Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Abstract contrasts persist alongside acoustic sensitivity: An MEG study of fricative voicing in Dutch
Poster Session F, Friday, October 2, 2:45 - 4:45 pm, Wangari Maathai
Anna Mai1, Andrea Martin1,2; 1MPI Psycholinguistics, 2Donders Center for Cognitive Neuroimaging
What information does the brain track to understand speech? Are acoustic distinctions enough to facilitate word recognition, or do abstract sound categories (phonemes) also play a role? Previous work shows that both matter (Mai et al., 2024; Tezcan et al. 2023; DiLiberto et al. 2015), but since phonemes are often acoustically distinct, some argue that phoneme effects could be artifactual (Daube et al. 2019). This project uses cases where distinct phonemes have similar acoustics to disentangle abstract and acoustic processing. In Dutch, voiced fricatives (v, z, ɣ) are produced as voiceless (f, s, x) when they occur following a voiceless obstruent. This creates voiceless fricatives that are acoustically similar to other voiceless fricatives, but share an abstract phonemic category with many voiced fricatives. By comparing the neural response to devoiced fricatives against 1. the response to fricatives that share similar acoustics (f,s,x) and 2. the response to fricatives that share the same abstract phonemic category (v,z,ɣ), this study disentangles the contributions of acoustics and abstract categories in the neural response to speech. Twenty-seven participants listened to 49 minutes of a Dutch speaker reading fables while their brain responses were recorded using 275-channel magnetoencephalography. While listening, each participant heard over 250 tokens of each fricative (v,z,ɣ,f,s,x). Neural epochs were defined for all fricatives as the 500ms following phone onset with a 100ms baseline prior to phone onset. A sliding-window one-way ANOVA comparing the band power of voiced, voiceless, and devoiced phone epochs was used to determine which sensors capture a phonologically sensitive vs. acoustically sensitive response. Comparisons were done separately for each power band (delta, theta, alpha, beta, gamma) and each place of articulation (labiodental, alveolar, velar). Phonologically-sensitive sites were defined as those for which there existed a 100ms window in which the neural response to voiced and voiceless sounds was significantly different, the response to voiceless and devoiced sounds was significantly different, and the response to devoiced and voiced sounds was NOT significantly different. Acoustically sensitive sites were those for which the response to voiced and voiceless sounds was significantly different, the response to voiced and devoiced sounds was significantly different, and the response to voiceless and devoiced sounds was NOT significantly different. For each power band and place of articulation, both more phonological sites and more acoustic sites were observed than predicted to occur by chance. To establish this, the sliding-window one-way ANOVA analysis was performed for 1000 random pairs of sounds. While an average of 807 phonological sites and 794 acoustic sites were observed across all participants for the true fricative devoicing comparisons, only 280 phonological sites and 180 acoustic sites were observed on average across all random comparisons. The observed values lie more than 3 standard deviations beyond the mean of the random distribution. From these results, we conclude that both acoustic and phonological contrasts are maintained during natural speech listening. Future work will localize the source vertices most responsible for these effects and assess the role of acoustic variability in the trial-by-trial neural response.
Topic Areas: Phonology, Speech Perception