Miroslav Pošta first posted his article July 4, 2017 on his blog, http://www.translatoblog.cz/prekladac-v-pocitaci/ as part of our Blogging Translator Review Program. This is Miroslav’s abstract of his original blog article for our English readers. Thank you Miroslav!
Miroslav Pošta’s English abstract
Have you ever wondered what it would be like if you had a little Google Translate in your computer that would learn from your translations and could be used to translate even those confidential texts that cannot leave your machine? With Slate™ Desktop, this is possible.
First, you have to feed your translation memories into the programme and let it process the data, ideally overnight. What you get is a customized translation engine which mimics your choices. It is hard to say how much data the programme needs – it depends on the language combination you work in and the type of the text you need to translate – but from users’ experience, engines built with memories of 80 thousand and more segments tend to produce good results. (The original article illustrates the quality of output by comparing results from two engines, one built with 214k segments, the other with 10.5k segments.) Anyway, this doesn’t mean that only size matters. Users of Slate™ Desktop report that subject-specific translation memories with 100,000 segments often work better than translation memories with a million of segments.