Multiple MT engines used at the same time. In older versions of memoQ you have to choose one single machine translation engine to be used during translation. With the new framework you can get matches from all of them at the same time, so if one is bad, you still can pick the good one.This was done, of course, in response to popular demand, and in this age of crowded schedules and crowdsourcing, the wise wannawanks know that crowds rule, especially in the Cloud, which in the Once and Future Middle Kingdom and nearby islands of past glory is usually known as the Crowd.
It's not enough to have one lump of crap generated by a computer algorithm as a restful alternative to the strain of thinking of a reasonable linguistic solution to the challenge of translation by an untrained mind. Because leading lingolemmings and their suppliers at SDL and other members in the localization cartels are fond of piling their MT higher and deeper, marketing logic at Kilgray dictated once again that mo be betta. But is it?
There's a little tactic I learned about many years ago in sales training. I try not to use it, because I consider it to be unethical in ordinary daily life. But car salesmen and some of the carnies I've seen operate in translation technology sales apply it quite effectively.
After you choose your new car, you are ushered into the sales office to discuss the details of the contract. What color do you want? Green, I'm a hunter. The 2.4 liter engine, right? No, 2.0 liters is plenty of power for me. Would you like that with a roof rack? No. What about a towing hitch? Well.... Here we have 20 models of car stereo to choose from, take your pick! Oh yes, and what kind of underbody anti-weathering treatment would you like? We have three options to choose from. What rim style would you like for the wheels? Would you like the three year extended warranty or the ten-year option with the alien abduction protection plan?
Question after question. Decision after decision. Decision fatigue.
At the end of a long day of applying his wisdom to the ordering of his kingdom, Solomon might have had some difficulties. Today's courtroom judges certainly do. Over the past year I've read interesting studies of how workload and diet affect the quality of sentencing decisions in the course of a day. I used to avoid 7 am classes in college, but if I ever go on trial for murder, I want to be the first case heard that day.
What's that got to do with memoQ and multiple MT suggestions displayed, or for that matter SDL and all the others who offer this particular path over the translation quality decision cliff?
If you're the guy who got that cool technical marketing job translating the newsletter for my favorite equipment manufacturer, I'll tell you what it means: mo be betta. Praise the Lord for this feature, which will help you do the job I deserve! If you further enhance your skills by adding a big portion of MT post-editing to your plate, reading and believing the content generated by TAUS, the Common (Non)sense Advisory and other advocates of the 7-Up Concept of Quality (extra points to the first one who groks that), then all will be well in my world of translation.
I had a problem with rats stealing my chicken food a few weeks ago. Finally, I made a reluctant visit to the local feed store and bought a box of tasty blue pellets, which were eagerly carried to the nest and consumed. For a whole week. I had to get an extra box, but just before it was finished, things went quiet. As the Germans say, gut Ding braucht Weil.
Thank you, Kilgray, for adding a feature in memoQ 6.2 which will make good translators stand out even more.
Tasteless machine translation with no hint of human intelect, shifts, lexical and grammatical devices or any other resource to make the text more colorful, like neutral pH?
ReplyDeleteI'm kinda hoping that Kilgray will go one better on "Don't Press This Button" and add a "Blender" button that will mix all the MT suggestions into a smooth hemlock milkshake. Just the right refreshment when surfing the Content Tsunami with today's technogurus.
ReplyDeleteI think if they really wanted to help the industry, one of the MT engines could be replaced by an appropriate adaptation of this: http://pdos.csail.mit.edu/scigen/
ReplyDeleteEntire computer manuals could be translated that way, and nobody would be able to tell the difference.
Of course they may have to pay a royalty to SDL, as the use of such a system is the only logical explanation for the MultiTerm so-called "documentation and help system"
I like the idea of the blender button for automated MT post-editing - no other human input required. Wonder if you could set it up as a batch task?
ReplyDeleteJayne, have you been intercepting my secret e-mail correspondence with the developers on The Red Computer? Why yes, it is configurable as a batch task, but only using the Client API, which is available only in the PM Version for the Few and Proud....
ReplyDeleteMy secret is revealed. All your base are belong to us.
ReplyDelete