Computing Word-Pair Antonymy

Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing, pages 982–991, Honolulu, October 2008. c⃝2008 Association for Computational Linguistics Computing Word-Pair Antonymy Saif Mohammad† Bonnie Dorr � Graeme Hirstφ † � Laboratory for Computational Linguistics and Information Processing † � Institute for Advanced Computer Studies and � Computer Science † � University of Maryland and � Human Language Technology Center of Excellence � saif,bonnie � @umiacs.umd.edu φDepartment of Computer Science University of Toronto gh@cs.toronto.edu Abstract Knowing the degree of antonymy between words has widespread applications in natural language processing. Manually-created lexi- cons have limited coverage and do not include most semantically contrasting word pairs. We present a new automatic and empirical mea- sure of antonymy that combines corpus statis- tics with the structure of a published the- saurus. The approach is evaluated on a set of closest-opposite questions, obtaining a preci- sion of over 80%. Along the way, we discuss what humans consider antonymous and how antonymy manifests itself in utterances. 1 Introduction Native speakers of a language intuitively recog- nize different degrees of antonymy—whether two words are strongly antonymous (hot–cold, good– bad, friend–enemy), just semantically contrasting (enemy–fan, cold–lukewarm, ascend–slip) or not antonymous at all (penguin–clown, cold–chilly, boat–rudder). Over the years, many definitions of antonymy have been proposed by linguists (Cruse, 1986; Lehrer and Lehrer, 1982), cognitive scien- tists (Kagan, 1984), psycholinguists (Deese, 1965), and lexicographers (Egan, 1984), which differ from each other in small and large respects. In its strictest sense, antonymy applies to gradable adjec- tives, such as hot–cold and tall–short, where the two words represent the two ends of a semantic dimension. In a broader sense, it includes other adjectives, nouns, and verbs as well (life–death, ascend–descend, shout–whisper). In its broadest sense, it applies to any two words that represent contrasting meanings. We will use the term de- gree of antonymy to encompass the complete se- mantic range—a combined measure of the contrast in meaning conveyed by two words and the tendency of native speakers to call them opposites. The higher the degree of antonymy between a target word pair, the greater the semantic contrast between them and the greater their tendency to be considered antonym pairs by native speakers. Automatically determining the degree of antonymy between words has many uses includ- ing detecting and generating paraphrases (The dementors caught Sirius Black / Black could not escape the dementors) and detecting contradictions (Marneffe et al., 2008; Voorhees, 2008) (Kyoto has a predominantly wet climate / It is mostly dry in Kyoto). Of course, such “contradictions” may be a result of differing sentiment, new information, non-coreferent mentions, or genuinely contradictory statements. Antonyms often indicate the discourse relation of contrast (Marcu and Echihabi, 2002). They are also useful for detecting humor (Mihalcea and Strapparava, 2005), as satire and jokes tend to have contradictions and oxymorons. Lastly, it is useful to know which words are semantically contrasting to a target word, even if simply to filter them out. For example, in the automatic creation of a thesaurus …

Content truncated to 500 words