Yanzeng Li's paper "Enhancing Chinese Pre-trained Language Model via Heterogeneous Linguistics Graph" has been accepted by ACL 2022.

　　By introducing the easy-to-get Chinese word segmentation information, we propose a Heterogeneous Linguistics Graph (HLG) to represent the hierarchical linguistic structure in Chinese language patterns. The proposed multi-step information propagation GCN is adopted to model the HLG and integrate it into the pre-trained language model as a plug-in adapter. The experimental results demonstrate the performance of the proposed method in 6 Chinese natural language processing tasks with 10 benchmark datasets. Compared with previous work, our method introduces only half the additional parameters while its training/inference speed is increased by more than 7x times.