Latin scansion with neural networks

Sequence-labelling models mark every syllable of a Latin verse as long —, short ∪ or elided ×. The demo then checks each line against dactylic hexameter and pentameter. Everything runs in your browser.

TransScan Transformers · by Yijun Wang

Baselines · by Nolden (2022)

F1 per label (long / short / elision) as reported in the TransScan paper (Table 4) and, for the CRF, Nolden's thesis.