Representational Interoperability of Linguistic and Collaborative Knowledge Bases

Creating a Natural Language Processing (NLP) application often requires to access lexical-semantic Knowledge Bases (KBs). Recently, Collaborative Knowledge Bases (CKBs) such as Wikipedia and Wiktionary1 have been recognized as promising lexicalsemantic KBs for NLP (Zesch et al., 2008b), complementing traditional Linguistic Knowledge Bases (LKBs). As CKBs differ significantly from LKBs concerning their content, structure and topological properties, the interoperability between CKBs and LKBs has become a major issue. To address this problem, we have developed a model of representational interoperability between LKBs and CKBs, which abstracts over the differences in their structures, and enables a uniform representation of their content in terms of entities and lexical-semantic relations between them. An entity consists of a set of lexeme–sense pairs along with a part-of-speech (PoS). The currently supported relations are the lexical relations synonymy and antonymy, as well as the semantic relations hypernymy, hyponymy, holonymy, meronymy and other, which covers any lexical-semantic relation other than the previously listed. NLP algorithms can thus be implemented in an one-time effort, as they only have to “know” about generalized entities and relations instead of being adapted to each KB individually. The KBs currently integrated are the LKBs WordNet (Fellbaum, 1998), GermaNet (Kunze, 2004), Cyc (Lenat and Guha, 1989), Roget’s Thesaurus (Jarmasz and Szpakowicz, 2003), Leipzig Annotation Project (Biemann, 2005), and the CKBs Wikipedia and Wiktionary, which are available for a large number of lan-