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

Tue Jul 27 23:38:29 UTC 2010

As an admin in Bengali wikipedia, I had to deal with this issue a lot
(some of which were discussed with the Telegraph (India) newspaper
article). But I'd like to elaborate our stance here:

(The tool used was Google Translation Toolkit. (not Google Translate).
There is a distinction between these two tools. Google Translation
Toolkit (GTT) is a translation-memory based semi-manual translation
tool. That is, it learns translation skills as you gradually translate
articles by hand. Later, this can be used to automate translation.)

Issues:
1. Community involvement: First of all, the local community was not at
all involved or informed about this project. All on a sudden, we found
new users signing up, dropping a large article on a random topic, and
move away. These users never responded to any talk page messages, so
we first assumed these were just random users experimenting with
wikipedia.

Even now, no one from Google has contacted us in Bengali wikipedia and
inform us about Google's intentions. This is not a problem by itself,
but see the following points.

2. Translation quality: The quality of the translations was awful. The
translations added to Bengali wikipedia were artificial, dry, and used
obscure words and phrases. It looked as if a non-native speaker sat
down with a dictionary in hand, and mechanically translated each
sentence word by word. That led to sentences which are hard to
understand, or downright nonsensical.

The articles were half-done. Numerals were not translated at all. The
punctuation symbol for Bengali language (the "danda" symbol: । ) was
not used. (apparently, GTT and/or the google transliteration tool does
not support that).

The articles were also full of spelling mistakes. The paid translator
misspelled many simple words, or even used different spellings for the
same word in different parts of the article.

Finally, different languages have different sentence structures.
Sometimes, a complex sentence is better expressed if broken up in two
sentences in another language. We found that the translators simply
translated sentences preserving their English language structure. This
caused the resulting Bengali sentences awkward and artificial to read.
For example, we do not write "If x then y" in Bengali just by
replacing if and then with the corresponding Bengali words. But the
translators did that, apparently this is an artifact of using GTT.

3. Lack of follow up: When we found the above problems, naturally, we
asked the contributor to fix them. Got no reply. It is NOT the task of
volunteers to clean up the mess after the one-night-standish paid
translators. Given the small number of volunteers active at any given
moment, it will take enormous efforts in our part to go through these
articles and fix the punctuation, spelling, and grammar issues. Not to
mention the awkward language style used by the translators.

So, after getting a cold shoulder from the paid translators about
fixing their mess, we had to ban such edits outright. We didn't know
who was behind this, until the Wikimania talk from Google. Not that it
matters ... even now, we won't allow these half done and badly
translated articles on bengali wikipedia.

Bengali wikipedia is small (21k articles), but we do not want to
populate it overnight with badly translated content, some of which
won't even qualify as grammatically correct Bengali. While wikipedia
may be a perpetual work in progress, that does not mean we need to be
guinea-pigs of some careless experiments. So, our stance is, "Thanks,
but NO Thanks!". Unless, of course, they can put enough commitment
into the translations and fix mistakes.

We welcome automation in translation, but not at the expense of
introducing incorrect and messy content on wikipedia. We'd rather stay
small and hand-craft than allow an experimental tool and unskilled
paid translators creating a big mess.

Thanks

Ragib (User:Ragib on en and bn)

--
Ragib Hasan, Ph.D
NSF Computing Innovation Fellow and
Assistant Research Scientist

Dept of Computer Science
Johns Hopkins University
3400 N Charles Street
Baltimore, MD 21218

Website:
http://www.ragibhasan.com

On Sun, Jul 25, 2010 at 2:12 AM, Shiju Alex <shijualexonline at gmail.com> wrote:
> Hello All,
>
> Recently there are lot of discussions (in this list also) regarding the
> translation project by Google for some of the big language wikipedias. The
> foundation also seems like approved the efforts of Google. But I am not sure
> whether any one is interested to consult the respective language community
> to know their views.
>
> As far as I know only Tamil, Bengali, and Swahili Wikipedians have raised
> their concerns about Google's project. But, does this means that other
> communities are happy about Google efforts? If there is no active community
> in a wikipedia how can we expect response from communities? If there is no
> response from a community, does that mean that Google can hire some native
> speakers and use machine translation to create articles for that wikipedia?
>
> Now let us go back to a basic question. Does WMF require a wiki community to
> create wikipedia in any language? Or can they utilize the services of
> companies like Google to create wikipedias in N number of languages?
>
> One of the main point raised by the supporters of Google translation is
> that, Google's project is good *for the online version of the language*.That
> might be true. But no body is cared to verify whether it is good for
> Wikipedia.
>
> As pointed out by Ravi in his presentation in Wikimania, (
> http://docs.google.com/present/view?id=ddpg3qwc_279ghm7kbhs), the Google
> translation of wikipedia articles:
>
>   - will affect the biological growth of a Wikipedia article
>   - will create copy of English wikipedia article in local wikis
>   - it is against some of the basic philosophies of wikipedia
>
> The people outside wiki will definitely benefit from this tool, if Google
> translation tool is developed for each language. I saw the working example
> of this in Poland during Wikimania, when some people who are not good in
> English used google translator to communicate with us. :)
>
> Apart from the points raised by Ravi in his presentation, this will affect
> the community growth.If there is no active wiki community, how can we expect
> them to look after all these junk articles uploaded to wiki every day. When
> all the important article links are already turned blue, how we can expect
> any future potential editors. So according to me, Google's project is
> killing the growth of an active wiki community.
>
> Of course, Tamil Wikipedia is trying to use Google project effectively. But
> only Tamil is doing that since they have an active wiki community*. Many
> Wiki communities are not even aware that such a project is happening in
> their wiki*.
>
> I do not want to point out specific language wikipedas to prove my point.
> But visit the wikipedias (especially wikipedias* that use non-latin scripts*)
> to view the status of google translation project.  Loads of junk articles
> are uploaded to wiki every day. Most of the time the only edit in these
> articles is the edit by its creator and the  inter language wiki bots.
>
> This effort will definitely affect community growth. Kindly see the points
> raised by a Swahali
> Wikipedian<http://muddybtz.blog.com/2010/07/16/what-happened-on-the-google-challenge-the-swahili-wikipedia/>.
> Many Swahali users (and other language users) now expect a laptop or some
> other monitory benefits to write in their wikipedia. That affects the
> community growth.
>
> So what is the solution for this? Can we take lessons from
> Tamil/Bengali/Swahili wikipedias and find methods to use this service
> effectively or continue with the current article creation process.
>
> One last question. Is this tool that is developing by Google is an open
> source tool? If not, we need to answer so many questions that may follow.
>
> Regards
>
> Shiju Alex
> http://en.wikipedia.org/wiki/User:Shijualex
> _______________________________________________
> foundation-l mailing list
> foundation-l at lists.wikimedia.org
> Unsubscribe: https://lists.wikimedia.org/mailman/listinfo/foundation-l
>