Datasets:
Tasks:
Summarization
Sub-tasks:
news-articles-summarization
Languages:
Galician
Size:
10K<n<100K
License:
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,65 +1,108 @@
|
|
| 1 |
---
|
|
|
|
|
|
|
|
|
|
| 2 |
license: cc-by-4.0
|
| 3 |
task_categories:
|
| 4 |
-
- summarization
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
- gl
|
| 8 |
tags:
|
| 9 |
-
-
|
| 10 |
-
-
|
| 11 |
-
-
|
| 12 |
-
-
|
|
|
|
|
|
|
| 13 |
size_categories:
|
| 14 |
-
- 10K<n<100K
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
---
|
| 16 |
-
# Dataset Card for summarization_gl
|
| 17 |
|
| 18 |
-
|
| 19 |
|
| 20 |
-
|
| 21 |
-
It has a total of 80.829 items. Each item contains an id, a summary and its corresponding text, see below for more details.
|
| 22 |
|
|
|
|
| 23 |
|
| 24 |
-
- **
|
| 25 |
-
- **
|
| 26 |
-
- **
|
| 27 |
|
| 28 |
-
|
| 29 |
|
| 30 |
-
|
| 31 |
|
| 32 |
-
|
| 33 |
|
| 34 |
-
|
|
|
|
| 35 |
|
| 36 |
-
|
| 37 |
|
| 38 |
-
## Dataset Structure
|
| 39 |
|
| 40 |
-
|
| 41 |
|
| 42 |
-
The dataset is
|
| 43 |
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
|
|
|
|
| 48 |
|
| 49 |
-
|
| 50 |
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
|
| 55 |
-
Example
|
| 56 |
|
| 57 |
-
```
|
| 58 |
{
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
}
|
| 63 |
```
|
| 64 |
-
##
|
| 65 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
language:
|
| 3 |
+
- gl
|
| 4 |
+
pretty_name: summarization_gl
|
| 5 |
license: cc-by-4.0
|
| 6 |
task_categories:
|
| 7 |
+
- summarization
|
| 8 |
+
task_ids:
|
| 9 |
+
- news-articles-summarization
|
|
|
|
| 10 |
tags:
|
| 11 |
+
- galician
|
| 12 |
+
- summarization
|
| 13 |
+
- news
|
| 14 |
+
- journalism
|
| 15 |
+
- low-resource-nlp
|
| 16 |
+
- jsonl
|
| 17 |
size_categories:
|
| 18 |
+
- 10K<n<100K
|
| 19 |
+
configs:
|
| 20 |
+
- config_name: default
|
| 21 |
+
data_files:
|
| 22 |
+
- split: train
|
| 23 |
+
path: "train.jsonl"
|
| 24 |
+
- split: validation
|
| 25 |
+
path: "validation.jsonl"
|
| 26 |
+
- split: test
|
| 27 |
+
path: "test.jsonl"
|
| 28 |
---
|
|
|
|
| 29 |
|
| 30 |
+
# summarization_gl
|
| 31 |
|
| 32 |
+
## Dataset Summary
|
|
|
|
| 33 |
|
| 34 |
+
summarization_gl is a Galician summarization dataset built from news articles and automatically extracted summaries from three Galician news sources:
|
| 35 |
|
| 36 |
+
- **Nós Diario**
|
| 37 |
+
- **Que Pasa na Costa**
|
| 38 |
+
- **Praza Pública**
|
| 39 |
|
| 40 |
+
The dataset contains **80,829 instances** in total. Each instance includes a news text and its associated summary.
|
| 41 |
|
| 42 |
+
## Dataset Description
|
| 43 |
|
| 44 |
+
summarization_gl consists of pairs of:
|
| 45 |
|
| 46 |
+
- `summary`: an automatically extracted summary
|
| 47 |
+
- `text`: the corresponding full news article
|
| 48 |
|
| 49 |
+
The dataset includes materials from different news outlets with varying summary quality.
