A Manual Annotation for the Task of Pronoun Resolution for Events

In this paper, we add an annotation for the task of Chinese pronoun resolution for events to the existing dataset. We first segment the article by sentence and parse the sentence into consistency parsing. Then we collect the pronouns from the consistency parsing. After that, we annotate them manually. Finally, we employ an existing model which contains BERT, LSTM, and MLP modules to test the performance of the annotated data. We also use two strategies to balance the number of positive and negative samples, making the F1 metric reach 0.537.