Dec
2
The gift of gas
Filed Under ReadyMade, Satire, Recent stories | Leave a Comment
I’m told that the holidays are “just around the corner,” a time when we are all looking for that “special something.” Inspired by an informative holiday press release, I’ve thus put together a collection of practical gifts in latest issue of ReadyMade, with something for “everyone on your list.”
This entry was posted on Saturday, December 2nd, 2006
del.icio.us Digg RedditDec
1
What’s the meaning of this?
Filed Under Technology, Wired, Recent stories | Leave a Comment
My most recent feature for Wired, which has been out on the newsstands for a couple weeks, went up online today, here. It’s about a New York company, Meaningful Machines, that has come up with a novel technology for doing machine translation — software-based translation of text from one language to another. I wrote about the company briefly in Safe, when they were at a much earlier stage. The original ideas behind the software, which are simple but fairly ingenious, sprung from the mind of Eli Abir, a former used car salesman in New Jersey with no college education and no formal training in computers, artificial intelligence, or linguistics.
Machine Translation, or MT, is a thorny problem, and the field has gone through years of eye-rolling hype and resultant vaporware. For a great comprehensive look at the area, check out Steve Silberman’s Wired story from a few years ago. But expectations are so low now (given the limitations of older systems like those currently used by Babelfish and most of Google Translate), that perhaps some new advances are being overlooked. The hottest area developing over the past few years has been statistical MT, the use of parallel translated text to train MT systems using statistical algorithms. As the story mentions, the outfits getting the best results with that method are Google (who hasn’t yet put many of their research systems up on Google Translate), Language Weaver, and IBM. Both Language Weaver’s and IBM’s are being used extensively by the military and a growing number of businesses.
Unfortunately the latest results from the NIST evaluation, which looks at MT systems head-to-head, weren’t out when the story closed. You can now find those here. Google, for the second year running, came out on top, and Language Weaver fared quite well. (Despite NIST’s new, bright-red disclaimer that its evaluation is not a competition, everyone in the MT research world still considers it the big show).
Meaningful Machines has yet to enter the NIST (non)competition, which is the most common criticism I heard about them. So there’s really no way yet to know whether their early promise will translate into a system that trumps the state-of-the-art, or they’ll end up on the scrap heap of MT failures. They, at least, believe they are on their way to human-quality translation.
This entry was posted on Friday, December 1st, 2006
del.icio.us Digg Reddit