Posted by: patenttranslator | May 6, 2010

Is Machine Translation a Threat to Human Translators?

But How Do I Do That?

Helpful sign at an airport in Japan

A few times a year I receive a call from a lawyer who wants to know how much would I charge to “clean up” a “pretty good” machine translation of a Japanese patent. My answer is always the same. Unfortunately, I do not edit machine-translated patents because it usually takes longer to do that than to retranslate the patent from scratch. I provide only human translation.

I wrote several articles about machine translation of patents for Translation Journal and other publications almost a decade ago, including “Reflections of  a Human Translator on Machine Translation” and “Useful Translation of Japanese Patents Have Become a Reality“. These articles were published online and discussed on websites in several languages, such as Arabic, Hungarian, Spanish and Chinese, as well as translated into Polish and other languages.

The Japan Patent Office has been providing a free machine translation service on its website since March 2000. So did this machine translation service influence the life of this human translator of patents from Japanese? I think that it did. For instance, that lawyer is calling me on the phone because he knows about the free translation tool on the JPO website, and possibly also because he came across one of my articles on the Web. Sometime I do end up retranslating the patent. So that would be a positive influence of machine translation on my business. If machine translation could work the way people who don’t know much about translation think it should and could work, after 10 years of free software that can translate in the same manner as humans, but for free, I should be out of business. But I am still here. Some years I am very busy and some years are kind of slow, but this has nothing to do with the availability of free or cheap machine translation and everything to do with the ebb and flow of technical translation which basically follows general economic trends.

I think that the free machine translation tool available on the JPO website often makes it possible for potential clients who do not read Japanese to discover Japanese patents of interest to them that would otherwise be hidden. And some of them will be translated by human translators such as yours truly.

The issue of machine translation is also hotly debated on discussion groups for translators as many translators are afraid that machines could replace them in the same manner as machines replaced humans in so many areas: human bank tellers were replaced by ATMs, some cashiers were replaced by check-out scanners, and many mid-level managers were replaced by sophisticated software. It seems that a piece of software can be more sophisticated than a mid-level manager.

There was an article recently in New York Times about pharmacists who provide medicine labels in Spanish to their customers in areas with a large Spanish speaking population (40% of the population in that area speaks Spanish according to this article). According to this article “about three-quarters did so. Among these pharmacies, nearly 90 percent used computers to translate labels from English into Spanish, 11 percent used staff members, and 3 percent used professional interpreters”. The reporters then looked at 76 medicine labels the pharmacists had generated using 13 of the 14 computer programs pharmacists reported using for translation. They found that half of all the labels contained serious mistakes. Thirty-two of the labels included incomplete translations and six contained major spelling or grammatical errors. Examples of translation errors included translating “once a day” into “eleven times a day”, replacing “by mouth” with “by the little”, and translating “two times” as “two kiss”. Just imagine, in order to save money, a pharmacist who does not speak Spanish uses a machine to translate instructions on how to take a medication, which can be potentially deadly if it is taken incorrectly, into Spanish and then gives these instructions to customers who speak only Spanish. If you were a pharmacist in an area with a lot of people who speak only Spanish, would you not hire a bilingual person or pay a human translator for this service? I wonder how they do this in other countries with large minorities of people who do not speak the language of the country, such as Holland or France.

I am sure that machine translation is sometime also used in order to save money by other businesses, for example by owners of small hotels in Europe called “pensions”, to translate the content of their websites into English and other languages. But from what I have seen (I look at these sites because I go to Europe quite often), unlike the pharmacists in New York, these businesses mostly use non-native human translators for translation into English. The translations are often clumsy, but usually not silly and incomprehensible which is typically the result of machine translation.

