File size: 9,040 Bytes
368cf34
9ee40c6
 
 
 
 
 
 
bfeed7e
 
 
 
 
 
9ee40c6
 
bfeed7e
9ee40c6
 
 
 
bfeed7e
 
9ee40c6
cdfa371
368cf34
9ee40c6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
14183bf
9ee40c6
14183bf
 
 
 
 
9ee40c6
14183bf
9ee40c6
14183bf
9ee40c6
 
 
 
 
 
 
 
 
 
 
 
7994923
 
 
 
 
9ee40c6
 
 
e918566
9ee40c6
14183bf
 
9ee40c6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
45d3804
9ee40c6
14183bf
9ee40c6
0affa61
 
 
 
 
 
 
14183bf
0affa61
14183bf
aab5e45
9ee40c6
14183bf
 
 
 
 
 
 
 
 
 
 
9ee40c6
 
 
cdfa371
 
 
9ee40c6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
14183bf
9ee40c6
 
 
 
 
 
 
 
 
 
 
 
 
14183bf
9ee40c6
 
 
aab5e45
024af4d
47159fb
024af4d
47159fb
024af4d
47159fb
024af4d
9ee40c6
 
 
 
 
97aad7b
 
 
 
 
 
 
 
 
9ee40c6
 
14183bf
e0b8488
cdfa371
9ee40c6
e0b8488
9ee40c6
faf6f3c
a096895
14183bf
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
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
---
language:
- es
- ca
- gl
- eu
multilinguality:
- translation
annotations_creators:
- found
- expert-generated
- machine-generated
language_creators:
- crowdsourced
pretty_name: escagleu-64K
size_categories:
- 10K<n<100K
source_datasets:
- original
task_categories:
- translation
- audio-to-audio
- automatic-speech-recognition
task_ids: []
license: cc-by-4.0
---

# Dataset Card for escagleu-64K corpus

## Table of Contents
- [Dataset Description](#dataset-description)
  - [Dataset Summary](#dataset-summary)
  - [Supported Tasks](#supported-tasks-and-leaderboards)
  - [Languages](#languages)
- [Dataset Structure](#dataset-structure)
  - [Data Splits](#data-instances)
- [Dataset Creation](#dataset-creation)
  - [Source Data](#source-data)
  - [Personal and Sensitive Information](#personal-and-sensitive-information)
- [Considerations for Using the Data](#considerations-for-using-the-data)
  - [Social Impact of Dataset](#social-impact-of-dataset)
  - [Discussion of Biases](#discussion-of-biases)
  - [Other Known Limitations](#other-known-limitations)
- [Additional Information](#additional-information)
  - [Author](#author)
  - [Contact Information](#contact-information)
  - [Copyright](#copyright)
  - [Licensing information](#licenciung-informatrion)
  - [Funding](#funding)

## Dataset Description

### Dataset Summary

This is the second version of escagleu-64k, a parallel corpus containing approximately 64k sentences translated across Spanish, Catalan, Valencian Catalan, Galician, and Basque.

The original sentences are in Spanish and are sourced from the [Spanish Common Voice Corpus](https://github.com/common-voice/common-voice/tree/main/server/data/es).

This corpus was prepared with the goal of creating a parallel speech dataset for these languages using the [Common Voice](https://commonvoice.mozilla.org) platform as part of the [Ilenia](https://proyectoilenia.es/) project.

In the first version of the corpus, translations into Galician and Basque were carried out via machine translation, and several errors were identified in the source sentences.

In this second version, all translations have been reviewed by a professional team, and the source sentences have been corrected. During the manual revision process, 199 sentences were removed. However, to maintain continuity for future reference, we have preserved the original sentence indexing from v1. As a result, some index numbers are now missing.

Version 2 of the corpus includes 63,892 parallel sentences.

### Supported Tasks and Leaderboards

This dataset can be used for training Machine Translation (MT) models, Speech-to-Text translation models, and Speech-to-Speech translation models.

### Languages

This multilingual dataset is in Spanish (es), Catalan (ca), Valencian Catalan (ca-va), Galician (gl) and Basque (eu).

## Dataset Structure
Five separated tsv files are provided with the sentences sorted in the same order:

* escagleu-64k_ca.tsv: contains 63,892 sentences translated into Catalan.
* escagleu-64k_ca_va.tsv: contains 63,892 sentences adapted into Valencian Catalan.
* escagleu-64k_es.tsv: contains 63,892 sentences in Spanish (authentic).
* escagleu-64k_gl.tsv: contains 63,892 sentences translated into Galician using MT and human post-edited. 
* escagleu-64k_es.tsv: contains 63,892 sentences translated into Basque using MT and human post-edited.
    
In addition, a tsv is provided with the sentences aligned in all the languages:

* escagleu-64k.tsv: the sentences in all the languages of the corpus.

