PDA

View Full Version : Language Translating Machines



aostling
01-28-2008, 06:35 PM
I've taken up the hobby of digiscoping, taking photographs by mounting a P&S digital camera behind the eyepiece of a spotting scope. I'm in the early learning phase for hardware and technique, and for knowledge of the local Arizona birds.

Here's a reply (#2) I made to a recent post on the digiscoping forum http://www.birdforum.net/showthread.php?t=106255. The Italian, Michele, used translating software which I think failed to get even the gist of his question across. However, the Finn seems to have made a better guess than I did.

I have little experience with language translating software. Has the state-of-the art progressed no further than this?

Evan
01-28-2008, 06:54 PM
The question is, "do we have intelligent machines yet?" The answer is no, not even close. Language doesn't have rigorous rules, especially English.

Machine translation will never be very good unless we can develop intelligent machines. There isn't a single shred of evidence that that is possible.

pntrbl
01-28-2008, 07:03 PM
The question is, "do we have intelligent machines yet?" The answer is no, not even close. Language doesn't have rigorous rules, especially English.

Machine translation will never be very good unless we can develop intelligent machines. There isn't a single shred of evidence that that is possible.

I dunno Evan. I think some of your machines are smarter than me already! :D

SP

rotate
01-28-2008, 08:38 PM
Machine translation will never be very good unless we can develop intelligent machines. There isn't a single shred of evidence that that is possible.

Do you mean that it's not possible in principle, or that all the progress that's been made in AI doesn't mount up to much?

I think there's some real renewed optimism in AI research that's coming from neuroscience and microbiology. Trying to model a CNS of a slug or a snail is a good start, instead of some stupid s/w that tries to emulate a psychologist. If you believe that mind is the what the brain does, then it's only a matter of time before we can reverse engineer it.

DancingBear
01-28-2008, 09:00 PM
The free online translators are lousy. But I doubt if they're state of the art. Does anyone have experience with the ones that actually cost money? I'd definitely expect better from them, and I suspect I'd get it or they'd be out of business by now.

But I also expect it would be a poor substitute for fluency in both languages.

Walt

Evan
01-28-2008, 09:23 PM
There may be renewed optimism but I haven't heard of any developments to support it. It more like a loss of collective memory of all the attempts and utter failures over the last half century.

Machine intelligence is an all or nothing thing. It's very easy to discern if you are conversing with a machine at this time and always has been. The Turing Test has only succeeded when the subject matter is very tightly restricted both in nature and scope.

To convince even the skeptics an intelligent machine will have to produce original commentary on it's own that is relevant to it's experience. It will need to exhibit insight, predictive ability and most especially, the ability to draw valid conclusions from information that appears to be contradictory and insufficient.

A classic example from the task of machine translation is the phrase "Time flies like an arrow". How is this to be interpreted? It all depends on the context it is given in and without reference to such context there is no certainty possible as to the correct translation of the phrase.

Does it mean that a particular species of flies called "time flies" are fond of arrows? Does it mean that time proceeds in the same metaphorical manner as arrows do, quickly and in a more or less straight line? Does it mean to measure the speed of flies in the same way one measures the speed of arrows?

We have no machine that can decipher that phrase today any better than one could in the 1950's. It is only possible by using an explicit rule based translation and then it is a tossup what will result. What all attempts to produce machine intelligence have failed to produce is the single most important part of intelligence, understanding.

rotate
01-28-2008, 09:58 PM
Computers may one day figure out what "Time flies like an arrow" mean, but it will be a great deal longer before it can laugh at "but fruit flies like banana". :D

barts
01-29-2008, 01:45 AM
Reminds me of the classic: Machine translation from English to German and back of the following phrase:

"Out of sight, out of mind"

yielded:

"Invisible Idiot"

Years ago, I worked in the robotics field, building custom robotic cells for welding, inspection, etc. We had a large lab, full of all sorts of robots, computers, machine vision systems, etc. When AI became a buzzword, management hired a bunch of cognitive psych. folks, and those folks placed a sign at their (small) end of the lab that read "Artificial Intelligence Laboratory".
Since we'd actually delivered working production robotic systems rather than one canned demo after another, we took umbrage and put up a sign at our end of the lab: "Real Intelligence Laboratory". That lasted about two days; both signs were then removed :-).

