Generating Lexical Analogies Using Dependency Relations

Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pp. 561–570, Prague, June 2007. c⃝2007 Association for Computational Linguistics Generating Lexical Analogies Using Dependency Relations Andy Chiu, Pascal Poupart, and Chrysanne DiMarco David R. Cheriton School of Computer Science University of Waterloo Waterloo, Ontario, Canada {pachiu,ppoupart,cdimarco}@uwaterloo.ca Abstract A lexical analogy is a pair of word-pairs that share a similar semantic relation. Lex- ical analogies occur frequently in text and are useful in various natural language pro- cessing tasks. In this study, we present a system that generates lexical analogies au- tomatically from text data. Our system dis- covers semantically related pairs of words by using dependency relations, and applies novel machine learning algorithms to match these word-pairs to form lexical analogies. Empirical evaluation shows that our system generates valid lexical analogies with a pre- cision of 70%, and produces quality output although not at the level of the best human- generated lexical analogies. 1 Introduction Analogy discovery and analogical reasoning are ac- tive research areas in a multitude of disciplines, in- cluding philosophy, psychology, cognitive science, linguistics, and artificial intelligence. A type of anal- ogy that is of particular interest in natural language processing is lexical analogy. A lexical analogy is a pair of word-pairs that share a similar semantic rela- tion. For example, the word-pairs (dalmatian, dog) and (trout, fish) form a lexical analogy because dal- matian is a subspecies of dog just as trout is a sub- species of fish, and the word-pairs (metal, electric- ity) and (air, sound) form a lexical analogy because in both cases the initial word serves as a conductor for the second word. Lexical analogies occur fre- quently in text and are useful in various natural lan- guage processing tasks. For example, understanding metaphoric language such as “the printer died” re- quires the recognition of implicit lexical analogies, in this case between (printer, malfunction) and (per- son, death). Lexical analogies also have applica- tions in word sense disambiguation, information ex- traction, question-answering, and semantic relation classification (see (Turney, 2006)). In this study, we present a novel system for gen- erating lexical analogies directly from a text cor- pus without relying on dictionaries or other seman- tic resources. Our system uses dependency relations to characterize pairs of semantically related words, then compares the similarity of their semantic rela- tions using two machine learning algorithms. We also present an empirical evaluation that shows our system generates valid lexical analogies with a pre- cision of 70%. Section 2 provides a list of defini- tions, notations, and necessary background materi- als. Section 3 describes the methods used in our system. Section 4 presents our empirical evalua- tion. Section 5 reviews selected related work. Fi- nally, Section 6 concludes the paper with suggested future work and a brief conclusion. 2 Definitions A word-pair is a pair of entities, where each entity is a single word or a multi-word named entity. The underlying relations of a word-pair (w1, w2) are …

Content truncated to 500 words