Datasets:
MLRS
/

Modalities:
Text
Languages:
Maltese
Libraries:
Datasets
License:
File size: 3,039 Bytes
28269ab
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d23bae7
 
28269ab
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d23bae7
 
6b6614b
8d28244
6b6614b
e9823db
8d28244
6b6614b
d23bae7
 
 
 
 
 
 
 
8d28244
195713f
 
 
 
 
 
 
 
 
 
 
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
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
---
pretty_name: Korpus Malti
languages:
- mt
multilinguality:
- monolingual
size_categories:
- 10M<n<100M
annotations_creators:
- no-annotation
language_creators:
- found
source_datasets:
- original
task_categories:
- text-generation
- fill-mask
task_ids:
- language-modeling
- masked-language-modeling
licenses:
- cc-by-nc-sa-4.0
---
# Korpus Malti 🇲🇹

General Corpora for the Maltese Language.

This dataset is composed of texts from various genres/domains written in Maltese.


## Configurations

### Shuffled data

The default configuration (`"shuffled"`) yields the the entire corpus from all genres:
```python
import datasets

dataset = datasets.load_dataset("MLRS/korpus_malti")
```

All sentences are combined together and shuffled, without preserving the sentence order.
No other annotations are present, so an instance would be of the following form:
```json
{
  "text": "Din hija sentenza."
}
```

The training/validation/testing split is what was used to train the [BERTu](https://huggingface.co/MLRS/BERTu) model.

### Domain-split data

All other configurations contain a subset of the data.
For instance, this loads the Wikipedia portion:
```python
import datasets

dataset = datasets.load_dataset("MLRS/korpus_malti", "wiki")
```

For these configurations the data is not shuffled, so the sentence order on a document level is preserved.
An instance from these configurations would take the following form:
```json
{
  "text": ["Din hija sentenza.", "U hawn oħra!"],
}
```

The raw data files contain additional metadata.
Its structure differs from one instance to another, depending on what's available from the source.
This information was typically scraped from the source itself & minimal processing is performed on such data.


## Additional Information

### Dataset Curators
The dataset was created by [Albert Gatt](https://albertgatt.github.io), Kurt Micallef, [Marc Tanti](https://www.um.edu.mt/profile/marctanti), [Lonneke van der Plas](https://sites.google.com/site/lonnekenlp/) and [Claudia Borg](https://www.um.edu.mt/profile/claudiaborg).

### Licensing Information
This work is licensed under a
[Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License][cc-by-nc-sa].
Permissions beyond the scope of this license may be available at [https://mlrs.research.um.edu.mt/](https://mlrs.research.um.edu.mt/).

[![CC BY-NC-SA 4.0][cc-by-nc-sa-image]][cc-by-nc-sa]

[cc-by-nc-sa]: http://creativecommons.org/licenses/by-nc-sa/4.0/
[cc-by-nc-sa-image]: https://licensebuttons.net/l/by-nc-sa/4.0/88x31.png

### Citation Information
```bibtex
@inproceedings{bertu:deeplo2015,
    Author = {Micallef, Kurt and Gatt, Albert and Tanti, Marc and van der Plas, Lonneke and Borg, Claudia},
    Booktitle = {Proceedings of the 3rd Workshop on Deep Learning for Low Resource NLP (DeepLo)},
    Publisher = {Association for Computational Linguistics},
    Title = {Pre-training Data Quality and Quantity for a Low-Resource Language: New Corpus and BERT Models for Maltese},
    Year = {2022}
}

```