SPOKEN WORD ACCESS PROCESSES (SWAP)

May 29-31, 2000
Nijmegen, The Netherlands

Language-universal constraints on the segmentation of English

Dennis Norris (1), Anne Cutler (2), James M. McQueen (2), Sally Butterfield (1), Ruth Kearns (1)

(1) MRC Cognitive and Brain Sciences Unit, Cambridge
(2) Max Planck Institute for Psycholinguistics, Nijmegen

Norris, McQueen, Cutler and Butterfield (1997) proposed that listeners employ a Possible Word Constraint (PWC) during word recognition in continuous speech. The PWC disfavours interpretations which would leave an unparseable residue between the end of a candidate word and a "known boundary". Such an unparseable residue would be any string that could not constitute a possible word in the language. For example, in English it would not be possible for a word to consist solely of a sequence of consonants, or solely of an open syllable with a lax vowel. This constraint offers a powerful method for inhibiting activation of spuriously embedded words in spoken utterances. Thus the input "the committee met a fourth time" might activate, inter alia, "for" and "metaphor", but their activation could be reduced on the grounds that they would leave a single consonant as residue (th), and a single consonant cannot constitute a word. 

Norris et al.'s experiments used the word-spotting task (detection of any real word, in short nonsense strings), and successfully demonstrated that words are harder to spot when they would leave a single consonant as a residue (e.g. apple in fapple, or sea in seash) than when they would leave a syllable as residue (e.g. apple in vuffapple, sea in seashub). In this presentation we will describe two further experiments which addressed the issue of whether the PWC is language-specific or language-universal. If it is language-specific, then an open syllable with a lax vowel, for instance, will form a difficult context for word-spotting in English, since such a string cannot be a real word of that language. Other languages (e.g. French, Japanese) do, however, allow words of this structure; thus if the PWC is language-universal, it must allow possible words of this type, and therefore an open syllable with a lax vowel will constitute an easy context for word-spotting in all languages, including English. In the first experiment we compared the spotting of words in three contexts: open syllable with long vowel (canal in zeecanal, with the same vowel as in peek), open syllable with short lax vowel (zEcanal, with the same vowel as in peck), and consonant (scanal). Word-spotting was significantly harder in the consonant context than in the syllable contexts, which did not differ; this result is consistent with a language-universal form of the PWC.

In the second experiment we compared contexts containing a strong (full) versus weak (reduced) vowel -- bell in bellshif (with the same vowel as in pick) or bellsh@f (where @ represents schwa); again these were contrasted with a consonant context (bellsh). In this experiment, word-spotting was again significantly harder in the consonant context than in both syllable contexts, which did not differ. Other studies have shown that boundaries before weak syllables tend to be overlooked in listening to English, but this language-specific effect appears not to affect the PWC. Although the "known boundaries" on which the PWC operates appear to be determined by language-specific (e.g. rhythmic and phonotactic) properties of the language, the PWC itself may be a language-universal constraint, disfavouring interpretations which leave any nonsyllabic residue.

Full Paper

Bibliographic reference.  Norris, Dennis / Cutler, Anne / McQueen, James M. / Butterfield, Sally / Kearns, Ruth (2000): "Language-universal constraints on the segmentation of English", In SWAP-2000, 43-46.