Quantifying Gender Bias in Different Corpora

By Marzieh Babaeianjelodar, Stephen Lorenz, Joshua Gordon, Jeanna Neefe Matthews, Evan Freitag. 2020

50 citations recorded by the source corpus.

Word embedding models have been shown to be effective in performing a wide variety of Natural Language Processing (NLP) tasks such as identifying audiences for web advertisements, parsing resumés to select promising job candidates, and translating documents from one language to another. However, it has been demonstrated that NLP systems learn gender bias from the corpora of documents on which they are trained. It is increasingly common for pre-trained models to be used as a starting point for building applications in a wide range of areas including critical decision making applications. It is also very easy to use a pre-trained model as the basis for a new application without careful consideration of the original nature of the training set. In this paper, we quantify the degree to which gender bias differs with the corpora used for training. We look especially at the impact of starting with a pre-trained model and fine-tuning with additional data. Specifically, we calculate a measure of direct gender bias on several pre-trained models including BERT’s Wikipedia and Book corpus models as well as on several fine-tuned General Language Understanding Evaluation (GLUE) benchmarks. In addition, we evaluate the bias from several more extreme corpora including the Jigsaw identity toxic dataset that includes toxic speech biased against race, gender, religion, and disability and the RtGender dataset that includes speech specifically labelled by gender. Our results reveal that the direct gender bias of the Jigsaw toxic identity dataset is surprisingly close to that of the base pre-trained Google model, but the RtGender dataset has significantly higher direct gender bias than the base model. When the bias learned by an NLP system can vary significantly with the corpora used for training, it becomes important to consider and report these details, especially for use in critical decision-making applications.

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

Underspecification Presents Challenges for Credibility in Modern Machine Learning (2020, background): “Finally, we note that the particular axes along which we observe sensitivity depends on the dataset used to train the embeddings, and could vary on embeddings trained on other corpora (Babaeianjelodar et al., 2020).”
Towards Accountability for Machine Learning Datasets: Practices from Software Engineering and Infrastructure (2020, background): “ning data are sometimes used interchangeably, for example biases in models are frequently seemingly identified with training data bias, or used to abductively infer biases in the training data (e.g., [11, 38, 77]). Despite the critical role data plays, the systemic devaluation of data work is in pla”
Adapting Coreference Resolution for Processing Violent Death Narratives (2021, background): “Likewise, even when evaluating a model on a general set, Babaeianjelodar et al. (2020) showed that many general-domain datasets include as much bias as datasets designed to be toxic and biased.”
SqueezeBERT: What can computer vision teach NLP about efficient neural networks? (2020, methodology): “The gender-related biases of our pretraining corpora (Wikipedia and BooksCorpus) are investigated by Tan et al. [61], and the gender biases of BERT and GPT models ﬁnetuned on the GLUE tasks are investigated by Babaeianjelodar et al. [62].”