A Computational Model for Resolving Arabic Anaphora using Linguistic Criteria

Abstract

Anaphora resolution is seen to be a very challenging and complex problem in the NLP. A majority of the NLP applications used for question answering, information extraction, and text summarisation, need a proper resolution and identification of the anaphora. Despite the fact that several authors have published studies for anaphora resolution in many European languages, including English, very few studies have been published for anaphora resolution in the Arabic language. In our study, we have proposed a novel model for the Arabic pronominal anaphora resolution. Our model contains several steps. In the first step, we have identified the pronouns and removed the non-anaphoric pronouns. In the second step, we have identified a list of the candidates from the context around the anaphora. Lastly, we selected the most probable candidates for every identified anaphoric pronoun. In our study, we have determined the proper rules which can be used for this task. The different linguistic rules depend on the morphological, lexical, heuristic, syntactic, and the positional constraints. We have assessed the performance of our proposed model using the Quran corpus, which was annotated with the pronominal anaphora. Our experimental results indicated that our proposed model was able to yield good results and could also choose the appropriate antecedents with 84.43% accuracy.