A Coarse-to-Fine Labeling Framework for Joint Word Segmentation, POS Tagging, and Constituent Parsing

Proceedings of the 25th Conference on Computational Natural Language Learning (CoNLL), pages 290–299 November 10–11, 2021. ©2021 Association for Computational Linguistics 290 A Coarse-to-Fine Labeling Framework for Joint Word Segmentation, POS Tagging, and Constituent Parsing Yang Hou1, Houquan Zhou1, Zhenghua Li1∗, Yu Zhang1, Min Zhang1, Zhefeng Wang2, Baoxing Huai2, Nicholas Jing Yuan2 1School of Computer Science and Technology, Soochow University, China 2 Huawei Cloud, China 1{yhou1,hqzhou}@stu.suda.edu.cn 1{zhli13,minzhang}@suda.edu.cn 1yzhang.cs@outlook.com 2{wangzhefeng, huaibaoxing, nicholas.yuan}@huawei.com Abstract The most straightforward approach to joint word segmentation (WS), part-of-speech (POS) tagging, and constituent parsing (PAR) is converting a word-level tree into a char-level tree, which, however, leads to two severe challenges. First, a larger label set (e.g., ≥ 600) and longer inputs both increase computational cost. Second, it is difficult to rule out illegal trees containing conflicting production rules, which is important for reli- able model evaluation. If a POS tag (like VV) is above a phrase tag (like VP) in the output tree, it becomes quite complex to decide word boundaries. To deal with both challenges, this work proposes a two-stage coarse-to-fine labeling framework for joint WS-POS-PAR. In the coarse labeling stage, the joint model outputs a bracketed tree, in which each node corresponds to one of four labels (i.e., phrase, subphrase, word, subword). The tree is guaranteed to be legal via constrained CKY decoding. In the fine labeling stage, the model expands each coarse label into a final label (such as VP, VP∗, VV, VV∗). Experiments on Chinese Penn Treebank 5.1 and 7.0 show that our joint model consistently outperforms the pipeline approach on both settings of without and with BERT, and achieves new state-of-the-art performance. 1 Introduction As shown in Figure 1(a), the results of word seg- mentation (WS), part-of-speech (POS) tagging, and constituent parsing (PAR) can be organized in a unified hierarchical tree, where leaf nodes include words and their POS tags, and non-terminal nodes correspond to phrases (or constituents) with their syntactic label. Before the deep learning (DL) era, there had been intensive research interest in jointly modeling of the three tasks, i.e., WS-POS- PAR (Luo, 2003; Qian and Liu, 2012; Zhang et al., ∗Corresponding author IP PU 。 .4 VP VP VA 美味 Delicious3 ADVP AD 很 Very2 NP NN 巧克力 Chocolate1 (a) Word-level tree IP PU 。 .7 VP VP VA 味 Taste6 美 Beautiful5 ADVP AD 很 Very4 NP NN 力 -late3 克 -co-2 巧 Cho-1 (b) Char-level tree Figure 1: Example WS-POS-PAR trees. The English translation is “The chocolate is very delicious.” 2013; Wang et al., 2013). The motivations are three-fold: 1) alleviating error propagation, 2) pro- moting knowledge sharing and interaction during inference, 3) simplifying system architecture by training a single model. In contrast, there have been few work on joint WS-POS-PAR in the DL era, with the exception of Zheng et al. (2015). Similar to this work, they treated the joint problem as a char-level parsing task, which is obviously the most straightforward way for Chinese, as illustrated in Figure 1(b). How- ever, they showed that …

Content truncated to 500 words