People who develop and keep improving machine translation have been trying to put people like me out of business for more than 60 years now. There has been some improvement in the software. Google is working on a new “statistical” model of machine translation software which is now offered for 52 languages. Google Translate is also used by the World Intellectual Property Organization (WIPO) and by the European Patent Office (EPO). An insightful article on the issue of machine translation in general and of the approach used for the “Google Translate” machine translation tool, written by David Bellos, director of the Program in Translation and Intercultural Communication at Princeton, is available online here. My niece in Europe is using machine translation software, probably Google, to try to figure out what is on my blog. As the software improves, it is quite possible, even likely, that the best product of the best machine translation software will become almost indistinguishable from the worst product of non-native amateurs who call themselves translators. But that is probably the best that buyers of machine translation products can hope for. The promises made by public relation departments of companies selling the software product as a replacement for costly human translation are mostly corporate propaganda.

The problem is, machines can translate words, but they cannot translate meaning. By definition, a machine or a piece of software will never understand the meaning of anything. And you cannot translate if you don’t understand the meaning. I can imagine a society that uses machines instead of judges, politicians and presidents. You just feed in the information, the machine will check a database of laws and legal precedents and spit out the result, which could be reviewed by a law clerk or a politician’s aid.

But I cannot imagine a machine or a piece of software that understands the meaning of anything.

http://www.youtube.com/watch?v=3I24bSteJpw

Advertisements

Responses

  1. @patenttranslator: human bank tellers were replaced by ATMs, …, and many middle managers were replaced by sophisticated software.
    @patenttranslator: And you cannot translate if you don’t understand the meaning.

    Do ATMs understand what they do? Does the software understand what it does? For translation, why does software need to understand text?

    Like

    • “why does software need to underst

      Like

    • I didn’t have a look at the whole answers and my comment may be redundant.

      “why does software need to understand” did Mike say.
      From my point of view: nothing, BUT the lawyers MUST, just to judge infringements.

      Like

  2. I am not sure what you mean.
    Obviously, software does not understand anything. It only does what we tell it to do. And that is the problem with machine translation. It only replaces words by other words without understanding the meaning because unlike for instance dogs, machines by definition do not understand anything.

    Like

  3. Software does not understand anything, but software has replaced humans for many tasks. You gave some examples.

    What is special about the translation task? Why does software need to understand the meaning to translate a text?

    Like

    • Basically, the short answer is that one word can mean many things. A human translator can chose the right word because he or she understands the context. Software does not understand the meaning of the context. You can try to program software to “understand” context. Machine translation developers have been trying to do that for about 70 years now. Since it does not seem to work, Google Translate is trying a new approach which attempts to circumvent meaning in the translation process (they must have read my articles). It seems to work better, but even this software cannot replace human intelligence which is what you really need. You need to be able to translate the meaning rather than keep replacing words in one language by words in another language. But software will never understand the meaning of anything. You need human brain to understand meaning and machine translation will never get around this problem.

      Like

  4. @patenttranslator: You need to be able to translate the meaning rather than keep replacing words in one language by words in another language.

    Yes, I agree.

    @patenttranslator: But software will never understand the meaning of anything. You need human brain to understand meaning…

    Yes, I agree.

    However, software does many complex tasks that previously were done by people. Software does not understand the meaning of those tasks. I do not understand why language translation is a special case.

    Like

    • Mike, here’s how I think about it:

      Machines are great at commands. You tell it to do X, and it does X (assuming it had a good programmer!). Interactions with ATMs or the self-checkout at a grocery store work perfectly well, because you are dealing with a string of commands—deposit this money, withdraw that money; scan this product, subtract the amount of this coupon, charge my credit card; and so on. The software doesn’t have to understand what is happening, because it is just following a string of commands.

      On the other hand, translation demands understanding. Machine translation works with commands, just like any software—look up this word, change the word order, find that string in another translated text, etc. But human translation means understanding the text and producing a new text in another language that conveys the same meaning. It isn’t just a bunch of tasks that can be programmed, because there is always a moment (or many!) where the translator gets creative. And software isn’t creative.

      (Not to say that machine translation can’t do pretty impressive things. Custom machine translation is pretty spiffy, because it involves a lot of custom data input to achieve better results. However, as functional as it is, it still doesn’t recreate the human translation process or provide the same results.)

