What's the book corporate America doesn't want you to read? Find out now -- or you could get scammed.

Speech recognition revisited

July 19, 1999

Regular readers of this column know that I’m no fan of speech recognition technology. The last time I installed one of the programs, I suffered a catastrophic systems failure from which my PC is still recovering.

It would probably be no stretch to say that you’d have to torture or bribe me into trying another speech-recognition program, especially on my desktop computer. Dragon Systems shipped me three of its top-selling software applications after my last story in an effort to coax me back into the fold, but the boxes are still collecting dust on my shelf.

Now I’m thinking of giving in. No, not because two suspicious-looking characters from Dragon’s security department showed up at my front door carrying long wires and every conceivable kind of adapter. No one has visited me – yet.

Instead, the news that Dragon’s speech recognition engine had been coupled to a new translation program made me take notice and reconsider my ban on the software. Simple speech recognition is a wonderful concept for travelers who lack the time or skills to type the text for themselves. But imagine taking the recognition to the next level by interpreting the speech and outputting it as another language.

Cool, huh?

In a nutshell, that’s what Transparent Language’s promises to do. Its Desktop Translator is a $99 application that will translate Spanish, French, German, English, Italian and Portuguese. The program, which will be available about a week from now, will handle a variety of documents, including plain text, e-mail messages, Web pages, and wordprocessor documents.

I’d like to interrupt this column to address an issue that must be bugging the language purists among you by now. My interchangeable use of the words interpretation and translation is probably irritating the college English majors out there who know that there¹s a distinction. What do you call a machine rendering synthesized text into another language? I think an argument can be made for using either word.

I don’t know how well Desktop Translator works yet, although Transparent’s engine is available for testing on its Web site to find out what it would do with your speech, once recognized. My biggest worry – and I think it’s one shared by other frequent travelers – is how the program will run on a portable computer.

Can’t forget that the limited amount of memory on my computer made another speech recognition program hiccup so badly that it threw my system out of whack. Could adding another layer of programming just compound the difficulties? I asked product manager Jay Marciano.

“No problem,” he said. “It can run on a laptop computer. The memory requirements are minimal.” (To be specific, it calls for a Windows 95, 98, or NT; 200MHz Pentium with 80MB of RAM and 210MB of disk space as well as a 16-bit sound card or built-in audio system and a high-quality noise-canceling microphone.)

In fact, Desktop Translator is aimed at small-business and home users who don’t have the big bucks for a minicomputer or a workstation. Marciano’s company offers two higher-end products that require more computing power but are probably of little use to mobile travelers.

I’m intrigued by the possibilities of Translator. I could use the program to dictate a speech from English into German or to communicate with a source who only spoke Spanish. With the right setup, I could even deliver a presentation in Portuguese without the benefit of a human interpreter.

I may be getting a little too excited, according to Marciano and others. “As with any speech recognition software, it has to be trained by the speaker. So you couldn’t have someone in the audience stand up and ask you a question in French and have that translated into English,” he says.

Another limitation is the context. A computer doesn’t know if the trunk you’re talking about is the back of the car, the front of an elephant or something you pack clothes into. Even asking Translator to handle a simple phrase like “fruit flies like bananas,” would give the machine a conniption fit because it doesn’t know if you’re saying that fruit flies or if you’re referring to an insect.

“We’re still literally 50, 60, maybe even 100 years away from having a product that can tell the difference,” says Kevin Hendzel, a computational language specialist at Aset International Services, a translation and interpreting company in Arlington, Va. “You can get the gist of what someone is saying in another language, but it’s more or less unintelligible English.”

How high-tech will a true interpreter be? Think Star Trek. Hardcore trekkers will note that the translators used by the crew of the USS Enterprise read brain waves and interpreted them as speech. We’re still light-years away from being able to do that. The more I read about the emerging discipline of machine translation, the more I long for a simpler solution.

Given all the hassles of speech recognition and translation (or interpretation), why not just learn the language? A language immersion class will put the knowledge directly into your head rather than having to haul some hardware around that only gives you a rough translation. It’s just a thought. How do you deal with language barriers when you travel.

Christopher Elliott is the author of Scammed: How to Save Your Money and Find Better Service in a World of Schemes, Swindles, and Shady Deals. Critics have called it “eye-opening” and “inspiring” — it’ll “grab your attention and won’t let go.” Order your copy now on Amazon, Barnes & Noble or iTunes.

Be the first to comment

Previous post:

Next post: