[54] NATURAL LANGUAGE UNDERSTANDING SYSTEM
[75] Inventors: Kathleen Dahlgren. Edward Stabler.
both of Los Angeles. Calif.
[73] Assignee: Intelligent Text Processing, Inc.. Santa Monica. Calif.
[21] Appl. No.: 943,069
[22] Filed: Oct 2,1997
Related U.S. Application Data
[63] Continuation of Ser. No. 369,034, Jan. 4, 1995, abandoned.
[51] Int. CI.6 G06F 9/45
[52] U.S. CI 395/708; 395/705; 395/707
[58] Field of Search 395/705.707,
395/708
[56] References Cited
U.S. PATENT DOCUMENTS
4,887,212 12/1989 Zamora et al 364/419.08
4,974,191 11/1990 Amnirghodsi et al 364/419.08
5,062,074 10/1991 Heinberger 395/600
5,200,893 4/1993 Ozawa et al 364/419.1
5,265,065 11/1993 Turtle.
5,331,554 7/1994 Graham 364/419.07
5,331,556 7/1994 Black, Jr. et al 395/759
5,369,577 11/1994 Kadashevich et al 364/419.13
5,404^95 4/1995 Katz et al 364/419.19
5,418,948 5/1995 Turtle 395/600
5,488,725 1/1996 Turtle 395/600
5,519,608 5/1996 Kupiec 364/419.08
5,526,259 6/1996 Kaji 364/419.03
5,541,836 7/1996 Church et al 364/419.07
OTHER PUBLICATIONS
Dahlgren, K., "Naive Semantics for Natural Language Understanding," Kluwer Academic Publishers, Norwell, MA, CH's 1-8, 1988.
Primary Examiner—Emanuel Todd Voeltz
Assistant Examiner—Peter J. Corcoran. Ill
Attorney, Agent, or Finn—Hecker & Harriman
[57] ABSTRACT
The present invention interprets natural language input using common sense reasoning. The invention avoids the combinatorial explosion that has occurred in other natural language understanding systems. The invention uses modules for parsing, disambiguation, formal semantics, anaphora resolution, and coherence, and a naive semantic lexicon. The naive semantic lexicon is consulted by the parsing, disambiguation, formal semantics, anaphora resolution, and coherence modules to determine whether an interpretation alternative is plausible based on the world knowledge contained in the naive semantic lexicon. The parsing module employs both a top-down and bottom-up parsing strategy. The parsing module consults the naive semantic lexicon to build a structure from natural language input that has both semantic and pragmatic plausibility. The invention uses a psychologically-motivated naive semantic ontology that provides a means for classifying concepts. The lexicon relates word senses to the ontological concepts, contains word sense-specific common sense knowledge, and connects syntactic information with the meaning of each word sense. Using the natural language understanding of the present invention, a process is used for retrieving text that includes the steps of: 1) natural language understanding of a document base. 2) natural language understanding of a text retrieval request (i.e.. query), 3) a comparison of the output of steps 1 and 2.
30 Claims, 6 Drawing Sheets
