MTData

MTData tool automate the dataset collection and preparation for machine translation research. It provides CLI and python APIs, so it can be used as a standalone tool or embedded into python apps for preparing MT experiments.

This tool knows:

• From where to download data sets: WMT tests and devs for [2014, 2015, ... 2020], Paracrawl, Europarl, News Commentary, WikiTitles, Tilde Model corpus ...
• How to extract files : .tar, .tar.gz, .tgz, .zip, ...
• How to parse .tmx, .sgm and such XMLs, or .tsv ... Checks if they have same number of segments.
• Whether parallel data is in one .tsv file or two sgm files.
• Whether data is compressed in gz, xz or none at all.
• Whether the source-target is in the same order or is it swapped as target-source order.
• How to map code to ISO language codes! Using ISO 639_3 that has space for 7000+ languages of our planet.
• (And more of such tiny details over the time.)

MTData is here to:

• Automate the MT training data creation by taking out human intervention. Inspired by SacreBLEU that takes out human intervention in evaluation stage.
• A reusable tool instead of dozens of use-once shell scripts spread across multiple repos.

Limitations (as of now):

• Only publicly available datasets that do not need login are supported. No LDC yet.
• No tokenizers are integrated. (It should be fairly easy to get those integrated)

Installation

# from the source code on github 
git clone https://github.com/thammegowda/mtdata 
cd mtdata
pip install --editable .

# from pypi ; do not do this yet, since the code is evolving faster than releases
pip install mtdata  

Current Status:

These are the summary of datasets from various sources (Updated: May 10 2020). The list is incomplete and meant to see as start. Here I (/TG) have picked some commonly used datasets that I use for my work, you are welcome to add more.

^$1 - Currently excluded JW300 from Opus due to parsing issues. JW300 source alone contains 46,000+ bitexts.

CLI Usage

• After pip installation, the CLI can be called using mtdata command or python -m mtdata
• There are two sub commands: list for listing the datasets, and get for getting them

mtdata list

Lists datasets that are known to this tool.

mtdata list -h
usage: mtdata list [-h] [-l L1-L2] [-n [NAME [NAME ...]]]
                   [-nn [NAME [NAME ...]]] [-f]

optional arguments:
  -h, --help            show this help message and exit
  -l L1-L2, --langs L1-L2
                        Language pairs; e.g.: deu-eng (default: None)
  -n [NAME [NAME ...]], --names [NAME [NAME ...]]
                        Name of dataset set; eg europarl_v9. (default: None)
  -nn [NAME [NAME ...]], --not-names [NAME [NAME ...]]
                        Exclude these names (default: None)
  -f, --full            Show Full Citation (default: False)

# List everything
mtdata list

# List a lang pair 
mtdata list -l deu-eng

# List a dataset by name(s)
mtdata list -n europarl_v9
mtdata list -n europarl_v9 news_commentary_v14

# list by both language pair and dataset name
mtdata list -l deu-eng -n europarl_v9 news_commentary_v14 newstest201{4,5,6,7,8,9}_deen

# get citation of a dataset (if available in index.py)
mtdata list -l deu-eng -n newstest2019_deen --full

mtdata get

This command downloads datasets specified by names for languages to a directory. You will have to make definite choice for --train and --test arguments

mtdata get -h
usage: mtdata get [-h] -l L1-L2 [-tr [NAME [NAME ...]]]
                  [-tt [NAME [NAME ...]]] -o OUT

optional arguments:
  -h, --help            show this help message and exit
  -l L1-L2, --langs L1-L2
                        Language pairs; e.g.: deu-eng (default: None)
  -tr [NAME [NAME ...]], --train [NAME [NAME ...]]
                        Names of datasets separated by space, to be used for *training*.
                          e.g. -tr news_commentary_v14 europarl_v9 .
                          All these datasets gets concatenated into one big file.
                           (default: None)
  -tt [NAME [NAME ...]], --test [NAME [NAME ...]]
                        Names of datasets separated by space, to be used for *testing*.
                          e.g. "-tt newstest2018_deen newstest2019_deen".
                        You may also use shell expansion if your shell supports it.
                          e.g. "-tt newstest201{8,9}_deen."  (default: None)
  -o OUT, --out OUT     Output directory name (default: None)

Example

See what datasets are available for deu-eng

$ mtdata list -l deu-eng  # see available datasets
    europarl_v9	deu-eng	http://www.statmt.org/europarl/v9/training/europarl-v9.deu-eng.tsv.gz
    news_commentary_v14	deu-eng	http://data.statmt.org/news-commentary/v14/training/news-commentary-v14.deu-eng.tsv.gz
    wiki_titles_v1	deu-eng	http://data.statmt.org/wikititles/v1/wikititles-v1.deu-eng.tsv.gz
    wiki_titles_v2	deu-eng	http://data.statmt.org/wikititles/v2/wikititles-v2.deu-eng.tsv.gz
    newstest2014_deen	deu-eng	http://data.statmt.org/wmt20/translation-task/dev.tgz	dev/newstest2014-deen-src.de.sgm,dev/newstest2014-deen-ref.en.sgm
    newstest2015_ende	en-de	http://data.statmt.org/wmt20/translation-task/dev.tgz	dev/newstest2015-ende-src.en.sgm,dev/newstest2015-ende-ref.de.sgm
    newstest2015_deen	deu-eng	http://data.statmt.org/wmt20/translation-task/dev.tgz	dev/newstest2015-deen-src.de.sgm,dev/newstest2015-deen-ref.en.sgm
    ...[truncated]

