On the implementation of Latin part-of-speech taggers in intertextuality analysis: TreeTagger, CLTK, Cracovia system, LatinCy, and ChatGPT compared.

The article focuses on evaluating Latin part-of-speech (POS) taggers to improve digital-assisted intertextuality analysis by filtering text matches based on grammatical criteria derived from Mare Revellio's historical text-reuse grammar (HTRG). The study compares several POS taggers—including the transformer-based Cracovia system, the probabilistic TreeTagger, the Classical Language Toolkit (CLTK), LatinCy, and ChatGPT—using Latin texts by Virgil and Jerome. Results show that the Cracovia system achieves the highest accuracy (F2-measure of 0.9627), benefiting from transformer architecture and enriched training data, while ChatGPT, despite high precision, has lower recall and is less reliable for this specific task. The authors recommend applying POS tagging to whole sentences with context-aware comparison (implementation 3b) to best approximate traditional hermeneutic methods and enhance citation detection in ancient Latin literature.