In order to expand the scale of manual annotated data and thereby improve model performance, we attempt to make full use of existing heterogeneous annotations to learn model parameters. We extend coupled sequence labeling model proposed by Li et al. (2015) under the BiLSTM-based deep learning framework. The neural coupled model learn its parameters directly on two heterogeneous training data, and predicts two optimal sequences simultaneously during the test phase. A lot of experiments have been conducted on the part-of-speech (POS) tagging task and the joint word segmentation and POS (WS&POS) tagging task. The results show that neural coupled approach is superior to other methods for exploiting heterogeneous lexical data, including the multi-task learning method and the traditional discrete-feature coupled model. Neural coupled model achieves higher performance on both scenarios, i.e., annotation conversion and boost the final target-side tagging accuracy by exploiting heterogeneous data.

Considering the close correlation between syntactic and semantic structures, this paper attempts to add syntactic information into the universal conceptual cognitive annotation (UCCA) semantic parsing model to enhance the performance of semantic parsing. Based on the state-of-the-art graph-based UCCA semantic parser, we propose and compare four different approaches for incorporating syntactic information. Experiments are conducted on the English benchmark dataset for the semantic parsing shared task of the SemEval-2019 conference. The results on both the in-domain and out-domain evaluation data show that syntax-enhanced methods can achieve significant improvements of UCCA parsing. After utilizing BERT, syntactic information is still beneficial to some extent.