Strong Generative Capacity and the Empirical Base of Linguistic Theory.

Dennis Ott

doi:10.3389/fpsyg.2017.01617

Abstract

This Perspective traces the evolution of certain central notions in the theory of Generative Grammar (GG). The founding documents of the field suggested a relation between the grammar, construed as recursively enumerating an infinite set of sentences, and the idealized native speaker that was essentially equivalent to the relation between a formal language (a set of well-formed formulas) and an automaton that recognizes strings as belonging to the language or not. But this early view was later abandoned, when the focus of the field shifted to the grammar's strong generative capacity as recursive generation of hierarchically structured objects as opposed to strings. The grammar is now no longer seen as specifying a set of well-formed expressions and in fact necessarily constructs expressions of any degree of intuitive “acceptability.” The field of GG, however, has not sufficiently acknowledged the significance of this shift in perspective, as evidenced by the fact that (informal and experimentally-controlled) observations about string acceptability continue to be treated as bona fide data and generalizations for the theory of GG. The focus on strong generative capacity, it is argued, requires a new discussion of what constitutes valid empirical evidence for GG beyond observations pertaining to weak generation.

Highlights

There exists a contradiction between the near-universal acceptance of acceptability judgments as a source of data for Generative Grammar (GG) on the one hand and the theory’s express focus on strong generative capacity on the other
Perhaps the first clear articulation of this shift appears in Chomsky (1980), where we find the assertion that “[a GG] does not in and of itself determine the class of what we might choose to call ‘grammatical sentences’ [...],” an unremarkable conclusion “once we recognize that the fundamental concepts are grammar and knowing a grammar, and that language and knowing a language are derivative” (p. 126)
In many cases “acceptability judgments” are shorthand for judgments about such correlations—we can say that Hei likes Johni is “unacceptable,” or that it lacks the intended reading; we can say that (2) above is “deviant,” with an implicit understanding that we’re referring to the absence of an interpretation analogous to (1)

Summary

INTRODUCTION

There exists a contradiction between the near-universal acceptance of acceptability judgments as a source of data for Generative Grammar (GG) on the one hand and the theory’s express focus on strong generative capacity on the other. While linguists agree on this focus, they tend to uncritically assume that judgments of the acceptability of strings constitute data for GG. The set of sequences so determined “corresponds to the ‘intuitive sense of grammaticalness’ of the native speaker” (LSLT, 95); “the sequences generated by the grammar as grammatical. Generative Capacity and Linguistic Theory sentences must be acceptable, in some sense, to the native speaker [...]” (LSLT, 101). “the linguist’s task [is] that of producing [...] a grammar [that generates] all and only the sentences of a language [...]” (SS, 85) On this Early View (EV), the idealized native speaker is the human equivalent of an automaton in the theory of formal languages, which accepts (recognizes) or rejects a given string depending on whether or not it is part of the set of legal sequences. A true shift in perspective, took place later, when the notion of sentence, understood as sequence in L, was eliminated altogether from the theory

A SHIFT IN PERSPECTIVE

CONCLUSION