Abstract:

An approach for semiautomatic grammar acquisition from un-annotated corpus about a specific domain is presented. Its grammar is produced by an iterative procedure, it spatially and temporally clusters the words from a domain-specific corpus. Temporal clustering can discover the fragment’s syntactic structure. Spatial clustering can discover the fragment’s semantic category. Finally, the resultant grammar is post-processed by hand-editing. The preliminary experimental result shows that the method is effective and practical.

Key words: dialog system, semantic grammar, kullback-leibler divergence, Mutual Information