Evaluation of Named Entity Recognition in Historical Argentinian Documents

Resumen

Research over historical text volumes can be performed by means of automatic tools that help historians achieve more abstract and aggregated points of view. Tasks such as Information Extraction or Text Mining can be performed more efficiently if Machine Learning models are employed. We propose the evaluation of different state-of-the-art models over a new dataset for Named Entity Recognition. The dataset was built over a History texts volume about General G¨uemes, a national Argentinian independence hero. The results show that some models perform better in terms of precision, recall and f1-score for most types of entities. Specifically, pretrained language models fine-tuned for this particular task show considerably higher performance than classical models based on word embeddings and other kinds of representations and models. Besides, statistical tests are provided to ensure the significance in the differences of the performance values attained. Hence, the contribution of this work is twofold, on the one hand a new corpus and dataset for Named Entity Recognition and a complete statistical assessment of performance values of state-of-the-art models over the generated dataset.