Components of a Production Model

There are several ways to approach the construction of a production model. One might take generative grammar as one’s starting point, assume that it has been internalized by the speaker, and wonder what kind of mechanism is required to make use of this internalized knowledge in speaking and comprehending speech. Another approach is to concentrate on writing computer programs that can analyze or produce at least a small subset of natural language sentences. It is also possible to be skeptical regarding both these approaches. As for generative grammar, it is becoming more and more clear that its relation to the speaker–hearer’s performance is much less straightforward than was once fondly hoped. Current formulations of generative grammar have turned out to not adequately predict the complexity of processing sentences. Moreover, there are a priori reasons why the rules according to which a performance model operates are not likely to be identical to those formulated by the linguist ( Fodor, 1971 ; Watt, 1970 ). One may have similar misgivings regarding attempts to model the human speaker by computer programs. The routines which seem most appropriate for the computer may turn out to differ from the processes occurring in the speaker-hearer, who after all operates under quite different constraints and has different resources at his disposal (see, e.g., Keyser & Petrick, 1967 ). While admitting that work in artificial intelligence may in principle lead to a valid model, one may feel that recent attempts at simulation were premature.