Posted by: patenttranslator | May 9, 2012

Advances of Google Translate Put a Premium on Human Translation

The title of my post may sound a little counterintuitive as I can see frequently on blogs and discussion groups of translators that many people are scared silly that Google Translate will soon phase human translators out of existence.

But I have to say, I love Google Translate. Instead of having to look for words in heavy dictionaries, I can now found answers to my questions with a few keystrokes on my keyboard.

Google Translate and other MT programs made life much easier for human translators in particular in my field of technical translation. These tools also made life easier for monolingual people, such as our clients who need quick access to information in foreign languages. With Google Translate the access is instantaneous and so far it has been free.

But fortunately for me and other translators, Google translate does not provide access to accurate translation. The words and sentences magically appearing on the screen in English when I type something in Japanese, French, German, or Russian or Czech are accurate only when I say that they are accurate. Incidentally, this is the same decision that translators have to make when we look at a word suggested to us by a dictionary. In the twentieth century we were using mostly dictionaries printed on paper, in the twenty first century we will be probably using mostly smart dictionaries online which are much easier to update and which can provide much more context.

Because I have been studying the languages that I translate for decades, I understand them fairly well. Unlike my clients who have been studying other things than languages, I can see the mistakes in machine translations because I can compare the target language to the source language. I can usually even tell why the machine made a mistake.

I can also see that Google Translate is able to translate languages that are similar to English, such as French or German, much better than Japanese. This was to be expected, of course. English, French and German grammar is based on rules that were developed by Latin grammarians more than a thousand years ago. These rules include concepts such as subject, singular and plural, future tense, or a single word as a unit, and many other concepts that most English, French, Russian or German speakers take for granted. But these logical building units used in European languages either don’t exist at all or exist in a very different form in Japanese.

Unlike every European language that I translate, Japanese grammar dares to completely ignore the rules of ancient Latin grammarians, which makes it a very cool language in my book.

For example: Japanese has a special kind of subject called “wadai”, which can be translated roughly as “topic” and which does not exist in any European language as far as I know, plural is usually not specified (Japanese peoples say that they don’t need it), there is really no future tense, only a “probable” tense (they again say that they don’t need it), there are no spaces between words (who needs spaces when you can use elegant Chinese characters?), etc. I agree, they don’t need to know what is a word. There are more important things in life.

Of course, the Japanese language has many other features lacking in European languages. From the viewpoint of a Japanese speaker, English must be in many respects a pitifully inadequate language incapable of expressing basic concepts that are very important for everyone’s life.

As a classical example, there are many ways to say “I” in Japanese, and each of them means something else: (watakushi, watashi, washi, boku, ore, atashi, uchi, kochira, [which really means "here"]) … I don’t remember them all. But unlike in English, this “I”, is usually missing in a Japanese sentence, while in English, I cannot say anything without using the word “I”.

The point in Japanese is to make a distinction between different types of “I” depending on whether I am a man or a woman and depending on my relationship to you, while English and many other European languages are much more “I”-centered, without offering any particular context.

How can software cope when there are no equivalents for so many words and so many logical building blocks of other languages? How can a human translator translate without numerous highly intrusive “translator’s notes”? A real translation is almost an impossibility even for a human, and yet, arrogant humans assume that a piece of software can translate anything.

These are just a few example of many traps in one language that make it an extremely difficult for machine translation programmers to figure out algorithms that would work for translation.

Even when I look at Google’s machine translation of a patent from German or French, there are usually many mistakes in every sentence that will often render the meaning in English impossible to grasp. I don’t pretend to have an objective opinion, but I do think that Google Translate is no threat to translators who work from other languages either and probably never will be.

The problem is, although machine translation makes more sense now, you really have to be a translator, and a good one, if you want to understand machine translation.

Thank you, Google Translate. I for one appreciate all you have been doing for us human translators, and I hope that you will continue educating our clients through practical experience with machine translation about the difference between machine translation and real translation, also known as human translation.

About these ads

Responses

  1. I like the way you see the matter of MT. Google Translate is not that bad for us translators at all, but it is for translators who possess sufficient knowledge in languages. “A real translation is almost an impossibility even for a human, and yet, arrogant humans assume that a piece of software can translate anything.”

    Some people would like to believe that the Babel Tower is near, but we know that it has been destined to collapse. Your example with the Japanese 話題/wadai is a good one. Even Chinese who usually believe that Japanese stems from Chinese would have a lot of problems with the usage of は and が just because there is no correspondence in Chinese for the difference between an usual subject and a subject of the kind of wadai. (Both “月は美しいですね” and “月が美しいですね” are translated by Google Translate as “The moon is beautiful.” I wonder that anyone who has no idea of Japanese could ever figure out what difference it is between these two simple, distinct sentences.)

