Hunting Elusive Metaphors Using Lexical Resources.

Proceedings of the Workshop on Computational Approaches to Figurative Language, pages 13–20, Rochester, NY, April 26, 2007. c⃝2007 Association for Computational Linguistics Hunting Elusive Metaphors Using Lexical Resources Saisuresh Krishnakumaran∗ Computer Sciences Department University of Wisconsin-Madison Madison, WI 53706 ksai@cs.wisc.edu Xiaojin Zhu Computer Sciences Department University of Wisconsin-Madison Madison, WI 53706 jerryzhu@cs.wisc.edu Abstract In this paper we propose algorithms to automatically classify sentences into metaphoric or normal usages. Our algo- rithms only need the WordNet and bigram counts, and does not require training. We present empirical results on a test set de- rived from the Master Metaphor List. We also discuss issues that make classification of metaphors a tough problem in general. 1 Introduction Metaphor is an interesting figure of speech which expresses an analogy between two seemingly un- related concepts. Metaphoric usages enhance the attributes of the source concept by comparing it with the attributes of the target concept. Abstrac- tions and enormously complex situations are rou- tinely understood via metaphors (Lakoff and John- son, 1980). Metaphors begin their lives as Novel Poetic Creations with marked rhetoric effects whose comprehension requires special imaginative leap. As time goes by, they become part of general use and their comprehension becomes automatic and idiomatic and rhetoric effect is dulled (Nunberg, 1987). We term such metaphors whose idiomatic ef- fects are dulled because of common usage as dead metaphors while metaphors with novel usages as live metaphors. In this paper we are interested only in identifying live metaphors. ∗ The fi rst author is currently affi liated with Google Inc, Mountain View, CA. Metaphors have interesting applications in many NLP problems like machine translation, text sum- marization, information retrieval and question an- swering. Consider the task of summarizing a parable which is a metaphoric story with a moral. The best summary of a parable is the moral. Paraphrasing a metaphoric passage like a parable is difficult with- out understanding the metaphoric uses. The per- formance of the conventional summarizing systems will be ineffective because they cannot identify such metaphoric usages. Also it is easy to create novel and interesting uses of metaphors as long as one con- cept is explained in terms of another concept. The performance of machine translation systems will be affected in such cases especially if they have not en- countered such metaphoric uses before. Metaphor identification in text documents is, however, complicated by issues including con- text sensitiveness, emergence of novel metaphoric forms, and the need for semantic knowledge about the sentences. Metaphoric appeal differs across lan- guage or people’s prior exposure to such usages. In addition, as (Gibbs, 1984) points out, literal and fig- urative expressions are end points of a single con- tinuum along which metaphoricity and idiomaticity are situated, thereby making clear demarcation of metaphoric and normal usages fuzzy. We discuss many such issues that make the task of classifying sentences into metaphoric or non- metaphoric difficult. We then focuses on a sub- set of metaphoric usages involving the nouns in a sentence. In particular, we …

Content truncated to 500 words