Anaphora Resolution in Hindi

Abstract

Extensive research works in the development of computational solutions to automate anaphora resolution in English, and many other languages have been carried out in the past few decades. Methods ranging from linguistics-oriented, rule based approaches to corpus oriented, machine-learning approaches have been employed for resolving anaphora. However, limited research is available for Hindi language. The present work addresses this gap. Both, rule-based and a machine-learning anaphora resolution approach are presented in this work. The primary motivation for this work comes because of the following: newlinea. English is a Subject-Verb-Object (SVO) language while Hindi is a morphologically newlinerich subject-object-verb (SOV) and relatively of free word-order language. Hindi also has a rich case marking system, although case marking is not obligatory. Historically the grammar for Hindi is assumed to have been derived from Sanskrit grammar developed by Panini who standardized the details of pronominal system for Sanskrit. Linguistic feature Pro-drop (zero pronoun) is important and needs to be handled in Hindi unlike in English. All these features (free word order, case inflection and zero pronoun) make Hindi a rich language from linguistic point of view. Various existing approaches developed for other languages are to be seen whether they are suitable for Hindi or need suitable adaptation and enhancement. newlineb. Although few independent studies for anaphora resolution in Hindi have been conducted in literature, their solutions have never been evaluated and compared with each other. newline