This is the repo for the code to reproduce the experiments and results of the system entry C2L2 in the CoNLL shared task.
For more up-to-date parser under maintenance, please see https:/tzshi/cdparser.
- DyNet 1.1 (tested with commit b7e4f4e1d9129b6b4cef8c3c85da58dac5adc392 )
Example training script calling from bash:
MKL_NUM_THREADS=2 python -m cdparser_multi.cdparser \
- build-vocab $TRAIN_FILE \
- create-parser \
- init-model \
- train $TRAIN_FILE --dev $DEV_FILE \
--utag True --mst True --aedp True --ahdp True \
- finish --dynet-mem 2000
Example test script calling from Python environment:
parser = CDParser()
parser.load_model(model_file, verbose=False)
parser.predict(graphs, aedp=True)
If you make use of this software in your research, we appreciate you citing the following papers:
@InProceedings{shi-etal2017conll,
author = {Shi, Tianze and Wu, Felix G. and Chen, Xilun and Cheng, Yao},
title = {Combining Global Models for Parsing Universal Dependencies},
booktitle = {Proceedings of the CoNLL 2017 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies},
month = {August},
year = {2017},
address = {Vancouver, Canada},
publisher = {Association for Computational Linguistics},
pages = {31--39},
url = {http://www.aclweb.org/anthology/K17-3003}
}
@InProceedings{shi+huang+lee2017exact,
author = {Shi, Tianze and Huang, Liang and Lee, Lillian},
title = {Fast(er) Exact Decoding and Global Training for Transition-based Dependency Parsing via a Minimal Feature Set},
booktitle = {Proceedings of the Conference on Empirical Methods in Natural Language Processing},
month = {September},
year = {2017},
address = {Copenhagen, Denmark},
publisher = {Association for Computational Linguistics},
pages = {(To appear)}
}
When implementing the first-order graph-based algorithm, we referenced the BiST parser: https:/elikip/bist-parser.