Machine translation gets it spectacularly right [entries|reading|network|archive]
simont

[ userinfo | dreamwidth userinfo ]
[ archive | journal archive ]

Thu 2011-04-14 17:15
Machine translation gets it spectacularly right

In a debate at work about development workflow, I somewhat whimsically responded to a suggestion by writing ‘Jira issues non sunt multiplicanda praeter necessitatem’. (For those lucky enough not to have encountered it, ‘Jira’ is the name of a bug tracker.) I wondered briefly whether that would be too incomprehensible to post (but decided not, and posted it anyway).

Chris had the obvious idea that if someone didn't understand it, they might try pasting it into Google Translate, and tried it. Google Translate renders the sentence into English as ‘Jira issues must not be multiplied beyond necessity’. Very good; I'd have preferred should, but must is acceptable.

But what's impressive is that if you switch the target language to something else, it turns out that it's not (as I initially assumed) identifying the language as Latin, applying a Latin-to-English translation step, and giving up on the English fragment and passing it through unchanged. No: it actually managed to translate the English part from English, and the Latin part from Latin, both into the specified target language, and as far as I could tell it even got the two fragments in their correct grammatical relationship. I may be out of date with the state of the art, but I was impressed by that!

[xpost |http://simont.livejournal.com/232244.html]

LinkReply
[personal profile] ptc24Thu 2011-04-14 16:41
Hmmm, given that the sorts of English texts that are routinely translated (e.g. Canadian Hansards make wonderful parallel corpora) often include Latin phrases anyway, it could be that this is an unexpected fringe benefit of statistical machine translation.
Link Reply to this
[personal profile] ewxThu 2011-04-14 17:54
Could it be translating Latin to (say) French by translating Latin to English and then English to French?
Link Reply to this
[personal profile] jackThu 2011-04-14 18:41
"must" might be better :)
Link Reply to this
[identity profile] https://me.yahoo.com/a/yCTlOyQXx.q6YbKK1vGXgmS38A--#a37f0Fri 2011-04-15 00:33
I tried a few more languages; when I asked for Swedish I got this:

Roundup frågor skall inte multipliceras utöver nödvändighet

[In case you hadn't guessed, Roundup is a different issue tracking system]
Link Reply to this
navigation
[ go | Previous Entry | Next Entry ]
[ add | to Memories ]