This dataset contains 91 pairs of newspaper articles from the Nineteenth Century Serials Edition (NCSE). The articles are the original OCR from the NCSE and the transcribed equivalent. The data was used in "CLOCR-C: Context Leveraging OCR Correction with Pre-trained Language Models" to demonstrate that pre-trained language models are able to perform post-OCR correction improve the accuracy of corrupted OCR text. The paper is can be found on arxiv at https://arxiv.org/abs/2408.17428
The data set comes from 6 different publications, and is made up of 91 articles, containing a total of 40712 words distributed across the 19th Century.
The dataset is zip file made up of two sub-files containing 91. Each file shares its name with a corresponding file in the other folder.