Exploiting Multiple Sources for Open-Domain Hypernym Discovery

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1224–1234, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Exploiting Multiple Sources for Open-domain Hypernym Discovery Ruiji Fu, Bing Qin, Ting Liu∗ Research Center for Social Computing and Information Retrieval School of Computer Science and Technology Harbin Institute of Technology, China {rjfu, bqin, tliu}@ir.hit.edu.cn Abstract Hypernym discovery aims to extract such noun pairs that one noun is a hypernym of the other. Most previous methods are based on lexical patterns but perform badly on open- domain data. Other work extracts hypernym relations from encyclopedias but has limited coverage. This paper proposes a simple yet ef- fective distant supervision framework for Chi- nese open-domain hypernym discovery. Giv- en an entity name, we try to discover its hy- pernyms by leveraging knowledge from mul- tiple sources, i.e., search engine results, ency- clopedias, and morphology of the entity name. First, we extract candidate hypernyms from the above sources. Then, we apply a statistical ranking model to select correct hypernyms. A set of novel features is proposed for the rank- ing model. We also present a heuristic strate- gy to build a large-scale noisy training data for the model without human annotation. Exper- imental results demonstrate that our approach outperforms the state-of-the-art methods on a manually labeled test dataset. 1 Introduction Hypernym discovery is a task to extract such noun pairs that one noun is a hypernym of the other (S- now et al., 2005). A noun H is a hypernym of an- other noun E if E is an instance or subclass of H. In other word, H is a semantic class of E. For instance, “actor” is a hypernym of “Mel Gibson”; “dog” is a hypernym of “Caucasian sheepdog”; “medicine” is a hypernym of “Aspirin”. Hypernym discovery is an important subtask of semantic relation extraction ∗Email correspondence. and has many applications in ontology construction (Suchanek et al., 2008), machine reading (Etzion- i et al., 2006), question answering (McNamee et al., 2008), and so on. Some manually constructed thesauri such as WordNet can also provide some semantic relations such as hypernyms. However, these thesauri are lim- ited in its scope and domain, and manual construc- tion is knowledge-intensive and time-consuming. Therefore, many researchers try to automatically ex- tract semantic relations or to construct taxonomies. Most previous methods on automatic hypernym discovery are based on lexical patterns and suffer from the problem that such patterns can only cov- er a small part of complex linguistic circumstances (Hearst, 1992; Turney et al., 2003; Zhang et al., 2011). Other work tries to extract hypernym rela- tions from large-scale encyclopedias like Wikipedia and achieves high precision (Suchanek et al., 2008; Hoffart et al., 2012). However, the coverage is limit- ed since there exist many infrequent and new entities that are missing in encyclopedias (Lin et al., 2012). We made similar observation that more than a half of entities in our data set have no entries in the en- cyclopedias. This paper …

Content truncated to 500 words