# "Short is the Road that Leads from Fear to Hate": Fear Speech in Indian WhatsApp Groups

**Authors:** Punyajoy Saha, Binny Mathew, Kiran Garimella, Animesh Mukherjee  
**arXiv:** [2102.03870](https://arxiv.org/abs/2102.03870)  
**Date:** 2021-02-07  
**Summary page:** https://www.summarizepaper.com/en/paper/2102.03870v1/

## Key points

- WhatsApp is a popular and cost-effective tool for political campaigning in India
- However, it has become a breeding ground for harmful speech against protected groups and religious minorities
- Such messages instill fear among the population about specific minority communities, which could lead to real-world violence
- Researchers conducted the first large-scale study on fear speech across thousands of public WhatsApp groups discussing politics in India
- They curated a new dataset and characterized fear speech from this dataset, observing that users writing such messages use various events and symbols to create an illusion of fear about a target community
- Current state-of-the-art natural language processing (NLP) models do not perform well at detecting fear speech messages due to their low toxic nature
- Fear speech messages tend to spread faster and may go undetected by classifiers built to detect traditional toxic speech
- Using a novel methodology involving Facebook ads, the researchers conducted a survey among users of these WhatsApp groups to understand the types of users who consume and share fear speech
- This work opens up new research questions that differ from tackling hate speech traditionally addressed by the research community; thus providing an opportunity for further exploration into how best to tackle this issue.

## Summary

The popularity of WhatsApp as a messaging app has made it a powerful and cost-effective tool for political campaigning, particularly during the 2019 Indian general election. However, reports have emerged that WhatsApp has also become a breeding ground for harmful speech against various protected groups and religious minorities. Such messages often attempt to instill fear among the population about specific minority communities, which could lead to real-world violence according to research on inter-group conflict. In response, researchers conducted the first large-scale study on fear speech across thousands of public WhatsApp groups discussing politics in India. They curated a new dataset and characterized fear speech from this dataset, observing that users writing such messages use various events and symbols to create an illusion of fear about a target community. The researchers built models to classify fear speech but found that current state-of-the-art natural language processing (NLP) models do not perform well at this task. Fear speech messages tend to spread faster and may go undetected by classifiers built to detect traditional toxic speech due to their low toxic nature. Using a novel methodology involving Facebook ads, the researchers conducted a survey among users of these WhatsApp groups to understand the types of users who consume and share fear speech. This work opens up new research questions that differ from tackling hate speech traditionally addressed by the research community; thus providing an opportunity for further exploration into how best to tackle this issue.