Click the star to add/remove an item to/from your individual schedule.
You need to be logged in to avail of this functionality.

Accepted Paper:

Investigating content moderation systems for the Maghrebi Arabic  
Mona Elswah (Center for Democracy and Technology)

Short abstract:

Using qualitative methods, this study investigates content moderation biases for non-English languages in the Global South, focusing on the Maghrebi Arabic as a case study

Long abstract:

Efforts to develop accurate and fair content moderation systems have primarily focused on English-language content. Prior research found that Maghrebi Arabic (the dialect spoken in Tunisia, Algeria, and Morrocco) has been even more neglected compared to other Arabic dialects, which have already faced bias and discrimination (Elswah, 2024). Hence, we study how tech companies create mechanisms to moderate Maghrebi Arabic and how these impact users in the Maghreb region. We interview NLP researchers, AI developers, and content creators to scrutinise how content moderation operates across platforms in the Maghrebi Arabic case. Preliminary findings reveal several key points: a) Maghrebi Arabic speakers have evolved the language and adopted what they call “Arabizi”, writing Arabic words using Latin letters. This colonial effect on the language has imposed a new challenge on technology companies and AI developers. This Westernized version of the language has no standard form of writing and no rules, posing a new challenge for developing accurate AI models. B) technology companies are badly moderating content from languages in the Global South. This bad moderation includes over-moderation and under-moderation of content at the same time., c) researchers in this region invest a lot of effort in collecting and storing datasets that could be used to develop AI models and applications. Thus, it is important for large technology companies to involve local researchers from this region to improve the accuracy of their models.

Traditional Open Panel P224
Big data and artificial intelligence global asymmetries: infrastructures, skills, uses, value and side effects
  Session 2 Friday 19 July, 2024, -