Multilingual Grammar Induction with Continuous Language Identification

This is PyTorch implementation of the paper:

Multilingual Grammar Induction with Continuous Language Identification
Wenjuan Han, Ge Wang, Yong Jiang, Kewei Tu
EMNLP 2019

The code performs unsupervised grammar learning in the dataset of 15 languages selected from UD treebanks v1.4.

Please concact [email protected] or [email protected] if you have any questions.

Environments

• Python 2.7
• PyTorch >=1.0
• Numpy 1.15.4

Data

run:

python ml_dmv_parser.py --cvalency 2 --do_eval --neural_epoch 1 --function_mask --lr 0.001 --epoch 70 --use_neural --embed_languages --em_type em --child_only --language_predict

The grammar induction is by default set to be performed on 15 selected languages, you can customize your own language selection by changing the contents in the language_list file

Reference

@inproceedings{han2019multilingual,
  title={Multilingual grammar induction with continuous language identification},
  author={Han, Wenjuan and Wang, Ge and Jiang, Yong and Tu, Kewei},
  booktitle={Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)},
  pages={5728--5733},
  year={2019}
}