New Features for FrameNet - WordNet Mapping

Proceedings of the Thirteenth Conference on Computational Natural Language Learning (CoNLL), pages 219–227, Boulder, Colorado, June 2009. c⃝2009 Association for Computational Linguistics New Features for FrameNet – WordNet Mapping Sara Tonelli and Daniele Pighin FBK-Irst, Human Language Technologies Via di Sommarive, 18 I-38100 Povo (TN) Italy {satonelli,pighin}@fbk.eu Abstract Many applications in the context of natural language processing or information retrieval may be largely improved if they were able to fully exploit the rich semantic information an- notated in high-quality, publicly available re- sources such as the FrameNet and the Word- Net databases. Nevertheless, the practical use of similar resources is often biased by the limited coverage of semantic phenomena that they provide. A natural solution to this problem would be to automatically establish anchors between these resources that would allow us 1) to jointly use the encoded information, thus possibly over- coming limitations of the individual corpora, and 2) to extend each resource coverage by ex- ploiting the information encoded in the others. In this paper, we present a supervised learn- ing framework for the mapping of FrameNet lexical units onto WordNet synsets based on a reduced set of novel and semantically rich features. The automatically learnt mapping, which we call MapNet, can be used 1) to ex- tend frame sets in the English FrameNet, 2) to populate frame sets in the Italian FrameNet via MultiWordNet and 3) to add frame labels to the MultiSemCor corpus. Our evaluation on these tasks shows that the proposed approach is viable and can result in accurate automatic annotations. 1 Introduction In recent years, the integration of manually-built lexical resources into NLP systems has received growing interest. In particular, resources annotated with the surface realization of semantic roles, like FrameNet (Baker et al., 1998) or PropBank (Palmer et al., 2005) have shown to convey an improve- ment in several NLP tasks, from question answer- ing (Shen and Lapata, 2007) to textual entailment (Burchardt et al., 2007) and shallow semantic pars- ing (Giuglea and Moschitti, 2006). Nonetheless, the main limitation of such resources is their poor cov- erage, particularly as regards FrameNet. Indeed, the latest FrameNet release (v. 1.3) contains 10,195 lex- ical units (LUs), 3,380 of which are described only by a lexicographic definition without any example sentence. In order to cope with this lack of data, it would be useful to map frame information onto other lexical resources with a broader coverage. We be- lieve that WordNet (Fellbaum, 1998), with 210,000 entries in version 3.0, can represent a suitable re- source for this task. In fact, both FrameNet and WordNet group together semantically similar words, and provide a hierarchical representation of the lex- ical knowledge (in WordNet the relations between synsets, in FrameNet between frames, see Ruppen- hofer et al. (2006)). On the other hand, WordNet provides a more extensive coverage particularly for adjectives and nouns denoting artifacts and natural kinds, that are mostly neglected in FrameNet. In this paper, we present an approach using Sup- port Vector Machines (SVM) to map FrameNet …

Content truncated to 500 words