Paul Alciatore
01-29-2008, 11:10 AM
You expect good translations from machines? You can't even get fair ones from people who make their living that way.

I once had an equipment maintenance manual from a major Japanese manufacturer with the title "Color Handy Lookie System". It was for a hand held, color TV camera. If you think the title was bad, you should have tried following the procedures inside it. Groups of us (professional TV engineers) would sit around discussing each step in the procedures trying to discern what was intended. It took a lot of time and experimentation to figure out even a fair percentage of it.

This book was translated by people who were professionals in the field and who had a good understanding of the original Japanese and of the subject matter. They probably also spoke and wrote passable English. What was lacking was a good understanding of the English language usage in the same profession. What was really needed was two professionals, one of each native language and both with experience in the particular field.

Good luck putting all that in a machine.

Consider, there are English speaking (or at lease writing) persons of many countries on this board. Even between the English English and the US English there are often different ways of saying things. Others like the Australians, the Scotch, etc. are quite colorful. I love it, but it can make for problems. And that is with the "same" language.

I love computers, but personally feel that we will need expert, HUMAN translators for many decaces to come. Perhaps even for centuries. Perhaps forever. AI? Remember HAL.

Evan
01-29-2008, 12:12 PM
Do you realize that HAL is IBM shifted one letter over?

andy_b
01-29-2008, 01:41 PM
A classic example from the task of machine translation is the phrase "Time flies like an arrow". How is this to be interpreted? It all depends on the context it is given in and without reference to such context there is no certainty possible as to the correct translation of the phrase.


i don't think your phrase is a valid example. how would YOU interpret "Time flies like an arrow"? i am not a computer and i have no idea what you mean. you would judge the meaning by what else was said in the conversation. in fact, i would counter that the phrase has NO meaning and could be ignored. please use the phrase in an example of normal conversation and prove me wrong.

computers have already mastered chess. while there are set rules as to how pieces can move in chess, the game also consists of intuition and guessing what your opponent's moves will be. language is the same. there are set rules (albeit many of them), and part of carrying on a conversation is making adjustments on the fly in reaction to what the person you are conversing with says. i think there will come a day in the future (and probably in my children's lifetime) where a human will be able to carry on a relatively normal conversation with a machine.

andy b.

Evan
01-29-2008, 02:08 PM
i don't think your phrase is a valid example. how would YOU interpret "Time flies like an arrow"? i am not a computer and i have no idea what you mean. you would judge the meaning by what else was said in the conversation. in fact, i would counter that the phrase has NO meaning and could be ignored. please use the phrase in an example of normal conversation and prove me wrong.
You have just proven my point, thanks. It's a line that not everyone is familiar with. Without that familiarity you don't know how to interpret it so how could a computer be expected to do so? It was first said by Groucho Marx in a film but it was a linguist who siezed on the implications on seeing his movie.

You say you would judge the meaning by what was said in a conversation and that is of course partly correct. However, most of what we say and write also has a broader underlying cultural context that we automatically presume the listener/reader knows. If I glance at my watch and say "it's time to make tracks" you will have no doubt that I mean it's time I left, not build a railroad even though the conditions are such that I won't be leaving tracks. This is so even if I made no reference to having a reason to leave. Even without the body language you would almost certainly interpret the statement correctly.

We already know from previous learning and conversational experience that "making tracks" means to depart even though no tracks are made. Our language is filled with such phrases where the meaning can only be correctly discerned via a previously accquired knowledge of the liguistic ideomatics and traditions. This is why true intelligence is required to properly translate from one language to another. Even then it also requires a cultural background in the target language as well as the original if the translation is to be entirely satisfactory. It's why we see so much "chinglish" in manuals for Chinese products. They don't have ready access to translators that are well grounded in North American cultural norms and language.

andy_b
01-29-2008, 11:00 PM
hmmm....
i am going to have to watch the Groucho Marx film he says that in.
at some point i will probably post back to this thread, as i have other AI ideas.

andy b.