    Google Translate is a nice tool when we translate from a non-native language of which we have sufficient knowledge, but it is the translator who makes the translation “human” when he knows the subject matters and other aspects of human transactions and interactions. The further MT develops, the more some of us translators can find benefits in it. Yet, there is still a danger, that is, some translators could easily be degraded to wheel treading hamsters of posteditors.

    BTW, thanks for the music of Kitaro (喜多郎/高橋正則). It was the theme of the 12-part documentary NHK特集 シルクロード first aired in 1980. The advisors for the making of it were the writers 井上靖、司馬遼太郎、陳舜臣、加藤九祚. The documentary is fantastic, worth collecting. Just imagine how many translators there were to have achieve that story of mankind. What if they had had Google Translate?

    Like

  2. “Yet, there is still a danger, that is, some translators could easily be degraded to wheel treading hamsters of post-editors.”

    This is the “pium desiderium” of brave new entrepreneurs who were probably shipping slaves from Africa to America in their previous incarnation 200 years ago.

    But I have a feeling it’s not going to work for them the way they imagine it this time around.

    I watch the Silk Road series on NHK when I lived in Tokyo.

    Like

  3. “But I have a feeling it’s not going to work for them the way they imagine it this time around.”

    The paragraph before the quoted one reminds me of a translation ageny owner who wrote:

    “I’ve been running a small translation agency for close to 15 years. Yes, there is still money to be made in this business, but it is a dying business. Very soon there will be a special on Discovery about this threatened species: the lone freelance translator. Running down the tundra only to be shot down by poachers (that is, us, the agencies, the end customers, and specially by the technology). My advice is: get specialized. Very specialized. Find a few customers that are not tech savvy (a tall order these days) and stick to them. Kiss them. Sleep with them. Marry them. This may give you another 10-15 years in this field, and if you are lucky enough to be over 40, that can bring you to retirement or something else. If you are younger, go back to school NOW and learn another profession. You will need it before you know it. Don’ t believe me? Read below.
    That took 0.32 seconds and ZERO dollars. Yes, it is not perfect, but I assure you: about a billion people can read it and understand it.
    [Some funny translation of the above 'pium desiderium' of one of those guys spat out of Google Translate...]”

    I hope that our translation colleagues will vehemently resist becoming wheel-treading hamsters, as Miguel Llorens call those unfortunate ones. Miguel’s blog and yours are both of the most educational ones for translation colleagues. You are superb thinkers and writers.

    Ah, yes, I remember that you wrote somewhere that you move to San Franscisco in 1983. When NHK was airing Silk Road in 1980, I was living in Germany and watched a TV sequel by NHK, 人間の條件, kind of a piece of story of mankind in Manchuria. Marvellous that some German translation colleague had made it to synchonize it in German. This is why I want to stay a translator before I die. This is one way to make sure how alike we humans are and how different we are from machines.

    Like

  4. As far as I can tell, translation business is doing very well in my field.

    Machine translation is no threat to patent translators. Free MT has been available in the patent translation field for well over a decade. One difference in the status before MT and after MT including Google Translate is that unnecessary translation can now be avoided, which reduces the amount of available work.

    On the other hand, more patents can be now identified for human translation with MT, which increases the amount of available work.

    I think that the overall result is more work rather than less work for human translators in my field because MT does not and probably never will provide accurate translation, which is what you need if you want to file or litigate patents.

    Like

  5. Although I enjoy reading your post, it is entirely wrong in many aspects.

    First of all, Google translate is not based on grammar, but on statistical machine translation. This means that some differences in grammar don’t degrade translation quality, others do. For example when you translate from a morphologically rich language (say French) to a morphologically poorer language (say English), the difference is not a problem.

    Secondly, patent language is a technical descriptive language. You will rarely encounter a first person, future tense, politeness or social relationship.

    Thirdly, there is a big chance that patent machine translation will reduce the need for human translations, at least between European languages.Ever heard of London Agreement and Unitary Patent ?

    Finally, the purpose of machine translation is to translate for people not knowing the other language. For most people, any machine translation from Chinese to English is so much better for them than unreadable Chinese characters. Although human translation is of course much better, first of all it is a question of money and timeliness, and secondly you cannot search crosslingually by a human translator.

    Don’t take me wrong: I like the skills of human translators, and I know the weak points of Google’s approach (completely different word order source language vs. target language, relationship between words which are not close to each other in a sentence, availability of similar translated texts).

    However, there is definitively a place for machine translation, and yes it will reduce the work for human translators.

