GOOGLE TRANSLATE.

May 3, 2010 by 64 Comments

I’ve been reluctantly impressed with the results Google Translate (Wikipedia) gives me (reluctantly because my default assumption has long been that automatic translation is No Damn Good), and I was interested to read a couple of pieces about it online. Here is Mark Phillips at NPR’s All Tech Considered, and here is a longer piece by Philip Bethge at Der Spiegel (translated from the German by Christopher Sultan, an actual human). Thanks, Sven!

Comments

  1. bulbul says
    May 3, 2010 at 3:17 pm

    Reluctantly impressed describes my feelings as well. My thrice accursed job now involves comparing various MT solutions and though I haven’t delved into it too deeply yet, so far, Google Translate always comes out on top.

  2. Martin says
    May 3, 2010 at 4:13 pm

    “Christopher Sultan, an actual human”
    As opposed to … an intelligent entity that’s not an actual human? As we approach the singularity, or whatever, will “human” join the list of retronyms — words that need an adjective for clarity, like analog watch, acoustic guitar, hardcover books, standard transmission? (Examples picked from a Safire column on the topic: http://www.nytimes.com/2007/11/18/magazine/18wwln-safire-t.html)
    (And of course “human translation” as distinct from “machine translation” is in the same category.)

  3. AJP Coming up next: Skijumps says
    May 3, 2010 at 5:11 pm

    And then there’s “inhuman”, which usually refers to something done by humans.

  4. AJP Iceskating? After this. says
    May 3, 2010 at 5:13 pm

    (As does “subhuman”.)

  5. Nijma says
    May 3, 2010 at 5:52 pm

    Google support for romanized Arabic:
    If you set the languages to translate from Arabic to English, you can type phonetically on an English keyboard, hit the space bar, and it will render what you wrote into some sort of Arabic script. I’ve used it as a starting point before going to a regular dictionary, but you have to be careful. If I type in “kelp [SPACE]” I get back قلب the word for heart (an endearment) and not كلب dog (an affront), “gelb ” brings back جلب, and “galb ” yields غالب. But if you are googling something in Arabic you probably already know about kaaf, qaaf, and jeem.

  6. dearieme says
    May 3, 2010 at 6:08 pm

    Awa’ tae Freuchie whaur the Froggies bide.

  7. boo says
    May 3, 2010 at 9:00 pm

    Statistical Machine Translation is obviously a few orders of magnitude better than the retarded “dictionary and grammar” approaches used by older machine translation companies. This has been clear for several years already.

  8. Giuseppe M Brescia says
    May 3, 2010 at 9:02 pm

    Yes, it’s true, I also had to reluctantly praise Google Translate, when a while back I ran a simple test on several automatic translators and Google really stood out (you’ll find a link to my post below).
    That said, though, any slightly more convoluted sentence from the average newspaper is still likely to be teeming with mistakes. I’ll follow up the old post soon to make that point… After all, as Phillips says in the article you linked,
    “I’m skeptical that “gestations” will be enough. Much of what we read on the web is written beautifully or full of nuance and software will never be able to translate that.”
    Anyway, check out how Google outshines the competition:
    http://smuggledwords.wordpress.com/2010/04/15/machine-translation-the-baton-of-italian-fan-and-the-ready-era/

  9. David L says
    May 3, 2010 at 9:31 pm

    But the reason Google Translate works so well is that it can draw upon a large body of existing translations performed by, as you put it, actual human beings. As GT grows, so that increasing quantities of the material it uses as a resource were in fact created by GT itself — well then, we’re back in begging-the-question territory, are we not?

  10. Bathrobe says
    May 3, 2010 at 10:39 pm

    Yes, Google Translate leaves the others in the dust, but as a person who uses it every day to translate English into Chinese, I can tell you, oftentimes the results resemble a バラバラ殺人事件 — a murder where the body has been dismembered and the body parts hidden in different places. Coherent sentences are converted into total nonsense, where subjects mysteriously turn up at the end of the sentence and numbers are even occasionally split into two parts (like, 5,800 split into 5 in one part of the sentence and 800 somewhere else). True, Google Translate does a fairly good job of choosing appropriate vocabulary, which saves time spent in looking up the right word, and in some cases the sentences produced are almost perfect, but I can assure you, you pay for your laziness with a very laborious and frustrating bout of post-editing.

  11. Alexiphy says
    May 3, 2010 at 11:53 pm

    As translation services on the computers evolve, people will get much more accurate translation via computer systems. I don’t think it will ever be perfect, but it will certainly get close to it. Especially since they are doing more -phrase translation- versus word-for-word.

