Making a MIRACL: Multilingual Information Retrieval Across a Continuum of Languages
Paper
•
2210.09984
•
Published
•
2
query-id
string | corpus-id
string | score
int64 |
|---|---|---|
0
|
151236#1
| 1
|
0
|
151236#11
| 1
|
0
|
151236#0
| 0
|
0
|
3333928#1
| 0
|
0
|
969813#11
| 0
|
0
|
149159#11
| 0
|
0
|
3924773#4
| 0
|
0
|
3562274#11
| 0
|
0
|
3565056#5
| 0
|
0
|
3645940#4
| 0
|
0
|
151236#8
| 0
|
0
|
151236#10
| 0
|
29
|
233622#0
| 1
|
29
|
361673#0
| 1
|
29
|
817516#0
| 0
|
29
|
5111285#0
| 0
|
29
|
817516#1
| 0
|
29
|
1768231#0
| 0
|
29
|
441065#0
| 0
|
29
|
296569#0
| 0
|
29
|
2696053#0
| 0
|
29
|
233622#5
| 0
|
31
|
3966510#0
| 1
|
31
|
316572#11
| 0
|
31
|
316572#4
| 0
|
31
|
316572#9
| 0
|
31
|
316572#8
| 0
|
31
|
316572#5
| 0
|
31
|
316572#14
| 0
|
31
|
118473#0
| 0
|
31
|
79182#23
| 0
|
31
|
3745440#0
| 0
|
32
|
1803#4
| 1
|
32
|
1803#0
| 1
|
32
|
1803#3
| 0
|
32
|
1803#6
| 0
|
32
|
2968024#1
| 0
|
32
|
1560307#9
| 0
|
32
|
334972#0
| 0
|
32
|
2180742#0
| 0
|
32
|
527857#0
| 0
|
32
|
3061087#1
| 0
|
32
|
513431#1
| 0
|
34
|
48533#16
| 1
|
34
|
48533#0
| 1
|
34
|
48533#2
| 0
|
34
|
543393#1
| 0
|
34
|
44375#67
| 0
|
34
|
48533#3
| 0
|
34
|
1253821#1
| 0
|
34
|
83885#3
| 0
|
34
|
553887#1
| 0
|
34
|
48533#6
| 0
|
35
|
2902421#1
| 1
|
35
|
29072#6
| 0
|
35
|
1890060#7
| 0
|
35
|
3901933#1
| 0
|
35
|
2108916#0
| 0
|
35
|
4422993#7
| 0
|
35
|
94626#3
| 0
|
35
|
605042#2
| 0
|
35
|
4600292#8
| 0
|
35
|
2433431#3
| 0
|
57
|
4237722#0
| 1
|
57
|
2495134#4
| 1
|
57
|
1547903#7
| 1
|
57
|
1547903#8
| 0
|
57
|
1547903#0
| 0
|
57
|
494425#0
| 0
|
57
|
1099345#74
| 0
|
57
|
1180098#0
| 0
|
57
|
1301999#1
| 0
|
57
|
16703#27
| 0
|
60
|
517434#0
| 1
|
60
|
3355342#0
| 0
|
60
|
3356141#0
| 0
|
60
|
443019#0
| 0
|
60
|
4116733#0
| 0
|
60
|
517434#1
| 0
|
60
|
4848434#3
| 0
|
60
|
3143634#0
| 0
|
60
|
3926422#0
| 0
|
60
|
517434#9
| 0
|
77
|
908725#0
| 1
|
77
|
3696639#0
| 0
|
77
|
313679#0
| 0
|
77
|
3898107#0
| 0
|
77
|
1343#55
| 0
|
77
|
14942#17
| 0
|
77
|
317431#9
| 0
|
77
|
453772#13
| 0
|
77
|
2796682#0
| 0
|
77
|
3697569#0
| 0
|
77
|
317431#0
| 0
|
88
|
887694#6
| 1
|
88
|
1048463#9
| 1
|
88
|
2652154#8
| 0
|
88
|
1565130#2
| 0
|
88
|
35421#6
| 0
|
88
|
3886#9
| 0
|
This is a Parquet-converted version of mteb/miracl-hard-negatives, compatible with the latest HuggingFace datasets library (4.0+).
The original mteb/miracl-hard-negatives uses a Python script-based loader, which is no longer supported in datasets >= 4.0.0. This dataset provides the same data in standard Parquet format.
MIRACL (Multilingual Information Retrieval Across a Continuum of Languages) is a multilingual retrieval dataset that focuses on search across 18 different languages.
The hard negatives version was created by pooling the top 250 documents per query from:
This makes the retrieval task more challenging compared to the standard MIRACL dataset.
| Code | Language |
|---|---|
| ar | Arabic |
| de | German |
| en | English |
| es | Spanish |
| fa | Persian |
| fi | Finnish |
| fr | French |
| hi | Hindi |
| id | Indonesian |
| ja | Japanese |
| ko | Korean |
| ru | Russian |
| te | Telugu |
| th | Thai |
| zh | Chinese |
from datasets import load_dataset
# Load English data (original config naming convention)
corpus = load_dataset("datalama/miracl-hard-negatives", "corpus-en", split="corpus")
queries = load_dataset("datalama/miracl-hard-negatives", "queries-en", split="queries")
qrels = load_dataset("datalama/miracl-hard-negatives", "en", split="dev")
print(f"Corpus: {len(corpus)} documents")
print(f"Queries: {len(queries)} queries")
print(f"Qrels: {len(qrels)} relevance judgments")
queries-{lang})
| Column | Type | Description |
|---|---|---|
_id |
string | Query ID |
text |
string | Query text |
corpus-{lang})
| Column | Type | Description |
|---|---|---|
_id |
string | Document ID |
title |
string | Document title |
text |
string | Document text |
{lang})
| Column | Type | Description |
|---|---|---|
query-id |
string | Query ID |
corpus-id |
string | Document ID |
score |
int | Relevance score |
@article{zhang2022miracl,
title={MIRACL: A Multilingual Retrieval Dataset Covering 18 Diverse Languages},
author={Zhang, Xinyu and Thakur, Nandan and Ogundepo, Odunayo and Kamalloo, Ehsan and Alfonso-Hermelo, David and Li, Xiaoguang and Liu, Qun and Rezagholizadeh, Mehdi and Lin, Jimmy},
journal={arXiv preprint arXiv:2210.09984},
year={2022}
}
Apache 2.0 (same as the original dataset)