CORAL: Joint research on language models

The CORAL (Constrained Retrieval-Augmented Language Models) research project was launched in October 2024. It is funded by the BMBF. Together with the Universities of Leipzig and Kassel and Anhalt University of Applied Sciences, the DNB is researching how language models can be trained with text corpora that are subject to certain restrictions (constraints) - for example, copyright. These constraints affect a large part of the German National Library's collection.

The project investigates whether and how large language models can be trained with derived text formats - i.e. texts with reduced information content from which the original text can no longer be reconstructed. Statements in generated texts should also be transparently comprehensible by indicating sources. CORAL thus contributes to making work with language models legally more secure and qualitatively better in the long term.

Research with Twitter: conference and data sprint

Another cooperation with the research community was related to Twitter. Social media are a source of data and a subject for different research approaches in various disciplines. This is one of the reasons why the DNB set itself the task of archiving German-language Twitter (now “X”) in 2023.

Photo: DNB, Stephan Jockel

Photo: DNB, Stephan Jockel

In March 2024, the DNB hosted the first social media conference in Germany, which was attended by libraries and archives as well as researchers. The conference was followed by a two-day Twitter data sprint. Here, researchers were able to work on their research questions using Twitter data. Three extensive and unique data corpora were made available for this purpose.

New research network: EHRI (European Holocaust Research Infrastructure)

Since 2022, the German Exile Archive has been cooperating with the EHRI (European Holocaust Research Infrastructure). EHRI aims to support Holocaust research, network the research community and provide access to archive holdings and cutting-edge research. In November 2024, the Exile Archive and the Center for Holocaust Studies at the Leibniz Institute for Contemporary History jointly organized the EHRI seminar “Holocaust and Exile. Approaches, Sources, Methodologies”. Thirteen scholars from Germany, Israel, Italy, Austria, Portugal and Serbia took part in the seminar.

During the four-day event, the participants were able to learn more about the approaches and methods of exile research. They were also given a detailed insight into the work of the German Exile Archive. The Exile Archive offered the participants guided tours of its exhibitions and a city tour on “Child emigration from Frankfurt”. As part of a hands-on workshop, the researchers worked with archive materials from the Exile Archive on the topic of “Last messages into exile” and also dealt with the cooperation project with the Arolsen Archives to record the expatriation card file.

Towards a data competence center: HERMES

The DNB not only learns from other cultural institutions, but also from and with the scientific community. Since November 2023, it has been participating in the HERMES joint project. The project is funded by the Federal Ministry of Education and Research (BMBF). HERMES stands for Humanities Education in Research, Data, and Methods. The aim is to establish a data competence center for the humanities and cultural sciences. The project gained momentum in 2024: Students and employees from libraries, archives, and museums (international: GLAM) now come together in a transfer workshop. They discuss how the GLAM sector can better support research projects in the digital humanities in the future and what digital skills and infrastructures are needed to achieve this. The project also addresses new job profiles that are emerging as a result of the digital transformation, such as data steward, data librarian, and embedded librarian.

News from the Committee for Library Standards

In addition to collaborating in research networks, the DNB also cooperates with other organisations and institutions in the field of standardisation. In the Committee for Library Standards (STA), for example, it works together with partners from culture and science on uniform standards for cataloguing, interfaces and formats to enable better networking. The STA's rules of procedure were updated in November 2024. With the STA Community Forum, there is now also a virtual cooperation space in which STA groups and communities outside the STA can exchange information.

Furthermore, the STA's documentation platform was given an editorial environment last year. This makes it easier for the specialist and working groups to develop and coordinate changes and additions to the recording rules. The agreed changes are published on the STA-Plattform in half-yearly releases.

GND and European Thesaurus on International Relations and Area Studies

The Integrated Authority File (GND) also benefited from a cooperation in 2024. The entries of the multilingual European Thesaurus on International Relations and Area Studies (ETIRAS) are now linked to the Integrated Authority File. Terminology experts from the fiv (The German Information Network for International Relations and Area Studies) have created a mapping to the GND for this purpose. The dataset comprises around 8,000 mappings of terms relevant to political science. It will be jointly maintained by the DNB and the fiv-representatives, in particular the Franco-German Institute and the German Institute for International and Security Affairs (Stiftung Wissenschaft und Politik).