huggingface / huggingface/datasets
Missing multilingual directions in IWSLT2017 dataset's processing script
- Dominant language
- Python
- Stars
- 22k
- Forks
- 3.4k
- Avg merge
- 5d 7h
- Merged PRs (30d)
- 17
Description
### Describe the bug
Hi,
Upon using `iwslt2017.py` in `IWSLT/iwslt2017` on the Hub for loading the datasets, I am unable to obtain the datasets for the language pairs `de-it`, `de-ro`, `de-nl`, `it-de`, `nl-de`, and `ro-de` using it. These 6 pairs do not show up when using `get_dataset_config_names()` to obtain the list of all the configs present in `IWSLT/iwslt2017`. This should not be the case since as mentioned in their original paper (please see https://aclanthology.org/2017.iwslt-1.1.pdf), the authors specify that "_this year we proposed the multilingual translation between any pair of languages from {Dutch, English, German, Italian, Romanian}..._" and because these datasets are indeed present in `data/2017-01-trnmted/texts/DeEnItNlRo/DeEnItNlRo/DeEnItNlRo-DeEnItNlRo.zip`.
Best Regards,
Anand
### Steps to reproduce the bug
Check the output of `get_dataset_config_names("IWSLT/iwslt2017", trust_remote_code=True)`: only 24 language pairs are present and the following 6 config names are absent: `iwslt2017-de-it`, `iwslt2017-de-ro`, `iwslt2017-de-nl`, `iwslt2017-it-de`, `iwslt2017-nl-de`, and `iwslt2017-ro-de`.
### Expected behavior
The aforementioned 6 language pairs should also be present and hence, all these 6 language pairs' IWSLT2017 datasets must also be available for further use.
I would suggest removing `de` from the `BI_LANGUAGES` list and moving it over to the `MULTI_LANGUAGES` list instead in `iwslt2017.py` to account for all the 6 missing language pairs (the same `de-en` dataset is present in both `data/2017-01-trnmted/texts/DeEnItNlRo/DeEnItNlRo/DeEnItNlRo-DeEnItNlRo.zip` and `data/2017-01-trnted/texts/de/en/de-en.zip` but the `de-ro`, `de-nl`, `it-de`, `nl-de`, and `ro-de` datasets are only present in `data/2017-01-trnmted/texts/DeEnItNlRo/DeEnItNlRo/DeEnItNlRo-DeEnItNlRo.zip`: so, its unclear why the following comment: _`# XXX: Artificially removed DE from here, as it also exists within bilingual data`_ has been added as `L71` in `iwslt2017.py`). The `README.md` file in `IWSLT/iwslt2017`must then be re-created using `datasets-cli test path/to/iwslt2017.py --save_info --all_configs` to pass all split size verification checks for the 6 new language pairs which were previously non-existent.
### Environment info
- `datasets` version: 3.5.0
- Platform: Linux-6.8.0-56-generic-x86_64-with-glibc2.39
- Python version: 3.12.3
- `huggingface_hub` version: 0.30.1
- PyArrow version: 19.0.1
- Pandas version: 2.2.3
- `fsspec` version: 2024.12.0
Contributor guide
Assessment
This issue has not been assessed yet.