Statistical learning of multiple speech streams: A challenge for monolingual infants

Area of Science:

• Developmental Psychology
• Cognitive Science
• Linguistics

Background:

• Language acquisition relies on detecting statistical properties of speech.
• Learners must adapt to changing speech statistics from different sources.
• Tracking multiple, overlapping speech streams presents a significant challenge.

Purpose of the Study:

• To investigate 8-month-old infants' ability to track statistical regularities in two artificial speech streams.
• To determine if infants can differentiate and track separate speech streams when they share overlapping phonetic features.
• To assess the effectiveness of contextual cues (pitch, accent) in aiding stream separation.

Main Methods:

• Four experiments were conducted with 144 monolingual 8-month-old infants.
• Infants were exposed to individual artificial speech streams, followed by sequential presentation of two streams.
• Contextual cues (pitch, accent) were introduced to aid stream separation in later experiments.

Main Results:

• Infants demonstrated difficulty in tracking the statistical regularities of two sequentially presented speech streams.
• This difficulty was observed even when the streams shared common syllables.
• The addition of pitch and accent cues did not significantly improve infants' ability to separate the streams.

Conclusions:

• Eight-month-old infants face challenges in simultaneously tracking the statistical properties of multiple, overlapping speech streams.
• The findings suggest limitations in infants' ability to segment and process concurrent auditory information with distinct statistical structures.
• Implications for understanding early language development and the mechanisms of auditory scene analysis in infants.