[Home]PartialTranslation

MeatballWiki | RecentChanges | Random Page | Indices | Categories

Translation follows the Quality <-> Time <-> Cost <-> Quality triangle so common in all engineering efforts. You can only maximize two of the three. While in commercial translation, cost is money, cost for a volunteer organizations typically means people, however time and cost are confounded for volunteer organizations. Like all such endeavors, though, technology--machine translation--has optimized translation quite dramatically; however, machines still lack the intuitive sense that humans possess, or at least the full comprehension of natural language and context (cf. WhatIsContext) necessary to translate not just lexical and syntactic forms, but the semantic meaning. Nonetheless, while perfect machine translation is an unrealistic goal for the perceptible future, PartialTranslation can replace many of the repetitive tasks that human translators would otherwise have to waste time doing manually.

A HumaneInterface for translation would maximize the time spent actually doing the translation and minimize the time spent doing work other than translations. The software should ideally only require the operator to translate words that require contextual translation as those are the words that cannot be machine translated with reasonable accuracy. On average a good translator can contextually translate 2000 words per day (four pages), which isn't a lot, so optimal software should be very good at selecting what words need human intervetion.

The previous translation model was for the human to translate the entire work, word for word. Now the workflow can be partially optimized, putting the operator in a more efficient passive post-editing role. Further, the technology can optimize dictionary, phrase, and lexicon look ups by reducing the space the human needs to search for contextually dependent words--say by showing all possible translations of a contextually dependent word to prompt the human to choose the correct phrase, or providing definition and verb table look ups, or by automatically verifying tenses.

Of course, like all technologies, operators may rely too heavily on the quality of the machine translation and fail to properly PeerReview the material in post-editing. They may also rely too heavily on the grammar check and the dictionaries as is frequently the case for people using MS Word. For a wiki, however, the many eyes principle of PeerReview will hopefully catch enough of these errors over time. But once again the triangle rules: the push for high quality on a wiki comes at the cost of people and time.


I copied this here and to CommunityWiki:PartialTranslation from PhpWiki http://phpwiki.sourceforge.net/phpwiki/MultiLingualWiki with permission of MichaelVanDam? http://phpwiki.sourceforge.net/phpwiki/MichaelVanDam. --MattisManzel

This is quiete elemental. It is actually the basis for world wide justice. I can write in whatever language. The page shows me all original text and all translated and corrected text, but it shows no pure machine translation. I can though request machine translation of any part of text to whatever language. After the machine translation I can correct this text. When I correct the text is a corrected translation and appears as such when the page is called up. If I don't correct it does not. Superior are my selections of languages in my preferences. I choose English and German, I do not choose Chinese. A Chinese-only speaker writes a page in Chinese. I see: there's a new page in chinese, as I can't read it, it does not interest me. A Chinese/English now speaker reads it, thinks it's good, and corrects the machine translation. I see: The Chinese page has been translated to English and now I can read it and - if I think it's worth it - translate it to German. A German-only speaker could now read it and answer. I think it's a good answer and translate it to English. The Chinese/English speaker thinks so as well and translates it to Chinese. Now the Chinese only speaker can read the German-only speakers answer. It is a little exiting to think this on a bigger scale, isn't it? ?CollectiveIntelligence. MattisManzel

This sounds like an interesting idea. If I am reading correctly, you are suggesting every time you write a new (and I guess edit an old) page, that automatically it generates machine translated versions into your chosen languages. If you edit these, they are saved also, if you do not edit them, they aren't saved. I have one question though: what if there is an english page called 'abc' which also has a german translation. Now suppose I edit page 'abc' by adding a *lot* of stuff. Suppose I don't really speak any other language and can't translate very well. Now the german version of 'abc' doesn't match the english version of 'abc'. How is this resolved, or how are german-and-english speakers notified? (I guess this applies to any translation scheme, so it is not meant to be a question about MattisManzel's idea in particular.) --MichaelVanDam?

