Learning grammatical stucture using statistical decision-trees

In this paper, I describe SPATTER, a statistical parser based on decision-tree learning techniques which avoids the difficulties of grammar development simply by having no grammar. Instead, the parser is driven by statistical pattern recognizers, in the form of decision trees, trained on correctly parsed sentences. This approach to grammatical inference results in a parser which constructs a complete parse for every sentence and achieves accuracy rates far better than any previously published result.