Linguistically-Informed Self-Att

1. jointly predict parts of speech and predicates
   • parts of speech 词性标注
   • predicates 谓语标注，是Semantic Role Labeling的一个子任务，把句子中的谓词标注出来，用于后续SRL。
2. perform parsing and attend to syntactic parse parents
   • Syntactically-informed self-attention，源自于论文[2]，进行 dependency parsing。需要注意，这一层是有监督信号的，根据attention A_parse 计算每一个token的head，同时利用Q_parse 和 K_parse 计算dependency label。
3. assigning semantic role labels
   • 根据给定的predicate，预测每个token与predicate的关系。

reference：

1. Linguistically-Informed Self-Attention for Semantic Role Labeling
2. Deep Biaffine Attention for Neural Dependency Parsing
3. Deep Semantic Role Labeling: What Works and What’s Next
4. http://www.hankcs.com/nlp/parsing/deep-biaffine-attention-for-neural-dependency-parsing.html
5. https://blog.csdn.net/mingzai624/article/details/78061506