Reserating the awesometastic: An automatic extension of the WordNet taxonomy for novel terms

Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 1459–1465, Denver, Colorado, May 31 – June 5, 2015. c⃝2015 Association for Computational Linguistics Reserating the awesometastic: An automatic extension of the WordNet taxonomy for novel terms David Jurgens School of Computer Science McGill University jurgens@cs.mcgill.ca Mohammad Taher Pilehvar Department of Computer Science Sapienza University of Rome pilehvar@di.uniroma1.it Abstract This paper presents CROWN, an automatically con- structed extension of WordNet that augments its taxonomy with novel lemmas from Wiktionary. CROWN fills the important gap in WordNet’s lexi- con for slang, technical, and rare lemmas, and more than doubles its current size. In two evaluations, we demonstrate that the construction procedure is accu- rate and has a significant impact on a WordNet-based algorithm encountering novel lemmas. 1 Introduction Semantic knowledge bases are an essential, enabling component of many NLP applications. A notable exam- ple is WordNet (Fellbaum, 1998), which encodes a tax- onomy of concepts and semantic relations between them. As a result, WordNet has enabled a wide variety of NLP techniques such as Word Sense Disambiguation (Agirre et al., 2014), information retrieval (Varelas et al., 2005), semantic similarity (Pedersen et al., 2004; B¨ar et al., 2013), and sentiment analysis (Baccianella et al., 2010). However, semantic knowledge bases such as WordNet are expensive to produce; as a result, their scope and domain are often constrained by the resources available and may omit highly-specific concepts or lemmas, as well as new terminology that emerges after their construction. For ex- ample, WordNet does not contain the nouns “stepmom,” “broadband,” and “prequel.” Because of the coverage limitations of WordNet, sev- eral approaches have attempted to enrich WordNet with new relations and concepts. One group of approaches has enriched WordNet by aligning its structure with that of other resources such as Wikipedia or Wiktionary (Ruiz- Casado et al., 2005; Navigli and Ponzetto, 2012; Miller and Gurevych, 2014; Pilehvar and Navigli, 2014). How- ever, because these approaches identify corresponding lemmas with identical lexicalizations, they are often un- able to directly add novel lemmas to the existing taxo- nomic structure. The second group of approaches per- forms taxonomy induction to learn hypernymy relation- ships between words (Moro and Navigli, 2012; Meyer and Gurevych, 2012). However, these approaches often produce separate taxonomies from WordNet, which are also generally not readily accessible as resources. We introduce a new resource CROWN (Community- enRiched Open WordNet) that extends the existing WordNet taxonomy, more than doubling the existing number of synsets, and attaches these novel synsets to their appropriate hypernyms in WordNet. Novel sense data is extracted from Wiktionary, a large-scale collaboratively-constructed dictionary, and attached us- ing multiple heuristics. CROWN fills an important gap in WordNet’s limited coverage of both domain-specific lem- mas and slang terminology and idioms.1 In two experi- ments, we demonstrate that (1) our construction process accurately associates a novel sense with its correct hy- pernym and (2) the resulting resource has an immediate benefit for existing WordNet-based applications. Impor- tantly, CROWN v1.0 …

Content truncated to 500 words