Search Abstracts | Symposia | Slide Sessions | Poster Sessions
The developmental trajectory of word-onset encoding in running speech
Poster Session A - Sandbox Series, Thursday, October 24, 10:00 - 11:30 am, Great Hall 3 and 4
This poster is part of the Sandbox Series.
Zhen Zeng1, Patrick Wong1, Xiangbing Teng1; 1The Chinese University of Hong Kong
Decades of developmental research established that infants’ sensitivity to native speech increase rapidly in the first year of life. They can track speech envelope from birth (Barajas et al., 2021) and are able to categorize phonemes on place and manner of articulation by 3 months (Gennari et al., 2021). Yet when and how infants learn to segment words remains an open question. Longitudinal evidence, using Mis-match Responses (MMR) (Werwach et al., 2022) and temporal response functions (TRFs) (Di Liberto et al., 2023), has demonstrated that the EEG amplitudes indexing discrimination and encoding of acoustic and linguistic features undergo non-linear changes in the first year of life. Specifically, Di Liberto et al. (2023), using repeated nursery rhymes, has unveiled that while infants’ neural encoding of linguistic features grows linearly from 4 months onwards, their encoding of spectral information demonstrates non-linear growth that peaked at 4 months. This process may be largely driven by decreasing delta power from 4-6 months due to maturation (Chu et al., 2014). Following the latest development in measuring the neural encoding of running speech, the current study investigates infants’ word onset tracking in running speech. We hypothesized that word onset encoding improves in earlier months before decrease. The temporal dynamics of word onset encoding may involve both acoustics and higher-level processing (Karunathilake et al., 2023). Here we report data from infants aged newborns (N = 17), 1-3 months (N = 25), 6-8 months (N = 30), as well as adults (N = 30) who underwent ~20-30 mins EEG recording. The stimuli consist of 5 children stories in adult-directed speech, recorded from a female native Cantonese speaker. The stories were then cut into 8 2-3-min-long chunks, with random order of presentation for each participant. EEG data were filtered at 0.5-30Hz for preprocessing. We calculated the word onset TRFs (Crosse et al., 2016) in a -100-800ms sliding time window. Results showed that the word onset TRFs unfolds with a single positive peak around 400-500ms for newborns, 300-400ms for 1-3 month-olds, 250-350ms for 6-8 month-olds, suggesting a single stage acoustics processing and decreasing peak latency for infants. Permutation test showed significant encoding for all four groups. One-way ANOVA comparing the positive peak amplitudes for the three infant groups revealed a significant effect, and pairwise comparison showed that peak amplitudes were the smallest in newborns and largest in 1-3 months. The non-linear amplitude changes may involve maturational changes non-specific to word onset processing (Chu et al., 2014). In contrast, the amplitudes in adults showed dramatic qualitative difference to the infant groups: positive peaks around 100ms and around 250ms as well as a negative peak around 400-600ms, echoing separate processing stages: early acoustics processing and other higher level linguistic processing (Karunathilake et al., 2023). Overall, our findings showed that infants are able to track word onsets in running speech from birth. Participants’ EEG peak latency decreased with age. In contrast, amplitudes showed non-linear changes with age, suggesting that the increase of amplitudes may not directly reflect better neural encoding for young infants.
Topic Areas: Language Development/Acquisition, Computational Approaches