  12. boo says
    May 4, 2010 at 1:21 am

    I’m not qualified as a linguist, but from my naive point of view it seems that the difference in results between the old approach of programming grammars into a machine and plugging in vocabulary, versus Google’s statistical approach, can say a lot about Chomsky’s Universal Grammar.
    If the Universal Grammar hypothesis were valid, I would suppose that the grammar/vocabulary approach would yield the better results. Any ideas?

  13. Mike Unwalla says
    May 4, 2010 at 4:31 am

    @Languagehat: I’ve been reluctantly impressed with the results Google Translate (Wikipedia) gives me (reluctantly because my default assumption has long been that automatic translation is No Damn Good)
    Some months ago, I paid professional translators to evaluate translations from Google Translate. For English to Spanish, and for English to Norwegian, the translations were satisfactory. For English to Welsh, the translations were not good. For details, see http://www.international-english.co.uk/mt-evaluation.html.
    @Giuseppe M Brescia: … any slightly more convoluted sentence from the average newspaper is still likely to be teeming with mistakes.
    I agree. The source text for the evaluations was optimised for machine translation. For example, the sentence structure is simple and words are used with their literal meanings.

  14. dale says
    May 4, 2010 at 2:07 pm

    As an ex-programmer, I was a bit startled too by the implication that a computer program was a non-human intelligence. Computer programs *are* human intelligence. (All too human, in fact.)

  15. language hat says
    May 4, 2010 at 3:52 pm

    As an ex-programmer, I was a bit startled too by the implication that a computer program was a non-human intelligence. Computer programs *are* human intelligence.
    I fear you are suffering from what the French, in an untranslatable little joke, call déformation professionelle. To non-programmers, the idea that “computer programs are human intelligence” seems perverse in the extreme. They are a product of human intelligence, just like socket wrenches, but they no more are human intelligence than are wrenches.

  16. marie-lucie says
    May 4, 2010 at 4:07 pm

    LH: déformation professionelle
    The non-native always sees things that the native does not. I had never thought of associating déformation professionelle (professional skewing?) with formation professionelle (job training). I am sure the joke was unintentional.
    Déformation professionelle could apply to physical adaptation of the body to habitual gestures or positions (eg the bigger and stronger hitting arm of a tennis player) or more often to common mental habits and kneejerk reactions as a result of one’s profession (the attitude shown in English by: if you have a hammer, everything looks like a nail).

  17. John Emerson says
    May 4, 2010 at 5:07 pm

    Kenneth Burke referred to “occupational psychosis” and “trained incapacity”, crediting these concepts to Dewey and veblen.
    I agree with Dale, pretty much. Computers are devised by humans and are not inhuman, any more than any tool is. They certainly transform human life, often in unexpected ways, but they’re not inhuman.
    Not quite the same point, but when Deep Blue defeated Kasparov or whoever it was, there was a whole team of three-dimensional meatspace chess master coaching the machine.

  18. komfo,amonan says
    May 4, 2010 at 5:17 pm

    I come for the linguistic nuggets & stay for the spontaneous outbursts of Scots verse.
    To non-programmers, the idea that “computer programs are human intelligence” seems perverse in the extreme. They are a product of human intelligence, just like socket wrenches, but they no more are human intelligence than are wrenches.
    This doesn’t sound right to me. Programs are an extension of human intelligence, are they not? They can solve problems in minutes that we could solve, but only in years, decades, centuries. And I imagine, due to the amount of data they can process, they can solve problems we can’t, or point to solutions we wouldn’t otherwise see.

  19. Zack says
    May 4, 2010 at 7:11 pm

    Chomsky’s “universal” grammar is quite thoroughly discredited, altho his acolytes do a superb job of obscuring that fact. I have read that for every single claim it makes, there exists at least one known human language that provides a counterexample.

  20. language hat says
    May 4, 2010 at 7:59 pm

    Computers are devised by humans and are not inhuman, any more than any tool is.
    Programs are an extension of human intelligence, are they not?
    I’m not sure why you guys are being so stubborn about this, and I’m not sure what you think you’re arguing against. I’m not saying, nor do I think, that computer programs have nothing to do with humanity, having been dropped on earth by the same aliens who built the pyramids. My original statement was that the Spiegel piece was translated by “an actual human” as distinguished from a computer program, and I continue to see nothing whatever controversial in that statement. Unless you are claiming that because computer programs are created by humans, their translations are ipso facto indistinguishable from those produced by actual human beings, I’m not sure what your point is. You remind me of the Supreme Court insisting that corporations should be able to vote, marry, and be president of the U.S. (I may be exaggerating a bit here.)