Get these datasets and store under dir deu-eng

$ mtdata get --langs deu-eng --train europarl_v10 wmt13_commoncrawl news_commentary_v14 --test newstest201{4,5,6,7,8,9}_deen --out deu-eng
    # ...[truncated]   
    INFO:root:Train stats:
    {
      "total": 4565929,
      "parts": {
        "wmt13_commoncrawl": 2399123,
        "news_commentary_v14": 338285,
        "europarl_v10": 1828521
      }
    }
    INFO:root:Dataset is ready at deu-eng

To reproduce this dataset again in the future or by others, please refer to <out-dir>>/mtdata.signature.txt:

$ cat deu-eng/mtdata.signature.txt
mtdat get -l deu-eng -tr europarl_v10 wmt13_commoncrawl news_commentary_v14 -ts newstest2014_deen newstest2015_deen newstest2016_deen newstest2017_deen newstest2018_deen newstest2019_deen -o <out-dir>
mtdata version 0.1.1

See what the above command has accomplished:

$ find  deu-eng -type f | sort  | xargs wc -l
    3003 deu-eng/tests/newstest2014_deen.deu
    3003 deu-eng/tests/newstest2014_deen.eng
    2169 deu-eng/tests/newstest2015_deen.deu
    2169 deu-eng/tests/newstest2015_deen.eng
    2999 deu-eng/tests/newstest2016_deen.deu
    2999 deu-eng/tests/newstest2016_deen.eng
    3004 deu-eng/tests/newstest2017_deen.deu
    3004 deu-eng/tests/newstest2017_deen.eng
    2998 deu-eng/tests/newstest2018_deen.deu
    2998 deu-eng/tests/newstest2018_deen.eng
    2000 deu-eng/tests/newstest2019_deen.deu
    2000 deu-eng/tests/newstest2019_deen.eng
 1828521 deu-eng/train-parts/europarl_v10.deu
 1828521 deu-eng/train-parts/europarl_v10.eng
  338285 deu-eng/train-parts/news_commentary_v14.deu
  338285 deu-eng/train-parts/news_commentary_v14.eng
 2399123 deu-eng/train-parts/wmt13_commoncrawl.deu
 2399123 deu-eng/train-parts/wmt13_commoncrawl.eng
 4565929 deu-eng/train.deu
 4565929 deu-eng/train.eng

ISO 639 3

Internally all language codes are mapped to ISO-639 3 codes. The mapping can be inspected with python -m mtdata.iso

$ python -m mtdata.iso -h
usage: python -m mtdata.iso [-h] [langs [langs ...]]

ISO 639-3 lookup tool

positional arguments:
  langs       Language code or name that needs to be looked up. When no
              language code is given, all languages are listed.

optional arguments:
  -h, --help  show this help message and exit

# list all 7000+ languages and their 3 letter codes
$ python -m mtdata.iso 
...

# lookup codes for some languages
$ python -m mtdata.iso ka kn en de xx english german
Input   ISO639_3        Name
ka      kat     Georgian
kn      kan     Kannada
en      eng     English
de      deu     German
xx      -none-  -none-
english eng     English
german  deu     German

How to extend, modify, or contribute:

Please help grow the datasets by adding missing+new datasets to index module. Here is an example listing europarl-v9 corpus. Note: the language codes such as de en etc will be mapped to 3 letter ISO codes deu eng internally

from mtdata.index import INDEX as index, Entry
EUROPARL_v9 = 'http://www.statmt.org/europarl/v9/training/europarl-v9.%s-%s.tsv.gz'
for pair in ['de en', 'cs en', 'cs pl', 'es pt', 'fi en', 'lt en']:
    l1, l2 = pair.split()
    index.add_entry(Entry(langs=(l1, l2), name='europarl_v9', url=EUROPARL_v9 % (l1, l2)))

If a datset is inside an archive such as zip or tar

from mtdata.index import INDEX as index, Entry
wmt_sets = {
    'newstest2014': [('de', 'en'), ('cs', 'en'), ('fr', 'en'), ('ru', 'en'), ('hi', 'en')],
    'newsdev2015': [('fi', 'en'), ('en', 'fi')]
}
for set_name, pairs in wmt_sets.items():
    for l1, l2 in pairs:
        src = f'dev/{set_name}-{l1}{l2}-src.{l1}.sgm'
        ref = f'dev/{set_name}-{l1}{l2}-ref.{l2}.sgm'
        name = f'{set_name}_{l1}{l2}'
        index.add_entry(Entry((l1, l2), name=name, filename='wmt20dev.tgz', in_paths=[src, ref],
                             url='http://data.statmt.org/wmt20/translation-task/dev.tgz'))
# filename='wmt20dev.tgz' -- is manually set, because url has dev.gz that can be confusing
# in_paths=[src, ref]  -- listing two sgm files inside the tarball
# in_ext='sgm' will be auto detected fropm path. set in_ext='txt' to explicitly set format as plain text 

Refer to paracrawl, tilde, or statmt for examples.

If citation is available for a dataset, please incl.deu

cite = r"""bib tex here""
Entry(... cite=cite)

For adding a custom parser, or file handler look into parser.read_segs() and cache for dealing with a new archive/file type that is not already supported.

Developers:

• Thamme Gowda