Eyetracked Multi-Modal Translation (EMMT)

Release version 1.0

We present Eyetracked Multi-Modal Translation (EMMT), a dataset containing monocular eye movement recordings, audio data and 4-electrode wearable electroencephalogram (EEG) data of 43 participants while engaged in sight translation task supported by an image.

The full description of the experiment design is in the paper accompanying the EMMT dataset:

EMMT (Eyetracked Multi-Modal Translation), a simultaneous eye-tracking, 4-electrode EEG and audio corpus for multi-modal reading and translation scenarios
Sunit Bhattacharya, Věra Kloudová, Vilém Zouhar, Ondřej Bojar
2022

Description of the data collected

The participants processed inputs in four stages: (1) reading an English sentence aloud, (2) translating it to Czech, (3) observing an image, related or unrelated to the sentence, (4) producing the same or a new translation.

Three kinds of data were recorded during the course of the experiment: eye-tracking (gaze), EEG and audio.

For the experiment, each participant was presented with a set of 32 stimuli, each containing the text of the English sentence and the image. Each such set of 32 stimuli is referred to as a "probe". 20 such probes were used for the purpose of the experiment. EMMT contains the gaze and audio recordings of 43 participants (P01-P43). EEG data of 28 (out of 43) participants was collected and accordingly included in EMMT.

Description of the folder structure

This is the overall folder structure used here:

EMMT/
    - probes/
        - probe<xy>
            - P<participant_id>-<order_of_presentation>-S<sentence_id>-<cond>-<cong>-i<image_id>.eeg
            - P<participant_id>-<order_of_presentation>-S<sentence_id>-<cond>-<cong>-i<image_id>.et
            - P<participant_id>-<order_of_presentation>-S<sentence_id>-<cond>-<cong>-i<image_id>.events
            - P<participant_id>-<order_of_presentation>-S<sentence_id>-<cond>-<cong>-i<image_id>.mp3
            - ...
            - probe
        - ...
        - Participants.csv
        - Sentences.csv

    - images/
        - a<image_id>.jpg
        - a...   
        - u<image_id>.jpg
        - u...
        - 100.jpg

The folder probes contain the recordings corresponding to the probes presented to the participants. The probes are numbered in the format:probe<xy> with <xy> ranging from 01 to 20. The details of the 32 stimuli in the probe can be found in the file called probe inside each probe<xy> directory.

Inside each probe<xy> directory, there are four types of files corresponding to each stimuli presented to the participants:

.events contains the overall timing log
.et is the gaze file
.eeg is the EEG recording (missing for some participants)
.mp3 is the recording of the spoken translations and anything else said by the participant or the experiment moderator

The filenames have the following format: P<participant_id>-<order_of_presentation>-S<sentence_id>-<cond>-i<cond><image_id> and then the file type suffix.

<participant_id> corresponds to the unique participant ID accorded to each participant. The anonymized participant names and the <participant_id> corresponding to them can be found in the file probes/Participants.csv.
<order_of_presentation> refers to the actual order in which the stimuli were presented to the participant. Note that the stimuli listed in the probe file were presented in a random order to each participant during the experiment. <order_of_presentation> contains the information about that ordering.
<sentence_id> is the unique ID assigned to the English sentence. All sentences used in the experiment with their IDs can be found in the file probes/Sentences.csv.
<cond> refers to the condition captured by the textual stimulus. <cond> has the values of A or U corresponding to an ambiguous or unambiguous sentence, respectively.
<cong> refers to the congruency between the textual stimulus and the visual stimulus. <cong> has the values of C, I or M corresponding to Congruent (text and image are related), Incongruent (text and image are unrelated) and Missing (image is missing, i.e. an empty image shown).
<image_id> refers to the ID of the actual image presented to the participant as part of the stimulus. The image ID distinguishes between images serving for ambiguous sentences (a...) and unambiguous sentences (u...). The images can be found in the images/ directory.

Authors

Sunit Bhattacharya
Věra Kloudová
Vilém Zouhar
Ondřej Bojar

License

This project is licensed under a Creative Commons Attribution 4.0 International License. - see the LICENSE.md file for details

Acknowledgment