  21. marie-lucie says
    May 4, 2010 at 9:21 pm

    Chomsky’s “universal” grammar is quite thoroughly discredited, … for every single claim it makes, there exists at least one known human language that provides a counterexample.
    In addition, as more and more examples to not fit with the theory, it has to be tweaked and complexified so that it gets farther and farther away from providing a believable model of human language usage and processing.
    I was teaching French syntax (not my specialty) the last few months, with a Chomskyan oriented textbook (there are few others, and this was the best one I could find). I did not use many examples from the textbook (usually dreamed up by the authors), instead I used my own examples, and also actual texts from the press or from literature. In almost every case, the real life examples were more complex to analyze than the textbook examples. Also, several important features of French syntax were barely mentioned, if at all, in the textbook, because the theory arose from English syntax, and features which are not part of English syntax tend to be overlooked or just glossed over in adaptations of the theory to other languages.

  22. komfo,amonan says
    May 4, 2010 at 9:49 pm

    I was responding only to your response to Dale, El Hat. And as far as socket wrenches go, they’re a product of human intelligence with goal of devising a different type of hand, while (some) programs have the goal of devising a different type of brain, an enhanced, specialized subset of human intelligence. But I’m not claiming that machine translation is the same as human translation, & agree with you about corporations (and it looks like that ship has sailed).

  23. Ran says
    May 4, 2010 at 11:06 pm

    Speaking as a professional programmer, I’m with you, Hat. There are respects in which a program is part of its programmer (e.g., ethically: “he wrote a program to steal your credit card numbers, and it did so for him” entails “he stole your credit card numbers”), but human intelligence and machine intelligence are very different things, without even much overlap.

  24. Ran says
    May 4, 2010 at 11:10 pm

    @komfo,amonan: So would you say that a socket wrench is “an enhanced, specialized subset of human” torque-applying ability?

  25. Grumbly Stu says
    May 5, 2010 at 12:31 am

    m-l: I had never thought of associating déformation professionelle (professional skewing?) with formation professionelle (job training). I am sure the joke was unintentional.
    marie-lucie, even in my German eyrie ey have encountered déformation professionelle more than once, for instance in sociology books and internet articles in French. I was pleased as Polichinelle to have understood it. The joke is quite deliberate, and may have been around for some time. I think one of those books was from the seventies, either by Atlan or Morin.
    dale: Computer programs *are* human intelligence. (All too human, in fact.)
    This is an example of a category mistake. The very fact that you feel it necessary to stress “*are*” to make your point demonstrates the opposite of what you claim. Computer programs are made by humans, just like socket wrenches, but they are not human. Hat’s socket-wrench argument is short, sweet and incontrovertible.
    komfo,amonan: Programs are an extension of human intelligence, are they not? They can solve problems in minutes that we could solve, but only in years, decades, centuries. And I imagine, due to the amount of data they can process, they can solve problems we can’t, or point to solutions we wouldn’t otherwise see.
    Programs don’t solve problems, and are not an extension of any kind of intelligence. They merely execute. If programs are programmed to write programs, then they do that too, but that’s no big deal. The people who use programs use them in such a way that what the programs do is interpretable as “solving problems”. Without such interpretation and specific, controlled use in specific environments, there is nothing to say one way or the other about what computer output consists in.
    Output is usually “incorrect” during program development. Most programs are tinkered with in their productive phase because the requirements of users change, or the output turns out to be not quite what is now wanted. Computers and programs require as much maintenance as cars. People, in contrast, pretty much take care of themselves.
    As a programmer and software architect, I continually have to struggle against simple-minded programmers who think anthropomorphically, and so are unable to design good software. They think in terms of “I”: “When I call the server, the remoting service serializes the answer back to me”, which leads to procedural, process-operating-on-data software. What makes software so useful is precisely its non-human capabilities, especially the fact that they have no mind of their own. Human is just not good enough.
    Computers, engines and other artefacts of human technology are the least interesting kind of machines that we know of, see Edgar Morin’s books.

  26. Grumbly Stu says
    May 5, 2010 at 12:35 am

    correction towards the end of my last comment: “… especially the fact that it has no mind of its own”

  27. bruessel says
    May 5, 2010 at 4:20 am

    All I can say about Google Translate is that whenever I go to the language section of Yahoo Answers, it is full of people asking native speakers to help them make sense of the machine translation or if the translation into a foreign language is correct, and most of the time it isn’t, especially when German and French are involved.

  28. Eimear Ní Mhéalóid says
    May 5, 2010 at 5:36 am

    It does a very bad job of translating into Irish. The vocabulary is mosly ok but the sentence structure and verb format is completely wrong. What’s irritating is that a lot of commercial websites are generating versions of themselves in this pidgin-Irish and swamping the results of Google searches for actual text in Irish.

