Pages

Wednesday, July 3, 2013

Machine and Human Translation Based Humor

As the momentum around machine translation technology continues to build, and we see more and more discussion about post-editing at conferences, (I recall a roomful of people gasping in horror when I said “Post-editing is in your future” at a Localization World conference in Montreal in 2006) I thought it would be fun to see if I could dig up more funny stuff about MT. Especially since my blog analytics tell me that my original post on Mocking MT has been particularly popular of late.

As MT enters the mainstream of business translation we still see a lot of misinformation and it is amazing to me how still so few understand that all MT engines are not equal and that success with MT requires expertise that is not easily available or rapidly acquired. However, it is encouraging to see more translators urging peers to learn and adapt rather than resist and malign the technology that actually does make sense to many and is used by millions every day.

The most destructive myth about MT that I think undermines its long-term use and potential, is that anybody can just whip out an MT system in an instant (why wait?) and it will be immediately useful. Getting good MT systems that provide long-term strategic advantage is still a complex and challenging undertaking or “really hard work” as George W. Bush used to say. The ROI is not well understood and I plan to write more about this.

We also continue to see blatantly wrong information (read the comments on the post) being disseminated in the mainstream translation press, allegedly disguised as expertise, so we are still a long way from widespread, reliable MT technology in professional use. Just as the quality and competence of language service agencies vary greatly, MT solutions and expertise also vary greatly and unfortunately there is a growing community of quick-fix MT experts out there. While technology agnostic approaches can sometimes make sense for buyers with a very clear domain focus, more often than not general “technology agnostic” approaches equate to technology ignorance or conclusions drawn on miniscule or biased samples.

The language industry apparently has 25,000 translation agencies, many with questionable credentials and competence, so as we saw from a recent buyer presentation, it is possible to get quotes that range from $310 to $10,430 for the exact same translation job description. So given the cottage-industry or wild west character of the broader “translation industry”, all customers need to be on guard against misinformation and false prophets and promises.

Just to be clear, while I occasionally do mock machine translation errors,  I have no doubts that in the right hands this technology can solve real business problems and provide real business leverage.

This is ‘Japanese Titanic‘ with a script that is made up entirely of lines generated by a free online translator after turning the original English into Japanese and back again. I have always felt that doing this is a particularly pointless way to assess MT system competence since one can be assured of poor results. While this is funny I don’t think it is quite as funny as the “movie” or the other examples in the original post. My personal favorite is still the Bollywood video pseudo-translation which is really way funnier, since it is a guy who is just saying that the sub-titles are what he thinks he is hearing.

Script produced by running some lines repeatedly through MT

If that was not enough for you you could also check out the outtakes and deleted scenes here. Avoid the link if you  find sophomoric humor offensive or think that four letter words have the power to send you to darkness or fiery places.

Here is an example from a Japanese translation that is warning label on a massage towel, that is hopefully the result of MT and not human translation.

 

It starts off normally enough, warning us to keep the towel away from naked flames and telling us that it is effective in the removal of dirt and impurities from the skin.

But then things get a little bit odd:

“Skiing and snowboarding are so cold; it makes me not want to go outside.”

O-kaaay. Rather a strange time to come clean about your feelings towards winter sports, but everyone needs a canvas to express themselves on, we suppose.

But then it gets worse. The label suddenly starts spouting menacing prose like some kind of maniacal fortune-teller:

“You who selected the snowman, frozen over but enduring the cold, have something in your past. You value your parents’ opinions more than your own.”

Er, what? The label starts to repeat its frightening message before cutting itself off and giving a direct order:

“Look at your partner! “You were raised with many burdens upon you.”

Smile and nod, smile and nod. When it’s not looking back slowly out of the room…

Source: ねとらぼ Title image: @AK3ono

Here are some examples that I think are mostly human translation errors, and it is funny how similar they are to machine translation errors.  The problem is that often amateur translators (I presume) involved in these examples, focus on literal strings of words, and do not really speak the target language or trust a friend who assures them that they do indeed speak the target language fluently.

