I know that it was predictable that a product like this would eventually appear; but I'm stunned to be seeing it introduced so early in my life. The world is funny like that, you're waiting for one thing and a dozen other inventions are made in the mean-time that you weren't expecting.

IBM laptops with ‘multilingual, automatic speech-to-speech translators’

Now, with regards to peace ... there HAS to be a benefit with being able to communicate. I expect there to be hiccups; but we can expect this type of technology to become more streamlined with time.

I thought it would look more fish-y for some reason...

I'll have my Babel fish fried with a side order of the prime directive, well done.

I spent about 6 years involved in the military linguistics field (2000-2006), and I'll tell you one conclusion I've come to in that time.

Machine translators will probably never reach a level at which they replace a human speaker. I say this from experience, having worked with some MIT guys on a speech recognition program and having tested several "top of the line" machine translation tools.

Now, am I saying that there aren't any that work well? No. They're great for doing word by word translations, and sometimes they even get the gist of something fairly close. But when you're talking about politics, peace, governments, those are the areas where the way you roll your R can be the difference between a treaty and a war. You do not want a machine doing that.

Of course technology will advance. It may even get "there" at some point in the future. I just don't think that anything we're capable of at this moment or in the immediate future is going to do it.

Agreed.

This is an amazing tool for sending along with a team when going in to foreign territory.

This is not what you hand your Secretary of State, and say "Hey, no need for a translator any more, woo!"

I'll echo that. It doesn't replace a good translator, but in situations where you would otherwise not have one, this is better than a phrase-book or nothing.

What we seem to be discussing is the difference between translators and interpreters.

This is not a universal translator.

A universal translator is capable of translating languages, that are not known beforehand!

Big difference!

Bye
Thanee

Uhh, not that it matters much, but can you support that claim? Outside of Star Trek conventions, the term could be interpreted a lot of ways. It's just kind of a bold statement to assert without any explanation.

(I have a universal joint on my transaxle, but I doubt that it is capable of joining devices that have not been invented yet.)

Stupid thing to argue about, I know.

Anyways, this device looks just like the application of technology I've seen before: I've seen language translators of various quality on the internet, and voice recognition/production has been around for years, also often of questionable
quality. This device really doesn't seem all that remarkable to me.

Further to what I said before, a translator in one situation might report. "I invite you to perform gentle oral massage of the root of me. You have removed me on the path." But an interpreter would say, "He's mad because you cut him off back there."

Translators always do very poorly when it comes to idioms, and English especially is laced with idioms. Interpreters account for idioms, body language, sarcasm, context and so on. I'm not saying that technology couldn't one day create an interpreter, but I'm certain that it has not yet.

Kage: Yeah, I had originally said only "never", changed it to "probably never", and then just realized that there wasn't a way to get my point across without leaving it open to someone nitpicking the semantics, so I just posted.

Fact is, if it's not in our lifetime, we're pretty well okay when using "never" for stuff like this, because the word works, and conveys the meaning. I could say "won't replace us for a million years", or "won't replace us for 10^N", but what would be the point, really? I wouldn't be any more objectively and perfectly correct, and my meaning would be much less clear and to the point.

And yeah, while you're on the track of what I think he meant, I can't know what Dog meant without his telling me/us.

And I love that discussions about this topic are always their own evidence. It's interesting.

Well, I personally think we're likely to see effective machine translators (not necessarily better than a well-trained human, but an acceptable replacement) within the next forty years. I wouldn't bet more than a few thousand on it, though, and only that much because anything less probably wouldn't be worth tracking for forty years (plus I'd have the interest on the money for all that time).

~J

Manufacturing a bunch of electronic translators is probably a lot cheaper than training a bunch of people, so in logistical terms, they're probably very useful.

eidolon: I was friends with an ASL interpreter who would get pissed if you called her a translator. She would say that a translator shoots for literal accuracy, where as an interpreter listens for the gist of things, then explains the point being made. My dictionary describes translating as "putting into other words" and interpreting as "explaining the meaning of." Seems to me that a translator's aim is to convey the message, but an interpreter's goal is to convey the idea. Translators will tell the joke, but an interpreter will explain it.
When you watch a DVD, turn on the subtitles, watch a scene and compare the words in Spanish or whatever to what the actors are saying in English. You'll spot some points where the words used are significantly different, but still convey the emotion or thought accurately. This is an example of the subtitle guys interpreting instead of just translating.
If a 1st 2nd or 3rd ed. shadowrunner says "Frag this drek," a translator would tell a 4th ed shadowrunner it means "Fuck this shit." An interpreter might say it means "I've had enough of this."
Did I help, in my usual, rather long winded way?

