Datasets:
Updated description of the dataset format and content
Browse files
README.md
CHANGED
@@ -22,7 +22,7 @@ license: cc-by-nc-sa-4.0
|
|
22 |
|
23 |
### Dataset Summary
|
24 |
|
25 |
-
The CA-PT Parallel Corpus is a Catalan-Portuguese dataset of
|
26 |
Machine Translation.
|
27 |
|
28 |
|
@@ -39,11 +39,9 @@ The sentences included in the dataset are in Catalan (CA) and Portuguese (PT).
|
|
39 |
|
40 |
### Data Instances
|
41 |
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
- ca-pt_2023_09_01_full.pt: contains 9.892.953 Portuguese sentences.
|
47 |
|
48 |
|
49 |
### Data Fields
|
@@ -63,29 +61,19 @@ This dataset is aimed at promoting the development of Machine Translation betwee
|
|
63 |
### Source Data
|
64 |
|
65 |
#### Initial Data Collection and Normalization
|
|
|
|
|
|
|
66 |
|
67 |
-
|
68 |
-
|
69 |
-
| Dataset | Sentences |
|
70 |
-
|:-------|-------:|
|
71 |
-
| CCMatrix v1 | 3.765.459 |
|
72 |
-
| WikiMatrix | 317.649 |
|
73 |
-
| GNOME | 1.752 |
|
74 |
-
| KDE4 | 117.828 |
|
75 |
-
| OpenSubtitles | 235.604 |
|
76 |
-
| GlobalVoices | 3.430 |
|
77 |
-
| Tatoeba | 723 |
|
78 |
-
| Europarl | 1.631.989 |
|
79 |
|
80 |
-
|
81 |
-
The Europarl corpus is a synthetic parallel corpus created from the original Spanish-Catalan corpus by [SoftCatalà](https://github.com/Softcatala/Europarl-catalan).
|
82 |
-
|
83 |
-
The remaining **3.733.322** sentences are synthetic parallel data created from a random sampling of the Spanish-Portuguese corpora
|
84 |
available on [Opus](https://opus.nlpl.eu/) and translated into Catalan using the [PlanTL es-ca](https://huggingface.co/PlanTL-GOB-ES/mt-plantl-es-ca) model.
|
85 |
|
86 |
All datasets are deduplicated and filtered to remove any sentence pairs with a cosine similarity of less than 0.75.
|
87 |
This is done using sentence embeddings calculated using [LaBSE](https://huggingface.co/sentence-transformers/LaBSE).
|
88 |
-
The filtered datasets are then concatenated to form
|
89 |
|
90 |
#### Who are the source language producers?
|
91 |
|
@@ -102,7 +90,7 @@ The dataset does not contain any annotations.
|
|
102 |
#### Who are the annotators?
|
103 |
|
104 |
[N/A]
|
105 |
-
|
106 |
### Personal and Sensitive Information
|
107 |
|
108 |
Given that this dataset is partly derived from pre-existing datasets that may contain crawled data, and that no specific anonymisation process has been applied,
|
|
|
22 |
|
23 |
### Dataset Summary
|
24 |
|
25 |
+
The CA-PT Parallel Corpus is a Catalan-Portuguese dataset of created to support Catalan in NLP tasks, specifically
|
26 |
Machine Translation.
|
27 |
|
28 |
|
|
|
39 |
|
40 |
### Data Instances
|
41 |
|
42 |
+
A single tsv file is provided with the sentences sorted in the same order and
|
43 |
+
a header containing the two-letter ISO language code for the language in each column:
|
44 |
+
ca-pt_2023_09_01_full.tsv.
|
|
|
|
|
45 |
|
46 |
|
47 |
### Data Fields
|
|
|
61 |
### Source Data
|
62 |
|
63 |
#### Initial Data Collection and Normalization
|
64 |
+
@alogin4 dataset_to_hf]$ /bin/python /gpfs/scratch/bsc88/bsc088063/dataset_to_hf/txt2parq
|
65 |
+
The first portion of the corpus is a combination of the following original datasets collected from [Opus](https://opus.nlpl.eu/):
|
66 |
+
CCMatrix, WikiMatrix, GNOME, KDE4, OpenSubtitles, GlobalVoices, Tatoeba.
|
67 |
|
68 |
+
Additionally, the corpus contains synthetic parallel data generated from the original Spanish-Catalan Europarl
|
69 |
+
made public by [SoftCatalà](https://github.com/Softcatala/Europarl-catalan).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
70 |
|
71 |
+
A last portion of the dataset is composed by synthetic parallel data generated from a random sampling of the Spanish-Portuguese corpora
|
|
|
|
|
|
|
72 |
available on [Opus](https://opus.nlpl.eu/) and translated into Catalan using the [PlanTL es-ca](https://huggingface.co/PlanTL-GOB-ES/mt-plantl-es-ca) model.
|
73 |
|
74 |
All datasets are deduplicated and filtered to remove any sentence pairs with a cosine similarity of less than 0.75.
|
75 |
This is done using sentence embeddings calculated using [LaBSE](https://huggingface.co/sentence-transformers/LaBSE).
|
76 |
+
The filtered datasets are then concatenated to form the final corpus.
|
77 |
|
78 |
#### Who are the source language producers?
|
79 |
|
|
|
90 |
#### Who are the annotators?
|
91 |
|
92 |
[N/A]
|
93 |
+
@alogin4 dataset_to_hf]$ /bin/python /gpfs/scratch/bsc88/bsc088063/dataset_to_hf/txt2parq
|
94 |
### Personal and Sensitive Information
|
95 |
|
96 |
Given that this dataset is partly derived from pre-existing datasets that may contain crawled data, and that no specific anonymisation process has been applied,
|