Some Effects of Semantic and Grammatical Context on the Production and Perception of Speech

Three speakers read aloud meaningful grammatical English sentences at a fast rate. Some of these sentences contained common maxims and stereotyped phrases. Other sentences which were less familiar contained, in similar phonetic environments, certain test words that also occurred in the stereotyped sentences. The test words were “excised”, i.e., gated out of all the sentences, and listening tests were performed by 43 listeners. Quite apart from these listening tests two operational measures of redundancy were derived. One measure was based on a Markovian model of perception and readers were asked to guess what word n would be after they had read all the words through word n–1 in a sentence. The second measure was based on the hypothesis that people reserve their final decision on the recognition of each word until they perceive an entire sentence. Readers read an entire sentence in the midst of which a word was represented by a dash. They were then asked to guess what the missing word was. Two groups of thirty readers each performed these written tests on the sentences from which the test words had been excised. Indexes of redundancy were computed for each test word from the percent of correct guesses that occurred for the two written tests. The results of the listening tests show that the intelligibility of the excised words is inversely proportional to the redundancy index obtained from the total sentence context. Two acoustic correlates of linguistic stress (duration and amplitude) were measured for each excised word. Stress is also apparently inversely proportional to the sentence context redundancy measure. The results of this experiment support the hypothesis that both the acoustic realization and auditory perception of a given word in a meaningful sentence may be a function of the speaker's and listener's knowledge of the semantic and grammatical information contained in the entire sentence. The experimental results also indicate that, in some circumstances, a listener may be able to identify a word only after he adapts to the speaker's voice.