Authors: , Kezheng Xiang, Nikhil Deb and Mohammad Ruhul Amin
Abstract: Newspapers are considered to be the mirrors reflecting what is happening within different societies. Thus, analyzing online newspaper contents across languages can help us to understand the language-dependent implicit biases, i.e. gender bias at a global scale. Implicit biases occur when someone consciously rejects stereotypes and supports anti-discrimination efforts but also holds negative associations in his/her mind unconsciously. Gender bias can be quantified reliably by using the contextual representations of words, i.e. word embeddings, created from massive corpora. In this research, we present a comparative study of gender bias in the published news from conservative (Arabic peninsula, and Pakistan), semi or lower-conservative (Bangladesh, Indonesia, West Bengal, and India) and western (USA, Canada and UK) countries. We collected newspapers data from the online sources for the regions of our interest, such as Urdu, Arabic, Indonesian, Indian Bengali, Hindi, English and Bangla. As a result of the study, we present the current scenario of gender bias in different occupations in the above-mentioned countries. We show that Urdu language embedding has the minimum correlation with English language and highest correlation with Arabic language, showing that despite having lower gender gap, stereotypical gender biases in Pakistan is similar to the Arab societies. Interestingly, despite being the largest Muslim nation, Indonesian language has a higher correlation with English language than that of Urdu or Arabic language. On the other hand, Bangla language still has higher correlation to the diverse languages of surrounding societies such as Hindi and Indian Bengali language. We evaluate the results of computational methods in the light of recent literature discussing gender biases in the regions of our interest. Finally, we present a list of occupations in which gender bias is more prevalent in those countries. We believe that this study will help to better understand the new development efforts we need to eradicate inequalities from the current world.