Generative and Discriminative Learning in Semantic Role Labeling for Italian

摘要

In this paper, we present a Semantic Role Labeling tool for Italian language for the FLaIT competition at Evalita 2011. This tool adopts a double level architecture, based on a discriminative and a generative approach, to resolve the different sub-tasks that composed the SRL task. We apply a discriminative model for the boundary detection task based on lexical and syntactical features. A distributional approach to modeling lexical semantic information, instead, for the Argument Classification sub-task is applied in a semi-supervised perspective. Few labeled examples are generalized through a semantic similarity model automatically acquired from large corpora. The combination of these models achieved interesting results in the FLaIT competition.