Building a BioWordNet Using WordNet Data Structures and WordNet’s Software Infrastructure–A Failure Story

Software Engineering, Testing, and Quality Assurance for Natural Language Processing, pages 31–39, Columbus, Ohio, USA, June 2008. c⃝2008 Association for Computational Linguistics Building a BIOWORDNET by Using WORDNET’s Data Formats and WORDNET’s Software Infrastructure — A Failure Story Michael Poprat Elena Beisswanger Jena University Language & Information Engineering (JULIE) Lab Friedrich-Schiller-Universit¨at Jena D-07743 Jena, Germany {poprat,beisswanger,hahn}@coling-uni-jena.de Udo Hahn Abstract In this paper, we describe our efforts to build on WORDNET resources, using WORDNET lexical data, the data format that it comes with and WORDNET’s software infrastructure in order to generate a biomedical extension of WORDNET, the BIOWORDNET. We began our efforts on the assumption that the soft- ware resources were stable and reliable. In the course of our work, it turned out that this belief was far too optimistic. We discuss the stumbling blocks that we encountered, point out an error in the WORDNET software with implications for research based on it, and con- clude that building on the legacy of WORD- NET data structures and its associated soft- ware might preclude sustainable extensions that go beyond the domain of general English. 1 Introduction WORDNET (Fellbaum, 1998) is one of the most au- thoritative lexical resources for the general English language. Due to its coverage – currently more than 150,000 lexical items – and its lexicological rich- ness in terms of definitions (glosses) and semantic relations, synonymy via synsets in particular, it has become a de facto standard for all sorts of research that rely on lexical content for the English language. Besides this perspective on rich lexicological data, over the years a software infrastructure has emerged around WORDNET that was equally ap- proved by the NLP community. This included, e.g., a lexicographic file generator, various editors and visualization tools but also meta tools rely- ing on properly formated WORDNET data such as a library of similarity measures (Pedersen et al., 2004). In numerous articles the usefulness of this data and software ensemble has been demonstrated (e.g., for word sense disambiguation (Patwardhan et al., 2003), the analysis of noun phrase conjuncts (Hogan, 2007), or the resolution of coreferences (Harabagiu et al., 2001)). In our research on information extraction and text mining within the field of biomedical NLP, we sim- ilarly recognized an urgent need for a lexical re- source comparable to WORDNET, both in scope and size. However, the direct usability of the original WORDNET for biomedical NLP is severely ham- pered by a (not so surprising) lack of coverage of the life sciences domain in the general-language English WORDNET as was clearly demonstrated by Burgun and Bodenreider (2001). Rather than building a BIOWORDNET by hand, as was done for the general-language English WORDNET, our idea to set up a WORDNET-style lexical resource for the life sciences was different. We wanted to link the original WORDNET with various biomedical terminological resources vastly available in the life sciences domain. As an obvious candidate for this merger, we chose one of the ma- jor high-coverage umbrella systems for biomedical ontologies, the OPEN …

Content truncated to 500 words