Facebook supports 111 languages, but translates content rules to just 41, and its moderation staff speaks ~50; deficit leaves it struggling to monitor content
NAIROBI/SAN FRANCISCO (Reuters) - Facebook Inc's struggles with hate speech and other types of problematic content are being hampered … Tweets: @yolandavalery and @bafeldman See also Mediagazer Tweets: Yolanda Valery / @yolandavalery : “Facebook's 15,000-strong content moderation workforce speaks about 50 tongues, though the company said it hires professional translators when needed. Automated tools for identifying hate speech work in about 30” http://www.reuters.com/... Brian Feldman / @bafeldman : facebook has hamburger menus in 111+ languages, and rules about not endorsing genocide in less than half that http://www.reuters.com/... http://twitter.com/... See also Mediagazer
Context & Ripple Effects
The arithmetic behind this report has been years in the making: Facebook rebuilt its translation backend on neural networks in 2017 to handle billions of daily translations for the product itself, but the governance layer never scaled to match. Even inside well-resourced markets, ProPublica found hate speech rules enforced unevenly; Reuters now quantifies the structural version of that problem — 111 interface languages against 41 translated rulebooks, a 15,000-person moderation workforce speaking about 50 languages, and automated hate-speech detection covering roughly 30.
Facebook's own disclosures show the gap is not hypothetical: it pointed to rising proactive AI removals in Myanmar as evidence of progress, yet leaked documents later revealed that some European markets had no automated moderation at all — Finland was covered by just 11 Berlin-based moderators. The Reuters numbers explain why those per-country shortfalls keep surfacing: they are symptoms of a single design decision to internationalize the product faster than the policy apparatus.
First-order effects
- Users posting in the roughly 70 languages outside the 41 translated rule sets are held to policies they cannot read, while reviewers who do not speak a post's language must rely on automation that itself covers only ~30 languages or on professional translators hired case-by-case.
- Markets like Myanmar, where Facebook touted AI-driven removal gains, sit at the sharp end of the deficit: detection quality tracks directly with whether a language made the ~30-language cut for automated tools.
Second-order effects
- Governments and civil society in low-coverage language markets gain concrete evidence for demanding locally staffed, locally language-capable moderation, turning per-market localization from a discretionary spend into a recurring regulatory cost for Facebook.
- Each new interface language Facebook adds widens the enforcement surface faster than hiring can close it, so the deficit compounds with growth rather than shrinking with scale.
Third-order effects
- If the pattern holds, global platforms harden into a two-tier system in which product features ship in 100+ languages by default while safety infrastructure ships in dozens — entrenching unequal protection by language rather than by law, and giving regulators a ready-made argument for mandating local-language compliance.
The trend: Consumer platforms are internationalizing their products far faster than their governance, leaving language coverage of enforcement — not feature rollout — as the binding constraint on global scale.