Please note that, in relation to v1, 199 sentences have been removed, so the corresponding numbers are missing from the index.


### Data Splits

The dataset contains a single split.

## Dataset Creation

### Curation Rationale

We created this corpus with the goal of establishing a parallel speech dataset among Spanish, Catalan, Galician, and Basque using the Common Voice platform. The resulting dataset will be employed to train Speech-to-Text and Speech-to-Speech translation models.

### Source Data

The original sentences are in Spanish and come from the [Spanish Common Voice Corpus](https://github.com/common-voice/common-voice/tree/main/server/data/es).

#### Initial Data Collection and Normalization

We extracted a set of 223,261 sentences from the [Spanish Common Voice Corpus](https://github.com/common-voice/common-voice/tree/main/server/data/es) v.07.

An automatic filtering process was first applied to the source sentences, removing those that:

*  Were repeated.
*  Exceeded 14 words in length.
*  Ended with a preposition.
*  Were in lowercase but Out of Vocabulary (OOV), utilizing the Hunspell dictionary.
*  Contained repeated words.
*  Featured characters not existing in Spanish or sequences of characters impossible in Spanish.

This filtering resulted in the selection of 64,091 sentences, which constitute v1 of the corpus.

For v1, the selected sentences were translated from Spanish into Catalan by a professional translation company and subsequently adapted into Valencian Catalan by an expert team from the Universitat d'Alacant.
The same corpus of sentences was translated into Galician using [NOS-MT-OpenNMT-es-gl](https://huggingface.co/proxectonos/Nos_MT-OpenNMT-es-gl) and into Basque using [itzuli](https://www.euskadi.eus/itzuli/).

In this second version, the following improvements have been performed:

* Revision of the Galician and Basque translations by specialized companies.
* Revision of the original Spanish text and its Catalan (general and Valencian) translations by our specialist team.
* Removal of ambiguous sentences.
* A general review to ensure consistency across all language versions.

During this manual revision process, 199 sentences were removed. However, to maintain continuity for future reference, we have preserved the original sentence indexing from v1. As a result, some index numbers are now missing.

Version 2 of the corpus includes 63,892 parallel sentences.


#### Who are the source language producers?

The [Common Voice](https://commonvoice.mozilla.org) is a corpus designed for speech-related tasks. 
The recorded sentences are sourced from diverse origins. 
For more information, please refer to the [project repository](https://github.com/common-voice).

### Annotations

This corpus doesn't have annotations.

#### Annotation process

N/A

#### Who are the annotators?

N/A

### Personal and Sensitive Information

The original sentences are in Spanish and come from the [Spanish Common Voice Corpus](https://github.com/common-voice/common-voice/tree/main/server/data/es). The sentences have been manually reviewed and no personal information has been detected. 

## Considerations for Using the Data

### Social Impact of Dataset

We expect that this corpus will contribute to the development of speech technologies in the targeted languages.

### Discussion of Biases

We are aware that some of the sentences in the corpus may convey stereotypes. Nonetheless, we have not applied any steps to reduce their impact.

### Other Known Limitations

The translations made automatically in v1 have been manually reviewed. Likewise, errors detected in the source sentences and their translations have been corrected. Finally, a final review has been carried out to ensure consistency between the languages.

## Additional Information

### Authors
[Grupo de Procesamiento del Lenguaje Natural y Sistemas de Información, Centro de Inteligencia Digital (CENID, Universidad de Alicante)](https://cenid.es/),

[HiTZ Center - Aholab, University of the Basque Country UPV/EHU](https://www.hitz.eus/), 

[Language Technologies Unit (LangTech) at the Barcelona Supercomputing Center (BSC)](https://www.bsc.es/discover-bsc/organisation/research-departments/language-technologies-unit),

[Proxecto Nós (Universidade de Santiago de Compostela)](https://nos.gal/).

### Contact information
For further information, please send an email to langtech@bsc.es.

### Copyright

[Grupo de Procesamiento del Lenguaje Natural y Sistemas de Información, Centro de Inteligencia Digital (CENID, Universidad de Alicante)](https://cenid.es/), (2023).

[HiTZ Center - Aholab, University of the Basque Country UPV/EHU](https://www.hitz.eus/),  (2023).

[Language Technologies Unit (LangTech) at the Barcelona Supercomputing Center (BSC)](https://www.bsc.es/discover-bsc/organisation/research-departments/language-technologies-unit), (2023).

[Proxecto Nós (Universidade de Santiago de Compostela)](https://nos.gal/),  (2023).


### Licensing information

This dataset can be used for any purpose, whether academic or commercial, under the terms of the [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). 
Give appropriate credit, provide a link to the license, and indicate if changes were made.


### Funding

This work is funded by the Ministerio para la Transformación Digital y de la Función Pública - Funded by EU – NextGenerationEU within the framework of the [project ILENIA](https://proyectoilenia.es/) with reference 2022/TL22/00215337