Language model debiasing has emerged as an important field of study in the NLP community. Numerous debiasing techniques were proposed, but bias ablation remains an unaddressed issue. We demonstrate a novel framework for inspecting bias in pre-trained transformer-based language models via movement pruning. Given a model and a debiasing objective, our framework finds a subset of the model containing less bias than the original model. We implement our framework by pruning the model while fine-tuning it on the debiasing objective. Optimized are only the pruning scores - parameters coupled with the model's weights that act as gates. We experiment with pruning attention heads, an important building block of transformers: we prune square blocks, as well as establish a new way of pruning the entire heads. Lastly, we demonstrate the usage of our framework using gender bias, and based on our findings, we propose an improvement to an existing debiasing method. Additionally, we re-discover a bias-performance trade-off: the better the model performs, the more bias it contains.
Masked language models pick up gender biases during pre-training. Such biases are usually attributed to a certain model architecture and its pre-training corpora, with the implicit assumption that other variations in the pre-training process, such as the choices of the random seed or the stopping point, have no effect on the biases measured. However, we show that severe fluctuations exist at the fundamental level of individual templates, invalidating the assumption. Further against the intuition of how humans acquire biases, these fluctuations are not correlated with the certainty of the predi…
Neural machine translation has significantly pushed forward the quality of the field. However, there are remaining big issues with the output translations and one of them is fairness. Neural models are trained on large text corpora which contain biases and stereotypes. As a consequence, models inherit these social biases. Recent methods have shown results in reducing gender bias in other natural language processing tools such as word embeddings. We take advantage of the fact that word embeddings are used in neural machine translation to propose a method to equalize gender biases in neural mach…
Colombian Waitresses y Jueces canadienses: Gender and Country Biases in Occupation Recommendations from LLMs
Erasmus Mundus Master in Language and Communication Technologies（埃拉斯谟 Mundus 语言与沟通技术硕士项目）；HiTZ Center - Ixa, University of the Basque Country (UPV/EHU)（巴斯克国家大学（UPV/EHU）HiTZ中心 - Ixa）；Cohere（Cohere公司）；Saarland University（萨尔兰州大学）
Department of Computer Science, Bar-Ilan University（巴伊兰大学计算机科学系）；Allen Institute for Artificial Intelligence（艾伦人工智能研究所）
Word embeddings are widely used in NLP for a vast range of tasks. It was shown that word embeddings derived from text corpora reflect gender biases in society. This phenomenon is pervasive and consistent across different word embedding models, causing serious concern. Several recent works tackle this problem, and propose methods for significantly reducing this gender bias in word embeddings, demonstrating convincing results. However, we argue that this removal is superficial. While the bias is indeed substantially reduced according to the provided bias definition, the actual effect is mostly h…
UNC Chapel Hill（北卡罗来纳大学教堂山分校）；University of California, Santa Cruz（加利福尼亚大学圣克鲁兹分校）；University of British Columbia（不列颠哥伦比亚大学）
Pre-trained language models learn socially harmful biases from their training corpora, and may repeat these biases when used for generation. We study gender biases associated with the protagonist in model-generated stories. Such biases may be expressed either explicitly ("women can't park") or implicitly (e.g. an unsolicited male character guides her into a parking space). We focus on implicit biases, and use a commonsense reasoning engine to uncover them. Specifically, we infer and analyze the protagonist's motivations, attributes, mental states, and implications on others. Our findings regar…
Automatic machine learning systems can inadvertently accentuate and perpetuate inappropriate human biases. Past work on examining inappropriate biases has largely focused on just individual systems. Further, there is no benchmark dataset for examining inappropriate biases in systems. Here for the first time, we present the Equity Evaluation Corpus (EEC), which consists of 8,640 English sentences carefully chosen to tease out biases towards certain races and genders. We use the dataset to examine 219 automatic sentiment analysis systems that took part in a recent shared task, SemEval-2018 Task …
University of Maryland（马里兰大学）；Microsoft Research（微软研究院）
Correctly resolving textual mentions of people fundamentally entails making inferences about those people. Such inferences raise the risk of systemic biases in coreference resolution systems, including biases that can harm binary and non-binary trans and cis stakeholders. To better understand such biases, we foreground nuanced conceptualizations of gender from sociology and sociolinguistics, and develop two new datasets for interrogating bias in crowd annotations and in existing coreference resolution systems. Through these studies, conducted on English text, we confirm that without acknowledg…
