Search Abstracts | Symposia | Slide Sessions | Poster Sessions
The spatio-temporal dynamics of phonetic encoding in aging and aphasia
Poster Session D, Saturday, October 26, 10:30 am - 12:00 pm, Great Hall 3 and 4
Jill Kries1, Pieter De Clercq2, Maaike Vandermosten2, Laura Gwilliams1; 1Stanford University, 2KU Leuven
Comprehending natural speech entails rapidly converting continuous acoustic input into discrete units, such as phonemes and words. In younger adults, this process entails neural encoding of phonetic features, whereby different neural ensembles are recruited across time – termed “Hierarchical Dynamic Coding”. The duration of phonetic encoding has been linked to resolving lexical uncertainty, and the speed of encoding dynamics with the speed of incoming inputs. Given that disorders like post-stroke aphasia can affect phoneme identification and decrease neural encoding of acoustic features and phoneme onsets, it is important to establish whether difficulty in speech comprehension arises from distorted dynamics of phonetic encoding. Here, we tested (1) whether dynamic encoding of phonetics can be reproduced with EEG, using a relatively short recording duration (~25 mins), in healthy older adults, and (2) whether the dynamics of phonetic encoding in aphasia differs from healthy older adults. 39 individuals with aphasia in the chronic phase after stroke (> 6 months after onset; with left-hemispheric or bilateral lesions) and 24 age-matched healthy control participants listened to a narrative for 25 minutes while EEG data was recorded. The narrative was annotated for phonetic features, allowing time-resolved decoding of these features from the EEG signals. The decoding was performed using one-versus-all logistic regression and 5-fold cross-validation for 18 phonetic features, including manner, place, voicing, roundness, and front-back-ness of consonants and vowels. Temporal generalization was implemented by testing how a classifier trained on time t performs on data from subsequent and preceding timepoints. Phonetic features were decodable above chance in healthy older adults from -0.04 to 0.49 seconds relative to phoneme onset (p<.001). Comparing healthy older adults to individuals with aphasia, we observed significantly lower decoding accuracy from 0.09 to 0.26 seconds after phoneme onset (p<.001), indicating a shorter duration of phonetic encoding. No significant group difference was found in the temporal generalization, with an average generalization time of 139ms in the aphasia group and 132ms in the control group. Spatially, we identified a cluster of 22 EEG sensors (p<.001) over which individuals with aphasia had lower decoding accuracy of phonetic information than healthy controls, particularly over the left auditory cortex and posterior sensors bilaterally. A potential explanation is that this is related to structural and functional changes following stroke. In conclusion, we replicate robust phonetic encoding in older healthy adults using 25 minutes of EEG data. This suggests that the analytical approach is robust, and can be applied to relatively short recording durations to study different populations. In comparing healthy controls and individuals with aphasia, we find that the primary marker of the language disorder lies in the shorter duration of phonetic encoding, which spatially localized to left-hemispheric auditory sensors. The neural pattern of phonetic encoding evolved at the same rate for both groups, as evidenced by the generalization analysis, not providing evidence for distinct dynamics of information evolution in aphasia. Overall, our study suggests that speech comprehension challenges in aphasia may be related to difficulties in maintaining lower-order phonetic information long enough to facilitate lexical recognition.
Topic Areas: Phonology, Disorders: Acquired