Matching Pattern Acquisition Approach for Ancient Chinese Treebank Construction

Matching Pattern (MP) is a sequence of words or part-of-speech (POS), sampled from clauses, and MP acquisition is an effective approach for ancient Chinese treebank construction. This approach uses the typical characteristics of ancient Chinese short-clauses and strong-patterns, and lays down the syntactic annotation process of the treebank construction in three stages. These stages involve: (1) obtaining weighted MPs with a syntactic skeleton; (2) applying these MPs to match the clauses; and (3) generating syntactic structures of these clauses according to the syntactic skeleton of the MP. The syntactic skeletons are constructed based on the Sentence-based Grammar in our experiments. The MP-based parsing procedures are implemented on both clause and fragment units. Experiments on corpora extracted from Yili and Zuozhuan show that an integrated algorithm, involving both clause and fragment units, can achieve a performance of 99.07%/82.76% and 97.25%/77.77% for coverage/precision, respectively.