The Center for Digital Technology (CDT) has released a report titled "Moderating Maghrebi Arabic Content on Social Media," highlighting the challenges and implications of moderating content in Maghrebi Arabic dialects in North Africa. The report is part of a CDT series investigating biases and disparities in content moderation within the Global South.
Language plays a crucial role in expressing desires, emotions, and ideas. However, languages from the Global North, particularly English, dominate knowledge-sharing and technology spaces. This dominance has led to the marginalization of Global South languages in digital environments, especially regarding content moderation.
Content moderation involves mechanisms that govern participation in online communities and control visible content. These processes aim to remove harmful content to ensure positive user experiences, safeguard platform reputations, comply with legal requirements, and increase advertising revenues. Initially dependent on human moderators, the vast amount of online content necessitates automated systems. However, these systems often struggle with Global South languages due to inadequate training data and a lack of cultural understanding.
The CDT report uses interviews, focus groups, and online surveys to examine these issues specifically for Maghrebi Arabic dialects. It finds that most US-based social media companies use global strategies for content moderation, applying uniform policies worldwide. In contrast, TikTok adopts localized approaches tailored to each region's cultural matters.
Maghrebi Arabic users have developed tactics like "algospeak" to evade moderation algorithms because they believe they are censored for political reasons. They also resort to mass reporting mechanisms due to ineffective standard reporting tools.
The report highlights that the lack of diversity in natural language processing teams at social media companies affects the accuracy of automated moderation systems. Insufficient training datasets for Maghrebi Arabic dialects and recruiting non-native annotators contribute to this issue.
Content moderators face harsh working conditions and are assigned tasks from any Arab country despite varying cultural and linguistic nuances across regions. This can lead to errors in some instances.
For more detailed insights into these findings, readers are encouraged to access the full report.
