Xuefeng Bai

The Wechat 2-D code is overdue, could you upload a new one? :)

Hi, Thanks for your interest. For coreference, we append the speaker names to the beginning of each utterance before applying the neural coreference tool. To ensure the quality, we set...

Hi, In our scripts, we set BPE num_operation as 10000. We have tried BERT_tokenizer, but we don't observe significant improvements.

Sorry for my late reply, I can't find the exact codes file now, you can try [this one](https://1drv.ms/u/s!ArC7JSpdBblgqBM6cWY-hZIwuLN5?e=erzz03). Also, please note that we do not apply bpe on AMR concepts...

I have uploaded it to google drive [here](https://drive.google.com/file/d/19epz-bH1wEAH-ArAN02g5U7_25xJwP73/view?usp=sharing), please have a try.

好的，那我再研究下，谢谢了

Hi, @Sreyan88 Please refer to [this repository](https://github.com/goodbai-nlp/amr-masking) for a offline version of graph masking. By the way, we did not find significant improvements when applying offline graph masking.

Hi, @LTC232 Please refer to [issue#1](https://github.com/goodbai-nlp/Sem-Dialogue/issues/1) and [issue#2](https://github.com/goodbai-nlp/Sem-Dialogue/issues/2).

I met the same problem and solved it by upgrading pytorch from 0.4.0 to 0.4.1.

I also found that beam_search always get much worse results than greedy decoding.