tartuNLP/est-roberta-vm-morph-homonym-tagging

Model description

est-roberta-vm-morph-homonym-tagging ( a.k.a Ro-v2-H ) is an Est-RoBERTa based expert model specifically finetuned for tagging homonymous word forms in Estonian. The model uses partofpseech and morphological features from Vabamorf's tagset and focuses on disambiguating homonymous words from inflectional types 1, 16, 17 and 19. The model has been created by Sander Saska (2026), and the finetuning and evaluation code is available here.

Important: This model has specialized knowledge of homonymy disambiguation, but its general morphological tagging ability can be lacking (likely due to catastrophic forgetting). Therefore, do NOT use this model alone for general text; use MoE variant or as an expert component.