license: cc-by-nc-sa-4.0
task_categories:
- text-classification
language:
- pl
pretty_name: DARIAH Year Prediction Dataset
size_categories:
- 100K<n<1M
DARIAH Year Prediction Dataset
Table of Contents
- DARIAH Year Prediction Dataset
Dataset Description
Dataset Summary
DARIAH Year Prediction is a dataset comprising 677665 paragraphs of noisily OCR-ed Polish-language texts from scans of historical documents along with the corresponding date on which they were published, accurate up to the day of publication achieved by converting the date to a real number (float) format.
Supported Tasks and Leaderboards
text-classification
: This dataset was originally designed for a task best described as text regression, that is, predicting a continuous value given a piece of text. Considering that there is currently no such task listed on HuggingFace, the closest alternative is text classification since the labels are easily representable as integers at the cost of a relatively small (less than a year) loss of temporal precision.
Languages
The dataset contains only Polish-language texts.
Dataset Structure
Data Instances
{
text: 'Że płóczka mogła przywiązać rudnika do miej sca pierwotnego dowodzi fakt, iż w niektórych miejscowościach istnieją groble spiętrza jące, zbudowane całkowicie z żuż la dymarskiego. Grobletakie sąwPiasku i w Witowie nad rzeką Młynkowską, w Błotnicy nad rzeką Krasną oraz w kilku innych miejscowo ściach. Niejednokrotnie też spotykał autor na łą kach eksploatowane złoża żużla dymarskiego w miejscach, gdzie ledwie zarysowane na powierzchni łąki ślady spiętrzenia naprowadzały przebiegłych eksploatatorów na myśl, że tam może się żużel znajdować. Tak było np. w Szałasie. W tych wy padkach jak w Piasku, Błotnicy itp. nie sposób so bie wytłumaczyć budowanie grobli inaczej, jak tym, że zwały te gromadzone były wiekami, początkowo spiętrzenie było niewielkie, następ nie, z biegiem czasu, rosło. Zbędny żużel wywożo no na groblę, bo ta świetnie wstrzymywała napór wody. Powoli spiętrzenie należało podnosić, gdy przybywały nowsze urządzenia i maszyny. Ale umiejscowienie pieców wymagało innej jeszcze oko liczności : upłynnienia żużla. Pierwotnie żużel pozo stawiano wraz z piecem, od spodu wygrzebywano tylko łupkę. Tutaj natomiast należało żużel spuścić w stanie płynnym, przed wyjęciem łupki, by za prawę zbudowaną z kamieni użyć do następnego wytopu. Wszystko więc wskazuje na to, że ta technika zasadniczo różni się od opisanej poprzednio techniki pierwotnej wytopu; jest ona po czątkiem techniki dymarskiej, dokładnie znanej z XVI wieku. Jakim nazwom każda z tych technik odpowia dała? Czy jest w języku polskim jakiś znany ter min, który by posiadał wszelkie cechy prawdopo dobieństwa, że one tak się właśnie nazywały. Na określenie zakładu, wytapiającego rudę znane są następujące nazwy: ruda, rudnica, rud nia, rudnik. Wszystkie one oznaczają jedno i to samo, przeważnie jednak używa się ich dla ozna czenia ogólniejszego pojęcia.'
'year': 1937.49999996829
}
Data Fields
- text (string): OCR-ed text fragment
- year (float): year with a fraction representing a specific day on which a text fragment was written
Data Splits
This dataset has no canonical splits. The user is free to create his own splits suitable for the specific task at hand.
DARIAH Year Prediction Challenge
There exists a machine learning challenge based on this dataset accessible through AmuEval - Adam Mickiewicz University's challenge platform for machine learning research, competition, cooperation, and reproducibility. You can access it here.
Dataset Creation
Curation Rationale
This dataset was developed as part of the DARIAH-PL project at the Faculty of Mathematics and Computer Science of Adam Mickiewicz University in Poznań, Poland. The goal was to adapt several existing resources to the format of a single machine learning challenge that could be used as a benchmark for comparing an existing text dating model.
Source Data
The data comes from OCR-ed digitized Polish historical documents (mostly periodicals from the 1800-2000 year range) taken from collections of several Polish digital libraries.
Initial Data Collection and Normalization
The data was selected from two existing datasets available on the Gonito machine learning competition platform: retroc2 and Diachronia year prediction.
Who are the source language producers?
As the data comes from historical documents, it is fully human-generated. Nevertheless, it contains many artifacts typical for OCR outputs of inherently hard-to-OCR material, which makes it look less "natural" than standard written language.
Additional Information
Dataset Curators
All credit goes to @filipg7777 (Filip Graliński) (et al.) for gathering and compiling all of the data used in this dataset.
The dataset itself was curated by @kdudzic (Kacper Dudzic).
This research has been carried out within the DARIAH-PL project, a part of the European network DARIAH and funded by the Intelligent Development Operational Programme, Polish National Centre for Research and Development, ID: POIR.04.02.00-00-D006/20.
Licensing information
This dataset is available under the CC BY-NC-SA 4.0 license.
Citation information
As there is currently no publication associated with this dataset, please cite its HuggingFace dataset card:
@misc {dudzic2025dariahyearprediction,
author = {Kacper Dudzic and Filip Graliński},
title = {DARIAH Year Prediction Dataset},
howpublished = {\\url{https://huggingface.co/datasets/amu-cai/dariah-year-prediction-dataset}},
url = {https://huggingface.co/datasets/amu-cai/dariah-year-prediction-dataset},
type = {dataset},
year = {2025},
month = {May},
}