Update README.md

README.md

@@ -10,19 +10,14 @@ Submitted to LREC 2026
 ## Model Description
 
 BERnaT is a family of monolingual Basque encoder-only language models trained to better represent linguistic variation—including standard, dialectal, historical, and informal Basque—rather than focusing solely on standard textual corpora. Models were trained on corpora that combine high-quality standard Basque with varied sources such as social media and historical texts, aiming to enhance robustness and generalization across natural language understanding (NLU) tasks.
-
 **Model Types**: Encoder-only Transformer models (RoBERTa-style)
-
 **Languages**: Basque (Euskara)
 
 ## Training Data
 
 The BERnaT family was pre-trained on a combination of:
-
 - Standard Basque corpora (e.g., Wikipedia, Egunkaria, EusCrawl).
-
 - Diverse corpora including Basque social media text and historical Basque books.
-
 - Combined corpora for the unified BERnaT models.
 
 Training objective is masked language modeling (MLM) on encoder-only architectures across medium (51M), base (124M), and large (355M) sizes.