Linguistics Colloquium: Thomas Graf, Associate Professor, Linguistics Stony Brook University, SUNY

APR 08, 2021 | 4:15 PM TO 6:15 PM



Online Event


April 08, 2021: 4:15 PM-6:15 PM




Thomas Graf, Associate Professor, Linguistics Stony Brook University, SUNY presents:

A computational Minimalist program for syntax and sentence processing

Abstract: Minimalism emphasizes the simplicity of UG and the role of computational efficiency as a third factor that shapes grammar. But how does one turn this big-picture idea into a concrete proposal with specific predictions? This talk presents a computational answer in the form a simple processing model for Minimalism that 1) unifies a broad range of phenomena from relative clause processing to heavy NP shift and quantifier scope preferences, and 2) suggests hard limits on the complexity of syntax, thereby deriving a number of island constraints.

The model combines multiple components: Minimalist grammars as a computational counterpart to Minimalism, a sound and complete top-down parser, and simple memory usage metrics that quantify the processing load of any given parse. The model is simplistic to a fault as it deliberately ignores lexical and structural frequencies, morphology, all ambiguity, context, world-knowledge, and much more. Nonetheless it makes the correct predictions for a surprising number of seemingly unrelated phenomena, based purely on their syntactic structure.

If this top-down parsing model is on the right track, one has to wonder what syntax would have to look like in order to facilitate top-down parsing. Subregular complexity provides an intriguing answer in the form of Sensing Tree Automata: they allow for deterministic (and hence highly efficient) top-down processing while still being able to track the c-command relations that form the backbone of syntactic dependencies. Curiously, though, these automata also require movement to be top-down deterministic in a manner that immediately rules out extraction from specifiers, adjuncts, and conjuncts, but not across-the-board extraction from all conjuncts. Not only then does a computationally informed perspective turn Minimalist syntax into an empirically viable model of sentence processing, it also lets us reason in the other direction to derive syntactic laws from computational efficiency considerations for parsing.

This is an online event. Register online to receive the Zoom link. All are welcome!