2025 – 2028
nfants acquire language on the basis of an acoustic signal – the sounds of speech – which is complex and uninterrupted. We do not pause (remain silent) between each word, and yet a young child aged 2 already understands around a hundred words. Young children must therefore learn to segment this continuous stream of speech in order to extract meaningful units – words. Among the mechanisms involved in this segmentation is statistical learning, which enables them to detect patterns in language. This statistical learning is made possible in particular by the probabilities of transition between syllables. For example, in French, the syllables ‘ma’ and ‘man’ occur more frequently together than ‘ta-ma’ or ‘ma-ma’, which might appear in a sentence such as ‘ta maman’ or ‘ma maman’. We know that from birth, very young children are sensitive to these statistical regularities between syllables; however, the mechanisms that enable them to extract these patterns are still poorly understood. This learning process could, for example, be compromised by a deterioration in the acoustic signal. In this new project, we aim to gain a better understanding of the role of acoustic information in speech within this statistical learning of words. Are 10-month-old infants able to segment speech under different acoustic conditions, and in particular, when certain acoustic features of the voice have been degraded? It seems essential to investigate whether this learning process is impaired under these poor acoustic conditions, as they correspond to the sound conditions experienced by hearing-impaired infants and those with cochlear implants in their everyday lives.