The scientific community is increasingly aware of the necessity to embrace pluralism and consistently represent major and minor social groups. Currently, there are no standard evaluation techniques for different types of biases. Accordingly, there is an urgent need to provide evaluation sets and protocols to measure existing biases in our automatic systems. Evaluating the biases should be an essential step towards mitigating them in the systems. This paper introduces WinoST, a new freely available challenge set for evaluating gender bias in speech translation. WinoST is the speech version of W…
Facebook AI Research（Facebook人工智能研究院）；Laboratoire Lorrain d’Informatique et Applications (LORIA)（洛林计算机科学及其应用实验室）
Models often easily learn biases present in the training data, and their predictions directly reflect this bias. We analyze gender bias in dialogue data, and examine how this bias is actually amplified in subsequent generative chit-chat dialogue models. We measure gender bias in six existing dialogue datasets, and focus on the most biased one, the multi-player text-based fantasy adventure dataset LIGHT, as a testbed for our bias mitigation techniques. The LIGHT dataset is highly imbalanced with respect to gender, containing predominantly male characters, likely because it is entirely collected…
Human gender bias is reflected in language and text production. Because state-of-the-art machine translation (MT) systems are trained on large corpora of text, mostly generated by humans, gender bias can also be found in MT. For instance when occupations are translated from a language like English, which mostly uses gender neutral words, to a language like German, which mostly uses a feminine and a masculine version for an occupation, a decision must be made by the MT System. Recent research showed that MT systems are biased towards stereotypical translation of occupations. In 2019 the first, …
Multiple studies have demonstrated that behavior on internet-based social media platforms can be indicative of an individual's mental health status. The widespread availability of such data has spurred interest in mental health research from a computational lens. While previous research has raised concerns about possible biases in models produced from this data, no study has quantified how these biases actually manifest themselves with respect to different demographic groups, such as gender and racial/ethnic groups. Here, we analyze the fairness of depression classifiers trained on Twitter dat…
Despite their prevalence in society, social biases are difficult to identify, primarily because human judgements in this domain can be unreliable. We take an unsupervised approach to identifying gender bias against women at a comment level and present a model that can surface text likely to contain bias. Our main challenge is forcing the model to focus on signs of implicit bias, rather than other artifacts in the data. Thus, our methodology involves reducing the influence of confounds through propensity matching and adversarial learning. Our analysis shows how biased comments directed towards …
Tokyo Institute of Technology（东京工业大学）；University of Liverpool（利物浦大学）；Amazon（亚马逊）
Different methods have been proposed to develop meta-embeddings from a given set of source embeddings. However, the source embeddings can contain unfair gender-related biases, and how these influence the meta-embeddings has not been studied yet. We study the gender bias in meta-embeddings created under three different settings: (1) meta-embedding multiple sources without performing any debiasing (Multi-Source No-Debiasing), (2) meta-embedding multiple sources debiased by a single method (Multi-Source Single-Debiasing), and (3) meta-embedding a single source debiased by different methods (Singl…
Wharton School, University of Pennsylvania（宾夕法尼亚大学沃顿商学院）
Cultural items like songs have an important impact in creating and reinforcing stereotypes, biases, and discrimination. But the actual nature of such items is often less transparent. Take songs, for example. Are lyrics biased against women? And how have any such biases changed over time? Natural language processing of a quarter of a million songs over 50 years quantifies misogyny. Women are less likely to be associated with desirable traits (i.e., competence), and while this bias has decreased, it persists. Ancillary analyses further suggest that song lyrics may help drive shifts in societal s…
Word embedding has become essential for natural language processing as it boosts empirical performances of various tasks. However, recent research discovers that gender bias is incorporated in neural word embeddings, and downstream tasks that rely on these biased word vectors also produce gender-biased results. While some word-embedding gender-debiasing methods have been developed, these methods mainly focus on reducing gender bias associated with gender direction and fail to reduce the gender bias presented in word embedding relations. In this paper, we design a causal and simple approach for…
Recent research has demonstrated that vector space models of semantics can reflect undesirable biases in human culture. Our investigation of crosslinguistic word embeddings reveals that topical gender bias interacts with, and is surpassed in magnitude by, the effect of grammatical gender associations, and both may be attenuated by corpus lemmatization. This finding has implications for downstream applications such as machine translation.
