Rapid learning of a phonemic discrimination in the first hours of life

Human neonates possess remarkable linguistic sensitivity, exhibiting preferences for speech sounds over non-linguistic sounds and even showing a preference for their mother's voice. A crucial aspect of this linguistic ability is phoneme discrimination, the ability to distinguish the smallest units of sound in speech. While it's generally accepted that neonates can discriminate phonemes at birth, the neural mechanisms underlying this ability and the extent of postnatal plasticity remain unclear. This study aimed to investigate the neural dynamics of postnatal phonological learning immediately after birth, focusing on the discrimination of forward versus backward vowels – a contrast unlikely to have been learned prenatally. Understanding this rapid learning process is crucial for elucidating the foundations of language development and identifying potential risks for neurodevelopmental disorders.

Previous research has demonstrated neonates' ability to discriminate phonemes across various languages, even showing sensitivity to both native and non-native vowels. Studies measuring sucking rates revealed that infants increased sucking amplitude when exposed to non-native vowels, indicating both innate sensitivity and the influence of prenatal learning. Evidence suggests that the auditory system becomes functional as early as 24 weeks of gestation, allowing for in utero exposure to shape auditory perceptual representations, including speech representations. A seminal study by Cheour et al. (2002) showed that exposure to speech sounds can affect the neural dynamics associated with phoneme discrimination immediately after birth, using mismatch negativity (MMN) as a measure. However, the neural mechanisms and dynamics associated with postnatal phonological learning immediately after birth remained poorly understood, prompting the current investigation using fNIRS, a method offering relatively high spatial resolution and tolerance for infant movement.

Seventy-five healthy full-term neonates were randomly assigned to three groups: an experimental group, an active control group, and a passive control group. fNIRS data were collected at three time points: baseline (T0), 5 hours after birth (T1), and 2 hours after T1 (T2). The experimental group underwent 5 hours of training with forward and backward vowels of Mandarin Chinese. The active control group received similar training but with different vowels. The passive control group received no specific stimulation. Stimuli consisted of six native Mandarin vowels and their reversed counterparts, carefully controlled for prosodic variation. fNIRS data were collected using a cap with 20 emitters and 16 detectors, covering temporal and frontal regions. Data were recorded at 10 Hz and pre-processed to remove artifacts and convert optical intensity data into changes in oxyhemoglobin (Δ[HbO]) and deoxyhemoglobin (Δ[Hb]) concentrations. Linear mixed effects regression analysis was used to model Δ[HbO] mean amplitude and peak latency as functions of stimulus type, participant group, and test phase. Resting-state functional connectivity analysis was performed to examine changes in neural synchronization between brain regions.

Analysis of oxyhemoglobin concentration amplitude revealed a significant three-way interaction between stimulus type, group contrast (active control vs. experimental), and test phase contrast (T1 vs. T2). The experimental group showed greater differences in [HbO] between forward and backward vowels at T2 compared to the active control group at T1. This effect was maximal over the superior temporal (ST) and supramarginal (SM) regions bilaterally and the left inferior parietal (IP) region. Analysis of oxyhemoglobin concentration peak latency showed a significant three-way interaction between stimulus type, group contrast, and test phase contrast (T0 vs. mean (T1,T2)). The experimental group showed greater differences in [HbO] peak latencies between forward and backward vowels at T1 and T2 compared to the active control group at T0. This effect was maximal over bilateral inferior frontal (IF) regions. Resting-state functional connectivity analysis revealed a significant two-way interaction between group contrast (passive control vs. mean of active control and experimental groups) and test phase contrast (T1 vs. T2). Both experimental and active control groups showed stronger increases in connectivity after T1 (post-sleep), particularly between IF and ST regions.

The findings indicate that neonates exhibit rapid learning of phonemic discrimination within the first hours of life. The experimental group's superior performance in distinguishing forward from backward vowels after 5 hours of exposure, contrasted with the active control group's lack of such changes, suggests that the observed effects reflect specific vowel acquisition rather than general exposure to speech sounds. The involvement of ST, SM, and IP regions in amplitude changes and IF regions in latency changes points to a network reminiscent of the mirror neuron system, connecting speech perception and production. The increased functional connectivity between IF and ST regions after a period of sleep further supports the role of sleep in consolidating these neuroplastic changes. These findings align with theories proposing that innate neural connectivity facilitates rapid sensorimotor learning of speech from birth.

This study demonstrates the remarkable capacity of human neonates for ultra-fast tuning to natural phonemes, highlighting the early development of a speech acquisition network involving IF, ST, SM, and IP regions. Future research should further explore the role of this network in sensorimotor learning and perceptual narrowing, and its potential use in identifying newborns at risk for neurodevelopmental disorders.

The study used a relatively small sample size and a limited set of vowel sounds. The use of a single token for each vowel, potentially minimizing prosodic variation, may have influenced the results. Generalizability to other languages and phonemic contrasts requires further investigation. Finally, while polysomnography data indicated that neonates were mostly asleep during consolidation, the exact sleep stages were not comprehensively analysed.