Publishing Data, Mediapipe, AI Good practice

Service and Support from Project S

Job Schepens

Universität zu Köln

8 July 2026

Overview

Three parts:

• Part 1 — Publishing Data licensing, FAIR principles, data journals, and choosing a repository (Zenodo, OSF, LAC, etc.)
• Part 2 — MediaPipe Group quick recap of what the group is, and a call for contributions
• Part 3 — AI Good Practice in Daily Workflows workshop at the upcoming CRC 1252 retreat (5–7 October 2026)

Part 1 · Publishing Research Data

Part 1: Publishing Data

Project S

We help with, e.g.:

• Choosing a repository: Zenodo, OSF, LAC, etc.
• Choosing a License: choosing a licence, identifying responsible parties (you, PI, UzK)?
• Sensitive data: determining access, secure storage, encryption, GDPR
• Data Management Plan: drafting, reviewing
• Data paper: finding a suitable journal (e.g., JOHD), documentation

Speaker Notes: - audio/video recordings, corpora, and experimental/behavioural data; sensitive-data advice includes steering people away from sciebo for identifiable recordings; DMP support is in line with DFG expectations for CRC 1252.

Good Scientific Practice

As open as possibe:

• Open Science Make data usable for reuse and replication
• All Rights Reserved : Without a license, nobody may legally copy, reuse, or redistribute your data, even if it is in a public GitHub repo

As closed as necessary:

• Ethics GDPR, informed consent, and sensitive data

Who Decides?

• The person accountable / grant holder / the PI / the PI’s

Speaker Notes:

• visible is not automatically reusable.
• Data is subject to both copyright and the EU database right (protecting the investment in compiling it).
• PIs/institutions may hold rights to data collected under employment or grant contracts.

Removing Unnecessary Barriers

FAIR principles:

• F — Findable: metadata, DOI

• A — Accessible: clear access conditions

• I — Interoperable: standard format

• R — Re-useable: what may others do? + provide documentation

• Also sensitive data can be FAIR

Data Licenses

Creative Commons is the most common licensing framework for research data. It has a modular structure, e.g.:

• CC0 (Public Domain): “Waiver”, waives all copyright and database rights.
• BY (Attribution): Users must give you credit.
• SA (ShareAlike): Adaptations must be licensed under the same terms.
• NC (Non-Commercial): Only non-commercial use is permitted (not recommended).

       CC BY (Attribution)
       CC BY-SA (Attribution-ShareAlike)
       CC BY-NC (Attribution-NonCommercial)
       CC BY-NC-SA (Attribution-NonCommercial-ShareAlike)

Licenses for software (e.g., MIT/GPL) don’t address database rights, personal-data protection, or citation norms that apply to research data. We will discuss licences in detail in the next RDM session (July 15th)

Speaker Notes:

• license controls what others may do with your data
• others include Open Data Commons (ODC), seperates database structure rights from content rights.
• NC creates compatibility problems (Non-commercial is not clearly defined, cannot combine a CC BY-NC with CC BY into one, no Search Indexing, nc is not fully open, not allowed in schoolbooks, not with commercial publishers, NC is not unrestricted.)
• In the EU, database rights protect investment in collecting data, even if the data lacks creative expression. CC 4.0 explicitly addresses and waives these rights, unlike CC 3.0.

Documentation: Data Journals

A data paper in a peer-reviewed data journal can help to properly document a dataset. Examples:

• Journal of Open Humanities Data Ubiquity Press / Brill
• Journal of Open Psychology Data Ubiquity Press / Brill
• Research Data Journal for the Humanities and Social Sciences Brill
• Scientific Data Springer Nature
• Data in Brief Elsevier

Speaker Notes:

• A “data paper” describes how a dataset was created and how it can be reused — it does not report new analyses or findings.
• These are peer-reviewed, indexed, and citable, which is a strong incentive for early-career researchers to invest time in preparing data properly.
• JOHD explicitly welcomes linguistics submissions and has run special collections on language dataset reuse — natural venue for CRC 1252 datasets.

Storing and Publishing Sensitive Data

GDPR: Extra care in anonymization/pseudonymization, informed consent, and storage:

• Sciebo (or similar) is not built for sensitive data
• Consider purpose-build access-controlled repositories (e.g., CLARIN/Language Archive Cologne, Dataverse/TROLLing, etc.).
• Institutional network drives with an access log
• Encryption

Speaker Notes:

• repositories like LAC supports restr access with access tiers.
• Always reach out to Project S / C3RDM / DCH / Data protection officer before collecting data.
• repository controls where it lives, how long it survives, and how discoverable it is.
• storage/handling vs publishing/licensing
• Project S can help reviewing, e.g.: consent forms, etc..

Choosing a Repository

Important criteria include Visibility (in community) and Openness of the License e.g. OSF does not require an explicit license.

• Open Science Framework (OSF) helpful concepts for experimental data (e.g., structured project components, preregistration). Supports granular licensing (e.g., project page CC BY, data component CC0). Full project lifecycle.
• Zenodo — general purpose; funded by the EU, operated by CERN.
• lac.uni-koeln.de — good for audio-visual data, including annotations alongside recordings; operated by the Data Center for the Humanities (DCH)
• site.uit.no/trolling — built on the Dataverse platform, recommended by journals (e.g., Glossa)

Checklist

• Ask Project S
• Don’t leave a dataset unlicensed
• As open as possible, as closed as necessary
• Check GDPR/consent
• Use secure storage when necessary
• Choose a suitable repository
• Consider a data paper

Part 2: MediaPipe

Part 2: Mediapipe

MediaPipe Special Interest Group

What?

• Ongoing, informal, discussion and exchange group
• Focus on video-based motion tracking of gesture, sign, gaze, etc. using tools such as MediaPipe, OpenPose, OpenFace
• Original idea is to support projects studying e.g. multimodal prosody and signal, code, and interactional prominence

How?

• 5 meetings in the past 12 months based on interest and available speakers.
• Vadim Kimmelman (March 2026), Anna Kuder et al. (Nov 2025), Fabian Eckert & Aviad Albert et al. (Sept 2025), Klym Myslyvyi & Janne Lorenzen et al. (July 2025), and an open discussion after PaPe/EnvisionBox (July 2025).

Speaker Notes:

• Point new members to the mailing list link.
• Briefly mention Kimmelman’s talk as an example of the kind of methodological exchange the SIG supports — he is PI of an ERC project on universal properties of nonmanual gestures in sign languages.

Call for Contributions

• Propose a talk: work-in-progress, a problem you’re stuck on, etc.
• Suggest a speaker: know someone (internal or external) working on multimodal prosody, gesture, or gaze methods?
• Suggest a topic / question: work on problems together
• Join the mailing list.

Part 3: AI

Part 3: AI

AI Good Practice Norms

• Currently: ad hoc individual practices, for example, is it OK to use AI to summarize papers you haven’t read?
• What: responsible use of LLMs, disclosure and documentation norms, data privacy, AI openness questions.
• Starting point: UzK KI-Richtlinie (27 January 2026)
• When: the CRC 1252 retreat, 5–7 October 2026.

Speaker Notes:

• AI as part of everyday workflows (transcription, coding, writing, analysis), the CRC wants shared, informed norms across projects rather than ad hoc individual practices.
• Also connects to CC/FAIR discussion from Part 1.

Thanks for your attention!