I think the experience I've had with translating matches up well with
the conclusions James has outlined. Even though I'm more likely to
translate content into English rather than out of English, the
principles still hold.
Trying to produce a translation without quality content in the original
article is a frustrating and pointless exercise for the translator.
Unless the original meets certain standards, it would be better and
easier to write the article from scratch in the "destination" language
and translate it back to the "source" language.
Assuming we have a good article in the original language, I definitely
encourage translators to use editorial judgment in what they carry over.
Focusing on the lead section is one possible approach. In general,
because we are trying to translate information and not literature, we
should have different priorities. It is more important that the
translation maintain fidelity to the facts than to the language and
structure of the article. Sometimes it makes sense to pass over certain
details, even a beginning-to-end translation might come out a bit
condensed. As one reason for this, making some details accessible to the
cultural audience in the new language can at times require a fair amount
of elaboration, more than may be ideal for the context under discussion.
The best approach to use is one of adaptation as much as translation.
I don't have strong feelings about whether a paid model will work, or
work better than purely volunteer activity, but I would be open to
seeing a trial. The essential thing is that we find translators who can
understand and apply standards of quality in their work, much like we
would expect if they were editors writing entirely new articles.
On 2/24/2018 5:26 AM, James Heilman wrote:
We learned a few things during the medical translation
started back in 2011:
1) You must start with high quality content and thus all articles are
extensively improved before being proposed for translation.
2) A lot of languages want "less" content than is present on EN WP. Thus we
moved to just improving and suggesting for translation the leads of the
3) The "Content Translation" tool developed by the WMF made efforts more
efficient than handing around word documents. Would love to see that tool
improved further such as having it support specific lists of articles that
are deemed ready for translation by certain groups. Would also love the
tool to have tracking metrics for these types of projects.
4) We used volunteer translators mostly associated with our partner
Translators Without Borders. One issue we found was that languages in which
their are lots of translators such as French, Spanish, and Italian there is
often already at least some content on many of the topics in question. The
issue than becomes integration which needs an expert Wikipedia. And for
languages in which we have little content there are often few avaliable
5) With respect to "paying per word" the problem is this would require
significant checks and balances to make sure people are taking the work
seriously and not simple using Google translate for the 70 or so languages
in which it claims to work. We often had translations undergo a second
review and the volunteers at TWB have to pass certain tests to be accepted.
6) I hired a coordinator for the translation project for a couple of years.
The translators at TWB did not want to become Wikipedians or learn how to
use our systems. The coordinator created account like TransSW001 (one for
each volunteer) and preloaded the article to be translated into Content
Translation. They than gave the volunteer translator the user name and
password to the account.
7) Were are we at now? There are currently just over 1,000 leads of
articles that have been improved and are ready for translation. This
includes articles on the 440 medications that are on the WHO Essential
List. We have worked a bit in some 100 languages. The efforts have resulted
in more than 5 million works translated and integrated into different
Wikipedias. The coordinator has unfortunately moved on to his real job of
teaching high school students.
8) The project continues but at a slower pace than before. The Wikipedian
and retired orthopedic surgeon Subas Chandra Rout has basically single
handedly translated nearly all 1,000 leads into Odia a language spoken by
40 million people in Eastern India. The amazing thing is that for many of
these topics this is the first and only information online about it. Google
translate does not even claim to work in this language. Our partnerships
with WMTW and medical school in Taipai continue to translate into Chinese.
There the students translate and than their translations are reviewed by
their profs before being posted. They translate in groups using hackpad to
make it more social.
I am currently working to re invigorate the project :-)
On Sat, Feb 24, 2018 at 5:51 AM, John Erling Blad <jeblad(a)gmail.com> wrote:
This discussion is going to be fun! =D
A little more than seventy Wikipedia-projects has more than 65k articles,
the remaining two hundred or so are pretty small.
What if a base set of articles were opened for paid translators? There are
several lists of such base sets. We have both the thousand articles from
"List of articles every Wikipedia should have" and and the ten thousand
articles from the expanded list.
Lets say verified good translators was paid about $0.01 per word (about $1
for a 1k-article) for translating one of those articles into another
language, with perhaps a higher pay for contributors in high-cost
countries. The pay would also have to be higher for languages that lacks
good translation tools.
I believe this would be an _enabling_ activity for the communities, as
without a base set of articles it won't be possible to build a community at
all. By not paying for new articles, and only translating well-referenced
articles, some of the disputes in the communities could be avoided. Perhaps
we should also identify good source articles, that would be a help.
Translated articles should be above some minimum size, but they does not
have to be full translations of the source article.
A real problem is that our existing lists of good articles other projects
should have is pretty much biased towards Western World, so they need a lot
of adjustments. Perhaps such a project would identify our inherit bias?
Wikimedia-l mailing list, guidelines at: https://meta.wikimedia.org/
wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/
New messages to: Wikimedia-l(a)lists.wikimedia.org