An abstract is not available for this content so a preview has been provided. Please use the Get access link above for information on how to access this content.

However the human listener manages to parse sentences, he or she must frequently manage to do so without ever having simultaneous access to the entire surface of the input sentence. This follows from two very simple and undisputed facts. First, the physical speech signal fades extremely rapidly. The complex wave form that carries speech is damped almost immediately upon delivery. Mercifully, perhaps, the air at the surface of the earth has no resonant properties by which to provide echoes of words, phrases, or sentences. With the exception of caves and canyons, the physical circumstances of speech ordinarily guarantee a transient life for the speech signal. To a certain extent, the human cognitive apparatus compensates for the rapid demise of the physical signal. Short term memory provides some of the resonant capacity that the speech environment lacks. However, the second fact intrudes for human short term memory is notoriously limited. Although exact estimates of short term memory capacity disagree, still there are no estimates of the capacity of immediate memory that would even remotely suggest that the human listener can hold in mind (say) the entire 15-25 words of a modestly complicated spoken sentence. There are two consequences that follow directly from the fact that the human parser's simultaneous access to the surface of the sentence—“window-is frequently too narrow to encompass a sentence, which it nevertheless parses successfully.” These consequences take the form of constraints that any empirically adequate model of parsing must obey. The first such constraint might be called the weak left-to-right constraint. The second constraint might be called the left accessibility constraint.