ABSTRACT
The recognition of these lexical entries depends upon identification of where each corresponding unit in the input begins and ends—that is, on the recognition of word boundaries. The strategy that English listeners apparently use to solve the lexical segmentation problem in speech recognition exploits the rhythmic structure of the English language. The "metrical segmentation strategy" exploits stress rhythm in the following way: English listeners assume that every strong syllable is the onset of a new content word. The misperceptions of speech that are of relevance to lexical segmentation are those which involve word boundaries. The continuous temporal nature of speech is best suited to a bottom-up analysis. Evidence from paradigms such as phonetic categorisation indicates that the lexicon does not influence pre-lexical processing. On the basis of evidence from tasks requiring explicit phonetic decisions, then, it would appear that autonomous theories of speech processing are more plausible than interactive theories.