  29. Grumbly Stu says
    May 5, 2010 at 6:37 am

    swamping the results of Google searches for actual text in Irish.
    It is getting ever harder to find anything useful, in any language. Everybody and his dog publishes rubbish in pidgin-X that happens to contain my search words, whether X is German, French or English.
    I expect that the internet will have become effectively unusable for searching purposes within 2-3 years. Already the net primarily catches old boots.

  30. language hat says
    May 5, 2010 at 7:27 am

    The joke is quite deliberate
    Yes, m-l, I’m quite sure déformation professionelle was originally a joke, but it’s such a useful phrase it crops up without any intended reference to the original one.

  31. Zackary Sholem Berger says
    May 5, 2010 at 9:01 am

    Not bad for Hebrew – I mean, the nuance is wrong, and so is the word order, and the word meaning, but the gist is there somewhere amid the garble.
    But the Yiddish is execrable. I don’t know why that is (maybe the quality is worse because there is less of a corpus available).

  32. marie-lucie says
    May 5, 2010 at 10:03 am

    Grumbly and LH, I was not implying that déformation professionelle, which I often heard in my parents’ speech (long before the 70’s), was a new phrase. Actually I think that I have encountered it more often than formation professionnelle. It is only that it never occurred to me that the two were deliberately related.

  33. MMcM says
    May 5, 2010 at 10:59 am

    From GB, it looks like it was used in a literal sense, as a medical condition due to occupation. (The apparent earlier uses are bad metadata: 1843 is 1943 and 1593 is the wrong book.)

  34. marie-lucie says
    May 5, 2010 at 11:24 am

    MMcM, thank you for searching. It is quite evident from the context that the literal physical meaning (which I mentioned in my first comment above) has nothing to do with “formation professionnelle”, let alone being a jocular derivation from it. From a literal to a metaphorical meaning is but a short step, which does not imply going through “formation”. If some people saw a joke, it must be posterior to the metaphorical extension of the meaning of “déformation”.

  35. Nijma says
    May 5, 2010 at 12:07 pm

    As someone who worked in the hardware side of computer manufacturing R&D and field service for some years, I have to second (third?) Hat and Grumbly that computers are not intelligent and can not solve problems. I would compare them to a light switch. A light switch is either on or off, corresponding to a one or a zero. The computer is just a huge array of on and off switches with an instruction set that defines which combinations of on and off on the input results in which combinations of on and off on the output. More switches tell the processor whether to treat a string of ones and zeros as instructions or as data. They’re very sophisticated, high speed semi-conductor switches, but they’re switches just the same. They don’t ask and answer questions. All they do is crunch numbers. It’s an enhanced abacus, not an enhanced brain.
    But as komfo,amonan intuits, due to the amount of data they can process, they can allow humans to solve statistical problems that were unsolvable before, because even in the eighties, the problem of manipulating huge databases had not yet been solved. Computers just weren’t big enough or fast enough. But a human still has to define the problem, set it up, and find a place where meta data has already been collected. Then, as now, it’s “garbage in, garbage out.”
    It’s only natural to anthropomorphize machines though. People have probably been talking to their cars for at least as long as cars have had radios.

  36. language hat says
    May 5, 2010 at 1:01 pm

    From GB, it looks like it was used in a literal sense, as a medical condition due to occupation.
    Well, I’ll be darned. You learn something every day!

  37. marie-lucie says
    May 5, 2010 at 1:11 pm

    LH, What did I tell you!

  38. Nijma says
    May 5, 2010 at 1:26 pm

    Wikipedia says “déformation professionnelle” is a pun on “formation professionnelle,” although there is some discussion about lack of citations.

  39. marie-lucie says
    May 5, 2010 at 1:55 pm

    Nijma, Just because you read it on Wikipedia does not mean that it is correct.

  40. Dima K. says
    May 5, 2010 at 3:49 pm

    Thanks for sharing this article. I have been noticing the quality of Google Translate improving over the last year or so; this seems to prove my observations.
    Also, the article had an absolute gem in it: “the German word “Schloss” means both castle and lock”. As you may or may not know, the Russian word ‘замок’ also means both ‘castle’ and ‘lock’. I wonder if there is an actual explanation for this or if it’s just an amazing coincidence. (see my LiveJournal post on this: http://induke.livejournal.com/66190.html).

