基于图神经网络的汉语依存分析和语义组合计算联合模型(Joint Learning Chinese Dependency Parsing and Semantic Composition based on Graph Neural Network)
Kai Wang (汪凯), Mingtong Liu (刘明童), Yuanmeng Chen (陈圆梦), Yujie Zhang (张玉洁), Jinan Xu (徐金安), Yufeng Chen (陈钰枫)
Abstract
组合原则表明句子的语义由其构成成分的语义按照一定规则组合而成, 由此基于句法结构的语义组合计算一直是一个重要的探索方向,其中采用树结构的组合计算方法最具有代表性。但是该方法难以应用于大规模数据处理,主要问题是其语义组合的顺序依赖于具体树的结构,无法实现并行处理。本文提出一种基于图的依存句法分析和语义组合计算的联合框架,并借助复述识别任务训练语义组合模型和句法分析模型。一方面图模型可以在训练和预测阶段采用并行处理,极大缩短计算时间;另一方面联合句法分析的语义组合框架不必依赖外部句法分析器,同时两个任务的联合学习可使语义表示同时学习句法结构和语义的上下文信息。我们在公开汉语复述识别数据集LCQMC上进行评测,实验结果显示准确率接近树结构组合方法,达到79.54%,而预测速度提升高达30倍。- Anthology ID:
- 2020.ccl-1.19
- Volume:
- Proceedings of the 19th Chinese National Conference on Computational Linguistics
- Month:
- October
- Year:
- 2020
- Address:
- Haikou, China
- Editors:
- Maosong Sun (孙茂松), Sujian Li (李素建), Yue Zhang (张岳), Yang Liu (刘洋)
- Venue:
- CCL
- SIG:
- Publisher:
- Chinese Information Processing Society of China
- Note:
- Pages:
- 195–206
- Language:
- Chinese
- URL:
- https://aclanthology.org/2020.ccl-1.19
- DOI:
- Cite (ACL):
- Kai Wang, Mingtong Liu, Yuanmeng Chen, Yujie Zhang, Jinan Xu, and Yufeng Chen. 2020. 基于图神经网络的汉语依存分析和语义组合计算联合模型(Joint Learning Chinese Dependency Parsing and Semantic Composition based on Graph Neural Network). In Proceedings of the 19th Chinese National Conference on Computational Linguistics, pages 195–206, Haikou, China. Chinese Information Processing Society of China.
- Cite (Informal):
- 基于图神经网络的汉语依存分析和语义组合计算联合模型(Joint Learning Chinese Dependency Parsing and Semantic Composition based on Graph Neural Network) (Wang et al., CCL 2020)
- PDF:
- https://preview.aclanthology.org/naacl-24-ws-corrections/2020.ccl-1.19.pdf