The Estonian languge specific BERT model EstBERT was trained as part of the EKTB11 project. This page is dedicated to benchmarking EstBERT on various Estonian language NLP tasks.
The reports in both Estonian and English are in progress.
Text Classification tasks
The experiments are conducted on the Estonian Valence Dataset. The paragraphs in the corpus originate from aricles of different rubrics and are annotated with discrete sentiment labels (positive, negative, neutral, ambiguous).
- Fine-tuned Rubric classification model in Huggingface