← Leibniz

Methodology

An honest account of how this artifact was constructed, what decisions were made, and where it falls short. Read this before citing or critiquing the ontology proper.

Source text

The corpus is George Martin Duncan's The Philosophical Works of Leibnitz, second edition (1890), digitized by Cornell University Library and made available through the Internet Archive. Duncan's translation is in the public domain. The volume contains thirty-six selections in roughly chronological order, spanning Leibniz's mature philosophical career from 1679 to 1716.

The Akademie edition of Leibniz's collected works (in progress since 1923) is the standard scholarly resource and is far more comprehensive than Duncan, including correspondence, drafts, and manuscripts not yet published when Duncan compiled his anthology. This ontology is bounded by Duncan's selection and inherits the editorial judgments embedded in that selection. A more comprehensive ontology drawing on the Akademie edition would be a substantial follow-up project.

Where Duncan's translation is unclear by current scholarly standards, this affects what the proposer extracted and how. OCR artifacts in the digitized text occasionally produced garbled passages; these were typically caught and rejected during extraction but a small number may have produced atypical class commitments.

Pipeline

The ontology was constructed using BFO-Agent, the same dialogue architecture used to produce the companion Spinoza ontology. The architecture combines large language model extraction with reasoner-based validation:

Extraction

The corpus was chunked and fed sequentially to the proposer (Anthropic Claude, Sonnet variant). For each chunk, the model identified candidate metaphysical claims, proposed BFO-typed class structures to formalize them, and emitted source quotes establishing provenance.

Validation

Each proposal was tested against the existing ontology by the HermiT description-logic reasoner. Proposals producing inconsistency, or violating BFO's standard disjointness axioms, were rejected automatically. Of 1,476 candidate claims, 1,382 were committed; 66 were rejected as inconsistent and 28 produced extraction errors.

Commit

Accepted proposals were merged into the working ontology and committed to git, with the source quote and claim text as commit metadata.

Finalization

The ontology was finalized as a frozen artifact on May 7, 2026. Finalization marks the ontology as read-only.

Decisions worth flagging

The Leibnizian vocabulary presented several BFO-mapping challenges that the Spinoza extraction did not. The decisions made here are tentative and would benefit from scholarly review.

Monads

Monads — the simple, indivisible, soul-like substances that for Leibniz constitute the ultimate furniture of reality — were typed predominantly as BFO independent continuants. This captures their status as bearers of qualities and dispositions and their persistence through time. The proposer's classification reflects, where applicable, the further distinction between simple monads, dominating monads, animal souls, and rational souls (a four-fold typology Leibniz makes in the late writings).

Perception and appetition

The two activities Leibniz attributes to all monads — perception and appetition — were typed as BFO dispositions or processes depending on the local commitment. Where a passage treats perception as the intrinsic capacity of a monad, it is rendered as a disposition. Where a passage treats it as an actual unfolding event, it is rendered as a process.

Pre-established harmony

The relation between mind and body, and more generally between any two monads in the Leibnizian universe, was extracted but not given a clean BFO mapping. Pre-established harmony is a relational structure across all of creation that does not localize cleanly to any standard BFO category. The proposer typically rendered specific harmonic correspondences as relational qualities or as participation in shared processes; readers should be cautious about reading these as Leibniz's full doctrine.

Possible worlds and contingency

Leibniz's modal commitments — possible worlds, contingency, the principle of sufficient reason — were partially extracted as classes but the ontology does not capture modal logic structurally. OWL 2 DL has no native modal operators. Where Leibniz says "this proposition is necessary," the ontology has a class NecessaryTruth, but the modal structure that would distinguish necessity from contingency at a logical level is absent. Readers interested in formal modal analysis of Leibniz should treat this ontology as a starting catalogue, not a structural rendering.

The continuum

Leibniz's distinction between the metaphysical, mathematical, and physical continua was preserved through separate classes where the source text made the distinction explicit. Where the text speaks of "the continuum" without specifying which sense, the proposer typed it as the metaphysical continuum by default. This default may be incorrect for some passages.

A known issue: degraded provenance

This corpus has a meaningful difference from the Spinoza site. Due to a logging regression in BFO-Agent during this run, the per-class provenance was not preserved. The OWL artifact and its 1,837 classes / 722 individuals reflect 1,382 successful extractions — but the bidirectional navigation between source passages and the OWL classes they produced is unavailable.

What this means in practice:

The git commit history and the job file's claim records together preserve enough information to confirm that 1,382 specific source quotes produced specific committed proposals. What was lost is the commit-by-commit record of which OWL entities each proposal added. Reconstructing that mapping post hoc would require either re-running the extraction (with proper logging) or applying inferential heuristics that would introduce uncertainty into the displayed provenance. The decision was made to ship the artifact with the missing information honestly disclosed rather than reconstructed approximately.

The companion Spinoza site has full per-class provenance and demonstrates what the intended user experience looks like. Future ontologies will be extracted with corrected logging and will match Spinoza in fidelity.

Limitations

Translation drift. Duncan's 1890 translation is older than current scholarly conventions; some terminology and emphasis is now considered misleading. The ontology inherits these limitations.

Selection bias. Duncan's anthology reflects a particular nineteenth-century reading of what counts as Leibniz's important philosophical work. Other selections would produce other ontologies.

Single rendering. The ontology represents one reading of Leibniz, by one extractor, validated by one reviewer. Different scholars would produce different ontologies.

Reasoner caveats. HermiT validates consistency under standard OWL 2 DL semantics. Leibnizian metaphysics involves modal claims, infinitesimal magnitudes, and relational structures that OWL 2 DL renders only approximately.

Proposer biases. The language model used for extraction has prior associations with Leibniz's text from training data. Where the model misreads a passage, the misreading propagates to a proposal. Author review caught most such errors but not all.

Provenance. As described above, per-class provenance is not navigable for this corpus. This is a real limitation of the artifact and is not present in the companion Spinoza ontology.

References

Arp, R., Smith, B., and Spear, A. D. (2015). Building Ontologies with Basic Formal Ontology. MIT Press.

Duncan, G. M., ed. (1890). The Philosophical Works of Leibnitz, second edition. New Haven: Tuttle, Morehouse, & Taylor. Public domain. Sourced via Internet Archive.

Glimm, B., Horrocks, I., Motik, B., Stoilos, G., and Wang, Z. (2014). HermiT: An OWL 2 reasoner. Journal of Automated Reasoning 53(3), 245–269.

Smith, B. and Ceusters, W. (2010). Ontological realism: A methodology for coordinated evolution of scientific ontologies. Applied Ontology 5(3-4), 139–188.