OneStop: A 360-Participant English Eye Tracking Dataset with Different Reading Regimes

📄 Paper | 📚 Documentation | 💾 Data | 🔬 More from LaCC Lab

Example

Overview

OneStop Eye Movements (in short OneStop) is a large-scale English corpus of eye movements in reading with 360 L1 participants, 2.6 million word tokens and 152 hours of recorded eye tracking data. The dataset was collected using an EyeLink 1000 Plus eyetracker (SR Research).

The dataset release includes Interest Area Reports (features aggregated at the word level), Fixation Reports (features aggregated at the level of fixations/saccades), raw data in edf and ASCII formats, and a detailed participant questionnaire. To facilitate analyses, we further provide precomputed text annotations: word length, frequency and surprisal (GPT2), as well as part-of-speech tags and syntactic dependency trees.

OneStop comprises four sub-corpora, one for each of the following reading regimes:

• Ordinary reading for comprehension Download this data if you are interested in a general-purpose eye tracking dataset (like Dundee, GECO, MECO and others).
• Information seeking
• Repeated reading
• Information seeking in repeated reading

We provide the entire dataset, as well as each of the sub-corpora separately.

Key Features

Texts and Reading Comprehension Materials

• Taken from the OneStopQA dataset (Paper Data 🤗)
• 30 articles with 162 paragraphs in English from the Guardian.
• Annotations of part-of-speech tags, syntactic dependency trees, word length, word frequency and word surprisal.
• Each paragraph has two versions: an Advanced version (original Guardian text) and a simplified Elementary version.
• Extensively piloted reading comprehension questions based on the STARC (Structured Annotations for Reading Comprehension) annotation framework.
  • 3 multiple-choice reading comprehension questions per paragraph.
  • 486 reading comprehension questions in total.
  • Auxiliary text annotations for answer choices.

Statistics

Statistics of OneStop and other public broad-coverage eyetracking datasets for English L1.

Category	Dataset	Subjects	Age	Words	Words Recorded	Questions	Subjects per Question	Questions per Subject
Reading Comprehension	OneStop	360	22.8±5.6	19,425 (Advanced) 15,737 (Elementary) 19,221 (QA)	2,632,159 (Paragraphs) 1,311,752 (QA)	486	20	54
	SB-SAT L1	66	NA	2,539	167,574	20	95	20
Passages	Dundee	10	NA	51,502	307,214	NA	10	NA
	GECO L1	14	21.8±5.6	56,410	774,015	NA	14	NA
	Provo	84	NA	2,689	225,624	0	0	0
	MECO En	46	21.0±2.2	2,109	83,246	48	46	48
Sentences	CELER L1	69	26.3±6.7	61,233	122,423	78	69	78
	ZuCo	18	34.3±8.0	15,138	272,484	42	18	42
	UCL	43	25.8±7.5	1,932	81,144	110	43	110

'Reading Comprehension' are datasets with a substantial reading comprehension component over piloted reading comprehension materials. The remaining datasets are general purpose datasets over passages or individual sentences. 'Words' is the number of words in the textual corpus. 'Words Recorded' is the number of word tokens for which tracking data was collected. 'NA': data not available.

Controlled experimental manipulations

1. Reading goal: ordinary reading for comprehension or information seeking.
2. Paragraph difficulty level: original Guardian article (Advanced) or simplified (Elementary).
3. Question identity: one of three possible questions for each paragraph.
4. Prior exposure to the text: first reading or repeated reading.

Experiment Structure

Each participant reads 10 Guardian articles paragraph by paragraph, and answers a reading comprehension question after each paragraph. After reading a 10-article batch, participants read two of the previously presented articles for a second time. Half of the participants are in an information seeking regime where they are presented with the question prior to reading the paragraph.

Trial Structure

A single experimental trial consists of reading a paragraph and answering one reading comprehension question about it. Trials have the following Interest Periods, corresponding to pages in the experiment:

1. Question Preview (only in the information seeking regime) - the participant reads the question (self-paced).
2. Paragraph - the participant reads the paragraph (self-paced).
3. Question - the participant reads the question again (self-paced).
4. QA - retains the question, and also displays the four possible answers in a cross arrangement. Participants are then required to choose one of the answers and confirm their choice (self-paced).
5. Feedback - informs the participants on whether they answered the question correctly (presented for one second).

Pages presented only in the information seeking regime are depicted in green.

Overall, the dataset includes 19,438 regular trials, 9,720 in the information seeking regime and 9,718 in the ordinary reading regime. The dataset also includes 3,888 repeated reading trials, split equally between the two reading regimes.

