Re: How longe before a wearable electronic translator is produced?

Robert,

Like most all "When will X happen?" questions, "When wearable translators will come into existence?" is not so easy to answer. As you point out, someone could lash together existing technologies and, basically, build one today. In fact, except for the "wearable" part, people have actually done so...and the goal of at least some of them is to get to wearable. For example, see http://www .cam.sri.com/html/research_information.html#slt and http://www.is.cs.cmu.edu/js/janus.html As you note, three kinds of software are required to build a speech-to-speech translator:

• A speech recognition program that recognizes speech in the source language and converts it into something that the next stage can use. If we're talking about lashing together existing technologies, the result will probably be text since that's what machine translators generally process. If we're not trying to lash together existing technology, we'd probably want, in addition to the text, to pass prosodic information, such as where phrase boundaries occur, to help the next step.
• A machine translator that can convert what the first stage produces into what the last stage wants. If we're building on existing technology the output  would probably be text in the target language. If we aren't trying to lash together existing technologies we'd probably want to pass on some information about the meaning of the text to help the next step with cadence and intonation.
• A vocalizer that can turn what the second stage produces into speech.

And, of course, hardware sufficient to run it on. The result currently yields something that works well enough to be interesting for research purposes and for actual use in particular domains (http://www.eurekalert.org/releases/cmu-cms071399.html travel planning seems to be a popular one; link defunct as of 8/8/2006) but isn't exactly wearable. With additional work on the software and the passage of time (during which Moore's law will make the hardware cheaper, faster, and less energy consumptive) things will improve. But even so, I'd be really surprised to see a commercially-viable, general-purpose, wearable translator -- one that Joe and Jill Tourist could afford and would actually use -- arrive on the scene for at least ten years. Maybe much longer.

The hard problem is the middle step in the software. I know people who have been working on it for many (more that 20) years. Over the years they've pushed the state of the art, but, even so, the best machine translators come nowhere near the flexibility and accuracy of a human translator. And really big breakthroughs haven't happened that I am aware of. The fundamental problem is that to do a human-like job of translation, the machine needs to "understand" the material being translated. Ad hoc or "shallow" approaches to translation work, but only to a limited degree. (Though, to be fair, just how far one can go with these approaches seems to have surprised most researchers in the field: The quality of commercially-available machine translators based on these methods is much better than many thought it would be.)

For the other technologies, the picture is much better. Speech recognition, particularly when near-field microphones and speaker-specific training are used, is now quite good for the major languages. (Though you can still confuse them with pathological input. Try getting one to deal correctly with an American speaker saying, "How to wreck a nice beach.") Text to speech is, similarly, not bad. What comes out is pretty "mechanical" sounding, mostly because getting the cadence and intonation seem to require at least some understanding of what's being spoken. And computer hardware is definitely getting better-cheaper-faster, as we've all seen.

The bottom line is that English teachers are still "safe"...at least for the foreseeable future, I'd bet.

I hope this helps!

______

For further information on speech-to-text try one of the search engines. I used http://www.altavista.com/ and asked to search for

+"speech-to-text"

This returned over 3000 articles, most of which were on topic.

For further information on text-to-speech, try asking altavista for

+"text-to-speech"

For me this returned about 34,000 pages. Obviously I didn't look at them all but the ones at the front of the list were quite interesting.

For information on speech-to-speech translation try.

+"speech-to-speech" -disabled -relay

This returned hundreds of pages, most all of which were right on topic. (The "-disabled -relay" part excludes many articles on speech relay services for the disabled.)