Named-entity recognition for early modern textual documents: a review of capabilities and challenges with strategies for the future

Abstract

PurposeBy mapping-out the capabilities, challenges and limitations of named-entity recognition (NER), this article aims to synthesise the state of the art of NER in the context of the early modern research field and to inform discussions about the kind of resources, methods and directions that may be pursued to enrich the application of the technique going forward.Design/methodology/approachThrough an extensive literature review, this article maps out the current capabilities, challenges and limitations of NER and establishes the state of the art of the technique in the context of the early modern, digitally augmented research field. It also presents a new case study of NER research undertaken by Enlightenment Architectures: Sir Hans Sloane's Catalogues of his Collections (2016–2021), a Leverhulme funded research project and collaboration between the British Museum and University College London, with contributing expertise from the British Library and the Natural History Museum.FindingsCurrently, it is not possible to benchmark the capabilities of NER as applied to documents of the early modern period. The authors also draw attention to the situated nature of authority files, and current conceptualisations of NER, leading them to the conclusion that more robust reporting and critical analysis of NER approaches and findings is required.Research limitations/implicationsThis article examines NER as applied to early modern textual sources, which are mostly studied by Humanists. As addressed in this article, detailed reporting of NER processes and outcomes is not necessarily valued by the disciplines of the Humanities, with the result that it can be difficult to locate relevant data and metrics in project outputs. The authors have tried to mitigate this by contacting projects discussed in this paper directly, to further verify the details they report here.Practical implicationsThe authors suggest that a forum is needed where tools are evaluated according to community standards. Within the wider NER community, the MUC and ConLL corpora are used for such experimental set-ups and are accompanied by a conference series, and may be seen as a useful model for this. The ultimate nature of such a forum must be discussed with the whole research community of the early modern domain.Social implicationsNER is an algorithmic intervention that transforms data according to certain rules-, patterns- or training data and ultimately affects how the authors interpret the results. The creation, use and promotion of algorithmic technologies like NER is not a neutral process, and neither is their output A more critical understanding of the role and impact of NER on early modern documents and research and focalization of some of the data- and human-centric aspects of NER routines that are currently overlooked are called for in this paper.Originality/valueThis article presents a state of the art snapshot of NER, its applications and potential, in the context of early modern research. It also seeks to inform discussions about the kinds of resources, methods and directions that may be pursued to enrich the application of NER going forward. It draws attention to the situated nature of authority files, and current conceptualisations of NER, and concludes that more robust reporting of NER approaches and findings are urgently required. The Appendix sets out a comprehensive summary of digital tools and resources surveyed in this article.

MELHISSA: a multilingual entity linking architecture for historical press articles

“…To improve the accuracy of the non-NIL candidates provided by the EL system, we use a post-processing filter 14 based on heuristics and data provided by Wikidata and DBpedia. The goals of the filter are to: (1) Remove candidates which are unlikely such as disambiguation pages or people born after the document publication; (2) Verify that the tokens of a particular named entity are linked to the same candidates;…”

“…However, this has not been the case for historical documents, due to certain characteristics, which make their processing particularly difficult. A few exceptions exist [9,13,14], but in far smaller numbers than for contemporary documents. Among the challenges, such tools need to be able to deal with errors produced by OCR systems, to manage some specific vocabulary, and also to handle spelling variations with respect to modern standards.…”

Development and application of a digital humanities research platform for biographies of Malaysian personalities

“…For example, in this case, the GIS system had mislabeled the coordinates of Thean Hou Temple in Kuala Lumpur as the Taichung Mazu Temple in Taiwan. In Europe and the USA, this problem is mostly solved by combining the annotations on geographical coordinates with the authoritative tables provided by a Gazetteer, such as Getty Thesaurus of Geographic Names , Pleiades and so forth (Humbel et al , 2021). Thus, it is suggested that the geographic information regarding the Malaysian Chinese organization should also be established with the same kind of authoritative table from sources, such as the Roster of Malaysian Chinese Associations , to reduce the errors in geographic information.…”

“…The researcher pointed out that although CNER can successfully identify most of the character entities, it is difficult to directly identify the characters’ real names as some of them are aliases or honorifics. In the future, a much more reliable glossary of names should be established by making use of the organizational registers and donation registers of local Chinese societies to enhance the recognition accuracy of named entities (Humbel et al , 2021; Xu et al , 2020). Furthermore, future research may focus on collecting more external information, such as character representations and cross-lingual information, and maintaining more and higher quality CNER data sets like ENER, to improve the model performance of CNER (Liu et al , 2022).…”

Application of named entity recognition method for Indonesian datasets: a review

“…There are four approaches to NER: i) a rule-based approach, which does not require annotated data because it relies on artificial rules; ii) an unsupervised learning approach; iii) a feature-based supervised learning approach that relies on supervised learning algorithms with careful feature engineering; and iv) a deep-learning-based approach, which automatically finds the required representation for detecting or classifying raw input in an end-to-end manner [3], [4]. NER is a straightforward process for humans because many named entities are self-names, and most of them have initial capital letters and can be easily recognized, but for machines, it is very difficult [5]. Information extraction often uses data available on social media, online news, and e-commerce [3].…”