the implementation part of clarification in code?

[crescent73]

Hello, I found that in the “clarification” methods mentioned in Section 2.2 of the paper, I did not see the implementation in the code?

[raspberryice]

Hi Cresent73, It is true that the current code base does not cover the entity type statement. It was originally implemented as a postprocessing script that takes multiple decoding results and reranks them based on the probability of the decoding result + type statement. I'm working on another deadline now and will add more details on this over the weekend.

[crescent73]

Thanks for you reply. I still wonder that the clarification is only implemented by postprocessing script? I think it was a part of the model output(argument + type), so model can learn the type information during train.Is there something wrong with my understanding? Thanks again, and looing forward to your details explationation and implementation.

[raspberryice]

It's generally not a good idea to make BART output very long sequences (if you think of concatenating the filled templated with all of the type statements). One workaround is to simply modify the templates to make the argument types more explicit. I think this is what they did in this paper: https://arxiv.org/pdf/2108.12724.pdf