Evaluation of Different Tagging Schemes for Named Entity Recognition in Handwritten Documents.

Performing Named Entity Recognition on Handwritten Documents results in categorizing particular fragments of the automatic transcription which may be employed in information extraction processes. Different corpora employ different tagging notations to identify Named Entities, which may affect the performance of the trained model. In this work, we analyze three different tagging notations on three databases of handwritten line-level images. During the experimentation, we train the same Convolutional Recurrent Neural Network (CRNN) and n -gram character Language Model on the resulting data and observe how choosing the best tagging notation depending on the characteristics of each task leads to noticeable performance increments.