      Like

  5. Software is based on the principle of: if the answer is yes, the result is “1”, if the answer is no, the result is “0”. When we talk about complexity of sophisticated software, what we really mean is how many possible options are taken into account. The underlying principle, even of the most sophisticated software, is “1” or “0”. But language is as complex as human thinking, and so is translation. You can take into account the entire range of possible options for ATMs or even for more sophisticated computers on Wall Street. But how do you take into account the entire range of possible options in a human language, which is a reflection of human thinking?

    I don’t think it can be done because the range of options is infinite. And unless you do that (cover the entire range of infinite possibilities), machine translation will still read like … machine translation, a text that is full of really weird mistakes.

    Like

  6. Rachel, patenttranslator,

    Thank you for your replies.

    I think that we have a conflict of belief. I believe that if software is sufficiently good, and assuming that the writer is not deliberately ambiguous, software can translate technical texts such as patents. (Certainly, for creative texts such as poetry and literature, machine translation will never give good translations.)

    Like

  7. Then how would you answer the question in my article, how is it possible that I am still in business? Software that specializes in the translation of patents has been available for many years. Japan Patent Office made this software available for free to anybody. Why do patent lawyers and inventors pay people like me money when they can have patents translated for free by machines on Internet?

    I think that non-translators don’t realize that technical translation is just like any other creative activity. Once we invent machines that can write music that Beethoven would be proud of and mystery novels that keep you turning pages until the morning, we will be able to design software that provides real translation, as good as human translation.

    This will never happen, of course. Machines cannot replace human brain. Machines can only replace words by words. No understanding and no creativity is involved on the part of the machine.

    Like

  8. In my previous comment, I wrote, “I believe that if software is sufficiently good, and assuming that the writer is not deliberately ambiguous, software can translate technical texts such as patents.”

    I did not mean to imply that software is sufficiently good now. A clearer version of my comment is, “If software were sufficiently good… software would be able to translate technical texts.”

    I apologise for my unclear language.

    @patenttranslator: Then how would you answer the question in my article, how is it possible that I am still in business?

    Software is not sufficiently good now to put you out of business.

    Like

  9. If software were sufficiently good, it could write itself and you would be out business too.

    For some reason, software is not sufficiently good now to put you out of business either.

    Like

  10. I came across this and remembered your article from 2000 or 2001.

    I cannot believe how naive translators are with respect to coming MT. “how is it possible that I am still in business? ”

    You won’t be in business by 2015. Serious MT has only been around in Japanese since 2007. Give it 5 more years, and let us know how all is going in your new line of work.

    Like

    • Could you explain what you mean by “Serious MT has only been around in Japanese since 2007?”

      Or are you just throwing it out there without any substantiation?

      I have been following MT issues for 30 years, but maybe I missed a crucial development.

      Like

  11. I forgot about this.

    It has been ten years since you wrote your article for TJ, and there have been major gains in J/E MT. Do you have any samples from 2001? Kanji everywhere and often even the topic wasn’t obvious. But try to put a Japanese news story into free Google Translate and look at the stark improvement despite the errors.

    Now Google has 1.5 million Eurpean previous patents from the EPO. As computer power continues to increase exponentially, there will be better and better translations. Humans make J/E errors all the time as well.

    Won’t Google obtain a huge stack of J/E patents as well? When that happens, expect a major leap in quality and probably a push to modify claims to allow for multiple sentences.

    These are the final years for the human translator. What isn’t clear is when J/E turns into mostly editing and how much globalization will push down rates.

    Like

  12. You are entitled to your opinion, but the facts are what they are.

    I have not seen any real improvement in machine translation in the last ten years.
    Globalization may be pushing rates down for some people, of course, but that is a separate issue from machine translation.

    The results of Google Translate are just as pathetic as MT programs that were used 10 years ago, see my test of Google Translate software here:

    https://patenttranslator.wordpress.com/2010/08/29/a-short-test-of-the-google-translate-function-on-a-pct-patent-application-published-in-japanese-on-the-wipo-website/

    Machines will never write mystery novels or compose operas, and machine translation will never replace human translators, because you cannot translate anything if you don’t understand the meaning of the original text. Machine translation will only bring more work to capable human patent translators because it uncovers patents that were previously hidden.

