Slone, L. , Abney, D. , Smith, L. & Yu, C.
Indiana University, Bloomington
One proposal in statistical language-learning is that early auditory environments are structured and that early learning is supported by acquiring regularities in speech. Statistical learning research supports this notion, demonstrating infants' acquisition of statistical speech properties at the level of sounds, words, and syntax. This talk will demonstrate that structure is also present at higher levels of speech - utterances - and that this structure facilitates early language learning. We focus on one type of temporal structure: Burstiness. Burstiness is a characteristic of communications involving a large amount of data sent in a short time - in bursts - rather than as a continuous stream. Burstiness dynamics have been found across communicative phenomena such as linguistic interactions and dyadic problem-solving tasks. We used Goh and Barabási's (2008) burstiness metric to describe the temporal structure of parents' utterances to their infants during 190 sessions of dyadic toy play, and found that parents' speech to infants is temporally structured, and exhibits a predominantly bursty signal relative to a rhythmic periodic signal. We will discuss how the temporal structure of speech may support statistical learning of object names and the utility of the burstiness metric for quantifying temporal structure in early learning environments.