    Like

  6. Dear “WT”:

    I really don’t know what to say to you.

    My post is “entirely wrong in many aspects”?

    How sad.

    So I will just let your comment stand.

    Like

  7. Steve — I’ve been in the same line of work as you for two decades (J>E IP translation) and just recently discovered your blog. I consider myself fortunate to have become acquainted, if only through your blog, with such a shrewd observer of our (very narrow) field. I really mean it–it’s a rare blog on any topic that rewards regular reading.

    On the subject of machine translation, I was startled by how good Google Translate became this past winter after Google gained access to the EPO database, and alarmed when an old contact of mine at a PTO subcontractor told me this spring that their business had fallen by two-thirds after the PTO decided to rely more on machine translation. (I haven’t worked for them for quite some time, but send her an email each year on her birthday.)

    I had nightmares for about a week after receiving this news. The nightmare was always the same: I entered a paragraph or two from a Japanese patent application into Google Translate, and the translation it spit back was flawless.

    Perfect. For free. In seconds. I’m ruined.

    I’d wake up in a cold sweat and have to calm myself with clear-headed reason before going back to sleep.

    During waking hours, of course, the limitations of machine translation are readily apparent, and it helps that business has not suffered in the slightest.

    While I largely agree with your assessment of machine translation, I’ve recently been viewing the situation from a different angle. I’m more concerned about machine translation lowering the barrier to entry into our field by making barely adequate translators appear to be better than they are. As I’m sure you’d agree, much of the knowledge and “tricks of the trade” that has accrued to us over the years and that has set us apart from the newbies and mediocrities is now a mouse click away on Google Translate or a simple Google search. In such a price-sensitive field, this could have ripple effects on our business.

    I suppose as long as demand keeps pace with the supply of translators, all will be well with us. It also helps that the only other place with a sizeable number of J-E translators is Japan, which is, of course, the high-cost producer.

    I also tend to agree with you that the supply of work will probably increase in our field on the high-end. I’ve been trying to swim upstream for a while now, and do more translating of patent applications for filing in the U.S. Also, cost-cutting by examiners at the PTO will probably result in more private work for us as more of their decisions are contested by applicants.

    Anyway, sorry to ramble. I wanted to get something off quick before having lunch, napping and getting back to this way-cool motorcycle patent application I’ve been working on.

    Keep up the good work.

    Like

  8. Hi Furanku (Frank)?

    Thank you for your comment.

    1. I was startled by how good Google Translate became this past winter after Google gained access to the EPO database, and alarmed when an old contact of mine at a PTO subcontractor told me this spring that their business had fallen by two-thirds after the PTO decided to rely more on machine translation.

    I translated a couple of months ago a long Japanese patent (prior art) for an inventor who was about to file his new patent application. He was very happy that the examiner was using a machine translation because it was not very good, while he had access to the real thing through me, although it cost him almost a couple of thousand dollars.

    His lawyer thus had a big advantage over the PTO examiner.

    Some clients, those who are very price sensitive, will go for free MT, probably quite a few. But once they realize that the reason why they are losing in court is that they don’t have a good translation, they will probably start ordering real translations again.

    As you said in your comment, it is important to try to specialize in the “high end” of clients.

    2. Google Translate has already lowered the barrier to entry, but I see it mainly as a good thing.

    Talented translators can use it to gain experience in fields that they would not be able to handle a few years ago. But you still need to know your languages. People who don’t really know much German or Japanese will not be able to translate just because they can now use GT.

    Especially in languages such as Japanese or Chinese, the result of GT is still really bad.

    The emphasis is now more on linguistic education and linguistic skill instead of mainly on knowledge of the correct technical terms, because you can usually get those from MT.

    I hope your nightmares won’t last too long.

    Like

    • Especially in languages such as Japanese or Chinese, the result of GT is still really bad.

      I totally agree. I guess one of the reasons I tend to overestimate the powers of GT is that I subconsciously correct some of the better results in my mind, owing to my proficiency in both languages. The following is somewhat related to what I mean:

      Aoccdrnig to a rscheearch at an Elingsh uinervtisy, it deosn’t mttaer in waht oredr the ltteers in a wrod are, the olny iprmoetnt tihng is taht frist and lsat ltteer is at the rghit pclae. The rset can be a toatl mses and you can sitll raed it wouthit porbelm. Tihs is bcuseae we do not raed ervey lteter by it slef but the wrod as a wlohe.

      (H/T: Andrew Sullivan)

      – Frank

      Like

    • He was very happy that the examiner was using a machine translation because it was not very good, while he had access to the real thing through me, although it cost him almost a couple of thousand dollars.

