Wikipedia as Frame Information Repository

Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 276–285, Singapore, 6-7 August 2009. c⃝2009 ACL and AFNLP Wikipedia as Frame Information Repository Sara Tonelli FBK-irst I-38100, Trento, Italy satonelli@fbk.eu Claudio Giuliano FBK-irst I-38100, Trento, Italy giuliano@fbk.eu Abstract In this paper, we address the issue of au- tomatic extending lexical resources by ex- ploiting existing knowledge repositories. In particular, we deal with the new task of linking FrameNet and Wikipedia us- ing a word sense disambiguation system that, for a given pair frame – lexical unit (F, l), finds the Wikipage that best ex- presses the the meaning of l. The mapping can be exploited to straightforwardly ac- quire new example sentences and new lex- ical units, both for English and for all lan- guages available in Wikipedia. In this way, it is possible to easily acquire good-quality data as a starting point for the creation of FrameNet in new languages. The evalua- tion reported both for the monolingual and the multilingual expansion of FrameNet shows that the approach is promising. 1 Introduction Many applications in the context of natural lan- guage processing or information retrieval have proved to convey significant improvement by ex- ploiting lexical databases with high-quality anno- tation such as FrameNet (Fillmore et al., 2003) and WordNet (Fellbaum, 1998). Nevertheless, the practical use of similar resources is often biased by their limited coverage because manual anno- tation is time-consuming and requires a relevant financial effort. For this reason, some research ac- tivities have focused on the automatic enrichment of such resources with annotated information in (near) manual quality. The main strategy proposed was the mapping between resources in order to reciprocally enrich different lexical databases by linking their information layers. This has proved to be useful in several tasks, from verb classifica- tion (Chow and Webster, 2007) to semantic role labeling (Giuglea and Moschitti, 2006), open text semantic parsing (Shi and Mihalcea, 2004) and textual entailment (Burchardt and Frank, 2006). In this work, we focus on the automatic enrich- ment of the FrameNet database for English and we propose a new framework to extend this procedure to new languages. While similar works in the past have mainly proposed to automatically extend the FrameNet database by mapping frames and Word- Net synsets (Shi and Mihalcea (2005), Johans- son and Nugues (2007), and Tonelli and Pighin (2009)), we present an explorative approach that for the first time exploits Wikipedia to this pur- pose. In particular, given a lexical unit l belong- ing to a frame F, we devise a strategy to link l to the Wikipedia article that best captures the sense of l in F. This is basically a word disam- biguation (WSD) problem (Erk, 2004) and to this purpose we employ a state-of-the-art WSD sys- tem (Gliozzo et al., 2005). The mapping between (F, l) pairs and Wikipedia pages could then be ex- ploited for three further subtasks: (a) automati- cally extract from Wikipedia all sentences point- ing to the Wikipage mapped with …

Content truncated to 500 words