Verified65% confidenceFactTime unknown
English data accounts for an overwhelming proportion of mainstream large language model training corpora; taking public datasets such as Common Crawl as an example, English content typically constitutes 40%–60% of the total, while languages such as Vietnamese and Chinese often account for less than 5%.
3
Sources
65%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedEnglish content in mainstream large language model training corpora typically exceeds 50%, leading to significant performance gaps for non-English languages.81% similarUnverifiedMeta Llama 2的2万亿token训练数据中英语占比约89.7%，欧洲语言合计约8%，其余所有语言共享约2%75% similarUnverifiedIn general-purpose corpora such as Common Crawl, English typically accounts for more than 45%, while Portuguese accounts for less than 2%.74% similarUnverifiedSafety alignment training data is most comprehensive in English, and may have significant coverage gaps in languages such as Arabic and Urdu, constituting a multilingual jailbreak vulnerability.72% similarVerifiedThe safety alignment of many models degrades significantly in non-English languages, and low-resource languages are especially easy to exploit because the human annotation data for RLHF alignment training is highly concentrated in English.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/31032API
curl https://kongchang.com/api/v1/knowledge/claims/31032MCP
get_claim(id=31032)