1. "the joint task" folder contains all the source code for predicting the joint task of word segmentation, POS tagging, and constituent parsing. "the single task" folder contains all the source code for predicting the single task of constituent parsing.

2. In the project, the code released from the Berkeley parser [1] is employed as the basic architecture, and our implementation of the proposed recursive semi-Markov model is incorporated by adding or modifying the following files. 
my_chart_helper.py (newly added)
global_para.py (newly added)
myutil.py (newly added)
evaluate.py (newly added in the joint task only)
transformer.py (modified)
parse_nk (modified)
main.py (modified)
trees.py (modified in the joint task only)

[1] Kitaev, Nikita; Cao, Steven; and Klein, Dan. 2019. Multilingual Constituency Parsing with Self-Attention and Pre-Training. In Proc. of ACL 2019, 3499–3505.