Queens are Powerful too: Mitigating Gender Bias in Dialogue Generation

TL;DR

This paper investigates gender bias in dialogue generation, analyzes its amplification in models, and proposes techniques like data augmentation and bias-controlled training to reduce bias while maintaining dialogue quality.

Contribution

It introduces effective bias mitigation methods for dialogue models, especially in highly biased datasets like LIGHT, and demonstrates their success through multiple evaluation metrics.

Findings

Bias mitigation techniques reduce gendered language in generated dialogues.

Combined methods are more effective than individual ones.

Models maintain engagement while generating less biased responses.

Abstract

Models often easily learn biases present in the training data, and their predictions directly reflect this bias. We analyze gender bias in dialogue data, and examine how this bias is actually amplified in subsequent generative chit-chat dialogue models. We measure gender bias in six existing dialogue datasets, and focus on the most biased one, the multi-player text-based fantasy adventure dataset LIGHT, as a testbed for our bias mitigation techniques. The LIGHT dataset is highly imbalanced with respect to gender, containing predominantly male characters, likely because it is entirely collected by crowdworkers and reflects common biases that exist in fantasy or medieval settings. We consider three techniques to mitigate gender bias: counterfactual data augmentation, targeted data collection, and bias controlled training. We show that our proposed techniques mitigate gender bias in LIGHT by balancing the genderedness of generated dialogue utterances and are particularly effective in combination. We quantify performance using various evaluation methods---such as quantity of gendered words, a dialogue safety classifier, and human studies---all of which show that our models generate less gendered, but equally engaging chit-chat responses.