Unsupervised Extraction of False Friends from Parallel Bi-Texts Using the Web as a Corpus

International Conference RANLP 2009 - Borovets, Bulgaria, pages 292–298 Unsupervised Extraction of False Friends from Parallel Bi-Texts Using the Web as a Corpus Svetlin Nakov and Preslav Nakov∗ Department of Mathematics and Informatics Sofia University “St Kliment Ohridski” 5, James Bourchier Blvd.,1164 Sofia, Bulgaria {svetlin.nakov, preslav.nakov}@fmi.uni-sofia.bg Elena Paskaleva Linguistic Modeling Department Institute for Parallel Processing Bulgarian Academy of Sciences 25A, Acad. G. Bonchev St., 1113 Sofia, Bulgaria hellen@lml.bas.bg Abstract False friends are pairs of words in two languages that are perceived as similar, but have different meanings, e.g., Gift in German means poison in English. In this paper, we present several un- supervised algorithms for acquiring such pairs from a sentence-aligned bi-text. First, we try dif- ferent ways of exploiting simple statistics about monolingual word occurrences and cross-lingual word co-occurrences in the bi-text. Second, using methods from statistical machine translation, we induce word alignments in an unsupervised way, from which we estimate lexical translation prob- abilities, which we use to measure cross-lingual semantic similarity. Third, we experiment with a semantic similarity measure that uses the Web as a corpus to extract local contexts from text snippets returned by a search engine, and a bi- lingual glossary of known word translation pairs, used as “bridges”. Finally, all measures are com- bined and applied to the task of identifying likely false friends. The evaluation for Russian and Bulgarian shows a significant improvement over previously-proposed algorithms. Keywords Cognates, false friends, cross-lingual semantic similarity, Web as a corpus, statistical machine translation. 1 Introduction Words in two languages that are orthographically and/or phonetically similar are often perceived as mu- tual translations, which could be wrong in some con- texts. Such words are known as cognates1 when they ∗Also: Department of Computer Science, National Uni- versity of Singapore, 13 Computing Drive, Singapore 117417, nakov@comp.nus.edu.sg 1 We should note that linguists define cognates as words de- rived from a common root regardless of whether they differ in meaning or not. For example, the Electronic Glossary of Linguistic Terms gives the following definition [3]: are mutual translations in all contexts, partial cognates when they are mutual translations in some contexts but not in other, and false friends when they are not mutual translations in any context. For example, the Bulgarian sl�nce (slynce) and the Russian solnce (solnce) are cognates, both mean- ing sun. However, the Bulgarian sin (sin) and the Russian syn (syn) are only partial cognates: while they can both mean son in some contexts, the Bul- garian word can also mean blue in other. Finally, the Bulgarian bistrota (bistrota) and the Russian bystrota (bystrota) are false friends meaning clear- ness and quickness, respectively. False friends are important not only for foreign language learning, but also for various natural lan- guage processing (nlp) tasks such as statistical ma- chine translation, word alignment, automated transla- tion quality control, etc. In this paper, we propose an unsupervised approach to the task of extracting pairs of false friends from a sentence-aligned corpus (a bi-text). While we experi- ment with Bulgarian …

Content truncated to 500 words