Keynote CLARIAH-AT Summer School 2026 - Machine Learning for Digital Scholarly Editions

This contribution offers a methodological reflection on context engineering applied to digital scholarly editing, drawing on experiences from the Italian research landscape: primarily CNR-ILC, CLARIN-IT, Ca' Foscari University of Venice, and the University of Pisa, with connections to the Universities of Catania and Naples Federico II, the Italian Institute of Germanic Studies (IISG, Rome), and others.

Context engineering denotes the systematic design of the informational environment where language models operate: sources, constraints, terminologies, standards, and tools that guide and delimit their contribution within controlled philological workflows.

Three levels are discussed. The first concerns workflows: documented pipelines from manuscript to edition (HTR/ATR, assisted correction, TEI encoding, collation), where AI assistance is restricted to verifiable tasks. The second concerns infrastructures: the role of CLARIN, DARIAH, E-RIHS, OPERAS, and knowledge centres, as providers of data, standards, and services constituting a shared, reusable context for these workflows. The third, more theoretical, proposes modal logics as a meta-level for making explicit the bouletic, epistemic, and doxastic attitudes — establishing, admitting, preferring, explaining, interpreting — of human editors and AI agents within textual communities.

The aim is to outline transferable, assessable practices consistent with established standards in digital philology.