A Novel Approach Based on Reinforcement Learning for Anaphora Resolution

Abstract:

This paper focuses on the anaphora resolution task in Arabic texts. The proposed approach includes the following steps: identifying the anaphora, removing the non-referential ones, identifying the lists of candidate antecedents and choosing the best of them for each processed anaphora. The last step can be regarded as a dynamic and probable process that consists of a sequence of decisions. Thus, it could be modeled by a Markov Decision Process (MDP). We propose a new approach based on reinforcement learning as it is an effective method for learning in such uncertain and stochastic environment and for resolving MDPs. When evaluating our approach, we have obtained encouraging results that can reach about 80% of accuracy.