FYI.

This story is over 5 years old.

Identity

Why Google Is Adding Gender Pronouns to Bahasa Indonesia Translations

It's not the technology. It's us.
Illustration by Dini Lestari

The Indonesian language is pretty vague, a fact can be seen as both a blessing a curse, depending on who you ask. Sure, the language's grammatical rules are pretty flexible, and its vocabulary makes for sentences that can be, at times, light on the specifics. But it's also gender-neutral. Or at least it is until someone (or some algorithm) tries to translate it into English.

A Facebook post by Dina Utami, a PhD candidate in Computer Science at Northeastern University, pointed out how Google Translate's algorithm suffers from some of the same gender biases as our society does—rendering "dia" and either "he" or "she" depending on the career or activity that follows. The Facebook post shows how a sentence like "Dia seorang president," is translated as "he's a president," while "Dia menangis" becomes "she's crying." The thing here is that in Bahasa Indonesia, none of these sentences specify whether "dia" is a male or a female. Google's algorithm does it instead.

Advertisement

Dina told me that she wasn't surprised at all.

“I’m not shocked that Google Translate has a bias,” she told me over the phone.

But it's still weird. Google Translate takes a language that lacks gender-specific pronouns and uses an internal bias to split "dia" into "he" and "she." Meanwhile, some activists and academics in the US have been working to de-gender the English language. The Associated Press Style Guide, which a lot of newspapers worldwide use as the standard, recently added a gender-neutral singular "they" to a list of approved pronouns. And in romance languages like Spanish where there are masculine and feminine forms of words, some have adopted the word "latinx," instead of the gendered "latina" and "latino" as a way to weed-out societal gender stereotypes.

So, in some regards, we were already ahead of the curve with the simplicity of Bahasa Indonesia.

“In many European languages, it’s almost impossible to speak in gender neutral terms," said Katrin Bandel, a lecturer on postcolonial studies at Sanata Dharma University, in Yogyakarta. "It’s a lot of work to modify a language so it’s not biased. That doesn’t apply to Indonesian."

In Katrin's book Sastra Nasionalisme Pascakolonialitas, she wrote that while language itself is incredibly political, it's often difficult to de-gender a language after-the-fact. And it's not easy keeping a gender-neutral language neutral either.

Advertisement

That's where the Google Translate problem comes in. The company has never released a statement about how its algorithm works, but most coders agree that it uses a form of Statistical Machine Translation to figure out how to write a sentence in more than 100 languages. Unlike, say, a traditional translation service, Google Translate doesn't deal with grammatical rules, dictionaries, or the advice of linguists. It, instead, learns by analyzing a huge database of human-made translations to see how sentences are most-accurately translated, according to Carlos Alberto Gómez Grajales, an expert in applied mathematics, analysis, and statistics.

This allowed Google Translate to be far more accurate that similar services back when it launched. And it only gets better over time as it learns from more-and-more translations. But when an algorithm learns from humans it picks up human biases as well. So, over time, all doctors become men and all dancers women if thats the way native speakers of the language typically use their own language.

Google Translate is a mirror of our own internal biases. Dina told me that she was inspired to do her study after reading a similar post from Alex Shams, an Iranian-American academic at the University of Chicago, on Google Translating Turkish.

“Google Translate magnifies biases that exist in our culture—like unequal access to employment for women in all fields, or culturally constructed notions that women are unhappy, lazy, hopeless romantics always looking for husbands—and transforms them into translations,” Shams wrote on Facebook.

It's another reminder that a gender-neutral language doesn't necessarily mean a gender-neutral society. Indonesian society is pretty patriarchal and these kinds of beliefs rise to the surface regardless of our language.

“Of course it doesn’t mean that people who use Indonesian language aren’t sexist," Katrin told VICE. "It doesn’t matter how ‘neutral’ a language is, as long as the content is sexist, the society and culture are sexist. So we need to address the gender problems in our society, not only in the language.”