    Like

  13. […] Is Machine Translation a Threat to Human Translators? May 2010 15 comments 3 […]

    Like

  14. You wrote that you haven’t seen any improvement in J/E MT for ten years, but you didn’t put up any examples from 2000 or 2005.

    And how do you back up your statement that computers can’t translate because they can’t understand when Google Translate is often doing trasnlations that rival human translators? This was posted two years ago:

    “Out of three sample texts that Cris submitted to Google Translate, two would have failed the ATA certification exam and one would have passed (for what it’s worth, I think that 30% is a better passing rate than what is achieved by the humans who take the ATA exam!). So, what’s the outlook for MT? Friend? Foe? Colleague?”

    Google passing an ATA exam in late 2008, even if just once out of three times, would have been laughed at in 2000. And computers keep getting more powerful. Human translation from scratch is toast within the next few years. It will be interesting to see what happens to editors in terms of wages. I suspect lower than today but to what extent is unknown.

    Why not take three samples of a patent and each January, post the result of either Google Translate or whichever MT is used most often in among J/E translators. We can watch it get better and better each year.

    It is wishful thinking for translators to assume they won’t be editors, although this will happen sooner with Spanish, French and German than with Japanese.

    Like

  15. P.S. I just read your sample from 2000, and I’d argue that there has been an improvement. But you didn’t include the Japanese in your article so that one could put it into Google now, ten years later, to compare. Do you still have that to try out?

    It is also important to have more than one example to see improvements.

    Like

  16. 11 years ago it was very difficult to use Japanese on a PC. I was actually using a special word processor on a Mac, which is why I don’t have the Japanese text. But I can type in any language now on a PC, of course.

    Like

  17. Google Translate might have passed an ATA test, but it sure did not pass my test. If a machine can pass the ATA test, my guess would be that there is probably either something wrong with the test, or with the person who evaluated the test.

    The results of my test of Google Translate were pretty awful, see my post on my simple test of the Google Translate software here:

    https://patenttranslator.wordpress.com/2010/08/29/a-short-test-of-the-google-translate-function-on-a-pct-patent-application-published-in-japanese-on-the-wipo-website

    Do you really think that a machine will be ever able to translate a text without understanding the meaning? I don’t. The problem with machine translation is that nobody can design an algorithm for meaning. That’s why Google went for the statistical approach, but as far as I can tell, it’s not working either.

    Machine translation is very different from human translation and always will be. It has a very useful function, though. I use it quite frequently when I am looking for patents in foreign languages. As far as I am concerned, MT is the greatest invention since sliced bread. It brings me more work and it makes my work easier.

    People who think that machine translation will “one day soon” replace human translators are naive and uninformed. But this misconception is understandable. There is a lot of propaganda from vendors of machine translation. Machine translation has been “almost as good as human translation” for the last three decades or so.

    You can buy a sex toy from a porn shop. I never tried it, but I’m sure that vendors of these toys say that it’s almost like the real thing. My guess would be that just like with machine translation vs. human translation, the sex toy is really not very similar to the real thing at all, without, you know, the human input.

    Best regards,

    Steve Vitek

    Like

  18. Hi again Steve,

    1) Your statement that vendors and others have claimed: “Machine translation has been “almost as good as human translation” for the last three decades or so.” is simply false. Please show us *one* example from the 1980s and *one* example from the 1990s where somone has claimed that.

    2) It is always possible to raise the bar to where your quality MT standard is of a J/E patent translator who has been in the business over a decade, but that isn’t what is needed for this to switch into editing, and quite soon. And later this decade, computers will be powerful enough to require little editing.

    3) You also dismiss the person who judged the Potuguese /English translation using Google Translate to have passed the ATA one out of three times as incompetent to judge or the test was bad. Well, how in the world do you know that?

    4) Yes, I think machines will translate BECAUSE THEY ALREADY ARE. Translator’s Journal had a letter from a disgruntled translator because a $5,000 tech job was granted to Google Translate instead of him at the last minute. The Fire Ant dude said the same thing you did, that it can’t possibly be correct so the company likely lied and gave the job to another translator. Gimme a break. The ostrich approach lasts only so long.

    5) The uninformed ones are translators like you, who do not understand how powerful computers are becoming , even if Google Translate is now able to pass the ATA, which would have been impossible in 2000. Yes, translation gets easier with MT (although many still laugh at any usage of it) but over the next few years easier and easier no longer works in the translator’s favor. They become editors, and likely for lower pay.

    Will you still stay in J/E when you are an editor? That is what all translators will need to consider much sooner than they seem to think.

    Jeb

    Like

  19. Hello Jeb:

    Thank you for your response. I am not going to answer your points 1 through 4. Obviously, I am not going to convince you about anything. In any case, I am working on a rush translation of several Japanese patents and I don’t have time for this. The patent lawyer who sent me the work (or rather his client) could have chosen to spend only about a hundred dollars to have all of the documents translated with machine translation, but for some reason, his client does not mind spending several thousand dollars instead for human translation. I suspect the reason here is that they need to know what these patents really say, and they know that MT produces very, very inferior product, which cannot be really called translation. So they have no choice – if they don’t spend the money on human translation now, although it is very expensive, somewhere down the line they might lose a lot of money if they don’t get the information they need right now.

    But you have not addressed my main issue: how is MT going to get around the category of meaning? This is a problem that cannot be solved with machines using algorithms. It can only be solved by human brain. Google realized this some time ago, which is why they are trying the statistical approach. However, based on my little simple test explained in a post on this blog, it does not work either. The problem is that, again, they can’t get around the category of meaning, even with the statistical approach.

    I will address your fifth point. No, I do not expect to become a human editor of machine translated patents in the years to come, for the reasons explained above. And I feel sorry for translators who are planning to become “machine translation editors”. I don’t think that “machine translation editors”, if this occupation exists one day, would be able to make a livable wage. I think that they would be exploited to no end by corporations bent on trying to find cheaper and cheaper labor, the way blue collar workers have been exploited in US in recent years so that eventually, their jobs were “outsourced” to ultra low cost countries such as China.

    It would drive me crazy if I had to earn a living “fixing” the garbage that MT produces! What a horrible way to live, or rather to die.

    Got to go. Work is waiting. It’s a rush job and I am charging a very nice rush rate.

    Take care.

    Steve Vitek

    Like

  20. Steve,

    You realize that to make your case the computers can’t trasnlate, you need to resort to assuming the translator who said that in his opinion that Google Translate passed an ATA test with Portuguese, isn’t good enough to make that judgement. You could always email the blogger who reported that story. You need to assume that the TJ letter was faked or that the translator was lied to about ebing replaced by Google Translate for a $5,000 etch job. Quite a reach….

    Obviously you will still get work translating patents because it is only 2011, and computers are not powerful enough, nor does Google have a huge database of Japanese/English patents as it now has 1.5 million European languange patents. But they’ll get those as well, and then we’ll see. What is so difficult about understanding that the present and future aer not the same?

    When translators turn into editors over the next few years, it won’t be editing horrible MT but very good quality. Some people will choose to stay in, and it looks like you will prefer to do something else. The future wage rate is unkown, but certainly lower than today. Translators need to get their heads out of the sand and stop assuming they will be translating as today in 2013, 2015 or 2017. The year of the editor isn’t clear either, but it is absolutely coming in a few years.

    I have answered your question several times. MT doesn’t need to think to translate well. Japanese is much further from English than Portuguese, so obviously Google Translate can’t pass a J/E ATA test — yet. You haven’t explained how a non thinking device could pass that test. You don’t find that a but odd? It is like isnisting “humans can’t fly because we are not birds” yet despite a human flying above your head in metal shaped like wings, you still insist “humans can’t fly because we are not birds.”

    Anyway, let’s see what happens when Google gets thousands of J/E patents. Computerrs may still not be powerful enough to turn you into an editor that same year, but watch your future examples godd MUCH better when this happens.

    Like

  21. OK, you win.

    In a few years, my job will be gone.

    I mean, how could you possibly be wrong when a machine passed the ATA (American Translators Association) test? It failed miserably my test, but … that must be just some sort of a weird coincidence.

