The digital editing of early modern correspondence requires precise modeling of historical texts as well as the integration of modern tools for annotation and analysis. The project “Die sozinianischen Briefwechsel” (http://www.sozinianer.de

) provides a digital edition of letters from the circle of the anti-Trinitarian Socinians of the seventeenth century. Based on the ATAG model (Applied Text as Graph), texts are represented in a labeled property graph down to the level of individual characters, enabling flexible annotations, recursive comments, and algorithmic analyses (cf. Kuczera 2024; Haentjens/Birnbaum 2017).

This article presents an innovative approach to the automated annotation of these letters: the integration of Anthropic’s language model Claude via the interface module MCP (Model Context Protocol) enables the identification, typing, and semantic enrichment of annotatable text passages while considering only the already existing metadata on persons, places, and objects. The annotations thus generated are integrated as annotation nodes into the existing Neo4j graph database and are subsequently available both in the web frontend and for algorithmic analysis. During the annotation process, Claude can access previously annotated texts and registers and orient itself accordingly.

The aim of this article is to present the technical and methodological workflow combining ATAG, Claude with MCP, and Neo4j, and to analyze it with regard to editorial practice and annotation fidelity. The original contribution lies in the technical integration of an LLM annotator into a granularly structured editorial system with direct graph anchoring, which for the first time enables direct semantic enrichment at the character level.

This paper proposes extending the scope of digital humanities textual scholarship to the formal sciences — logic, mathematics, and computer science — and argues for the adoption of formal methods as a class of digital methods suited to this domain that have thus far been absent from DH practice.

The proposed application context is the study of concept and theory genesis in the formal sciences, a domain analogous to the history of concepts (Begriffsgeschichte) in historiography or motif research in literary studies. Just as certain motifs recur across literary traditions, particular concepts and theories pervade the history of mathematics. What distinguishes mathematics from literary culture is that its canonicity is constituted through formalization — and formalized texts are amenable to formal methods operating at the symbolic level that offer a precision unavailable to conventional DH approaches. Unlike statistical methods such as large language models, formal methods remain transparent and interpretable, and are free from the hallucination problem.

The paper argues for a two-component model of digital editions of mathematical texts, comprising a philological component — encoding the natural-language text according to standard editorial principles — and a formalist component that captures the conceptual and theoretical content in a fully formalized representation accessible to formal methods. TEI is proposed as the framework for the former; OMDoc (Kohlhase, 2006) for the latter. Existing digital editions of mathematical and scientific texts, such as the Newton Project and Bernoulli Euler Online, are shown to fall short of this standard by representing formulae only at the level of visual surface structure, rendering them unsuitable for formal analysis. While a fully formalized edition component remains a long-term goal requiring expertise in automated theorem proving beyond current DH capacities, the paper argues that this should not preclude first steps in this direction.

Digital edition projects have expanded significantly but remain fragmented, hindering collaboration and knowledge integration. While diversity in scholarly approaches is valuable, incompatible metadata standards (e.g., TEI, Dublin Core, project-specific schemas) and monolithic front end designs create technical barriers, preventing seamless data aggregation and cross-referencing. This isolation forces researchers to manually reconcile differences, duplicating effort and limiting comparative analysis.

To address these challenges, a federated interoperability framework is proposed, built on three pillars: (1) semantic mapping layers using linked data to translate between metadata schemas without requiring standardization; (2) standardized API contracts (REST/OpenAPI) to expose digital objects in machine-readable formats; and (3) distributed APIs for search, entity linking, and web annotations, enabling cross-platform scholarly discourse.

The reference implementation includes a Python-based back-end with a JSON database to accommodate heterogeneous metadata. The TextAPI organizes resources in a flexible hierarchy (collection-manifest-item), while complementary APIs (SearchAPI, IndexAPI, AnnotationAPI) support discovery and annotation workflows. The front end application, TIDO (Text Viewer for Digital Objects) , allows customizable views while maintaining consistency across projects. Configured views can be shared via URLs using IIIF-inspired content state encoding.

Key challenges include representing textual variants efficiently. Instead of storing variants as separate items, a token-level annotation strategy (using the output of CollateX) encodes differences as Web Annotations, though scalability issues arise with highly fragmented texts.

The framework preserves heterogeneity while enabling interoperability, transforming isolated editions into a networked scholarly infrastructure. By adopting modular components, institutions can integrate existing projects without sacrificing autonomy. The resulting platform is designed to receive updates and enhancements beyond the project’s conclusion, ensuring long-term sustainability and continued evolution of the system as scholarly practices and technical standards develop.