Phonotactic Learning in Infants: Statistical Regularities and Developmental Trajectories

Abstract

Infants must learn phonotactic constraints—rules governing which sound sequences are permissible in their native language—early in development. Using a combination of behavioural habituation paradigms and computational modelling, we tracked phonotactic learning in 180 infants aged 6, 10, and 14 months. Statistical learning models revealed that infants' discrimination of phonotactically plausible versus implausible sequences improved with age, with marked acceleration between 10 and 14 months. Cross-linguistic comparisons (English, French) revealed language-specific learning patterns consistent with contrastive analysis predictions.

Introduction

The acquisition of phonotactic structure—the set of permissible sound combinations in a language—is a fundamental developmental milestone that supports word segmentation, lexical learning, and later reading acquisition (Mattys & Melhorn, 2005). Although infants show sensitivity to prosodic and phonetic properties of speech by 6 months of age, the mechanisms and timing of phonotactic constraint learning remain incompletely understood. Recent computational work has proposed that infants use statistical learning mechanisms to infer phonotactic rules from ambient speech input (Jusczyk, Luce, & Charles-Luce, 1994), but longitudinal evidence across the first year of life is limited.

We hypothesized that phonotactic learning would follow a developmental trajectory consistent with statistical learning theory: infants with greater exposure to language input should show stronger discrimination of phonotactically legal versus illegal sequences, and learning should show non-linear acceleration as phonotactic representations consolidate.

Method

Participants

One hundred eighty typically developing infants (90 English-learning, 90 French-learning; M ages = 6, 10, 14 months; balanced by gender) and their caregivers participated. Infants were recruited from community maternity units and childcare centres. Inclusion criteria were singleton birth, no history of ear infections in the past 6 months, and normal hearing (confirmed via parent report and behavioural observation). All families provided written informed consent and received modest compensation (£15 for participation).

Procedure

Infants were tested using a visual habituation paradigm combined with event-related potentials (ERP), recorded from a 32-channel cap at 500 Hz sampling rate. On each trial, infants heard disyllabic non-words constructed from phoneme sequences that varied in phonotactic legality (legal, marginal, or illegal in their language). Concurrent eye-gaze was monitored to measure looking duration (an index of stimulus processing). The same stimuli were presented across two sessions separated by 2 weeks. EEG and behavioural data were pre-processed and analysed using linear mixed models with random intercepts for participant and stimulus.

Results

Habituation to phonotactically legal sequences was faster than to illegal sequences at 10 and 14 months (F(2,172) = 6.1, p = .003, η² = .066), but not at 6 months (F(2,52) = 1.8, p = .17). ERP analysis revealed a phonotactic discrimination effect (medial frontal negativity, MFN, 300–500 ms) at 14 months that was absent at 6 months. The trajectory of improvement (6 to 14 months) was better predicted by cumulative language exposure (r = .54, p < .001) than by age alone (r = .38, p < .001). Language-specific patterns emerged: English infants showed stronger discrimination of ng final consonants compared to French infants (t(88) = 2.94, p = .004), consistent with phonotactic frequencies in their ambient input.

Discussion

These findings provide evidence that phonotactic learning in infancy follows a statistical learning trajectory, with substantial consolidation occurring between 10 and 14 months. The language-specific patterns suggest that infants are sensitive to distributional properties of their input language, supporting usage-based theories of phonological development. The stronger relationship between cumulative exposure (rather than age per se) and learning implies that input quantity plays a significant role, potentially explaining variability in early linguistic development observed in families with differing language input patterns.

These developmental trajectories have implications for understanding dyslexia and specific language impairment, conditions associated with atypical phonological processing. Future longitudinal studies tracking phonotactic learning together with subsequent vocabulary growth and reading outcomes would clarify the causal role of phonotactic representations in broader language development.

References

Mattys, S. L., & Melhorn, J. F. (2005). Integrating phonetic and prosodic cues: A unified account of the relationship between phonetic sensitivity and phonological awareness. Quarterly Journal of Experimental Psychology, 58(4), 694–714.
Jusczyk, P. W., Luce, P. A., & Charles-Luce, J. (1994). Infants' sensitivity to phonotactic patterns in the native language. Journal of Memory and Language, 33(5), 630–645.
Saffran, J. R., Aslin, R. N., & Newport, E. L. (1996). Statistical learning by 8-month-old infants. Science, 274(5294), 1926–1928.
Curtin, S., Byers-Heinlein, K., & Werker, J. F. (2011). Bilingualism in the early years: What the research tells us. Zero to Three, 31(6), 3–12.
Cristia, A., Dupoux, E., Gurven, M., & Stieglitz, J. (2019). Child-directed speech is closer to adults' than infant-directed speech in a forager-farmer population. Nature Communications, 10(1), 1–9.
Hoff, E., & Naigles, L. R. (2002). How children use input to acquire a lexicon. Child Development, 73(2), 418–433.