Basic Principles for Segmenting Thai EDUs

Basic Principles for Segmenting Thai EDUs Nalinee Intasaw Department of Linguistics, Faculty of Arts, Chulalongkorn University, Bangkok 10330, THAILAND nalinee.int@gmail.com Wirote Aroonmanakun Department of Linguistics, Faculty of Arts, Chulalongkorn University, Bangkok 10330, THAILAND awirote@chula.ac.th Abstract This paper proposes a guideline to determine Thai elementary discourse units (EDUs) based on rhetorical structure theory. Carson and Marcu’s (2001) guideline for segment- ing English EDUs is modified to propose a suitable guideline for segmenting EDUs in Thai. The proposed principles are used in tagging EDUs for constructing a corpus of discourse tree structures. It can also be used as the basis for implementing automatic Thai EDU segmentation. The problems of deter- mining Thai EDUs both manually and auto- matically are also explored and discussed in this paper. 1 Introduction Elementary discourse unit or EDU is a building block that can combine together to form a larger unit or structure in discourse. It is significant to applications that process discourse such as text summarization, machine translation, text genera- tion, and discourse parsing. In some applications e.g. text summarization and machine translation, an EDU is suitable to be used as an input than a sentence or a paragraph since it is smaller and contains a single piece of information, In addi- tion, in languages in which sentence boundaries are not clearly marked like Thai, determining an EDU would be more practical and more useful since an EDU serves as a building block for con- structing the discourse structure. However, little study has been devoted to Thai elementary dis- course unit. Previous research on Thai discourse structure (Charoensuk, 2005; Sinthupoun, 2009; Katui et al., 2012) did not clearly discussed how to determine an EDU in Thai. Determining an EDU is not an easy task. As a result, Carson and Marcu (2001) had developed a guideline for determining an EDU in English, which is used for tagging discourse tree structure. In this paper, our objective is to propose a guideline to deter- mine Thai EDUs. The proposal is grounded on the framework of rhetorical structure theory by Mann and Thomson (1988). The background knowledge related to our paper will be discussed in section 2. Data used in this work is described in section 3. In section 4, principles for segment- ing Thai EDU are proposed. Problems arisen with Thai EDU segmentation are explored in section 5. The last section will be the conclusion. 2 Background knowledge To analyze the structure of text, the text has to be segmented into pieces of information and linked together to reflect the coherence of text. Rhetorical structure theory (RST), one of the most widely used in both linguistics and compu- tational linguistics, was proposed by Mann and Thomson (1988) to explain discourse structure of written texts. Briefly, RST explains the discourse relation of two spans of texts. It explains how parts of text are organized and formed into a larger structure of text which can be represented as a tree structure. For any two spans of text, one of them will …

Content truncated to 500 words