知识卡片
语言概率不保证文本纯单语
语言识别器可能给混合语段一个很高的单语言分数,因此“英语概率超过阈值”不等于每个片段都是英语。短文本、借词和代码混合更需检查;单一整段标签也不足以证明模型的全部多语能力来源。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 预训练数据
原书出处 · 1 处
- · 查看原书依据
Tools like langdetect and langid are helpful for speedy determination of the language in which the text is written at scale, but how do they deal with code-switched text (text in multiple languages, where English is often interspersed with a local
参考来源
- 位置:《Designing Large Language Model Applications》第02章《预训练数据》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch02.html)
- 结论依据:摘录中关于语言概率不保证文本纯单语的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。
- 原始内容:Tools like langdetect and langid are helpful for speedy determination of the language in which the text is written at scale, but how do they deal with code-switched text (text in multiple languages, where English is often interspersed with a local