What Psycholinguists Know About Chemistry: Aligning Wiktionary and WordNet for Increased Domain Coverage

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 883–892, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP What Psycholinguists Know About Chemistry: Aligning Wiktionary and WordNet for Increased Domain Coverage Christian M. Meyer and Iryna Gurevych Ubiquitous Knowledge Processing Lab Technische Universit¨at Darmstadt Hochschulstraße 10, 64289 Darmstadt, Germany http://www.ukp.tu-darmstadt.de Abstract By today, no lexical resource can claim to be fully comprehensive or perform best for every NLP task. This caused a steep increase of resource alignment research. An important challenge is thereby the alignment of differently represented word senses, which we address in this paper. In particular, we propose a new automatically aligned resource of Wiktionary and Word- Net that has (i) a very high domain cov- erage of word senses and (ii) an enriched sense representation, including pronunci- ations, etymologies, translations, etc. We evaluate our alignment both quantitatively and qualitatively, and explore how it can contribute to practical tasks. 1 Introduction Though WordNet has been extensively used in knowledge-rich natural language processing (NLP) systems, there is no best lexical resource for all purposes. Jarmasz and Szpakowicz (2003), for example, found better results for solving word choice problems when using Roget’s thesaurus in- stead of WordNet. There is indeed a large number of different lexical resources: The ACL Special Interest Group on the Lexicon1 lists, for instance, more than 40 different lexical resources on their homepage that have been proposed as a source of background knowledge for different NLP tasks. These resources typically differ in two ways: (i) They have a different coverage of words and word senses, and (ii) they encode heterogeneous types of information that is attached to their words and word senses. This heterogeneity ranges from very fundamental differences, like the distinction between lexicographic and encyclopedic know- ledge to more specific ones, such as one re- 1http://www.siglex.org/, accessed 2011-05-10 Figure 1: Wiktionary article ‘plant’ source encodes semantic frames, while another focuses on subsumption relations between word senses. Using WordNet without further consid- erations thus limits the performance of a system, since each resource has its individual advantages. This has caused increasing research in the area of lexical resource alignment. It has been shown that aligned resources yield synergies, which lead to better performance than using the resources in- dividually. For instance, Shi and Mihalcea (2005) improve semantic parsing using the knowledge of an aligned resource of FrameNet, WordNet, and VerbNet. Recently, Ponzetto and Navigli (2010) observed improvements for coarse-grained and domain-specific word sense disambiguation using an alignment between WordNet and Wikipedia for adding new relations to WordNet. In another line of research, the community based online dictionary Wiktionary has been suc- cessfully applied in several NLP tasks, such as cross-lingual image retrieval (Etzioni et al., 2007), named entity recognition (Richman and Schone, 2008), or synonymy mining (Navarro et al., 2009; Sajous et al., 2010). Zesch et al. (2008b) compare different semantic relatedness measures using ei- ther WordNet, Wikipedia, or Wiktionary and find 883 the best results for Wiktionary. Besides its large …

Content truncated to 500 words