Bangla is a gender-neutral language [1]
Spoken by more than 228 million people, it is a language that does not use gendered pronouns and verbs. However, when translation technology is developed using publicly sourced biased datasets, we lose this linguistic equality to gender-biased results.
Let us look at a few examples. The Bengali sentence তিনি অধ্যাপক (Tini adhyāpaka ) can be translated to English as both “She is a professor” as well as “He is a professor.” But Google translates it to “He is a professor” every single time.
Similarly, সে একটি শিশুকে খাওয়াচ্ছে (Sē ēkaṭi śiśukē khā'ōẏācchē) can mean both “she is feeding a baby” and “he is feeding a baby” and yet, the Google translation for the sentence only shows “She is feeding a baby.” These results do not depend on the location or the gender of the user.
Translations like the ones above are a reflection of the existing gender roles in society men have jobs and earn money, while women cook and take care of children at home. These results also tell us how certain qualities like bravery are associated with men, while emotions like shame are associated with women. They reinforce the gender binary and create a bias in results that is not present in the language itself.
We are aware of the linguistic complexity involved in translations. But for the problems that we are already aware of, more inclusive solutions can be devised. Google boasts of its efforts to achieve diversity and equity in everything it does.[2] According to data from 2018, Google translates around 143 billion words per day[3], and so Google has a responsibility to all those who use its service to be free of sexism.
The technological solution to this problem can be highly complex for now, but that doesn’t mean that there’s nothing we can do. We’ve been in contact with people who have been working on the social aspect of the solution and together, we suggest that Google India and Google Inc:
Use ‘He/She/They’ instead of one pronoun separately for languages with no gendered pronouns.
Start using gender-balanced data to train Google Translate models for gender-neutral languages, retrain translation models for languages with ungendered pronouns.
Work with organizations working on gender-based issues to continuously review the trained models.
[1] A Beauty Spot of Bangla Language- Daily Star
[2] Build for Everyone- Google Diversity
[3] Google CEO Sundar Pichai revealed a jaw-dropping fact about its translation app: Business Insider India
Bangla is a gender-neutral language [1]
Spoken by more than 228 million people, it is a language that does not use gendered pronouns and verbs. However, when translation technology is developed using publicly sourced biased datasets, we lose this linguistic equality to gender-biased results.
Let us look at a few examples. The Bengali sentence তিনি অধ্যাপক (Tini adhyāpaka ) can be translated to English as both “She is a professor” as well as “He is a professor.” But Google translates it to “He is a professor” every single time.
Similarly, সে একটি শিশুকে খাওয়াচ্ছে (Sē ēkaṭi śiśukē khā'ōẏācchē) can mean both “she is feeding a baby” and “he is feeding a baby” and yet, the Google translation for the sentence only shows “She is feeding a baby.” These results do not depend on the location or the gender of the user.
Translations like the ones above are a reflection of the existing gender roles in society men have jobs and earn money, while women cook and take care of children at home. These results also tell us how certain qualities like bravery are associated with men, while emotions like shame are associated with women. They reinforce the gender binary and create a bias in results that is not present in the language itself.
We are aware of the linguistic complexity involved in translations. But for the problems that we are already aware of, more inclusive solutions can be devised. Google boasts of its efforts to achieve diversity and equity in everything it does.[2] According to data from 2018, Google translates around 143 billion words per day[3], and so Google has a responsibility to all those who use its service to be free of sexism.
The technological solution to this problem can be highly complex for now, but that doesn’t mean that there’s nothing we can do. We’ve been in contact with people who have been working on the social aspect of the solution and together, we suggest that Google India and Google Inc:
Use ‘He/She/They’ instead of one pronoun separately for languages with no gendered pronouns.
Start using gender-balanced data to train Google Translate models for gender-neutral languages, retrain translation models for languages with ungendered pronouns.
Work with organizations working on gender-based issues to continuously review the trained models.
[1] A Beauty Spot of Bangla Language- Daily Star
[2] Build for Everyone- Google Diversity
[3] Google CEO Sundar Pichai revealed a jaw-dropping fact about its translation app: Business Insider India
Next step - spread the word
Gender bias in Google translate