Presentation
Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Lightning Talks
Neural Prediction of Early Language Development is Language Independent: Evidence from Neural Speech Encoding
Poster E91 in Poster Session E, Thursday, October 26, 10:15 am - 12:00 pm CEST, Espace Vieux-Port
This poster is part of the Sandbox Series.
Shaoqi PAN1, Nikolay Novitskiy1, Ching Man Lai1, Gangyi Feng1, Patrick CM Wong1; 1Brain and Mind Institute, The Chinese University of Hong Kong, Hong Kong SAR, China
Early development of neural speech encoding, as reflected by the early-latency EEG responses (e.g., the frequency following responses [FFR]), has been found to be independent of the nativeness of the speech signal (Novitskiy et al., 2022). This differs from long-latency cortical responses which are sensitive to the infant’s native language experiences starting from mid-infancy (Kuhl, 2010). These findings support a model that postulates a two-stage developmental process of speech, with the two stages corresponding to early- and long-latency neural responses. While infants’ subcortical structures contribute to the faithful neural encoding of the incoming speech signal regardless of its native status, cortical structures are developed to attend to native speech. The current study seeks to obtain converging evidence for this model by examining whether neural speech encoding predicts early language development regardless of whether the speech stimuli are native or non-native to the child’s immediate language environment. To test this hypothesis, we recruited 300 medically healthy infants (157 males) from Cantonese-speaking families and evaluated their speech encoding of Cantonese tones 2 and 4 (native) and Mandarin tone 3 (non-native) using EEG. We quantified neuronal phase-locking to the speech stimulus frequencies by extracting FFRs from the EEG. As the fundamental frequencies of all tones were above 120 Hz, the neural generator of the FFRs was likely (though not exclusively) below the cortex (Bidelman, 2018). Each infant underwent an EEG procedure while naturally sleeping (0.75–24 months, M=5.67 SD=4.41). The MacArthur-Bates Communicative Development Inventories (MCDI)-Cantonese version was used to evaluate language and communication outcomes up to 26 months (0–26.86 months, M=12.06, SD=6.88) after EEG testing. Random Forest with out-of-bag validation was used to construct a binary prediction model whereas children were classified as below or above 1 SD (~16th percentile) of the mean on the MCDI (n=300). Neural predictive models were constructed separately for Cantonese (native) and Mandarin (non-native) stimuli. An external validation procedure was also implemented by dividing the sample into separate sub-samples (n=204, 102 children acted as the unseen data). FFR encoding accurately predicted future language outcomes. The out-of-bag validation and validation with unseen data had sensitivity values of at least 0.8. Other model performance indexes such as diagnostic accuracy and area under the curve are similarly high. The sensitivity of the neural predictive models constructed using the native stimuli of tone 2 and tone 4 were .83±0.018 and .86±0.018 respectively, and .84±0.015 for the non-native Mandarin tone 3. The predictive performance of these models did not significantly differ from each other. (tone 2/4 vs tone 3, all p values of performance indexes were > 0.05). These results support our hypothesis that FFR encoding of speech predicts early language development independent of the languages that these infants learned and perceived. Our findings speak to a potentially auditory-general rather than a language-dependent speech encoding process of early spoken language development. Future research will examine the contribution of cortical processes in the neural prediction of language development as well as whether these findings can generalize to non-tone-learning infants.
Topic Areas: Language Development/Acquisition, Computational Approaches