iMinerva: a mathematical model of distributional statistical learning.

Area of Science:

• Cognitive Science
• Psycholinguistics
• Computational Linguistics

Background:

• Statistical learning identifies input structure via statistical properties.
• Distributional regularities (frequency, variability) are key to linguistic structure.
• These regularities are implicated in phonetic categories, word learning, and non-adjacent relation discovery.

Purpose of the Study:

• To demonstrate a unified computational framework for statistical learning across diverse linguistic tasks.
• To investigate the role of domain-general memory characteristics in language acquisition.

Main Methods:

• Utilized a computational framework to model statistical learning processes.
• Applied the framework to tasks including phonetic categorization, word learning, and non-adjacent relation discovery.

Main Results:

• A single computational framework successfully accounts for learning across multiple linguistic tasks.
• Evidence suggests a shared set of underlying processes drives distributional statistical learning.
• Results indicate domain-general memory characteristics contribute to language learning.

Conclusions:

• Distributional statistical learning can be explained by a common set of cognitive processes.
• Language acquisition draws upon domain-general memory mechanisms, not solely domain-specific ones.