Closed MichaelCaohn closed 12 months ago
Thank you for your interest in our work.
I think you are correct in your interpretation. The evidence in this task is indeed an incomplete sentence. It's a result of how the data is processed for this task. The goal here is to provide enough context for the model to make the correct inference despite the incomplete sentence.
If you have further queries related to the data, I would recommend referring to the retrieval algorithm in the paper "Evidence-based Factual Error Correction". It better explains how the evidence and claims are generated and used.
Best Regards
Hi authors,
Thank you very much for the great work. I am very interested in it. I have looked into the t5_train.json file downloaded from the google drive link shared by you.
However, I have problems in interpreting the content. For example this line of input:
If we treat the content after "[evidence]" as the evidence, and the content after the "[claim]" to be the claim, the result would be:
evidence:
claim:
I think the claim should be correct, however, I think my interpretation has some problem with getting the evidence. As in the above interpretation, the evidence seems to be an incomplete sentence.
Could you help me on how to extract the evidence and the claim for this example?
Thank you in advance.
Best regards, Michael