Language learning as language use: A cross-linguistic model of child language development

Area of Science:

• Computational Linguistics
• Developmental Psychology
• Cognitive Science

Background:

• Usage-based theories of language development are supported by computational studies.
• A key challenge is modeling language learning as language use.
• Existing models often lack broad linguistic coverage or a unified framework for comprehension and production.

Purpose of the Study:

• To present a usage-based computational model of language acquisition.
• To model language learning as language use through incremental, online processing.
• To achieve broad, cross-linguistic coverage uniting comprehension and production.

Main Methods:

• Developed an incremental, usage-based computational model utilizing chunking for online processing.
• Designed the model with memory constraints and inspired by psycholinguistic evidence on distributional properties and shallow processing.
• Trained the model on child-directed speech corpora and evaluated its shallow parsing and production capabilities across numerous child-specific datasets in multiple languages.

Main Results:

• The model demonstrated high performance on both comprehension (shallow parsing) and production tasks.
• Evaluated across 79 English, French, and German corpora for shallow parsing.
• Validated production performance on over 200 corpora across 29 languages from the CHILDES database, capturing children's complex sentence production.

Conclusions:

• Children's early linguistic behavior can be supported by item-based learning via online processing of distributional cues.
• Language acquisition can be understood as learning to process language.
• The model provides a unified framework for understanding language acquisition, integrating comprehension and production.