Research on a Genre Taxonomy of Western Classics Presented at IPSJ SIG-CH 141

On May 16, 2026, at the 141st meeting of the IPSJ Special Interest Group on Computers and the Humanities (SIG-CH), held at Osaka University’s Toyonaka Campus, Yuki Suzuki (doctoral student at Nagoya University and research assistant on the Humanitext project) and Naoya Iwata presented their paper, “Constructing a Genre Taxonomy of Western Classical Literature Using Large Language Models” (IPSJ SIG Technical Report, Vol. 2026-CH-141, No. 11, pp. 1–6).

Existing databases of Western classical literature — including Wikidata, Perseus, and the Thesaurus Linguae Graecae — mix several distinct axes in their genre classifications, such as form (verse or prose), subject matter, and metre. This inconsistency hampers both search and analysis. The study reported an attempt to use LLMs to classify roughly 2,500 classical works according to an original two-level genre taxonomy.

What distinguishes the approach is that the LLM serves not merely as a classification engine but as a tool for testing and improving the taxonomy itself. The model supplies a rationale and confidence score for each classification; low-confidence cases and unstable assignments can then reveal problems such as uneven category granularity or blurred boundaries, which in turn inform iterative refinement. The resulting genre metadata is intended to support cross-corpus search and large-scale analysis in Humanitext Antiqua.