  41. Nijma says
    May 5, 2010 at 5:00 pm

    Just because you read it on Wikipedia does not mean that it is correct.
    But of course I understand it is not correct. But not knowing French myself, I had used Google (and not Google Translate) in order to follow the discussion here, which is how I knew there was an associated wiki article that had bogged down at exactly that point. If I felt qualified and knew what to do, I would correct it myself, especially since there is now a citation on this thread.

  42. language hat says
    May 5, 2010 at 5:07 pm

    Also, the article had an absolute gem in it: “the German word “Schloss” means both castle and lock”. As you may or may not know, the Russian word ‘замок’ also means both ‘castle’ and ‘lock’. I wonder if there is an actual explanation for this or if it’s just an amazing coincidence.
    That struck me too, and I went straight to Vasmer, who says замок /zámok/ ‘castle’ is from Polish zamek, which is a calque on German Schloss, whereas замок /zamók/ ‘lock’ is a native derivation from замкнуть. So, mostly coincidence.

  43. marie-lucie says
    May 5, 2010 at 6:19 pm

    Nijma, I understand what you say. I would go and correct it myself, but I don’t seem to be able to follow the instructions properly so that my contribution stays, so I have given up on trying to make corrections (if I succeeded, I would probably spend far too much time trying to correct things on Wiki).

  44. Dima K. says
    May 5, 2010 at 6:48 pm

    LH, thanks for doing the research, although you’ve got your accents reversed – /zámok/ is for castle, an /zamók/ is for lock.
    I wonder if other such coincidences exist in any other languages.

  45. language hat says
    May 5, 2010 at 7:22 pm

    you’ve got your accents reversed
    D’oh! That’s what comes of trying to do brain surgery, I mean transliterate Russian, after a day of babysitting. I’ll fix my comment so it doesn’t mislead anyone.
    m-l: It should be very easy to make changes; just go to the Edit page, change what you want, hit “Show preview” to make sure it reads the way you want it to, and then hit “Save page.” Presto, it’s changed.

