Improving Low-resource Dependency Parsing Using Multi-strategy Data Augmentation

• Published in: Ji suan ji ke xue Vol. 49; no. 1; pp. 73 - 79
• Main Authors: Xian, Yan-tuan, Gao, Fan-ya, Xiang, Yan, Yu, Zheng-tao, Wang, Jian
• Format: Journal Article
• Language: Chinese
• Published: Chongqing Guojia Kexue Jishu Bu 01.01.2022; Editorial office of Computer Science
• Subjects: Data augmentation; dependency parsing|low-resource language|mixup data augmentation|synonym substitution|multi-strategy; Feature extraction; Information retrieval; Machine translation; Performance enhancement; Syntax; Training
• ISSN: 1002-137X
• DOI: 10.11896/jsjkx.210900036

Dependency syntax analysis aims to identify the syntactic dependencies between words in a sentence. Dependency syntax can provide syntactic features for tasks such as information extraction, automatic question answering and machine translation, and improve model performance. The lack of training data will bring serious problems of unknown words and model overfitting. In this paper, a variety of data enhancement strategies are proposed for the low-resource-dependent syntactic analysis problem. The proposed method effectively expands training through synonym replacement. The problem of unknown words is alleviated. Through a variety of data enhancement strategies of Mixup, the problem of model overfitting is effectively alleviated and the generalization ability of the model is improved. The experimental results on the (Universal Dependencies treebanks, UD treebanks) data set show that , the proposed method effectively improves the performance of Thai, Vietnamese and English dependent syntactic analysis under the