BleuScore Online

Crate PyPI NPM GitHub

N-gram Order

Smoothing

Enable

Reference Length Mode

Reference length mode

Shortest uses the shortest reference for brevity penalty and matches HuggingFace evaluate / TF NMT behavior.

Closest uses the reference length closest to each prediction, with shorter winning ties, matching SacreBLEU / NIST mteval-v13a.

Predictions & References

Add your machine translation predictions and their corresponding reference translations.

Calculating BLEU score...

Error

0.0000

-

Detailed Metrics

BLEU Score -

Brevity Penalty -

Length Ratio -

Translation Length -

Reference Length -

N-gram Precisions

Ready to Calculate

Add your predictions and references, then click "Calculate BLEU Score" to see the results.

Created by Mathew Shen • Powered by Rust + WebAssembly