Syntactic Graph Convolution in Multi-Task Learning for Identifying and Classifying the Argument Component

摘要

In this study, we focus on conducting fundamental research for combining syntactic knowledge with neural studies that utilize syntactic information in the argument component identification and classification (AC-I/C) tasks in argument mining. Our approach examines the role of syntactic graph convolutional networks (GCNs) used in multi-task learning for AC-I/C. Our contributions are as follows: 1) we propose a method to incorporate the dependency-based syntactic GCN into multitask learning models for AC-I/C tasks; 2) we demonstrate the effectiveness of the proposed syntactic GCN through experiments involving a few renowned argument mining datasets. The experimental results demonstrate that syntactic GCNs are effective when applied to specific datasets. Furthermore, we observe that the proposed syntactic GCN is promising for a lexically independent scenario. Additionally, our experimental code is available for reproducibility.¹