Page MenuHomePhabricator

Start from a non-blank page when creating an article
Open, NormalPublic

Description

The data that is displayed on the placeholder should be reflected on the edit page the user gets to when clicking "create article"

Event Timeline

Lucie created this task.Aug 1 2016, 2:54 PM
Restricted Application added a subscriber: Aklapper. · View Herald TranscriptAug 1 2016, 2:54 PM

How to generate the text is a very large problem. Different languages have very different grammar, and not every wikis have adequate infoboxes; making it agnostic will cause very large lua module that is hard to maintain and transplant.

Esc3300 added a subscriber: Esc3300.Aug 2 2016, 4:23 PM

It should be possible to define the QID when creating the article.

Otherwise an infobox would initially appear empty when a new article includes it.

Similarly, some Wikipedias could decide to generate basic text based on the QID/properties on that item.

Danny_B renamed this task from [Story] Start from a non-blank page when creating an article to Start from a non-blank page when creating an article.Aug 9 2016, 9:30 PM
Danny_B added a project: Story.
Lucie moved this task from Incoming to Backlog on the ArticlePlaceholder board.Aug 23 2016, 10:57 AM
thiemowmde triaged this task as Normal priority.Sep 5 2016, 3:21 PM
Addshore removed Lucie as the assignee of this task.Nov 3 2018, 2:49 PM
jeblad added a subscriber: jeblad.EditedNov 3 2018, 3:19 PM

Why not use a simple NLG solution where statements or group of statements are reformulated as simple text fragments, and then leave it to the editor how to use, abuse, or remove them? Check out Dale, Robert; Reiter, Ehud (2000). Building natural language generation systems. Cambridge, U.K.: Cambridge University Press. ISBN 0-521-02451-X.

Note also my previous Wikimedia-l post Captioning Wikidata items? This touches on how to use statements from an item to generate prose. This is now a 20+ year old technique, except for attention steering, that is more like 10+ years old.

It should probably not be done in MW-core, aka web servers, as it is compute heavy. Ordinary NLG isn't that compute heavy, but use of RNNs to generate prose is rather compute heavy.