# Multi-Annotation fMRI Dataset from Extensive Drama and Film Viewing by Hiroto Q. Yamaguchi, Naoko Koide-Majima, Rieko Kubo, Tomoya Nakai, and Shinji Nishimoto 

The dataset is designed to investigate the brain representations underlying narrative understanding. We measured fMRI BOLD responses from six participants as they watched nine narrative movies (~9 hours total). Additionally, BOLD responses were recorded during several functional localizer runs. Detailed annotations of the movie contents are also provided. 

	•	Six participants: sub-S01, sub-S02, …, sub-S06
	•	Nine titles of narrative movies: breakingbad, bigbangtheory, crown, heroes, suits, dreamgirls, glee, mentalist, gis 

The "derivatives" directory contains the following six subdirectories, storing data used in the analytical procedures on our GitHub page (https://github.com/nkmjm/fMRI_Narrative_movie/). These procedures include extracting large language model (LLM) latent features from the annotations, encoding model fitting using the LLM latent features, and visualizing encoding accuracies:

	•	preprocessed_data: Preprocessed fMRI BOLD responses.
	•	annotation: Three types of annotations. In each annotation file, the text in each row corresponds to the annotation for each second. 
	•	feature: GPT-2 latent features in the 6th, 12th, 18th, and 24th layers, extracted from the annotations.
	•	pycortex_db: Pycortex database for visualizing analytical results on the cortical surface.
	•	localizer: Statistics and estimates derived from nine types of functional localizer data (visual, auditory, motor, and cognitive tasks) used to create brain masks for various regions of interest (ROIs).
	•	accuracy: Prediction accuracies of an encoding model using latent features from the three types of annotations.

For detailed instructions on how to use the data, please refer to our GitHub page (https://github.com/nkmjm/fMRI_Narrative_movie).

***Caution*** This dataset is for research purposes only. The data have been anonymized, and users must not perform analyses aimed at re-identifying individual subjects.