Session 12 – Data from Annotated Texts
The sixth day of the Data for History conference started with Cristina Vertan’s (University of Hamburg) presentation “Representing vague and uncertain historical data on places people and events in historical texts – a case-study of historical texts of Dimitrie Cantemir”. The main focus was on the texts and translations of the universal researcher Dimitrie Cantemir, who was the only one to write a history on the Balkan region from a western perspective. The original work was written in Latin and translated into other languages only in the 18th century. Since there was no word for word translation, the project is now dedicated to such translations. It combines hermeneutic methods with the methods of computer science. There are three directions of investigation: reliability, consistency and vagueness. First, a model with a knowledge base was developed, which also takes into account vague or uncertain places and persons. After that, the detection was done, e.g. to annotate linguistic marks for uncertain places in each of the available languages, followed by the visualization. One of the central points of the modeling was the creation of an ontology. With this ontology, one can now answer queries about the translations and texts, while taking ambiguities into account. The presentation highlighted the complexity of working with texts and translations.
The next presentation featured Deborah Leems, Julianne Nyhans and Antonis Bikakis (University College London) project “Sir Han Sloane’s Information Architecture: Lessons Learned From A Data Driven Research“. This research project deals with five works by physicist Sir Han Sloane that have been digitized, transcribed, and encoded with TEI. The key idea of the Collection as data initiative is to encourage computational use of the digitized sources instead of viewing them as mere surrogates. The project promotes interoperability and compatibility. The project can be broken down into three steps: First automatic extraction of named entities eg. Places and person names. Then assigning a unique ID to each place and person identified, and finally linking them to CIDOC-CRM and Wikidata, to make the TEI-XML more interoperable. This is an iterative process. During the project, the participants encountered different problems. One of them was dealing with various spellings of person and place names. To solve this and simplify the search, expanding persons and place name lists in XML with the different name variations were created.
The discussion explored the question of how to incorporate various background information into the projects. There was also the question of how to model hierarchical relations between place names, e.g. in the case of provinces. Another point that was discussed was the question of how to trace the development of toponyms, used in the documents and to what extent these are correct.
The third keynote “Extracting and Representing Time, Places and Agents from Text & their representation” was given by Antske Fokkens (Vrije Universiteit Amsterdam). As a computational linguist, she is heavily involved in the use of language-based technologies in the humanities. In her keynote, she highlighted the benefits – and the potential problems – of using language tools for historical research. For her, dealing with Language Tools (LT) means two things in particular: the numerous possibilities, but also the awareness that these tools are not error-free, meaning that one always has to deal critically with their use as well. These technologies have the advantage of being able to process larger amounts of data and detect patterns that are difficult or impossible for humans to discern. On other hand, technologies and processed data can contain or even amplify biases, since computers are still bad at interpreting natural language. In order to use LT, it is important to know what these tools can and cannot do and how that relates to the type of question they are intended to help answer.
For this to happen, different disciplines need to work together and share information and data. LT engineers can contribute knowledge of how these tools work and where possible pitfalls are when they are used for certain questions. Historians, on the other hand, know exactly what problems can arise when working with historical data, such as multiple perspectives, changing concepts, and vagueness in language, and ambiguities of terms.
Her conclusion is that computational linguists can provide tools to extract, analyse and present historical data, while also taking uncertainties into account. However, it is difficult to do this in a reliable way and to communicate clearly about it, as analyses can be very diffuse. Therefore, technical challenges and possible adjustments need to be communicated to users so that they know where to be cautious and where possible flaws or tool errors can arise, and then decide how to make the data accessible in a way that reflects the data that is needed.
This requires collaborations between disciplines, an understanding of different ways of working, and knowledge about the sources and the tools we use.
At the end of the presentation, she asked the audience how to communicate about uncertainty, both on a source basis and on a tool basis, and how to formalize interpretations and methods.
In the following discussion, the first question was raised on how to make contradictory sources visible. For Antske Fokkens it is important to refer to the source as well as to add annotations to include such contradictions in a model.
Likewise, multiple questions were asked, how different expert opinions can be handled and whether it is also possible to include connotations. Here Antske Fokkens showed a part of the GRaSP model (a generic framework for modeling perspectives and their sources she is working on), which fulfills exactly these requirements. In the last part of the discussion, it was discussed, how to deal with different interpretations of time and the solution for it in the GRaSP model, and finally how to deal with “re-classification” of sources, meaning sources that turn out to be unreliable. This last requirement can also be represented in the Modell.
digitalhistory: Julia Pabst, Anna Grönig: Report on the conference – “Data for History 2021” – Day 6 (23 June 2021), in: Digital History Berlin (Blog), erschienen am: 12. Juli 2021, Internet: https://dhistory.hypotheses.org/722.
About the authors:
Julia Pabst studies history at the Humboldt-Universität zu Berlin in the master’s program with focus on Digital History.
Anna Grönig has a bachelor’s degree in German language and literature and history and is now studying for a master’s degree in history with a focus on Digital History at the Humboldt-Universität zu Berlin.