From Detection to Counterspeech: Auditing AI Moderation and Fact-Checking Practices in Ethiopia’s Multilingual Online Sphere
Published in Media and Communication, 2026
This article examines how AI-assisted hate-speech detection operates in Ethiopia’s multilingual online sphere, with particular attention to Amharic and Afan Oromo. It combines a computational audit of three classifiers against 838 hand-annotated posts from 2020–2025, analysis of platform self-descriptions and transparency reports, and 20 interviews with Ethiopian fact-checkers, volunteer flaggers, and counter-speakers.
The study finds that the most widely used generic classifier cannot process either language natively and, when run on English translations, recovers only about a tenth of hate speech. Locally oriented classifiers perform substantially better on Amharic but fail sharply on Afan Oromo. Across the tools examined, under-detection is the dominant error.
The article argues that these technical failures do not eliminate human moderation labor; they displace it onto an unpaid civic ecology of fact-checkers, volunteer flaggers, and counter-speakers who absorb political and psychological costs that platforms benefit from but rarely acknowledge.
Recommended citation: Chala, Endalkachew H. (2026). "From Detection to Counterspeech: Auditing AI Moderation and Fact-Checking Practices in Ethiopia’s Multilingual Online Sphere." Media and Communication, 14. https://doi.org/10.17645/mac.12653
Download Paper