The pandemic presents the biggest language translation challenge for COVID-19 info, as Wikipedia supports 309 languages, Google Translate 109, and Bing 71
Services like Google Translate only support 100 languages, give or take. What about the thousands of other languages—spoken by people just as vulnerable to this crisis? Tweets: @appenglobal , @backchnnl , @professorolsen , @itiuk , @bentranslates , and @wired Tweets: Appen / @appenglobal : We are proud to be working with @TranslatorsWB alongside many tech leaders on the Translation Initiative for #COVID19 to translate Covid-related messaging to 36 languages. https://www.wired.com/... Backchannel / @backchnnl : Services like Google Translate only support 100 languages, give or take. What about the thousands of other languages—spoken by people just as vulnerable to this crisis? https://www.wired.com/... Barry S. Olsen / @professorolsen : Covid-19 Is History's Biggest #Translation Challenge (And it's biggest interpreting challenge too, I would add.) #xl8 https://www.wired.com/... @itiuk : Covid information from reputable sources has been created in over 500 languages and counting, including over 400 videos in over 150 languages. https://www.wired.com/... @bentranslates : “There is a recognition that language is a vital part of the #Covid response, an understanding that's come from hard-gained experience.” Thanks @GretchenAMcC for this great article https://www.wired.com/... #t9n @wired : In a pandemic, the challenge isn't just translating one or a handful of primary languages in a single region—it's on a scale of perhaps thousands of languages. https://www.wired.com/...
Context & Ripple Effects
In mid-2020 the translation layer for the pandemic is visibly thinner than the crisis itself: Wikipedia carries COVID-19 information in 309 languages, but Google Translate reaches 109 and Bing just 71 — a gap the Translation Initiative for COVID-19, run by Appen with Translators Without Borders, tries to narrow by targeting messaging into 36 languages while public health organizations have produced material in over 500. The stakes are clinical as well as informational, since researchers have already documented serious errors when doctors rely on machine translation with non-English-speaking patients (researchers flag inaccurate Google Translate results in medical settings).
The arc that follows shows both progress and a new risk: months later the WHO grants Wikipedia free use of its published data to spread COVID information across nearly 200 languages (WHO's free-data agreement with Wikipedia), and by 2024 Google's largest-ever expansion pushes Translate to 243 languages via its PaLM 2 model (Google's PaLM 2 expansion of Translate). But the same period brings warnings that machine-translated pages feeding AI training corpora may entrench errors and further marginalize vulnerable languages (the risk of an AI translation 'doom spiral').
First-order effects
- Speakers of the thousands of languages outside Google's 109 and Bing's 71 depend on human efforts like Appen's 36-language initiative and Wikipedia's volunteer base, making Wikipedia's editor community the de facto distribution network for multilingual COVID guidance.
- Appen and Translators Without Borders' tech partners take on immediate triage work: deciding which of the 500-plus-language body of public health content gets professionally translated first.
Second-order effects
- Google responds over time with capability rather than curation — its PaLM 2-driven jump to 243 languages narrows the coverage gap while shifting the bottleneck from language count to translation accuracy, the exact problem clinicians already hit with patient conversations.
- Health authorities lean on institutional channels like the WHO–Wikipedia data deal to guarantee authoritative text in many languages, reducing reliance on consumer translation tools whose error rates are documented in medical contexts.
Third-order effects
- If low-resource languages are served mainly by machine translation trained on machine-translated text, errors compound across the corpus — the dynamic MIT Technology Review describes as a doom spiral — meaning the languages added fastest may be the least reliably served.
- The pattern points toward a two-tier global information infrastructure: high-resource languages get platform-scale automated translation while vulnerable languages stay dependent on volunteer and NGO pipelines like Translators Without Borders.
The trend: Global health communication is outgrowing commercial machine-translation coverage, forcing a division of labor between AI-expanded platforms, Wikipedia's volunteer base, and NGO translation networks — with accuracy, not reach, becoming the binding constraint for vulnerable languages.