Introduction

About this repository

.├── README.md├── mtnt_approp_annotated.tsv # pre-filtered MTNT dataset with annotated appropriateness (See Appendix A)├── proper│   ├── proper.alignment # translations of targeted expressions│   ├── proper.en # references│   ├── proper.ja # source sentences│   └── proper.tsv├── abbrev│   ├── abbrev.alignment│   ├── abbrev.en│   ├── abbrev.norm.ja # normalized source sentences│   ├── abbrev.orig.ja # original source sentences│   └── abbrev.tsv├── colloq│   ├── colloq.alignment│   ├── colloq.en│   ├── colloq.norm.ja│   ├── colloq.orig.ja│   └── colloq.tsv├── variant│   ├── variant.alignment│   ├── variant.en│   ├── variant.norm.ja│   ├── variant.orig.ja│   └── variant.tsv└── src    └── calc_acc.py # script for calculating translation accuracy

Basic statistics and examples from the dataset

- Abbreviated Nounoriginal source : 地味なアプデ (apude, meaning update) だがnormalized source : 地味なアップデート (update) だがreference : That’s a plain update thoughalignment : update- Colloquial Expressionoriginal source : ここまで描いて飽きた、かなちい (kanachii, meaning sad)normalized source : ここまで描いて飽きた、かなしい (kanashii)reference : Drawing this much then getting bored, how sad.alignment : sad

Citation

@inproceedings{fujii-etal-2020-phemt,    title = "{P}he{MT}: A Phenomenon-wise Dataset for Machine Translation Robustness on User-Generated Contents",    author = "Fujii, Ryo  and      Mita, Masato  and      Abe, Kaori  and      Hanawa, Kazuaki  and      Morishita, Makoto  and      Suzuki, Jun  and      Inui, Kentaro",    booktitle = "Proceedings of the 28th International Conference on Computational Linguistics",    month = dec,    year = "2020",    address = "Barcelona, Spain (Online)",    publisher = "International Committee on Computational Linguistics",    url = "https://www.aclweb.org/anthology/2020.coling-main.521",    pages = "5929--5943",}

Reference