Zero Pronoun Resolution in Thai: A Centering Approach

Since pronouns can be dropped in Thai, a natural language processing system for Thai must be able to resolve referents of the missing pronouns. One of several approaches that have been used for reference resolution is Centering Theory. Centering Theory is a focusing process in which salience of discourse entities is being kept track of. Referents of pronouns or zero pronouns are usually entities that are in focus. However, centering model can resolve only pronouns or zero pronouns whose antecedents are in the immediately preceding utterance. In this study, we indicate that antecedents of Thai zero pronouns are not always in the immediately preceding utterance. Discourse structure is hypothesized to be relevant for resolving zero pronouns, and centering model is extended to work with the hierarchical structure of discourse. To investigate whether hierarchical structure of discourse is relevant for zero pronoun resolution in Thai, the extended centering and the existing centering algorithms were tested on the same corpus. The results indicate that the extended model did not perform better than the existing model because most of antecedents are in the immediately preceding utterance. A few are in a distant utterance. Coreferences of these zeroes could be explained in terms of hierarchical structure of clauses, which seems to operate at the sentence level rather than at the discourse level. However, the number of examples found in this study are too small to make a strong conclusion. Further research should be pursued on a larger corpus to see whether the hierarchical structure of discourse is relevant for the resolution.