A metric that evaluates text similarity by comparing contextual embeddings from BERT rather than exact word matches.