software-store-developers team mailing list archive

[Gary Lasker <gary.lasker@xxxxxxxxxxxxx>]

Hey Aaron,

Thanks for sending this! I have comments in-line.

On 06/16/2011 06:38 PM, Aaron Peachey wrote:
> Hi guys,
> 
> An interesting idea on bug # 780504 to use a google translate api python
> library to translate reviews into a user's local language.
> 
> http://www.catonmat.com/blog/python-library-for-google-translate
> 

Yes, I noticed this bug comment today also, it's definitely a cool idea!
I'm thinking that this might be something that we'd implement on the
server side actually. We could translate reviews on demand from the
client and cache translations on the server such that we build up a set
of translated reviews that we can serve up on request. Or we could even
translate a review at the time it's entered into the n most common
languages to seed the server cache (and then translate into other
languages on request from clients, also adding those translations to the
cache).

Stepping back a bit, though, it seems to me that a very basic problem
with the idea of automatically translating reviews is the potential for
a given translation to obscure or even modify the reviewer's original
intent. I'm not sure how serious a problem this might be. At the very
least, I'd think any translated review would naturally be somewhat
suspect. For this reason I think we'd want to clearly indicate reviews
that have been translated with a comment, something like "Translated
from the original German courtesy Google Translate".

It also occurs to me that having to moderate multiple translations for
every review would likely complicate that team's work dramatically.

> The library looks like it would be quite easy to utilise for this and
> may be a more elegant solution than the one we discussed at UDS. Or
> perhaps the two could go hand in hand. (e.g. "Show reviews in other
> languages, and translate for me")
> 
> Some considerations would be
> - the licence (it's MIT) and whether we can use that in this project
> - the speed - would possibly be two hits on the api per review (summary
> and review body) which could add up as we load 10 at a time (we may be
> able to counter this with an opt-in to translate approach that pre-warns
> of the potential speed hit and network activity), and
> - relying on an external service provider's free service over which we
> have no control

Ok, hmmmm, since I wasn't sure about the MIT license, I took a quick
look at the Google Translate API itself to see the feasibility of
writing our own small Python library to access it. I found that, in
fact, the API is *deprecated* and will apparently be shut down
completely on December 1st of this year. Take a look:

  http://code.google.com/apis/language/translate/overview.html

D'oh! Seems as though the last consideration you listed above turns out
to have been spot-on.

Cheers,
Gary