  46. Lee Dembart says
    May 5, 2010 at 10:02 pm

    I got the article on Google Translate from Der Spiegel in the original German, and I ran it through Google Translate. It was very fast and much better than other machine translation I’ve tried over the years. But don’t fire human translators just yet.
    Here’s how Google Translate put the article into English:
    Googling in tongues
    By Philip Bethge
    A German researcher has developed one of the first translation software for everyday. Brute force calculation helps the Google software to astonishingly good results.
    It is a good sign when the creator of a software that finally even used. In Japan, Franz Och was recently on tour. In the restaurant, he pulled out his cell phone and decrypted the menu. Even local news he read without difficulty. For the translations he received in seconds.
    For six years the 38-year-old in the Google headquarters in Mountain View, California developed the Translation Program “Google Translate”, “and so far I’ve actually never really used it myself,” admits Och. But now the change of heart: “I am very happy with what we have achieved.”
    The German is the secret star of a software industry that has no less a goal than to tear down the global language. Och Google wrestles with case records, conjunctiva and auxiliary verbs. The result is an affront to all linguists. For it is not language know-how, but brute computing power allowed the automated Dolmetscherei.
    52 languages mastered the system already. The databases are like 296 more under construction. Among them there are exotics like Sardinian, West Frisian, or Zulu.
    “Machine translation has reached a new level”
    Complete websites, theses, even love letters translated “Google Translate” in no time – and it often provides surprisingly useful results. For Google the benefit is obvious: Even more Internet surfers than previously can be with such a useful and free them also apply to the company’s Web site.
    “Machine translation is arrived at a new level,” says Oh, “people share our program is now a solid and the software has reached the real world.”
    “What Google does here, is very impressive,” said Alon Lavie from Carnegie Mellon University in Pittsburgh. The computer sees the industry moving. The market for translation software is growing rapidly: “There are very exciting times.”
    The era of machine translation has begun. Programs such as Google Translate exhibit “the way to a future where everyone can talk at the touch of tongues. The ultimate goal of Sprachtüftler is an electronic version of that Babel fish, which was invented by the British author Douglas Adams in his science-fiction classic “The Hitchhiker’s Guide to the Galaxy”: used her ear, simultaneously translated the blutegelartige creature each language. Even the crude Vogon poetry of the Romantic hero can understand Arthur Dent.
    “Words often have multiple meanings, and the number of combinations is almost infinite”
    So far there have been developers in the real world yet though. But already there are iPhone apps like “Jibbigo” quick as the spoken English translated into Spanish. Creator of the software is Alex Waibel, computer scientist at the University of Karlsruhe and at Carnegie Mellon University. Waibel can already translate many of his lectures of computers simultaneously. Even in parliamentary debates, he tested the technique already.
    In the interpreter from the computer lab has long been a bold dream. How do example, the machine know that the English are zerdeppert “breaking records” no records? In translating the phrase “we meet at the Castle” again, the software has the word “lock” with “castle” and not to “lock” (lock), which would be quite stupid.
    For a long time tried the computer, einzubimsen programs such world knowledge about a complex set of rules. But even if the software ran perfectly formulated texts often in happy delirium. A “nightmare” about Swamy Viswanathan calls from the U.S. company Language Weaver tried, such as the English language with all its nuances to squeeze into rules. “Words often have multiple meanings, and the number of combinations is almost endless,” laments Viswanathan.
    The experts at Language Weaver pursued so early on a different concept. They fed their systems with many texts from the Internet that already existed in multiple languages. The approach of the experts: almost every sentence, every phrase has long since been translated several times. Pure statistics is therefore sufficient to decode a language.
    2nd Part: Och has perfected this statistical procedure for Google now
    At about the example of “we’ll meet in the castle to interpret”, the program was searching its database for texts of converging “and” castle “near froze. Then it thinned out the translations of these texts and there is often the word “castle”. Consequently, the machine, “we meet in the castle” and not “we meet in the lock” off.
    Oh this statistical procedure for Google has been perfected. During his PhD specializing in the Franke on speech recognition. He then went to the University of Southern California. Soon the Pentagon was interested in his work. After 9 / 11 did the U.S. intelligence control exacerbated Arab newspapers, chat rooms and Web sites.
    But in 2004 attracted the Sprachbezwinger Google in Mountain View. Och since there is benefit the awesome computing power of the Internet empire. Figures may not call Och. For many language pairs, however, store trillion messages in Google’s databases. Important resources for the Archive include the word translated into numerous languages Bible, transcripts of the United Nations or the 23-language EU documents.
    These kind of “parallel texts” are something like the Rosetta stone of the digital age: The ancient model bears the same inscription in Greek, demotic and hieroglyphics. 1822 lifted the Egyptologist Jean-Francois Champollion with the help of the secret of hieroglyphics.
    So it now makes Ochs software. The strengths of the system: The same code works for all languages. There must be only enough translated text.
    Sawaf also criticized that Ochs system only works online.
    A letter eater as a universal translator? Many linguists consider such slide rule tricks for Tinnef. “The statistical translation quickly reaches its limits,” the linguist says about Martin Kay of Stanford University, “the approach ignores the complex structure of language.” When used in German position of verb and auxiliary verb fail about the technology. Even the distinction between subject and object, they have to scrimp.
    “For really good results, we must delve deeper into the language,” says Hassan Sawaf, head of U.S. software developers forge Apptek. The company is a middle ground. Besides using statistical algorithms Sawaf and classical grammar rules: “This improves the sentence structure and readability significantly.”
    Sawaf also criticized that Ochs system only works online, “Who is working offline, you can ‘Google Translate forget’.” Waibel is also skeptical. “Imagine, you are abroad want to talk to a seller and first have to search a network and then also pay high roaming fees – it is not practicable.”
    Indeed, the power dependence as a major weakness of the Google translation engine, however, its Californians do not deviate from their course. Already they develop a special version with integrated speech recognition for the company’s own mobile operating system “Android.” And soon it should be possible to have texts translated instantaneously Photos. This is how the future travelers unfamiliar photographed in China a sign bearing the words – and knew immediately that he is on his way to Beijing.
    “He who speaks no English, can only use a fraction of the Internet”
    Another money printing machine for the Internet giant seems to mature. But Och dismissive gesture. How many Google employees he thinks he is better on a campaign for freedom and equality in the net. “Anyone who speaks no English, can only use a fraction of the Internet,” he says. It is true that diversity make available to all.
    There is no indication of the noble intentions of the programmer it anyway. Och and his team have developed a special software, feeding with the help of interpreters on their own translations into the system can – and even for very exotic idioms such as the Bantu language Xhosa, the language of living in Japan Ainu or the Inuit language Inuktitut. In this way, the software developers want to give also to those tongues hearing community already forgotten. The New Zealand Te Taka Keegan computer engineering from the University of Waikato has already tested the program for the Maori language. Keegan recently spent six months at Google, to explore whether the digital language of miracle from Mountain View, the idiom of the New Zealand natives could keep them from disappearing. His experiences are very positive.
    “The number and quality of Maori translations are growing with the help of this tool,” says Keegan. A digital archive arises, the grant of the language major boost.
    “Our children grow into a digital world,” said Keegan. “Only if we manage to Maori to make it part of the world, the language will survive.”

