The Battle of the Translators: Man vs. Machine

FunnyTranslator.com demonstrates the weaknesses of machine translations by translating phrases back and forth 56 times, and showing readers each step along the way. “Will you translate this?” becomes “Payment?” “To be or not to be, that is the question” becomes “Ask.” Apparently there’s still a need for human translators.[%comments]


kureshii

Translation Party (http://translationparty.com) does something similar: it translates back and forth between English and Japanese, seeking equilibrium. Although restricted only to 2 languages, the results are no less hilarious.

Eric M. Jones

"When I use a word," Humpty Dumpty said in rather a scornful tone, "it means just what I choose it to mean - neither more nor less."

Why am I underwhelmed by the foibles of translations? This seems expected in a field as complex as utterances of the human brain.

But take heart, all tasks that can be stated as algorithms can now more easily be done by computer.

Some cultures have tries to structure their languages along rigid guidelines. But people don't pay much attention to the rules.

Thank God the French don't have any word for entrepreneur.

Which became: "Indeed, in some font in 1956 compared with him."

A fair comparison to reality would be to involve 56 professional translators in a similar study.

Rosewood

I love google voice's voice to text feature but it has a hard time translating most everyone who calls. What I do find interesting is that it does a really remarkable job when I get calls from anyone with an indian accent. Similarly the Nintendo Brain Age game always had a problem understanding the word "blue." Many people suggested saying it as "brew" and sure enough, that worked every time.

I wonder if these cultural biases that exist in speech to text programming (and so many other things in life) also exist in language translation.

I always found it interesting when we would have to do translations in my upper level spanish classes how many different results there were. So if 15 students working on their Spanish major can't come up with the same Spanish to English translation, I'm not sure how I can expect a computer to do much better.

Brett

I wonder what kind of results you'd get if you took "To be or not to be, that is the question" and had 56 professional translators (preferably ones unfamiliar with Shakespeare) shift the phrase back and forth between languages...

Adam Kolber

It would be interesting to see how well human translators do at the same task. Fifty six translations seem like an awful lot. I imagine that, at least for some phrases, the human results would also be amusing (kind of like a translation version of the "telephone game").

Alex Buran

Thanks guys for the review. I am excited and humbled by that!
By the way, that tool is multilingual. If you paste the text in any other language other than English, it will produce the funny mistranslation back in the same language!

Richard

Eric, are you trying to be ironic/sarcastic?

BTW, Dubya didn't say that (though he is dumb).

Marc-O

Translation, even human translation, is a lossy process. That means that is it often not possible to translate an expression with 100% accuracy from one language to another. While we're at it, putting thoughts into words is also a lossy process, and can often lead to misunderstandings, without crossing any language barrier. Many words have more than one meaning, sometimes ambiguous even with context. The same can be said of syntactic structures.

Between two languages, it's common to find an instance of a perfectly precise expression in one language that will have more than one (imperfect) match in the second language. Now back to the first, and you've already lost some precision.

A bit of mathematics now. Assume for each and every translation, we have the same probability of getting the right translation, say 98% (which is really high for automatic tools), then it takes 35 steps to come to less than 50% accuracy. Now take into account that for most of the language pairs in those samples, accuracy is most likely much worse than the 98% we used. It might never ever attain this level, even for human translation.

To be really fair in the comparison, we should put human translators to the same task, each one unaware of the game being played. I'd be surprised if we got great results, for anything more complex than trivial words or well known sentences (like 'To be or not to be'). Most likely, we'd get something funny in the end, like in a game of broken telephone.

Of course, this would be way to hard and costly to organize, and tricks like FunnyTranslator are only possible because automatic translation is cheap and rapid.

Now, it's not really a proof for anything. Multiple back-and-forth translations are hardly a credible use case. Since most people are not close to master their own language's subtleties, let alone know and speak two different languages, we should probably be grateful for the relative quality of free and rapid automatic tools. For their usual job of doing only one translation, and the result being used to understand a text rather than blind copying for print (I might be wrong on that though), automatic translation tools are still useful, and often, good enough.

Read more...

jimbino

A great part of the machine translation problem has to do with the fact that words are often misspelled, the subjunctive mood is abused, "due to" is misused, and political correctness requires abortions like "everybody needs to bring their own towel," and so on ad nauseam.

Indeed, I find it much easier to converse with a machine than to endure listening to a product of our public school system or, god forbid, read something he has written.

notmelbrooks

Is there a similar website for translating Sarah Palins verse?

chosunking

Its not a translation thats important but interpretation.
Knowing the language AND culture is whats important in communication.
And the nuances therin.

John

Machine translation has taken great strides over the past two decades, and translation-memory (TM) software has also improved considerably, but the human factor in translation and interpreting are still so vital that it's hard to imagine a world without them.

One of my favorite examples from English is the verb "to get", which when coupled with prepositions (i.e. get up, get out, get over, get by, etc) can be translated into dozens and dozens of verbs in other languages. Add in to that potential potential nouns (a get-up) and you've got a programmer's nightmare.

TM software is a big help to those of us in the industry (disclosure: I'm a professional translator), but for the foreseeable future, I don't think that we're in any danger of losing our jobs.

Nikki

John,

do you think you could have announced to the world that you are a professional translator without making it sound like TM is a sort of MT? TM tools, in essence, can't replace translators any more than Photoshop can replace designers. MT is a different story.

Nor does the problem have anything to do with getting computers to check whether there is a preposition after a verb and adjust translation accordingly.

Bobby G

What's that old Italian saying? "Traduttore, traditore" ... which is supposed to mean "Translator, traitor". In other words (no pun intended, given the topic), those who make translations inherently lose a part of what is being translated during the process, "betraying" the original message.

I can only imagine what the Italians would say about 56 computerized translations...*

*: this was translated into

"In addition, Italy, in 1956 the group who can not think of anything to sell, 'he said.' "

Ben D

Ever heard of the game "telephone?" Obviously humans have similar issues.

Steve

Input:
Twas brillig, and the slithy toves
Did gyre and gimble in the wabe;
All mimsy were the borogoves,
And the mome raths outgrabe

56 Translations Later:
Output:
Blood crystal and the average Closing zhimbl slithi: If you mimuzusupesumausu borogoves.

Amazing that it hasn't lost any meaning!

Paul

If machine translation were perfect, we would see it integrated everywhere... unfortunately, that's not yet the case...and probably won't be for a while. There are some interesting companies trying to bridge the gap between Human and Machine translation.

An Economist article covered http://www.speaklike.com/ a few months ago. SpeakLike crowdsources the translation process from thousands of qualified translators from around the world. At the end of the day, seems like language is still human!

Luke

Somewhat on the topic ... an article was reviewed by http://www.translationperfect.com (in their blog section) where some hot shot proposes to use computers and poor students to do "quality" translations. Fun reading too.

Jessica

"It was the best of times, it was the worst of times"...56 translations later, it became "Ndoni time"

What?!