搜索结果：Woman

When we train models on biased datasets, they not only reproduce data biases, but can worsen them at test time - a phenomenon called bias amplification. Many of the current bias amplification metrics (e.g., BA (MALS), DPA) measure bias amplification only in classification datasets. These metrics are ineffective for image captioning datasets, as they cannot capture the language semantics of a caption. Recent work introduced Leakage in Captioning (LIC), a language-aware bias amplification metric that understands caption semantics. However, LIC has a crucial limitation: it cannot identify the source of bias amplification in captioning models. We propose Directional Bias Amplification in Captioning (DBAC), a language-aware and directional metric that can identify when captioning models amplify biases. DBAC has two more improvements over LIC: (1) it is less sensitive to sentence encoders (a hyperparameter in language-aware metrics), and (2) it provides a more accurate estimate of bias amplification in captions. Our experiments on gender and race attributes in the COCO captions dataset show that DBAC is the only reliable metric to measure bias amplification in captions.

As the deployment of large language models (LLMs) expands, there is an increasing demand for personalized LLMs. One method to personalize and guide the outputs of these models is by assigning a persona -- a role that describes the expected behavior of the LLM (e.g., a man, a woman, an engineer). This study investigates whether an LLM's understanding of social norms varies across assigned personas. Ideally, the perception of a social norm should remain consistent regardless of the persona, since acceptability of a social norm should be determined by the region the norm originates from, rather than by individual characteristics such as gender, body size, or race. A norm is universal within its cultural context. In our research, we tested 36 distinct personas from 12 sociodemographic categories (e.g., age, gender, beauty) across four different LLMs. We find that LLMs' cultural norm interpretation varies based on the persona used and the norm interpretation also varies within a sociodemographic category (e.g., a fat person and a thin person as in physical appearance group) where an LLM with the more socially desirable persona (e.g., a thin person) interprets social norms more accuratel

Data on historical populations often extends no further than numbers of people by broad age-sex group, with nothing on numbers of births or deaths. Demographers studying these populations have experimented with methods that use the data on numbers of people to infer birth and death rates. These methods have, however, received little attention since they were first developed in the 1960s. We revisit the problem of inferring demographic rates from population structure, spelling out the assumptions needed, and specialising the methods to the case where only child-woman ratios are available. We apply the methods to the case of Maori populations in nineteenth-century Aotearoa New Zealand. We find that, in this particular case, the methods reveal as much about the nature of the data as they do about historical demographic conditions.

This paper investigates the subtle and often concealed biases present in Large Language Models (LLMs), focusing on implicit biases that may remain despite passing explicit bias tests. Implicit biases are significant because they influence the decisions made by these systems, potentially perpetuating stereotypes and discrimination, even when LLMs appear to function fairly. Traditionally, explicit bias tests or embedding-based methods are employed to detect bias, but these approaches can overlook more nuanced, implicit forms of bias. To address this, we introduce two novel psychological-inspired methodologies: the LLM Implicit Association Test (IAT) Bias and the LLM Decision Bias, designed to reveal and measure implicit biases through prompt-based and decision-making tasks. Additionally, open-ended generation tasks with thematic analysis of word generations and storytelling provide qualitative insights into the model's behavior. Our findings demonstrate that the LLM IAT Bias correlates with traditional methods and more effectively predicts downstream behaviors, as measured by the LLM Decision Bias, offering a more comprehensive framework for detecting subtle biases in AI systems. Thi

No woman of science has lived a more controversial life nor possessed a most contrasting character than Gabrielle Emilie Le Tonnelier, Marquise du Chatelet. One one hand, she was a woman of great intelligence, a philosopher of science, a student of mathematics, and she was an ardent supporter of Newton and his new laws of physics. At the same time, Emilie du Chatelet was an aristocrat society woman who gambled, enjoyed parties, and had several extramarital affairs, provoking numerous scandals in her native Paris. She was a passionate woman who was at ease conversing with the nobles at the court and with the most renowned scholars of her time. Emilie du Chatelet did not develop theorems and she did not discover new scientific principles. However, she studied mathematics with Maupertuis and Clairaut to better understand the geometrical language in Newton's Principia. In this article, we review some important aspects of this controversial woman of science, exploring her relationship with the greatest scholars of her time.

