The scalability of language and conversations
Nokia's Jan Chipchase on the evolution of mobile

The scalability of language; machine translation

iStock_000001812081XSmall.jpgThe scalability of language is something that's been on my mind a lot lately. In my post yesterday, I looked at the top issues that the language hurdle presents with digital content and conversations.

Today, I want to look at the error of using machine translations. If you're not sure what I mean, let's look at an example. Let's say you wanted to go to and get it translated into English. You would go to a site like Google Translate and enter the URL into the "translate a web page" option. Google then goes through the site and does a literal translation for you.

The problem is that the translations don't account for many factors including local dialect, slang and trendy language. As Mario Vellandi put it in his comment on my last post:

"I'm trilingual and must say that although I adore languages, especially English, there are many phrases that are culturally bound and cannot be translated without some serious human assistance to the computer. Sometimes it's metaphors and similes. Other times, humor and irony are at play. Then there's slang and trend/fad language...It's crazy, but fun."

The real test
The true test of any language translation service is the ability to translate language two ways. What I mean is take a sentence that you want to use. Translate that phrase to the other language and then take the translated version and translate it back to the original.

Here is an example:
Take the first two sentences from yesterday's post:

There is a game that every American kid plays in school at one point in time called telephone (also known as Chinese whispers). The idea is that all of the kids line up in a single-file line and then the person on one end of the line whispers a sentence to the person next to them like "Steve Jobs is the CEO of Apple, pass it on".

Now I'll translate it into Spanish using Google Translate:

No es un juego que cada chico juega en América en la escuela en un momento llama por teléfono (también conocido como chino susurra). La idea es que todos los niños de la línea en una sola línea del archivo y, a continuación, la persona en un extremo de la línea susurra una frase a la persona junto a ellos como "Steve Jobs es el CEO de Apple, dárselo".

And now let's look at it when we translate it back to English using Google Translate:

It is not a game that every kid plays in americas in school at one point called by phone (also known as Chinese whispers). The idea is that all children of the line in a single file line, and then the person at one end of the line whispers a phrase to the person next to them as "Steve Jobs is the CEO of Apple, giving it".

Pretty clear isn't it? The overall meaning is totally gone (even reversed in this case). Just imagine what would happen if you were trying to do real-time translations. Google does take steps toward humanizing the machine with their "Suggest a better translation" link which lets native speakers contribute a new, more accurate translation. Here is a screen shot of that process:

Picture 13.png

The point here is that if you automate translation, you are not going to communicate clearly to your audience. It's worth the expense and effort to make sure that your key information is translates by a native-speaking human being.

Technorati Tags:
, , , , ,

If you liked this post, you can subscribe to the RSS feed or sign up to get updates FREE by email.