Metaphor Detection with Cross-Lingual Model Transfer

Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, pages 248–258, Baltimore, Maryland, USA, June 23-25 2014. c⃝2014 Association for Computational Linguistics Metaphor Detection with Cross-Lingual Model Transfer Yulia Tsvetkov Leonid Boytsov Anatole Gershman Eric Nyberg Chris Dyer Language Technologies Institute Carnegie Mellon University Pittsburgh, PA 15213 USA {ytsvetko, srchvrs, anatoleg, ehn, cdyer}@cs.cmu.edu Abstract We show that it is possible to reliably dis- criminate whether a syntactic construction is meant literally or metaphorically using lexical semantic features of the words that participate in the construction. Our model is constructed using English resources, and we obtain state-of-the-art performance relative to previous work in this language. Using a model transfer approach by piv- oting through a bilingual dictionary, we show our model can identify metaphoric expressions in other languages. We pro- vide results on three new test sets in Span- ish, Farsi, and Russian. The results sup- port the hypothesis that metaphors are conceptual, rather than lexical, in nature. 1 Introduction Lakoff and Johnson (1980) characterize metaphor as reasoning about one thing in terms of another, i.e., a metaphor is a type of conceptual mapping, where words or phrases are applied to objects and actions in ways that do not permit a literal inter- pretation. They argue that metaphors play a fun- damental communicative role in verbal and writ- ten interactions, claiming that much of our every- day language is delivered in metaphorical terms. There is empirical evidence supporting the claim: recent corpus studies have estimated that the pro- portion of words used metaphorically ranges from 5% to 20% (Steen et al., 2010), and Thibodeau and Boroditsky (2011) provide evidence that a choice of metaphors affects decision making. Given the prevalence and importance of metaphoric language, effective automatic detec- tion of metaphors would have a number of ben- efits, both practical and scientific. Language pro- cessing applications that need to understand lan- guage or preserve meaning (information extrac- tion, machine translation, dialog systems, senti- ment analysis, and text analytics, etc.) would have access to a potentially useful high-level bit of in- formation about whether something is to be under- stood literally or not. Second, scientific hypothe- ses about metaphoric language could be tested more easily at a larger scale with automation. However, metaphor detection is a hard problem. On one hand, there is a subjective component: hu- mans may disagree whether a particular expres- sion is used metaphorically or not, as there is no clear-cut semantic distinction between figurative and metaphorical language (Shutova, 2010). On the other, metaphors can be domain- and context- dependent.1 Previous work has focused on metaphor identi- fication in English, using both extensive manually- created linguistic resources (Mason, 2004; Gedi- gian et al., 2006; Krishnakumaran and Zhu, 2007; Turney et al., 2011; Broadwell et al., 2013) and corpus-based approaches (Birke and Sarkar, 2007; Shutova et al., 2013; Neuman et al., 2013; Shutova and Sun, 2013; Hovy et al., 2013). We build on this foundation and also extend metaphor detec- tion into other languages in which …

Content truncated to 500 words