In this paper, we propose a data-driven method to measure the impact of the 'woman card' exchange between Hillary Clinton and Donald Trump. Building from a unique dataset of the two candidates' Twitter followers, we first examine the transition dynamics of the two candidates' Twitter followers one week before the exchange and one week after. Then we train a convolutional neural network to classify the gender of the followers and unfollowers, and study how women in particular are reacting to the 'woman card' exchange. Our study suggests that the 'woman card' comment has made women more likely to follow Hillary Clinton, less likely to unfollow her and that it has apparently not affected the gender composition of Trump followers.

Wikipedia -- like most peer production communities -- suffers from a basic problem: the amount of work that needs to be done (articles to be created and improved) exceeds the available resources (editor effort). Recommender systems have been deployed to address this problem, but they have tended to recommend work tasks that match individuals' personal interests, ignoring more global community values. In English Wikipedia, discussion about Vital articles constitutes a proxy for community values about the types of articles that are most important, and should therefore be prioritized for improvement. We first analyzed these discussions, finding that an article's priority is considered a function of 1) its inherent importance and 2) its effects on Wikipedia's global composition. One important example of the second consideration is balance, including along the dimensions of gender and geography. We then conducted a quantitative analysis evaluating how four different article prioritization methods -- two from prior research -- would affect Wikipedia's overall balance on these two dimensions; we found significant differences among the methods. We discuss the implications of our results, i

Prior work shows that men and women speak with different levels of confidence, though it's often assumed that these differences are innate or are learned in early childhood. Using academic publishing as a setting, we find that language differences across male and female authors are initially negligible: in first drafts of academic manuscripts, men and women write with similar levels of uncertainty. However, when we trace those early drafts to their published versions, a substantial gender gap in linguistic uncertainty arises. That is, women increase their use of cautionary language through the publication process more than men. We show this increase in the linguistic gender gap varies substantially based on editor assignment. Specifically, our author-to-editor matched dataset allows us to estimate editor-specific fixed effects, capturing how specific editors impact the change in linguistic uncertainty for female authors relative to male authors (the editor's author-gender gap). Editors' author-gender gaps vary widely, and correlate with observable editor characteristics such as societal norms in their country-of-origin, their work history, and the year that they obtained their PhD.

Analogies such as "man is to king as woman is to X" are often used to illustrate the amazing power of word embeddings. Concurrently, they have also been used to expose how strongly human biases are encoded in vector spaces built on natural language, like "man is to computer programmer as woman is to homemaker". Recent work has shown that analogies are in fact not such a diagnostic for bias, and other methods have been proven to be more apt to the task. However, beside the intrinsic problems with the analogy task as a bias detection tool, in this paper we show that a series of issues related to how analogies have been implemented and used might have yielded a distorted picture of bias in word embeddings. Human biases are present in word embeddings and need to be addressed. Analogies, though, are probably not the right tool to do so. Also, the way they have been most often used has exacerbated some possibly non-existing biases and perhaps hid others. Because they are still widely popular, and some of them have become classics within and outside the NLP community, we deem it important to provide a series of clarifications that should put well-known, and potentially new cases into the

We present a systematic study of biases in natural language generation (NLG) by analyzing text generated from prompts that contain mentions of different demographic groups. In this work, we introduce the notion of the regard towards a demographic, use the varying levels of regard towards different demographics as a defining metric for bias in NLG, and analyze the extent to which sentiment scores are a relevant proxy metric for regard. To this end, we collect strategically-generated text from language models and manually annotate the text with both sentiment and regard scores. Additionally, we build an automatic regard classifier through transfer learning, so that we can analyze biases in unseen text. Together, these methods reveal the extent of the biased nature of language model generations. Our analysis provides a study of biases in NLG, bias metrics and correlated human judgments, and empirical evidence on the usefulness of our annotated dataset.

Word embeddings are the standard model for semantic and syntactic representations of words. Unfortunately, these models have been shown to exhibit undesirable word associations resulting from gender, racial, and religious biases. Existing post-processing methods for debiasing word embeddings are unable to mitigate gender bias hidden in the spatial arrangement of word vectors. In this paper, we propose RAN-Debias, a novel gender debiasing methodology which not only eliminates the bias present in a word vector but also alters the spatial distribution of its neighbouring vectors, achieving a bias-free setting while maintaining minimal semantic offset. We also propose a new bias evaluation metric - Gender-based Illicit Proximity Estimate (GIPE), which measures the extent of undue proximity in word vectors resulting from the presence of gender-based predilections. Experiments based on a suite of evaluation metrics show that RAN-Debias significantly outperforms the state-of-the-art in reducing proximity bias (GIPE) by at least 42.02%. It also reduces direct bias, adding minimal semantic disturbance, and achieves the best performance in a downstream application task (coreference resolutio