Using Cognates in a French-Romanian Lexical Alignment System: A Comparative Study

Proceedings of Recent Advances in Natural Language Processing, pages 247–253, Hissar, Bulgaria, 12-14 September 2011. Using Cognates in a French - Romanian Lexical Alignment System: A Comparative Study Mirabela Navlea Linguistique, Langues, Parole (LiLPa) Université de Strasbourg 22, rue René Descartes BP, 80010, 67084 Strasbourg cedex navlea@unistra.fr Amalia Todiraşcu Linguistique, Langues, Parole (LiLPa) Université de Strasbourg 22, rue René Descartes BP, 80010, 67084 Strasbourg cedex todiras@unistra.fr Abstract This paper describes a hybrid French - Roma- nian cognate identification module. This mod- ule is used by a lexical alignment system. Our cognate identification method uses lemma- tized, tagged and sentence-aligned parallel corpora. This method combines statistical techniques, linguistic information (lemmas, POS tags) and orthographic adjustments. We evaluate our cognate identification module and we compare it to other methods using pure sta- tistical techniques. Thus, we study the impact of the used linguistic information and the or- thographic adjustments on the results of the cognate identification module and on cognate alignment. Our method obtains the best results in comparison with the other implemented sta- tistical methods. 1 Introduction We present a new French - Romanian cognate identification module, integrated into a lexical alignment system using French - Romanian pa- rallel law corpora. We define cognates as translation equivalents having an identical form or sharing orthographic or phonetic similarities (common etymology, borrowings). Cognates are very frequent between close languages such as French and Romanian, two Latin languages with a rich morphology. So, they represent important lexical cues in a French - Romanian lexical alignment system. Few linguistic resources and tools for Romanian (dictionaries, parallel corpora, MT systems) are currently available. Some lexically aligned cor- pora or lexical alignment tools (Tufiş et al., 2005) are available for Romanian - English or Romanian - German (Vertan and Gavrilă, 2010). Most of the cognate identification modules used by these systems are purely statistical. As far as we know, no cognate identification method is available for French and Romanian. Cognate identification is a difficult task due to the high orthographic similarities between bilin- gual pairs of words having different meanings. Inkpen et al. (2005) develop classifiers for French and English cognates based on several dictionaries and manually built lists of cognates. Inkpen et al. (2005) distinguish between: - cognates (liste (FR) - list (EN)); - false friends (blesser (‘to injure’) (FR) - bless (EN)); - partial cognates (facteur (FR) - factor or mail- man (EN)); - genetic cognates (chef (FR) - head (EN)); - unrelated pairs of words (glace (FR) - ice (EN) and glace (FR) - chair (EN)). Our cognate detection method identifies cog- nates, partial and genetic cognates. This method is used especially to improve a French - Roma- nian lexical alignment system. So, we aim to ob- tain a high precision of our cognate identification method. Thus, we eliminate false friends and unrelated pairs of words combining statistical techniques and linguistic information (lemmas, POS tags). We use a lemmatized, tagged and sentence-aligned parallel corpus. Unlike Inkpen et al. (2005), we do not use other …

Content truncated to 500 words