Building Specialized Bilingual Lexicons Using Large Scale Background Knowledge

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 479–489, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Building Specialized Bilingual Lexicons Using Large-Scale Background Knowledge Dhouha Bouamor1, Adrian Popescu1, Nasredine Semmar1, Pierre Zweigenbaum2 1 CEA, LIST, Vision and Content Engineering Laboratory, 91191 Gif-sur-Yvette CEDEX, France; firstname.lastname@cea.fr 2LIMSI-CNRS, F-91403 Orsay CEDEX, France; pz@limsi.fr Abstract Bilingual lexicons are central components of machine translation and cross-lingual infor- mation retrieval systems. Their manual con- struction requires strong expertise in both lan- guages involved and is a costly process. Sev- eral automatic methods were proposed as an alternative but they often rely on resources available in a limited number of languages and their performances are still far behind the quality of manual translations. We intro- duce a novel approach to the creation of spe- cific domain bilingual lexicon that relies on Wikipedia. This massively multilingual en- cyclopedia makes it possible to create lexi- cons for a large number of language pairs. Wikipedia is used to extract domains in each language, to link domains between languages and to create generic translation dictionaries. The approach is tested on four specialized do- mains and is compared to three state of the art approaches using two language pairs: French- English and Romanian-English. The newly in- troduced method compares favorably to exist- ing methods in all configurations tested. 1 Introduction The plethora of textual information shared on the Web is strongly multilingual and users’ information needs often go well beyond their knowledge of for- eign languages. In such cases, efficient machine translation and cross-lingual information retrieval systems are needed. Machine translation already has a decades long history and an array of commercial systems were already deployed, including Google Translate 1 and Systran 2. However, due to the intrin- sic difficulty of the task, a number of related prob- lems remain open, including: the gap between text semantics and statistically derived translations, the scarcity of resources in a large majority of languages and the quality of automatically obtained resources and translations. While the first challenge is general and inherent to any automatic approach, the second and the third can be at least partially addressed by an appropriate exploitation of multilingual resources that are increasingly available on the Web. In this paper we focus on the automatic creation of domain-specific bilingual lexicons. Such resources play a vital role in Natural Language Processing (NLP) applications that involve different languages. At first, research on lexical extraction has relied on the use of parallel corpora (Och and Ney, 2003). The scarcity of such corpora, in particular for spe- cialized domains and for language pairs not involv- ing English, pushed researchers to investigate the use of comparable corpora (Fung, 1998; Chiao and Zweigenbaum, 2003). These corpora include texts which are not exact translation of each other but share common features such as domain, genre, sam- pling period, etc. The basic intuition that underlies bilingual lexi- con creation is the distributional hypothesis (Harris, 1954) which puts that words with …

Content truncated to 500 words