ISCA Archive SWAP 2000
ISCA Archive SWAP 2000

Syllabic cues to word segmentation

Cecilia Kirk

The Shortlist model of spoken word recognition proposes that the lexical competition process is modulated by word boundary information. Listeners do not assume that words can begin at any point in the input, but instead they use their knowledge of what constitutes a possible word in their language to limit the number of competitors (Norris et al., 1997). It has been shown that listeners use metrical structure (Cutler et al., 1988) and phonotactic knowledge (McQueen, 1998) to select the most likely locations for word boundaries. The present study uses the wordspotting task to investigate the role of the syllable in the segmentation of English for those cases where syllabification is not dictated by phonotactics.

Listeners appear to follow a strategy of maximizing the onset when segmenting the input. The word smooth was spotted much more quickly and accurately when embedded in voosmooth than spotting of neat embedded in voosneat. Wordspotting in the latter case is more difficult since onset maximization ensures that the syllable boundary is misaligned with the word boundary even though there is no phonotactic requirement that s and the following nasal be syllabified together. The mean percentage of missed targets and mean reaction times for each of the five hypotheses under investigation are summarized below.

Another segmentation strategy used by English listeners is the attraction of consonants to stressed syllables. A consonant which occurs with a vowel on either side is parsed as belonging to an adjacent stressed syllable. Listeners found it easier to spot the word edge embedded in ZEENedge than when embedded in zeeNEDGE (upper case indicates stress).

The devoicing of a glide following an aspirated stop in the same syllable gives an important cue to a speaker's intended syllabification. If wine is preceded by k syllabified together with the glide in the onset, as in zee.khwine, it is more difficult to spot than when the k is syllabified as a coda and no devoicing occurs.

When alveolar stops occur together with [r] in the syllable onset, they are very commonly pronounced with a strongly retroflexed articulation. This retroflexion provides important information about possible word boundaries. The word rock is much more difficult to spot when the preceeding d in fidrock is retroflexed than when it is not.

Words must contain at least one vowel, and the rime of English monosyllabic words must contain either a tense vowel or a lax vowel and coda consonant. This restriction on possible words in English causes the lax vowel in the first syllable of vesland to attract the s to the coda resulting in alignment of both the syllable and word boundary. Listeners were faster to spot land when embedded in vesland than when embedded in voosland, although they were not more accurate.

These findings indicate that sensitivity to syllabic structure, in addition to knowledge about phonotactics and metrical structure, helps to restrict the list of candidates entered into the competition process.


Cite as: Kirk, C. (2000) Syllabic cues to word segmentation. Proc. Spoken Word Access Processes (SWAP), 131-134

@inproceedings{kirk00_swap,
  author={Cecilia Kirk},
  title={{Syllabic cues to word segmentation}},
  year=2000,
  booktitle={Proc. Spoken Word Access Processes (SWAP)},
  pages={131--134}
}