Computational Linguistics Colloquium

Thursday, 5th July, 16:15
Conference Room, Building C7 4

Sentence production and grammatical cognition: The representation and computation of linear order

Sentence production phenomena are an important source of constraints on models of grammar and of grammatical processing that aim for cognitive plausibility. I start with a critical examination of experimental and speech error data that are often interpreted as supporting the distinction between a functional and a positional stage of sentence production. I will argue that the available data do not justify this distinction.

Another widely held assumption concerns the existence of syntactic movement operations and of syntactic traces (gaps) left behind by such operations. A review of the sentence production evidence reveals that traces/gaps are real, at least in the case of fronted Wh-constituents. However, I will outline a model of syntax (called Performance Grammar, PG) that generates traces/gaps without moving constituents around.

The upshot of the foregoing is that the computations giving rise to functional (hierarchical, dominance) structures should be viewed as distinct from, but intimately intertwined with, linear order computations.

Finally, I address the time course of linear order computations. In the PG formalism, they involve rules that "pipeline" lexical items to particular slots in one-dimensional arrays called topologies. Very often, these rules can predict the target slot of potentially upcoming constituents at a point in time where the type an/or the lexical content of these constituents is yet to be determined. I will briefly discuss a set of perception-to-production priming RT experiments whose results are compatible with this property of linear order computing.

If you would like to meet with the speaker, please contact Matthew Crocker.