Google Does Care - We Heart Google

As al-Fil is currently in the desert, I promised him I'd actually post stuff while he was away. Well, here's an update on the Google mini-scandal from mid-July which began with the finding of a derogatory translation of "gay" and an unsympathetic letter from Google, but is now A-OK:

In the MEGJournal e-mail, al-Fil received this from Google:

Dear Sir,

I have just seen your complaint re the translation tool offered by Google. First of all please accept Google's apology if this translation has offended you. As you may well be aware this tool is still in Beta phase and hence some bugs or incorrect translations will occur. During the Beta phase many of our users provide us feedback on issues such as this, so we can take corrective action. I thank you for bringing this to our attention and strongly encourage you to provide us with extensive feedback of any other mistranslations you might come across.

Also, allow me to take a few minutes of your time to explain the mechanics of how the translation engine works. Our solution is built around a statistical model that depends on previously translated material to statistically determine translations for new sentences. The system continuously learns to update and improve translations. As a starting point we have ingested a number of translated documents and use this to provide the service you now see. Unfortunately many sources on the web use the translation you have seen.

[Here is a link he put in the e-mail here, but is so long, it messes up our formatting, and I saw the complaints about our formatting before, thank you very much!]

As we continuously improve the service and add more documents such occurrences should decrease. In the meantime we are working on fixing this error in the very near future and hope you accept our apology and understand the nature of this service and how it provides translations based on parallel data and not through human intervention.

Once again please accept my thanks for highlighting this issue in our feedback and feel free to provide us with any feedback you feel might improve or enhance our translation service.

Best Regards

Sherif R. Iskander
Regional Business Manager
Middle East and North Africa

Is that sweet or what? Pink News also ran a story on the issue. It said:

Despite the abundance of more derogatory slang in Arabic, Ali Asali, administrator of GayEgypt.com, one of the Middle East’s leading pro-gay websites, agrees that the term [luti] is unsuitable, he said: “It's not the term used on the street for abuse, there are hundreds of these which vary from country to country and indeed from region to region within countries. You could argue that the terms “khawal” in Egypt, “pédé” in Algeria and “ajala” (meaning bicycle) in upper Egypt and I could list many more, are much more abusive. However the term looti is still inappropriate.”

The controversy over “luti” arose about a week ago when the administrator of a blog called The Middle East Gay Journal wrote an open letter to Google upon his discovery that the international company's translation tools translated the word "gay" derogatorily into Arabic. Upon receiving a perfunctory, perhaps automated, response, the administrator was irked and spread the word to numerous other blogs, which spawned more letters to Google.

From his office in Egypt, Sherif Iskander, Google’s business manager for the Middle East and North Africa, told PinkNews.co.uk that he would fix the problem. He said that he had been out of the country for a few days and had learned of the problem upon his return.

“The machine is learning,” he said, emphasising that Google’s translation tools were still in their early phases, and they often went into the system to re-teach it better translations. “Several examples like this have come to my attention,” he said, adding, "Issues like that should not stay in the system." He said that the problem should be fixed in a few days.

Nevertheless, Mr Iskander welcomed the input, "We totally depend on user feedback to fix issues," he said, adding that when problems with translation are reported to Google, it allows them to improve the system.

Google’s translation tools use an approach similar to the methods used to decipher the Egyptian hieroglyphs on the Rosetta Stone – they take identical bodies of work in two languages and compare them side by side.

In Google’s case, they use the immense corpa of the United Nations. Despite using documents that totalled over 200 billion words, however, there were still some terms unknown to the tools.

To solve this quandary, the Google tools access online dictionaries to search for translations. “This is where most of the problems arise,” said Mr Iskander, indicating that the dictionaries often offered inadequate or imprecise translations, without context. Sadly, many of these online dictionaries employ “luti.”

Mr Iskander reiterated that Google’s translation services are a “very powerful tool” that is “opening up the Middle East” to non-Arabic speakers.

He said that the translations are far from ideal, but are meant to give people an idea of what is being written in other languages, without having to actually learn to speak them, "It's like a five-year-old that knows two languages…it's better being stuck with a five-year-old than someone who speaks only one language," he explained.

On checking Google's translation page today, I remarked that when "gay" is put into the system, "مثلي الجنس" is now returned instead of "اللوطي". Thanks, Google!

I must say that Google did a great thing. I doubt, although I have no proof to back it up, other companies would be so quick to change such an error. Anyway, kudos.