Topic on Talk:Content translation

Apologies if it was requested before.

At present the machine translation implemented in "Context translation" tool, produces typical machine translation gibberish, doesn't convert citations etc. This, in itself, does not make the tool useless (it is good for creating links, moving pictures etc.) and it can be a nice help for an experienced (and sensible) editor.

There is one big "bug" in the tool: the "translations" are, by deafault, published into main space of the target wiki. Yes, one can redirect it into one's own sandbox, but this needs a) some precaution b) some common sense c) a bit of additional tinkering. If any of the three are not implemented, a horrible junk of mumbo-jumbo lands in the main, requiring clean-up, often by other users.

The Content translation tool should by default publish into user's sandbox. Then, when he, using the frame of links etc., translates the article, and is satisfied with the results, he can move it to the main.

Of course, this won't be a bullet-proof against self-satisfied stupidity, but nothing is. For avoiding plain errors, and helping with more careful editing, I believe it will be helpful,

And if somebody is stupid enough to manually redirect the translation output at once to the main, well, his doing entirely, he was not aided by features of CTX tool.

This is actually how it used to work in early 2015, when it was first deployed, and the result was that most translations were never moved out of the draft space and into the article namespace.

If publishing is done directly to the namespace, then yes—sometimes the published page will be bad, and it will be sometimes deleted or moved to the user space, but most of the time neither fo these things happen, and the page stays in the the main space and is improved there. The percent of pages that are created using Content Translation and subsequently deleted is lower then the percent of pages that are created using the source editor or the visual editor and subsequently deleted.

No page is perfect when it is first created.

And what is the problem with bad translation remaining in the drafts...? (except for server's diskspace). User got the translation, realized he actually had to do some work and it was not a magic trick, abandoned the half-created article and good riddance. As long as it is not in the article namespace, who cares.

Taking into account how many VE or text-editor created articles consist od "qqqjajalkapdk", "Jackie has nice legs" and obvious copyright violations, which are all speedily deleted (I'm deleting a few daily), not surprising the percentages of deletions are higher.

Deleted pages are not a problem; not deleted, but shabbily done articles are a problem. So original policy was much more reasonable, IMNSHO :) and should be re-impemented. Let's encourage good work not facilitate shabby work.

PS. Perfect no... But from start to DYK on the Main Page in a couple of edits is not especially remarkable ;)

The problem is not bad translation remaining in drafts. The problem is good translation remaining in drafts. A bad translation will not remain in the main space for long—it will be deleted or moved. A good translation remaining in drafts will be forgotten. And that's what was happening.

Could you substantiate this claim a bit? E.g. point me to these good and abandoned drafts (btw if these drafts were found, checked to be good and checked to be abandoned, they could be moved to the main. I can do).

I don't know how long is "long", but it took me less then 10 minutes (an I had to learn about searching with tags, something I had never done before), to find a couple of pretty bad articles in EnWiki (not gibberish, but pretty bad). None more than 30 days old, but that's because I used Recent Changes for searching.

Content Translation is nearing 200,000 translations. This was true for the first few hundreds. The development team was checking every single one and talking to the users about why aren't they moving them out of the draft. Turned out that they were perfectionists—they were reluctant to move a ~90% done page out of the draft status. Clearly, if they would move it, it would improve more quickly; the perfect is the enemy of the good.

Of course it is easy to find bad articles created with Content Translation; but we agree that it is even easier to find bad articles not created by Content Translation. If you like moving almost-ready articles out of draft status, you'll find thousands and thousands of those in English Wikipedia's Articles for creation, in the Draft namespace in languages where it exists, and of course in the user space in all languages.

Well, I myself shudder at the thought of publishing article with just 10% of words misspelled; call me anal-retentive ;) . Anyway, perfectionists are not a problem - they will find a way to return the article to their namespace, if they consider it too bad to be published.

Now. sloppiness is a problem, because one sloppy editor can easily overwhelm many good ones with his fast produced, large amounts of dubious quality writing. And I don't think that encouraging the attitiude: "yeah, sure, write whatever, publish, somebody will maybe correct it sometime, readers are not paying for it, anyway, so why bother..." is correct one.

Publishing into the sandbox is like saying, ok, please have one more look into the article. Are you sure it is correct? If yes, go ahead, move it to the main article namespace. Perfectionists will tinker no matter what tool they use. But giving sloppy editors a tool to create more slop without even having to type it, and to publish it directly into the main....

Finally, we have a number of well-meaning, not-very-experienced editors, who can a) learn to use their sandbox, get into habit of careful reading etc. or b) publish and don't give a damn (and then be offended if somebody points out their mistakes).

Of course, all of the above applies only, if we concern ourselves in any degree with quality of the articles. If not, than please disregard my remarks entirely.