We examined the effects of acoustic noise on the relative weightings of different segment types during lexical learning and lexical retrieval. In previous research, we showed that learners more often confused newly learned words disambiguated only by their vowels ("consonant-matched") than words disambiguated only by consonants ("vowelmatched"). This suggests that, at least within the English phonological system, consonants are more integral to lexical identity than vowels.
However, in natural listening situations segmental information is often obscured by various types of noise. In such situations, the learner/perceiver might benefit from the use of different, more reliable cues to word identity, rather than a fixed set of weights assigned to cues in low-noise conditions.
We designed an artificial lexicon consisting of 16 CVCV words, using 4 consonants and 4 vowels. Each C and V occurred equally often in each word position across the set. For each word (dabo), another word shared the same consonants (dubeI) and a third had the same vowels (gapo). Listeners learned these words as names for 16 black-and-white shapes. There were 384 exposure trials (24 per word); in each, a picture was presented, its name was spoken, and the participant clicked on the picture to proceed to the next trial. We then tested listeners in a 4AFC task, where 4 pictures were presented and the name of one of them was spoken. The participant was asked to click on that picture. On each trial, one picture had a name with the same vowels or same consonants as the target picture; the other two pictures had phonologically unrelated names. Learning and test stimuli were presented either without noise or embedded in white noise (3 dB SNR).
When words were learned and tested without noise, consonant-matched items were more often confused with one another than vowel-matched items, replicating our earlier findings. When learning and test items were embedded in noise, overall error rates did not increase compared to the no noise condition. However, the pattern of errors did change: vowel-matched items were much more often mistakenly selected, indicating that vowels were more robustly used to identify words in the noise condition than in the no noise condition. Thus, consonant and vowel information was weighted differently in the presence of noise. Our results support a mechanism of adaptive plasticity whereby acoustic/phonetic cues are re-weighted to maximize the intelligibility of words during lexical learning and lexical retrieval.