This is the reimplement code[BERT + BiLSTM] for the paper:
ToNy: Contextual embeddings for accurate multilingual discourse segmentation of full documents
Pytorch 1.0.0:
https://download.pytorch.org/whl/cpu/torch_stable.html
DISRPT2019 Shared Task:
https://github.com/disrpt/sharedtask2019
We only prepare the RST Tree Bank.
https://catalog.ldc.upenn.edu/LDC2002T07
BERT
https://github.com/huggingface/transformers
eng.rst.rstdt | P | R | F |
---|---|---|---|
Orignal Paper Test | 95.29 | 96.81 | 96.04 |
Our Test | 95.49 | 96.55 | 96.02 |