Clean Sheet?

Peter Damoc pdamoc at gmx.net
Sat Jan 15 06:15:28 CST 2005


On Sat, 15 Jan 2005 03:16:28 -0800, Jonathon Blake <jonathon.blake at gmail.com> wrote:

> Peter wrote:
>
>> Better exposure which might bring better accuracy.
>
> There have been errors in the French L10N version of OOo since day
> one.  They _might_, but probably won't be fixed for OOo 2.0. I doubt
> that OOo is the only FLOSS project where the UI has had known language
> errors for years.
>
>> Think about a centralised translation system where you get
> suggestions for a string based on previous project
>
> You're assuming the same word gets translated the same way every time.
>  That is precisely why the French version of OOo has the errors it
> does have.

Nope, I wan't assuming that... I used the word "suggestions" as in the translator can see how others translated the string in his/hers language or in a language that he knows maybe grouped in "Same project - other languages", "Same language other projects" or "Same project, same language, other place"

> If your L10N project is for a language like Toki Pona, then MAT is the
> fastest way to ensure that the translations are meaningless.
>

Again... I wasn't talking about enforced fully automated memory translations but about "assisted" translations.
Ok... so a full automated should be available but that should no be the "recomanded" way...
An automated translation will not yield a good translation but for most of the languages I think it might provide at least some clue To a pearson that doesn't speak any of the available languages even a crooked translation might help. How much will it help... well that's debatable... depends on the project, on the language and... maybe on luck. :)

>> i track that wrong suggestion
>
> The word might well be correct in one instance, and wrong in another
> instance.  The big issue in translating the UI, is seeing the strings
> in context.  In document translation, everything is in context.
>
>> A centralised translation system might mean "more" information for a translator.
>
> If one of two conditions apply:
> i) The translator can see the strings in context.
> ii) The L10N team is creating the technical vocabulary, as it
> translates the documentation, User Interface, etc.
> then the additional information is useful.  Otherwise it is of
> questionable utility.
>
>> you join the translation community and you might even get an entire
> program automatically translated in a bunch of languages due to
> "common" strings.
>
> And the project lead gets visions of the strings being translated as
> "Invisible, Insane", when it should read "Out of Sight, Out of Mind".
>
> All it takes is for a couple of very badly translated strings to come
> up, and the Project Lead will nix the tool, and go back to using
> OmegaT, Translation Tool Kit, or PO.  With those tools, the
> translation won't be corrupted by other software projects.

Again... the automation should be available as a translation solution for the parts that aren't translated something like.
"there is no po file for X language... would you like to create one via automatic translation system?" and the system should create one based on previous translations... maybe assign fuzzy flag to all strings pending a translator review.

> Go back to how material is typically checked for translation accuracy.
> A third party translated the translated material back into the
> original language.  If the results correspond, the translation is
> good.  If they don't, start again from the beginning.

> Have an L10N project for a "nice" language like Toki Pona.  The back
> translation is not going to be anywhere close to the original, even
> with skilled translators doing both translations.  Add translation
> memory to the mix, and the back translation is going to be pure
> gibberish.  [Granted, Toki Pona exploits some interesting aspects of
> the Sapir-Whorf Hypothesis.]
>
>> I don't think that such a system needs to be incorporated in the website to accomplish CVS synchronisation,
>
> i) Security.
> ii) Current strings being translated.
>
> Do you really want to spend 500 hours translating something, to
> discover that it won't be incorporated into any build?
>
>> again with the exposure.
>
> Exposure means nothing if the results are inaccurate.  And since
> you're relying on machine translation, the results will be inaccurate.
>

in my view exposure means more eyeballs... this should improve accuracy. Again... projects should rely on machine translation ONLY when there is no other alternative... like at the start of a translation.

>> they in turn could benefit from other translation projects.
>
> There is a lot more to a translation project than translating the
> strings of the software.  That probably is the least of what is
> involved.  The big effort is in producing _localized_ documentation.
> It doesn't matter whether it is created from scratch, or translated
> from existing material.  Running it through anything that has a
> Translation Memory is asking for bad translations.  It typically takes
> translators twice as long to fix errors in bad translations, than to
> translate from scratch.

Here I agree completely..

>
>> a web of  translations could provide a lot of automagic.... even for
> big projects like GNOME or KDE.
>
> No automagic.  Still have to have the L10N teams.

automagic for the L10N teams ;)

>> celebrity of the translation project.
>
> Translation projects are celebrity projects when one of three things happen.
> i) It is the first software of its type to be localized for the language.
> ii) The FLOSS project as a whole, is available with significantly more
> localizations than the equivalent commercial software.
> iii) A local political authority shuts down the translation project,
> because it promotes the use of a minority language. [Name a country,
> and it has prohibited a minority language from having a public voice.]
>
> In none of those instances is the use of a web portal in translating
> going to be relevant to the celebrity status of the translation
> project.

Actually I was referring to celebrity as in recognition, like Sourceforge.net is famous for being a good place to start a FOSS project.

>> assigning translation homework to their students.
>
> Like INGOTS does?
>
> No need for any web based tools for that.
>
> Like a wrote elsewhere, the tools have to fit the need of the project.
>  For most L10N projects, if they have an existing site,it makes more
> sense to add the tool to their site, than use still another website.

I disagree... if a website is seen as providing a better service it should be used even if that means that some connection between that site and the project's CVS should be established.

> xan
>
> jonathon

FWIW I do agree that I'm overly naive :)

-- 
Peter Damoc
jack of all trades, master of none
http://www.sigmacore.net/



More information about the rosetta-users mailing list