Unverified50% confidenceFactExact time
孟加拉语-英语的高质量平行数据仅有数百万级别，而英法语对拥有数千万句对齐语料
1
Sources
50%
Confidence
Long-term
Relevance
8/21/2026
First Seen
Sources
Related Claims
UnverifiedEnglish speech datasets can reach hundreds of thousands of hours, while many Indian languages have only hundreds to thousands of hours of high-quality annotated data72% similarUnverified主流大语言模型训练语料中，英文数据占据压倒性比例，以Common Crawl等公开数据集为例，英文内容通常占总量的40%-60%，而越南语、中文等语言的占比往往不足5%70% similarUnverified手语翻译的基准数据集（如 PHOENIX-2014T、CSL-Daily、How2Sign）规模通常仅有数千到数万条样本，远小于机器翻译领域数百万句对的平行语料69% similarUnverified词库语种越多、发音质量越高，价格和体积越大；纯英语学习者选专精英语的机型比选'十国语言'的通用机型更划算，多语种机型往往每个语种的词库深度和例句都打折扣66% similarUnverified支持语言数量超过100种的机型多为营销噱头，实际小语种（如斯瓦希里语、乌尔都语）准确率往往低于60%，真正决定体验的是主流20种语言的翻译质量而非语种总数64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/784041API
curl https://kongchang.com/api/v1/knowledge/claims/784041MCP
get_claim(id=784041)