A resource-rational model of human processing of recursive linguistic structure

Area of Science:

• Psycholinguistics
• Cognitive Science
• Computational Linguistics

Background:

• Linguistic expectations (surprisal) predict processing difficulty, but fail for nested structures.
• Nested structures are explained by working memory constraints, but a unified theory is lacking.

Purpose of the Study:

• To propose a unified theory of language processing that integrates expectation-based and memory-based accounts.
• To computationally implement this theory using machine learning and test its predictions.

Main Methods:

• Developed a unified theory modeling a rational trade-off between memory precision and prediction ease.
• Created a computational implementation using contemporary machine learning techniques.
• Conducted comprehension and sentence completion experiments.

Main Results:

• The unified theory accurately predicts difficulty patterns in nested recursive structures.
• Experimental results in English, Spanish, and German confirmed the theory's distinctive predictions.
• The model outperforms theories relying solely on surprisal or memory constraints.

Conclusions:

• A unified framework explains how memory constraints and prediction interact in language.
• This approach provides computationally explicit methods for studying language comprehension and production.
• The findings advance psycholinguistic theory by bridging expectation and memory mechanisms.