Skip to content
Open Imaging Index
MRI Brain

Narratives (ds002345)

The Narratives fMRI dataset for evaluating models of naturalistic language comprehension

fMRI of healthy adults listening to spoken stories, with audio and time-stamped transcripts, a benchmark for models of language and narrative comprehension. 345 people from Princeton across 891 runs and 27 stories, with anatomical scans, in BIDS under CC0.

Go to dataset Edit Report

Overview

Narratives pools functional MRI recorded while adults listened to spoken stories in the Hasson and Norman labs at the Princeton Neuroscience Institute between October 2011 and September 2018. The authors shared it on OpenNeuro as ds002345 under a CC0 waiver and described it in Scientific Data in 2021. It serves as a benchmark for models of language and narrative comprehension and for naturalistic analyses such as intersubject correlation.

Composition

Snapshot 1.1.4 covers 345 adults aged 18 to 53 (mean 22.2 years, 204 reporting female). Subjects heard one or more of 27 stories lasting from about 3 to 56 minutes, about 4.6 hours of unique audio in total, and the paper counts 891 functional scans. Subjects who took part in several sub-studies keep one identifier, and participants.tsv lists the stories, experimental conditions and comprehension scores per subject. Every subject has a T1-weighted image, and 46 also have a T2-weighted image. The release includes the audio files, plain transcripts and word- and phoneme-level time stamps.

Acquisition

All scans come from the Scully Center for Neuroimaging in Princeton with a 1.5 s repetition time. The earlier sub-studies used a 3 T Siemens Skyra with a 20-channel head coil and 3 x 3 x 4 mm EPI. Later ones used a 3 T Siemens Prisma with a 64-channel coil and multiband EPI at 2 or 2.5 mm, and T2-weighted images were only acquired in the 2018 sub-studies. Images were converted with dcm2niix, defaced with pydeface and organized in BIDS 1.2.1.

Annotations

There are no image labels. Annotations are on the stimulus side: transcripts aligned in time to the audio. The full DataLad distribution adds fMRIPrep outputs, AFNI-postprocessed time series and MRIQC quality metrics.

Known limitations

  • The sample is mostly young adults from a university community, and both native and non-native English speakers are included.
  • Acquisition parameters differ between sub-studies, so data from the Skyra and Prisma groups need harmonization.
  • The authors recommend excluding specific scans, flagged by their intersubject correlation checks and listed in code/scan_exclude.json.
  • The paper counts 891 functional scans, while the snapshot 1.1.4 file listing holds 873 BOLD NIfTI files.
  • Ages in participants.tsv are given per story, so a subject scanned across years has several ages.

Cohort

Aggregate numbers from the sources below. Bars are relative to the 345 subjects.

Sex

  • Female 204 59%
  • Male 141 41%

Methods: Participants in Nastase et al. 2021, The "Narratives" fMRI dataset for evaluating models of naturalistic language comprehension, Scientific Data; sex column in participants.tsv of ds002345 1.1.4 (CC0), counted per row from the sex column

Condition

Groups can overlap

  • Healthy control 345 100%

Methods: Participants in Nastase et al. 2021, The "Narratives" fMRI dataset for evaluating models of naturalistic language comprehension, Scientific Data

Scanner vendor

  • Siemens Healthineers 345 100%

Methods: MRI data acquisition in Nastase et al. 2021, The "Narratives" fMRI dataset for evaluating models of naturalistic language comprehension, Scientific Data

Country

  • United States 345 100%

Methods: Participants in Nastase et al. 2021, The "Narratives" fMRI dataset for evaluating models of naturalistic language comprehension, Scientific Data

Age

mean 22.2 ± 5.1, range 18 to 53

No age bins reported.

Methods: Participants in Nastase et al. 2021, The "Narratives" fMRI dataset for evaluating models of naturalistic language comprehension, Scientific Data

Contrast / sequence

Groups can overlap

  • BOLD fMRI 345 100%
  • T1-weighted 345 100%
  • T2-weighted 46 13%

sub-*/func/*_bold.nii.gz, sub-*/anat/*_T1w.nii.gz, sub-*/anat/*_T2w.nii.gz in File listing of ds002345 1.1.4 (CC0), counted from sub-*/anat and sub-*/func NIfTI files

Field strength

  • 3 T 345 100%

Methods: MRI data acquisition in Nastase et al. 2021, The "Narratives" fMRI dataset for evaluating models of naturalistic language comprehension, Scientific Data

Contrast combinations

How many subjects have exactly each set of contrasts.

T1wT2wboldSubjects with exactly this set
299
46

sub-*/anat and sub-*/func in File listing of ds002345 1.1.4 (CC0), counted from sub-*/anat and sub-*/func NIfTI files

License and access

Our reading of the license, not legal advice. Before you use the data, read the original license and confirm that your use is allowed. We take no responsibility for how you use a dataset. Full disclaimer

Access
Open download

Download without an account

Access page

Creative Commons Zero 1.0 Universal

Public domain dedication. Do anything with the data, including commercial use, without asking and without having to give credit.

dataset_description.json of snapshot 1.1.4 states "License" CC0; the paper states that all MRI data and metadata are released under CC0.

Original license text Version read: 1.0 Checked 2026-10-07

What you can do

  • Yes
  • Yes
  • Yes
  • Yes
  • Yes

What you can share

  • Yes
  • Yes
  • Yes

What you must do

  • No
  • Share alike No
  • No
  • No
  • Manuscript review No
  • Release code No
  • Return results No
  • Delete after use No

Limits

  • No
  • Location limits No

Citation

Nastase SA, Liu YF, Hillman H, et al. The "Narratives" fMRI dataset for evaluating models of naturalistic language comprehension. Scientific Data 8, 250 (2021). doi:10.1038/s41597-021-01033-3

Sources

Every number on this page comes from one of these documents. Each chart names the table or page it is taken from. The raw numbers are in stats.csv.