If any extant generative grammar were seriously proposed as a model of speaker- hearer performance, it would immediately face several obvious insufficiencies in principle: (1) Grammars, by virtue of their recursive properties, can produce sentences of indefinite length and complexity; speakers and hearers are finite devices, and their limitations must be incorporated in any performance theory. (2) Grammars are not time-bound in generating sentences; speakers and hearers operate within time constraints, and sentences must be created and understood on a ‘ left-to-right ’ basis. (3) Grammars deal with idealized speaker-hearers who have perfect knowledge of the rules of their language (la langue); performance theories must deal with real speaker-hearers who are both fallible and variable, i.e., perform probablistically (la parole). (4) Grammars contain no learning principles; given the obvious arbitrariness of both lexical and grammatical rules for particular languages, which must be acquired via experience by their speakers, any performance theory must account for such learning. (5) Grammars do not (and need not) provide any account of selection among alternative expansions of the left-hand terms in re-write rules (e.g., of the fact that NP may be rewritten as T + N, T + A + N, T + N + WH + S and so on ad libidum); any performance model must account for the antecedents of such ‘ decisions ’. (6) Grammars do not (and need not) account for the non-linguistic antecedents in speaking or subsequents in understanding of the superordinate symbol S in the generation of sentences; performance theories of creating and interpreting sentences must provide an account of such non-linguistic antecedents and subsequents of sentences.

This last requirement for performance theories is, of course, the question of ‘ where sentences come from and go to ’. The process of Simply Describing, as I have illustrated it here, is a very ordinary and familiar psycholinguistic performance of real speakers. It involves the ability to paraphrase, in language, entities and events (perceptual signs and their interrelations) which are essentially non-linguistic in nature. That speakers of a language do have this ability is evident in the highly similar information content across the sentences produced by the 26 speakers in each of the preceding demonstrations. Possession of this paraphrasing ability implies both (a) that perceptual signs and events must themselves have meaning and (b) that perceptual and linguistic signs must share a common representational (semantic) and organizational (deep syntactic) system - otherwise, signals in the perceptual and linguistic systems would ‘ pass each other like ships in the night ’. Furthermore, both the recent linguistic evidence on ‘presuppositions’ (reviewed earlier in this paper) and the detailed ways in which perceptual antecedents in my demonstrations literally ‘drive’ the form and content of sentential describings strongly suggest that this shared representational and organizational system is not linguistic but cognitive in nature.

If this is the case, then the constructs and rules of generative grammars are, in principle, incapable of accounting for Simply Describing Things. The only conceivable approach via generative linguistics would be an extreme form of Whorfian psycholinguistic relativity - the assumption that people literally impose their ‘ sentences ’ upon the real world, perceiving entities and events in it in terms of what their grammars permit them to say about them. Not only is it counter-intuitive (to me, at least) that I must generate an implicit sentence, the man is holding a small black ball, before I can perceive and understand that this is in fact the state of affairs in the real world, but even casual observations of the behaviors of pre-linguistic children and non-linguistic higher animals contradict any such linguistic restriction upon ability to comprehend. More than this, such extreme psycholinguistic relativism would imply that the basic cognitive abilities of perceptual grouping, contrasting and categorizing are determined by language rather than being determinants of language. As I indicated earlier, the absurdity of this position becomes apparent when one conjures with the number of antecedent sentences which would be required to fulfill the ‘ happiness conditions ’ for a simple sentence like Please shut the door.