---
annotations_creators:
- expert-generated
language:
- ca
- es
language_creators:
- found
license: apache-2.0
multilinguality:
- multilingual
pretty_name: ner-cat
size_categories:
- 1K<n<10K
source_datasets:
- original
tags:
- NER
- Catalan
- NLP
- television transcriptions
- manual annotation
- GLiNER
task_categories:
- text-classification
- token-classification
task_ids:
- multi-label-classification
- named-entity-recognition
---

# NERCat Dataset

### Dataset Summary

The NERCat dataset is a manually annotated collection of Catalan-language television transcriptions, designed to improve Named Entity Recognition (NER) performance for the Catalan language. The dataset covers diverse domains such as politics, sports, and culture, and includes 9,242 sentences with 13,732 named entities annotated across eight categories: Person, Facility, Organization, Location, Product, Event, Date, and Law. The dataset was created to address the lack of high-quality annotated data for Catalan, a low-resource language, and to support the development of NLP applications in Catalan media, governance, and cultural domains.


### Languages

The dataset primarily contains **Catalan** (ca) language text, with occasional **Spanish** (es) code-switching segments. The language is represented in its standard form, with no specific regional or dialectal variations highlighted.


### Label Descriptions

- **Person**: Names of individuals, including fictional and historical figures.
  
- **Facility**: Physical infrastructures such as airports, bridges, and stadiums.

- **Organization**: Entities such as companies, institutions, political parties, and media outlets.

- **Location**: Geographical places, including cities, regions, and landmarks. This also includes names of countries, mountain lakes, or planets.

- **Product**: Commercial products, such as branded phones or soft drinks. This category also includes cultural products, such as the titles of films, books, and songs.

- **Event**: Named occurrences like festivals, elections, or historical battles.

- **Date**: Temporal expressions indicating specific days, months, years, or time periods.

- **Law**: Names of laws, legal frameworks, and judicial matters, including constitutions, court rulings, and high-profile judicial cases.

## Dataset Structure


### Data Instances

The dataset uses a tokenized format that is directly compatible with the GLiNER framework, which is optimized for training and inference tasks in NER. Below is an example of a JSON-formatted instance:

```json
{
    "tokenized_text": [
        "La",
        "Universitat",
        "de",
        "Barcelona",
        "és",
        "una",
        "de",
        "les",
        "institucions",
        "educatives",
        "més",
        "importants",
        "de",
        "Catalunya",
        "."
    ],
    "ner": [
        [1, 3, "Organization"],
        [13, 13, "Location"]
    ]
}

```
### Explanation of the Fields:

- **tokenized_text**: This is the sentence broken down into individual tokens (words).
  
- **ner**: This is a list of tuples, where each tuple represents a named entity in the sentence:
  - The first and second elements represent the token indices in the `tokenized_text` list (e.g., `[1, 3]` means the entity spans tokens at positions 1 to 3).
  - The third element is the label for that entity.

### Example of No Named Entities Detected:

If a sentence doesn't contain any recognized named entities, the `ner` list will be empty. In such cases, it’s important to include all possible labels in the `label` field to maintain consistency in your data.

Example with no named entities:

```json
{
    "tokenized_text": [
        "El",
        "director",
        "de",
        "la",
        "pel·lícula",
        "és",
        "molt",
        "talentós",
        "i",
        "ha",
        "guanyat",
        "diversos",
        "premis",
        "."
    ],
    "ner": [],
    "label": [
        "Person",
        "Facility",
        "Organization",
        "Location",
        "Product",
        "Event",
        "Date",
        "Law"
    ]
}
```



### Annotations

All the processes involved in creating the dataset, including data collection and annotation, were carried out manually by experts.


## Supported Tasks

 **Named Entity Recognition (NER):** The dataset can be used to train and evaluate models for Named Entity Recognition in Catalan. Success on this task is typically measured by achieving high precision, recall, and F1-scores. The fine-tuned NERCat model achieves near-perfect scores in categories like Person, Organization, and Location, with significant improvements in underrepresented categories such as Facility, Product, and Law.
  - **Metrics:** Precision, Recall, F1-score
  - **Model:** [NERCat Fine-Tuned GLiNER Model](https://huggingface.co/Ugiat/NERCat)


## Other


### Citation Information

```
@misc{article_id,
  title        = {NERCat: Fine-Tuning for Enhanced Named Entity Recognition in Catalan},
  author       = {Guillem Cadevall Ferreres, Marc Bardeli Gámez, Marc Serrano Sanz, Pol Gerdt Basullas, Francesc Tarres Ruiz, Raul Quijada Ferrero},
  year         = {2025},
  archivePrefix = {arXiv},
  url          = {https://arxiv.org/abs/2503.14173}
}
```