Detecting Hate Speech in Tamil Language using Natural Language Processing

A global team of 50 AI changemakers has been working to detect hate speech in Tamil language. In the span of 2 months, the team successfully leveraged the existing knowledge in the domain of hate speech and was able to deliver significant results. The collaborators used language modeling to power an API that can constantly segregate a sentence as a positive or a negative hate speech.

The partner for this challenge is Sri Lanka based social enterprise DreamSpace Academy (DSA). The project has been supported by the NYU Center on International Cooperation and the Netherlands Ministry of Foreign Affairs.

The problem areas

This project has focused on identifying the various types of hate speech, including community-based hate speech, religion-based hate speech, gender-based hate speech, and political hate speech.

The project outcomes 

As a result of this project, an AI-enabled system has been developed that generates graphs and statistics for lexicon reports on hate speech used in Sri Lanka in the Tamil Language. The deliverables comprise of an AI model for identifying hate speech in Tamil, a classification system for categorizing hate speech based on religion, gender, community, and politics, and the potential for retraining the model and utilizing insights gained from this initiative to design a similar tool for the Sinhala language.