Listening through voices: Infant statistical word segmentation across multiple speakers

Area of Science:

• Cognitive Science
• Developmental Psychology
• Linguistics

Background:

• Infants must discern patterns in variable environments for learning.
• Early language acquisition presents significant learning challenges due to pervasive acoustic variation.
• Statistical learning is a proposed mechanism for infants to segment words.

Purpose of the Study:

• To investigate if infants utilize statistical learning to segment words amidst acoustic voice variation.
• To determine the impact of acoustic variation in speech on infants' word segmentation abilities.
• To assess the scalability of statistical learning mechanisms in naturalistic language learning conditions.

Main Methods:

• Infants (8- and 10-month-olds) listened to continuous speech streams with varying numbers of female voices.
• Transitional probability patterns were analyzed as cues for word segmentation.
• Word segmentation success was tested under conditions of high (8 voices) and low (2 voices) acoustic variation.
• Generalization of learning to novel, acoustically distinct voices was assessed.

Main Results:

• Infants successfully segmented words when exposed to high acoustic variation (8 voices).
• Eight-month-olds generalized word segmentation learning to a new, distinct voice.
• Infants failed to segment words when acoustic variation was low (2 voices).
• Low acoustic variation appeared to impede infants' word segmentation efficiency.

Conclusions:

• Infants effectively employ statistical learning for word segmentation in high-variability speech.
• Acoustic variation is crucial for efficient word segmentation in early language acquisition.
• Statistical learning mechanisms can adapt to varying degrees of acoustic complexity in natural speech environments.