Speech perception

Phonetic context effects

As a result of working in John Kingston‘s lab, I’m interested in the explanation for phonetic context effects in speech perception. One clearly established effect of this kind is that a phonetic continuum from [da-ga] tends to be perceived differently depending on the context that precedes it. After [al], “ga” responses are favored, while “da” responses predominate after [ar]. Since the effect can be found with listeners of diverse language backgrounds, we assume this effect and variants on it are not due to the application of language-specific knowledge. (Analogous effects obtain in judgments of a [ta-ka] continuum presented after [s] vs. [ʃ].)

The contexts and target intervals contrast in their place of articulation and in their acoustics: [da] = coronal and spectrally high, [ga] = velar and spectrally low. [al] = tongue blade is high front and spectrally high, [ar] = tongue blade retracted/retroflexed and spectrally low. This raises the following question:

Are the objects of perception the auditory qualities of speech, or the articulations that produce them?

If articulations are directly perceived, then listeners are compensating for perceived coarticulation with the context. Ambiguous [da/ga] tokens may have been pulled forward after [al] and backward after [ar], and so are categorized in the opposite direction to make up for it.

If auditory qualities are the objects of perception, the context effect on [da-ga] categorization may be due to an auditory contrast mechanism that exaggerates the physical differences between the context and target intervals. Thus, ambiguous [da/ga] stimuli are more likely to be labelled “ga” after [al] and “da” after [ar]. (See Lori Holt’s webpage for sound files that illustrating this effect.)

We extended this finding in a series of discrimination experiments that favor the auditory account over the articulatory alternative:

Kingston, John, Kawahara, Shigeto, Chambless, Della, Key, Michael, Mash, Daniel, and Sarah Watsky (2014). “Context effects as auditory contrast” Attention, Perception, & Psychophysics 76(5). 1437-1464.

Article [PDF]

If listeners perceive coarticulation rather than acoustics, context effects in non-speech analogues require another explanation — auditory masking. That means that energy in an interval at one frequency obscures energy at similar frequencies in an adjacent interval. We tested the masking hypothesis directly by varying (i) the intensity of the context interval, (ii) the silent gap between it and the target interval, and (iii) and how closely the frequencies in the two inter- vals matched; all of these manipulations should produce effects if masking is responsible for context effects. Only (i) had an effect, which shows that context effects cannot be reduced to auditory masking.

Poster [PDF]

Linguistic mediation of context effects

Are the auditory processes that are responsible for context effects affected by higher-level linguistic processing? The slides below review the long series of experiments, including one carried out at UMass, that have found evidence for both answers to this question.

Presentation [PDF]

Note: Several of the figures in the slides show an idealized version of experimental results. These figures were only intended as visual aids in a psychology seminar.

Comments are closed.