The ZuCo benchmark on cross-subject reading task classification with EEG and eye-tracking data.

Area of Science:

• Computational neuroscience
• Cognitive science
• Machine learning

Background:

• Advancing research at the intersection of computational language processing and cognitive neuroscience requires robust benchmarks.
• Distinguishing between normal and task-specific reading paradigms using neurophysiological and eye-tracking data is a key challenge.

Purpose of the Study:

• To introduce a novel machine learning benchmark for reading task classification.
• To facilitate advancements in electroencephalography (EEG) and eye-tracking research.

Main Methods:

• Utilized the Zurich Cognitive Language Processing Corpus (ZuCo 2.0) for simultaneous eye-tracking and EEG signals during natural English sentence reading.
• Developed a cross-subject classification task to differentiate between normal and task-specific reading.
• Provided baseline methods and a publicly available training dataset with a hidden test set.

Main Results:

• Established a new benchmark for reading task classification.
• Demonstrated the feasibility of distinguishing reading paradigms using EEG and eye-tracking data.
• Released code, an evaluation interface, and a public leaderboard for community engagement.

Conclusions:

• The developed benchmark provides a valuable resource for the research community.
• Future work can build upon this benchmark to explore more sophisticated models and further understand reading processes.
• Encourages further investigation into the integration of computational language processing with cognitive neuroscience.