A Model of Learning Syntactic Comprehension for Natural and Artificial Grammars

Abstract

One important function of syntactic analysis of speech is the assignment of thematic roles to noun phrases. Thematic roles of noun-phrases in canonical sentences are assigned in “default” orderings (e.g. Agent Object Recipient in English for the canonical sentence “John gave the ball to Mary.”). This ordering is transformed in non-canonical sentences (e.g. “The ball was given to Mary by John.”), whose thematic role assignment is guided, in part, by function items (e.g. prepositions “to” and “by”). Agrammatic patients are impaired in the syntactic comprehension of non-canonical sentences. It is not clear whether this is due to a failure to process function items, a failure to assign thematic roles, or both. We have recently studied artificial grammar learning in a recurrent network model in which serial surface structure, and abstract transformational rules are represented by separate systems. We now examine the behavior of this dual system model in syntactic comprehension of canonical and non-canonical sentences. Function items are represented by the “surface” system and guide the application of transformational rules that are represented by the “abstract” system. After learning the structural regularities of the target language, the model predicts that failure in either of these systems, i.e. in the representation of function items or in the representation of syntactic knowledge will impair syntactic comprehension. The model also predicts that the expression of these impairments should not be restricted to performance with natural grammars but should also be manifest in artificial grammar conditions. With respect to this second prediction, we have recently confirmed that agrammatic patients indeed fail to process non-canonically ordered sentences both in natural and artificial grammars. In the artificial grammar sequences, this failure is purely related to application of the transformation, as no function items are involved in the task. The relevance of these results to current linguistic theory will be discussed.