A Reproducible Universal Dependencies-Style Pipeline for Katharevousa Greek Parliamentary Text

In three linesReproducible pipeline for building a Universal Dependencies parsing resource for Katharevousa Greek (post-junta parliamentary texts). Dataset of 1,697 annotated sentences, comparison of 6 architectures (spaCy, Stanza, XLM-R, mBERT, etc.). Best model (XLM-R): 0.5162 LAS vs 0.4183 baseline. Code and annotations released open-access.

Summary generated by Claude — human-verified