News for hackers. Stuff that matters.
Re:Why Deep Learning Cannot Be Applied to Natural Languages ... by nl on Monday January 02, 2017 @09:29PM (#13306200)
Every translation is done by a single human

No, the "perfect score" is an "averaged" translation by multiple humans, measured by BLEU score.

Re:Why Deep Learning Cannot Be Applied to Natural Languages ... by ma2rten on Monday January 02, 2017 @09:57PM (#13306332)
No, they asked humans to rate translation quality from 0 to 6. See section 8.2 of the paper.
Re:Why Deep Learning Cannot Be Applied to Natural Languages ... by nl on Monday January 02, 2017 @10:12PM (#13306399)
Oh, you are right, sorry. I'm confusing it with a different paper.