Tuesday, August 30, 2005

The downside of project memories

I find project memories fairly useless: it is true that they include fuzzy matches, and permit a translation company to limit the size of the memories sent to translators, but they are of very limited usefulness as regards terminology, since the concordance feature will only find terms that are in either the 100% or fuzzy matches, but will miss any term that appears in segments where the fuzzy threshold is not reached.

Post script:
One thing that was probably not clear in my original post: I think that project memories are almost useless, but cerainly not translation memory. Project memories are created by analyzing a project against an existing (full) translation memory, so that they may contain only the segments from the full memory that match (either 100% or as fuzzy matches) the segments contained in the text to be translated.
My issue with project memories is that any term contained in a segment of the full translation memory that is not similar enough to a segment in the text to be translated as to be included in the project memory is not going to appear in the project memory, and will, therefore, not appear in a concordance search.

...for example, like...

The ubiquitous "like" has, long taken over most illiterate conversation here in the United States.

Now it's rearing its ugly head even in technical writing, where the writers (one would think) are supposed to know better:

"There are a few things, for example, like some editing functions and the Help files, that are only available in the xxx program on the desktop."

Either use "...for example, some editing functions...", or "...like some editing functions...".

"...for example, like..." is redundant.

Monday, August 29, 2005

One reason I believe the sooner Trados disappears, the better...

...is the very poor quality of their matching engine. For example, I'm currently translating a list of software strings for a telephony system.

For the following string:

468:"Enable DUALmode"

I get a 62% fuzzy match, the original SL of which was:

451:"New building"

Very useful: the number "4", the colon and both quotes are, indeed, identical.

Even better:

For the string:

519:"Value"

I get a 74% fuzzy match, the original SL of which was:

142:"Permanent"


Hopefully, SDLX will be a change for the better.