    Better get those 4 Japanese patents that I still have here translated and delivered before some machine steals them from me and turns me into a homeless bum without an income.

    Best regards,

    human translator with three decades of completely useless experience, a relic from the past.

    Like

  22. Google translate can’t _currently_ pass an ATA test in J/E because Portuguese is much closer to English than Japanese. In 2005, you wrote: “But I will go ahead and make the following prediction: Because machine translation will be always hampered by the fact that machines don’t understand the meaning of anything, machine translation will never really make sense.”

    Since a tranaslator judged that Google Translate was at an equivalent level of one who has passed the Portuguese/English ATA test, it is clear that your prediction was wrong just 5 years later. MT does make sense, just not 100% of the time. Many translators make errors as well.

    I don’t know at what point you will no longer have an income through translation. You will still be able to make money as an editor for a while. The problem is that job satisfaction will likely be lower for many and wages will likely sharply decline, but still be a middle class American wage.

    It isn’t as if translators or editors can’t find other work to do. The good news is that while we are in the final years of human translation, J/E translators will have more time to prepare to adjust than those translating most other languages.

    Jeb

    Like

  23. “It isn’t as if translators or editors can’t find other work to do. The good news is that while we are in the final years of human translation, J/E translators will have more time to prepare to adjust than those translating most other languages.”

    Man, what can I say?

    You’re crazy.

    Please let’s not continue this nonsense. (I have a feeling you will want to have the last word, and I will let you have it, no double entendre here).

    Take care, Jeb, whoever you are and whatever it is that you do. Let’s hope that it has nothing to do with translation because you don’t seem to be able to grasp really important concepts involved in this field.

    Steve Vitek

    Like

  24. I don’t think I’m crazy (at least about this), but let’s see what happens with patent translation for European languages now that Google has 1.5 million pairs.

    I consider it an important sign when free MT passes an ATA test in Portuguese. I definately wouldn’t want to be a Portuguese /English translator after hearing that.

    You don’t seem to understand how much more powerful computers will be in coming years.

    MANY translators are worried about MT, and they should be. Burying our heads in the sand isn’t a long term solution. Translators will be able to edit for a few years but within this decade, we all need a new line of work. Yet isn’t it also exciting to go into new directions?

    Since nobody but us is reading this old post anyway, let’s see how things go later in the year with Google’s 1.5 million patent translations.

    Take it easy,
    Jeb

    Like

  25. […] patent translators, will be soon replaced by machine translation software. See for example discussion on my blog here. But it is of course wishful thinking. There is no software substitute for real knowledge, […]

    Like

  26. […] Most days I have quite a few queries about whether MT poses a major threat to human translation, presumably from translators who are wondering how long will they be able to ply their trade. The most popular article on my blog so far is a piece that I called Is Machine Translation a Threat to Human Translators? […]

    Like

  27. I haven’t read the last few comments because they seem to lose focus but regarding your article and argumentation, you have one important flaw in your thinking: you seem to assume that humans have some magical power and some inherent genes for languages. But that’s not the case. Why are humans able to use languages? Why are they able to express meaning? And why are some humans apparently more capable of doing so than others? Because humans LEARN. When we don’t know a word, we have to look it up. We don’t automatically know it. The same is true for grammar. Before we are able to avoid grammar mistakes, we need to LEARN the rules. And machines are more than capable of learning, too. Does “machine learning” ring a bell?
    Machines don’t need to be able to inherently understand words – they need to be able to learn! And yes, they are learning a lot these days.

    Now, having said that: does this mean that human translators will go out of business? No, certainly not. At least not in our times. As has already been pointed out, literature and poetry, art, marketing – all these texts often use language in a creative way. And they also need to consider cultural issues. Many, many years will have to pass until any machine will have acquired enough knowledge to produce a remotely creative text.

    However, many rather simple texts (I can’t speak about patents, but think along the lines of hotel descriptions, user manuals, general instructions, software licenses, etc.) will be translated by machines at some point. This will mean that *some* translators will go out of business or have to find another specialization.

