[lac-discuss-en] Status of translation software
kent
kent at icann.org
Tue Sep 20 19:00:57 UTC 2011
Hello all
As you know, the Google interface we were using with some success a few
months ago was discontinued. Google released a commercial version not
too long ago, and I have been working on incorporating that into the
translation software. This has been a bigger deal than I would have
thought, for several reasons.
The previous version of the software used a general purpose translation
interface that happened to support Google, but unfortunately that
general purpose interface did not, last time I looked, work with the new
commercial API. So it was necessary to re-code the system to directly
use the new API. This was probably a good thing to do on general
principles, but it wasn't as easy as one would like. However, I now
have this working.
However, a further problem emerged in my testing: the new API is a
commercial product. They carefully meter how much text you have
translated, and they are stricter about the amount of text that you can
translate in a single request. In particular, the default limit for a
request is now around 2000 bytes, and a significant part of that is
consumed in the protocol boilerplate. Moreover, the transmitted text
must be encoded in a way that can significantly expand the number of
bytes sent (for those that care, the interface is a REST interface, the
text must be URI-escaped, and all the other parameters count in the
limit - it is 2K for the entire URL).
Previously we just returned an error when the text to be translated was
too large, but this new limit is so small that it really is necessary to
handle the error in a more graceful way. I am implementing this change
at the moment, and I expect it to be done today. If things go as
planned, I will convert the list to use the new code tomorrow.
However, please bear in mind that there have been significant internal
changes, so there may well be issues. Please contact me
(kent at icann.org) with any problems you may find.
Best Regards
Kent Crispin
More information about the lac-discuss-en
mailing list