ModelRefs / BERTScore — AI Glossary
BERTScore — AI Glossary
A semantic similarity metric using contextual BERT embeddings to compare generated and reference texts beyond surface n-gram overlap.
Overview
BERTScore (Zhang et al. 2019) computes token-level cosine similarity between reference and candidate embeddings, taking maximum similarity for each reference token (recall) and candidate token (precision). Correlates better with human judgment than BLEU/ROUGE on summarization and translation. Computed using transformers library.
Reference details
| Topic | evaluation |
|---|---|
| Last reviewed | 2026-06-24 |
Related terms
Primary source
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to BERTScore — AI Glossary.
Frequently asked questions
What is BERTScore?
A semantic similarity metric using contextual BERT embeddings to compare generated and reference texts beyond surface n-gram overlap.
What concepts are related to BERTScore?
Closely related concepts include bleu, rouge, embedding model.