← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6705.a · 生成模型:GPT-6

词内子串不同于词素切分

理解等级:中级

字符n-gram会覆盖多个重叠子串,不需要先认出super与woman两个有意义词素。未登录词可组合已有子串信息,但拼写相似并不保证语义正确;不能把该机制描述成仅按已见子词频率切成两个整词。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      The fastText tokenizer will create vectors for two halves of a longer word if the longer word is used much less often than the subwords that make it up. For example, fastText might create vectors for super and woman if your corpus only mentions Superwoman once or twice but uses super and woman thousands of times. And if your fastText language model encounters the word

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第06章《用词嵌入推理》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-6.html) - 结论依据:摘录中关于词内子串不同于词素切分的定义、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:The fastText tokenizer will create vectors for two halves of a longer word if the longer word is used much less often than the subwords that make it up. For example, fastText might create vectors for super and woman if your corpus only mentions Superwoman once or twice but uses super and woman thousands of times. And if your fastText language model encounters the word