GOSt-MT: A Knowledge Graph for Occupation-related Gender Biases in Machine Translation
National Technical University of Athens（雅典国立技术大学）；Panteion University of Social and Political Sciences（潘泰翁社会政治大学）；National and Kapodistrian University of Athens（雅典国立卡波迪斯特里安大学）
Tokyo Metropolitan University（东京都立大学）；University of Liverpool（利物浦大学）
Word embeddings learnt from massive text collections have demonstrated significant levels of discriminative biases such as gender, racial or ethnic biases, which in turn bias the down-stream NLP applications that use those word embeddings. Taking gender-bias as a working example, we propose a debiasing method that preserves non-discriminative gender-related information, while removing stereotypical discriminative gender biases from pre-trained word embeddings. Specifically, we consider four types of information: \emph{feminine}, \emph{masculine}, \emph{gender-neutral} and \emph{stereotypical},…
CENIDET（墨西哥国家数字教育技术中心）；COLMEX（墨西哥学院）；IIM（印度管理研究所）；Université de Montréal（蒙特利尔大学）；EBSI（图书馆与信息科学学院）；Université d’Avignon（阿维尼翁大学）；LIA（阿维尼翁计算机科学实验室）
This work aims to evaluate the ability that both probabilistic and state-of-the-art vector space modeling (VSM) methods provide to well known machine learning algorithms to identify social network documents to be classified as aggressive, gender biased or communally charged. To this end, an exploratory stage was performed first in order to find relevant settings to test, i.e. by using training and development samples, we trained multiple algorithms using multiple vector space modeling and probabilistic methods and discarded the less informative configurations. These systems were submitted to t…
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…
University of Southern California（南加州大学）；Carnegie Mellon University（卡内基梅隆大学）
Numerous works have analyzed biases in vision and pre-trained language models individually - however, less attention has been paid to how these biases interact in multimodal settings. This work extends text-based bias analysis methods to investigate multimodal language models, and analyzes intra- and inter-modality associations and biases learned by these models. Specifically, we demonstrate that VL-BERT (Su et al., 2020) exhibits gender biases, often preferring to reinforce a stereotype over faithfully describing the visual scene. We demonstrate these findings on a controlled case-study and e…
The rise of concern around Natural Language Processing (NLP) technologies containing and perpetuating social biases has led to a rich and rapidly growing area of research. Gender bias is one of the central biases being analyzed, but to date there is no comprehensive analysis of how "gender" is theorized in the field. We survey nearly 200 articles concerning gender bias in NLP to discover how the field conceptualizes gender both explicitly (e.g. through definitions of terms) and implicitly (e.g. through how gender is operationalized in practice). In order to get a better idea of emerging trajec…
Centre for Artificial Intelligence Research (CAiRE)（人工智能研究中心（CAiRE））；Hong Kong University of Science and Technology（香港科技大学）
Abusive language detection models tend to have a problem of being biased toward identity words of a certain group of people because of imbalanced training datasets. For example, "You are a good woman" was considered "sexist" when trained on an existing dataset. Such model bias is an obstacle for models to be robust enough for practical use. In this work, we measure gender biases on models trained with different abusive language datasets, while analyzing the effect of different pre-trained word embeddings and model architectures. We also experiment with three bias mitigation methods: (1) debias…
Wangxuan Institute of Computer Technology, Peking University（北京大学王选计算机研究所）；The MOE Key Laboratory of Computational Linguistics, Peking University（北京大学计算语言学教育部重点实验室）
Many studies have revealed that word embeddings, language models, and models for specific downstream tasks in NLP are prone to social biases, especially gender bias. Recently these techniques have been gradually applied to automatic evaluation metrics for text generation. In the paper, we propose an evaluation method based on Word Embeddings Association Test (WEAT) and Sentence Embeddings Association Test (SEAT) to quantify social biases in evaluation metrics and discover that social biases are also widely present in some model-based automatic evaluation metrics. Moreover, we construct gender-…
All AI models are susceptible to learning biases in data that they are trained on. For generative dialogue models, being trained on real human conversations containing unbalanced gender and race/ethnicity references can lead to models that display learned biases, which we define here broadly as any measurable differences in the distributions of words or semantic content of conversations based on demographic groups. We measure the strength of such biases by producing artificial conversations between two copies of a dialogue model, conditioning one conversational partner to state a name commonly…
