Computational Language Modeling and the Promise of In Silico Experimentation.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Neuroscience of Language

Background:

• Current language neuroscience research primarily uses controlled experiments with hand-designed stimuli or natural stimulus experiments.
• Both paradigms have complementary strengths but also inherent limitations in studying the neurobiology of language.

Purpose of the Study:

• Introduce and evaluate a third experimental paradigm: in silico experimentation using deep learning-based encoding models.
• Demonstrate the potential of this new approach to bridge the interpretability of controlled studies with the generalizability of naturalistic ones.

Main Methods:

• Utilized deep learning-based encoding models for in silico experimentation in language neuroscience.
• Simulated four distinct language neuroscience experiments using this computational approach.

Main Results:

• The in silico paradigm demonstrated the potential to combine interpretability with broad scope and generalizability.
• Four examples illustrate the application and feasibility of simulating language neuroscience experiments.

Conclusions:

• In silico experimentation represents a promising third paradigm in language neuroscience.
• This approach offers a novel way to investigate the neurobiology of language, leveraging advances in cognitive computational neuroscience.