• Tobias Hodel (University of Bern):
  
  Machine Learning Approaches to Making and Using Digital Editions: Premodern Documents, Text, and Named Entity Recognition

  Machine learning capabilities and scholarly edition workflows seem to fit almost perfectly together at first sight. Text and entities (like persons and places) must be recognized and digitally processed. The presented paper focuses on premodern documents and depicts the current state-of-the-art in combining machine learning approaches for handwritten text and named entity recognition. Both text recognition and language model-based entity recognition are still developmental or need specific training when applied to unformalized (read premodern) textual entities; statistical scores are relatively low compared to standardized systems for modern languages. Ongoing developments in the digital humanities and the understanding to use available technology cautiously make for an optimistic outlook, even for languages and language forms with only sparse data reservoirs.

  The use of deep learning offers in parallel insights into the consequences for scholarly editions, whereby the need to focus on deliberately chosen documents is minimized, and post-processing steps require to be intensified. The use of machine learning for editions leads thus to a reflective stance about digital scholarly editions and the insight that cooperation is critical in developing robust and consciously built machine learning systems that benefit the academic edition community.

• Richard Hadden, Nina Rastinger, Matthias Schlögl (Austrian Academy of Sciences):
  
  Data Mining and Data Sowing: Automated Methods for Building a Digital Scholarly Edition of Historical Newspapers

  This article presents ongoing work on the DIGITARIUM, a project to digitise the Wien[n]erisches Diarium newspaper. The information contained within the Diarium is an invaluable resource for historical researchers, as well as for linguistic and other analysis. Here, we consider the process of digitisation, and the progress towards annotating the digitised texts, both using machine learning technologies. While this article focuses on the digitised texts themselves, considering them as a form of digital scholarly edition, both the process of creation and augmentation, and digital humanities-based research into the text can be seen as interdependent (even symbiotic): historical research employing machine learning techniques is fed back into the edition itself, forming layers of annotation, while simultaneously improving machine learning models. Such an approach raises questions regarding traditional approaches to textual scholarship and the status and utility of such a digital edition, which we shall consider.

• Sandra Bläß, Marie Flüh, Julia Nantke (Hamburg University), David Maus (University Library Hamburg):
  
  Quality Management for Machine Generated Data in Digital Scholarly Editions – Possibilities and Challenges

  In our paper, we present the automatisation-oriented workflow we apply in our project Dehmel digital to transcribe and index 35,000 documents (mainly handwritten letters). Our approach includes image digitisation, data transfer, HTR and transcription, tokenizing, NER, and disambiguating the entities. We combine manual, semi-automatic, and automatic resp. algorithm-driven methods by using diverse Machine Learning techniques. The goal is to process the enormous amount of analog letters so they can be presented in a DSE. A main issue throughout the whole process is the question of how we assure scholar quality of all of our produced data. By applying the quality management dimensions of Askham et al. – Completeness, Uniqueness, Timeliness, Validity, Accuracy, Consistency –, we examine in our paper how to measure and define quality at each work step, at which points in the workflow corrective interventions are possible without rendering the data structure unusable for subsequent processing steps, and how to deal with inaccuracies. A key interest here is also the role of the editor: How does it change when editorial work becomes algorithm-driven?

• Daniel Stökl Ben Ezra (École Pratique des Hautes Études), Hayim Lapin (University of Maryland):
  
  From HTR to Digital Critical Scholarly Edition: Reflexions on the Use of Machine Learning, Computational and Digital Humanities in the Sofer Mahir Project

  The present paper will give an overview about the use of machine learning as well as vintage algorithms from natural language processing (NLP) and computer vision (CV) and diverse digital humanities (DH) tools in the preparation of a series of digital editions of the major texts from the earliest phase of rabbinic Judaism. We employed these methods for page layout segmentation (PLS), handwritten text recognition (HTR), tokenization, lemmatization, named entity recognition (NER), text-reuse detection and text alignment. We are currently in the final stages of the data curation and preparation of the TEI-Publisher customization for the visualisation of the data.

