As Large Language Models (LLMs) are increasingly deployed to handle various natural language processing (NLP) tasks, concerns regarding the potential negative societal impacts of LLM-generated content have also arisen. To evaluate the biases exhibited by LLMs, researchers have recently proposed a variety of datasets. However, existing bias evaluation efforts often focus on only a particular type of bias and employ inconsistent evaluation metrics, leading to difficulties in comparison across different datasets and LLMs. To address these limitations, we collect a variety of datasets designed for the bias evaluation of LLMs, and further propose CEB, a Compositional Evaluation Benchmark with 11,004 samples that cover different types of bias across different social groups and tasks. The curation of CEB is based on our newly proposed compositional taxonomy, which characterizes each dataset from three dimensions: bias types, social groups, and tasks. By combining the three dimensions, we develop a comprehensive evaluation strategy for the bias in LLMs. Our experiments demonstrate that the levels of bias vary across these dimensions, thereby providing guidance for the development of specific bias mitigation methods.

Recent studies show that Natural Language Processing (NLP) technologies propagate societal biases about demographic groups associated with attributes such as gender, race, and nationality. To create interventions and mitigate these biases and associated harms, it is vital to be able to detect and measure such biases. While existing works propose bias evaluation and mitigation methods for various tasks, there remains a need to cohesively understand the biases and the specific harms they measure, and how different measures compare with each other. To address this gap, this work presents a practical framework of harms and a series of questions that practitioners can answer to guide the development of bias measures. As a validation of our framework and documentation questions, we also present several case studies of how existing bias measures in NLP -- both intrinsic measures of bias in representations and extrinsic measures of bias of downstream applications -- can be aligned with different harms and how our proposed documentation questions facilitates more holistic understanding of what bias measures are measuring.

"Mia" is the virtual assistant introduced by Australian digital bank Ubank in February 2019. These aren't the members of a new, all-female rock group, but names that several large banks have been giving to their automated digital assistants. So-called chatbots have become a useful cost-cutting tools for companies with large subscriber bases (think banks, insurance firms and mobile phone operators). As they replace human call-center workers, such bots will help save banks an estimated $7.3 billion in operational costs by 2023, Juniper Research predicts. But the proliferation of bots with female names raises questions about whether they might also perpetuate gender stereotypes, particularly around the notion of women in the role of assistants.

Understanding an emotion isn't as simple as noticing a smile-- but we still look to facial movements for everything from navigating everyday social interactions to the development of emotionally attuned artificial intelligence. According to a July 2019 study from researchers at Northeastern and the California Institute of Technology, facial expressions only reflect the surface of emotions: The culture, situation, and specific individual around a facial expression add nuance to the way a feeling is conveyed. For example, the researchers note that Olympic athletes who won medals only smiled when they knew they were being watched by an audience. While they were waiting behind the podium or facing away from people, they didn't smile (but were probably still happy). These results reinforce the idea that facial expressions aren't always reliable indicators of emotion.

While harms of allocation have been increasingly studied as part of the subfield of algorithmic fairness, harms of representation have received considerably less attention. In this paper, we formalize two notions of stereotyping and show how they manifest in later allocative harms within the machine learning pipeline. We also propose mitigation strategies and demonstrate their effectiveness on synthetic datasets.