Machine-Learning & Corpus Study

THE GENTILESCHI DEBATE THROUGH MACHINE EYES

Rank signals; test assumptions; keep connoisseurship in the loop.

Research in brief

what this actually does
  • Corpus: structured lists for Artemisia and Orazio—split into unquestioned, questioned, and workshop—plus gender-sliced views of a broader Baroque set.
  • Models: (1) an artist classifier (Artemisia vs. Orazio) to read painterly decisions; (2) a gender pattern classifier (Female vs. Male) to check for broad stylistic tendencies, not identity.
  • Output: a ranked list of high-confidence reattributions/confirmations and a bucket of disagreements that warrant human review. It’s a filter, not a verdict.
Everything is static HTML; the analysis report includes per-work modals and training diagnostics.

Explore the corpus

dataset views

Workshop policy

why they’re excluded from training
  • Ambiguity: workshop labels can mix hands (master, assistants, copyists). Training on them bakes in noise.
  • Leakage risk: repeated cartoons/templates across both artists can let a model memorize composition rather than painterly decisions.
  • Policy: workshop images are shown for context, not used for training. Examine them after you’ve reviewed higher-confidence signals.

How to read the results

what to trust first
  • Agreement + high confidence: top of the queue for re-examination.
  • Disagreement: often workshop influence, copies, or transitional periods. Use the modal bars (artist vs gender) to see why.
  • Always cross-check: compare with the static galleries and published scholarship before changing a catalogue entry.
Open the ML report → tabs: Strong Reattribution / Models Disagree / Confirmed.

Scholarship & sources

blunt acknowledgments

Working lists and attributions reference:

  • R. Ward Bissell, Orazio Gentileschi and the Poetic Tradition in Caravaggism, 1981.
  • Keith Christiansen (ed.), Orazio and Artemisia Gentileschi, Metropolitan Museum of Art / Rome, 2001.

Notes: the Orazio pages explicitly cite these; this homepage routes to those files directly.