Background

Humanities research on canonical texts often requires scholars to move between primary sources and extensive secondary literature. This is especially visible in text-based areas of philosophy, where work on texts such as Wittgenstein's On Certainty or Kant's Critique of Pure Reason often depends on tracking how commentators discuss particular passages, propositions, arguments, or concepts. In such cases, researchers are not only concerned with whether a work is cited, but also with where and how it is cited: which passage is invoked, what interpretive role it plays, and how that use relates to broader debates.

Existing bibliographic and citation-analysis tools provide only limited support for this passage-level reconstruction of scholarly engagement. Large-scale computational approaches, such as the in-text reference network analysis of Becker and Culotta (2025), can illuminate broad intellectual lineages across philosophical corpora, but they do not aim to represent the local interpretive function of individual citations. Bibliographic utilities manage citation metadata, but they do not capture how a cited work engages with specific passages or interpretive positions.

The problem is not new. Wittgenstein scholars have produced manual aids for navigating On Certainty, such as Martin's detailed index of terms and contextual uses (1983), and recent surveys map the main interpretative readings of the work (Moyal-Sharrock and Pritchard 2025). Our contribution is not to replace such scholarship, but to provide infrastructure for making some of its working materials explicit, structured, and reusable.

Methodology

We present a pipeline for constructing structured annotated bibliographies that support fine-grained scholarly analysis. The workflow combines bibliographic collection, source alignment, extraction, enrichment, local validation, and publication.

First, candidate secondary works are collected from services such as PhilPapers, PhilArchive, journal platforms, and library catalogues in Zotero, then exported as a local BibTeX file. This file provides the bibliographic spine of the project: citation keys, titles, authors, publication data, DOIs, URLs, and source-specific identifiers. The corresponding PDFs are aligned with these BibTeX records and stored in paper-specific local directories keyed to the bibliography, so renamed downloads, duplicate files, and records from different services can be reconciled against manually checked records rather than automatically extracted metadata.

Second, the aligned PDFs are processed with complementary tools. Docling is used for body-text reconstruction, producing JSON and Markdown with page-level provenance and document blocks such as headings, paragraphs, captions, and footnotes. GROBID (Lopez 2009) is used for reference-list structure, producing TEI from which bibliographic entries can be recovered. OpenAlex, OpenCitations, Wikidata, and WorldCat links are then used as enrichment and verification layers, but only after being checked against the locally extracted reference lists and, where necessary, corrected in the editing interface.

Third, citation contexts are organized as a local knowledge layer from which draft summaries, labels, and interpretive descriptors are generated. These outputs are analytical aids, not final interpretations. They are validated in a local editing interface that displays Docling-derived body text, GROBID-derived references, and the source PDF side by side. Researchers can correct extracted text, add lightweight TEI-style markup such as quote and footnote links, and preserve edits as local annotation data.

Finally, selected datasets are transformed into viewer-ready formats and published through Bibliographic Information Hub (BIH), a lightweight framework for publishing curated bibliographies as stable collections linked to identifiers, records, and resources (https://bibliographic-information-hub.github.io/bih-site/).

Results and Discussion

We apply the pipeline to secondary scholarship on Wittgenstein's On Certainty (Wittgenstein 1969). The resulting passage-centered bibliography helps trace how passages such as OC §115 and §341 are cited, what interpretive roles they play, and where clusters or divergences emerge. For example, the data makes visible a contrast between uses of OC §115 to frame sceptical challenges and the grammar of "knowing," and uses of OC §341 to articulate hinge-like background commitments that structure inquiry.

The case study shows three practical outcomes. First, beginning with Zotero-exported BibTeX keeps the pipeline close to existing scholarly practice while adding a local data layer that connects each record to its PDF, extracted text, reference list, and annotations. Second, the division of labor between automated extraction and the local editing layer reduces manual work by confining intervention to targeted corrections and lightweight semantic links. Third, publication through BIH supports data circulation and reuse. A sample OC §341 collection is available at https://cm3.github.io/bih-cm3/oc341/.

The pipeline occupies a methodological middle ground between macro-scale corpus analysis and micro-level interpretation. Future work will test its generalizability in other text-based humanities fields, including literary studies, where scholars also track how passages, motifs, or concepts are discussed across secondary literature.

Acknowledgements: This work was supported by NIHU Digital Humanities Projects.