Named entity recognition specialised for Portuguese 18th century History research

Abstract

This paper presents the construction of a corpus and the respective models learned for the Named Entity Recognition (NER) task, specialised for historical research. The entity categories were adapted based on the objectives of the historical analysis of the 18th-century text. We trained and evaluated traditional neural networks and the new Large Language Models (LLMs) for the NER task. In total, we assessed six language models, where the results of traditional architectures were superior to LLMs.