Two Attempts to Do Without Mental Phrase Markers

Abstract

I cast doubt on two proposals for doing without mental phrase markers (MPMs). The first is due to Roger Schank and his colleagues at Yale, who constructed comprehension models that relied almost exclusively on semantic and pragmatic resources. I rehearse the striking and pervasive failures of such models and suggest that similar problems will likely plague newer incarnations in the connectionist tradition. The second proposal for doing without MPMs is Devitt’s “brute-causal” conception of language processing, which sees comprehension as a reflex like, associative mapping directly from words and sentences to concepts and thoughts. I argue on empirical grounds that language comprehension, even at the earliest stages, is neither reflex-like nor associative. Setting that aside, I examine Devitt’s distinction between responding to a property and representing something as having it, and show that the operations of the HSPM cannot be mere “responses” in the relevant sense. Parsing requires keeping track of prior context, and the clues that are relevant for a successful parse may be arbitrarily far back and in discontinuous regions of the stimulus stream. To explain the HSPM’s sensitivity to contextually relevant factors, and the resulting flexibility of its decision-making and behavior, we must posit inference-like transitions between MPMs.