J Tiers
01-29-2008, 11:17 PM
BTW............

The reason (or ONE reason) that English is so inconsistent, is that IT IS NOT ENGLISH......

NOPE..

What we CALL "English" is an aboriginal language of some obscure islanders (not in "England" by the way) , grossly modified by additions from EVERY other language on earth.

Do we translate words and then use the "english" version? Sometimes.......

But, Gestalt, dashiki, karma, smorgasbord, restaurant, pasta, tofu, ballet, rendezvous and thousands of other words are all directly from other languages, and retain their meanings in general.

Grammatical constructions and pronunciations are in many cases derived from the use of so many other languages , each with its own rules...... Small wonder that a machine has trouble with the rules and meanings.

THEN you get into the "folk sayings", which are generally based on meanings which are now obscure.......

Just today, a guy at work, when he was told to do something a way that was a bit of a "kludge", protesting this, replied "that isn't very 'downtown' ".

It took me a second to realize what he meant, and that I had heard it before.

tell your machine to "translate that"......................

HTRN
01-30-2008, 01:07 AM
J Tiers, I was under the impression that English originated in Northern Germany sometime with the three Germanic Tribes - Angles, Saxons and Jutes(Yep, The Angles and Saxons later went on to become the AngloSaxens, the basis for a hefty part of English Culture until the arrival of the Norman French in 1066 under William the Conqueror).


HTRN

Evan
01-30-2008, 03:10 AM
You are both correct and this is yet another example of the problems. The Islanders Jerry mentions are the Frieselanders from an area that today is the region of northern Germany and southern Denmark that is Holland.

You wouldn't recognize it at all unless you speak either Danish, low German or Dutch. I speak Danish and German and between the two languages they contain about 2/3rds of the same basic words as English in either identical or nearly the same form. The rest is largely French followed by Spanish and then everything else.

barts
01-30-2008, 03:33 AM
You are both correct and this is yet another example of the problems. The Islanders Jerry mentions are the Frieselanders from an area that today is the region of northern Germany and southern Denmark that is Holland.

You wouldn't recognize it at all unless you speak either Danish, low German or Dutch. I speak Danish and German and between the two languages they contain about 2/3rds of the same basic words as English in either identical or nearly the same form. The rest is largely French followed by Spanish and then everything else.

My mother speaks Frisian; a friend of mine w/ a Phd in Medieval English Literature (he built armor for a living...) remarked how similar the language was to Old English as he could puzzle out several of her Frisian sayings.
I speak passable Dutch and some German, and Frisian is rather different yet.

- Bart

juergenwt
02-28-2008, 01:30 AM
Having translated technical German into English and the other way around for many years,I have only one answer - It's a "No Go" for computer translation. For general use conversation it works some times, but be very careful. You can make a fool of yourself. If you are a tourist people will forgive you and you will make friends.
For technical translations you must know the technical terms in a particular field in both languages. Don't try dictionaries or computer translations unless you just need a little help in technical terminology. It could be very costly!!! Don't ask someones Grandmother from another country to do a technical translation for you or a carpenter to translate a CNC- machine manual.
Remember the old Japanese instruction manuals?
There is a book out on some of the biggest mistakes made in translations by some of our biggest corporations ( it is funny), and the news media is full of translation errors. You wonder at times why these people think the can ask someone with a few hours ( or month, or years) of lessons in a foreign language to do a correct translation. Usually people can laugh about it but some times it is outright embarrassing. In the technical field it is unforgiving.
The US Government spent zillions of dollars to develop a super computer to translate instantly and accurately from other languages into English. So far - nothing.
So don't call your Grandmother Helga to translate a machine repair manual for you. Call on someone who has a working knowledge in that particular field in both languages and has the skill to give you an accurate translation.
Remember Pres. Kennedy's famous words:" Ich bin ein Berliner" - I am a jelly filled pancake. (correct: "Ich bin Berliner") but people knew what he wanted to say and loved him for it.
So long - Juergenwt.

