JU: A Supervised Approach to Identify Semantic Relations from Paired Nominals

Proceedings of the 5th International Workshop on Semantic Evaluation, ACL 2010, pages 206–209, Uppsala, Sweden, 15-16 July 2010. c⃝2010 Association for Computational Linguistics JU: A Supervised Approach to Identify Semantic Relations from Paired Nominals Santanu Pal Partha Pakray Dipankar Das Sivaji Bandyopadhyay Department of Computer Science & Engineering, Jadavpur University, Kolkata, India santanupersonal1@gmail.com,parthapakray@gmail.com, dipankar.dipnil2005@gmail.com,sivaji_cse_ju@yahoo.com Abstract This article presents the experiments carried out at Jadavpur University as part of the participation in Multi-Way Classification of Semantic Relations between Pairs of Nomi- nals in the SemEval 2010 exercise. Separate rules for each type of the relations are iden- tified in the baseline model based on the verbs and prepositions present in the seg- ment between each pair of nominals. Inclu- sion of WordNet features associated with the paired nominals play an important role in distinguishing the relations from each other. The Conditional Random Field (CRF) based machine-learning framework is adopted for classifying the pair of nominals. Application of dependency relations, Named Entities (NE) and various types of WordNet features along with several com- binations of these features help to improve the performance of the system. Error analy- sis suggests that the performance can be im- proved by applying suitable strategies to differentiate each paired nominal in an al- ready identified relation. Evaluation result gives an overall macro-averaged F1 score of 52.16%. 1 Introduction Semantic Relations describe the relations between concepts or meanings that are crucial but hard to identify. The present shared task aims to develop the systems for automatically recognizing semantic relations between pairs of nominals. Nine relations such as Cause-Effect, Instrument-Agency, Product- Producer, Content-Container, Entity-Origin, En- tity-Destination, Component-Whole, Member- Collection and Message-Topic are given for Se- mEval-2010 Task #8 (Hendrix et al., 2010). The relation that does not belong to any of the nine re- lations is tagged as Other. The first five relations also featured in the previous SemEval-2007 Task #4. The present paper describes the approach of identifying semantic relations between pair of nominals. The baseline system is developed based on the verbs and prepositions present in the senten- tial segment between the two nominals. Some WordNet (Miller, 1990) features are also used in the baseline for extracting the relation specific at- tributes (e.g. Content type hypernym feature used for extracting the relation of Content-Container). The performance of the baseline system is limited due to the consideration of only the verb and preposition words in between the two nominals along with a small set of WordNet features. Hence, the Conditional Random Field (CRF) (McCallum et al., 2001) based framework is considered to ac- complish the present task. The incorporation of different lexical features (e.g. WordNet hyponyms, Common-parents, distance), Named Entities (NE) and syntactic features (direct or transitive depend- ency relations of parsing) has noticeably improved the performance of the system. It is observed that nominalization feature plays an effective role for identifying as well as distinguishing the relations. The test set containing 2717 sentences is evaluated against four different training sets. Some of the relations, e.g. Cause-Effect, Member-Collection …

Content truncated to 500 words