Gender bias in machine translation: an analysis of Google Translate in English and Spanish
María López Medel · Academia Letters · 2021
More than seventy years after Warren Weaver's prediction, automatic translation has become a part of our daily lives, both personally and professionally, even for language specialists.Translation engines are used by millions of people every day, delivering the best results between grammatically-similar languages with larger multilingual corpora available, like the English-Spanish pair.Big translation departments also rely heavily on machine translation based on custom memories and databases, and translation processes have dramatically changed.For non-professional users, a service like Google Translate offers draft-quality texts that are free, almost unlimited and faster than human translations.But automation comes at a price, specially with regard to unwanted bias, gender-wise and other.Although Google has attempted on several occasions to remove gender bias from its free online translation service, it still tends to exhibit predominantly masculine options and shows a tendency towards perpetuating or exaggerating sexist stereotypes, which adds to other flaws like a failure to notice text formality, typos and nuances.The technology on which its machine translation system is based, that feeds from statistics and large multilingual corpora such as the UN and the European Parliament, is supposed to reflect the gender divide in society, including men's and women's professional quotas, but the results can show even greater inequalities than in real life.In 2018 Google introduced a development that showed gender-specific translations in a selection of languages for single words or even short phrases and sentences for fewer languages, like Spanish, but it was not devoid of a masculine default.To counteract it, the company launched an update in 2020 based on its neural translation technology that generates a default translation, rewrites results that are gendered and checks Academia