These are not the Stereotypes You are Looking For: Bias and Fairness in Authorial Gender Attribution

By Corina Koolen, Andreas van Cranenburgh. 2017

45 citations recorded by the source corpus.

Stylometric and text categorization results show that author gender can be discerned in texts with relatively high accuracy. However, it is difficult to explain what gives rise to these results and there are many possible confounding factors, such as the domain, genre, and target audience of a text. More fundamentally, such classification efforts risk invoking stereotyping and essentialism. We explore this issue in two datasets of Dutch literary novels, using commonly used descriptive (LIWC, topic modeling) and predictive (machine learning) methods. Our results show the importance of controlling for variables in the corpus and we argue for taking care not to overgeneralize from the results.

Subfields: Artificial Intelligence, Gender Studies, 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

1. Yejin Choi
2. Rachel Rudinger
3. Benjamin Van Durme
4. Eduard H. Hovy
5. Yonatan Belinkov
6. Samuel R. Bowman
7. Mohit Bansal
8. Shrimai Prabhumoye

In Gender Studies

1. Lizhen Qu
2. Jarrod Knibbe
3. Rachel Rudinger
4. Zili Wang
5. Lucie Flekova
6. Qiongkai Xu
7. Yolande A. A. Strengers

In Natural Language Processing

1. Rachel Rudinger
2. Benjamin Van Durme
3. Yejin Choi
4. Eduard H. Hovy
5. Yonatan Belinkov
6. Samuel R. Bowman
7. Shrimai Prabhumoye
8. Mohit Bansal

In Safety, Robustness & Interpretability

1. Rachel Rudinger
2. Benjamin Van Durme
3. Eduard H. Hovy
4. Lizhen Qu
5. Jarrod Knibbe
6. Brian Leonard
7. Zili Wang
8. Lucie Flekova

Citations in the corpus

• Gender Bias in Coreference Resolution (2018, background): “Both Larson (2017) and Koolen and van Cranenburgh (2017) offer guidelines to NLP researchers and computational social scientists who wish to predict gender as a variable.”
• Returning the N to NLP: Towards Contextually Personalized Classification Models (2020, background): “However, albeit more accurate thanks to capturing stylistic variation elements, statistical models are prone to stereo-type propagation as well (Costa-juss`a et al., 2019; Koolen and van Cranenburgh, 2017).”
• Adhering, Steering, and Queering: Treatment of Gender in Natural Language Generation (2020, background): “Similar results are echoed in literary fic-tion [36], demonstrating that these are not differences between lay and expert writers.”
• Five sources of bias in natural language processing (2021, background): “…by demographic biases, such as unequal performance for different user groups (Tatman, 2017), mis-identification of speakers and their needs (Criado Perez, 2019) or the proliferation of harmful stereotypes (e.g., Agarwal et al., 2019; Koolen & van Cranenburgh, 2017; Kiritchenko & Moham-mad, 2018).”
• MSnet: A BERT-based Network for Gendered Pronoun Resolution (2019, methodology): “Inspired by Lee et al. (2017), I adopt the hidden states of transformers of the pre-trained BERT as the contextual representation.”