---
license: mit
task_categories:
- token-classification
language:
- ro
tags:
- ner
- flair
---

# HistNERo (CoNLL)

This repository contains the Historical Romanian Named Entity Recognition (HistNERo) dataset in a CoNLL-compatible format. 

The dataset uses the reference dataset from [`avramandrei/histnero`](https://huggingface.co/datasets/avramandrei/histnero) and introduces the following features:

* CoNLL-like dataset splits (see [`train.tsv`](train.tsv), [`dev.tsv`](dev.tsv) and [`test.tsv`](test.tsv)), including metadata
* Document-wise grouping
* Flair Dataset Loader, see the [`FlairDatasetExample.ipynb`](FlairDatasetExample.ipynb) example notebook

The dataset has 10,026 sentences (i.e. 8,020 train, 1,003 valid and 1,003 train) annotated with five named entities:

* `PERSON`
* `ORGANIZATION`
* `LOCATION`
* `PRODUCT`
* `DATE`

More details can be found in the [HistNERo: Historical Named Entity Recognition for the Romanian Language](https://arxiv.org/abs/2405.00155) paper.

# Caveats

We provide a CoNLL-compatible format of the original HistNERo dataset. Thus, there could still be some errors in the dataset transformation. Please open an issue if you
find any problems!

# License

We use the original MIT license from the authors' dataset repository.

# Acknowledgements

We heavily thank Andrei-Marius Avram, Andreea Iuga, George-Vlad Manolache, Vlad-Cristian Matei, Răzvan-Gabriel Micliuş, Vlad-Andrei Muntean, Manuel-Petru Sorlescu, Dragoş-Andrei Şerban, Adrian-Dinu Urse, Vasile Păiş and Dumitru-Clementin Cercel for releasing this great dataset resource!