BORIS - An In-Dept Understander of Narratives

BORIS -- AN IN-DEPTH UNDERSTANDER

OF NARRATIVES

Michael G. Dyer, Thomas C. Wolf, Martin Korsin

Computer Science Department Yale University New Haven, CT 06520

BORIS [4) is an experimental computer program designed to understand and answer questions about complicated narratives. By 'complicated' we mean narratives which require the specification and interaction of multiple sources of knowledge, including: scripts, plans, goals, interpersonal relationships, object pr1m1t1ves, affects, settings, social acts, physical states, roles, etc.

In addition, BORIS partially implements a theory of HOPs (Memory Organization Packets) [5) which is used to represent and process such abstrac'.ions as: legal disputes, contracts, and personal favors.

While an earlier version of BORIS read a kidnapping story [2), the specific knowledge currently maintained in BORIS allows it to process two narratives, each involving a divorce. Although these narratives vary significantly (i.e. lexically, syntactically, and in terms of their outcomes and causal connections), the knowledge structures used by BORIS are general enough to

'handle both versions.

The demo will show an execution trace of BORIS comprehending a port ion of DIVORCE-2, a six paragraph story which begins:

George vas having lunch with another teacher and grading hoaevork aeeigoaeot1 vheo the vaitreu accidentally knocked a glau of coke oo hia. George vas very annoyed and left refusing to pay the check. Be decided to drive .boae to get out of hie vet clothe·.

When be got there, be found hie wife Aoo and another aao io bed. George became extremely upeet and felt like going out and getting plastered ···

and answering such questions as:

• Q: Why didn't George pay the check?
• A: Becauee the vaitre·· epilled coke oo hia.
• Q: Bov did Aoo feel vheo George caught her cheating oo bia?
• A: She vae aurpri1ed.

This work supported in part by Advanced Research Projects Agency contract N0014-75-C-lll and National Science Foundation contract IST7918463.

Instead of skimming [I) many stories, BORIS tries to understand a few complicated stories in as great a depth as possible. To accomplish in-depth understanding, BORIS employs the following:

• A memory overlay technique which allows BORIS to model multiple levels and multiple perspectives of knowledge.
• A theory of TAUs (Thematic Affect Units) [3) [4), designed for handling expectation failures and norm violations as they occur within narratives.

Another general issue examined in BORIS is processing integration. By 'integration' we mean that episodic memory is searched, and often created, while each word in a sentence is being parsed. One consequence of this approach (during question answering) is that BORIS often knows the answer to a quest io'n before the quest ion has been completely understood.

R!F!II.ENCES

• (1] DeJong II, Gerald F. "Skimming Stories in Real Time: An Experiment in Integrated Understanding." Technical Report 158, Yale University. Department of Computer Science,
• 1979.
• [2) Dyer, Michael G. and Wendy G. Lehnert. "Organization and Search Processes for Narratives." Technical Report 175, Yale University. Department of Computer Science, April 1980.
• Dyer, Michael G. "The Role of TAUs in Narratives." In Proc. of Cognitive Science Conference. Held at University of California, Berkeley. August 1981.
• [ 4) Wendy G. Lehnert, Michael G. Dyer, Peter N. Johnson, CJ Yang, and Steve Harley. "BORIS -- An Experiment in In-Depth Understanding of Narratives." Technical Report 188, Yale University. Department of Computer Science,
• December 1980.
• Schank, Roger C. "Reminding and Memory Organization: An Introduction to HOPs." In W. Lehnert and H. Ringle, editors, Strategiee for Natural Language Proceeeiog. Lawrence Erlbaum Assoc. (in press), 1981.