Page MenuHomePhabricator

Invite volunteers to review Tone language model (v2)
Open, Needs TriagePublic

Description

Tone check and Improve Tone Suggested edit are based on a model that detects terms that may violate Wikipedia's neutrality policies.

The purpose of this task is to invite volunteers to evaluate the model's assessments so that we can identify if the model matches the prediected tresholds.

The evaluation will start on October 3, 2025.

This is a second model evaluation, the first one was covered by T388471. The end goal remains to provide Tone check to all wikis.

If you want to help us to review of the Tone language detection model, please signup here.

Planning

  1. Update https://www.mediawiki.org/wiki/Edit_check/Tone_Check/Model_evaluation with the new list of wikis
  2. Ask the Machine learning team about when the links to each model will be available. This would define the start date.
    • The dates are available in T394448
  3. [milestone] Inform the concerned communities that we need volunteers to evaluate the model, by signing up on the Model Evaluation page (link above).
    • Publish a message on Tech News
    • Check with other Ambassadors and Movement Comms colleagues, as they may have heard that their communities want to test Tone check.
  4. To the users who signed-up, send them the link. We need a minimum 3 users/wikis. The more, the better, as some users won't have time to participate, and diversity is a good bonus.
    • The signup page has a list of volunteers ready to help; these users should be contacted directly.
  5. [milestone] Define the end date for the evaluation.
    • Users have 10 days to review the model (at least one week that includes a full week-end)
  6. The ML team has data on how many diffs have been reviewed. Posting a reminder may be needed for some languages.
  7. Thank the users who participated.
Ressources

List of wikis

The list of languages for this phase is final. Please avoid adding another language.

Details

Other Assignee
Trizek-WMF

Event Timeline

Per what Benoît and I discussed offline today, I'm boldly assigning this task to Johan as he'll be supporting Samuel and Habib in completing this work.

Trizek-WMF updated the task description. (Show Details)

The evaluation will start on October 3, 2025.

I set up the invite page, as I shared the invite at the CEE meeting, where many users who speak the languages we target were present.

By the date of Thursday October, 2nd, all communities and volunteers listed on the Mediawiki page were contacted and notified by the start of the evaluation.

We finished this evaluation, with all languages listed participating (except one).

4 languages will have a follow up, as the data provided had some issues.