Reinforcement-learning signals support dynamic adaptive control during language switching.

Area of Science:

• Neuroscience
• Cognitive Psychology
• Computational Linguistics

Background:

• Bilinguals adapt language control based on context, but learning mechanisms are unclear.
• The Adaptive Control Hypothesis (ACH) describes context-dependent control but not strategy acquisition.
• Reinforcement learning (RL) theory offers a framework for understanding adaptive strategy adjustments.

Purpose of the Study:

• Investigate if reward prediction error (RPE) influences voluntary language switching in bilinguals.
• Examine the neurocomputational basis of learning and automating language control strategies.
• Apply RL principles to high-level cognitive control beyond simple stimulus-response.

Main Methods:

• Chinese-English bilinguals performed a voluntary picture-naming task with probabilistic reward feedback.
• Computational modeling analyzed how RPE updated abstract switch policies.
• Representational similarity analysis (RSA) and connectome-based predictive modeling (CPM) investigated neural representations and network dynamics.

Main Results:

• RPE dynamically updated a generalizable language switch policy.
• The medial temporal lobe (MTL) represented policy value, while the dorsolateral prefrontal cortex (dlPFC) represented RPE.
• Connectome-based predictive modeling (CPM) revealed a multi-stage network process: exploration, exploitation, automation, and error monitoring.

Conclusions:

• Adaptive language control is a value-based RL process.
• Findings provide a neurocomputational framework for cognitive strategy learning.
• RL principles extend to high-level cognitive control, explaining strategy automation and error monitoring.