    For some languages, this might happen sooner. For other languages it might take more time. I translate into German and I’ve run some tests with Google myself. Sometimes, the result is scarily good.

    Am I worried? No. I believe a true translator should be able to do more than just translate user manuals and hotel descriptions. Sure, these kinds of texts are easy money and it will be sad to lose some opportunities here. But we can adjust and the good ones will remain in business.

    But arguing along the lines of “hey, I have 30 years of experience and nothing will change because machines cannot think” is a very dangerous attitude. I’m sure you’ll make it safely to retirement and you’ll stay in business. Japanese is a rather different language. But trust me, sooner or later, certain translators will have to face the truth and rethink their business strategy.

    But maybe your strategy is to calm others down and make sure that other translators don’t take MT seriously? So that you’ll have an advantage over them? Sneaky, sneaky! 😀 (just kidding! I found your article and your experience from the Japanese-English perspective quite insightful.)

    Like

  28. But “learning” implies thinking and machines are not thinking at all. The human programmers who are writing the software for a machine are thinking, but the machine (the computer) is incapable of thinking.

    And human language is so complicated, even when it comes to simple instructions for hotel guests, for example, that you simply cannot write software sophisticated enough to provide real translation that would be as good as human translation. You need to be able to think to translate. There will be further minor improvement in MT technology, but the barrier of meaning will never be broken by machine translation. MT will only work if it is post-processed by human translators, and preferably also pre-processed before it is processed by a machine.

    Occasionally, a sentence will translated correctly by a machine, more or less by accident.

    But most of the time it will not be translated correctly because translating correctly means understanding the meaning of the words in any context and under any circumstances.

    This is something that cannot be programmed into a piece of software because contrary to what you are saying, unlike people, dogs, and even frogs probably, machines are not capable of thinking.

    Can you agree with that?

    Like

  29. […] A few times a year I receive a call from a lawyer who wants to know how much would I charge to “clean up” a “pretty good” machine translation of a Japanese patent. My answer…  […]

    Like

  30. Imagine a document is translated from Japanese into English and the client says, this is OK but it needs to sound more formal and businesslike, or, this is too formal can you make it more informal and localize it for Canada, or, this is an brochure for a Medieval themed roller coaster can you throw in some ye olde style English.

    How would a computer deal with that?

    And with Japanese in particular, the subject is often inferred after it’s mentioned once. Sometimes the subject and the object are inferred in a sentence. It can be difficult for a human to infer the correct subject and object sometimes, with a full understanding of the meaning, the wider context and the information in the previous sentences?

    Do machine translations take context into account? Do they take into account the subject and object mentioned in previous sentences when translating following sentences? If not it’s difficult to see how they can produce accurate translations, no matter how much ‘power’ they have.

    Not to mention prose written with style and panache, idioms, slang, sarcasm etc etc. Using a sophisticated statistical model to accurately guess the correct word in all of these cases would be extremely difficult, especially where the word isn’t even there, it’s just inferred.

    As globalization increases the demand for translations with increase exponentially. MT will be useful for tourists, people talking on skype etc. But if you’re translating your annual report, a medical document, an advertising brochure or anything where a bad translation could have serious financial or legal repercussions you’d want to spend money on a good human translator.

    The amount of J/E translation work has increased dramatically over the last few years. If anything demand is set to increase.

    Computer ‘power’ is not an issue. We already have super computers that can predict weather patterns and develop artificial ecosystems. Computers are plenty powerful enough already.

    Like

  31. […] a brainless, heartless, bloodless, and pulse-less machine even passed a few years ago a Portuguese to English translation test administered by the American Translators Association to people who want to be accredited by the ATA. However, this interesting fact may be saying more […]

    Like

  32. […] While the statement that humans will never be able to develop a sense of smell that would rival the sense of smell of their four-legged companions would not be surprising to most people, thanks to the hype surrounding machine translation, the statement that machines will never be able to translate anything, (if by translation we mean interpretation of the real meaning of what is being said), is surprising to most people, and people will argue ad nauseam that human translators will one day soon be replaced by machines. […]

    Like


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Categories

%d bloggers like this: