Ancient Greek Macronizers

Enter Ancient Greek text to have the alphas, iotas and ypsilons marked as long or short.

Choose between four models. The default, Ericu950/Stoicheia-macronizer, is a 405M character-level model. Ericu950/oga-macronizer-char is a fast 0.9M character-level model. Two older syllable-level options are available for comparison: Ericu950/SyllaMoBert-grc-macronizer-v1 (ModernBERT 0.1B) and Ericu950/macronizer_mini (RoBERTa 11M).

Made by Albin Thörn Cleland (Lund university) and Eric Cullhed (Uppsala university). Training the models was made possible by resources provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS), partially funded by the Swedish Research Council (grant agreement no. 2022-06725).

Read the paper: Automatic Annotation of Ancient Greek Vowel Length.

Model

Stoicheia Macronizer OGA character model Pretrained ModernBERT Fine-tuned RoBERTa

Greek Lines

Try examples

Plain Output

Python API

from gradio_client import Client

client = Client("Urdatorn/macronizer")
result = client.predict(
    text="μῆνιν ἄειδε θεὰ Πηληϊάδεω Ἀχιλῆος",
    model="stoicheia",
    api_name="/macronize",
)
print(result)

The default is model="stoicheia". Other options are model="oga", model="modernbert", and model="roberta".