Do Neural Language Models Overcome Reporting Bias?

By Vered Shwartz, Yejin Choi. 2020

67 citations recorded by the source corpus.

Mining commonsense knowledge from corpora suffers from reporting bias, over-representing the rare at the expense of the trivial (Gordon and Van Durme, 2013). We study to what extent pre-trained language models overcome this issue. We find that while their generalization capacity allows them to better estimate the plausibility of frequent but unspoken of actions, outcomes, and properties, they also tend to overestimate that of the very rare, amplifying the bias that already exists in their training corpus.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Large Language Models, Natural Language Processing

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. Hinrich Schütze
2. Yoav Goldberg
3. Mohit Bansal
4. Yonatan Belinkov
5. Graham Neubig
6. Eduard H. Hovy
7. Noah A. Smith
8. Yejin Choi

In Computer Vision and Pattern Recognition

1. Hinrich Schütze
2. Dacheng Tao
3. Mohit Bansal
4. Eduard H. Hovy
5. Luc Van Gool
6. Yoav Goldberg
7. Trevor Darrell
8. Xiaogang Wang

In Large Language Models

1. Hinrich Schütze
2. Yoav Goldberg
3. Cyprien de Masson d'Autume
4. Nora Kassner
5. Abhilasha Ravichander
6. Yanai Elazar
7. Shauli Ravfogel
8. Sang-Woo Lee

In Natural Language Processing

1. Yoav Goldberg
2. Hinrich Schütze
3. Eduard H. Hovy
4. Mohit Bansal
5. Graham Neubig
6. Yonatan Belinkov
7. Yejin Choi
8. Mirella Lapata

Citations in the corpus

• Measuring and Improving Consistency in Pretrained Language Models (2021, background): “, 2016), and overestimating events and actions (reporting bias) (Shwartz and Choi, 2020).”
• What Changes Can Large-scale Language Models Bring? Intensive Study on HyperCLOVA: Billions-scale Korean Generative Pretrained Transformers (2021, background): “Many studies have already reported that these biases from training data have significant influences on large-scale language models as well (Abid et al., 2021; Garrido-Muñoz et al., 2021; Shwartz and Choi, 2020).”
• A Systematic Investigation of Commonsense Understanding in Large Language Models (2021, background): “Previous work has probed for commonsense knowledge in large language models using knowledge base completion (Petroni et al., 2019; Davison et al., 2019) or manually designed probing tasks (Weir et al., 2020; Shwartz and Choi, 2020).”