Simulating Language-specific and Language-general Effects in a Statistical Learning Model of Chinese Reading.

Area of Science:

• Cognitive Science
• Computational Linguistics
• Psycholinguistics

Background:

• Theoretical models often assume distinct reading processes for different writing systems.
• Specifically, Chinese reading is sometimes thought to lack sub-lexical print-to-sound processing.

Purpose of the Study:

• To test if a unified connectionist model can learn Chinese print-to-sound mappings.
• To investigate if the same learning mechanisms apply to diverse writing systems like Chinese and English.
• To identify language-specific effects in Chinese reading within a computational framework.

Main Methods:

• Developed a connectionist model using established architecture and learning rules applied to English.
• Trained the model on Chinese character print-to-sound mappings.
• Conducted behavioral naming experiments using model-generated test items.

Main Results:

• The model predicted and behavioral experiments confirmed an interaction between item frequency and print-to-sound consistency, similar to English.
• A language-specific regularity effect unique to Chinese was identified.
• The model utilized 'phonetic components' for probabilistic pronunciation information.

Conclusions:

• Adult reading across different writing systems may rely on shared learning mechanisms.
• The input statistics of a writing system, shaped by culture and communication needs, guide these universal learning processes.
• Connectionist modeling provides a unified framework for understanding reading across diverse orthographies.