      Ah, yet another example of the time-honored American custom of passing costs that should be borne by the public sector onto the private sector. My sons play Little League baseball at a very expensive private facility because the town is too cheap to build one for their own kids.

      Like

  9. “Talented translators can use it to gain experience in fields that they would not be able to handle a few years ago. But you still need to know your languages… The emphasis is now more on linguistic education and linguistic skill instead of mainly on knowledge of the correct technical terms, because you can usually get those from MT.”

    Steve, I notice the shift to linguistic education and linguistic skill for quite a while. However, it is not simply a matter of languages when it comes to translation. Knowledge in the subject matters and writing skill contribute fundamentally to the translation quality. That is why we can easily figure out what’s wrong or right with a piece of imperfect machine or human translation.

    MT helps to ensure correct and consistent technical terms, but it is always the translator who decides on the correctness and consistency.

    Who is afraid of MT? I don’t think those who are “specialized” in the “high end” would be afraid of MT or become merely post-editors, because there is much more in translation which we consider good quality than just having texts translated.

    BTW, thanks for the change to モスクワ郊外の夕べ. It takes me back to my childhood, when I had to listen to Radio Moscow in a sound isolated room because communists, Russian or Chinese, were enemies and we were not supposed to learn their language to understand their cause in Taiwan. But I have always been fond of Russian folk songs and that was reason why I came to learn the language.

    Like

  10. Hi Wenjer:

    1. But detailed knowledge of the field can be also acquired by “non-technical” translators who have been immersed in highly specialized translation fields for many years. Many are doing just that because this is where the money is.

    I think it that would be probably more difficult for “techies” to learn a complicated language such as Chinese, Japanese or even German or French through constant exposure to Google Translate. I think they would need to first become really fluent in these languages.

    2. Actually, I had to change the music video because Youtube disabled Kitaro for embedding in blogs. They keep doing that for some reason which is why I have to use a different video every now and then.

    Like

  11. For Japanese and Chinese, Google translate is not yet trained with parallel patent data.

    This will change in 2013 or 2014. You may expect a significant quality increase by then.

    Quality will also increase for some European languages in 2012 and 2013, in particular Italian, Greek , Dutch, Hungarian etc. Russian is also in the pipeline.

    Like

  12. I think you are misinformed.

    Google Translate is already used in conjunction with other MT programs to translate patents from and into Japanese and Chinese according to announcements by EPO and WIPO.

    One WIPO official who works on translation issues also left a comment explaining how this works on the WIPO website on this blog.

    I don’t remember where it is but you should be able to find it using the search function on my blog.

    The main problem is not “having access to a huge amount of data”, that is only one part of the problem, namely the part that can be solved relatively easily as the available computing power is virtually unlimited.

    The real problem is that you cannot “train” a computer to think, no matter how much data you throw at it.

    The algorithms for something like that don’t exist and never will. Translation of European languages into English can be simulated to some extent because the grammar and the structures of sentences are similar. But with languages such as Chinese and Japanese, the linguistic issues are much more complicated and based on the results so far, it appears that Google is mostly ignoring linguistic issues as it is based on the statistical approach.

    Like

  13. Sorry, indeed WIPO allows you to translate into Chinese. However, to my knowledge the training material was not very complete, so more training data is expected leading to quality increase.

    The EPO-SIPO deal was signed only in Nov. 2011:

    http://www.epo.org/news-issues/news/2011/20111129.html

    So far you can’t translate in espacenet from EN into ZH:

    http://www.epo.org/searching/free/patent-translate/faq.html

    I fully agree that within our lifespan computer translations will remain of much lower quality than the translation of a professional translator.

    Like

  14. “Sorry, indeed WIPO allows you to translate into Chinese. However, to my knowledge the training material was not very complete, so more training data is expected leading to quality increase.”

    I hope the quality will increase. Better quality MT is a very useful tool for us human translators.

    But as you said, within our lifespan, MT will not replace high quality human translation, and I would add that this is something that can probably never happen, regardless of how much data you through at MT because nobody can design an algorithm that could replace thinking.

    Like

  15. Ditto,

    “The words and sentences magically appearing on the screen in English when I type something in Japanese, French, German, or Russian or Czech are accurate only when I say that they are accurate.”

    Like

  16. thanks for the list. i have been searching for do follow blogs from long time.this list you shared has made my work easy.
    Technical Language Translation

    Like

  17. I agree with steve, i think automatic translations are no threat to human translators. And also i believe that this will be the case for many more years to come..

    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

Follow

Get every new post delivered to your Inbox.

Join 1,352 other followers

%d bloggers like this: