Infants' vocalizations analyzed with an articulatory model: A preliminary report

Articulatory exploration enables the infant to discover abilities of the vocal tract and learn relationships between movements and percepts. However, neither direct measurements nor transcription methods have access to tongue configurations in pre-linguistic infant vocalizations. A statistical articulatory–acoustic model integrating the non-linear growth of the human vocal tract was used to describe infant behavior before and at the beginning of canonical babbling. Analyses were developed to assess from a set of (F1, F2) formant frequencies reported at 4 and 7 months in two separate corpora the most likely articulatory degrees of freedom of the model. Results indicate that exploration in the 4-month corpus is centered around a neutral configuration. It involves at least three articulatory parameters, including at least one for the tongue. The jaw seems to play a minor role in this exploration. In contrast, in the 7-month corpus, the exploration range increases: in this case the jaw plays a dominant role, leading to a large exploitation of the open–close contrast and associated F1 diversification in formant space. The simulation of co-occurrences between closants and vocants from the 7month corpus in the framework of the Frame-Content theory provides a portrait largely consistent with previously reported experimental data. Locus scatter-plots were also simulated and compared to available data on development of coarticulation in CV syllables. This kind of analysis could be applied to corpora of infants’ vocalizations at various ages to understand the development of speech production in relation to the growth of the human vocal tract. r 2006 Elsevier Ltd. All rights reserved.