fg-zettelkasten

Language Disparities in Moderation Workforce Allocation by Social Media Platforms

4 min read

Tonneau, M., Liu, D., McGrady, R., Zheng, K., Schroeder, R., Zuckerman, E., & Hale, S. (2025). Language Disparities in Moderation Workforce Allocation by Social Media Platforms. https://doi.org/10.31235/osf.io/amfws_v1

View paper

Summary

This paper offers the first comparative empirical analysis of how six major social media platforms — YouTube, Meta, TikTok, Twitter/X, Snapchat, and LinkedIn — allocate human content moderators across languages, leveraging transparency data mandated by the EU’s Digital Services Act. The authors document substantial cross-lingual disparities in both language coverage (whether any moderators exist) and intensity (moderators relative to content volume). They find that smaller platforms leave millions of EU users without moderation in their national language, and that languages predominantly spoken in the Global South — Arabic, Spanish, Portuguese — receive consistently fewer moderators per unit of content than English. The paper argues that current DSA reporting, while a pivotal advance, is insufficient for independent scrutiny, and calls for globally inclusive transparency requirements.

Key Contributions

  • First comparative empirical analysis of cross-lingual moderator workforce allocation across six major DSA-regulated platforms.
  • A novel methodology pairing DSA transparency disclosures with independently constructed, calibrated content-volume estimates to normalize moderator counts by language.
  • Quantification of the number of EU users left without national-language moderation on specific platforms.
  • Empirical evidence that extends and nuances earlier reports (e.g., Haugen disclosures, Global Witness) about underinvestment in non-English moderation.
  • Concrete policy recommendations: reporting content volume per language, moderator workload capacity, harmful-content prevalence, and extending transparency obligations beyond the EU.

Methods

The authors collected per-language moderator counts from DSA-mandated transparency reports (Summer 2023–Fall 2024) for the six platforms, averaging across reporting periods and treating non-reported EU languages as having zero moderators. To normalize counts by content volume, they built independent representative datasets — the TwitterDay corpus (375M tweets from a single day in 2022) and a random sample of ~26,000 YouTube videos. Language identification used fastText, with inference scores calibrated via isotonic regression against native-speaker annotations, and bootstrap resampling (1,000 samples) to estimate per-language counts with confidence intervals. Twitter/X user locations were mapped to countries via the Google Geocoding API, and DSA-reported EU user counts were used to estimate how many users lack moderators for their national language.

Findings

  • YouTube, Meta, and TikTok cover nearly all EU official languages; Twitter/X, LinkedIn, and Snapchat have multiple blind spots, especially in Southern, Eastern, and Northern Europe.
  • Roughly 16M EU Twitter/X users (14%), 8M LinkedIn users (16%), and 7M Snapchat users (7%) have no moderators for their national language.
  • Languages under Twitter/X blind spots represent, on average, 31% of tweets in countries where they are official.
  • On Twitter/X, only Bulgarian and German have more moderators than English relative to content volume; Italian and Bulgarian have similar counts despite Italian content being 78× more prevalent.
  • On Twitter/X, Portuguese, Arabic, and Spanish receive just 9%, 7%, and 7% of English’s per-content allocation respectively.
  • On YouTube, most covered EU languages receive proportionally more moderators than English — only Spanish and Portuguese receive less, signaling reduced investment in Latin American user bases.
  • Global South languages average 55% of English’s allocation on YouTube down to 7.5% on Twitter/X.
  • Interface language support and moderation staffing are mismatched: some UI-supported languages (Greek, Czech, Romanian) lack dedicated moderators, while some moderated languages (Tagalog) lack UI support.

Connections

This paper sits within the platform-governance-data-access literature that treats DSA transparency reporting as a rich but flawed empirical resource. It connects most directly to critical assessments of DSA transparency data quality and utility, such as Rieder2025-ju, Rieder2026-pp, and Votta2025-xz, and to work on platform accountability and access under the DSA regime like Ahuja2025-ku. Its focus on linguistic and Global South inequities in moderation complements broader concerns about uneven platform governance across regions and communities.

Graph View

Backlinks