Modeling children's early grammatical knowledge.

Area of Science:

• Linguistics
• Developmental Psychology
• Computational Linguistics

Background:

• Theories of grammatical development vary in the abstract knowledge attributed to young children.
• This study investigates an alternative model of child grammar acquisition based on concrete linguistic items.

Purpose of the Study:

• To evaluate the explanatory power of item-based grammars versus abstract rule-based grammars in early child language.
• To model grammatical development using computational methods applied to child speech data.

Main Methods:

• A Bayesian procedure was used to extract item-based grammars from over 28 hours of speech from two children at ages 2 and 3.
• These extracted grammars were then used to parse unique multiword utterances from separate recordings of the same children.

Main Results:

• At age 2, the item-based model demonstrated good coverage and predictive fit, indicating limited child productivity.
• Adding parts of speech (nouns, verbs) improved the model's performance at age 3, coinciding with increased child productivity.

Conclusions:

• Early child grammars may be better explained by concrete item-based knowledge with local abstractions.
• Grammatical productivity and the utility of abstract categories like nouns and verbs increase significantly between ages 2 and 3.