Semantic Inference for Anaphora Resolution: Toward a Framework in Machine Translation

Anaphora is a discourse-level linguistic phenomenon. There is consensus that anaphora resolution should rely on prior sentences within the context of the discourse. We propose to cast anaphora resolution as a semantic inference process in which a combination of multiple strategies, each exploiting different aspects of linguistic knowledge, is employed to provide a coherent resolution of anaphora. A framework which encompasses several salient linguistic parameters such as grammatical role, proximity, repetition, sentence recency and semantic cues is demonstrated. This work also shows how an anaphora-resolution algorithm can be embedded within a framework which captures all the above salient parameters, as well as remedies some of the inadequacies found in any monolithic resolution system. A language-neutral semantic representation characterized by semantic cues is presented in order to capture the distilled information after resolution. The effectiveness of the language-neutral representation, both for machine translation and anaphora resolution, is demonstrated through a set of simulations and evaluations.