  47. Grumbly Stu says
    May 5, 2010 at 11:27 pm

    This Google Translate version of the Spiegel article is not a whit better than what is produced by someone with a German-English dictionary and a minimal knowledge of German. The English is weird, and German words have been left in it. This is typical of someone who thinks that she is translating something of significance, not all of which she could understand. So she falls back on imitating syntax and reproducing lexical items from the source language, in the hope that these things somehow carry significance per se and will help readers to understand what is meant. There are thousands of “translation services” that produce exactly such results.
    On the evidence of this Englishing of the Spiegel article, I am fairly astonished by the hype about Google Translate, and by people who find “improvements” over other translations whether done by software or by hand. Such comparisons remind me of the urban legend about a question supposedly put to Tolstoy: “what is the difference between state violence and revolutionary violence ?”. He answered: “the difference between cat shit and dog shit”.

  48. Bathrobe says
    May 5, 2010 at 11:42 pm

    The main difference between a Google version and one by someone with a German-English dictionary and a minimal knowledge of German is that Google takes a matter of seconds. You don’t have to pay someone a lot of money, or force someone to spend hours of their time in order to get an idea of what the German is saying. It’s done for free and in an instant. For that it’s quite useful. But if you need to understand what’s actually being said rather than just being presented with a mishmash of (often false) impressions and engaging in a lot of guesswork, you need a real translator.

  49. Grumbly Stu says
    May 5, 2010 at 11:43 pm

    Possibly the excitement over Google Translate is just another by-product of belief in progress. This is similar to belief in God, and seems to have replaced it in Western countries – as many people over the last 300+ years have remarked.

  50. Grumbly Stu says
    May 5, 2010 at 11:50 pm

    if you need to understand what’s actually being said
    That is usually why I read a Spiegel article, if I read one at all (once every three years, say). It’s also the reason I read anything.
    Of course such portable translators are useful for tourists inspecting menus (“fried enema”) and understanding the signs on monuments. For years now you can buy such translators for 20 euros. Google Translate requires too much computing power, and produces results no better than the portable devices.

  51. Grumbly Stu says
    May 6, 2010 at 12:01 am

    I wonder whether the garbled stuff that comes out of machine translators only reinforces the idea many people have that furriners have a strange way of talking – like the Captain in the Katzenjammer Kids.

  52. Bathrobe says
    May 6, 2010 at 12:42 am

    Grum, for people like us who are In The Know, Google Translate will merely give rise to frustration. But for those who know nothing of foreign languages, it is quite possible that it will reinforce stereotypes and cause people to stop trying to listen.
    The only people who can be expected to make a serious attempt to understand are those with vital interests at stake — for instance, businessmen, whose minds are wonderfully concentrated by the prospect of making a buck.
    Grum, für Leute wie uns, die in die wissen, wird Google Translate lediglich Anlass zu Frustration. Aber für diejenigen, die nichts von Fremdsprachen zu beherrschen, ist es durchaus möglich, dass es Stereotypen verstärken und dazu führen Menschen aufhören zu versuchen, um zuzuhören.
    Die einzigen Leute, die voraussichtlich einen ernsthaften Versuch zu verstehen, machen können, werden diejenigen mit vitalen Interessen auf dem Spiel steht – zum Beispiel, Geschäftsleute, deren Geist wunderbar durch die Aussicht auf einen Buck konzentriert.

  53. Bathrobe says
    May 6, 2010 at 12:52 am

    Translation Party has now told me what I really wanted to say but was too clumsy to express it properly:
    Moody’s, knows people like us, Google’s translation can lead to frustration. The order of these languages, my next step, please thank the director is known to increase. Stereotyped sequence of stops. Enough people interested in stock – for example, can be expected to understand the important issues to focus attention on the business behind it.

  54. John Cowan says
    May 6, 2010 at 3:25 am

    “Cash registers don’t really add and subtract; they just grind their gears. But then, they don’t really grind their gears either; they just obey the laws of physics.”
    By the same token, computers only have derived intentionality: they do what we mean them to do. But is human intentionality really anything different? We attribute intention to human beings, as Dennett says, because it’s a simpler approach for prediction and planning than trying to understand human beings as objects simply obeying the laws of physics.

  55. Grumbly Stu says
    May 6, 2010 at 5:13 am

    Dennett might also have said: we attribute intention to objects when it seems simpler than speaking as if they were things simply obeying [obeying !] the laws of physics. For instance “computers solve problems”, “enzymes control DNA replication”.
    The trouble is, we tend to get stuck in such discursive conventions without noticing it – whether these are anthropomorphic, “objective” or otherwise.

