This story, “Google Translate AI invents its own language to translate with” caught my eye for an odd reason.
Long-time Saskatchewan residents will recognize the word “GigaText.” As I’ve noted elsewhere, I’m working on a book about the Progressive Conservative government of Grant Devine, which held power in Saskatchewan from 1982 to 1991. One of the boondoggles that government mistakenly invested in was a company called GigaText, which claimed it could use computers to translate Saskatchewan laws into French.
The government had to comply with a ruling by the Supreme Court that Section 221 of the Northwest Territories Act contained French-language guarantees that were still valid in Saskatchewan, and thus had to be either respected or repealed. As a result, The Court stated, all of Saskatchewan’s English-only laws were invalid, and the provincial government had to either translate all laws into French within a reasonable time, or invalidate, as soon as possible, the right to French-language laws. The government chose to do both: invalidating the right to French-language laws, but voluntarily translating some key laws into French as well.
In all the stories I’d ever read about GigaText (and I was a newspaper editor at the the time), nobody had ever bothered to explain the technology behind it. Clearly there had to be something real there, even though the company never managed to do what it said it could (and had a principal investor, a Montreal businessman named Guy Monpetit, who it turned out was as shady as they come.)
I wanted to put something in MY book about the actual GigaText technology. A little Internet digging scared up an old paper that described the theoretical underpinning of it.
If I may quote myself from the book’s current draft, at the heart of GigaText was:
“…a computer translation system being developed in Winnipeg by Dr. Douglas A. Young, a professor in the Department of Computer Science at the University of Manitoba…In 1983 Dr. Young published a theory that he believed could be the basis for effective computerized language translation—something that did not yet exist. Having noted that much of our knowledge is acquired through our senses and actions, rather than in the form of words, he theorized that the meanings of words and sentences, no matter what language they originated in, could be represented by something he called Cognitive Modalities (CM). He and his research team developed a system to define any given word in the English language using about 10 “modal codes,” each represented by three or six characters: an English-to-CM dictionary. In principle, if you then developed similar CM dictionaries for other languages, such as French, a computer should be able to translate between any two languages simply by matching the CM codes from one language’s dictionary with the CM codes in another language’s dictionary…
“Dr. Young set up a small company to develop machine translation based on his theory, and produced a demonstrator that could translate a specific sample of legal text from English into French via the CM codes at an average speed of about one word every 15 milliseconds, with reasonable accuracy (which in practice meant “only” about half the text did not require editing). The programs were written in a language called Zeta Lisp, the preferred language of artificial intelligence researchers at the time, and ran on specialized Lambda computers, made by a company called Lisp Machines, Inc. (LMI), which was formed in 1979 by Richard Greenblatt of MIT’s Artificial Intelligence Laboratory.”
It looks to me like Dr. Young’s GigaText technology was similar to what is described in the article above. It just came 30 years too soon.