Marinus Vesseur wrote:
I noticed, when calculating for a big project, that the analysis starts counting 'matches' from 50% upwards. At the same time the standard setting in the workbench is 70% (I think). Any phrase below 70% is not entered into the open unit. Which means I'll be translating the whole thing. Which corresponds to a 'no match'.
Is there a flaw in my logic? Okay, you can adjust the 'match' level in workbench. Does anyone ever do that? If not, then I don't really need the 50-74% bracket in the analysis, do I? Three ranges, say 60-79, 80-89 and 90-99 would do, wouldn't it? Or even less.
Curious for your comments.
My match setting is usually at 50% or even lower. This does often result in not exactly useful matches.
On the other hand, it doesn't take much to turn almost identical translations into matches of 50% or below - different tags, slightly different word order, missing or additional punctuation, different capitalization, etc. So a lower setting can provide useful additional leverage. And even if the overall matched translation is not that great, it often makes the additional use of the concordance unnecessary (i.e. it saves time).
For projects that started out as Word docs and were transformed into tagged html texts, which I sometimes get from one of my clients, even a setting as low as 30% produces useful results.