Alleviating Digitization Errors in Named Entity Recognition for Historical Documents

Proceedings of the 24th Conference on Computational Natural Language Learning, pages 431–441 Online, November 19-20, 2020. c⃝2020 Association for Computational Linguistics https://doi.org/10.18653/v1/P17 431 Alleviating Digitization Errors in Named Entity Recognition for Historical Documents Emanuela Boros1, Ahmed Hamdi1, Elvys Linhares Pontes1, Luis Adri´an Cabrera-Diego1, Jose G. Moreno1,2, Nicolas Sidere1, and Antoine Doucet1 1 University of La Rochelle, L3i, F-17000, La Rochelle, France {emanuela.boros,ahmed.hamdi,elvys.linhares pontes,luis.cabrera diego}@univ-lr.fr {nicolas.sidere,antoine.doucet}@univ-lr.fr 2 University of Toulouse, IRIT, UMR 5505 CNRS, F-31000, Toulouse, France jose.moreno@irit.fr Abstract This paper tackles the task of named entity recognition (NER) applied to digitized his- torical texts obtained from processing digital images of newspapers using optical character recognition (OCR) techniques. We argue that the main challenge for this task is that the OCR process leads to misspellings and linguistic er- rors in the output text. Moreover, historical variations can be present in aged documents, which can impact the performance of the NER process. We conduct a comparative evalua- tion on two historical datasets in German and French against previous state-of-the-art mod- els, and we propose a model based on a hi- erarchical stack of Transformers to approach the NER task for historical data. Our find- ings show that the proposed model clearly im- proves the results on both historical datasets, and does not degrade the results for modern datasets. 1 Introduction With the emergence of large scale archives of digitized contents, the need for efficient preser- vation and accessibility of historical documents through appropriate technologies increased expo- nentially. At the same time, there is a growing interest in extracting relevant information from his- torical sources. In this paper, we address the named entity recognition (NER) task which aims at identi- fying real-world entities, such as names of people, organizations, and locations within historical docu- ments. Since most of the state-of-the-art research fo- cuses on NER for modern available datasets, the performance of the NER systems grew at a fast pace, enabled by the representational capacity of neural networks and off-the-shelf pre-trained word embeddings (Ma and Hovy, 2016; Lample et al., 2016; Yadav and Bethard, 2018). More recently, NER models based on contextual word and sub- word representations provided by ELMo (Peters et al., 2018), Flair (Akbik et al., 2018), or BERT (Devlin et al., 2019), achieved impressive improve- ments. The Transformer-based (Vaswani et al., 2017) architectures for NER became popular since the release of the BERT (Bidirectional Encoder Representations from Transformers) model. However, while most NER systems have been developed to generally address contemporary data, NER systems for processing historical documents are less common. To extract entities from histor- ical documents, NER tools face additional chal- lenges. As the majority of these documents are hardcover, they are scanned and processed by an OCR to transcribe the text. However, an OCR tool can occasionally misrecognize letters and im- properly identify its textual content. This can be due to the level of degradation of the actual doc- ument being scanned, to the digitization artifacts and also to the quality of the OCR tool. This leads …

Content truncated to 500 words