Wide-Coverage Grammar Extraction from Thai Treebank

Parsing is an important step for natural language understanding, including phrase alignment for supporting statistical machine translation. Ability on analysing real text by parser strongly depends on grammar. Treebank could be one of the sources for grammar extraction. However, treebank construction largely relies on human annotators intuitions. Different intuitions from multiple annotators bring inconsistency to treebank construction. In this paper, we propose method to construct treebank with semi-automatic correction. Furthermore, we utilise extracted grammar from our corrected treebank to improve semi-automatic phrase structure annotation accuracy for next incremental treebank construction. Thus, it reduces wasting labour force on phrase structure annotation. Using corrected treebanks, we can extract wider coverage grammar from them for supporting parser.