Slide Slam O14
Extracting phonetic features from natural classes: A mismatch negativity study of Mandarin Chinese retroflex consonants
Zhanao Fu1, Philip J. Monahan1,2; 1University of Toronto, 2University of Toronto Scarborough
The brain codes phonetic features and categories (Mesgarani et al., 2014; Hestvik & Durvasula, 2016; Khalighinejad et al., 2017; Schluter et al., 2017). What is unknown is whether the brain rep-resents phonetic feature classes. Mandarin Chinese is an excellent test case. Mandarin has a rich inventory of retroflex consonants that span multiple manners of articulation: fricatives, affricates, approximants (Lee & Zee, 2003). A powerful tool for understanding how the brain represents speech sounds is the mismatch negativity (MMN; Näätänen, 2001). The MMN is a component in the event-related potential (ERP) that is elicited by a deviant stimulus embedded within a series of repeating standard stimuli. The MMN is typically concentrated over fronto-central electrode sites and peaks between 150–350ms (Näätänen, 2007). In all previous designs, however, any variation present in the standards was intra-category: All standards belong to the same phonetic category (Aulanko et al., 1993; Phillips et al., 2000; Scharinger et al., 2016). To test whether the brain repre-sents feature classes, we introduced inter-category variation in the standards. We observed an MMN only when the retroflex consonants were the standards, suggesting that Mandarin listeners extract the feature [retroflex]. Methods: Twenty-six native speakers of Mandarin Chinese participated in a many-to-one oddball auditory MMN paradigm using electroencephalography (EEG). All syllables were [Cɤː4]. The consonants were either the Mandarin retroflex consonants, i.e., [ʂ tʂ tʂʰ ɻ], or their non-retroflex counterparts, i.e., [s ts tsʰ l]. There were two blocks. Standard retroflex block: standard consonants share feature [retroflex], non-retroflex deviants (underlined), i.e., [… ʂɤ ɻɤ ʈʂɤ ʂɤ tsɤ tʂʰɤ ʂɤ …]. Non-retroflex standard block: non-retroflex standards, retroflex deviants, i.e., [… sɤ tsɤ lɤ tshɤ lɤ ʂɤ tsɤ …]. Participants passively listened to stimuli. We measured the difference in brain responses between the standard and deviant in each block. Permutation tests were used (pFDR < 0.05) to assess statistical differences. Results: We observed an MMN to the non-retroflex deviant embedded in the standard retroflex block. The MMN peaked approximately 300ms post-stimulus onset over fonto-central electrode sites. This suggests that listeners extract [retroflex] despite significant inter-category variation. In the non-retroflex standard block, no MMN was observed. This asymmetric MMN is consistent with previous findings that support underspecified feature representations (Cornell et al. 2013; Eu-litz & Lahiri, 2004; Hestvik & Durvasula, 2016; Scharinger et al. 2016). To determine whether these effects were driven by phonetic properties of the stimuli, we also calculated the identity MMN (iMMN; Pulvermüller & Shtyrov, 2006), which compared the retroflex consonants when they were the standards to when they were deviants. An iMMN to the non-retroflex consonants was observed over centro-parieto-occipital electrode sites instead of fronto-central electrodes. This shift in topography potentially reflects increased difficulty posed by the added variation in the standards. Discussion: Overall, we observed an MMN only when the standards were retroflex: Mandarin listeners grouped the retroflex sounds together and extracted the feature [retroflex]. The brain en-codes abstract phonetic feature classes, despite significant acoustic and phonological inter-category variation, so long as a shared feature binds them together.