Local and Global Context for Supervised and Unsupervised Metonymy Resolution

Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 183–193, Jeju Island, Korea, 12–14 July 2012. c⃝2012 Association for Computational Linguistics Local and Global Context for Supervised and Unsupervised Metonymy Resolution Vivi Nastase HITS gGmbH Heidelberg, Germany vivi.nastase@h-its.org Alex Judea University of Stuttgart Stuttgart, Germany alexander.judea@ims.uni-stuttgart.de Katja Markert University of Leeds Leeds, UK K.Markert@leeds.ac.uk Michael Strube HITS gGmbH Heidelberg, Germany michael.strube@h-its.org Abstract Computational approaches to metonymy res- olution have focused almost exclusively on the local context, especially the constraints placed on a potentially metonymic word by its grammatical collocates. We expand such approaches by taking into account the larger context. Our algorithm is tested on the data from the metonymy resolution task (Task 8) at SemEval 2007. The results show that incorpo- ration of the global context can improve over the use of the local context alone, depending on the types of metonymies addressed. As a second contribution, we move towards unsu- pervised resolution of metonymies, made fea- sible by considering ontological relations as possible readings. We show that such an unsu- pervised approach delivers promising results: it beats the supervised most frequent sense baseline and performs close to a supervised approach using only standard lexico-syntactic features. 1 Introduction With the exception of explicit tasks in metonymy and metaphor analysis, computational treatment of language relies on the assumption that the texts to be processed have a literal interpretation. This contrasts with the fact that figurative expressions are com- mon in language, as exemplified by the metonymy in the excerpt from a Wikipedia article in Exam- ple 1 and another in Example 2 from the SemEval 2007 metonymy resolution task (Markert and Nis- sim, 2009). (1) In the gold medal game, Canada defeated the American team 2-0 to win their third consecu- tive gold. (2) This keyword is only required when your rela- tional database is Oracle. The defeating in Example 1 will not be done by the country as such, but by a team represent- ing the country in a sporting event. Hence, in a metonymy a potentially metonymic expression or word (here Canada) stands for a conceptually re- lated entity (here, people of Canada). In the sec- ond Example, a company name (Oracle) stands for a product (database) developed by the company. Metonymy resolution can be important for a variety of tasks. Textual entailment may need metonymy resolution (Bentivogli et al., 2007): for example, we would like to be able to induce from Example 1 the hypothesis The Canadian team won .... Leveling and Hartrumpf (2008) show that metonymy recognition on location proper names helps geographical information retrieval by ex- cluding metonymically used place names from consideration (such as Example 1 or the use of Vietnam for the Vietnam war). Metonymies also fre- quently interact with anaphora resolution (Nunberg, 1995; Markert and Hahn, 2002), as in Example 1 where the metonymic use of Canada is referred to by a plural pronoun afterward (their). Metonymies can be quite regular: company …

Content truncated to 500 words