---
license: cc
task_categories:
- text-generation
language:
- eu
tags:
- Basque
- Euskara
pretty_name: Zelai Haundi
size_categories:
- 100M<n<1B
---
# Dataset Card for Zelai Handi

<!-- Provide a quick summary of the dataset. -->
Zelai-Handi: A Large Open Dataset to Feed the next Gen of Basque LLMs

## Dataset Details

### Dataset Description

<!-- Provide a longer summary of what this dataset is. -->

- **Curated by:** Orai NLP Technologies
- **Language(s) (NLP):** EU
- **License:** CC
- 

## Uses

<!-- Address questions around how the dataset is intended to be used. -->

### Direct Use

<!-- This section describes suitable use cases for the dataset. -->

[More Information Needed]

### Out-of-Scope Use

<!-- This section addresses misuse, malicious use, and uses that the dataset will not work well for. -->

[More Information Needed]

## Dataset Structure

<!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. -->

The corpus is released in jsonl format. Each document contains the following properties:


```json
{
        "source": "source", 
        "license": "document license", 
        "lang": "eu", 
        "url": "document url", 
        "title": "document title", 
        "author": "document author if known, source otherwise", 
        "date": "yyyy-mm-dd", 
        "text": "document content, ready to use raw text. Title text (for news and articles) is included as well.",
        "domain": "document domain. one of ['news', 'administrative', 'wikipedia', 'science', 'subtitles', 'literature', 'videogames']"
}
```


## Dataset Creation

### Curation Rationale

<!-- Motivation for the creation of this dataset. -->

[More Information Needed]

### Source Data

<!-- This section describes the source data (e.g. news text and headlines, social media posts, translated sentences, ...). -->
This is a corpus of Basque language. It contains 559.380.531 raw tokens. It was gathered from various web sources and contains data up to 2024/02/23. 
The corpus is comprised of documents from various sources. All documents included have a free license, but each source has its own license. The following table summarizes corpus statistics and license information. 


| **source**         | **tokens (M)** | **license**               |
|-------------------:|---------------:|--------------------------:|
| tokikom            | 165,75         | cc-by / cc-by-sa          |
| berria + egunkaria | 178,77         | cc-by-sa 4.0              |
| administrative     | 92,6           | public domain             |
| Wikipedia          | 63,8           | cc0                       |
| addi               | 17             | cc-by-nc-sa / cc-by-nc-nd |
| argia              | 17,44          | cc-by-sa 4.0              |
| subtitles          | 11             | cc-by-sa                  |
| zientzia.eus       | 8,3            | cc-by-sa                  |
| susa               | 5,8            | cc-by-sa 3.0              |
| zientzia kaiera    | 2              | cc-by-sa 3.0              |
| ekaia              | 1,8            | cc-by-nc-nd               |
| ikergazte          | 1,7            | cc-by-sa 3.0              |
| game-erauntsia     | 0,40           | cc-by-sa 4.0              |
| **Total**          | **566,36**     |                           |



#### Data Collection and Processing

Data collection efforts were performed between 2024/01/01-2024/02/23.

<!-- This section describes the data collection and processing process such as data selection criteria, filtering and normalization methods, tools and libraries used, etc. -->


#### Who are the source data producers?

<!-- This section describes the people or systems who originally created the data. It should also include self-reported demographic or identity information for the source data creators if this information is available. -->

[More Information Needed]

Hemen source guztiak zerrendatuko nituzke, tokikom-en kasuan, barruko guztiak zerrendatuz.

#### Personal and Sensitive Information

The dataset contains publicly available data that might be considered personal and sensitive (e.g., data that reveals addresses, uniquely identifiable names or aliases, racial or ethnic origins, sexual orientations, religious beliefs, political opinions, financial or health data, etc.).

[More Information Needed]

## Bias, Risks, and Limitations

<!-- This section is meant to convey both technical and sociotechnical limitations. -->
This corpus inherits the Basque speaking society's stereotypes (gender bias, ethnical bias, cultural bias, linguistic bias, ...), and it contains fictional text (literature and subtitles).

<!--Furthermore, text coming from susa (literature) and subtitles sources, contain hate speech, slurs or explicit language.
We decided to include them in the corpus, to ensure that the corpus contains conversational data.
We did not perform any cleaning or censoring of those, since it might be needed for tasks like hate speech detection.-->


### Recommendations

<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->

Users should be made aware of the risks, biases and limitations of the dataset. 
We suggest filtering sources as literature and subtitles that might contain hate speech, slurs or explicit language before pretraining an LLM for end tasks like interactive chat.

## Citation

<!--We plan to release a paper soon, but, for now, please, cite as follows:

**BibTeX:**

@misc{ZelaiHaundi,
 title     = "{Zelai-Haundi: A Big Open Dataset to Feed the next Gen of Basque LLMs}",
 author    = "{Iñaki, Gorka, Xabi, Ander}",
 url       = "https://huggingface.co/datasets/orai-nlp/ZelaiHaundi",    
 year = 2024 }
-->

## Dataset Card Contact

- Iñaki San Vicente (i.sanvicente@orai.eus)
- Gorka Urbizu (g.urbizu@orai.eus)