yndigo

Machine translation “gisting”

Gisting is the term applied in the translation industry to the output of certain Machine Translation (MT) programs — whether free on the internet or proprietary — that give a rough idea of the meaning of a text in a foreign language.

The merits of various MT programs or MT in general have been discussed for decades. Today, translators seem to fall into three camps: the first says that machine translation is decades away from being usable; others claim it will soon put us all out of work; still a third group believes it’s ushering in a new golden age that will uncover mountains of previously undiscovered text for human translation and generate even more work fine-tuning software and “post-editing.” I explored some of these ideas in an earlier post.

Here I just want to discuss the concept of gisting. At one agency I worked for, the term was applied to what is more commonly known as “draft translation” or “for informational purposes,” which means translation performed by a professional translator but not reviewed and not certified for accuracy or completeness.

Somehow the above definition of gisting makes more sense to me than the current term of art, partly because I believe the gist of the source document is produced more consistently by an unedited human translation than by raw machine output. This is not to say that machine translation does not have its place; its ever-growing role in the global economy is undeniable.

But the idea that a monolingual MT user will in fact get the gist from a machine translation depends on a lot of things, most notably the program used and the type of text. I’m thinking of a document review of boxes of colloquial, elliptical emails that paralegals and translators, traditionally, will review side-by-side, looking for responsive files. If, instead, an unassisted paralegal relies on machine translation, chances are good that the output will be awkward to the point that he won’t be able to follow the narrative thread for more than a couple of pages. The gist just might be there but making use of it may be more costly than the old-fashioned way.

Also the form of the word “gisting” is a misnomer in my opinion. As a present participle, it tries to answer the question, “what is the machine doing?” It’s gisting. But in fact it isn’t. It’s trying (in a metaphorical machine-like way of course) to translate very accurately; millions of dollars are spent every year to to get machines to translate, not gist. The fact that they miss the mark doesn’t mean they intended to, any more than a dart player is trying to see how many holes he can put in the surrounding wall.

Thus I think a better term would be “gisted translation” or “gist translation.” Some use the latter term but gisting seems to be here to stay, evidenced by its use among top industry players like SDL and Lionbridge.

Whatever the term used, gisting plays a fast-growing role in the translation industry — both for customers and translators. See Alan Cane’s discussion of this role in a recent Financial Times article.