Multiparty Dialogue Machine Reading Comprehension (MRC) differs from traditional MRC models as models must handle complex dialogue discourse structure . We present a discourse-aware dialogue graphneural network, DADgraph, which explicitly constructs the dialogue graph using discourse dependency links and discourse relations . Experiments on Molweni corpus, a large-scale MRC dataset built overmultiparty dialogue annotated with discourse structure, proved statistically significant improvements compared against strong neural network MRC baselines .

Author(s) : Jiaqi Li, Ming Liu, Zihao Zheng, Heng Zhang, Bing Qin, Min-Yen Kan, Ting Liu

Links : PDF - Abstract

Code :
Coursera

Keywords : dialogue - discourse - mrc - network - graph -

Leave a Reply

Your email address will not be published. Required fields are marked *