Evan
02-28-2008, 06:30 AM
The problem you mention was addressed quite well by Xerox in the 70s. The PARC research center developed a special limited vocabulary and a restricted syntax for Xerox technical manuals so that they could be machine translated with minimal cleanup by a human. They arranged the query structure so that it followed a binary tree. All troubleshooting steps followed a logical structure that consisted of a question about the result expected but the question was always worded as a statement, "Five volts is present on pin three?".

This system sometimes gave rather odd syntax in the target language but it was always readable without misinterpretation as statements are much easier to translate than questions.

tony ennis
02-28-2008, 08:11 AM
The "Time flies" is a silly statement. No one would use it in polite conversation.

:D

To show how sticky this is, I don't even know how to prove a computer can "understand." What does that mean? Does the savant that knows the day of week any date falls upon understand time? Does Deep Thought understand chess?

I think that requiring a machine to have the full capabilities of a person before we consider it to have 'understanding' is asking a lot - and showing a bit of hubris.

My dogs understand things. But they aren't going to be composing minuets any time soon. If I depend upon them for things they understand (like herding sheep and protecting my property) then I do quite well.

My point is that a machine that is more limited than a human could still 'understand' and still be useful. Even Google's translators do a reasonable job much of the time. It's true that the more specific the domain of the text then the worse the translation. No surprise there - people also fail in that situation as we have all seen. If you want to translate a machinery manual, then you better have the context and vocabulary of a machinist.

Regarding "Time flies", I am sure a competent translating engine would rate its own translation quality lowly. The statement is ambiguous (...and ridiculous, which is why it is amusing). Why penalize the translator for someone's poor writing! Besides, who would really expect that statement to translate?

-=-=-=-
All that being said, when machines truly understand and can produce ideas, IMO we're doomed. We'll lose our ability to think hard.

Evan
02-28-2008, 08:47 AM
But, the "time flies like an arrow" statement was not concocted just to confuse a computer. It was said by a person whose words were very likely translated to other languages. It's an appropriate example of the difficulty present in the task.

Machine translation can never be really successful without the ability to understand what is meant within the context of the writing. Languages have both formal rules and informal unwritten rules of usage. The latter cannot be codified because they depend on the circumstances in which something is said. It is possible to make up long lists of exceptional phrases using a direct mapping from one language to another but it cannot possibly cover anything close to all the possibilities.

Here is what Google has to say about it:



The translation quality isn't as good as I'd like it to be. Can you make it more accurate? We're constantly working on it. Even today's most sophisticated software, however, doesn't approach the fluency of a native speaker or possess the skill of a professional translator. Automatic translation is very difficult, as the meaning of words depends on the context in which they're used. While we are working on the problem, it may be some time before anyone can offer a quick and seamless translation experience. In the interim, we hope you find the service we provide useful for most purposes.

Pete H
02-28-2008, 10:31 AM
All that being said, when machines truly understand and can produce ideas, IMO we're doomed. We'll lose our ability to think hard.

Spot-on! (Another untranslatable phrase.)

- We've seen this happening already, with the proliferation of cheap hand-held calculators. When I was a kid, we learned the "Times Tables" from yellow cardboard sheets with "1x1 = 1" all the way to "12x12". Dull, yes. Repetitious, yes. We also had "mental arithmetic drills" where we had to add, subtract, multiply, divide various larger numbers. That was when the "Tables" drills paid off.

I'm nowhere near as facile as I was 55 years ago, but I still can do rough calculations in my head, to the amazement of some of my former (and much younger) colleagues, who carried little calculators from about 3rd grade upwards. Most of these kids don't know how to use a slide-rule, either.

I'm not decrying the use of calculators - I use them all the time, possibly more now that I'm "retired" and working on my own projects. But what happens when the battery goes dead in the middle of a job?

Pete in NJ

Evan
02-28-2008, 10:41 AM
But what happens when the battery goes dead in the middle of a job?

Same as what happens if the power to the machines goes out. It really isn't a very good argument anymore since we depend on so much technology everyday. It's more important to recognize when the technology is giving you wrong answers than it is to be able to calculate the right answer. Can you do arcsin in your head? I can't but I will know when the answer is silly because the calculator was left on radians instead of degrees.