BLEU (Bilingual Evaluation Understudy) is the standard automatic metric for evaluating machine translation quality. It measures n-gram precision — the overlap of contiguous word sequences (unigrams through typically 4-grams) between the model's translated output and one or more human reference translations — combined with a brevity penalty to discourage overly short translations that could otherwise achieve artificially high precision. BLEU scores range from 0 to 1 (or 0-100 as a percentage), with higher scores indicating closer alignment to reference translations.

The distractors represent metrics standard to other task families: F1 score (A) evaluates classification tasks by balancing precision and recall over discrete positive/negative predictions, ill-suited to open-ended text generation where there is no fixed set of "correct" tokens. Accuracy (B) similarly assumes a discrete correct/incorrect judgment, inappropriate for translation where multiple valid phrasings can convey the same meaning. Mean Absolute Error (C) is a regression metric measuring average magnitude of numeric prediction error, irrelevant to text output evaluation entirely.

It's worth noting BLEU has known limitations — it correlates imperfectly with human judgments of fluency and can penalize valid paraphrases — which has motivated complementary metrics like METEOR, ROUGE (more common for summarization), and learned metrics like BERTScore, though BLEU remains the benchmark most commonly referenced for translation specifically.