LT3: A Multi-modular Approach to Automatic Taxonomy Construction

Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 944–948, Denver, Colorado, June 4-5, 2015. c⃝2015 Association for Computational Linguistics LT3: A Multi-modular Approach to Automatic Taxonomy Construction Els Lefever LT3, Language and Translation Technology Team Department of Translation, Interpreting and Communication – Ghent University Groot-Brittanni¨elaan 45, 9000 Ghent, Belgium els.lefever@UGent.be Abstract This paper describes our contribution to the SemEval-2015 task 17 on “Taxonomy Extrac- tion Evaluation”. We propose a hypernym de- tection system combining three modules: a lexico-syntactic pattern matcher, a morpho- syntactic analyzer and a module retrieving hy- pernym relations from structured lexical re- sources. Our system ranked first in the compe- tition when considering the gold standard and manual evaluation, and second in the overall ranking. In addition, the experimental results show that all modules contribute to finding hy- pernym relations between terms. 1 Introduction Because of globalization and rapid technological evolution, it is no longer feasible to manually cre- ate and manage taxonomies for the large variety of scientific and technological (sub)domains. In ad- dition to domain-specific terminology, also com- panies desire to build their own mono- or bilin- gual taxonomies containing the relevant sector- and company-specific terminology. This clear need for automatisation has encouraged researchers to inves- tigate how terminological and semantically struc- tured resources such as taxonomies or ontologies can be automatically constructed from text (Bie- mann, 2005). Different approaches have been proposed to au- tomatically detect hierarchical relations between terms: pattern-based approaches (Hearst, 1992; Pan- tel and Ravichandran, 2004), statistical and machine learning techniques (Ritter et al., 2009), distribu- tional approaches (Caraballo, 1999; van der Plas and Bouma, 2005; Lenci and Benotto, 2012), morpho- syntactic approaches (Tjong Kim Sang et al., 2011) and word class latices (Navigli and Velardi, 2010). The SemEval-2015 “Taxonomy Extraction Eval- uation” Task (Bordea et al., 2015) is concerned with automatically finding relations between pairs of terms and organizing them in a hierarchical struc- ture. In this way, the task assumes that a list of do- main specific terms is already available in order to focus on the relation detection between these terms. To tackle this SemEval taxonomy learning task, we propose a multi-modular approach that combines lexico-syntactic, morphological and external struc- tured lexical information. We will describe our hy- pernym detection system in Section 2. The results of the evaluation are presented in Section 3, while Section 4 concludes this paper. 2 System Description Our hypernym detection system contains three main components: a lexico-syntactic pattern-based ap- proach, a morpho-syntactic analyzer and a module retrieving hypernym relations from a structured lex- ical resource. Each module takes as input a domain specific term list. 2.1 Pattern-based Approach The first module that automatically detects hyper- nym relations is a lexico-syntactic pattern-based ap- proach, based on Hearst (1992). These patterns are implemented as a list of regular expressions contain- ing lexicalized expressions (e.g. like), as well as iso- 944 lated Part-of-Speech tags (e.g. noun) and chunk tags, which represent different Part-of-Speech sequences (e.g. noun phrase (NP) = determiner + …

Content truncated to 500 words