Cognate Identification using Machine Translation

Cognate Identification using Machine Translation Shervin Malmasi Centre for Language Technology Macquarie University Sydney, NSW, Australia shervin.malmasi@mq.edu.au Mark Dras Centre for Language Technology Macquarie University Sydney, NSW, Australia mark.dras@mq.edu.au Abstract In this paper we describe an approach to automatic cognate identification in mono- lingual texts using machine translation. This system was used as our entry in the 2015 ALTA shared task, achieving an F1- score of 63% on the test set. Our pro- posed approach takes an input text in a source language and uses statistical ma- chine translation to create a word-aligned parallel text in the target language. A ro- bust measure of string distance, the Jaro- Winkler distance in this case, is then ap- plied to the pairs of aligned words to de- tect potential cognates. Further extensions to improve the method are also discussed. 1 Introduction Cognates are words in different languages that have similar forms and meanings, often due to a common linguistic origin from a shared ancestor language. Cognates play an important role in Second Lan- guage Acquisition (SLA), particularly between re- lated languages. However, although they can ac- celerate vocabulary acquisition, learners also have to be aware of false cognates and partial cognates. False cognates are similar words that have distinct, unrelated meanings. In other cases, there are par- tial cognates: similar words which have a common meaning only in some contexts. For example, the word police in French can translate to police, pol- icy or font, depending on the context. Cognates are a source of learner errors and the detection of their incorrect usage, coupled with correction and contextual feedback, can be of great use in computer-assisted language learning systems. Additionally, cognates are also useful for estimating the readability of a text for non-native readers. The identification of such cognates have also been tackled by researchers in NLP. English and French are one such pair that have received much attention, potentially because it has been posited that up to 30% of the French vocabulary consists of cognates (LeBlanc and S´eguin, 1996). This paper describes our approach to cognate identification in monolingual texts, relying on sta- tistical machine translation to create parallel texts. Using the English data from the shared task, the aim was to predict which words have French cog- nates. In §2 we describe some related work in this area, followed by a brief description of the data in §3. Our methodology is described in §4 and re- sults are presented in §5. 2 Related Work Much of the previous work in this area has re- lied on parallel corpora and aligned bilingual texts. Such approaches often rely on orthographic simi- larity between words to identify cognates. This similarity can be quantified using measures such as the edit distance or dice coefficient with n- grams. Brew and McKelvie (1996) applied such orthographic measures to extract English-French cognates from aligned texts. Phonetic similarity has also been shown to be useful for this task. Kondrak (2001), for example, proposed an approach that also incorporates …

Content truncated to 500 words