Resolving Pronouns for a Resource-Poor Language, Malayalam Using Resource-Rich Language, Tamil.

Proceedings of Recent Advances in Natural Language Processing, pages 611–618, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_072 611 Resolving Pronouns for a Resource-Poor Language, Malayalam Using Resource-Rich Language, Tamil Sobha Lalitha Devi AU-KBC Research Centre, Anna University, Chennai sobha@au-kbc.org Abstract In this paper we give in detail how a re- source rich language can be used for resolv- ing pronouns for a less resource language. The source language, which is resource rich language in this study, is Tamil and the re- source poor language is Malayalam, both belonging to the same language family, Dravidian. The Pronominal resolution de- veloped for Tamil uses CRFs. Our approach is to leverage the Tamil language model to test Malayalam data and the processing re- quired for Malayalam data is detailed. The similarity at the syntactic level between the languages is exploited in identifying the features for developing the Tamil language model. The word form or the lexical item is not considered as a feature for training the CRFs. Evaluation on Malayalam Wikipedia data shows that our approach is correct and the results, though not as good as Tamil, but comparable. 1 Introduction Natural language processing techniques of the present day require large amounts of manually- annotated data to work. In reality, the required quantity of data is available only for a few lan- guages of major interest. In this work we show how a resource-rich language, Tamil, can be lev- eraged to resolve anaphora for a related resource- poor language, Malayalam. Both Tamil and Mal- ayalam belong to the same language family, Dra- vidian. The methods we focus on exploits the sim- ilarity at the syntactic level of the languages and anaphora resolution heavily depends on syntactic features. If the resources available in one language (henceforth referred to as source) can be used to facilitate the resolution, such as anaphora, for all the languages related to the language in question (target), the problem of unavailability of resources would be alleviated. There exists a recent research paradigm, in which the researchers work on algorithms that can rap- idly develop machine translation and other tools for an obscure language. This work falls into this paradigm, under the assumption that the language in question has a less obscure sibling. Moreover, the problem is intellectually interesting. While there has been significant research in using re- sources from another language to build, for exam- ple, parsers, there have been very little work on utilizing the close relationship between the lan- guages to produce high quality tools such as anaphora resolution (Nakov, P and Tou Ng,H 2012). In our work we are interested in the follow- ing questions: If two languages are from the same language fam- ily and have similarity in syntactic structures and not in lexical form and script 1. Can the language model developed for one language be used for analyzing the other language? 2. How the lexical form difference can be resolved in using the language model? 3. How to overcome the challenges of script variation? In …

Content truncated to 500 words