Art Meets Science: The Early Years of MT Research

Top 2. Feb 6, 2001 11:35 AM

» NALocke - MT to replace humans...?

Je ne pense pas.

Even in 2001, MT can not replace humans. A quick trip to online translation sites will set any translator's mind at ease. And, I think you're on to something with the "natural food" analogy. Not all text can be synthesized.

On the other hand, MT has probably had an important impact on how translators do their work. I'd love to know what the pros and cons of computer-assisted translation are from the perspective of a translator (a human one, that is!).

Merci de votre intérêt!

-- posted by NALocke

Top 3. Feb 6, 2001 4:41 PM

» dtp_pro - MT...

Well, if we start talking about tools that help translators, that's a different thing.

To continue with an analogy, we, long time ago, discover that tractor that harvest the food for us does not influence the taste.

What I want to clarify is that MT gives a "taste" to a translation (not very tasty) that a human (the translator or the chef) will be more clever about it.

Salut!

-- posted by dtp_pro

Top 4. Feb 10, 2001 3:02 PM

» jyden23 - Re: Will the machine replace humans?

In response to message posted by dtp_pro:

Isn't it nice that all bilingual or trilingual can see it is close to impossible to create a MT which will work. I got a small free-bee of a so called pocket translator covering 10 Europeann languages, of which I can read 5 directly. I wrote a word in English, "Valve", and went through the list of languages, German, French Italian etc to end up in Swedish. In all but 2 the translation were correct, but in Dutch and Swedish, it came out with one of the slang words for penis in those languages. Not very clever, and an embarrassment for the people who had given the thing as a free-bee during the promotion of an encyclopedia it could have been a MT coding the chip inside.

-- posted by jyden23

Top 5. Feb 10, 2001 5:47 PM

» NALocke - Re: Re: Will the machine replace humans?

In response to message posted by jyden23:

Hello, Jyden,

Your experience seems to confirm the old adage "you get what you pay for." Actually, your frustration echoes the sentiment that doomed early MT research. People expected FAQT, fully-automated quality translated, that is, without human intervention.

Let's face it, at the time, visions of the future were closer to an episode of the t.v. series "The Jetsons" than realisable reality. (For the young and non-Americans, the futuristic, cartoon Jetsons lived in a world where people zipped around in mini-spaceships and absolutely everything was fully-automated.) In fact, chances are you may have "experienced" machine translation in some form or another and, because of the quality, you didn't even know it. Significant human intervention — by translators, engineers, editors, to name a few — ensures quality (or "tasty" translation to use dtp_pro's analogy).

It's a shame that a freebie, handed out as reliable a "instant" translation tool, should taint your perception of machine translation. Online automatic translation sites also tend to discourage visitors. Folks type in a few lines from "Jaberwocky" and giggle at the incoherent translation that results. To be fair though, other tools are far more refined. They may not fit in your pocket, they may cost a little more than nothing, and they may require human intervention to produce quality translation.

Will a machine ever produce a FAQT of "Jabberwocky"? Until human scholars can agree on its meaning, the prospect is highly unlikely.

Thanks for your message.

-- posted by NALocke

Top 6. Feb 13, 2001 10:51 AM

» cuzittt - The question remains:

The question remains: What is the upper limit that is trying to be reached by MT technology companies. If the answer is the "Star Trek Universal Translator" Quality, much more work needs to be done in the field of AI and probably linguistics. It is certainly unfeasable to expect MT to reach that level using dictionaries and parsing libraries.

Languages are difficult. Languages are evolving. Languages use words in multiple contexts, and translation of different context will be different words in different languages. This is where MT fails now, and I don't see a real solution in the near future.

Is MT used in business today? Of course. But, is Nancy's statement "In fact, chances are you may have "experienced" machine translation in some form or another and, because of the quality, you didn't even know it. Significant human intervention — by translators, engineers, editors, to name a few — ensures quality (or "tasty" translation to use dtp_pro's analogy).
" true. I'm not sure. I've no doubt I have read bad translations which were MT based. Have I read good translation that were originally MT based, and then modified? Probably, but the amount of human intervention was almost certain significant... and at that point, is it proper to call it MT?

MT, in it's purest form, cannot be used in any real public context, at least not without a large disclaimer. And, my experience with MT, makes me wary of how much it should be used in conjunction with the human translation industry.

However, the use of Translation Memory, is extremely helpful to translators. The reason it is helpful is the very fact that the glossaries used are specific to the product and/or company, and therefore terms are typically placed in the proper context (Similar to the example of "front" given in the original article). But, MT does not do that (At least not in it's core methodologies... it almost certainly can be made to do so).

On another tact completely, the prevalence of the free translation services (Systran's Babblefish and FreeTranslation.com) are almost certainly injuring the cause of MT and the underlying technologies in these engines (Systran and Transcend specifically). Because, the translations given are often poor or non-existent. The phrases do not have to be jaberwocky. These engines handle parsed/conjugated verbs very poorly (or not at all). They typically give one translation for a given word, even if it is plainly obvious that it should have multiple words attached (record which has two specifically different meanings, table- kitchen table or database table).

These specific context problems is what MT needs to work on now. It will then need to work on incorporating slang terminology and idioms. It will then need to work on parsing rules specific to word placement, after all, one can split an infinitive, or avoid the split, and the meaning is still the same.

This is difficult enough work to do for one language. Add 14 more languages, and you are now only covering half the worlds population... and still may not properly cover dialectical or regional differences.

MT is a work in progress. I have grave doubts that I will ever see MT replace humans in my lifetime, even for the major languages. I certainly don't expect MT to be able to work with the 18 official languages of India, let alone the 180+ other languages in use in India.

Brandon Magee
Localization Analyst
eventra

P.S. And what of the 800 languages of Papua New Guinea?

-- posted by cuzittt

Top 7. Feb 13, 2001 12:40 PM

» NALocke - Re: The question remains:

In response to message posted by cuzittt:

Brandon,

You ask, "What is the upper limit that is trying to be reached by MT technology companies?" If the history of science and technology (or just history) teaches us anything, I think the answer may still be the Trek-ian ideal. And, yes, if that is the objective, much more work needs to be done.

As for MT providing the desired results, you're right again. In its purest sense, MT is limited. When I used the term MT, in my response to Jyden, I was using it in a broader sense, as the original field of technological inquiry that has evolved, is still evolving, and has produced, among other things, TM tools.

As for Babelfish and FreeTranslate.com, they are sorry ambassadors for the technology they boast and the best p.r. for human translators. I had a funny (or not) experience the other day with an on-line translator announced in a press release. When I fed a paragraph from the press release into the maws of the machine and asked for a French translation it came out, well, perfect. Cynicism led me to submit another text, a straightforward paragraph from an unrelated news article. The results of the second test were, to say the least, less impressive. Coincidence? I think not.

In addition to all the languages, written and not (6,000+ is a figure that comes to mind), that MT or CAT or translation tools will never translate (for reasons that go well beyond technological limits), there are whole genres of writing that are simply not amenable to the technology. And even when translated by humans, there will always be readers (a show of hands, please) who will seek out works in their original because, frankly, always something precious is lost in a translation.

So, short answer to the original question posed by dtp_pro: Will machines replace humans? No.

Thanks for your input, Brandon.

-- posted by NALocke