Pronoun resolution with Markov logic networks

Abstract

Pronoun resolution refers to the problem of determining the coreference linkages between the antecedents and the pronouns. We propose to employ a combined model of statistical learning and first-order logic, the Markov logic network (MLN). Our proposed model can more effectively characterize the pronoun coreference resolution process that requires conducting inference upon a variety of conditions. The influence of different types of constraints are also investigated. © 2008 Springer-Verlag Berlin Heidelberg.