Datasets:

ArXiv:
File size: 3,803 Bytes
99f149a
63de5fa
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ef78ab9
99f149a
63de5fa
 
 
 
 
 
398c1eb
63de5fa
 
 
 
 
 
 
 
 
 
 
398c1eb
63de5fa
 
 
ef78ab9
63de5fa
 
 
ef78ab9
 
 
99f149a
ef78ab9
99f149a
1245e6f
 
99f149a
 
 
 
 
ef78ab9
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
# Dataset Summary

**mMARCO** is a multilingual version of the [MS MARCO passage ranking dataset](https://microsoft.github.io/msmarco/).
For more information, checkout our papers:
  * [**mMARCO: A Multilingual Version of the MS MARCO Passage Ranking Dataset**](https://arxiv.org/abs/2108.13897)
  * [**A cost-benefit analysis of cross-lingual transfer methods**](https://arxiv.org/abs/2105.06813)


There are two translated versions of mMARCO.

  * **v1**
  In v1 version, we use MarianNMT an open-source neural machine translation framework [made available](https://huggingface.co/Helsinki-NLP) by the Language Technology Research Group at the University of Helsinki for more than a thousand language pairs for translation. This version comprises 8 languages: Chinese, French, German, Indonesian, Italian, Portuguese, Russian and Spanish. In the paper, we refer to these models as "Helsinki".

  * **v2 (Recommended)**
  In v2 version, we use Google Translate to translate the dataset. In this commercial translation version, besides the 8 languages from v1, we add other 5 languages: Japanese, Dutch, Vietnamese, Hindi and Arabic.


### Supported languages

| Language name | Language code | v1 | v2 |
|---------------|---------------| ✓  | ✓  |
| English		| english		| ✓  | ✓  |
| Chinese		| chinese		| ✓  | ✓  |
| French		| french		| ✓  | ✓  |
| German		| german		| ✓  | ✓  |
| Indonesian	| indonesian	| ✓  | ✓  |
| Italian		| italian		| ✓  | ✓  |
| Portuguese	| portuguese	| ✓  | ✓  |
| Russian		| russian		| ✓  | ✓  |
| Spanish		| spanish		| ✓  | ✓  |
| Arabic        | arabic        |    | ✓  |
| Dutch         | dutch         |    | ✓  |
| Hindi         | hindi         |    | ✓  |
| Japanese      | japanese      |    | ✓  |
| Vietnamese    | vietnamese    |    | ✓  |


# Dataset Structure

You can load mMARCO dataset by choosing a specific language. We include training triples (query, positive and negative example), the translated collections of documents and queries.


#### Training triples

```
>>> dataset = load_dataset('mmarco', 'english')
>>> dataset['train'][1]
{'query': 'what fruit is native to australia', 'positive': 'Passiflora herbertiana. A rare passion fruit native to Australia. Fruits are green-skinned, white fleshed, with an unknown edible rating. Some sources list the fruit as edible, sweet and tasty, while others list the fruits as being bitter and inedible.assiflora herbertiana. A rare passion fruit native to Australia. Fruits are green-skinned, white fleshed, with an unknown edible rating. Some sources list the fruit as edible, sweet and tasty, while others list the fruits as being bitter and inedible.', 'negative': 'The kola nut is the fruit of the kola tree, a genus (Cola) of trees that are native to the tropical rainforests of Africa.'}
```

#### Queries

```
>>> dataset = load_dataset('mmarco', 'queries-spanish')
>>> dataset['train'][1]
{'id': 634306, 'text': '¿Qué significa Chattel en el historial de crédito'}
```

#### Collection

```
>>> dataset = load_dataset('mmarco', 'collection-portuguese')
>>> dataset['collection'][100]
{'id': 100, 'text': 'Antonín Dvorák (1841-1904) Antonin Dvorak era filho de açougueiro, mas ele não seguiu o negócio de seu pai. Enquanto ajudava seu pai a meio tempo, estudou música e se formou na Escola de Órgãos de Praga em 1859.'}
```


# Citation Information
```
@misc{bonifacio2021mmarco,
      title={mMARCO: A Multilingual Version of MS MARCO Passage Ranking Dataset}, 
      author={Luiz Henrique Bonifacio and Vitor Jeronymo and Hugo Queiroz Abonizio and Israel Campiotti and Marzieh Fadaee and  and Roberto Lotufo and Rodrigo Nogueira},
      year={2021},
      eprint={2108.13897},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}
```