关键词: 语音感知, 统计学习, 感觉-运动联结, 婴儿

Abstract: During the first year of life, infants’ speech perception undergoes a remarkable reorganization: their attention shifts from discriminating broad, non-native phonetic contrasts to focusing primarily on distinctions within their native language. How do infants develop this specialized sensitivity? The acquisition of speech perception is theorized to rely on two complementary mechanisms: statistical learning and sensorimotor coupling. Statistical learning emphasizes infants’ sensitivity to the distributional regularities of speech input, whereas sensorimotor coupling highlights the modulatory role of articulatory motor feedback. The present study examined the contributions of these two mechanisms during the sensitive period to determine whether speech perception arises from a single mechanism or from multiple mechanisms.
The present study examined the contributions and boundaries of statistical learning and sensorimotor coupling in 80 eight-month-old Mandarin-learning infants using an alternating/non-alternating visual fixation paradigm across two experiments. Experiment 1 manipulated the distributional properties of the familiarization input to evaluate statistical learning. Infants were exposed to either a unimodal or bimodal frequency distribution of speech tokens and subsequently tested for their ability to discriminate the Hindi dental-retroflex contrast (/ɖ/-/d̪/). Experiment 2 examined the contribution of sensorimotor coupling by using a teething toy that either restricted or did not restrict tongue-tip movement, following the paradigm of Bruderer et al. (2015). No additional auditory familiarization was provided, allowing us to assess the contribution of motor feedback to speech perception in the absence of short-term familiarization. Both experiments utilized speech and non-speech stimuli. The speech stimuli comprised the Hindi dental-retroflex contrast (/ɖ/-/d̪/), while the non-speech control stimuli consisted of synthesized tones based on piano timbre, designed to approximate selected temporal and spectral properties of the speech stimuli.
Across all conditions, infants’ looking times decreased progressively with repeated stimulus presentations, confirming a habituation effect. (1) After exposure to the bimodal distribution, infants exhibited significantly longer looking times to alternating speech streams than to non-alternating speech streams, indicating successful speech discrimination. Conversely, those in the unimodal condition, showed no significant differences between alternating and non-alternating speech streams, suggesting the lack of effective speech discrimination. When the stimuli were non-speech sounds, neither the bimodal nor the unimodal distribution elicited discrimination behaviour. (2) The results of Experiment 2 revealed that under the unrestricted tongue-tip movement condition, infants demonstrated significantly longer looking times to alternating speech streams than to non-alternating ones, indicating speech discrimination. In contrast, no significant discrimination effect was observed when tongue-tip movement was restricted. For non-speech stimuli, neither motor condition elicited discrimination behaviour.
The findings suggest that statistical learning and sensorimotor coupling may jointly contribute to early speech perception, revealing a multi-mechanistic architecture in infant language acquisition. Under the current experimental conditions, the distributional learning effect was primarily observed for speech stimuli, whereas the sensorimotor manipulation showed a similar but less conclusive trend. This relative advantage for speech stimuli may reflect infants’ selective attentional orientation towards speech and the preferential activation of sensorimotor couplings by speech signals, though this interpretation remains preliminary and requires validation from future larger-sample studies. Moreover, the role of environmental input in speech acquisition appears to be multilevel: while the sensorimotor mechanism may support discrimination sensitivity without short-term exposure to the specific speech contrast, the statistical learning mechanism may further refine and stabilize this sensitivity through continuous exposure to the ambient language environment.

Key words: speech perception, statistical learning, sensorimotor coupling, infants