metadata

task_categories:
  - translation
language:
  - en
  - fr
  - ja
multilinguality:
  - multilingual
pretty_name: MTNT
configs:
  - config_name: Parallel_EN_FR
    data_files:
      - split: train
        path: train/train.en-fr.tsv
      - split: validation
        path: valid/valid.en-fr.tsv
      - split: test
        path: test/test.en-fr.tsv
    sep: "\t"
  - config_name: Parallel_EN_JA
    data_files:
      - split: train
        path: train/train.en-ja.tsv
      - split: validation
        path: valid/valid.en-ja.tsv
      - split: test
        path: test/test.en-ja.tsv
    sep: "\t"
  - config_name: Parallel_FR_EN
    data_files:
      - split: train
        path: train/train.fr-en.tsv
      - split: validation
        path: valid/valid.fr-en.tsv
      - split: test
        path: test/test.fr-en.tsv
    sep: "\t"
  - config_name: Parallel_JA_EN
    data_files:
      - split: train
        path: train/train.ja-en.tsv
      - split: validation
        path: valid/valid.ja-en.tsv
      - split: test
        path: test/test.ja-en.tsv
    sep: "\t"
  - config_name: Monolingual_EN
    data_files:
      - split: train
        path: monolingual/train.en
      - split: validation
        path: monolingual/dev.en
      - split: test
        path: monolingual/test.en
  - config_name: Monolingual_FR
    data_files:
      - split: train
        path: monolingual/train.fr
      - split: validation
        path: monolingual/dev.fr
      - split: test
        path: monolingual/test.fr
  - config_name: Monolingual_JA
    data_files:
      - split: train
        path: monolingual/train.ja
      - split: validation
        path: monolingual/dev.ja
      - split: test
        path: monolingual/test.ja

Dataset origin: https://github.com/pmichel31415/mtnt

MTNT

MTNT is a collection of comments from the Reddit discussion website in English, French and Japanese, translated to and from English. The particularity of this dataset is that the data consists of "noisy" text, that exhibits typos, grammar errors, code switching and more. For more details, check out the paper.

Data

This folder should have the following structure:

MTNT
├ monolingual
│   ├ dev.en
│   ├ dev.fr
│   ├ dev.ja
│   ├ dev.tok.en
│   ├ dev.tok.fr
│   ├ dev.tok.ja
│   ├ train.en
│   ├ train.fr
│   ├ train.ja
│   ├ train.tok.en
│   ├ train.tok.fr
│   └ train.tok.ja
├ test
│   ├ test.en-fr.tsv
│   ├ test.en-ja.tsv
│   ├ test.fr-en.tsv
│   └ test.ja-en.tsv
├ train
│   ├ train.en-fr.tsv
│   ├ train.en-ja.tsv
│   ├ train.fr-en.tsv
│   └ train.ja-en.tsv
└ valid
    ├ valid.en-fr.tsv
    ├ valid.en-ja.tsv
    ├ valid.fr-en.tsv
    └ valid.ja-en.tsv

The monolingual data is distributed with and without tokenization, in raw text format. The parallel data is split into training, validation and test set. Each tsv file contains 3 columns:

Comment ID
Source sentence
Target sentence

Some source sentences are from a same original comment, and you can use the comment ID to group them together and leverage the contextual information.

If you're only interested in the source and target sentence, you can run the split_tsv.sh script to split the files into source and target files.

Code

The code to reproduce the collection process and the Machine Translation experiments is available on github.

Citing

If you use this dataset or the associated code, please cite:

@InProceedings{michel2018mtnt,
  author    = {Michel, Paul  and  Neubig, Graham},
  title     = {MTNT: A Testbed for Machine Translation of Noisy Text},
  booktitle = {Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing}
}

Contact

If you have any issue with the data, please contact pmichel1[at]cs.cmu.edu. For any question regarding the code, please open an issue on Github.

License

This data is released under the terms of the Reddit API.