AI Chat AI Image Generator AI Video Text to Speech

In-Depth Look at Word Filling Societal Bias Measures

02/24/2023

∙

by Matúš Pikuliak, et al.

∙

∙

Many measures of societal bias in language models have been proposed in recent years. A popular approach is to use a set of word filling prompts to evaluate the behavior of the language models. In this work, we analyze the validity of two such measures – StereoSet and CrowS-Pairs. We show that these measures produce unexpected and illogical results when appropriate control group samples are constructed. Based on this, we believe that they are problematic and using them in the future should be reconsidered. We propose a way forward with an improved testing protocol. Finally, we also introduce a new gender bias dataset for Slovak.

Matúš Pikuliak
6 publications
Ivana Beňová
1 publication
Viktor Bachratý
2 publications

research

∙ 05/30/2019

Reducing Gender Bias in Word-Level Language Models with a Gender-Equalizing Loss Function

Gender bias exists in natural language datasets which neural language mo...

0 Yusu Qian, et al. ∙

research

∙ 05/24/2023

Balancing the Picture: Debiasing Vision-Language Datasets with Synthetic Contrast Sets

Vision-language models are growing in popularity and public visibility t...

5 Brandon Smith, et al. ∙

research

∙ 07/18/2022

Selection Bias Induced Spurious Correlations in Large Language Models

In this work we show how large language models (LLMs) can learn statisti...

0 Emily McMilin, et al. ∙

research

∙ 05/22/2023

Keeping Up with the Language Models: Robustness-Bias Interplay in NLI Data and Models

Auditing unwanted social bias in language models (LMs) is inherently har...

6 Ioana Baldini, et al. ∙

research

∙ 03/22/2022

A Prompt Array Keeps the Bias Away: Debiasing Vision-Language Models with Adversarial Learning

Vision-language models can encode societal biases and stereotypes, but t...

2 Hugo Elias Berg, et al. ∙

research

∙ 12/20/2022

Trustworthy Social Bias Measurement

How do we design measures of social bias that we trust? While prior work...

0 Rishi Bommasani, et al. ∙

research

∙ 05/08/2023

ANALOGICAL - A New Benchmark for Analogy of Long Text for Large Language Models

Over the past decade, analogies, in the form of word-level analogies, ha...

2 Thilini Wijesiriwardene, et al. ∙