Abstract

AI's architecture is connectionist, not computational. Despite this, so the present work shows, AI's behavior is not only consistent with, but strongly supports, a number of shibboleths of classical semantics, each of which has always been assumed to align with computational conceptions of cognition. These include: (i) the contention that language divides neatly into syntax, semantics, and pragmatics; (ii) the contention that semantics and pragmatics are distinct and, in particular, that semantics is not an "abstraction" of pragmatics; (iii) the contention that meaning is compositional, along with the related contention that linguistic processing is at least partly recursive in nature, granting that it is not solely or even predominantly so. At the same time, (iv) AI-functioning is shown to warrant a complete rejection of the Russell-Frege position that grammar and logical form are misaligned. In the course of establishing (iv), it is shown exactly why, pace Russell-Frege, "everybody snores" and "Bob snores" parallel each other both in respect of grammar and logical form. By way of anticipation: "Bob snores" maps onto "Bob-singleton (smallest class containing Bob) is a subset of the class of snorers"; and "everybody snores" maps onto "the class of people is a subset of the class of snorers." This analysis is easily generalized to hold of all quantified generalizations and of all other sentences whose grammars allegedly misalign with their logical forms. And this last fact comes as a relief to researchers concerned with the cognitive feasibility of semantic theories, especially those familiar with the hundreds of psycholinguistic experiments showing that reaction times, event-related potentials, and error rates are totally inexplicable on the assumption that anything remotely similar to the Russell-Frege position to any degree describes any stratum of linguistic cognition.