A Richer-but-Smarter Shortest Dependency Path with Attentive Augmentation for Relation Extraction

Proceedings of NAACL-HLT 2019, pages 2902–2912 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 2902 A Richer-but-Smarter Shortest Dependency Path with Attentive Augmentation for Relation Extraction Duy-Cat Can1, Hoang-Quynh Le1∗, Quang-Thuy Ha1 and Nigel Collier2 1Faculty of Information Technology, VNU University of Engineering and Technology, Hanoi, Vietnam 2Department of Theoretical and Applied Linguistics, University of Cambridge, UK {catcd,lhquynh,thuyhq}@vnu.edu.vn, nhc30@cam.ac.uk Abstract To extract the relationship between two enti- ties in a sentence, two common approaches are (1) using their shortest dependency path (SDP) and (2) using an attention model to cap- ture a context-based representation of the sen- tence. Each approach suffers from its own dis- advantage of either missing or redundant in- formation. In this work, we propose a novel model that combines the advantages of these two approaches. This is based on the basic in- formation in the SDP enhanced with informa- tion selected by several attention mechanisms with kernel filters, namely RbSP (Richer-but- Smarter SDP). To exploit the representation behind the RbSP structure effectively, we de- velop a combined deep neural model with a LSTM network on word sequences and a CNN on RbSP. Experimental results on the SemEval-2010 dataset demonstrate improved performance over competitive baselines. The data and source code are available at https: //github.com/catcd/RbSP. 1 Introduction One of the most fundamental tasks in natural lan- guage processing, as well as in information extrac- tion, is Relation Extraction (RE), i.e., determining the semantic relation between pairs of named enti- ties or nominals in a sentence or a paragraph. Take the following sentences from the SemEval-2010 task 8 dataset (Hendrickx et al., 2009) as exam- ples: (i) We put the soured [cream]e1 in the butter [churn]e2 and started stirring it. (ii) The agitating [students]e1 also put up a [barricade]e2 on the Dhaka-Mymensingh highway. Here the nominals ‘cream’ and ‘churn’ in sentence (i) are of relation ∗∗Corresponding author Entity-Destination(e1,e2) while nom- inals ‘students’ and ‘barricade’ in sentence (ii) are of relations Product-Producer(e2,e1). The research history of RE has witnessed the development as well as the competition of a vari- ety of RE methodologies. All of them are proven to be effective and have different strengths by leveraging different types of linguistic knowledge, however, also suffer from their own limitations. Some early studies stated that the shortest de- pendency path (SDP) in dependency tree is usu- ally concise and contains essential information for RE (Bunescu and Mooney, 2005; Fundel et al., 2006). By 2016, this approach became dominant with many studies demonstrating that using SDP brings better experimental results than previous approaches that used the whole sentence (Xu et al., 2015a,b; Mehryary et al., 2016; Cai et al., 2016; Le et al., 2018). However, using the SDP may lead to the omission of useful information (i.e., negation, adverbs, prepositions, etc.). Recogniz- ing this disadvantage, some studies have sought to improve SDP approaches, such as adding the in- formation from the sub-tree attached to each node in the SDP (Liu et al., 2015) or applying a …

Content truncated to 500 words