Inferring semantic roles using sub-categorization frames andmaximum entropy model

摘要

In this paper, we propose an approachfor inferring semantic role using subcategorizationframes and maximum entropy model. Our approach aims touse the sub-categorization informationof the verb to label the mandatory argumentsof the verb in various possibleways. The ambiguity between theassignment of roles to mandatory argumentsis resolved using the maximumentropy model. The unlabelled mandatoryarguments and the optional argumentsare labelled directly using themaximum entropy model such that theirlabels are not one among the frame elementsof the sub-categorization frameused. Maximum entropy model is preferredbecause of its novel approachof smoothing. Using this approach,we obtained an F-measure of 68.14%on the development set of the dataprovided for the CONLL-2005 sharedtask. We show that this approach performswell in comparison to an approachwhich uses only the maximumentropy model.