This repository includes the code of the Semantic Role Labeling (SRL) Parser with label-aware graph convolutional network (LAGCN) based pointer networks of the AAAI 2021 paper: Encoder-Decoder Based Unified Semantic Role Labeling with Label-Aware Syntax.
pip install -r requirements.txt
Download them and put at ./data
folds.
To prepare the syntactic dependency features, deploy the CoreNLP:
wget https://nlp.stanford.edu/software/stanford-corenlp-latest.zip
nohup java -mx4g -cp "*" edu.stanford.nlp.pipeline.StanfordCoreNLPServer -port 8083 -timeout 15000 > 1.log 2>&1 &
See the example data in ./data/demo.
Step 1. To train the parser, you need to include the pre-trained word embeddings in the embs
folder and run the following script:
./scripts/run_parser.sh <model> <data>
To evaluate the best trained model on the test set, just use the official script to compute the F1 scores:
./scripts/eval.sh <best epoch> <data> <model>
@inproceedings{FeiGraphSynAAAI21,
author = {Hao Fei and Fei Li and Bobo Li and Donghong Ji},
title = {Encoder-Decoder Based Unified Semantic Role Labeling with Label-Aware Syntax},
booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence},
pages = {12794--12802},
year = {2021},
}
The code is released under Apache License 2.0 for Noncommercial use only.