On Formalization and Formal Linguistics

In NLLT 7,1 (1989), Geoffrey Pullum laments the impending doom of formal linguistics in favor of the "gentle, vague, cuddly sort of linguistics" that he feels I have been advocating since a shift of opinion that he traces to 1979. His account, however, is based on misreading and serious misunderstandings. I will keep to the latter. Two interrelated issues lie in the background: the status of formal linguistics and of formalization. A few comments follow on each. Pullum quotes approvingly a passage from my Syntactic Structures (1957, henceforth SS) on the value of replacing "obscure and intuition-bound notions" by "precise and technical development of linguistic theory," claiming that in the post-1979 heresy, I "flatly reject" my 1957 position. On the contrary, I have never questioned it, though a few years later I came to realize that the project under discussion namely, the one pursued in Logical Structure of Linguistic Theory (1955-6, LSLT) was premature and far too ambitious. One of the "obscure and intuition-bound notions" that should be clarified or eliminated is set of well-formed (grammatical) expressions (E-language, in the terminology of my Knowledge of Language (1986, henceforth KOL)). Though unproblematic (by stipulation) in the theory of formal languages, the notion remains obscure, perhaps lacking any empirical status, for natural language. In LSLT, a notion of E-language is indeed proposed, but as a high-level construct within the theory of strong generation of structural descriptions (SDs) (specifically, within the theory of derived constituent structure). Strong generation is the basic notion; weak generation of a designated set of strings (an Elanguage) is defined as an auxiliary concept, of marginal significance at best. The LSLT proposal was soon recognized to be faulty, and in my Aspects of the Theory of Syntax (1965), it is assumed that a generative grammar (an 1-language in the terminology of KOL) strongly generates an SD for every expression. The • only notion of E-language proposed is the trivial one: every I-language L weakly generates the fixed E-language E, where E is the set of all valid phonetic representations; each of these is assigned an SD by L. The set of SDs strongly generated (the structure of the language, in the suggested terminology) is a nontrivial set associated with L; the string set generated is not. See pp. 30f., and for further discussion, KOL, chapter 2, particularly note 17.