[Foundation-l] Is Google translation is good for Wikipedias?

[Mark Williamson]

That's absolutely a problem that should not be overlooked. Despite
what I said in the other thread about content equivalency across
languages, I think this is quite a different issue. A competent
translator must take into account context and fluency, and often
direct translations do not fit, even when they're grammatically
correct. Language is a living organism consisting of more than just
words and grammatical rules, we use lots of idioms and turns of phrase
that are unique to our languages (or even our local dialects).
Ignoring these things in a translation can generally give us output
that is understandable, but not necessarily "good" - it can come out
sounding stilted, awkward and contrived, at best.

The latest version of GTTK allows the merger of segments, i.e. two
sentences in the original can be merged into one and translated
accordingly. However, I think it's important to not lose sight of the
fact that GTTK is just that: a toolkit. It is not the end-all solution
for article creation on any Wiki, nor is it an evil entity that goes
around dumping poor-quality text on our projects. It is what we make
it - I can use GTTK to produce a translation that is a good, prosaic
article if I am willing to put in the time and effort to adapt text
from one language to another, which is really the job of the
translator anyhow.

This doesn't take away from the problem raised by M. Yahia about
community, but I do wonder about that. Informational cannibalism has
been common in our community between languages for a long time,
ranging from borrowed parts of articles to translations of full
articles. This hasn't seemed to be a problem in the past, before GTTK.
What struck me were the phrases "very bad sentence structures" and
"bad jargon translations". Aren't we talking about professional
translators here, people who do this for a living? An excellent
translator should not only know their source language well, they must
also be intimately familiar with the ins and outs of their target
language, beyond just the fact of being a native speaker. If a
translation doesn't sound natural in the target language, that's not
because it's a translation, it's because either 1) it's a poor
translation or 2) it wasn't natural sounding in the source language,
either! In most cases, I'd guess 1) since as a translator I'd rather
compensate for people's grammatical mistakes than attempt to re-render
them in another language.

The key to a great finished translation, in my opinion, is good
proofreading. Before proofreading, your translation is like a block of
unfinished wood. Rough, but still suitable for some uses. After
proofreading, it should be polished. A good translation should leave
the reader unable to tell whether the text was translated or if it was
originally written in the target language, with only very rare
exceptions.

-m.

On Thu, Jul 29, 2010 at 12:37 AM, Ziko van Dijk <zvandijk at googlemail.com> wrote:
> Has anybody more information about what Google exactly told the
> people? A link? To whom was this call for participation directed?
> This issue "Translation memory" is another problem, another divergency
> of interests. We Wikipedians want to write good articles in our
> languages, that often means that we do not translate 1:1 but shorten
> and customize. But Google wants 1:1 translations for its Translation
> memory. And, of course, its the big numbers Google is interested in to
> achieve better automatic translations in the end.
> Ziko
>
>
>
> 2010/7/29 Muhammad Yahia <shipmaster at gmail.com>:
>> On Wed, Jul 28, 2010 at 6:56 PM, Mark Williamson <node.ue at gmail.com> wrote:
>>
>>>
>>> I heard that for the Swahili Wikipedia contest at least, they gave
>>> away prizes... but perhaps they should've included a requirement that
>>> the articles they created be rated as "good" by the community, not
>>> full of errors and nonsense sentences, and that all project
>>> participants who want any chance at winning must respond to all
>>> talkpage messages within 72 hours (or something like that).
>>>
>>>
>>>
>> I have been involved with 2 big pushes by Google in the Arabic Wikipedia,
>> one of them was by professional paid translators, the other was done
>> completely by a volunteer organization in collaboration with Google. I
>> supported both efforts heavily. In the latter, they recruited university
>> students mostly to do the work and there was very little to earn beyond
>> recognition. All the problems mentioned above plagued both efforts, and
>> while the second one had slightly better results than the first, the vast
>> amount of translated articles lay ignored in the user space (that's what
>> the consensus on ar.wp was, confine them to their user space until deemed
>> good), the efforts to contact and teach either the volunteers or the paid
>> translators were futile, and the articles had some very awkward sentence
>> structures, some very bad jargon translation, etc.
>>
>> I have reached the opinion that the gradual nature of collaboration in
>> Wikipedia is what makes our good and excellent articles what they are. I
>> think a very little percent of wikipedians started by writing a full length
>> article, instead most of us started by a small edit in another article, and
>> a bigger edit after it and so on. By the time we began writing whole
>> articles, we had enough knowledge of the community and the wiki syntax to
>> produce good results. Whenever someone has a question about terminology, it
>> gets discussed on the VP, whenever someone is unsure, he recruits other
>> people to review or help. This was all missing from the effort and I think
>> what caused most of the problems.
>>
>> --
>> Best Regards,
>> Muhammad Yahia
>> _______________________________________________
>> foundation-l mailing list
>> foundation-l at lists.wikimedia.org
>> Unsubscribe: https://lists.wikimedia.org/mailman/listinfo/foundation-l
>>
>
>
>
> --
> Ziko van Dijk
> Niederlande
>
> _______________________________________________
> foundation-l mailing list
> foundation-l at lists.wikimedia.org
> Unsubscribe: https://lists.wikimedia.org/mailman/listinfo/foundation-l
>