Proceedings of the Annual Meeting of the Cognitive Science Society

Infants learn language by exposure to streams of speech pro-duced by their caregivers. Early on, they manage to segmentword forms out of this continuous input, which is either di-rectly addressed to them, or directed to other adults, thus over-heard. It has been suggested that infant-directed speech is sim-plified and could facilitate language learning. This study aimedto investigate whether features such as utterance length, seg-mentation entropy and lexical diversity could account for anadvantage in segmentability of infant-directed speech. A largeset of word segmentation algorithms was used on an ecolog-ically valid corpus, consisting of 18 sets of recordings gath-ered from French-learning infants aged 3-48 months. A se-ries of textual analyses confirmed several simplicity featuresof infant-, compared to adult-directed speech. A small seg-mentation advantage was also documented, which could notbe attributed to any of those corpus features. Some particular-ities of the data invite further research on more corpora.