Data Centric Domain Adaptation for Historical Text with OCR Errors
It addresses domain shift and OCR error issues for historical text processing, which is an incremental improvement in a domain-specific area.
The paper tackles Named Entity Recognition on historical Dutch and French texts with OCR errors by proposing methods for in-domain and cross-domain adaptation, achieving state-of-the-art results that outperform strong baselines.
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.