I sometime wonder what the programmers at SDL/Trados think that “similar” means, but I’m sure that what they think must be different from what most translators think.
Take for example the strings
- “- LEAD DESIGN -”,
- “- LEAD PROGRAMMER -”, and
- “Lead Design”.
Most translators, when asked to translate “- LEAD DESIGN -”, would find the translation of “Lead Design” more useful than knowing the translation of “- LEAD PROGRAMMER -”.
Seemingly, Trados programmers disagree: as you can see from this screenshot, 
Trados considers “- LEAD PROGRAMMER -” a 75% fuzzy match for “- LEAD DESIGN -”, while “Lead Design” only gets a 67% score.
How the program arrives at this result is clear: both strings 1 and 2 follow similar patterns (all caps, leading a trailing dashes), while 3 doesn’t.
But writing a more intelligent algorithm shouldn’t be all that difficult: a better algorithm would give more weight to the actual words, and not to such irrelevant characteristics as case or dashes.
Trados programmers should, in short, try to think of what is useful to translators, and implement that in new algorithms, rather than rely on old ones that have probably not changed in over ten years.