Obtaining the Data

Direct Download from OSF

The data is hosted on OSF, and can be downloaded here.

We provide the possibility to download four sub-corpora with eye movement recordings from paragraph reading, one for each of the following reading regimes:

1. OneStop Ordinary Reading - download this data if you are interested in a general-purpose eye tracking dataset.
2. OneStop Information Seeking
3. OneStop Repeated Reading
4. OneStop Information Seeking in Repeated Reading

Two versions of the complete dataset:

• OneStop All Regimes - paragraph reading for the entire dataset.
• OneStop Full - complete trials for the entire dataset, divided into title, question preview, paragraph, question, answers, QA (question + answers) and feedback.

The raw data:

• OneStop Raw - edf and ASCII files.

and the Metadata:

• OneStop Metadata includes the participant questionnaire and experiment summary statistics.

Python Script

The data can also be downloaded using the provided Python script. The script will download and extract the data files.

Basic usage to download the entire dataset:

0. Make sure you have Python installed.

   • If you don't have Python installed, you can download it from here.

1. Get the Code

   • Open your terminal/command prompt:

     • Windows: Press Win + R, type cmd and press Enter
     • Mac: Press Cmd + Space, type terminal and press Enter

   • Run this command to download the code:

     git clone https://github.com/lacclab/OneStop-Eye-Movements.git

   • Move into the downloaded folder:

     cd OneStop-Eye-Movements

2. Run the Download Script

   • Run this command to download the OneStop Ordinary Reading subcorpus:

     python download_data_files.py

   • The data will be downloaded to a folder called "data/OneStop"

   • Available options:

     • --extract: Extract downloaded zip files (default: False)
     • -o, --output-folder: Specify output folder (default: "data/OneStop")
     • --mode: Choose dataset subcorporus to download (default: "ordinary")
       • Options: "onestop-full", "onestop_all_regimes", "ordinary", "information_seeking", "repeated","information_seeking_repeated"

   • Example usage to download the information-seeking subcorporus:

     python download_data_files.py --mode information_seeking

pymovements integration

OneStop is integrated into the pymovements package. The package enables easy download of the paragraph data for the complete dataset. The following code snippet shows how to download the data:

First, install the package in the terminal:

pip install pymovements

Then, use the following python code to download the data:

import pymovements as pm
dataset = pm.Dataset('OneStop', path='data/OneStop')

dataset.download()

This will download the data to the data/OneStop folder. You can also specify a different path by changing the path argument.

Documentation

• Data Files and Variables: Detailed information about the data files and variables in the reports.

• Known Issues: Known issues with the dataset. If you identify an issue not listed here, please open a github issue specifying the problem.

• Scripts: Scripts for data preprocessing and reproducing the analysis in the dataset paper.

Release Versions

Releases correspond to the data release version.

Citation

Paper: OneStop: A 360-Participant English Eye Tracking Dataset with Different Reading Regimes

@Article{Berzak2025,
author={Berzak, Yevgeni and Malmaud, Jonathan and Shubi, Omer and Meiri, Yoav and Lion, Ella and Levy, Roger},
title={{OneStop}: A 360-Participant English Eye Tracking Dataset with Different Reading Regimes},
journal={Scientific Data},
year={2025},
doi={10.1038/s41597-025-06272-2},
}

Dataset Uses Examples

Text Simplification and Text Readability

• Eye Tracking Based Cognitive Evaluation of Automatic Readability Assessment Measures (arXiv 2025)
• The Effect of Text Simplification on Reading Fluency and Reading Comprehension in L1 English Speakers (CogSci 2025)

Reading Comprehension

• Fine-Grained Prediction of Reading Comprehension from Eye Movements (EMNLP 2024)
• Bridging Information-Seeking Human Gaze and Machine Reading Comprehension (CoNLL 2020)

Information Seeking

• Eye Movements in Information-Seeking Reading (CogSci 2023)
• Decoding Reading Goals from Eye Movements (ACL 2025)
• Decoding Open-Ended Information Seeking Goals from Eye Movements in Reading (arXiv 2025)

Repeated Reading

• Déjà Vu? Decoding Repeated Reading from Eye Movements (ACL 2025)
• Déjà Vu: Eye Movements in Repeated Reading (CogSci 2024)

Human-LLM Alignment and Memorization

• The Effect of Surprisal on Reading Times in Information Seeking and Repeated Reading (CoNLL 2024)

License

The eye tracking data, code, and anonymized participant questionnaire responses are released under a Creative Commons Attribution 4.0 International License. The underlying textual materials and auxiliary text annotations of OneStopQA are provided under the OneStopQA Creative Commons Attribution-ShareAlike 4.0 International License.