Domain Unconstrained Language Understanding Based on How-net

Domain Unconstrained Language Understanding Based on How-net *4. Jhing-Fa Wang, * Hsien-Chang Wang and * Chin-Nan Lee Department of Computer Science and Information Engineering + Department of Electrical Engineering National Cheng-Kung University, Taiwan, R.O.C. Abstract In this paper, we propose a method for domain unconstrained language understanding based on the How-net knowledge base. The goal is to construct a system that reads in an article and answers some related questions. For each sentence in the article, word segmentation is first applied. Then, the major components such as agent, theme, event, time, and place are extracted to construct a semantic-slot table and a semantic network. Answers of the questions are derived using two approaches, which are based on the relational and hierarchical relation among the major components. Our method is applied to the understanding of the primary-school textbook, and it is able to answer questions in the exercise of the textbook. 1. Introduction Currently, most available applications of natural language processing (NLP) are domain specific. In this paper, we propose a method for domain unconstrained language understanding. The knowledge base we used is the How-net knowledge base, which is constructed by researchers in Beijing. The description and content of How-net can be found in the following URL, http://www.how-net.com. How-net describes the relationship of objects using both concepts and attributes. Based on How-net knowledge base, we have some methods to analyze the sentences of an article. First, word segmentation is performed to find the corresponding word sequence. Then, major components (agent, event, time, place, and theme) conveyed in the sentence are extracted. Finally, the semantic table and semantic network are constructed for the understanding of the article. For the experiment, we try to answer the questions in the exercises of the primary-school textbook. Article of each lesson is used to construct the corresponding semantic table and semantic network. Answers of the questions are derived by measuring the likelihood of the major components. This paper is organized as follows. In Section 2, we introduce basic concept of the How-net knowledge base. In Section 3, the method to analyze sentences is described. In Section 4, we show how to answer the questions of the primary-school textbook. The conclusion is given in the final Section. 2. The Structure of How-net How-net is a Chinese knowledge base which describes the objects using concepts and attributes. The basic units in the How-net are physical and spiritual objects such as components, attributes, time, space, events, attribute values, and so on. Common and individual characteristics of concepts are both recorded in How-net. Consider the words "g-(doctor)" and "2ff(patient)" as example, the common characteristic of them is "people". This common characteristic is recorded in the How-net. On the other hand, "doctor" has the individual characteristic being the agent of "cure"; and "patient" has the individual characteristic being the experiencer of "suffer". The relations among different concepts and attributes are also described by How-net. The relations are shown below: (1) Upper-Down relation.� Ex. Father-Son, Father-Daughter, .. (2) Synonymous relation.� Ex. Good-Well, …

Content truncated to 500 words