A Study of Recognizing Implicit Discourse Relations in the Penn Discourse Treebank

Implicit discourse relation recognition has emerged to be a difficult problem due to lacking connectives in text. Traditional methods mainly adopted supervised methods which require a large amount of labeled training data. In this work, we explore two novel semi-supervised approaches to classify implicit discourse relations with a small amount of labeled data, i.e. self-training and co-training. Results on Penn Discourse Treebank (PDTB) 2.0 show that our proposed models outperform traditional methods with small training size. The discovery presented in this paper can be helpful for future research on recognizing implicit discourse relations.