JUNLP at SemEval-2016 Task 13: A Language Independent Approach for Hypernym Identification

Proceedings of SemEval-2016, pages 1310–1314, San Diego, California, June 16-17, 2016. c⃝2016 Association for Computational Linguistics JUNLP at SemEval-2016 Task 13: A Language Independent Approach for Hypernym Identification Promita Maitra and Dipankar Das Department of Computer Science Engineering, Jadavpur University Kolkata, India promita.maitra@gmail.com and dipankar.dipnil2005@gmail.com Abstract This paper describes our approach to build a language-independent hypernym extrac- tion system, based on two modules for the SemEval-2016 Task 13 on Taxonomy Extrac- tion Evaluation (TExEval-2). This task fo- cuses only on the hypernym-hyponym rela- tion extraction from a list of terms collected from various domains and languages. The first module of our system is built on the state- of-the-art system using BabelNet while the second one deals with the parts found within terms and which are useful to establish a hier- archical relation among them. Our system per- formed well in terms of recall in most of the domains irrespective of the languages; how- ever, the precision scores indicate a scope of improvement. In case of overall ranking, our present system stands fourth in monolingual (i.e. English) evaluation and second in multi- lingual (i.e. Dutch, Italian, French) setup. 1 Introduction The rapid growth in terms of digitalized texts in the recent years (specially, in the fields including scien- tific, clinical, enterprise, legal, and personal infor- mation management) has made the management of textual information increasingly important. In order to fulfill the need of having more structured data, on- tologies, taxonomy or hierarchical relations between ontological concepts are considered as useful tools for content organization, navigation, and retrieval, as well as to provide valuable input for semantically intensive tasks such as question answering and tex- tual entailment. This task specifically focuses on the identification of hypernym-hyponym relation among terms in four different languages (English, Dutch, Italian and French) and different domains. Typically, taxonomy construction has three basic steps: entity or concept identification, discover different relations among different entities and taxonomy construction. The challenge organizers have made it easy by al- ready providing us with a list of extracted entities (also called concepts/terms) for each of the domains in each of the languages. Our approach was to build a system that fits a multilingual setup and signifi- cantly reduces the computational time and complex- ity by taking existing resources into consideration. The rest of the paper is organized as follows: Section 2 describes the task at hand in brief along with the main challenges. Section 3 gives a brief overview of the work already done in this field. Sec- tion 4 contains the details of the modules we have used to address the problem. Section 5 presents the analysis of results and finally, the conclusion and fu- ture scopes are discussed in Section 6. 2 Problem Description SemEval-2016 Task 13: Taxonomy Extraction Eval- uation (TExEval-2) has its main focus on hypernym- hyponym relation extraction from given lists of terms collected from multiple domains like Food, Environment and Science (Bordea et al., 2016). This year, the task organizers have extended the problem setup to …

Content truncated to 500 words