/hosrl

Code for EMNLP-2020 paper "High-order Semantic Role Labeling"

Primary LanguagePython

HO-SRL

Code for EMNLP-2020-Findings paper "High-order Semantic Role Labeling"

Under Construction

1. Prepare the training data

python scripts/conll09_to_conllu.py ./data/conll09/conll09-english/conll09_ENG_train.dataset ./data/conll09-converted/conll09_ENG_train.conllu 0
python scripts/conll09_to_conllu.py ./data/conll09/conll09-english/conll09_ENG_dev.dataset ./data/conll09-converted/conll09_ENG_dev.conllu 0
python scripts/conll09_to_conllu.py ./data/conll09/conll09-english/conll09_ENG_test.dataset ./data/conll09-converted/conll09_ENG_test.conllu 0
python scripts/conll09_to_conllu.py ./data/conll09/conll09-english/conll09_ENG_test_ood.dataset ./data/conll09-converted/conll09_ENG_test_ood.conllu 0

2. Predict the predicate sense with given predicate

# Use any sequence labeling model

3. Train the SRL model

python -u main.py train GraphParserNetwork --config_file config_srl/glove_01lr_5decay_srl_switch_new1.cfg --noscreen > ./logs/sec_order_baseline.log 2>&1