Learning Co-relations of Plausible Verb Arguments with a WSM and a Distributional Thesaurus

摘要

We propose a model based on the Word Space Model for calculating the plausibility of candidate arguments given one verb and one argument. The resulting information can be used in co-reference resolution, zero-pronoun resolution or syntactic ambiguity tasks. Previous work such as Selectional Preferences or Semantic Frames acquisition focuses on this task using supervised resources, or predicting arguments independently from each other. On this work we explore the extraction of plausible arguments considering their co-relation, and using no more information than that provided by the dependency parser. This creates a data sparseness problem alleviated by using a distributional thesaurus built from the same data for smoothing. We compare our model with the traditional PLSI method.