[PS-2.8] Infants' learning of embedded regularities in multi-speaker environments

Potter, C. & Lew-Williams, C.

Princeton University

Infants have been shown to detect abstract structure (e.g., ABA vs. AAB) in a variety of paradigms and domains (e.g., Marcus et al., 1997; Saffran et al., 2007). However, in past studies, infants encountered fully deterministic input from a single source, which underestimates the challenge of real-world experience. In the current research, infants were presented with linguistic input from two speakers (one female, one male). 50% of the input followed a pattern (Target Stream); the other 50% had no learnable structure (Non-target stream). Infants heard four 30s blocks of each stream, presented in alternation. In Study 1, the Target stream was spoken entirely by the female speaker, while the male speaker produced the Non-target stream. Infants were tested using the Headturn procedure and reliably listened longer to test items that maintained the Target regularity compared to items that violated the pattern [t(16)=2.36, p=.03]. Thus, infants successfully discovered embedded statistical regularities when the streams were produced by different speakers. Study 2 tests whether infants can learn when each speaker produces half of each stream. Combined, this research explores the critical issue of how infants separate and learn from different sources of information, allowing them to navigate complex auditory and social environments.