Towards Building Semantic Role Labeler for Indian Languages

Towards Building Semantic Role Labeler for Indian Languages Maaz Anwar Nomani♣ and Dipti Misra Sharma♣ FC Kohli Center on Intelligent Systems (KCIS), IIIT-Hyderabad, India♣ maaz.anwar@research.iiit.ac.in, dipti@iiit.ac.in, Abstract We present a statistical system for identifying the semantic relationships or semantic roles for two major Indian Languages, Hindi and Urdu. Given an input sentence and a predicate/verb, the system first identifies the arguments pertaining to that verb and then classifies it into one of the semantic labels which can either be a DOER, THEME, LOCATIVE, CAUSE, PURPOSE etc. The system is based on 2 statistical classifiers trained on roughly 130,000 words for Urdu and 100,000 words for Hindi that were hand-annotated with semantic roles under the PropBank project for these two languages. Our system achieves an accuracy of 86% in identifying the arguments of a verb for Hindi and 75% for Urdu. At the subsequent task of classifying the constituents into their semantic roles, the Hindi system achieved 58% precision and 42% recall whereas Urdu system performed better and achieved 83% precision and 80% recall. Our study also allowed us to compare the usefulness of different linguistic features and feature combinations in the semantic role labeling task. We also examine the use of statistical syntactic parsing as feature in the role labeling task. Keywords: PropBank, Semantic Role labeling, Treebank, Dependency relations 1. Introduction We introduce a statistical semantic role labeler for Hindi and Urdu, two major Indian languages. A Semantic Role labeler (henceforth, SRL) automatically marks the argu- ments/valency of a predicate in a sentence. The proposed system is based on supervised machine learning approach on Hindi and Urdu PropBanks which are being built for these languages. The approach is a 2-stage architecture in which, first the arguments pertaining to a predicate in a sentence are iden- tified by the system and then those identified arguments are classified into one of the PropBank semantic labels. Our system uses a basic Logistic Regression machine learn- ing algorithm (Pedregosa et al., 2011) for identifying the predicates and Support Vector Machines (Pedregosa et al., 2011) to classify the arguments of a predicate into seman- tic labels. We have used 10 linguistic features, 6 of which have been derived from literature and 4 are novel features. Among the new features, we have used the karaka rela- tions/dependency relations (described later) which turns out to be the most discriminative feature among all. We have also experimented with the automatic parses which are used as features and derived from state-of-art Hindi and Urdu parsers. To the best of our knowledge, this is the first such attempt of building a semantic role labeler for any In- dian language. This paper is arranged as follows: Section 2 gives a brief introduction about Semantic Parsing. In Section 3, we give the data statistics of the language resources used. Section 4 talks about related work in semantic role labeling. In sec- tion 5, we briefly describe the Hindi and Urdu Propbanks and Treebanks which are being built for these two Indian languages. …

Content truncated to 500 words