Translation mishaps from around the world

There is a Facebook group that explores Linguistic Humor in much more interesting ways that some might find fun.  Anabela M. Barreiro  tries to keep it all clean and encourages contributors to not insult or humiliate anybody.

Here are some recent examples:

 

I just had to put this clip in here since it has a music track that has the language of tabla in it. Yes there is indeed a language that tabla players have.

Strange translations from around the world

Please let me know if you have found or know of funny MT based humor, especially short movie scripts that can sometimes be quiet hilarious. Have a happy holiday weekend for those in the US and a wonderful weekend to the rest of you who may not care about July 4th.

13 comments:

  1. From the Economist: http://www.economist.com/blogs/johnson/2013/07/botched-translation?fsrc=scn/tw_ec/mottakelse_to_new_york

    ReplyDelete
  2. Great article. Looking forward to your post about ROI in the context of machine translation. I've been musing about what magic combination of volume, frequency of updates, nr. of languages and/or other variables would justify the initial and ongoing investment into an MT system.

    By Ingrid Allsop

    ReplyDelete
  3. Hi, Kirti,
    In German we call this "ÜbeLsetzungen"...
    I found a rather good example myself in the restaurant of our hotel in the Dominicanian Republic in 2008: Papa frita -> German: "Gebratener Papst" (i.e "Fried pope") ;-))). I still have this photo and can send it to you.
    Cheers, Gabriele

    ReplyDelete
  4. Really it is a nice blog; I would like to tell you that you have given me much knowledge about it.
    machinery manufacturers

    ReplyDelete
  5. I totally agree with you that the most destructive myth about MT that I think undermines its long-term use and potential.

    ReplyDelete
  6. Yeah i agree with you there is very vast difference between machine translation and human translation but now all want work fast so most of are doing machine translation anyway really it is a nice blog.

    translator | translator services

    ReplyDelete
  7. Hello,
    Along with machine translation we need transliteration and pronunciation so English or (any language,e.g. Tamil,Telugu,Urdu) can be learned in your own language script and let machine do the rest.


    Needed Google Translate Improvements:

    1-Translate from English to Hindi or Gujarati
    2-Show translation in Roman script (to read Hindi or Gujarati in Roman script)
    3-Show English pronunciation in Hindi or Gujarati (to read English in Hindi or Gujarati Script)
    4-Make it reversible
    5- Scroll on English word to see translated word
    6-Can be used as instant Dictionary
    7-can be used as India's two languages state formula.
    8-can be used to learn English in your own script.

    Please work on this public project and challenge Google translate service.You may add other Indian languages

    what is your name?.................वॉट ईझ यॉर नेम?...................વૉટ ઈઝ યોર નેમ ?
    ................................................तुम्हारा नाम क्या है?.................તમારું નામ શું છે?
    ................................................Tumhārā nāma kyā hai ?.....Tamāruṁ nāma śuṁ chē?

    father........................................फाधर....................................ફાધર
    .................................................पिता.....................................પિતા
    ................................................ Pitā.....................................Pitā

    http://translate.google.com/#en/hi/

    If one can Read Hindi or Gujarati in Roman Script, Why one can't read / write English in Hindi or Gujarati Script ?

    ReplyDelete
  8. This comment has been removed by a blog administrator.

    ReplyDelete
  9. In my own opinion, machine translation isn't reliable when compared to human translation. Just my own thought.

    Technical Translation Company

    ReplyDelete
  10. I just found this blog and have high hopes for it to continue. Keep up the great work, its hard to find good ones. I have added to my favorites. Thank You.
    bobcat colombia

    ReplyDelete
  11. I like the in-depth explanations in your posts. It makes it easier to follow the posts for someone who is not as proficient with technology.

    ____________________
    Karolina Karczmarek-Giel
    Office Assistant
    www.wantwords.co.uk

    ReplyDelete
  12. Obviously a person also have to go through the considerable practical knowledge in addition to expertise of any interpretation organization. Review the actual qualifications in their human translation online thus you will be sure that they may become dealing with ones interpretation requires in the the majority of professional way.

    ReplyDelete
  13. That's actually been too much informative, was looking to find the best place for translation.

    ReplyDelete