Cross-lingual Synonymy Overlap

Proceedings of Recent Advances in Natural Language Processing, pages 147–152, Hissar, Bulgaria, Sep 7–9 2015. Cross-lingual Synonymy Overlap Anca Dinu1, Liviu P. Dinu2, Ana Sabina Uban2 1Faculty of Foreign Languages and Literatures, University of Bucharest 2Faculty of Mathematics and Computer Science, University of Bucharest anca_d_dinu@yahoo.com, liviu.p.dinu@gmail.com, ana.uban@gmail.com Abstract We investigate in this paper the degree of overlap between synonym sets of trans- lated word pairs across three languages: French, English and Romanian. We use for this purpose a French Synonym Dic- tionary, a Romanian Synonym Dictionary, Princeton’s WordNet and Google Trans- late API. We build a database contain- ing pairs of (translated) words from the three languages, along with their corre- sponding synonym sets. We use it in or- der to gain insight into the synonym over- lap for each language pair, and thus, into their degree of common concept lexical- ization, by various queries. While the overall percentage of common synonyms is (expectedly) quite small (averaging ~6% across all language pairs), the percentage of hard synonyms pairs (pairs that have at least one common synonym), reaching ~62%, is significant. This is encourag- ing for further use of this special kind of word translated pairs in tasks such as au- tomatic enhancement of lexical databases (such as WordNet) for less resourced lan- guages such as Romanian, based on cor- responding English versions of these lex- ical databases. Another interesting query topic was obtaining distributions of hard synonym pairs, function of their part of speech: hard synonyms were most fre- quent among verbs for English, and among adjectives for Romanian and French. Keywords: cross-lingual synonyms, French, Romanian, database 1 Introduction We investigate in this paper the degree of overlap between synonym sets of translated word pairs in three different languages, namely French, English and Romanian. The main idea is to test whether the synonym sets of pairs of translated words are still semantically related, that is to measure the de- gree of synonym overlap. Synonymy is a lexical semantic relation, that is, a relation between meanings of words. By def- inition, synonyms are ‘words or expressions of the same language that have the same or nearly the same meaning in some or all senses’ (Inc., 2004). Cross-linguistically, the question that we try to answer in this paper is how much of this common meaning is shared by pairs of translated words. Since synonymy closely associates differ- ent lexicalizations of the same concept (which is language-specific), the overlap between synonym sets across a pair of languages expresses a kind of concept lexicalization overlap. Cross-lingual synonym sets prove to be useful in tasks such as, for instance, automatic translation of web pages. Since search engines are using more of the Latent Semantic Indexing, which associates keywords of an article or a page with its synonyms within the domain covered by the keywords, one needs to take into consideration the synonym set of the translated keywords and the overlap of two languages synonym sets. 2 Related Works There are various NLP applications using syn- …

Content truncated to 500 words