BLEU Score

INFO

Evaluates the quality of machine translation by comparing output with reference translations.

How It Works

BLEU measures the precision of n-grams in the candidate translation that appear in the reference translation.
It includes a brevity penalty to discourage overly short outputs.

$$\text{BLEU}=BP\cdot \exp \left(\sum _{n=1}^N{w}_n\log p_n\right)$$

• $p_n$: Precision of n-grams of size $n$
• $w_n$: Weight for each n-gram level (typically uniform)
• $BP$: Brevity penalty to penalize short outputs

What to Look For

• Higher BLEU = better alignment with reference
• Sensitive to exact word matches, not semantics
• Best for machine translation, but also used in summarization and captioning

Application Models

• Transformer
• Recurrent Neural Network (RNN)