Towards Identifying Social Bias in Dialog Systems: Framework, Dataset, and Benchmark

• DOI: 10.18653/v1/2022.findings-emnlp.262
• Publication Date: 2023-08-04T20:21:02Z
• Authors: Jingyan Zhou; Jiawen Deng; Fei Mi; Yitong Li; Yasheng Wang; Minlie Huang; Xin Jiang; Qun Liu; Helen Meng
• Keywords: Benchmark (surveying); Offensive

Among all the safety concerns that hinder deployment of open-domain dialog systems (e.g., offensive languages, biases, and toxic behaviors), social bias presents an insidious challenge. Addressing this challenge requires rigorous analyses normative reasoning. In paper, we focus our investigation on measurement to facilitate development unbiased systems. We first propose a novel Dial-Bias Framework for analyzing in conversations using holistic method beyond lexicons or dichotomous annotations. Leveraging proposed framework, further introduce CDial-Bias Dataset which is, best knowledge, annotated Chinese dataset. also establish fine-grained benchmark conduct in-depth ablation studies shed light utility detailed annotations Finally, evaluate representative generative models with classifiers unveil presence these