Note: I (@SSalgaonkar-WMF) am writing this request on behalf of @KStoller-WMF. Please consider Kirsten the submitter and primary owner of this request.

Scoping details

• Use case: As described in the Growth Wiki https://www.mediawiki.org/wiki/Growth/Personalized_first_day/Structured_tasks/Copyedit, the goal is to improve an existing structured task that helps newcomers make simple copyedits to articles. As it exists today, the copyedits structured task relies completely on human intervention. First an editor or moderator must observe that an article needs copyediting, and then they must add the relevant Maintenance template to that article, in order for that article to appear as a copyediting structured task. We believe that language models can help us discover and surface a wider range of articles that are in need of copyediting.
• This use case for structured tasks is somewhat unique, as it will appear first in articles (and then later in the Suggested Edits module). At a high level, we expect the user flow to look like the following: (1) A newcomer arrives on an article page. (2) The newcomer starts reading the article and sees that a specific sentence, paragraph, or section of the article is highlighted. (3) When the newcomer hovers over the highlighted part of the article, they see a structured task that encourages them to copyedit that part of the article. (4) The newcomer can decide whether to accept or reject the structured task. If they reject the task, they can provide feedback about why they rejected it.

• Model purpose: The model should detect specific sentences and phrases that contain errors in spelling or grammar. For each prediction, the model should ideally provide a confidence score.

• Goal: Our main goal is to impact constructive activation, by giving newcomers easy and accessible structured tasks that will help them make successful first edits.

• Prior art: This feature will utilize new UX for structured tasks by making them visible during a newcomer's reading experience. We will also likely need to provide a new model to support this use case. The Research team has already done some exploration into what kinds of spelling and grammar tools could be best fit for this application, and their findings are available here: https://www.mediawiki.org/wiki/Growth/Personalized_first_day/Structured_tasks/Copyedit#Research_results

Prioritization details - Coming soon!

• Timing: When are you hoping to launch an experiment or feature using this model? How flexible is your timeline? Is there any other planned work that's blocked by this experiment or feature?
• KR impact: Which KRs are enabled by this project, and how critical is this project for moving the needle on those KRs?

Other comments

• [Optional] Model requirements: If you have any specific concerns around model performance (latency, cost, etc.) or model output quality (likelihood of false positives, ability to detect all possible instances, etc.), please note them here.
• [Optional] Is there anything else you'd like to share?