A Step toward Compositional Semantics: E-HowNet a Lexical Semantic Representation System

A Step toward Compositional Semantics: E-HowNet a Lexical Semantic Representation System* Keh-Jiann Chen a and Shu-Ling Huang b aCKIP, Institute of Information Science, Academia Sinica kchen@iis.sinica.edu.tw bDepartment of Chinese Language and Literature Studies, National Hsinchu University of Education slhuang@mail.nhcue.edu.tw Abstract. The purpose of designing the lexical semantic representation model E-HowNet is for natural language understanding. E-HowNet is a frame-based entity-relation model extended from HowNet to define lexical senses and achieving compositional semantics. The followings are major extension features of E-HowNet to achieve the goal. a) Word senses (concepts) are defined by either primitives or any well-defined concepts and conceptual relations; b) A uniform sense representation model for content words, function words and phrases; c) Semantic relations are explicitly expressed; and d) Near-canonical representations for lexical senses and phrasal senses. We demonstrate the above features and show how coarse-grained semantic composition can be carried out under the framework of E-HowNet. Possible applications of E-HowNet are also suggested. We hope that the ultimate goal of natural language understanding will be accomplished after future improvement and evolution of the current E-HowNet. Keywords: E-HowNet, compositional semantics, lexical semantic representations 1 Introduction E-HowNet is a lexical semantic representation model extended from HowNet and hence inherits the basic framework of HowNet. HowNet is an on-line common-sense knowledge base indexing relations of concepts gathered from lexicons of Chinese and English (Dong and Dong, 2006). Each concept is represented and understood by its definition and associated links with other concepts. For instance, the HowNet definition of Warrior|戰士 is as: {human|人:belong={army|軍隊}, {fight|爭鬥: agent={~}, domain={military|軍}}} The representation says that a warrior is a human in army who plays the role of agent in the event of military fighting. In HowNet, vocabulary of word sense definition is restricted to a set of around two thousands primitive concepts, called sememes. A word sense is defined by its hypernymy sememe and additional semantic features. HowNet has the following advantages over WordNet (Fellbaum, 1998): a) inherent properties of concepts are derived from encoded feature relations in addition to hypernym concepts, and b) information regarding conceptual differences between different concepts and information regarding morph-semantic structure are encoded. HowNet’s advantages make it an effective electronic dictionary for NLP applications. In recent years, HowNet has been applied to a variety of research topics including: word * This research was supported in part by the National Science Council under a Center Excellence Grant NSC 96- 2752-E-001-001-PAE and Grant NSC96-2221-E-001-023-MY3. Copyright 2009 by Keh-Jiann Chen and Shu-Ling Huang 1 23rd Pacific Asia Conference on Language, Information and Computation, pages 1–8 similarity (劉群, 2002), machine translation (Dong, 1999), and information retrieval (Dorr et al., 2000), etc. However, what interests us here is how to use HowNet to achieve mechanical natural language understanding. Computer understanding requires a representational framework which represents knowledge about lexical concepts and performs semantic composition and sense disambiguation. So far there has been little research on applying HowNet to semantic composition. We therefore propose a framework extending HowNet, called E-HowNet, to deal with this problem. In the …

Content truncated to 500 words