Proceedings

Delayed Interpretation Using Processing Notes

Download BibTeX Cite

Proceedings of the Fifth International Joint Conference on Artificial Intelligence. Specialized Systems. Pages 813--1.

© 1977 International Joint Conferences on Artificial Intelligence

Abstract

Full text extracted automatically from the PDF; figures, tables and equations may differ. The PDF is the version of record.

Christopher K. Riesbeck Yale University !lew Haven, CT 06511

Descriptive Keywo r:.ls and Phrases

natural langua?,e understanding - semantic repre­ sentat lon :nemory and inference knmlledge structures - memory uwdels

~a_ c_kg_ roun_d

ELI (Eng lish Language Interpreter), the natural language analyzer used by the Artificial Intelligence project at Yale (Riesbeck and Schank, 1976, Schank, 1976 ), builds Conceptual Dependency ( CD) forms (Schank, 1975) to represent its semantic interpretations of input texts. By including in the CD forms special eleoaents, called _proc_e_ss _in_g I!ot_es_ , ELI can indicate what parts of the CD form later inferencing progrruns must con plete. Three different notes have been developed so far.

The REF note

ELI uses the REF note to guide the instan­ tiatlon of noun phrases into memory tokens. For exar:1ple, "he" is analyzed into "(PERSON GENDCR ·-tALE REF DEF)" and "somone is analyzed into 111 (P~KSOtl GENDER HALE REF INDEF)". "REF DEF" in­ dicates that existing me1anry tokens should be looked at . "REF INDEF" indicates that a new to­ ken is needed.

_Ih_e_ S_PECH~Y n_o_te_

ELI uses the SPCCIFY note to indicate a hole tha t needs to be fi _lled in. For example, the question "1-lhere is he?" is represented as "I want you to tell me that he is in location SPECIFY ·" In CD this is:

(CON (ACTOR HEARER <=> HTAANS TO S PEAKEK HOBJt::CT (ACTOR (PERSON REl' DEF) IS (LOG VAL SPECIFY))) LEADTO (ACTOR SPEAKER TOWARD JOY INC 2))

_!h_e E_QU_lv_ n_~te

EI,.l uses the EQUIV note to say that an ob­ ject can be described with several different COs. For example, "Hary is the one who went to Boston" is represented as "the person named Mary is EQUIV to the person who 1~ent to Boston." In CD this is:

*This ~10rl<, was supported in part by the Advanced Research Projects Agency of the Department of Defense and monitored under the Off ice of Naval Research under contract N00014-75-C-llll.

(ACTOR (PERSON NAHE MARY) EQUI V (PEKSON REL (ACTOR PERSON <·> PTRANS TO (CITY NAME tiOSTON)))

A nore complex EQUIV note is used for sen­ tences of the form "Sl but S2" where Sl is an action from a knowledge structure (KS) such as a script or a plan (Schank and Abelson, in press). "But" in these sentences means that S2 is con­ trary tu the intentions of the KS. The inten­ tions of a KS are the goals associated with that KS, plus the actions that the KS says will lead to those goals.

The sentence fra~ae "Sl but S2" is repre­ sented as "52 is not EQUIV to any event leading to an event that is a goal of the KS in Sl." In

CD this is:

((CON S2 EQUIV (SPECIFYl

REL (CON SPECIFYl LEADTO (SPECIFY2 RE:L (KS

GOAL SPECIFY2)))))

~!ODE NEG)

Conclusion

The addition of processing notes like REF, SPECIFY, and EQUIV allows not only better commu­ nication between ELI and the inference programs, but a lso allm~s us to represent the meaning of words like "but" which make meta-comments about the texts in 1~hich they appear.

  • Riesbeck, C. and Schank, R. C. (1976) Compre­ hension by Computer: Expectation-based Analysis of Sentences in Context. Yale Uni­ versity, Department of Computer Science Re­ search Report 78.
  • Schank, R. C. (1975). Conceptual dependency: A theory of natural language understanding. ~o_g_I!:i,~i..~~ P.~y_t:_ho_l,_o_g_y_ 3(4), 552-631.
  • Schank, R. c. (1976). Research at Yale in Natural Language Processing. Yale Univer­ sity, Department of Computer Science Research Report 84.
  • Schank, R. C. and Abelson, R. P. (in Knowledge Structures. Lawrence Pr_e_s_s_ .- -Hillscii te" ,--N .-x:· press). Erlbaum
Welcome Future Conferences AI Journal Division IJCAI Proceedings About IJCAI IJCAI Awards