Datasets:

ArXiv:
License:
andreaschari commited on
Commit
44675f6
·
verified ·
1 Parent(s): 3e2a405

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +56 -0
README.md ADDED
@@ -0,0 +1,56 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ language:
4
+ - ca
5
+ - nl
6
+ - fr
7
+ - af
8
+ - ga
9
+ - gd
10
+ - ms
11
+ - oc
12
+ ---
13
+ # Dataset Summary
14
+ This is a variation of **mMARCO** by Bonifactio et al. used for the "Improving Low-Resource Retrieval Effectiveness using Zero-Shot Linguistic Similarity Transfer" paper
15
+
16
+ **mMARCO** is a multilingual version of the [MS MARCO passage ranking dataset](https://microsoft.github.io/msmarco/).
17
+ For more information, checkout our papers:
18
+ * [**mMARCO: A Multilingual Version of the MS MARCO Passage Ranking Dataset**](https://arxiv.org/abs/2108.13897)
19
+ * [**A cost-benefit analysis of cross-lingual transfer methods**](https://arxiv.org/abs/2105.06813)
20
+
21
+ # Dataset Structure
22
+
23
+ `queries/` contains both the translated dev queries used for evaluation under `/dev_small` and the translated train/judged queries used for training under `/train/`.
24
+
25
+ The dev queries are in the following languages: Afrikaans, Catalan, Scottish Gaelic, Malay, Occitan, Sicilian, Cantonese.
26
+ the train queries are in the following languages: Afrikaans, Catalan.
27
+
28
+ `train_with_negs` contains the triples (query, positive doc, negative doc) used for training the models our ECIR 2025 IR4GOOD paper.
29
+
30
+ # Licensing Information
31
+
32
+ This dataset is released under [Apache license 2.0](https://www.apache.org/licenses/).
33
+
34
+
35
+ # Citation Information
36
+
37
+ If you use this dataset, please cite:
38
+
39
+ ```
40
+ @article{DBLP:journals/corr/abs-2108-13897,
41
+ author = {Luiz Bonifacio and
42
+ Israel Campiotti and
43
+ Roberto de Alencar Lotufo and
44
+ Rodrigo Frassetto Nogueira},
45
+ title = {mMARCO: {A} Multilingual Version of {MS} {MARCO} Passage Ranking Dataset},
46
+ journal = {CoRR},
47
+ volume = {abs/2108.13897},
48
+ year = {2021},
49
+ url = {https://arxiv.org/abs/2108.13897},
50
+ eprinttype = {arXiv},
51
+ eprint = {2108.13897},
52
+ timestamp = {Mon, 20 Mar 2023 15:35:34 +0100},
53
+ biburl = {https://dblp.org/rec/journals/corr/abs-2108-13897.bib},
54
+ bibsource = {dblp computer science bibliography, https://dblp.org}
55
+ }
56
+ ```