9 - On the modularity of sentence processing: semantical generality and the language of thought

Summary

In memoriam Jacob Henry Atlas

Introduction

In section 1.1 of chapter 2 in this volume, Stephen Levinson reviews arguments that I and others have proffered in support of the conclusion that “[semantic representations] and [conceptual representations] are not only distinct kinds of representations they are also not isomorphic” (Levinson this volume). They are distinct kinds of representations, “contrary to assumptions in many diverse quarters, including not only Fodorians, Cognitive Linguists, and those of similar views, but also most branches of computational linguistics”. But the contrast that I defended was not one between types of representations; it was one between thoughts and English sentences, both ‘propositional’ or language-like in nature: thoughts that are true or false, and statements that are true or false. In the history of modern philosophy and psychology, from John Locke forward, philosophers have underemphasized the autonomy of the language faculty (Chomsky). My arguments were intended to redress the imbalance.

But I did not assume that thoughts were ‘representations’ in some computational sense. For those philosophers of mind and cognitive scientists who accept a version of a ‘computational theory of cognition’ (Cummins 1989) for mental states like belief, desire, etc. – or a Representational Theory of the Mind in general (Fodor 1975, among others) – the moral of my (Atlas 1975, 1977, 1978, 1979, 1989) argument for them was that many ‘sentences’ in Mentalese are not identical to, synonymous with, or translatable into any sentence in English.