Meta, the U.S. technology platform company behind Facebook, Instagram, and Threads, may misinterpret linguistic and cultural context when moderating content after publication. At the University of Szeged, Dr. Réka Brigitta Szaniszló, Senior Assistant Professor at the Institute of International and Regional Studies within SZTE’s Faculty of Law and Political Sciences, is researching the sociolinguistic shortcomings of moderation on these platforms.
The Digitalization and Democracy Research Group, part of the Digital Society Competence Center within the University of Szeged’s Center of Excellence for Interdisciplinary Research, Development, and Innovation (IKIKK), conducts research in public law and political science, with a particular focus on the transition to a digital society and its broader implications. Within this framework, Dr. Réka Brigitta Szaniszló examines the human rights challenges arising from the sociolinguistic shortcomings of Meta’s content moderation.
Meta operates several major social media platforms, including Facebook, Instagram, and Threads. Although the term media may imply some form of editorial review before publication, most user-generated content is made publicly available without prior screening. In practice, moderation generally takes place only after publication, although certain types of content may be blocked automatically during the upload process. At that stage, a post may be removed following user reports or through a combination of automated detection systems and human review. This delay can have important consequences. “By the time moderation steps in, content may already have spread widely – and once something is online, it stays there,” Dr. Réka Brigitta Szaniszló noted.
To provide independent scrutiny of certain content moderation decisions, Meta established the Oversight Board. Against this backdrop, Dr. Szaniszló’s research examines a crucial question: when cases reach the Board, are differences in context, language, and culture given sufficient consideration? Answering this question, however, is complicated by the relatively small body of available decisions. Only a small fraction of the millions of appeals and reports received by Meta’s content moderation system each year ever reach the Oversight Board. The scale of this gap is striking: during its first years of operation, the Board issued just 147 decisions.
The researcher initially expected linguistic and cultural misunderstandings to occur primarily in low-resource languages – languages with fewer technological resources and smaller datasets available for training language models and moderation systems. For certain language technology applications, Hungarian qualifies as a low-resource language. The research, however, revealed a much broader problem: similar misunderstandings also occur in high-resource languages such as English and French. This suggests that the issue is not simply a lack of linguistic data, but a more systemic difficulty within Meta’s content moderation processes in interpreting language and culture in context.
Dr. Réka Brigitta Szaniszló also examined whether Meta is more likely to acknowledge moderation errors when content is removed as a result of linguistic or cultural misunderstandings. Her findings indicate that this is not necessarily the case. “Based on my research, the Oversight Board’s greatest limitation lies not in its ability to assess individual cases, but in its inability to ensure that policy recommendations are implemented. As a result, systemic problems in content moderation can be addressed only to a limited extent,” she emphasized.
A case reviewed by the Oversight Board brings this challenge into sharp focus. An Iranian user posted the Persian phrase “Marg bar Khamenei” (“Death to Khamenei!”), which Meta interpreted as hate speech and removed. The Oversight Board later concluded that the post could not be assessed accurately on the basis of a literal translation alone: its linguistic, cultural, and political context also had to be taken into account. For Dr. Szaniszló, the case illustrates a central lesson of the research – on global platforms, understanding what words mean also requires an understanding of the ’world’ in which they are used.
Original Hungarian article by Helga Balog
Feature photo: Dr. Réka Brigitta Szaniszló at the Digital World (DIVIL) Conference. Photo: Dávid Miskolci
