Fatma Elsafoury

Fatma Elsafoury

This paper is a summary of the work done in my PhD thesis. Where I investigate the impact of bias in NLP models on the task of hate speech detection from three perspectives: explainability, offensive stereotyping bias, and fairness. Then, I discuss the main takeaways from my thesis and how they can benefit the broader NLP community. Finally, I discuss important future research directions. The findings of my thesis suggest that the bias in NLP models impacts the task of hate speech detection from all three perspectives. And that unless we start incorporating social sciences in studying bias in NLP models, we will not effectively overcome the current limitations of measuring and mitigating bias in NLP models.

Fatma Elsafoury

In this paper, we propose a new metric to measure the SOS bias in language models (LMs). Then, we validate the SOS bias and investigate the effectiveness of removing it. Finally, we investigate the impact of the SOS bias in LMs on their performance and fairness on hate speech detection. Our results suggest that all the inspected LMs are SOS biased. And that the SOS bias is reflective of the online hate experienced by marginalized identities. The results indicate that using debias methods from the literature worsens the SOS bias in LMs for some sensitive attributes and improves it for others. Finally, Our results suggest that the SOS bias in the inspected LMs has an impact on their fairness of hate speech detection. However, there is no strong evidence that the SOS bias has an impact on the performance of hate speech detection.

Fatma Elsafoury, David Hartmann

We know that language models (LMs) form biases and stereotypes of minorities, leading to unfair treatments of members of these groups, thanks to research mainly in the US and the broader English-speaking world. As the negative behavior of these models has severe consequences for society and individuals, industry and academia are actively developing methods to reduce the bias in LMs. However, there are many under-represented groups and languages that have been overlooked so far. This includes marginalized groups that are specific to individual countries and regions in the English speaking and Western world, but crucially also almost all marginalized groups in the rest of the world. The UN estimates, that between 600 million to 1.2 billion people worldwide are members of marginalized groups and in need for special protection. If we want to develop inclusive LMs that work for everyone, we have to broaden our understanding to include overlooked marginalized groups and low-resource languages and dialects. In this work, we contribute to this effort with the first study investigating offensive stereotyping bias in 23 LMs for 270 marginalized groups from Egypt, the remaining 21 Arab countries, Germany, the UK, and the US. Additionally, we investigate the impact of low-resource languages and dialects on the study of bias in LMs, demonstrating the limitations of current bias metrics, as we measure significantly higher bias when using the Egyptian Arabic dialect versus Modern Standard Arabic. Our results show, LMs indeed show higher bias against many marginalized groups in comparison to dominant groups. However, this is not the case for Arabic LMs, where the bias is high against both marginalized and dominant groups in relation to religion and ethnicity. Our results also show higher intersectional bias against Non-binary, LGBTQIA+ and Black women.

Fatma Elsafoury, Stamos Katsigiannis

Language models are the new state-of-the-art natural language processing (NLP) models and they are being increasingly used in many NLP tasks. Even though there is evidence that language models are biased, the impact of that bias on the fairness of downstream NLP tasks is still understudied. Furthermore, despite that numerous debiasing methods have been proposed in the literature, the impact of bias removal methods on the fairness of NLP tasks is also understudied. In this work, we investigate three different sources of bias in NLP models, i.e. representation bias, selection bias and overamplification bias, and examine how they impact the fairness of the downstream task of toxicity detection. Moreover, we investigate the impact of removing these biases using different bias removal techniques on the fairness of toxicity detection. Results show strong evidence that downstream sources of bias, especially overamplification bias, are the most impactful types of bias on the fairness of the task of toxicity detection. We also found strong evidence that removing overamplification bias by fine-tuning the language models on a dataset with balanced contextual representations and ratios of positive examples between different identity groups can improve the fairness of the task of toxicity detection. Finally, we build on our findings and introduce a list of guidelines to ensure the fairness of the task of toxicity detection.

Fatma Elsafoury, Gavin Abercrombie

In this paper, we trace the biases in current natural language processing (NLP) models back to their origins in racism, sexism, and homophobia over the last 500 years. We review literature from critical race theory, gender studies, data ethics, and digital humanities studies, and summarize the origins of bias in NLP models from these social science perspective. We show how the causes of the biases in the NLP pipeline are rooted in social issues. Finally, we argue that the only way to fix the bias and unfairness in NLP is by addressing the social problems that caused them in the first place and by incorporating social sciences and social scientists in efforts to mitigate bias in NLP models. We provide actionable recommendations for the NLP research community to do so.