🔍 BERTScore Demo

BERTScore evaluates the quality of generated text by comparing contextualized embeddings from models like BERT against a reference text.

Unlike n-gram metrics (e.g., BLEU), BERTScore focuses on semantic similarity
and is often better at capturing whether meaning is preserved.

1. Enter a reference text (ground truth).
2. Enter a candidate text (model output or paraphrase).
3. Click Compute BERTScore.