DAGA

This is the source code of our method proposed in paper "DAGA: Data Augmentation with a Generation Approach for Low-resource Tagging Tasks" accepted by EMNLP 2020.

Examples

flair_seq_tagger: sequense tagging model

cd flair_seq_tagger;

python train_tagger.py \
  --data_dir PATH/TO/TRAIN_DIR \
  --train_file  train.txt \
  --dev_file  dev.txt \
  --data_columns text ner \
  --model_dir ./model \
  --comment_symbol "__label__" \
  --embeddings_file PATH/TO/emb \
  --optim adam \
  --learning_rate 0.001 --min_learning_rate 0.00001 \
  --patience 2 \
  --max_epochs 100 \
  --hidden_size 512 \
  --mini_batch_size 32 \
  --gpuid 0

lstm-lm: LSTM language model

• train lstm-lm on linearized sequences

  
  cd lstm-lm;

python train.py \ --train_file PATH/TO/train.linearized.txt \ --valid_file PATH/TO/dev.linearized.txt \ --model_file PATH/TO/model.pt \ --emb_dim 300 \ --rnn_size 512 \ --gpuid 0

- generate linearized sequences

cd lstm-lm;

python generate.py \ --model_file PATH/TO/model.pt \ --out_file PATH/TO/out.txt \ --num_sentences 10000 \ --temperature 1.0 \ --seed 3435 \ --max_sent_length 32 \ --gpuid 0

## tools: tools for data processing
- preprocess.py: sequence linearization
- line2cols.py: convert linearized sequence back to two-column format

# Requirements
- flair\_seq\_tagger/requirements.txt
- lstm-lm/requirements.txt

# Citation
Please cite our paper if you found the resources in this repository useful.

@inproceedings{ding-etal-2020-daga, title = "{DAGA}: Data Augmentation with a Generation Approach for Low-resource Tagging Tasks", author = "Ding, Bosheng and Liu, Linlin and Bing, Lidong and Kruengkrai, Canasai and Nguyen, Thien Hai and Joty, Shafiq and Si, Luo and Miao, Chunyan", booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)", month = nov, year = "2020", address = "Online", publisher = "Association for Computational Linguistics", url = "https://www.aclweb.org/anthology/2020.emnlp-main.488", doi = "10.18653/v1/2020.emnlp-main.488", pages = "6045--6057", }