Search Abstracts | Symposia | Slide Sessions | Poster Sessions
The Neural Dynamics of Phonetic Feature Decoding in Developmental Dyslexia
Poster Session F, Friday, October 2, 2:45 - 4:45 pm, Wangari Maathai
Julia Schwarz1, Anastasia Klimovich-Gray2, Laura Gwilliams3, Marie Lallier1,4, Nicola Molinaro1,4; 1Basque Center on Cognition, Brain and Language, 2University of Aberdeen, 3Stanford University, 4Ikerbasque
Phonological deficits are the most commonly observed marker of developmental dyslexia, as evidenced by lower performance on phoneme awareness and letter-sound decoding, such as pseudoword reading (Ramus et al., 2013). Despite several decades of research on behavioural phonological deficits, we know very little about the neural dynamics of phonological processing in dyslexia. One possibility is that dyslexic listeners neurally decode phonetic-phonological information less accurately from speech, especially when other sources of information (e.g. semantic) are sparse. We tested this possibility using a machine learning model with binary classification of phonetic features, i.e. the articulatory and acoustic cues contributing to phonemic contrasts. Fourteen right-handed Spanish native adults with diagnosed dyslexia and 14 controls were matched on age and IQ. The dyslexic group scored significantly lower on word and pseudoword reading. Participants listened to 160 Spanish sentences while MEG was recorded. Half of the sentences were semantically predictable (i.e. highly constrained sentences with low semantic surprisal; M=2.5, SD=1.33), and half were semantically unpredictable (M=14.4, SD=2.34). Relative to phoneme onsets, we applied temporal decoding (logistic regression) of binary phonetic features (e.g., voiced/voiceless) to the MEG data to reveal subjects’ sensitivity to phonetic features over time. We selected International Phonetic Alphabet phonetic features pertaining to Castillian Spanish phonemes except those occurring in less than 10% of phonemes. The remaining features were all decodable above chance (cluster-based permutation; p<0.05) and averaged for further analysis. We compared groups with respect to decoding accuracy, duration, and topography (by decoding from each sensor individually); and correlated subjects’ individual pseudoword reading scores (accuracy/reading time*100) with their decoding accuracy at each time point. A two-tailed cluster-based permutation test revealed that dyslexic listeners decoded phonetic features less accurately than controls 232–297 ms after phoneme onset in semantically unpredictable sentences (Mean t = 8.44, p = .006). Topographic comparison suggested that this group difference was primarily driven by lower decoding in left temporal areas. Furthermore, a two-sample independent t-test showed that the decodable duration of phonetic features (i.e. number of time points where AUC > 0.5) was significantly shorter in the dyslexic group in both predictable (control: M=459ms, SD=94; dyslexic: M=362ms, SD=126; t(26)=2.23, p=.034) and unpredictable sentences (control: M=453ms, SD=64; dyslexic: M=351ms, SD=134; t(26)=2.49, p=.020). Finally, subjects’ pseudoword reading scores were significantly correlated with decoding accuracy in unpredictable sentences, peaking at 237 ms (Peak r = 0.7, fdr corrected p = .002). These results suggest that developmental dyslexia is associated with altered neural dynamics of phonetic feature processing during natural speech listening, characterised by reduced decoding accuracy and shorter sustained above-chance decoding. Jointly, the timing of the observed differences (232–297 ms), correlation with pseudoword reading scores, and localisation in left temporal areas provide converging evidence for a deficit in phonological computation, rather than purely acoustic encoding. However, decoding accuracy was only lower in semantically unpredictable sentences, which could suggest that dyslexic listeners can partially compensate for weaker phonetic processing with top-down semantic information, consistent with predictive coding frameworks of speech perception. This could explain why speech comprehension is relatively unaffected in dyslexia.
Topic Areas: Disorders: Developmental, Speech Perception