Acquisition of unbounded dependency using explanation-based learning

A natural language acquisition model using Explanation-Based Learning (EBL) had been proposed to acquire parsing-related knowledge which includes Context-Free grammar rules and syntactic and thematic features of lexicons. The domain theory that is assumed to be innate to the model includes the theta-theory and the universal feature instantiation principles in Generalized Phrase Structure Grammar (GPSG). In this paper, we show in particular how unbounded dependency may be acquired in the natural language acquisition model. The acquisition problem of unbounded dependency may be further divided into two sub-problems: detecting whether there are moved constituents and finding the places to which the constituents are moved. For these problems, the universal innate domain theory facilitates and constrains the acquisition process which is otherwise intractable.