Welcome to the machine….?
I have long thought that the idea of MT isn’t all that terrible, but it is a shame it’s a) largely promoted and sold by people I wouldn’t trust to honestly tell me the day of the week, and b) translations and data are always stored other than locally, which leads to potential confidentiality issues, especially given a).
(I could also acknowledge the point that if an idea is largely p. and s. by people meeting the above description, perhaps it isn’t that great an idea after all…)
As a spotty undergraduate, I once tried to write a program to translate/conjugate English & French verbs. Nothing mind-blowing. You’d stick in “je donnerai” and it would spew out “I will give”. But it was rules based, which if nothing else makes one realise how much work is involved in even simple tasks such as the example just given.
Thus when, decades (yes, decades) later, I first heard about MT, I was quite impressed someone else had stuck with it and made it work. Then I realised the commercial reality was just large scale processing of words based on a statistical probability of them meaning roughly the same thing. Or something.
Anyway, given that if one of the MT shysters moved into my street, I would be off down the estate agent before you could Google “TAUS”, my conclusion had been that I could only ever use MT if a standalone desktop application came out. I know there are desktop versions of MT engines, but they seem to run on Linux (or Unix-based machines) and require a level of interest and expertise in DIY configuration and building that I do not have and cannot be arsed to muster.
My interest was, therefore, definitely aroused when I heard about Slate, which is a Windows desktop MT application scheduled for release into the wild next year. I stress here I’m just interested, I’m endorsing nothing and no-one. But it does meet my criteria, and I doubt I’m alone in having those criteria, so I’m just mentioning it.
There was a webinar last week, hosted by the Alexandria Project (an organisation I know does not meet with universal approval) advertised on proz (likewise) which now has an active thread on the subject of Slate.
In essence, the idea is that it sits behind your CAT tool and suggests MT translations based on your own work, via an MT engine populated from your own TMs, when your CAT tool itself draws a blank. Akin to enabling the MT option in a CAT tool now, with none of the confidentiality issues (it will work with no internet connection at all). You do need a reasonably powerful laptop to set up the database, and you do need at least 130,000 TM segments (say, 3 years’ full-time work) to start getting useable results.
One could, nay should, wonder how much improvement on the results obtained from sub-segment matching is likely, especially as this latter feature develops in future. (And given that Slate Desktop is not cheap.)
One could, nay should, wonder about the claims it copies your “style”. I’m prepared to accept that it obviously won’t propose terms I don’t use because they won’t be in the TMs used to populate it. I’m doubtful it can replicate style in any meaningful sense just by grinding its way through source and target texts. And once people start making claims that are dubious in one regard….
One could, nay should, keep an eye on which CAT tools it will work with, because it won’t be all of them. Judging from the webinar, if yours isn’t one of them, you could use Slate to feed a source text in and get back a TMX file, which you could then use in your preferred CAT tool. Is that extra step worth it? (It does not, sadly, take much imagination to see how this feature could lead to more PEMT being offered by agencies who have whacked a text through Slate Desktop to provide a TMX, which then provides alleged “matches” for the source.)
But one could, and definitely should, be grateful that at least one person, somewhere, has recognised that the “cloud” (in its imprecise recent meaning of “not on your computer”) is not the be-all and end-all of computing in the 21st century and some of us need, or even just want, to keep our work local.
Anyway, some links:
Proz discussion here
Yahoo discussion here
Strange combo of early bird discount (like, really early!) and financially persuade the developers to build interface with your preferred CAT tool (although they don’t need the money, apparently) here. Don’t watch the video, it’s excruciating.