Modelling child learning and parsing of long-range syntactic dependencies.

Area of Science:

• Computational Linguistics
• Developmental Psychology
• Cognitive Science

Background:

• Child language acquisition involves learning complex linguistic phenomena, including long-range syntactic dependencies.
• Existing models often struggle to simultaneously learn word meanings and language-specific syntax.
• Understanding how children acquire language is a fundamental challenge in cognitive science.

Purpose of the Study:

• To develop a probabilistic computational model that learns linguistic phenomena from child-directed speech.
• To enable the model to simultaneously acquire word meanings and language-specific syntax.
• To test the model's ability to handle complex constructions like object wh-questions.

Main Methods:

• Training a probabilistic model on a corpus of transcribed child-directed speech paired with logical forms.
• Utilizing a meaning representation to guide the simultaneous learning of syntax and semantics.
• Developing algorithms for deducing parse trees and word meanings, and inferring meaning from strings.

Main Results:

• The model successfully learned a range of linguistic phenomena, including long-range syntactic dependencies.
• Simultaneous acquisition of word meanings and language-specific syntax was achieved.
• The model demonstrated the ability to deduce correct parse trees and infer meaning from given strings.

Conclusions:

• Probabilistic models can effectively capture key aspects of child language acquisition.
• Simultaneous learning of syntax and semantics is feasible within a computational framework.
• This model provides a novel approach to understanding the computational mechanisms underlying early language development.