODE:
TYPE: GOAL-EPISODE
GOAL SPECIFICATION:
TYPE: DELTA-GONT PLANNER: (JOHN)
OBJECT: (MONEY)
PLAN SPECIFICATION:
OBJECT: (HONEY REL (LIQUOR-STORE POSS t10NEY))
PLAN: THREATEN ACTION: ( ··· )
STATUS: FULFILLED
SUBEPISODES:
TYPE: GOAL-EPISODE
GOAL SPECIFICATION:
TYPE: DELTA-CONT PLANNER:
OB.JECT: (WEAPON)
PLAN SPECIFICIATION:
(JOHN)
TYPE: GOAL-EPISODE
*This work was supported in part by the Advanced Research Projects Agency of the Department of Defense and monitored under the Office of Naval Research .under contract N00014-75-C-llll.
GOAL SPECIFICATION: TYPE: DELTA-PROX PLANNER: (JOHN)
LOCATION: (PROX PART (HONEY))
PLAN SPECIFICIATIO!l:
RESPONSE:
TYPE: GOAL-GONFLICT
GOAL SPECIFICATION:
TYPE: PRESERVE-HEALTH
PLANNER: (OWNER REL LIQUOR-STORE)
STATUS: FULFILLED
GOAL SPECIFICATION:
TYPE: PRESERVE-GONT
PLANNER: (OWNER REL LIQUOR-STORE)
OBJECT:
(HONEY REL LIQUOR-STORE)
STATUS: ABANDONED
PLAN: COM.PL Y
ACTIO II: ( ··· )
The whole story is interpreted as a GOAL EPISODE, or sequence of events describing the pursuit of a goal. The goal is a DELTA-GmlT (a transfer of control type goal) of some money from its owner to John. The plan for the goal con sists of selecting some particular money (the 1 iquo r store's) and then threatening its owner. In implementing this plan, several subgoals had to be achieved: Being near the money (described in the abbreviated DELTA-PROX sub-episode), and possessing a weapon (the DELTA-GONT sub-episode). The ACTION field denotes a pointer to the CD representing the realization of the threat plan.
The RESPONSE shows that the threat caused the liquor store owner to have two conflicting goals, that of preserving his money and that of preserving his well-being. The status of the PRESERVE-GONT goal indicates that the 0\mer has abandoned this goal for the PRESERVE-HEALTH goal,
for 1~hich he has chosen to COUPLY with John's demand.
In order to produce this KSI, PAM stores its knowledge of goals and plans in the form of con dition-action rules. Typically, the application of a rule gives rise to an expectation that is used to interpret future sentences. For example, upon encountering sentence (2) above, PM! ac cesses the rule: "If a person obtains a func tional object, then he intends to use it." This rule gives rise to an expectation that John will use the gun to threaten or to overpower someone. When sentence (4) is encountered, this expecta tion causes PM! to interpret it as a realization of a THREATEN plan, as opposed to merely ex pressing one's desire. It is through the use of such expectations that PM1 effects contextual understanding.
Schank, R. C. and Ableson, R. P. (1977). ~'?_'!_l_s. ·. P_l._'!_'!S_!. ~<::_~!:.P_t:_~ <!.!!.<! U_l!c:!_~~t_a_~d-~:_ f§l. ~n_qu_i:_ry i,_n~~ l!!!_m_'!_l! ~o~l_~d.&~ ~t.~l!_C_t:_l!_r.~s..!. Erlbaum Press, N. J.
Wilensky, R. (1976). !!_s_i_n_g_ _I'_l,_'!_n~ .!:.~ Understand ~a.t:.l!.t"2.l. l::~'!&.'!a.Jl~· In ~r-~t:_~~c!!_n_g_~ ~{ -~tl~-~9:!~ .!!.O.l!S_t~l! '_7_6_.__ ACM, New York.
@inproceedings{ijcai1977p3,
title = {PAM - A Program That Infers Intentions},
author = {Wilensky, Robert},
booktitle = {Proceedings of the Fifth International Joint Conference on
Artificial Intelligence, {IJCAI-77}},
publisher = {International Joint Conferences on Artificial Intelligence},
editor = {Raj Reddy},
pages = {15---1},
year = {1977},
note = {Natural Language},
url = {https://ijcai.org/proceedings/1977/3},
}
TY - CPAPER TI - PAM - A Program That Infers Intentions AU - Wilensky, Robert T2 - Proceedings of the Fifth International Joint Conference on Artificial Intelligence, IJCAI-77 PB - International Joint Conferences on Artificial Intelligence PY - 1977 SP - 15 EP - -1 UR - https://ijcai.org/proceedings/1977/3 ER -Copied
Machine-readable versions of this paper, for tools and AI agents: