You are viewing the SNL 2017 Archive Website. For the latest information, see the Current Website.

Poster D59, Thursday, November 9, 6:15 – 7:30 pm, Harborview and Loch Raven Ballrooms

Mice can learn phonetic categories.

Michael Wehr1, Jonny Saunders1;1University of Oregon

How does the auditory system produce invariant perception of a phoneme despite highly variable acoustic information? Because phonemes cannot be identified by any invariant acoustic features, the brain must make use of multiple imperfect cues. Competing auditory speech perception theories for how this problem might be solved remain largely unconstrained by neurobiological data. One reason for this is that most current methods for measuring human brain activity lack the necessary spatiotemporal resolution: speech is too fast and the neural circuitry involved is too small. A growing array of genetic tools for probing neural circuit function are available in the mouse. Although early speech theorists proposed that "speech is special," today many argue that early stages of speech processing in the human brain are likely implemented by evolutionarily-conserved auditory processing mechanisms found in any mammalian auditory system. Here we demonstrate that mice are capable of learning generalizable phonetic categories, and are thus capable of serving as a model system for phonetic perception. A mouse model for phoneme categorization has the potential to powerfully augment the understanding already gained from non-human animal models of speech perception. Here we focus on an important test case, the categorization of consonants in different vowel contexts. We trained mice to discriminate between consonant-vowel pairs beginning with either /g/ or /b/. Mice performed a 2-alternative choice task for a water reward, progressing through the introduction of additional talkers (1 male, 1 female), 2 tokens, and 3 vowel contexts. In a final generalization stage, they were challenged on 20% of trials with 3 novel talkers, 2 novel tokens, and 3 novel vowels. Mice (n=17) progressed to the generalization phase in an average of 15.6 weeks. Mice demonstrated the ability to generalize consonant identity across novel vowel contexts and talkers, consistent with true category learning. Categorization accuracy for all generalization types was significantly greater than chance, indicating that mice learned to discriminate between the initial consonants themselves rather than memorizing the individual training tokens. Different training sets of consonant-vowel tokens from different talkers induced divergent but consistent patterns of novel token categorization, suggesting that the mice learned distinct phonetic ’templates’ or discrimination boundaries depending on their precise history of language exposure. These results suggest that mice are a suitable model for studying the neural mechanisms of phonetic categorization.

Topic Area: Perception: Speech Perception and Audiovisual Integration

Back to Poster Schedule