A Semi-shared Hierarchical Joint Model for Sequence Labeling

Multi-task learning is an essential yet practical mechanism for improving overall performance in various machine fields. Owing to the linguistic hierarchy, hierarchical joint model a common architecture used natural language processing. However, state-of-the-art models, higher-level tasks only share bottom layers or latent representations with lower-level thus ignoring correlations between at different levels, i.e., cannot be instructed by higher features. This paper investigates how advance among supervised end-to-end model. We propose semi-shared that contains cross-layer shared modules and layer-specific modules. To fully leverage mutual information we design four dataflows of Extensive experiments on CTB-7 CONLL-2009 show our approach outperforms basic models sequence tagging while having much fewer parameters. It inspires us proper implementation sharing residual shortcuts promising improve processing reducing complexity.