Abstract:  This paper sketches a solution to a problem which has been emphasized by Fodor. This is the problem of how to explain distinctively–human flexible cognition in modular terms. There are three aspects to the proposed account. First, it is suggested that natural language sentences might serve to integrate the outputs of a number of conceptual modules. Second, a creative sentence–generator, or supposer, is postulated. And third, it is argued that a set of principles of inference to the best explanation can be constructed from already–extant aspects of linguistic testimony and discourse interpretation. Most importantly, it is suggested that the resulting architecture should be implementable in ways that are computationally tractable.