Segmenting speech into words

Four experiments employed a listening for mispronunciations task to determine how listeners perceive an ordered series of words from a continuous, phonetically ambiguous stimulus. In experiments 1, 2, and 3, listeners’ reaction times to detect mispronunciations were obtained in phonological sequences that could be perceived as either one or two words (e.g., ’’cargo’’ or ’’car go’’ mispronounced ’’carko’’). In experiment 1, segmentation of the acoustic signal as one or two words was guided by the theme of a short story. For example, the sentence ’’They saw the carko on the ferry’’ was spliced into two stories—one about a shipment of cargo, and one about a car about to go on a ferryboat. The results showed significantly faster reaction times—by about 300 ms—when the mispronunciation was perceived as occurring in the second syllable of a word. In experiment 2, alternate segmentations of the same acoustic signal were constrained by the grammatical nature of the first few words in a sentence. Subjects were presented with sentences such as ’’The doctor said that nosetrops will help the cold’’ or ’’The doctor said he knows trops will help the cold.’’ By means of tape splicing, the final portions of the two sentences in each pair were acoustically identical. Reaction times were again found to be faster—by about 150 ms—when the mispronunciation was perceived as occuring in the second or third syllable of a word, rather than at the beginning of a word. Experiment 3 replicated these results using naturally recorded sentences having normal prosody. The results of experiments 1–3 demonstrated that prior context determines how syllables are recognized as words and that a mispronounced second syllable is detected faster than a mispronounced word‐initial syllable. It was argued that mispronunciations are detected more quickly in second syllables because the intended word has been accessed from its first syllable. This hypothesis was further supported in experiment 4. When subjects read each sentence before hearing it, so that all syllables were equally (and perfectly) predictable, mispronunciations were detected equally fast in first and second syllables.