BiaffineDependencyParsing

BERT+Transformer+Biaffine dependency parser

Update

[2020-04-23] 修复数据加载中一个bug，在use_cache=True时可大幅度缩短数据加载准备耗时

Result

以上结果均为在Semeval-2016-Test集上测试得到的LAS

详细结果见: metrics记录

Semeval-2016 Task9数据集

原始评测数据集：HIT-SCIR/SemEval-2016: SemEval-2016 Task 9: Chinese Semantic Dependency Parsing

原始数据集格式不符合conllu格式，你也可以使用我们处理过后的数据集，或者使用脚本自己处理

数据集处理完成之后需要在yaml配置文件中写入，例如：

data_set:

data_dir: 'dataset/sem16'

train_file: 'train/text_news.train.conllu'

dev_file: 'dev/sdp_news_dev.conllu'

test_file: 'test/sdp_news_test.conllu'

同时根据数据集的相关信息合理设置per_gpu_train_batch_size、max_seq_len等参数

使用

训练Train

查看帮助信息：python main.py train -h

CUDA_VISIBLE_DEVICES=0,1,2,3 python main.py train -c config_files/bert_biaffine.yaml

torch.distributed 分布式训练

distributedDataParallel VS dataParallel

当GPU卡数较多或数据量较大（或者max_seq_len较大时），推荐使用torch.distributed训练

CUDA_VISIBLE_DEVICES=0,1,2,3,5,6 python -m torch.distributed.launch --nproc_per_node=6 main.py -c config_files/bert_biaffine.yaml

验证Dev

查看帮助信息：python main.py dev -h

CUDA_VISIBLE_DEVICES=0 python main.py dev -m <模型保存路径> -i <输入gold conllu文件> -o <输出文件>

推理Inference

查看帮助信息：python main.py infer -h

CUDA_VISIBLE_DEVICES=0 python main.py infer -m <模型保存路径> -i <输入conllu文件> -o <输出文件>

References

[1] Ding, Y., Shao, Y., Che, W., Liu, T., 2014. Dependency Graph Based Chinese Semantic Parsing, in: Sun, M., Liu, Y., Zhao, J. (Eds.), Chinese Computational Linguistics and Natural Language Processing Based on Naturally Annotated Big Data, Lecture Notes in Computer Science. Springer International Publishing, Cham, pp. 58–69. https://doi.org/10.1007/978-3-319-12277-9_6

[2] Wang, Y., Che, W., Guo, J., Liu, T., 2018. A Neural Transition-Based Approach for Semantic Dependency Graph Parsing, in: AAAI.

[3] Dozat, T., Manning, C.D., 2018. Simpler but More Accurate Semantic Dependency Parsing, in: Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers). Presented at the ACL 2018, Association for Computational Linguistics, Melbourne, Australia, pp. 484–490. https://doi.org/10.18653/v1/P18-2077

[4] Li, Y., Li, Z., Zhang, M., Wang, R., Li, S., Si, L., 2019. Self-attentive biaffine dependency parsing, in: IJCAI International Joint Conference on Artificial Intelligence. pp. 5067–5073. https://doi.org/10.24963/ijcai.2019/704

[5] Kondratyuk, D., Straka, M., 2019. 75 Languages, 1 Model: Parsing Universal Dependencies Universally. pp. 2779–2795. https://doi.org/10.18653/v1/d19-1279

相关项目：

• https://github.com/Hyperparticle/udify
• https://github.com/nikitakit/self-attentive-parser
• https://github.com/WangYuxuan93/CLBT
• https://github.com/stanfordnlp/stanfordnlp