I’m currently running the neural coref model on newspaper articles, and it’s interesting that the model does not match the pronouns to the correct entity when they seem like rather easy coreferences such as the ones shown below.
Unlike the demo, the only references we get from the codes would be just “his: [he, his, he]” something like that. They are making coreferences but not in the way we would desire to as I want Matuszewski as the reference. what is going on here? I thought that the model would perform a lot better.