Implicit Discourse Relation Classification via Multi-Task Neural Networks

• DOI: 10.1609/aaai.v30i1.10339
• Publication Date: 2022-06-23T23:52:00Z
• Authors: Yang Liu; Sujian Li; Xiaodong Zhang; Zhifang Sui
• Keywords: FOS: Computer and information sciences; Computer Science - Computation and Language; Artificial Intelligence (cs.AI); Computer Science - Artificial Intelligence; 0202 electrical engineering, electronic engineering, information engineering; Computer Science - Neural and Evolutionary Computing; Neural and Evolutionary Computing (cs.NE); 02 engineering and technology; Computation and Language (cs.CL)

Without discourse connectives, classifying implicit relations is a challenging task and bottleneck for building practical parser. Previous research usually makes use of one kind framework such as PDTB or RST to improve the classification performance on relations. Actually, under different annotation frameworks, there exist multiple corpora which have internal connections. To exploit combination corpora, we design related tasks specific corpus, propose novel Convolutional Neural Network embedded multi-task learning system synthesize these by both unique shared representations each task. The experimental results relation demonstrate that our model achieves significant gains over baseline systems.