Cafetran is a CAT tool with translation memory, glossary lookup, and web searching. It also supports sourceless translation using an autocomplete function.
Cafetran is written in Java. Cafetran is not free, but a time-limited trial version can be downloaded from its web site.
What makes Cafetran interesting is that it uses XLIFF for its native TM format. Documents translated in Cafetran automatically generate an XLIFF file as you go along.
Cafetran is fairly complicated to use, and the menu bar is very non-standard. The help file contained on the web site is not very comprehensive or well organised.
Source files include HTML, OpenDocument, WordML (Microsoft Word 2007), etc.