search

vdobrovolskii / wl-coref

This repository contains the code for EMNLP-2021 paper "Word-Level Coreference Resolution"
MIT License
104 stars 37 forks source link

about how to predict with gold mention? #28

Closed leileilin closed 2 years ago

leileilin commented 2 years ago

hey, sorry to bother u again, i want to ask if i know mentions in a doc, how to predict with such gold mentions? thanks.

vdobrovolskii commented 2 years ago

You can run the model as usual and then filter the results, removing unwanted mentions and adjusting mention boundaries as needed.

leileilin commented 2 years ago

You can run the model as usual and then filter the results, removing unwanted mentions and adjusting mention boundaries as needed.

We know that the anaphora resolution model is generally composed of two modules: mention detection and mention pair detection, and the mention detection module is not abosolutely right. so if i have gold mentions with a doc, the menthod above proposed by u is still get little mention detection error, so i want to pass gold mention representations to model for better performance. thanks.

vdobrovolskii commented 2 years ago

A feature of this particular model is searching for coreference links between individual words and not pre-extracted mentions. The mentions are extracted after the coreference has been resolved, using span_predictor.py