Author(s) :

Aidan Han.

Volume/Issue :

Abstract :

This narrative review compares contemporary Latin Natural Language Processing (NLP) tools and their use cases for pedagogical applications. While NLP has advanced significantly for high-resource modern languages, Latin remains underrepresented due to its complex morphology, flexible word order, orthographic variation, and limited annotated corpora. To address this gap, this review examines five prominent Latin NLP systems, LatinBERT, Stanza, LatinCy, LemLat 3.0, and Lamon/LamonPy, across architecture, training data, task performance, and educational relevance. Drawing on published evaluation metrics from peer-reviewed sources, including part-of-speech tagging, lemmatization, dependency parsing, and word sense disambiguation, this study analyzes the strengths and limitations of each approach. Transformer-based models demonstrate strong contextual understanding and high accuracy in disambiguation tasks, while rule-based systems offer transparency and reliability for vocabulary learning. Pipeline models provide comprehensive syntactic analysis but show performance variability across text genres. The reviewed evidence suggests that no single model performs optimally across all tasks or corpora, and that performance is strongly influenced by alignment between training and evaluation data. The review concludes that an integrated, multi-tool approach is most effective for supporting Latin pedagogy and outlines directions for future research, including standardized benchmarking and classroom-based evaluation.