  56. language hat says
    May 6, 2010 at 8:15 am

    This Google Translate version of the Spiegel article is not a whit better than what is produced by someone with a German-English dictionary and a minimal knowledge of German.
    But that’s a pointless comparison. It’s a great deal better than any other free, virtually instantaneous translation tool, which is what the “excitement” is all about. And I’m not sure what this animus against the idea of progress implies: are you eager to go back to the caves? Don’t let me stop you from grumbling, but I’m not sure why this particular advance has set you off.

  57. Grumbly Stu says
    May 7, 2010 at 12:31 am

    I’m not sure what this animus against the idea of progress implies: are you eager to go back to the caves?
    Here in Germany, we have four dimensions in which to consider things – not a line on which, if you don’t move forward, you are moving backward. As far as civilization and science go, I don’t see anything that looks at all like cumulative movement in any direction or combination of them, but rather a kind of multifocal swelling and bursting, as in the mud geysers in Yellowstone. Something good turns up here or later, and is (necessarily, as far as I can see) accompanied by something bad there or now. We can be saved by medical techniques from an earlier death, only to die a prolonged death later in geistiger Umnachtung. Etc. etc. This is hardly a novel take on things.
    As far as Google Translate goes, it seems that I just don’t know what people have been trying to accomplish with “translation tools” in the past. The results seem to me pretty meager. Let’s consider a gatherer who has had to spend all day scrabbling in the undergrowth to find a few nuts. I suppose he considers it an improvement to find a tool that helps him collect the same number of nuts in half a day, so he can spend four additional hours staring at the walls of his cave. I am more of a hunter who prefers to ensnare rabbits and other hunters, and always has a full belly.
    There is so much meat in the German, English, French and German books I can choose to read that I can’t easily imagine what spiritual nourishment I might get from hamburgered “translations” of texts in languages I don’t know. Even if I had only one language, I wouldn’t have to live off scraps of intelligibility.
    Can you give me examples of what you have profitably used such “translation tools” for ?

  58. Nijma says
    May 7, 2010 at 1:35 am

    I have been using Google Translate for a long time to read Sig’s blog in French. It’s very time-consuming but without Google translate I wouldn’t be able to read it at all.

  59. Nijma says
    May 7, 2010 at 1:56 am

    Firefox has an excellent toolbar called FoxLingo that lets you paste whatever you want to translate with just one click. I use it more than I use the address bar with google, urban dictionary, amazon, wikipedia, and all the other search engines combined.

  60. Grumbly Stu says
    May 7, 2010 at 2:36 am

    I hope my sarcasms don’t come across as a multilingual superiority gambit. As I said, one language is quite good enough, since there are plenty of excellent translators and translations. I’m just surprised that Hat, who has several languages under his belt and does copy-editing, is not more critical about the quality of what comes out of these translation tools.
    BR’s point about understanding people well enough to make a buck off them is a good justification for wanting to use these translation tools. Maybe machine translation will someday work really well, but I’m not missing it.

  61. bruessel says
    May 7, 2010 at 4:11 am

    “I’m just surprised that Hat, who has several languages under his belt and does copy-editing, is not more critical about the quality of what comes out of these translation tools.”
    My sentiments exactly.

  62. language hat says
    May 7, 2010 at 8:36 am

    I’m just surprised that Hat, who has several languages under his belt and does copy-editing, is not more critical about the quality of what comes out of these translation tools.
    I’m very critical of them; where did you get the idea I thought they were peachy keen? If you recall, what I said was that I was “reluctantly impressed” with Google Translate; it goes without saying that its results are not comparable to those of a human translator, but they’re usually readable and comprehensible, as opposed to those of other automatic services. And I don’t understand your point about knowing languages: yes, I know several languages well enough to read them easily, but that leaves, oh, several thousand that I don’t, and if, say, No-sword links to an interesting-sounding article in Japanese, it’s a very good thing to be able to plug the URL into Google Translate and get the gist of it. Obviously I’m not about to use it to read Hungarian novels or Korean poetry, and if you read exclusively for “spiritual nourishment” I guess you wouldn’t have much use for it.

  63. Ø says
    May 7, 2010 at 10:34 am

    Even the crude Vogon poetry of the Romantic hero can understand Arthur Dent.
    This sentence nourishes me spiritually.

  64. Grumbly Stu says
    May 8, 2010 at 3:10 am

    spiritual nourishment
    My use of that expression was merely in line with the hunter-gatherer metaphor about what one prefers to eat. The expression sounds rather precious, so for that reason alone I am now obliged to renounce it.
    Let me just say that I am a natural-book freak. I prefer to fish my books straight out of the sea. Commercially processed, machine-translated scraps of intelligibility disagree with me and are probably oncogenic.

Speak Your Mind

*