Generalized Inference with Multiple Semantic Role Labeling Systems

摘要

We present an approach to semantic rolelabeling (SRL) that takes the output ofmultiple argument classifiers and combinesthem into a coherent predicateargumentoutput by solving an optimizationproblem. The optimization stage,which is solved via integer linear programming,takes into account both the recommendationof the classifiers and a setof problem specific constraints, and is thusused both to clean the classification resultsand to ensure structural integrity of the finalrole labeling. We illustrate a significantimprovement in overall SRL performancethrough this inference.