• Keynote: Roger Labahn (University of Rostock)

Dinner together

• Werner Scheltjens (University of Bamberg):
  
  Extracting Preindustrial Logistics Patterns From a Digital Edition: Reflections on a Workflow for Data Mining the Schenkenschans Customs Registers (1630-1810)

  This paper discusses preliminary results and forthcoming tasks of a project that started in the fall of 2021 at the University of Bamberg. The project aims to facilitate the study of logistics patterns in German-Dutch transport and trade on the River Rhine in the early modern period by means of a digital edition of the customs registers of the Schenkenschans (SSZ). Based on an ongoing pilot study with a sample of the SSZ registers, the paper discusses the use of tools for HTR as starting point for creating a digital edition of non-narrative sources such as the SSZ customs registers and discusses the possibility of using machine learning for the standoff annotation of register entries. The pilot study aims to find out whether a digital edition is at all feasible for the SSZ registers and what role automated procedures may play during the process.

• Manuel Portela (University of Coimbra):
  
  Text Mining the Book of Disquiet for Critical and Creative Explorations

  The LdoD Archive is a highly dynamic archive dedicated to Fernando Pessoa’s Book of Disquiet. It includes a scholarly edition of the work based on the autograph witnesses, a transcription of four canonical editions, and several interactive functions that allow subjects to explore multiple reading sequences, compare editorial versions, produce virtual editions of the work, and annotate texts in those virtual editions. These subject-oriented interactions are based on a combination of human-only and machine-assisted operations. This article discusses those machine-assisted interventions as forms of text mining for critical and creative explorations of the processes of reading and editing as interpretative actions. Thus the focus of the LdoD Archive is not on machine learning as a form of automation of text analysis per se, but rather on the uses of algorithmic procedures as contributions to the role-playing and gamification rationale of our editorial model.

• Ulrike Henny-Krahmer (University of Rostock), Frederike Neuber (Berlin-Brandenburg Academy of Sciences and Humanities):
  
  Topic Modeling in Digital Scholarly Editions

  Topic Modeling is a quantitative method of text analysis that has been increasingly used in recent years for the analysis of digital text collections in the humanities. In this paper, we investigate to what extent this also applies to text corpora that are being created as scholarly digital editions. Possible fields of application of topic modeling for the description and representation of contents in digital editions are discussed and from the perspective of the topic modeling workflow it is worked out which aspects have to be considered in particular when applying the method to scholarly edited texts. The method is applied to two corpora of correspondence of the German-language authors Jean Paul (1763-1825) and Uwe Johnson (1934-1984), which are being compiled in edition projects at the Berlin-Brandenburg Academy of Sciences and Humanities (BBAW). Topic modeling stands for a way of working with text that in many respects contrasts with critical editing. Nevertheless, the method can enrich the work on and presentation of digital scholarly editions in a thematic dimension if the different approaches to texts are explained and taken into account.

• Marc Lemke, Konrad Sperfeld, Jochen Zöllner (University of Rostock):
  
  Introducing NTEE: An Easy to Use Tool to Enrich TEI Files With Entities Based on State of the Art Neural Networks

  In this paper we introduce the open-source software NTEE which is designed for automatically enriching XML-TEI files with XML elements as part of a Named Entity Recognition (NER) and Named Entity Linking (NEL) task within a Digital Scholarly Edition workflow. As our starting point for the development of the software, we first outline the fact that NER and NEL are desired supporting technologies in digital edition projects, but are rarely used due to a lack of expertise and the lack of adaptability in existing software solutions to edition-specific guidelines. Referring to this, we then show how NTEE is specifically designed for adaptability and accessibility: Based on state of the art neural language models, users can train and then apply their own NER taggers to XML-TEI files. For the subsequent manual postprocessing, NTEE offers technical support on the one hand for evaluating the achieved annotation results and on the other hand by accessing databases using Semantic Web technology in order to clearly identify the recognized entities. All this together can be handled in NTEE by the users within a single graphical user interface.