Presentation
Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Lightning Talks
Time-course of neural computations supporting perception and misperception of degraded speech
There is a Poster PDF for this presentation, but you must be a current member or registered to attend SNL 2023 to view it. Please go to your Account Home page to register.
Poster E88 in Poster Session E, Thursday, October 26, 10:15 am - 12:00 pm CEST, Espace Vieux-Port
Máté Aller1, Connor Doyle1, Matthew H. Davis1; 1MRC Cognition and Brain Sciences Unit, Unvieristy of Cambridge
Humans utilize prior expectations to comprehend speech, but overreliance on these expectations can induce perceptual illusions if they mismatch with incoming sensory information. These misperceptions are especially likely in noisy acoustic environments or when there is partial overlap between expectations and sensory signals, which explains why misheard song lyrics are so common. Perceptual misperceptions may be explained by different theories: sharpening schemes in which neural representations matching prior knowledge are enhanced, and prediction error theories in which neural representations encode the difference between prior knowledge and sensory signals (Aitchison & Lengyel 2017). A previous fMRI study supported representations of prediction errors during perception of degraded speech (Blank et al. 2018). Our study aims to extend these findings using MEG/EEG to understand the time course of these computations during speech perception. Presentations of written words were used to manipulate prior knowledge while behavioural and neural (MEG/EEG) responses to noise-vocoded spoken words were recorded from 29 normal hearing listeners. The stimuli consisted of 32 monosyllabic words combined into four conditions: match (written “kip”, spoken “kip”), total mismatch (written “kip”, spoken “bath”), onset partial mismatch (written “kip”, spoken “pip”) and offset partial mismatch (written “kip”, spoken “kick”). Each trial started with the presentation of the written word for 500 ms, followed by a noise-vocoded spoken word. Participants indicated after each trial weather they perceived the written and spoken words as the same or different, as well as their confidence in their judgement (i.e., “definitely same”, "possibly same", "possibly different", "definitely different"). For the analyses presented below, responses were pooled over confidence levels within “same” and “different” judgements. As expected, participants correctly perceived word pairs in the match condition as “same” (P(same) = 0.909) and pairs in the total mismatch condition as “different” (P(same) = 0.005). However, perception in partial mismatch trials was more variable, with participants displaying frequent misperceptions (P(same) = 0.468 and 0.254 for onset and offset mismatch, respectively). Furthermore, the rate of misperceptions of particular partial mismatch word pairs (e.g., kit-pit) were better predicted by other word pairs sharing the deviating sounds (i.e., kip-pip, kitsch-pitch, kick-pick) than pairs sharing the matching sounds (i.e., pit-kit, wit-writ and writ-wit), pointing towards a prediction error mechanism (Blank et al. 2018). Preliminary MEG/EEG analyses in sensor space replicated previous findings that matching compared to totally mismatching text is associated with a reduction of evoked responses for magnetometers and gradiometers and an enhancement for EEG sensors (Sohoglu et al. 2012). Further univariate analyses in source space and multivariate representational similarity analyses are ongoing with the latter enabling specific conclusions concerning the timing, oscillatory correlates and perceptual contribution of sharpening and prediction error computations.
Topic Areas: Speech Perception,