mZPRT

Multi-domain Zero Pronoun Recovery and Translation Dataset

Citation

Title: A Benchmark for Zero Pronoun Recovery and Translation

If you use this benchmark, Please cite our paper:

@inproceedings{xu2022guofeng,
  title={GuoFeng: A Benchmark for Zero Pronoun Recovery and Translation},
  booktitle={Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing},
  pages={11266--11278},
  year={2022},
}

LICENSE

See LICENSE. We follow the data licensing plan as the same as the WMT benchmark.

Statistic

We release a benchmark for Zero Pronoun Recovery and Translation, this benchmark contains the zero pronoun annotations (handcraft) from five different source.

Catalog

.
├── 1_testset_mzprt            # Our benchmark testset
│   ├── processed              # tokenized and BPEed testset
│   │   ├── context-agnostic   # original/oracle testset for sent-level
│   │   ├── context-aware      # original/oracle testset for doc-level
│   │   ├── labeled-target     # target sequences with ZP-label
│   ├── raw                    # raw testset for each domain
│   ├── script                 # script for preprocessing and scoring bleu
├── 2_metric_azpt              # Our evaluation metric
│   ├── aZPT                   # evaluation toolkit
│   ├── aZPT_output            # output files in details of aZPT
│   ├── human_score            # human judgements on 6 systems
│   ├── scirpts                # scripts for getting alignment
├── 3_comparative_models       # Bechmark related resources
│   ├── mt                     # Machine translation task
│   │   ├── data               # training data for MT baseline
│   │   │   ├── FT_dataset     # domain-specific data for QA Forum and Web Fiction
│   │   │   ├── Movie_Subtitle # training data for Movie_Subtitle
│   │   │   ├── WMT2021        # training data for Others
│   │   ├── code               # training code
│   │   ├── model              # comparative models  
│   ├── zpr                    # Zero pronoun recovery task
│   │   ├── data               # training data for zpr
│   │   │   ├── Movie_Subtitle # domain-specific data for Movie Subtitle
│   │   │   ├── QA_Forum       # domain-specific data for QA Forum
│   │   │   ├── Others         # training data for Others
│   │   ├── scirpt             # Make training data from raw
│   │   ├── code               # training codes for zpr task
│   │   ├── model              # comparative_models  
│   └── zpt                    # Zero pronoun translation task
│   │   ├── context-aware      # Doc-level MT
│   │   │   ├── scirpt         # make doc-level data
│   │   │   ├── model          # comparative_models  
│   │   │   ├── codes          # training codes for doc-level mt
│   │   ├── reconstructor      # Reconstructor
│   │   │   ├── scirpt         # make training data
│   │   │   ├── model          # comparative_models  
│   │   │   ├── codes          # training codes for reconstructor
└── README.md

Contact information

Xu, Mingzhou : [email protected]
Wang, Longyue : [email protected]