How long is the sentence? Prediction and prosody in the on-line processing of language

The gating paradigm (Grosjean 1980) was used to determine whether subjects who are listening to the 'potentially last word' of a sentence (in this case, a noun before an optional prepositional phrase) can indicate whether the sentence is over or not, and if it is not over, how much longer it will last. Sentences that contained endings ranging in length from zero to nine words were gated on the object noun and presented to subjects who had to choose which of four sentences was being presented or press a key at a point in time when they felt the sentence would have ended had it been presented in full. Results showed that, basing themselves solely on prosodic cues, subjects were surprisingly accurate at predicting the length of the upcoming endings. An acoustic analysis of the test sentences showed a strong relationship between measures of fundamental frequency, amplitude, and duration and the experimental data. These findings are discussed in terms of the predictive and interpretative roles of prosody during the on-line processing of language. Much of current research in psycholinguistics is aimed at understanding the processes by which language is perceived in real time. Researchers are studying the time course of language perception and understanding; that is, how the listener goes from the acoustic signal to an interpretative representation of the message. Part of this research involves understanding how the processing of the acoustic signal interacts with the discourse context and the rules of the language as well as with pragmatic and environmental factors. A number of assumptions underlie this research. One is that language processing is obligatory and takes place online or in real time (Marslen-Wilson and Tyler 1980). Another is that the listener will use any and all information that may facilitate this processing. A third is that the various sources of information will interact — at some level at least — during processing. A fourth premise — and one that Linguistics 21 (1983), 501-529 0024^3949/83/0021-0501 $2.00