title
A Unified MRC Framework for Named Entity Recognition
Abstract:The task of named entity recognition (NER) is normally divided into nested NER and flat NER depending on whether named entities are nested or not. Models are usually separately developed for the two tasks, since sequence labeling models are only able to assign a single label to a particular token, which is unsuitable
fornested NER where a token may be assigned several labels.
concepts
实体重叠
可能会出现实体重叠的问题，即一个句子“席慕容散文集是我最喜欢的书”。在这个句子中“席慕容”和“席慕容散文集”都是实体，并且有重叠的部分。但传统做法无法解决此类问题，因为一个token只属于一个tag。
Qusetion
本文针对实体重叠问题提出了一种统一的框架，可以分别处理falt and nested NER task.
Strategy
与传统的序列标记问题不同的是，作者采用了MRC（machine reading comprehension)来进行任务完成。即提取两个实体需要提问两次得到答案。
