Title: Parallel corpora and WordSpace models: using a third language as an interlingua to enrich multilingual resources
Authors: Adil Toumouh; Dominic Widdows; Ahmed Lehireche
Addresses: Computer Science Department, Djillali Liabes University, Sidi Bel Abbes, 22000, Algeria. ' Google Pittsburgh, 6425 Penn Avenue, Pittsburgh, PA 15207, USA. ' Computer Science Department, Djillali Liabes University, Sidi Bel Abbes, 22000, Algeria
Abstract: In this paper we describe an automatic method for enriching multilingual resources by using WordSpace models and parallel corpora. This paper demonstrates that it is possible to enhance translation resources for a pair of language without direct parallel corpora, using instead a third language for which parallel corpora are available for both the target languages. First, we demonstrate the effectiveness of the proposed bilingual model, and then demonstrate how can this model been used for overcoming the handicap due to the absence of the resources like parallel dictionaries and corpora for less widespread languages by proposing an interlingua model. Our experiments show that precision for translation of word pairs through an interlingua model reaches ?85%, compared to ?90% for a direct bilingual model.
Keywords: WordSpace models; parallel corpora; random projection; multilingual ontology; synonymy; translation; word pairs.
DOI: 10.1504/IJICT.2011.043626
International Journal of Information and Communication Technology, 2011 Vol.3 No.4, pp.299 - 313
Published online: 21 Oct 2014 *
Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article