UNIQUENESS POINT
The point at which a spoken word becomes distinct from all others. Some accounts of spoken word recognition such as Cohort Theory adopt the premise that words are often recognised before their offsets, i.e. before the whole word has been uttered. As more and more of a word is heard, an initial set of likely word matches is narrowed down until, at the uniqueness point, a single match is identified.
The hypothesis derives from evidence that some listeners are capable of shadowing (repeating back running speech) at delays of around a quarter of a second behind the speaker. This is approximately the length of a syllable in English, suggesting that listeners are capable of early recognition of words before the phonetic information is complete. It has also been shown that co-articulatory information in vowels enables listeners to anticipate syllable endings before they hear them. According to some commentators, recognition is also supported by evidence from the general context.
‘Early recognition’ offers an account of how word boundaries are identified in connected speech. Having recognised a word early, the listener can anticipate its offset and this provides a marker for the onset of the word that follows.
However, evidence from a statistical analysis of the lexicon suggests that many words do not, in fact, have an early uniqueness point. Luce (1986) found that, with frequency weighting, only 39 per cent of English words in normal speech areunique before theiroffsets, and only another 23 per cent at offset. Many sequences that appear to constitute monosyllabic words may prove instead to be the initial syllables of polysyllabic ones, while 94 per cent of two-phoneme words and 74 per cent of three-phoneme words are potentially part of a longer word. Luce’s figures are probably an under-estimate since they do not take full account of suffixation (RUN is potentially the first syllable of RUNNING).
Experimental findings have also cast doubt upon the view that recognition is tied to a word’s uniqueness point. In the gating task, subjects are presented with progressively longer sections of input and asked to record their impressions of what they hear. This has shown that about half of low-frequency monosyllabic content words are not identified until after the word has finished– on the strength of subsequent phonetic (and possibly syntactic and semantic) informa tion. Words are more likely to remain unrecognised at offset if they are of short duration, occur early in an utterance or are functors. Overall, late recognition is quite common, and often takes the form of two words being recognised simultaneously. This presents a challenge to the view that we always process utterances linearly, from ‘left to right’; it suggests that much processing is retroactive.
See also: Cohort Theory, Lexical segmentation, Shadowing
Further reading: Luce (1986); McQueen et al. (1995)
