ISCA Archive SWAP 2000
ISCA Archive SWAP 2000

Allophonic influences on word-spotting experiments

Rachel Smith, Sarah Hawkins

The key role of prosody and phonotactics in word segmentation is well established; allophonic detail, acknowledged to play a part, has received less attention (Norris et al., 1997). Yet allophones, as positional variants of phonemes, convey prosodic and phonotactic information that should guide listeners. This work describes a first attempt to assess the role of allophonic detail in segmentation: in a word-spotting paradigm broadly following McQueen (1998), one critical allophone (CA) of the target word to be spotted was either appropriate or inappropriate for that word in citation form.

Seventy-two monosyllabic target words were embedded in pairs of nonsense contexts, one member containing the appropriate CA, and the other the inappropriate one. 26 "Onset" stimuli had targets in syllable 2, and CAs at syllable onset, e.g. reef in [vI"ri:f] (Appropriate) and [vI"kri:f] (Inappropriate). 16 "Coda" stimuli had targets in syllable 1, and CAs at syllable coda, e.g. bell in ["bELp@sIm]* (Appropriate) and ["bEl@sIm] (Inappropriate). "Morpheme" stimuli comprised a target word and a pseudomorphemic affix: 20 "Coda" (ill in ["ILl@s]* (Appropriate) and ["Il@s] (Inappropriate)) and 10 "Onset" (pip in [dIs"phIp] (Appropriate) and [dI"spIp] (Inappropriate)). Thirty listeners pressed a button upon spotting words and said the words aloud. Allophonic appropriateness interacted significantly with stimulus category in ANOVAs on reaction times and errors. Responses were slower and less accurate to Inappropriate than Appropriate allophones in "Onset" and "Morpheme" stimuli, but no difference was found with "Coda" stimuli.

Do allophones affect segmentation independently of the prosodic and phonotactic information they carry? The largest effects confounded allophones with phonological structure: targets in stimuli like [vI"kri:f] had to be extracted from syllables with complex onsets; stress was not varied, as target syllables were strong and nonsense contexts weak. Nonparametric tests on subsets of the data indicate that allophonic detail was important despite these confounds. Firstly, words seemed harder to spot when the CA in onset clusters was devoiced (e.g. /pl/ versus /bl/). Secondly, allophonic appropriateness affected "Coda Morpheme" items, which did not cluster CA with tautosyllabic consonants. Ongoing work will be presented which aims to disentangle metrical and allophonic factors, e.g. by varying allophones in weak syllables. Implications will be discussed with regard to phoneme-free processing, and episodic versus abstractionist approaches to sound categorisation.

Note: L designates dark /l/.

McQueen, J.M. (1998). Journal of Memory and Language, 39, 21-46.

Norris, D., J.M. McQueen, A. Cutler and S. Butterfield (1997). Cognitive Psychology, 34, 191-243.


Cite as: Smith, R., Hawkins, S. (2000) Allophonic influences on word-spotting experiments. Proc. Spoken Word Access Processes (SWAP), 139-142

@inproceedings{smith00_swap,
  author={Rachel Smith and Sarah Hawkins},
  title={{Allophonic influences on word-spotting experiments}},
  year=2000,
  booktitle={Proc. Spoken Word Access Processes (SWAP)},
  pages={139--142}
}