Word discount criteria when using CAT Tools

Quick question: Why are there discounts for words that have to be translated while the total gross amount of words is not fully paid?

Hi cesarm,
The discount in fuzzy, full matches and repetitions are due to the fact that you won't translate from scratch those parts. The matches come from the memory and they will help you face the words that are similar or identical even in context (101% matches) and the repetitions don't have to be put everywhere as Memsource like other CAT tools propagates them after the very first translation.

Thanks Dan! That makes sense...

And there are other tools and options that reduce the workload. For instance, memoQ has a Muses option that gives you inspiration instead of matches when there is similar content.

That sounds very interesting.

I've heard about MemoQ's great performance (and that it works quite well with XTRF).

Tell me more about these features that sound so... "poetic".

How do they work?

Yes, that one is very poetic, but like other CAT tools, the objective is to reduce the workload. The capacity of finding matches or recovering content from corpora is something that many companies are improving.

I see, but how is it different form what Memsource already does with regular matches? In what way are these "muses" different and how do they work?

I've never used them that much, but they are basically a corpus that suggests you content for the translation just because of the similarity. Matching is not applied in that case.
Every CAT tool has it's on way of counting words and finding matches in the memory and inside a segment. That's where you can find some differences that the companies keep secret ;)

Oh, I get it now.
It has to do with the way the app groups phrases within segments and how it relates them together. Sort of a different configuration of AI or, more precisely, the different algorithms each application uses to count words and see how they relate to each other.

Very interesting indeed!

It's important to note that first you need to "train" a Muse with a corpus of a few thousand segments.

Yes, like any other resource, a muse is a corpus and you need a good amount of segments or words. For instance, an engine for machine translation or a memory will give better results if they have more content. The settings for training come after that ;)