A VSM-based Statistical Model for the Semantic Relation Interpretation of Noun-Modifier Pairs

Proceedings of Recent Advances in Natural Language Processing, pages 636–645, Hissar, Bulgaria, Sep 7–9 2015. A VSM-based Statistical Model for the Semantic Relation Interpretation of Noun-Modifier Pairs Nitesh Surtani IIIT Hyderabad nitesh.surtani0606@gmail.com Soma Paul IIIT Hyderabad soma@iiit.ac.in Abstract The paper addresses the task of automatic in- terpretation of semantic relation in noun com- pounds. The problem has been attempted with both Ontology-based and Statistical ap- proaches, but both approaches having their own limitations. We present a novel VSM- based statistical model which represents each relation with a weighted vector of preposi- tional and verbal paraphrases. The model ranks the paraphrases on their relevance and assigns higher weights to more relevant para- phrases. The performance of the model is compared with the Ontology model and the results are quite encouraging. We finally pro- pose a Hybrid of the two models which com- pares on par with the best performing systems on Nastase and Szpakowicz (2003) dataset. 1 Introduction There has been an increased interest in discovering the semantics of Noun Compounds (NCs1). There are two reasons that make this task quite essential and interesting in text understanding: (i) their im- plicit nature, for instance the NC ‘monday meet- ing’ is the meeting scheduled on monday (Tem- poral), ‘teacher meeting’ is the meeting organized for teachers (Participant) and ‘NLP meeting’ is the meeting to discuss NLP topics (Quality-Topic); and (ii) their frequent and compounding behavior. NCs are very frequent in english and comprise of 3.9% and 2.6% of all tokens in the Reuters corpus and the British National Corpus (BNC) respectively (Baldwin and Tanaka, 2004). New NCs are very fre- quently constructed eg. website design, internet us- age, orange juice etc., and sometimes combine with other words to form longer compounds, e.g., orange juice company, orange juice company homepage etc. 1A noun compound (NC) is a sequence of nouns which act as a single noun (Downing, 1977), eg. sunday morning The frequency spectrum of NCs follows a Zipfian distribution (Séaghdha, 2008), where many NC to- kens belong to a long tail of low-frequency types. Over half of the two-type compounds in BNC occur just once (Kim and Baldwin, 2006). The research focusing on the semantic interpreta- tion of NCs has followed two directions: (i) Identi- fying the underlying semantic relation (Girju et al., 2005; Tratz and Hovy, 2010); and (ii) Paraphrasing the NC (Nakov, 2008; Butnariu and Veale, 2008; Butnariu et al., 2010). Consider the text: “A large student protest was carried out during monday evening by various engi- neering colleges to raise funds for research. This London protest saw tremendous par- ticipation by students from 14 colleges, see- ing to which R&D dept. agreed to increase the college funds to 10,000,000 GBP. ” The sequences marked in bold in the above ex- ample are Noun compounds (NCs). In the above text, some NCs are interpretable via paraphrasing: protest was carried out during evening, where ‘dur- ing’ defines the temporality of the protest. On the other hand, some NCs are not explicit: …

Content truncated to 500 words