Input Properties Shape Word Segmentation Performance Across Child Development: A Computational Modeling Study.

Area of Science:

• Developmental linguistics
• Computational linguistics
• Speech processing

Background:

• Child-directed speech (CDS) provides crucial input for language acquisition.
• Understanding how CDS properties change with child development is key to explaining language learning.
• Computational models can simulate and test hypotheses about language learnability.

Purpose of the Study:

• To investigate developmental changes in Korean child-directed speech (CDS).
• To analyze how these linguistic shifts affect word segmentation in computational models.
• To explore the relationship between input structure and language learnability.

Main Methods:

• Analyzed a cross-sectional corpus of Korean CDS for 35 children (6-30 months).
• Quantified linguistic properties: utterance/word length, lexical diversity (MATTR), entropy, word type frequency, and use of specific word types.
• Simulated word segmentation performance using computational models.

Main Results:

• Observed developmental shifts in CDS linguistic properties, including utterance length and lexical diversity.
• Found an inverse U-shaped trajectory in word segmentation performance, peaking in the early speech stage.
• Demonstrated that changes in CDS input properties largely explain the observed segmentation patterns.

Conclusions:

• Linguistic properties of child-directed speech systematically change across early development.
• These input structure shifts influence the learnability of linguistic units for infants.
• Modeling input-learnability dynamics is essential for understanding early language acquisition.