IntroductionWhen we understand language, we are, by conventional construal of the process, reconstructing the speaker's thought. The physical elements of an utterance are data that listeners use to identify and represent the relevant thought of the speaker. But, that is not all the 'data'. Indispensable additional information resides in the context of utterance and in background information in the mind of the listener. The dynamic combination of these two information sources with the physical form of the utterance determines the outcome of the comprehension process.The notion that the explicit form of an utterance is the primary driver of language processing is often described as a kind of 'bottom up' priority for language comprehension, and there is much in the experimental literature on language processing to support some version of that idea. One does, after all, want some defense against wholesale hallucination. But, one does not march very far into the web of language use before encountering equally strong evidence for constraints derived from the projection of several kinds of background information onto the acoustic (or orthographic, or signed) data stream. These range from the sublimely metaphoric to the most mundanely literal. How is the balance struck between form-driven and knowledge-driven processing in a way that combines veridical perception with our fluent appreciation of the force of an unexpected turn of phrase, a bit of whimsy, or a creative metaphor? We are inclined to the view that the answer lies in the multi-systems approaches that have emerged in many cognitive domains over the past couple of decades. Wherever the research probe has been stuck --memory, human judgment and decision making, mathematical reasoning, social reasoning, and so on, we find a collection of specialized systems from whose interaction the fabric of performance in that domain emerges. We will return to this issue.The infrastructure of phonetics, phonology, morphology, and phrasal syntax are the conservative focus of bottom up analysis approaches to language. Processes based on such structures deal in a sharply limited range of information, and their real-time application to the analysis of a signal is, while complex, demonstrably manageable. Matters become more difficult as syntax becomes more complex and as information not explicitly represented in standard grammar is required for capturing interpretive options. The difficulty is, as Fodor (1983) outlined the matter in his monograph on modularity, that the range of facts to which interpretation of a sentence is potentially responsive is open. This complicates enormously the computational problem that must be solved in real time. It is the limitation on access to information that motivates the informational encapsulation move in modularity theory. And, it is the unencapsulated nature of information relevant to the interpretive nexus of utterance, speaker and context that makes the pragmatic problem so acutely compelling. We know that people can solve this problem and th...