Training data behind Stoicheia-meter (macronization + scansion) and Stoicheia-macronizer.
Every label is silver: produced by rule, by constraint solving, or converted from published
markup — none of it is hand-annotated. The hand-annotated evaluation benchmark is Norma, in a
separate repository.
There is no gold/silver ambiguity about what is train and what is held out. The training code
(meter/encode.py, meter/train.py in the code… See the full description on the dataset page:
https://huggingface.co/datasets/anonymous-stoicheia/Stoicheia-meter-silver.
