Summary

  • A March 2010 update let contributors adopt an unowned sentence and edit it in place. Ownership carried practical editing rights.
  • After the January 2011 quality day, Tatoeba reported 2,448 fewer unowned sentences, alongside 422 corrections and 1,370 OK tags.

On Tatoeba, a sentence could be looking for a family. This was a surprisingly practical domestic arrangement: become its parent and you could fix its mistakes.

The project’s 6 March 2010 update explained the mechanics. A contributor normally owned a sentence they added. Unowned sentences could be adopted, giving the adopter permission to edit. Previously, clicking the adoption button had sent the volunteer to an information page. The update removed that detour: adopt, then correct, right there.

A tiny interface change shortened the journey from noticing a mistake to doing something about it. The punctuation could stay; the paperwork was getting lighter.

Five seconds, multiplied by a corpus

By 30 April, the project was proposing a checking team. Volunteers would inspect sentences in their native language, working through assigned lists and giving unowned material priority. Grammar and spelling came first; the accuracy of translations would have its own later round.

The proposal was specific about judgment. Preserve the meaning. Usually leave archaic language alone. Improve awkward phrasing where appropriate. If unsure, leave a comment for discussion.

Its optimistic arithmetic supplied the punchline. With roughly 370,000 sentences and five seconds allowed per check or correction, the job still came to more than 500 hours. Five seconds is a brisk glance. A corpus is very good at turning brisk glances into a working life.

The post explored private work lists, completed batches and future voting. These were plans being worked out in public. The immediate ambition—get every sentence a responsible reader—already exposed how much attention the collection needed.

The parent has left the building

Ownership also created a snag. A helpful visitor could spot a problem in somebody else’s sentence and have no ordinary right to fix it. The owner might never return.

That was the problem described in the 24 May moderator guide: correction and deletion requests were accumulating around unattended contributions. Moderators could edit or delete any sentence, and link or unlink sentences.

The guide gave these broad powers a deliberately slow fuse. Editing was a last resort, after the latest correction request had sat for at least two weeks without a correction. Before deletion, the moderator was to ask the owner. Public contribution logs helped make interventions visible.

A sentence had acquired both a parent and a procedure for dealing with the parent’s absence.

A Sunday for the repairs

The announcement for 23 January 2011 invited volunteers to improve the corpus together. They could adopt sentences, work through checking tags and discuss awkward cases in the project’s IRC channel. The announcement also warned that older quality articles were already out of date: this little maintenance system was still changing.

The next day’s results put numbers on the work. Unowned sentences fell from 254,779 to 252,331, a net decline of 2,448. Twenty-four contributors had adopted sentences; CK alone accounted for 873 adoptions.

The report separately counted 422 corrected sentences, 1,370 sentences tagged OK and 503 comments. CK’s comments frequently pointed out duplicates. A useful day could involve taking responsibility, changing words, checking them or telling another person where to look.

Tatoeba’s family vocabulary made the work sound cosy. Its records show the less cosy engineering underneath: someone needed permission, someone needed time, and somebody had to be available when the first person disappeared.