Chinese Complement (补语) Classification and Explanation

Official repository for Automatic Prediction and Linguistic Interpretation of Chinese Directional Complements(趋向补语) Based on BERT Model at Chinese Lexical Semantics Workshop 2021

• Authors: Younghoon Jeong, Ming Yue Li, Su Min Kang, Yun Kyung Eum, Byeong Kwu Kang
• This code is reimplementation of original code

Environment

conda create -f environment.yaml conda activate soc_env

How to run

1. Train

./scripts/run_train_buyu.sh {gpu_id}

• Automatically trains four 补语 classifiers (下去，下来，出来，起来)
• You can change hyperparameters from script

2. Explain

./scripts/run_explain_buyu.sh {gpu_id}

runs explanation with the test data

• --output_dir : directory of classifier model
• --lm_dir : directory of language model (different with classifier)
• --nb_range : window size to sample n
• --sample_n : number of sentences to sample
• --hiex : hierarchical explanation (default : sequential explanation)
• --output_filename : name of the explanation file

explanation result is saved in output_dir

3. Visualize

`./scripts/run_visualize_buyu.sh``

• --input_file : file dir to visualize (*.txt, *.heix)
• --buyu : buyu #
• --hiex : true if hierarchical explanation

---

Final output

Example of Sequencial Explanation

Example of Hierarchical Explanation

How to modify

1. Prepare data

|----data/buyu
|----|----train.json
|----|----dev.json
|----|----test.json

• Binary label

2. Modify data processor

loader/*.py

• modfiy class *Processor

---

Reference

@inproceedings{kennedy2020contextualizing,
   author = {Kennedy*, Brendan and Jin*, Xisen and Mostafazadeh Davani, Aida and Dehghani, Morteza and Ren, Xiang},
   title = {Contextualizing {H}ate {S}peech {C}lassifiers with {P}ost-hoc {E}xplanation},
   year = {to appear},
   booktitle = {Proceedings of the 58th {A}nnual {M}eeting of the {A}ssociation for {C}omputational {L}inguistics}
}