Lexical Context Effects in Speech Perception

Lexical Context Effects

Ganong (1980) Data

Knowledge of words has an important impact on the perception of phonemes. Ganong (1980) demonstrated a bias to perceive phonemes so they form words. For example sounds that could be either /d/ or /t/ tend to be heard as /t/ when followed by "ask" (to make the word "task") and as /d/ when followed by "ash" (to make the word "dash").

reference: Ganong, W. F. (1980). Phonetic categorization in auditory word perception. Journal of Experimental Psychology: Human Perception and Performance, 6, 110-125.

Try this example: identify the phoneme at the end of each utterance, is it /s/ (as in "lease") or /S/ (as in "leash")?

1. GOLDFI [?]

2. FORTRE [?]

It is likely that you heard the first as /S/ and the second as /s/.
However, they were identical sounds!
Your knowledge that "goldfish" and "fortress" are English words but "goldfiss" and "fortresh" are not, changed your perception of the sounds.

We believe this influence is due to direct interactions between word level processing and phoneme level processing. That is, as the word level converges on an interpretation of the input it provides support to the phoneme level units that are consistent with that interpretation.

One consequence of this feedback mechanism is that if the input is not the expected word, it should hinder phoneme recognition. Of course, we don't just hear what we expect to hear. So, the top-down feedback must be not strong enough to overrule the information from bottom-up acoustic input. But what would happen if the bottom-up information was partly consistent with the word level information? For example, consider two kinds of lexically inconsistent phonemes: a similar one (e.g., /k/ to /t/) and a dissimilar one (e.g., /k/ to /S/). Mirman, McClelland, & Holt (in prep) Behavioral Data

Mirman, McClelland, & Holt (in prep) Behavioral Data

Since /k/ and /t/ are similar, /k/ input will be partially consistent with /t/ and if there is top-down support for /t/ it might be hard to recognize the /k/. On the other hand, /k/ input is less consistent with /S/ so top-down support for /S/ will cause less delay in recognizing /k/. In other words, listeners should be slower to recognize phonemes when the lexical information specifies a similar phoneme (NNW) than when it specifies a dissimilar phoneme (DNW).

Here are some examples:

Target	NNW	DNW
/s/	diminiss	automatiss
/S/	apprentish	academish
/k/	deposik	blemik
/t/	arsenit	abolit

We have found that listeners are slower to detect phonemes that are lexically inconsistent when the target phoneme is similar to the lexically consistent phoneme than when it is not similar to the lexically consistent phoneme. That is, lexical feedback is able to delay phoneme recognition when the there is some support for the lexically consistent phoneme.

reference: Mirman, D., McClelland, J.L., and Holt, L.L. (2005). Computational and behavioral investigations of lexically induced delays in phoneme recognition. Journal of Memory and
Language, 52, 424-443.

Phonetic Context Effects
Lexical Context Effects
Audiovisual Effects
Spectrally Rotated Speach