Hungarian morphological generator model with mT5

For further models, scripts and details, see our demo site.

Pretrained model used: mT5
Prefix: "morph: "
UD-based generation

Limitations

max_source_length = 64
max_target_length = 32

Results

Model	emMorph	UD
mT5	95.53	94.66

Usage with pipeline

from transformers import pipeline

text2text_generator = pipeline(task="text2text-generation", model="NYTK/morphological-generator-ud-mt5-hungarian")

print(text2text_generator("morph: munka NOUN Case=Acc|Number=Sin")[0]["generated_text"])

Citation

If you use this model, please cite the following paper:

@inproceedings {morph-generator,
    title = {Neural Morphological Generators for Hungarian},
    booktitle = {XIX. Magyar Számítógépes Nyelvészeti Konferencia (MSZNY 2023)},
    year = {2023},
    publisher = {Szegedi Tudományegyetem, Informatikai Intézet},
    address = {Szeged, Hungary},
    author = {Laki, László János and Ligeti-Nagy, Noémi and Vadász, Noémi and Yang, Zijian Győző},
    pages = {331--340}
}