Sleeping neonates track transitional probabilities in speech but only retain the first syllable of words

Abstract

Extracting statistical regularities from the environment is a primary learning mechanism that might support language acquisition. While it has been shown that infants are sensitive to transition probabilities between syllables in speech, it is still not known what information they encode. Here we used electrophysiology to study how full-term neonates process an artificial language constructed by randomly concatenating four pseudo-words and what information they retain after a few minutes of exposure. Neural entrainment served as a marker of the regularities the brain was tracking during learning. Then in a post-learning phase, evoked-related potentials (ERP) to different triplets explored which information was retained. After two minutes of familiarization with the artificial language, neural entrainment at the word rate emerged, demonstrating rapid learning of the regularities. ERPs in the test phase significantly differed between triplets starting or not with the correct first syllables, but no difference was associated with subsequent violations in transition probabilities. Thus, our results revealed a two-step learning process: neonates segmented the stream based on its statistical regularities, but memory encoding targeted during the word recognition phase entangled the ordinal position of the syllables but was still incomplete at that age.

The power of rhythms: how steady-state evoked responses reveal early neurocognitive development

“…Beyond low-level auditory function, a few studies additionally investigated infants’ auditory statistical learning abilities using SS-EP ( Choi et al, 2020 ; Fló et al, 2022 ; Kabdebon et al, 2015 ), extending earlier adult studies. In these studies, participants were presented with a continuous stream of syllables presented at a fixed rate.…”

“…3A ). They can nonetheless be corrected by some estimate of the background noise, yielding a signal-to-noise ratio ( Fló et al, 2022 ).…”

Humans parsimoniously represent auditory sequences by pruning and completing the underlying network structure

“…Statistical learning has been proposed as a powerful general learning mechanism that might be particularly useful in language acquisition in order to extract words from the speech stream ( Saffran et al, 1996 ). However, the exact model explaining statistical learning remains under-specified: What is computed remains unclear ( Henin et al, 2021 ; Fló et al, 2022 ) and authors often tailored the computation to suit the paradigm (Transition probabilities in some studies, non-adjacent or backward transition probabilities in others, biased transitions probabilities in network studies…). We argue that the Free Energy Minimization Model (FEMM) is a more general model that, beyond explaining community separation, as shown above, can also account for results traditionally explained by the computation of local transitional probabilities and those that require the computation of long distance dependencies.…”

“…At first, the extraction of local regularities in auditory streams was proposed as a major mechanism to structure the input, available from an early age since Saffran et al (1996 ) showed that 8-month-old infants could use transition probabilities - P(A|B) - between syllables to extract words from a monotonous stream with no other available cues. Since then, the sensitivity of humans to local dependencies has been robustly demonstrated in the auditory and visual domain ( Fiser and Aslin, 2002 ), without the focus of attention ( Batterink and Paller, 2019 ; Benjamin et al, 2021 ; Batterink and Choi, 2021 ) and even in asleep neonates ( Fló et al, 2022 ; Benjamin et al, 2022 ). Moreover, it is not limited to adjacent elements but can be extended to non-adjacent syllables - P(A|XC) - that could account for non-adjacent dependencies in language ( Peña et al, 2002 ).…”

At Which Low Amplitude Modulated Frequency Do Infants Best Entrain? A Frequency Tagging Study

“…Saffran et al, (1999) demonstrated the first evidence of infant statistical learning of tone sequences in both infants and adults. Further studies have demonstrated that infants under the age of 12 months can use statistical learning to pick out words in regular computerised speech (Choi et al, 2020; Fló et al, 2022; Kabdebon et al, 2015), while cortical tracking studies have demonstrated that infants track sung speech (Attaheri et al, 2022), cartoons (Jessen et al, 2019) and infant directed and adult directed speech (Kalashnikova et al, 2018).…”