Evaluation of semantic role labeling based on lexical features using conditional random fields and support vector machine

The main objective of this paper is to identify the semantic roles of arguments in a sentence based on lexicalized features even if less semantic information is available. The semantic role labeling task (SRL) involves identifying which groups of words act as arguments to a given predicate. These arguments must be labeled with their role with respect to the predicate, indicating how the proposition should be semantically interpreted. The approach mainly focuses on improving the task of SRL by adding the similar words and selectional preferences to the existing lexical features, thereby avoiding data sparsity problem. Addition of richer lexical information can improve SRL task even when very little syntactic knowledge is available in the input sentence. We analyze the performance of SRL which use a probabilistic graphical model (Conditional Random Field) and a machine learning model (Support Vector Machines). The statistical modelling is trained by CONLL-2004 Shared Task training data.

I. Introduction

All NLP applications such as information extraction, question answering, summarization, speech recognition and part-of-speech tagging require some kind of semantic interpretation. Semantic role labeling is used to identify the relation of the verb with its participants. Semantic Role labeling (SRL) is performed in two steps, namely identification and classification. In the identification step, we find the role-bearing phrases, and in the classification step, we classify roles of the phrases. We have used Conditional Random fields and Support Vector Machines to perform SRL.