SALT-NLP/silent_signals_detection

Silent Signals | Human-Annotated Evaluation set for Detection (Synthetic Detection Dataset)

A dataset of human-annotated dogwhistle use cases to evaluate models on their ability to detect, identify, and define dogwhitles. 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. Dogwhistling historically originated from United States politics, but in recent years has taken root in social media as a means of evading hate speech detection systems and maintaining plausible deniability.

We developed an approach for word-sense disambiguation of dogwhistles from standard speech using Large Language Models (LLMs), and leveraged this technique to create a dataset of 16,550 high-confidence coded examples of dogwhistles used in formal and informal communication. Silent Signals is the largest dataset of disambiguated dogwhistle usage, created for applications in hate speech detection, neology, and political science.

Please note, this dataset contains content that may be upsetting or offensive to some readers.

Published at ACL 2024!

📄 Paper Link - Silent Signals, Loud Impact: LLMs for Word-Sense Disambiguation of Coded Dog Whistles
🗂️ Disambiguated Dogwhistle Dataset - Silent-Signals on HuggingFace
👔 Formal Potential Instance Dataset - Potential Dogwhistles from Congress
🏄‍♀️ Informal Potential Instance Dataset - Potential Dogwhistles from Reddit
💻 Dataset webpage - Coming soon 🚀

Dataset Schema

Field Name	Type	Example	Description
idx	int	"4"	Index.
dog_whistle	str	"illegals"	Dogwhistle word or term.
dog_whistle_root	str	"illegal immigrant"	The root form of the dogwhistle, as there could be multiple variations.
ingroup	str	"anti-Latino"	The community that uses the dogwhistle.
definition	str	"Latino, especially Mexican, immigrants regardless of documentation."	Definition of the dogwhistle, sourced from the Allen AI Dogwhistle Glossary.
example	str	"In my State of Virginia, the governor put a stop to the independent audits that were finding thousands of illegals on the roll."	Text containing the dog whistle.
label	str	"coded"	Denotes if the term is used as a coded dogwhistle ("coded") or not ("non-code").

NOTE: The dogwhistles terms and definitions that enabled this research and data collection were sourced from the Allen AI Dogwhistle Glossary.

Citations

MLA

Julia Kruk, Michela Marchini, Rijul Magu, Caleb Ziems, David Muchlinski, and Diyi Yang. 2024. Silent Signals, Loud Impact: LLMs for Word-Sense Disambiguation of Coded Dogwhistles. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 12493–12509, Bangkok, Thailand. Association for Computational Linguistics.

Bibtex

@inproceedings{kruk-etal-2024-silent,
    title = "Silent Signals, Loud Impact: {LLM}s for Word-Sense Disambiguation of Coded Dog Whistles",
    author = "Kruk, Julia  and
      Marchini, Michela  and
      Magu, Rijul  and
      Ziems, Caleb  and
      Muchlinski, David  and
      Yang, Diyi",
    editor = "Ku, Lun-Wei  and
      Martins, Andre  and
      Srikumar, Vivek",
    booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = aug,
    year = "2024",
    address = "Bangkok, Thailand",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.acl-long.675",
    pages = "12493--12509",
    abstract = "A dog whistle is a form of coded communication that carries a secondary meaning to specific audiences and is often weaponized for racial and socioeconomic discrimination. Dog whistling historically originated from United States politics, but in recent years has taken root in social media as a means of evading hate speech detection systems and maintaining plausible deniability. In this paper, we present an approach for word-sense disambiguation of dog whistles from standard speech using Large Language Models (LLMs), and leverage this technique to create a dataset of 16,550 high-confidence coded examples of dog whistles used in formal and informal communication. Silent Signals is the largest dataset of disambiguated dog whistle usage, created for applications in hate speech detection, neology, and political science.",
}