Automatic parsing of parental verbal input

Area of Science:

• Computational Linguistics
• Developmental Psychology
• Corpus Linguistics

Background:

• Evaluating child language acquisition theories requires analyzing large datasets of parsed child-directed speech.
• Manual parsing is time-consuming and impractical, leading to a lack of parsed corpora for child language input.
• Existing methods struggle with the coverage-ambiguity tradeoff inherent in parser design.

Purpose of the Study:

• To develop an automated system for syntactically parsing child-directed language.
• To create a reliable method for processing large corpora of child language input data.
• To facilitate the evaluation of theoretical proposals in child language acquisition.

Main Methods:

• Combined the MOR tagger, a rule-based parser, and statistical disambiguation techniques.
• Developed a specific processing sequence to manage the coverage/ambiguity tradeoff.
• Applied the system to the CHILDES database, an international corpus of child language transcripts.

Main Results:

• Achieved nearly 80% correct parses for sentences spoken to children.
• Demonstrated the system's effectiveness in automating the parsing of child language input.
• Successfully processed data from the CHILDES database.

Conclusions:

• The developed automated parsing system significantly aids research in child language acquisition.
• The system's high accuracy and availability address the need for parsed corpora.
• Freely available data and programs support further research using the CHILDES database.