Shallow Syntactic Processing
There are many points of disagreement between constraint-based models of language comprehension and other approaches such as the garden-path model, but on one assumption they entirely agree: Interpretations are built on syntactic frames. The comprehender computes either one (according to serial models) or activates even more than one (according to parallel models) syntactic structure as an essential part of the understanding process. Much recent work challenges this assumption, some of it conducted in our laboratory. Some kinds of syntactically demanding sentences are understood using fast and frugal heuristics rather than syntactic algorithms, and therefore are frequently misinterpreted. For sentences that are globally syntactically ambiguous people build an underspecified structure that does not commit itself to one analysis or another, and they resolve the ambiguity later semantically if it is necessary to do so. Extraneous material such as disfluencies (i.e., "uh"s) systematically influence interpretations, even though they are nonsyntactic and indeed perhaps nonlinguistic. Overall, a growing body of evidence supports the idea that language comprehension is often superficial and "good enough" rather than complete and ideal. This evidence requires substantial revision of existing models of language processing.