Start a new topic

Studio's upLift versus CafeTran's fragment matching

After reading this blog entry:, I wanted to replicate the findings in CafeTran.

Why doesn't CafeTran offer a fragment match for web content management?



(22.7 KB)


After setting the fuzzy match threshold to 10%, I get this:


Web content management is now identified as a fragment, though its translation isn't marked or even suggested. Let's see how this changes, when I translate some additional segments.

The translations for the added segments:



Two questions:

  1. Wouldn't it be clearer to colour-code the similarities, instead of the differences? (That is: colour web content management.)
  2. Would it be possible to let CafeTran also colour-code the corresponding fragment in the target language (here: gestion de contenu Web)?


Very interesting, all of this sub-segment stuff. I'm following this topic closely. Studio has some amazing features these days, just a shame the whole program is a massive pain in the ass to use and run by a company that is damaging our industry by forcing us to work with Studio packages/files.


Michael, Count of Hastings: Studio has some amazing features these days

In my view, Sudio is playing catch up, and has been doing so for years. And I wonder if upLift is doing a better job than DejaVu or CafeTran. A bit like Slate Desktop. Very promising at first...



Hmm. Well, one thing that Studio currently does quite a bit better than Déjà Vu, e.g., is auto-complete. I think CafeTran is just as good though. The short period I used Studio, I must say that I was very impressed: when typing, all kinds of useful things would be suggested, many of which are not suggested in Déjà Vu (although Atril tell me they are working on it). Really looking forward to the fabled next version of Déjà Vu, which people say is due to be released in the summer, so very soon. 

PS: luckily, I managed to sell my Slate Desktop licence on to someone else, so I didn't lose all too much money on that gamble. Also, colleagues tell me it actually works quite well, if you have large, clean TM databases.

The Count: Well, one thing that Studio currently does quite a bit better than Déjà Vu, e.g., is auto-complete

Studio's first implementation (way after CT) was a disaster. You had to extract a word list from a large TM (e.g. DGT) or buy one, and it still wasn't good enough. Maybe now it's better, but it's still... playing catch up.

I managed to sell my Slate Desktop licence

Good for you. Don't get me wrong, I like the idea, and I think I like Tom and his attitude towards his product, but I also think it's a waste of time and money for CT and DVV users.


CafeTran's fragment matching (was I criticized on the forum for choosing the now catchy-name "fragment matching" ages ago? :) is only a part of the complex interaction of various auto-translation methods being in focus over the years to provide the best translation of the current segment without interfering with translator's creativity. It also comprises auto-assembling, fuzzy matches with auto-correction (aka. repair), auto-completion and, to some extent, the online MT services. You cannot analyze (or compare) a piece of technology without taking the whole context into account. I believe it is one of the programs greatest features actively developed and refined since the very beginning. I don't know whether (or how) other tools catch up but I think their own solutions also need to be considered in a wider context, that is, how it integrates into their own translation system. Even the very promising Google's MT neural network approach is only a piece of technology often analysed out of context. For one thing, if the translator relies on Machine Translation accuracy too heavily, the MT will "dumb" him sooner or later, destroying his own creative powers. However, it is good to have them running in the background just like any other resource, and auto-typing a phrase or two comfortably if human memory fails. 

1 person likes this
Login to post a comment