metadata

language:
  - Arabic
  - Bengali
  - English
  - Finnish
  - Indonesian
  - Finnish
  - Korean
  - Russian
  - Swahili
  - Telugu
  - Thai
  - Japanese
license: Apache License 2.0

Dataset Summary

Mr. TyDi is a multi-lingual benchmark dataset built on TyDi, covering eleven typologically diverse languages. It is designed for monolingual retrieval, specifically to evaluate ranking with learned dense representations.

Dataset Structure

The only configuration here is the language, For each language, there are three splits: train, dev, and test. The negative examples from training set are sampled from the top-30 BM25 runfiles on each language. Specifically, we combine the training data for all languages under the combined configuration.

An example of train set looks as follows:

{
  'query_id': '1', 
  'query': 'When was quantum field theory developed?', 
  'positive_passages': [
    {
      'docid': '25267#12', 
      'title': 'Quantum field theory', 
      'text': 'Quantum field theory naturally began with the study of electromagnetic interactions, as the electromagnetic field was the only known classical field as of the 1920s.'
    },
    ...
    ]
  'negative_passages': [
    {
      'docid': '346489#8', 
      'title': 'Local quantum field theory', 
      'text': 'More recently, the approach has been further implemented to include an algebraic version of quantum field ...'
    },
    ...
  ],
}

An example of dev and test set looks as follows. We only provide the docid of positive passages here to save the space. Also no candidate passages are provided at this point. Note that to perform the retrieval, it need to be used together with castorini/mr-tydi-corpus

{
  'query_id': '0', 
  'query': 'Is Creole a pidgin of French?', 
  'positive_passages': [
    {
      'docid': '3716905#1',
      'title': '', 
      'text': ''
    },
    ...
   ]
}

Citation Information

@article{mrtydi,
      title={{Mr. TyDi}: A Multi-lingual Benchmark for Dense Retrieval}, 
      author={Xinyu Zhang and Xueguang Ma and Peng Shi and Jimmy Lin},
      year={2021},
      journal={arXiv:2108.08787},
}