Syntax-Aware Sentence Matching with Graph Convolutional Networks

Natural language sentence matching, as a fundamental technology for a variety of tasks, plays a key role in many natural language processing systems. In this article, we propose a new method which incorporates syntactic structure into "matching-aggregation" framework for sentence matching tasks. Our approach can be used in "matching-aggregation" framework efficiently. Concretely speaking, we introduce a multi-channel-GCN layer, which takes both words and the syntactic dependency trees of sentence pair as input to incorporate syntax information to the matching process. We also use a gating mechanism to dynamically combine the raw contextual representation of a sentence with the syntactic representation of the sentence to relieve the noise caused by the potential wrong dependency parsing result. Experimental results on standard benchmark datasets demonstrate that our model makes a substantial improvement over the baseline.