English

Data Centric Domain Adaptation for Historical Text with OCR Errors

Computation and Language 2021-07-05 v1 Machine Learning

Abstract

We propose new methods for in-domain and cross-domain Named Entity Recognition (NER) on historical data for Dutch and French. For the cross-domain case, we address domain shift by integrating unsupervised in-domain data via contextualized string embeddings; and OCR errors by injecting synthetic OCR errors into the source domain and address data centric domain adaptation. We propose a general approach to imitate OCR errors in arbitrary input data. Our cross-domain as well as our in-domain results outperform several strong baselines and establish state-of-the-art results. We publish preprocessed versions of the French and Dutch Europeana NER corpora.

Cite

@article{arxiv.2107.00927,
  title  = {Data Centric Domain Adaptation for Historical Text with OCR Errors},
  author = {Luisa März and Stefan Schweter and Nina Poerner and Benjamin Roth and Hinrich Schütze},
  journal= {arXiv preprint arXiv:2107.00927},
  year   = {2021}
}

Comments

14 pages, 2 figures, 6 tables

R2 v1 2026-06-24T03:50:09.342Z