UppsalaNLP at EvaLatin 2026: Multilingual parsing for Latin

Sara Stymne


Abstract
We describe the UppsalaNLP submission to the EvaLatin dependency parsing shared task. We explore using an out-of-the-box parser in combination with multi-treebank training on Latin and multilingual training on other ancient languages. Adding additional languages yields only small gains, but the results vary across treebanks and genres, with the largest positive effect for poetry. Our systems perform best in the shared task for prose but are less competitive for poetry, indicating the need for genre adaptation.
Anthology ID:
2026.lt4hala-1.44
Volume:
Proceedings of the Fourth Workshop on Language Technologies for Historical and Ancient Languages (LT4HALA 2026) @ LREC 2026
Month:
May
Year:
2026
Address:
Palma, Mallorca (Spain)
Editors:
Rachele Sprugnoli, Marco Passarotti
Venues:
LT4HALA | WS
SIG:
Publisher:
ELRA Language Resources Association (ELRA)
Note:
Pages:
429–436
Language:
External URL:
https://lrec.elra.info/lrec2026-ws-lt4hala-44
DOI:
10.63317/57u2jdmxj3ut
Bibkey:
Cite (ACL):
Sara Stymne. 2026. UppsalaNLP at EvaLatin 2026: Multilingual parsing for Latin. In Proceedings of the Fourth Workshop on Language Technologies for Historical and Ancient Languages (LT4HALA 2026) @ LREC 2026, pages 429–436, Palma, Mallorca (Spain). ELRA Language Resources Association (ELRA).
Cite (Informal):
UppsalaNLP at EvaLatin 2026: Multilingual parsing for Latin (Stymne, LT4HALA 2026)
Copy Citation: