Description:

Dr. Richard Lewis
Psychology, University of Michigan 

Toward Computationally Rational Language Processing and Language Design

Abstract: Characterizing sentence processing as rational probabilistic inference has yielded a number of theoretical insights into human language comprehension (e.g., surprisal theory (Hale, Levy) represents a simple and powerful formalization of incremental processing).  A major theoretical issue now facing the field is how to integrate rational theories with bounded cognitive mechanisms. The standard approach in cognitive science is to posit bounded mechanisms that approximate functions specified at a rational analysis level.  I will discuss a complementary approach, computational rationality, that uses assumptions about bounds in the rational analyses themselves.  The approach naturally admits of two kinds of analyses: derivations of architectural mechanisms that are optimal across a broad range of tasks, and derivations of programs that are optimal for specific task demands.  I will illustrate with three analyses.   (1) We will consider how the derivation of an optimal short-term memory system, given quite general assumptions about noisy representations, provides principled explanations of certain interference and speed-accuracy tradeoffs in sentence processing. (2) We will consider how the derivation of optimal eye-movement policies, given quite general assumptions about perceptual noise and oculomotor architecture, provides principled explanations of task and payoff effects in simple reading tasks. (3) More speculatively, we will consider how the approach may provide a way of theorizing about language "emergence" in a way that is close to the explanatory aims of minimalist theory.  This idea will be illustrated with a simple computational exploration of language emergence that treats language as a solution to a cognitive/motor control problem, abstracting away from mechanisms of biological or cultural evolution.