Researchers have introduced an interpretable parametric representation and an open dataset of 1,203 oracle-bone script glyph forms, establishing a structured alternative to opaque deep-learning techniques. In a study published in npj Heritage Science, authors Qing-sheng Li and Yu-lin Bian present a vector-based methodology designed to address key limitations in computational oracle bone script research. The new open-access release counters this trend by introducing an interpretable mathematical formulation that couples parametric character modeling with modern digital character assignments.
At the core of the framework is a stroke-level geometric formulation where each stroke sub-segment is encoded using three closed-form parameters: direction α, magnitude η, and signed curvature κ. Within the 1,203-glyph dataset, each entry is paired with a modern-character assignment, and where that modern character is officially encoded, the authors supply the corresponding Unicode code point.
To derive these parametric descriptions accurately, the authors introduced an error-driven adaptive arc-segmentation method that operates via residual-guided splitting. In addition to the adaptive segmentation method, the research team delivered a corpus-level structural characterisation of the script, an empirical controllability demonstration, and a resolution-independent Scalable Vector Graphics (SVG) glyph set.
Quantitative benchmarking confirmed that the parametric representation reconstructs ancient glyph forms with high precision. Across the digitised corpus, the representation achieves a median normalised reconstruction error of 0.0021 at approximately 10 sub-segments per glyph.
Beyond character reconstruction, the parametric corpus provides mathematical validation for established palaeographical observations. Structural measurements revealed that 72.0 percent of stroke sub-segments turn through less than 0.16 radians, quantitatively substantiating the traditional palaeographic perspective that oracle bone script strokes are predominantly simple, low-curvature arcs. Li and Bian emphasize, however, that the dataset operates under specific methodological boundaries. The collection is designed strictly as a single-representative-glyph-per-character vector resource, meaning it serves as an analytical complement to primary palaeographical materials and physical inscriptions rather than a definitive substitute for original archaeological sources.
Subject of Research: Parametric representation and vector dataset for oracle bone script
Article Title: An interpretable parametric representation and open dataset for oracle bone script: adaptive Arc segmentation and SVG resource
Article References: Li, Q.-S., & Bian, Y.-L. (2026). An interpretable parametric representation and open dataset for oracle bone script: adaptive Arc segmentation and SVG resource. npj Heritage Science. https://doi.org/10.1038/s40494-026-02995-6
Image Credits: AI Generated
DOI: 10.1038/s40494-026-02995-6
Keywords: oracle bone script, parametric representation, arc segmentation, vector dataset, SVG glyphs, residual-guided splitting, palaeography, stroke geometry, npj Heritage Science, closed-form parameters, curvature analysis, Unicode mapping, computational palaeography, reconstruction error, controllability demonstration, structural characterisation
Cite Scienmag News
Courtney Benton. (November 16, 2026). Parametric Model and Open Dataset Reconstruct Ancient Oracle Bone Script. Scienmag. https://scienmag.com/parametric-model-and-open-dataset-reconstruct-ancient-oracle-bone-script/
Courtney Benton. "Parametric Model and Open Dataset Reconstruct Ancient Oracle Bone Script." Scienmag, 16 November 2026, https://scienmag.com/parametric-model-and-open-dataset-reconstruct-ancient-oracle-bone-script/. Accessed 16 November 2026.
Courtney Benton. "Parametric Model and Open Dataset Reconstruct Ancient Oracle Bone Script." Scienmag. November 16, 2026. https://scienmag.com/parametric-model-and-open-dataset-reconstruct-ancient-oracle-bone-script/








