From: AAAI Technical Report WS-92-01. Compilation copyright © 1992, AAAI (www.aaai.org). All rights reserved. Refining Automatically-Discovered Lexical Relations: Combining Weak Techniques for Stronger Results Marti A. Hearst Gregory Grefenstette ComputerScience Division Department of Computer Science 571 Evans Hall 210 MIB University of California, Berkeley University of Pittsburgh Berkeley, CA94720 Pittsburgh, PA 15260 raa rti @cs.berkeley, edu
[email protected], edu Abstract level language processing techniques in isolation often do not suffice for a particular task; for this reason we Knowledge-poor corpus-based approaches to nat- are interested in finding ways to combine various ap- ural language processing are attractive in that proaches and improve their results. they do not incur the difficulties associated with complex knowledge bases and real-world infer- Accordingly, we conducted experiments to refine the ences. However, these kinds of language process- results of an automatic lexical discovery technique by ing techniques in isolation often do not suffice for makinguse of a statistically-based syntactic similarity a particular task; for this reason we are interested measure, and integrating them with an existing knowl- in finding ways to combine various techniques and edge structure. The discovery program uses lexieo- improvetheir results. syntactic patterns to find instances of the hyponymy Accordingly, we conducted experiments to refine (i.e., ISA) relation in large text bases. Once relations the results of an automatic lexical discovery tech- of this sort are found, they should be inserted into an existing lexicon or thesaurus. However, the terms in nique by making use of a statistically-based syn- the relation may have multiple senses, thus hampering tactic similarity measure.