Investigating the cross-linguistic potential of VerbNet-style classification

Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010), pages 1056–1064, Beijing, August 2010 Investigating the cross-linguistic potential of VerbNet -style classification Lin Sun and Anna Korhonen Computer Laboratory University of Cambridge ls418,alk23@cl.cam.ac.uk Thierry Poibeau LaTTiCe, UMR8094 CNRS & ENS thierry.poibeau@ens.fr C´edric Messiant LIPN, UMR7030 CNRS & U. Paris 13 cedric.messiant@lipn.fr Abstract Verb classes which integrate a wide range of linguistic properties (Levin, 1993) have proved useful for natural language pro- cessing (NLP) applications. However, the real-world use of these classes has been limited because for most languages, no resources similar to VerbNet (Kipper- Schuler, 2005) are available. We apply a verb clustering approach developed for English to French – a language for which no such experiment has been conducted yet. Our investigation shows that not only the general methodology but also the best performing features are transferable be- tween the languages, making it possible to learn useful VerbNet style classes for French automatically without language- specific tuning. 1 Introduction A number of verb classifications have been built to support natural language processing (NLP) tasks (Grishman et al., 1994; Miller, 1995; Baker et al., 1998; Palmer et al., 2005; Kipper-Schuler, 2005; Hovy et al., 2006). These include both syntactic and semantic classifications, as well as ones which integrate aspects of both. Classifications which in- tegrate a wide range of linguistic properties can be particularly useful for NLP applications suffer- ing from data sparseness. One such classification is VerbNet (Kipper-Schuler, 2005). Building on the taxonomy of Levin (1993), VerbNet groups verbs (e.g. deliver, post, dispatch) into classes (e.g. SEND) on the basis of their shared mean- ing components and syntactic behaviour, identi- fied in terms of meaning preserving diathesis al- ternations. Such classes can be identified across the entire lexicon, and they may also apply across languages, since their meaning components are said to be cross-linguistically applicable (Jack- endoff, 1990). Offering a powerful tool for generalization, ab- straction and prediction, VerbNet classes have been used to support many important NLP tasks, including e.g. computational lexicography, parsing, word sense disambiguation, semantic role labeling, information extraction, question- answering, and machine translation (Swier and Stevenson, 2004; Dang, 2004; Shi and Mihalcea, 2005; Abend et al., 2008). However, to date their exploitation has been limited because for most languages, no Levin style classification is avail- able. Since manual classification is costly (Kipper et al., 2008) automatic approaches have been pro- posed recently which could be used to learn novel classifications in a cost-effective manner (Joanis et al., 2008; Li and Brew, 2008; ´O S´eaghdha and Copestake, 2008; Vlachos et al., 2009; Sun and Korhonen, 2009). However, most work on Levin type classification has focussed on English. Large-scale research on other languages such as German (Schulte im Walde, 2006) and Japanese (Suzuki and Fukumoto, 2009) has focussed on se- mantic classification. Although the two classifica- tion systems have shared properties, studies com- paring the overlap between VerbNet and WordNet (Miller, 1995) have reported that the mapping is only partial and many to many due …

Content truncated to 500 words