In the preceding chapter, we described the role of syntax in language processing and the criteria to be met by an adequate theory of syntax. Then we introduced phrase structure grammar and discussed its inadequacies from the point of view of transformational theory. The syntactic mechanisms of transformational theory were then described.

Transformational theory incorporates a large number of different global operations, from the transformation rules themselves to the metarules that in turn control their application. As we discussed in Chapter 1, global operations are less satisfactory in a model than local rules. Global operations are usually expressed in general terms; to actually use them, a further set of procedures must be specified.

In a cognitive model, a global operation is an incomplete explanation. It does not explain how the disparate parts controlled by the global operation are recognized and operated on; it only states that they are. A model of a cognitive process that invokes complex global operations presupposes additional cognitive processes for the model to work. Such a model, in effect, requires a homunculus, a little person inside, who performs the additional operations that are needed to explain the cognitive processes. The model still has to explain how the homunculus operates. Explanation by homunculus is open to the kind of criticism that Miller, Galanter, and Pribram (1960) directed at Tolman's (1948) explanation of navigation behavior that posited a “map control room in the brain” and left unexplained the details of how the maps were drawn and used.