Computational Models and Psychological Reality

Abstract

The main claim of this chapter and the next is that all psychologically plausible parsing models either represent or embody a grammar. I substantiate this claim by surveying top-down, bottom-up, and left-corner parsing algorithms, illustrating the ways in which they can draw on explicit representations of grammatical principles. I then discuss the Parsing as Deduction approach, wherein a proof procedure takes the rules of a grammar as axioms and derives MPMs as theorems, using a subpersonal analogue of natural deduction. This constitutes the most concrete implementation of the idea that the HSPM draws on syntactic principles as data. Finally, I turn to three strategies for dealing with the massive structural ambiguity that any parser will encounter in the input stream. Resource-based approaches emphasize parsing heuristics that minimize the use of computational resources, like short-term memory. Frequency-based approaches use statistical analyses of corpuses and treebanks to guide parsing decisions. Grammar-based approaches appeal directly to Minimalist syntactic principles in accounting for the HSPM’s behavior in the face of ambiguity. The latter possibility is particularly exciting, as it would show that a Minimalist grammar is not only suitable for describing abstract formal relations, but also the real-time operation of psychological mechanisms.