|
| 50 |
|
|
|
|
| 51 |
|
| 52 |
+
## Dataset Structure
|
| 53 |
|
| 54 |
+
The dataset is distributed in **JSONL format** and is split into:
|
| 55 |
|
| 56 |
+
- **train**: 56,600 instances
|
| 57 |
+
- **validation**: 8,080 instances
|
| 58 |
+
- **test**: 16,200 instances
|
| 59 |
|
| 60 |
+
The data in each split has been shuffled.
|
| 61 |
|
| 62 |
+
Each instance contains the following fields:
|
| 63 |
|
| 64 |
+
- `id`: identifier of the summary-text pair; it typically indicates the news source and an internal numeric ID
|
| 65 |
+
- `summary`: summary of the news article
|
| 66 |
+
- `text`: full news article
|
| 67 |
|
| 68 |
+
## Example
|
| 69 |
|
| 70 |
+
```json
|
| 71 |
{
|
| 72 |
+
"id": "NOS_58435",
|
| 73 |
+
"summary": "O artista coruñés Pepe Galán leva desde os anos 70 vinculado o mundo da arte galega. Recoñecido popularmente polas súas esculturas e intervencións, cando se lle pregunta, prefire ser considerado como un \"artista multidisciplinar\" máis que cinguirse a unha única etiqueta como a de \"pintor\" ou \"escultor\".",
|
| 74 |
+
"text": "Indo polas rúas da casa ao taller e do taller á casa, desprázome co ritmo de liturxia aprendida sobre o empedrado e o formigón... onde case sempre atopo algún dato, algunha novidade... Chega a ser unha relación familiar coa veciñanza e a cidade. Pola mañá gozo da paisaxe urbana mais non podo evitar ser crítico co deseño do espazo común..."
|
| 75 |
}
|
| 76 |
```
|
| 77 |
+
## Intended Uses
|
| 78 |
+
|
| 79 |
+
This dataset can be used for:
|
| 80 |
+
|
| 81 |
+
- fine-tuning summarization models for Galician
|
| 82 |
+
- instruction tuning for summarization tasks
|
| 83 |
+
- evaluation of summarization systems in Galician
|
| 84 |
+
- research on low-resource summarization
|
| 85 |
+
- experiments on summary quality, faithfulness, and domain adaptation in journalistic text
|
| 86 |
+
|
| 87 |
+
## Limitations
|
| 88 |
+
|
| 89 |
+
- The summaries were extracted automatically and are not uniformly high-quality.
|
| 90 |
+
- Summary quality varies substantially depending on the source.
|
| 91 |
+
- Many examples from **Nós Diario** are closer to article introductions or interview leads than to full summaries.
|
| 92 |
+
- Many examples from **Que Pasa na Costa** are very short and may overemphasize the most click-attractive part of the article rather than the overall content.
|
| 93 |
+
- The **Praza Pública** subset is of better quality, but the dataset as a whole is heterogeneous.
|
| 94 |
+
- Because of this variability, the dataset may be more suitable for controlled experiments, filtering, or source-aware training than for direct use as a uniformly curated gold-standard summarization benchmark.
|
| 95 |
+
|
| 96 |
+
## Data Format
|
| 97 |
+
|
| 98 |
+
The dataset is provided in **JSONL** format, with one JSON object per line.
|
| 99 |
+
|
| 100 |
+
Main fields:
|
| 101 |
+
|
| 102 |
+
- `summary` (`str`): automatically extracted summary
|
| 103 |
+
- `text` (`str`): corresponding news article
|
| 104 |
+
- `id` (`str`, when available): identifier of the example
|
| 105 |
+
|
| 106 |
+
## Acknowledgements
|
| 107 |
+
|
| 108 |
+
This dataset was compiled within the Nós Project, funded by the Ministerio para la Transformación Digital y de la Función Pública - Funded by EU – NextGenerationEU within the framework of the project ILENIA with reference 2022/TL22/00215336.
|