A comparison of sequential and combined approaches for named entity recognition in a corpus of handwritten medieval charters

This paper introduces a new corpus of multilin-gual medieval handwritten charter images, annotated with fulltranscription and named entities. The corpus is used to com-pare two approaches for named entity recognition in historicaldocument images in several languages: on the one hand, asequential approach, more commonly used, that sequentiallyapplies handwritten text recognition (HTR) and named entityrecognition (NER), on the other hand, a combined approachthat simultaneously transcribes the image text line and extractsthe entities. Experiments conducted on the charter corpus inLatin, early new high German and old Czech for name, dateand location recognition demonstrate a superior performance ofthe combined approach.