yunan4nlp / NNEDUseg_conll

A RST EDU segmenter for conll format

RST EDU Segmenter

This is the reimplement code[BERT + BiLSTM] for the paper:
ToNy: Contextual embeddings for accurate multilingual discourse segmentation of full documents

Environment

Pytorch 1.0.0:
https://download.pytorch.org/whl/cpu/torch_stable.html

Data

DISRPT2019 Shared Task:
https://github.com/disrpt/sharedtask2019

We only prepare the RST Tree Bank.
https://catalog.ldc.upenn.edu/LDC2002T07

External Resource

BERT
https://github.com/huggingface/transformers

Performance

eng.rst.rstdt	P	R	F
Orignal Paper Test	95.29	96.81	96.04
Our Test	95.49	96.55	96.02