search

NLP-Discourse-SoochowU / sota_end2end_parser

According to my recent four years' research in DRS parsing, I will publish a pre-trained end-to-end DRS parser here for other researchers to directly apply it to specific NLP downstream tasks. Researchers using this project please cite our paper at xxx (coming soon).
Apache License 2.0
5 stars 3 forks source link

RST relation #3

Open NLP-Discourse-SoochowU opened 1 year ago

NLP-Discourse-SoochowU commented 1 year ago

In the code, each node in RST contains two fields about relations: rel and child_rel. What is the difference between them?

NLP-Discourse-SoochowU commented 1 year ago

Re: RST parsing has two main evaluation methods, 1. In RST-Parseval, the relation between two tree nodes is attached to both nodes (e.g., List-NN is the relation between them, we label 'List' and 'List' for both nodes' "rel"); 2. In the original Parseval, the relation between two tree nodes is attached to the parent node (e.g., List-NN is the relation between them, we label 'List' for the parent node's "child_rel").