Not quite like that. No automatic translation takes place unless I request it. No translation makes partof the text, that hasn't been edited. So there are two steps to do: 1.) select the desired part of the text and request automatic translation for it. 2.) edit this automatic translation. Only then the translated text appears as a part of the page from beginning on. I think this might be a way to select the important from the less important. An english-only speaker opening such a page will see: Aha, parts of a text have been translated, so it's possible that it's content in some or another way is more important than the other. If you just automatically translate everything it's one big babel and noone cares for the translation. Having to take two steps to make a translation part of the text, will improve quality. Someone who does that, took time and there must have been a reason for him to have done so. This mostly takes to run all multilingual parallel. Sentence by sentence. Not an easy task, I admit. Multiple languages text is easy to read if languages are in different text colors. The user should also be enabled to filter out content in other languages. Ways to see the page 1.) Only text in the chosen language(s), no translation indication 2.) Text in the chosen language(s), but an indication which parts of text has been translated to other (not chosen) languages. The indication could be a little flag and an slim underline or dotted line under the part of text translated. I.e. I see a text, my settings are: show English and German text, indicate all other translations. A good part of the text has been already translated to German. I read this first. As it interest me I read the complete English text. By the indication I see, parts which haven't been translated to German have been translated to Chinese others to Hungarian, which I both do not understand. I check these in English though, one sentence I think is important and should be translated to German, so I request the machine translation for it, I correct it, done. This isn't someting that works with few particpants, this takes many people. The idea behind is a common evaluation of content. That's the essential thing to do. Another way to do it would be by poll-tools. This contribution is one point, this one is seven points to me. It's my individual subjective vote. A vote can't be but subjective. Communities have to learn the use of poll tools, like babies have to lean to use their eyes. The sum of subjective votings makes common objectivity. --MattisManzel

Hmmm, I like the idea of partial translations... something I had not thought about. I think I understand more clearly what you are suggesting but I guess my original question still remains. If a portion of english text has been translated to german and chinese, and I edit that piece of english text, the german and chinese versions of that piece of text are no longer accurate translations (suppose that I don't have the fluency to attempt a good edit of the german or chinese versions). Are the german and chinese versions to be 'deleted', or could they somehow be flagged as 'out of sync' so someone with the necessary set of language skills can make the modifications? An even more complicated scenario: suppose I change the english version, someone else (who doesn't speak english) changes the german version. Now the changed english version may not correspond to the changed german version. Which takes priority? When someone wants to update the chinese translation, which one should they use as the new basis for the updated translation? Does this make sense? I think for relatively static pages this could work quite well and slowly evolve to a very good quality translation in many languages, but I wonder about a wiki with possibly very dynamic pages if things would remain in sync. I suppose they could if there are a lot of people continually monitoring updates. Also, it seems that wiki updates mostly (but not completely) seem to consist of *additions* to a page rather than corrections. Maybe the scenario I suggest wouldn't be all that common. --MichaelVanDam?

Maybe. I really can't tell, but I'd be very interested to try out what would happen, how people would behave. The technical aspect would have big influence on people's behaviour I think. How should it look like and worl like. Who could be in what way interested in it? As a matter of fact finding a way to deal with the synchronisation problems is finding a way for multilingual (global) collaboration. On whatever issue. Changes should be flagged, yes. The other text now "out of synch" shouldn't be be deleted, no. Will someone translate the new or translate back the old text?. Maybe he'll even write a third thought in another language. It depends on providing convenient translation tools (no automatisation though, two steps to be done) and, yes, the good will of the participating people. Maybe a bunch of tranlators and language interested people could get it going and devlop a certain common behaviour for it, I'd like to hope so. It's a community learning process just like the use of poll tools, which I'm interested in as well. I have been improvising on the software idea on http://wikifeatures.wiki.taoriver.net/moin.cgi/AutomaticTranslation#preview -- MattisManzel


Discussion

MeatballWiki | RecentChanges | Random Page | Indices | Categories
Edit text of this page | View other revisions | Search MetaWiki
Search: