Dijkstra-WSA: A Graph-Based Approach to Word Sense Alignment

Transactions of the Association for Computational Linguistics, 1 (2013) 151–164. Action Editor: Patrick Pantel. Submitted 12/2012; Revised 2/2013; Published 5/2013. c⃝2013 Association for Computational Linguistics. Dijkstra-WSA: A Graph-Based Approach to Word Sense Alignment Michael Matuschek ‡ and Iryna Gurevych †‡ † Ubiquitous Knowledge Processing Lab (UKP-DIPF), German Institute for Educational Research and Educational Information Schloßstr. 29, 60486 Frankfurt, Germany ‡Ubiquitous Knowledge Processing Lab (UKP-TUDA), Department of Computer Science, Technische Universit¨at Darmstadt Hochschulstr. 10, 64289 Darmstadt, Germany http://www.ukp.tu-darmstadt.de Abstract In this paper, we present Dijkstra-WSA, a novel graph-based algorithm for word sense alignment. We evaluate it on four different pairs of lexical-semantic resources with dif- ferent characteristics (WordNet-OmegaWiki, WordNet-Wiktionary, GermaNet-Wiktionary and WordNet-Wikipedia) and show that it achieves competitive performance on 3 out of 4 datasets. Dijkstra-WSA outperforms the state of the art on every dataset if it is com- bined with a back-off based on gloss similar- ity. We also demonstrate that Dijkstra-WSA is not only flexibly applicable to different re- sources but also highly parameterizable to op- timize for precision or recall. 1 Introduction Lexical-semantic resources (LSRs) are a corner- stone for many Natural Language Processing (NLP) applications such as word sense disambiguation (WSD) and information extraction. However, the growing demand for large-scale resources in dif- ferent languages is hard to meet. The Princeton WordNet (WN) (Fellbaum, 1998) is widely used for English, but for most languages corresponding re- sources are considerably smaller or missing. Col- laboratively constructed resources like Wiktionary (WKT) and OmegaWiki (OW) provide a viable op- tion for such cases and seem especially suitable for smaller languages (Matuschek et al., 2013), but there are still considerable gaps in coverage which need to be filled. A related problem is that there usually does not exist a single resource which works best for all purposes, as different LSRs cover differ- ent words, senses and information types. These considerations have sparked increasing re- search efforts in the area of word sense alignment (WSA). It has been shown that aligned resources can indeed lead to better performance than using the resources individually. Examples include seman- tic parsing using FrameNet (FN), WN, and VerbNet (VN) (Shi and Mihalcea, 2005), word sense disam- biguation using an alignment of WN and Wikipedia (WP) (Navigli and Ponzetto, 2012) and semantic role labeling using a combination of PropBank, VN and FN in the SemLink project (Palmer, 2009). Some of these approaches to WSA either rely heav- ily on manual labor (e.g. Shi and Mihalcea (2005)) or on information which is only present in few resources such as the most frequent sense (MFS) (Suchanek et al., 2008). This makes it difficult to apply them to a larger set of resources. In earlier work, we presented the large-scale re- source UBY (Gurevych et al., 2012). It contains nine resources in two languages which are mapped to a uniform representation using the LMF standard (Eckle-Kohler et al., 2012). They are thus struc- turally interoperable. UBY contains pairwise sense alignments between a subset of these resources, and this work also …

Content truncated to 500 words