Constituent Parser

This is a project of the following constituent parsers:

• Chart Parser (from A Minimal Span-Based Neural Constituency Parser).
• Top-Down Parser (from A Minimal Span-Based Neural Constituency Parser).
• Shift-Reduce Parser (from Span-Based Constituency Parsing with a Structure-Label System and Provably Optimal Dynamic Oracles).
• In-Order Parser (Ours).
• GNN Parser (Future work).

The shift-reduce parser code still has not been reconstructed.

Train

Full dataset:

python3 run/train.py --model InOrderParser

Full dataset (for more training):

python3 run/train.py --model InOrderParser --train_more --more_epoch 20

Small dataset (for code correctness test):

python3 run/train.py --model InOrderParser --train_file data/train_small.trees --dev_file data/dev_small.trees

Test

python3 run/test.py --model InOrderParser --dev_fscore xx.xx