Paper, Tags: #nlp
We retrain coreference resolution on manually-annotated twitter conversation data.
We use a state-of-the-art resolution system built with OntoNotes corpus and experiment with adding annotated TWitter conversations to the training data.
We show that the performance of a state-of-the-art 'standard' coreference resolution system run on Twitter can improve by 21.6% by adding in-domain data.