[Foundation-l] Free translation memory

[Jimmy O'Regan]

On Fri, 30 Jul 2010 23:22:00 +0200, Nikola Smolenski wrote:
> Дана Thursday 29 July 2010 10:38:20 Samuel Klein написа:
>> There is definitely a "free TM" project waiting to happen.  It would be
>> nice to see translatewiki [for instance] incorporate such a tool, but
>> it may be a nontrivial amount of work.
> 
> At Project Rastko for years now there is the idea of building something
> called Global Translation Project, where volunteers could
> collaboratively translate texts in a manner somewhat similar to
> Distributed Proofreaders.
> 
> To give some detail: the idea is to first parse the original text with a
> rule-based machine translation engine (of course this should be free
> software with free dictionary). 

Hi. I'm a contributor to Apertium (http://apertium.org), a Free Software 
RBMT system which... is exactly what you describe.

> The basic problem that these engines
> have is that they are unable to resolve ambiguities in the text (a
> classic example is sentence "Time flies like an arrow": does it means
> that time is flying as fast as an arrow or that there exist some insects
> called time flies (like there are fruit flies) which like some arrow?).
> This often ends in a mistranslation.
> 
> The crux of the idea is that it would be humans who resolve ambiguities
> in this step. For example, these two possible meanings of the sentence
> would in another language be translated to two completely different
> sentences. A human could then simply pick the correct one. After several
> people have done this for several independent languages, and their
> translations agree, the system would know what is the correct parsing of
> the original text. Then this parsing could be translated fully
> automatically to a large number of languages, and it will be highly
> likely that the translations will be close to correct.
> 

Apertium has a sister project, Tradubi (http://tradubi.com), which is 
developing exactly this.

> An offshoot of this is a crowdsourced dictionary project in GalaxyZoo
> style. Instead of doing battle with Wiktionary's or similar interface,
> volunteers could build a dictionary by solving various simple tasks
> (say, pick a word's gender, or verify that a word is correctly
> declined); if the supermajority of the volunteers gives the same answer,
> the word enters the dictionary.