once more on "mechanical language translation"... first, read:
http://www.nytimes.com/2006/04/05/technology/techspecial4/05lego.html
page 2 of this article talks about amazon's "mechanical turk", a _fascinating_ service unveiled recently which is an intermediary offering human piecemeal employees who will perform smallish tasks for smallish fees. these tasks -- which are called "hits", short for "human intelligence task" -- are administered via the web and picked up by willing workers; examples are "identifying objects in a photo" or answering "is there a human in this photo?" another example of a "hit" that's used by amazon is to "pick the one photo out of these three photos that best illustrates the storefront of this business". each "hit" that a worker performs receives a payment, which might range from as little as one-half a cent up to $.25 or more, depending on the nature of the "hit"... amazon developed this service for its own purposes, and then realized that they could make a business by offering it to other entities that could use it beneficially. amazon handles all the money transactions, which makes it possible to get such piecemeal work done _very_ cheaply. for information on the mechanical turk, read:
http://www.amazon.com/gp/browse.html/104-0655434-6846340?node=15879911
ok, back to the new york times article now, it says this:
One new start-up, Casting Words, is taking advantage of the Amazon service, known as Mturk, to offer automated transcription using human transcribers for less than half the cost of typical commercial online services.
reading further, "casting words" has its "mturk" workers download a digital audio file, then upload a transcript of it, so there's nothing really "revolutionary" about what they do, they're just taking advantage of a rather-cheap labor-force. (my guess is a lot of the people who are accepting "hits" are homebound, perhaps physically-handicapped people are stay-at-home moms looking to pick up some income.) however, there's no real reason that "casting words" couldn't come up with a more innovative methodology, one that would create a trio of computerized translations of a specific passage and then offer them as a "hit" in a "click the best one" fashion... this combination of computerized translation with human review might be a very potent means of obtaining high-quality translation at a relatively cheap price. and yes, someone could also set up a commercialized version of distributed proofreaders to get book-digitization done cheaply... (yes, d.p. does it for "free", but they also get their web-space and bandwidth donated, so their actual costs aren't readily apparent. it's also become increasingly clear they won't be able to keep up with the volume we're getting from the large scanning projects, so at some point, if we want proofed text for all of those scans, we're going to have to face the issue of whether we'll pay for it, or whether we'll decide we don't really want it all _that_ badly.) -bowerbird