Fun with Google Translate, MidEast way - who killed whom?

Inputing "Israeli kills Palestinian" in Arabic translates "Israeli killed by Palestinians"

While we all acknowledge that online translation websites are far from accurate, they generally do a good job giving an approximate rendition of the meanings of the original text.
And with its 51 supported languages, Google Translate has positioned itself as a leader in the online translator market.

Google Translate claims to take a "different approach" in creating its translation algorithms. One notable difference is the option to "contribute a different translation", offered at the bottom of each machine-generated translation page.

Last week surfaced what I first thought of as an amusing quirk. You'd enter the text "اسرائيلي يقتل فلسطيني" -- which is Arabic for "Israeli kills Palestinian".
It's a simple enough sentence. Subject, verb in the present tense, object. Surely Google Translate can handle that?

Well, if you hit "Translate", you get...

"Israeli killed by Palestinians".

The subject and object have been inverted; or to be more precise, the verb has shifted from the active to the passive form. And the one Palestinian is now many.
Pretty confusing to a potential reader who cannot decipher the original.

Funny quirk, I figured? At the suggestion of other friends, I tried a few other objects (and others did, too).

Entering, in Arabic, "اسرائيلي يقتل هندي" - "Israeli kills (an) Indian" (or "American", or "Brit", or "dog") - well, it's the poor who Israeli dies.

Apparently he can only take on the Brazilian fellow: "اسرائيلي يقتل برازيلي" ("Israeli kills Brazilian") is translated (almost) properly, with the verb shifting to the past tense ("Israeli killed Brazilian").
Not very flattering for Brazilians, eh! :)

Changing the subject of the sentence, however - when the murdered is then of any other nationality - the phrase translated properly. Funny, huh?

I did notice something else though. Dropping the object (so just subject+verb) gives very different translations. I tried "Palestinian kills" and "Israeli kills" in Arabic and this is what I got:



"Israeli killed", but "Palestinian kills".

Is that a joke of some sort? Some Google developer with an acute sense of humour and geopolitics? A coordinated user manipulation, the GT equivalent of a Google bomb? If many people decide to "Contribute a better translation" to the same sentence eventually GT will adopt the new sentence as correct.


I nevertheless decided to give Google Translate the benefit of the doubt. After all, I thought, it uses original text to compare its translations, and the English-based media seldom uses a sentence as straightforward as "Israeli kills Palestinian". When such event occurs - as it often does - the subject is usually "the IDF" or "Tsahal" or something...

So I tried a different pair of languages: Arabic and Hebrew. If, as I figure, the change was made by Israelis - I can imagine them, a class of Israeli kids who decided to pull a harmless joke on Google by submitting the same "better" translation dozens of times - then they're likely to have done it to this pair of languages, right?




Translating "Palestinian kills Israeli"(فلسطيني يقتل اسرائيلي) into Hebrew you get הורג פלסטיני ישראלי" - which is close to the original. (running the result in reverse leads to the original text).

But here's the fun bit: translate 'Israeli kills Palestinian' (اسرائيلي يقتل فلسطيني) and you get the unexpected but linguistically very correct (ישראלים שנהרגו בידי פלסטינים): "Israelis who were killed at the hands of Palestinians". Both subject and object now in plural, verb in passive form, with a few extra intermediary particles and words.


This cannot be a translation mistake. Another instance is here: until a few weeks ago, attempting to translate "long live Palestine" (تحيا فلسطين) from Arabic to Hebrew gave (זמן ישראל לחיות), which approximately read as "it's time for Israel to live!". Same: impossible to be a programming mistake.
(it has now been corrected).


Rather than a deficient algorithm, this result - and the linguistic precision of the erroneous wording -is assuredly human-caused. Which points out to a major deficiency in Google Translate: it trusts people.

The obvious correction to this problem is to have user-submitted "corrections" checked by a human translator. This might however turn out to be very costly and time-consuming... But until then, Google Translate will be open to user manipulation. It's the trade-off to make.