AlexaAI/TANGO

TANGO（Towards Centering Transgender and Non-Binary Voices to Measure Biases in Open Language Generation）是一个用于评估开放语言生成（OLG）模型中性别非肯定语言的数据集。它包含两个提示集：第一个提示集包含1,532,160个提示，用于测试模型对不同性别披露形式的反应；第二个提示集包含2,880个提示，用于评估生成文本中的误用性别代词情况。数据集的模板来源于非二元性别维基百科，并填充了各种指代形式和代词。该数据集旨在帮助评估模型在跨性别和非二元性别（TGNB）社区方面的偏见，促进模型构建的公平性和包容性，避免对TGNB社区造成伤害。

TANGO (Towards Centering Transgender and Non-Binary Voices to Measure Biases in Open Language Generation) is a dataset developed to evaluate gender non-affirming language in open language generation (OLG) models. It comprises two prompt sets: the first set contains 1,532,160 prompts intended to test model reactions to diverse forms of gender disclosure; the second set includes 2,880 prompts for assessing instances of misused gender pronouns within generated text. The templates of this dataset are sourced from Wikipedia entries on non-binary genders, and are filled with various reference forms and pronouns. This dataset is designed to facilitate the evaluation of model biases toward the transgender and non-binary (TGNB) community, advance fairness and inclusivity in model development, and prevent harm inflicted on the TGNB community.

原始信息汇总

TANGO数据集概述：

• 名称：TANGO
• 全称：Towards Centering Transgender and Non-Binary Voices to Measure Biases in Open Language Generation
• 目的：评估开放语言生成（OLG）中的性别非肯定性语言
• 组成：包含两组提示（prompts）