AI summary · may contain errors
Transformer-based large language models match human 7T fMRI sentence representations better than order-ignoring models, but all transformers poorly match sentences requiring semantic-role structural sensitivity.
- Open access
