A Dictionary-Based Approach for Evaluating Orthographic Methods in Cognates Identification

Proceedings of Recent Advances in Natural Language Processing, pages 141–147, Hissar, Bulgaria, 7-13 September 2013. A Dictionary-Based Approach for Evaluating Orthographic Methods in Cognates Identification Alina Maria Ciobanu Faculty of Mathematics and Computer Science University of Bucharest alina.ciobanu@my.fmi.unibuc.ro Liviu P. Dinu Faculty of Mathematics and Computer Science Center for Computational Linguistics University of Bucharest ldinu@fmi.unibuc.ro Abstract In this paper we propose a method for identifying cognates based on etymology and etymons. We employ this approach to evaluate the extent to which lexical simi- larity can be used for automatic detection of cognate pairs. We investigate some or- thographic approaches widely used in this research area and some original metrics as well. We apply this procedure for Ro- manian and its most closely related lan- guages, French and Italian, but our method is applicable to any languages. 1 Introduction and Related Work Cognates are words in different languages having the same etymology and a common ancestor. The task of cognates identification is widely used in historical and comparative linguistics, in the study of languages relatedness (Chin et al, 2010), phy- logenetic inference (Atkinson et al, 2005) and in identifying how and to what extent languages changed over time. Besides these research a- reas, in which the genetic relationships between words are extremely relevant, cognates have been successfully used in other fields, such as lan- guage acquisition, bilingual word recognition (Di- jkstra et al, 2012), corpus linguistics (Simard et al, 1992), cross-lingual information retrieval (Buck- ley et al, 1998) and machine translation (Knight et al, 2003). In these domains, the term “cognates” is usually used with a somewhat different meaning, denoting words with high orthographic/phonetic and cross-lingual meaning similarity, the condi- tion of common etymology being left aside. Kon- drak (2001) makes the distinction between the dif- ferent interpretations of the notion and Inkpen et al (2005) present the definition of “genetic cog- nates”. In this paper we focus on genetic relationships between words and we use the term “cognates” in a broader meaning, counting as cognates the word-etymon pairs as well. Our motivation is that these pairs of words also share a common ances- tor, thus complying with the cognates’ definition. For example, the Romanian word campion (mea- ning champion) has Italian etymology and the ety- mon campione, which has Latin etymology and the etymon campione(m). Thus, the Romanian word campion and the Italian word campione are cognates, as they share a common Latin ancestor. The paper is organized as follows: we intro- duce our approach to cognates identification in Section 2. We describe the corpus used for our research in Section 3. We present several ortho- graphic approaches used for cognates identifica- tion in Section 4. We evaluate these metrics and analyse the results of our experiments in Section 5. Finally, we draw some conclusion regarding our research in Section 6. 2 Our Approach We focus on the Romanian language and we inves- tigate its cognate pairs with two other Romance languages, French and Italian. We believe this comparison is …

Content truncated to 500 words