37.3022, FYI: Gesture Challenge 2026

[The LINGUIST List]

LINGUIST List: Vol-37-3022. Mon Sep 21 2026. ISSN: 1069 - 4875.

Subject: 37.3022, FYI: Gesture Challenge 2026

Moderator: Steven Moran (linguist at linguistlist.org)
Managing Editor: Valeriia Vyshnevetska
Team: Helen Aristar-Dry, Daniel Swanson
Jobs: jobs at linguistlist.org | Conferences: callconf at linguistlist.org | Pubs: pubs at linguistlist.org

Homepage: http://linguistlist.org

Editor for this issue: Daniel Swanson <daniel at linguistlist.org>

================================================================


Date: 21-Sep-2026
From: Sharjeel Ahmed Shaikh [shaikh1 at uni-potsdam.de]
Subject: Gesture Challenge 2026


The Envision Gesture Challenge 2026 has pooled seven annotated gesture
corpora into a single openly licensed dataset. We are releasing it
alongside an open benchmark for co-speech gesture detection, and we
would like help disseminating this challenge.
The challenge aims to address the bind we often find ourselves in:
annotated gesture corpora contain videos of identifiable people, which
limits their shareability. AI scientists can then only work with
limited data, which hampers reliability and hinders the development
of, for example, generalized co-speech gesture detection models.
So what changed? Pose estimation is now robust enough that we can
share kinematic landmarks and body-pose animations instead of video,
avoiding identity exposure. This motivated us to approach research
groups with restricted open-access datasets and ask whether they could
pool and release their annotated data in this form.  Together with
already available open datasets, we compiled the Envision Gesture
Challenge dataset (v1).
The dataset consists of 7 corpora, in 4 languages, with 109 unique
speakers, and roughly 34,000 Gesture, NoGesture, and Move clips (with
their dataset-specific sublabels; e.g., beat), under CC-BY-NC-SA 4.0.
The corpus breakdown, evaluation metrics, and submission requirements
are all on the challenge page (attached below).
We want to invite the community to:
1. Enter a model. The task is detecting where gestures occur, not
deciding what they mean — interpretation stays the analyst's work. Our
baselines achieve balanced accuracy of 58% (LightGBM) and 61% (CNN),
which is not yet good enough to rely on, so we invite the research
community to improve on this.
2. Use the data. It is a sizeable, openly licensed, multilingual
gesture corpus, and it is useful well beyond this challenge. For
example, to assess the natural statistics of gesture kinematics.
3. Contribute a corpus. We want to update the dataset in future
versions by including more datasets, and to organize new challenges
this way (e.g., gesture-type challenges or gesture-semantics
challenges). More languages, more settings, and more annotation
traditions would make the models more robust against biases they
inherit from nondiverse training datasets. Importantly, nothing needs
to leave your institution as video; we can provide the script and
setup support to generate the portable, de-identifiable data that can
be shared (with your annotations).
Submissions close on 30 November 2026.
Everything else — documentation, corpus details, metrics, sample
clips:
https://envisionbox.org/gesture_challenge_2026.html
You can also try one of our baseline gesture detection models running
live in your browser:
https://envisionbox.org/webcamgesturedetectdemo.html
This project emerged from our project to train a general co-speech
gesture classifier to be published as:
Pouw, W., Shaikh, S. Trujillo, J.,, Yung. B., Rueda-Toicen, A., De
Melo, G., & Owoyele, B. (2025). EnvisionHGdetector: A computational
framework for co-speech gesture detection, kinematic analysis, and
interactive visualization in Behavioromics: Semantic, Experimental,
and Computational Multimodal Interaction Studies' (eds. A. Lücking &
A. Mehler). doi: 10.30819/6158.07 (postprint)
Warmly,
The EnvisionBOX team

Linguistic Field(s): Cognitive Science
                     Computational Linguistics
                     Text/Corpus Linguistics




------------------------------------------------------------------------------

********************** LINGUIST List Support ***********************
Please consider donating to the Linguist List, a U.S. 501(c)(3) not for profit organization:

https://www.paypal.com/donate/?hosted_button_id=87C2AXTVC4PP8

LINGUIST List is supported by the following publishers:

Australian Linguistics Society https://als.asn.au/Home

Bloomsbury Publishing http://www.bloomsbury.com/uk/

Cambridge University Press http://www.cambridge.org/linguistics

Cascadilla Press http://www.cascadilla.com/

De Gruyter Brill https://www.degruyterbrill.com/?changeLang=en

Edinburgh University Press http://www.edinburghuniversitypress.com

European Language Resources Association (ELRA) http://www.elra.info

John Benjamins http://www.benjamins.com/

Language Science Press http://langsci-press.org

Lincom GmbH https://lincom-shop.eu/

MDPI Languages https://www.mdpi.com/journal/languages

MIT Press http://mitpress.mit.edu/

Multilingual Matters http://www.multilingual-matters.com/

Narr Francke Attempto Verlag GmbH + Co. KG http://www.narr.de/

Netherlands Graduate School of Linguistics / Landelijke (LOT) http://www.lotpublications.nl/

Peter Lang AG http://www.peterlang.com

SIL International Publications http://www.sil.org/resources/publications


----------------------------------------------------------
LINGUIST List: Vol-37-3022
----------------------------------------------------------