Author Topic: Use of Machine Translation in DVX2  (Read 1569 times)

spiros

  • Administrator
  • Hero Member
  • *****
  • Posts: 684074
  • Gender: Male
  • point d’amour
    • spiros.doikas
    • greektranslator
    • doikas
    • 102094522373850556729
    • lavagraph
    • Greek translator CV
Use of Machine Translation in DVX2
« on: 05 Oct, 2012, 12:49:14 »
Use of Machine Translation in DVX2

This is from Jost Zetzsche's newsletter No 214. I second his comments, sadly, to my knowledge, no other CAT tools have gone that far.

Here is the most intelligent and by far most productive way of using generic machine translation such as Google Translate or Microsoft Bing Translator. It almost always makes audiences of translators gasp in shock when I show it to them (I think they typically gasp because they simply can't believe that MT can actually be useful, but then maybe they just need oxygen to prevent falling asleep during my talk!).

To my knowledge, Déjà Vu X2 is the only tool out there that uses a feature called fuzzy match repair with machine translation. Fuzzy match repair has long been a specialty of Déjà Vu: Every time a fuzzy match is encountered (a fuzzy match is a translation memory match where the source segment in the text that needs to be translated is not quite the same as but very similar to a source segment in the TM), the internal processes of Déjà Vu try to identify what the differences are, whether it knows what the corresponding parts in the target segment are, and whether it can replace them with a translation that it knows.
In practice it works like this:
Imagine you have this segment that needs to be translated.
The TM contains:

Imagine you have this book that needs to be translated.

with the translation:

Stell dir vor, dass dieses Buch übersetzt werden muss.

That would be a fuzzy match. If the terminology database contains the entries segment and book Déjà Vu would automatically replace the translation of book with the translation of segment and come up with:

Stell dir vor, dass dieses Segment übersetzt werden muss.

and you end up with a repaired fuzzy match.

Naturally there are many variables that prevent this from always going so well (such as gender, etc.), but this is how it generally works. In many cases it does a great job, leaving nothing or very little for the translator to do.

This feature now has been enhanced with two processes that make it even more useful. The first is sub-segment matching, a process that virtually all translation environment tools use in some way or the other. What this would do in our example is that segment and book would not necessarily have to be separate entries anymore but, provided that they appear often enough in the translation memory, Déjà Vu would be able to extract that knowledge on its own.

The second feature is in a way a logical extension of that. Here only the old term needs to be known (in our case book) so that it can be marked for replacement. If the new term cannot be found, Déjà Vu goes out to a machine translation engine of your choosing (the available ones presently are Google Translate, Systran, Microsoft Translator, iTranslate4 or PROMT, provided you have the licenses), retrieves the translation for that term, and automatically uses it to replace just that part of the segment.

This works not only with single words but with phrases of any length as well.
Is this always useful? Nah, sometimes it creates more work than the good it does, but more often it's a very helpful and intelligent use of machine translation that I'm surprised more vendors don't offer (please correct me if I'm wrong). And it's definitely a sign of better things yet to come.


This is from Atril's site (as well as the attached guide):

Combine your databases with Machine Translation intelligently for more productivity

Machine Translation has now become incontrovertible in the translation industry. At ATRIL, we put all our efforts into providing the best possible help to all the players in translation industry. As a result, we have designed a Machine Translation module within Déjà Vu X2 that has a comprehensive set of options to provide full flexibility and customization. Machine Translation can be applied at different levels, and in particular you can combine your Translation Memories and Termbases with Machine Translation entries.

We consider that a match repaired with Machine Translation will usually be better than a normal fuzzy match, and better than an entire Machine Translated segment. Your own data is considered to be of higher quality and will always have priority over the Machine Translation results.

Advanced Machine Translation Module in Déjà Vu – Complete set of options for full flexibility and customization

Use Machine Translation at any level you'd like: project, file, segment, sub-segment and term level, after defining your favourite settings in a mere click.
Déjà Vu AutoSearch window can display Machine Translation results: insert it smoothly in a single click or shortcut if the translation is right.
Hit the road to success with the Translation Memory and Machine Translation smart combination: set up the Assemble and Fuzzy Match Repair options with Machine Translation and post-edit the results in one go with the new “Repaired” status filter.
 
NEW Integrations

The following Machine Translation Providers are now directly available in Déjà Vu. No need to install Add-ons to use:
SYSTRAN Enterprise Server
Microsoft Translator
« Last Edit: 22 Dec, 2012, 10:26:59 by spiros »


spiros

  • Administrator
  • Hero Member
  • *****
  • Posts: 684074
  • Gender: Male
  • point d’amour
    • spiros.doikas
    • greektranslator
    • doikas
    • 102094522373850556729
    • lavagraph
    • Greek translator CV
Re: Use of Machine Translation in DVX2
« Reply #1 on: 21 Oct, 2013, 17:23:35 »