Racial Disparity in Natural Language Processing: A Case Study of Social Media African-American English

By Su Lin Blodgett, Brendan T. O'Connor. 2017

166 citations recorded by the source corpus.

We highlight an important frontier in algorithmic fairness: disparity in the quality of natural language processing algorithms when applied to language from authors of different social groups. For example, current systems sometimes analyze the language of females and minorities more poorly than they do of whites and males. We conduct an empirical analysis of racial disparity in language identification for tweets written in African-American English, and discuss implications of disparity in NLP.

Subfields: Artificial Intelligence, Sociology and Political Science, Natural Language Processing, AI Ethics, Fairness & Society

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 Sociology and Political Science

In Natural Language Processing

In AI Ethics, Fairness & Society

Citations in the corpus

Racial Bias in Hate Speech and Abusive Language Detection Datasets (2019, methodology): “Following Blod-gett and O’Connor (2017), we call these datasets black-aligned and white-aligned tweets, reﬂecting 5 We use this threshold following Blodgett and O’Connor (2017) and after consulting with the lead author.”
Bias in Bios: A Case Study of Semantic Representation Bias in a High-Stakes Setting (2019, background): “In the context of language identification, researchers have also investigated racial bias, showing that African-American English is often misclassified as non-English [8].”
Perturbation Sensitivity Analysis to Detect Unintended Model Biases (2019, background): “Recent research has shown ample evidence that data-driven NLP models may inadvertently capture, reﬂect and sometimes amplify various social biases present in the language data they are trained on (Bolukbasi et al., 2016; Blodgett and O’Connor, 2017).”
Predictive Biases in Natural Language Processing Models: A Conceptual Framework and Overview (2019, background): “The same effect was reported for language identification difficulties for African-American Vernacular English (Blodgett and O’Connor, 2017; Jurgens et al., 2017).”
WILDS: A Benchmark of in-the-Wild Distribution Shifts (2020, neutral): “…models, but in large-scale commercial systems that are already widely deployed, e.g., in speech-to-text systems from Amazon, Apple, Google, IBM, and Microsoft (Tatman, 2017; Koenecke et al., 2020) or language identification systems from IBM, Microsoft, and Twitter (Blodgett and O’Connor, 2017).”
DoReMi: Optimizing Data Mixtures Speeds Up Language Model Pretraining (2023, background): “Large LMs have also been well-documented to have risks and biases [1; 33; 7; 6; 18].”
Algorithmic Bias in Education (2021, result): “This is in contrast to the relatively greater attention to learner dialect in research on algorithmic bias in domains other than education (Benzeghiba et al., 2007; Blodgett & O’Connor, 2017; Tatman, 2017).”
Exploring the Role of Grammar and Word Choice in Bias Toward African American English (AAE) in Hate Speech Classification (2022, background): “Prior work has noted biases towards AAE and towards African American names compared to European American names in sentiment analysis [16, 22] and in English language identification applications [2].”
Mitigation of Unintended Biases against Non-Native English Texts in Sentiment Analysis (2019, background): “This can lead to issues in automated speech recognition (ASR) [17], and to transfer of pronunciation into spelling [2].”
Twitter Universal Dependency Parsing for African-American and Mainstream American English (2018, background): “…trained on well-edited MAE text, their performance is degraded on Twitter, and even more so for AAE tweets compared to MAE tweets— gaps exist for part-of-speech tagging (Jørgensen et al., 2016), language identiﬁcation, and dependency parsing (Blodgett et al., 2016; Blodgett and O’Connor, 2017).”
Text and Causal Inference: A Review of Using Text to Remove Confounding from Causal Estimates (2020, background): “, 2015), and exhibit racial and gender disparity (Blodgett and O’Connor, 2017; Zhao et al., 2017).”
A Framework for the Computational Linguistic Analysis of Dehumanization (2020, background): “, 2018), language identification (Blodgett and O’Connor, 2017), and in many other areas (Sun et al.”
Quantifying Social Biases in NLP: A Generalization and Empirical Comparison of Extrinsic Fairness Metrics (2021, unknown): “, 2017), 11 (Blodgett and O’Connor, 2017; Bhaskaran and Bhallamudi, 2019), 12 (De-Arteaga et al.”
Fairness in Machine Learning: A Survey (2020, methodology): “AutoML abstracts much of the core methodological expertise (e.g., KDD [97] and CRISP-DM [60]) by automated feature extraction and training multiple models, often combining them into an ensemble of models that maximizes a set of performance measures.”
Dialect Diversity in Text Summarization on Twitter (2020, background): “We pre-process the dataset to ﬁlter and remove posts for which the probability of belonging to non-Hispanic African-American English language model or non-Hispanic White English language model is less than 0.99.”