Parsing Chinese by Examples

This paper presents a Chinese parser which has been derived from the Chinese treebank developed at CKIP, Academia Sinica. Contrary to previous approaches which aim at the conversion of a treebank into a parsers, we do not derive phrase structure rules of any type. Instead, the approach chosen relies on a fuzzy pattern matching strategy in order to extract relevant examples from the treebank. Via a set of adaptation mechanism, these examples are merged and modified so as to produce the best parse for the given set of examples. A detailed description of the parser is provided. The different modules of this parser are evaluated. It is shown that the parser is not only efficient and robust but provides a reasonable level of linguistic adequacy which can be improved upon by restricting the application domain or increasing the number of examples. Competitive approaches are presented and compared to the proposed approach.