Chunk-based incremental processing and learning: An integrated theory of word discovery, implicit statistical...

Area of Science:

• Cognitive Science
• Developmental Psychology
• Computational Linguistics

Background:

• Chunking theories propose children learn words by extracting subsequences from continuous speech.
• Existing chunking models often conflict with broader language development data.

Purpose of the Study:

• To propose a new theory connecting chunking with developmental literature.
• To introduce a computational model (CIPAL) for word discovery.

Main Methods:

• Developed the Chunk-Based Incremental Processing and Learning (CIPAL) computational architecture.
• Trained CIPAL on 70 child-directed speech corpora across 15 languages.
• Evaluated CIPAL's ability to simulate statistical learning effects from artificial languages.

Main Results:

• CIPAL successfully discovered words in diverse languages, showing cross-linguistic variation.
• Model processing times improved with experience, mirroring human developmental changes.
• CIPAL replicated seven key findings from statistical learning experiments, including preferences for words over nonwords and sublexical units.

Conclusions:

• Incremental chunking serves as an effective implicit statistical learning mechanism for vocabulary acquisition.
• The CIPAL model supports the integration of chunking with broader developmental and processing constraints.
• This framework offers a unified account of word discovery in early language development.