A Span-based Linearization for Constituent Trees

Yang Wei, Yuanbin Wu, Man Lan


Abstract
We propose a novel linearization of a constituent tree, together with a new locally normalized model. For each split point in a sentence, our model computes the normalizer on all spans ending with that split point, and then predicts a tree span from them. Compared with global models, our model is fast and parallelizable. Different from previous local models, our linearization method is tied on the spans directly and considers more local features when performing span prediction, which is more interpretable and effective. Experiments on PTB (95.8 F1) and CTB (92.4 F1) show that our model significantly outperforms existing local models and efficiently achieves competitive results with global models.
Anthology ID:
2020.acl-main.299
Volume:
Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics
Month:
July
Year:
2020
Address:
Online
Venue:
ACL
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
3267–3277
Language:
URL:
https://aclanthology.org/2020.acl-main.299
DOI:
10.18653/v1/2020.acl-main.299
Bibkey:
Cite (ACL):
Yang Wei, Yuanbin Wu, and Man Lan. 2020. A Span-based Linearization for Constituent Trees. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 3267–3277, Online. Association for Computational Linguistics.
Cite (Informal):
A Span-based Linearization for Constituent Trees (Wei et al., ACL 2020)
Copy Citation:
PDF:
https://aclanthology.org/2020.acl-main.299.pdf
Video:
 http://slideslive.com/38928752
Code
 AntNLP/span-linearization-parser
Data
Penn Treebank