This week I had a very interesting discussion with Boris Bastijanić, a Language Technology Expert who, amongst other things, specializes in terminology extraction. The discussion focused on whether all terms in a text should be extracted before starting a project, or whether they should be defined on the fly during the translation itself.
Having had some really bad experiences with automated terminology extraction tools myself, needing to define phrases like desk, office, desk in the office, desk in the office building, et cetera before being even allowed to start translating, I personally prefer to define terms on the fly. This also gives me the advantage of knowing the exact context of a term (which in games — my specialty — is always a surprise), so that I can get a term right in the first place.
Consistency first
There’s no doubt that consistent terminology is really important. You don’t want to translate item A as a table, only to call it a desk a few lines further down. The problem is that automated term extraction tools really can’t distinguish between useful and useless terminology, and also Boris stated they are useless and that agencies wanting to build glossaries before starting a project should use a specialized service, like the one he offers (and if you’re interested in that service, I definitely recommend you to contact him).
But as said, personally I prefer defining terms on the fly, and as a project manager, I expect my translators to do the same. Often this works out pretty well, but translators are only human beings and they don’t always define every term that needs a definition. So one of the aspects of my job is to check whether they haven’t forgotten anything and have translated all terms they defined consistently.
Performing this check manually is incredibly tedious work. You have to go over every line one by one, checking which term needs to be defined and which not. And this week I needed to deliver a 45,000-word project, and I really don’t feel like doing this meticulously check… for 3 languages in a row.
QA in Cattitude
As stated before, Cattitude is not only aimed at translators; it is also aimed a project managers. There’s a special QA mode that iterates over all lines one by one at an incredible speed, performing 33 quality checks per line. Only once all quality checks pass, Cattitude goes to the next line.
One of those checks for example is checking whether the number of [ symbols in the source text matches the number of [ symbols in the target text. Another check is to see whether all terms in the source text have been translated according to the defined glossary in the target text. But of course, this only works if said terms have actually been defined.
Watching Cattitude’s QA doing its work and simultaneously checking no terms have been left out is really important, but also very monotonous and tiring… and as I’m a really lazy person and hate monotony, this aspect of QA was just screaming for automation.
In this particular game, terms were quite easy to spot. They were either between [brackets] or — as often happens in games — Capitalized. So I wrote a new feature for project managers called The TERMinator that automatically spots non-defined terms on the fly.
Only the soldiers in the trenches know what weapons work and what weapons don’t.
Here you see an example of a term (Noodleman, a proper name) that should have been defined for Chinese, but wasn’t.

After defining said term, QA passes and Cattitude moves on to the next line.

This feature is incredibly useful. It allows me to define terms on the fly and it improves the consistency and thus quality of the translation, but there’s no way I could have come up with this idea without dogfeeding (using my own tool myself). Only when you constantly use certain software, you spot what needs to be improved, what can be done more efficiently, what needs to be changed and what needs to be added. Only the soldiers in the trenches know what weapons work and what weapons don’t. You don’t make up these things behind a drawing table.
Now putting terms between [brackets] is definitely not a standard way to define terms; this is really a custom and very project-dependent feature. I reckon that in the long term, this is something that can be defined per project. However, Strings Of Capitalized Words That Are Terms are very common if English is the source language, so this is definitely a feature that can be enabled by default.
Once The TERMinator was implemented, I went through 10,000 words in only a few minutes. There were only three false positives, and at the same time Cattitude found two inconsistencies. A lot better than the useless QA checks in many other CAT tools, that often come up with hundreds of false positives for the same amount of words.

