---
annotations_creators:
- expert-generated
language:
- ar
- bn
- en
- es
- fa
- fi
- fr
- hi
- id
- ja
- ko
- ru
- sw
- te
- th
- zh
- de
- yo
multilinguality:
- multilingual
pretty_name: MIRACL-corpus
source_datasets: []
task_categories:
- text-retrieval
license:
- apache-2.0
task_ids:
- document-retrieval
---

# Dataset Card for MIRACL (Topics and Qrels)


## Dataset Description
[Homepage](ttp://miracl.ai) | 
[Repository:](https://github.com/project-miracl/miracl) | 
[Paper](https://direct.mit.edu/tacl/article/doi/10.1162/tacl_a_00595/117438) | 
[ArXiv](https://arxiv.org/abs/2210.09984)

MIRACL 🌍🙌🌏 (Multilingual Information Retrieval Across a Continuum of Languages) is a multilingual retrieval dataset that focuses on search across 18 different languages, which collectively encompass over three billion native speakers around the world.

This dataset contains the collection data of the 16 "known languages". The remaining 2 "surprise languages" will not be released until later.

The topics are generated by native speakers of each language, who also label the relevance between the topics and a given document list.

This repository only contains the topics and qrels of MIRACL. The collection can be found [here](https://huggingface.co/datasets/miracl/miracl-corpus).

## Dataset Structure
1. To download the files:
Under folders `miracl-v1.0-{lang}/topics`,
the topics are saved in `.tsv` format, with each line to be:
```
qid\tquery
```

Under folders `miracl-v1.0-{lang}/qrels`,
the qrels are saved in standard TREC format, with each line to be:
```
qid Q0 docid relevance
```


2. To access the data using HuggingFace `datasets`:
```
lang='ar'  # or any of the 16 languages
miracl = datasets.load_dataset('miracl/miracl', lang, use_auth_token=True)

# training set:
for data in miracl['train']:  # or 'dev', 'testA'
  query_id = data['query_id']
  query = data['query']
  positive_passages = data['positive_passages']
  negative_passages = data['negative_passages']
  
  for entry in positive_passages: # OR 'negative_passages'
    docid = entry['docid']
    title = entry['title']
    text = entry['text']
```
The structure is the same for `train`, `dev`, and `testA` set, where `testA` only exists for languages in Mr. TyDi (i.e., Arabic, Bengali, English, Finnish, Indonesian, Japanese, Korean, Russian, Swahili, Telugu, Thai).
Note that `negative_passages` are annotated by native speakers as well, instead of the non-positive passages from top-`k` retrieval results.


## Dataset Statistics 
The following table contains the number of queries (`#Q`) and the number of judgments (`#J`) in each language, for the training and development set,
where the judgments include both positive and negative samples.

| Lang | Train |        |  Dev  |        |
|:----:|:-----:|:------:|:-----:|:------:|
|      | **#Q**| **#J** |**#Q** |**#J**  |
|  ar  | 3,495 | 25,382 | 2,896 | 29,197 |
|  bn  | 1,631 | 16,754 |   411 |  4,206 |
|  en  | 2,863 | 29,416 |   799 |  8,350 |
|  es  | 2,162 | 21,531 |   648 |  6,443 |
|  fa  | 2,107 | 21,844 |   632 |  6,571 |
|  fi  | 2,897 | 20,350 | 1,271 | 12,008 |
|  fr  | 1,143 | 11,426 |   343 |  3,429 |
|  hi  | 1,169 | 11,668 |   350 |  3,494 |
|  id  | 4,071 | 41,358 |   960 |  9,668 |
|  ja  | 3,477 | 34,387 |   860 |  8,354 |
|  ko  |   868 | 12,767 |   213 |  3,057 |
|  ru  | 4,683 | 33,921 | 1,252 | 13,100 |
|  sw  | 1,901 |  9,359 |   482 |  5,092 |
|  te  | 3,452 | 18,608 |   828 |  1,606 |
|  th  | 2,972 | 21,293 |   733 |  7,573 |
|  zh  | 1,312 | 13,113 |   393 |  3,928 |