SALT-NLP/formal_potential_dogwhistles

该数据集名为Silent Signals | Formal Potential Dogwhistles (Potential Instance Dataset)，主要包含美国国会记录中可能使用狗哨术语的文本片段。狗哨术语是一种编码通信形式，通常用于种族和社会经济歧视。数据集通过关键词搜索生成，包含超过100万条记录，但由于资源限制，仅采样了10万条记录。数据集用于通过大型语言模型（LLMs）进行狗哨术语的语义消歧。

A dataset of potential dogwhistle use cases in formal discourse. A dogwhistle is a form of coded communication that carries a secondary meaning to specific audiences and is often weaponized for racial and socioeconomic discrimination. This dataset contains over 1 million excerpts from U.S. Congressional Records that contain a dogwhistle term, created via keyword search on terms in the Allen AI Dogwhistle Glossary. The dataset was used as input to our approach for word-sense disambiguation of dogwhistles from standard speech using Large Language Models (LLMs). Although we collect over 7 million potential dogwhistle instances, due to resource constraints, we only sample 100,000 instances for the creation of the Silent Signals dataset. Therefore, we release the Potential Dogwhistle Dataset to enable the open sourced expansion of the Silent Signals dataset.