Yup. It turns out I guessed correctly after all, but it's always better to check (even if I sometimes don't ).

What you've said is exactly why machine translators have so much trouble. Take Chinese idioms for example. Four characters, but with the implied meaning of an entire parable/story/etc...well, "idiom" behind them. Drove us non-native speakers nuts! You can literally spend years studying them. It's a perfect example of the difficulties in mechanical translation.

Where a machine sees "man fish cut mountain", a fluent speaker would know that it means "with the essence of a thousand salmon, the river knifes through the valley", which actually means "to do something quickly and with vigor, against all odds". (By no means is that supposed to be a real Chinese idiom, to make sense, or to be anything other than an example.)

Now, yes, you could load up a database with all of the idioms, and write an algorithm that could probably match them up with input pretty reliably, but the program still doesn't understand them. At best, you're making an automated dictionary. It's still up to one person to enter it correctly and hit "go", and still relies on the listener/reader's ability to understand the output. Take that idea and spread it across the whole of a language, and you really start to see the issues of machine translation.

Other issues stem from the fact that language is ever evolving and liquid. Take the more recent English usage of nouns as verbs. "Mousing" is good one. To "Google" something. Taking a noun and using it as a verb form is really common in Chinese, and becoming ever more common in English. But how fast do dictionaries keep up with that? How fast, more importantly, do language courses pick up on that? I've seen ESL courses where they're still teaching formal polite early 20th century English, and Spanish courses in California being taught as if everyone spoke Castilian Spanish.

So are machines getting better? Yeah. Are they useful for certain applications and in certain situations? Yeah (several have been mentioned). I still don't see them replacing a human any time soon.

Kage, what are your reasons for choosing 40 as the number of years? What are some limiting factors now in machine translation that you see being overcome in that time frame, and how do you think we'll do it? (Not specifically, I know it's not your PhD area or anything, I mean stuff like "I think programming will advance significantly", etc.)

I do think your example is off somewhat.


Still find it odd how you could just pick a number. What do you speak outside of English (if anything)? (Not a snark or anything, just curious and can't recall if you've ever said.) I'm wondering if you're basing your viewpoint on a language that's mechanically easier than some others?

And surely you can agree that machine "learning" and machine "understanding" are two vastly different things. (Although this goes back to what I said in the spoiler tags about the necessity for understanding in communication.)

Latin, (ancient) Greek, and Japanese. Latin fairly proficiently, though with considerable cobwebs; Japanese at an extremely basic level. With regard to ancient Greek, my knowledge is both relatively basic and very rusty.

No, I don't agree that they're vastly different things. That isn't to say I think they're the same thing, I think we first need to define "understanding" before we can discuss this meaningfully at all. Care to propose one for the discussion?

(As for my number, I'm not basing that on any knowledge of language, I'm basing it on my personal guess as to when computers will duplicate human functionality, brain-wise. That would naturally include translation, as humans are able to translate.)

~J

The difference between learning and understanding, you mean? In that case, how do we test to determine which occurs?

~J

Originality is the key. A computer that truely understands must be able to create new, unique, and comprehensible work from the information in its databases and it must be able to do so reliably. It doesn't have to be good and it doesn't have to be big or long; but it does have to be the computer's own work.

How to test this does present a conundrum. Some tests are so easy that modern chatbots could defeat them and other tests are so difficult that few humans could pass them.

I had very good conversations with a german fellow, only using babelfish to translate our emails. What's amusing is how quickly we adapted to the translation software. We quickly adopted a much more simple language, and were able to communicate our ideas okay.

So, while a machine needs to learn our language, we also learn how to use the machine. Our intuitions are very powerful things, and we can adapt quickly to make good use of a tools; despite its weaknesses.

('univeral translator' was merely a cultural allusion)