Demoting Racial Bias in Hate Speech Detection

By Mengzhou Xia, Anjalie Field, Yulia Tsvetkov. 2020

140 citations recorded by the source corpus.

In the task of hate speech detection, there exists a high correlation between African American English (AAE) and annotators’ perceptions of toxicity in current datasets. This bias in annotated training data and the tendency of machine learning models to amplify it cause AAE text to often be mislabeled as abusive/offensive/hate speech (high false positive rate) by current hate speech classifiers. Here, we use adversarial training to mitigate this bias. Experimental results on one hate speech dataset and one AAE dataset suggest that our method is able to reduce the false positive rate for AAE text with only a minimal compromise on the performance of hate speech classification.

Subfields: Artificial Intelligence, Natural Language Processing, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Natural Language Processing

In Safety, Robustness & Interpretability

Citations in the corpus

Language Generation Models Can Cause Harm: So What Can We Do About It? An Actionable Survey (2022, background): “limitations or are known to exacerbate other kinds of harms (Xia et al., 2020), and thus, applying them to newer problems may lead to unseen issues.”
Anticipating Safety Issues in E2E Conversational AI: Framework and Tooling (2021, background): “For dialect bias, Sap et al. (2019) use race and dialect priming, while Xia et al. (2020) tackle the problem with adversarial training.”
Recipes for Safety in Open-domain Chatbots (2020, background): “019) showed that widely used hate-speech datasets contain correlations between surface markers of African American English and toxicity, and propose race and dialect priming as a way to mitigate this.Xia et al. (2020) tackle the same problem through adversarial training.Gencoglu(2020) proposes a cyb”
Guiding the Release of Safer E2E Conversational AI through Value Sensitive Design (2022, background): “Possible mitigations include using race and dialect priming (Sap et al., 2019), using adversarial training techniques (Xia et al., 2020), adding fairness constraints (Gencoglu, 2020), or relabeling data used during training (Zhou et al., 2021).”
Challenges in Automated Debiasing for Toxic Language Detection (2021, methodology): “We ﬁnd that dialectal biases are particularly challenging to address, as has also been shown by Xia et al. (2020).”
A Survey of Race, Racism, and Anti-Racism in NLP (2021, background): “…et al. (2016) does serve as a starting point and forms the basis of most current work assessing performance gaps in NLP models (Sap et al., 2019; Blodgett et al., 2018; Xia et al., 2020; Xu et al., 2019; Groenwold et al., 2020), but even this corpus is explicitly not intended to infer race.”
Fairness for Unobserved Characteristics: Insights from Technological Impacts on Queer Communities (2021, background): “Adversarial training can similarly assist by demoting the confounds associated with texts of marginalised groups [165].”
Latent Hatred: A Benchmark for Understanding Implicit Hate Speech (2021, background): “…models for coded language (Kambhatla et al., 2018; Qian et al., 2019), lifelong learning of hateful language (Qian et al., 2021), contextualized sarcasm detection, and bias mitigation for named entities in hate speech detection systems (Xia et al., 2020) and their connection with our dataset.”
ToxCCIn: Toxic Content Classification with Interpretability (2021, background): “For example, some slurs that frequently appear in African American English are usually picked as strong evidence of toxicity (Xia et al., 2020); these words are indeed innocuous and only confined within the black online community.”
Handling Bias in Toxic Speech Detection: A Survey (2022, methodology): “[98] used an adversarial training procedure [46] on the Founta100k dataset.”
Human-Machine Collaboration Approaches to Build a Dialogue Dataset for Hate Speech Countering (2022, background): “, 2021) and of problems such as bias replication (Binns et al., 2017; Davidson et al., 2019; Vidgen and Derczynski, 2020; Sap et al., 2019; Tsvetkov, 2020).”
Explaining Toxic Text via Knowledge Enhanced Text Generation (2022, background): “Considerations of algorithmic fairness should be taken into account (Corbett-Davies et al., 2017), as well as cultural differences (Oliva et al., 2020) and racial biases (Xia et al., 2020) which can lead to misclassifica-tions of offensiveness.”
Towards a Comprehensive Taxonomy and Large-Scale Annotated Corpus for Online Slur Usage (2020, background): “Furthermore, Davidson et al. (2019), Xia et al. (2020) and Sap et al. (2019) reveal how clas-siﬁers trained on data with systemic racial bias have a higher tendency to label text written in African-American English as abusive.”
On Transferability of Bias Mitigation Effects in Language Model Fine-Tuning (2021, methodology): “We implement UBM with two different bias mitigation algorithms in the upstream bias mitigation phase: explanation regularization (Kennedy et al., 2020), and adversarial de-biasing (Zhang et al., 2018; Madras et al., 2018; Xia et al., 2020 where α is a trade-off hyperparameter.”