Post date: Jan 08, 2013 10:46:51 AM
I have blogged about the problems associated with translation memory before (see post from 26 May 2011). So let me say right now that I use a leading translation memory program, and it is very useful etc. etc. etc.
However, the whole principle of translation memory appears to be based on the notion that the software somehow "knows" when a translation is correct. It doesn't: the program relies entirely - and I mean 100% - on the translator's ability. So if I translate "centrale nucléaire" as "carrot", the translation memory logs that as a correct translation. If a 100% match with a green background shows up the next time that chunk of text is translated, the translator will ignore it. Or maybe not even see it (if using the "translate to fuzzy" feature).
We have to challenge the notion that the translation memory software has some kind of innate capability for translation. It lulls us into a false sense of security, and then... boom! A bad translation is set in stone, and all the software does is recognize the match that we created in the first place.
I realised the full subtlety of this confidence trick when translating a specification this morning. The point is that we have to realize that translation memory is no substitute for good translation. Never let your customers forget that! Unless you are a bad translator, in which case, good luck....