Submissions:2025/Tone Check: a case study in developing ML to support editors

From WikiConference North America
Jump to navigation Jump to search

This submission has been noted and is pending review for WikiConference North America 2025.



Title:

Tone Check: a case study in developing ML to support editors

Type of session:

Lecture (15-30 min)

Session theme(s):

Credibility, Future of Wikipedia

Abstract:

In this session, we’ll talk about the development of Tone Check, a feature designed to help new Wikipedia editors identify and revise language in their edits that may violate community policies by being promotional, derogatory, or otherwise subjective. Tone Check continues to be built in partnership with volunteers, through sessions like this one. This project is part of an ongoing commitment to fostering a more welcoming editing experience and proactive moderation process by offering people feedback while they are editing - before changes are published. We will provide an overview of the end-to-end development process, including identifying contributor needs, model design and training, model evaluation and testing, and feature integration into user-facing workflows. While machine learning plays a central role in how the feature works, the session will focus on how Tone Check addresses real challenges in editing, and how collaboration with communities is shaping both its design and impact.

Author name(s):

Isaac Johnson, Peter Pelberg, Sucheta Salgaonkar

Wikimedia username(s):

Affiliated organization(s):

Wikimedia Foundation

Estimated length of session

30 minutes

Will you be presenting remotely?

I will present in-person

Okay to livestream?

Livestreaming is okay

Previously presented?

Not officially but there is existing documentation and prior discussions for Tone Check: https://www.mediawiki.org/wiki/Edit_check/Tone_Check#History

Special requests:

N/A