BLEU

# BLEU 점수 계산 예제
from nltk.translate.bleu_score import sentence_bleu, corpus_bleu
from nltk.translate.bleu_score import SmoothingFunction

# 1. 문장 단위 BLEU
reference = [['the', 'cat', 'is', 'on', 'the', 'mat']]
candidate = ['the', 'cat', 'sat', 'on', 'the', 'mat']

# 기본 BLEU (1~4-gram 균등 가중치)
score = sentence_bleu(reference, candidate)
print(f"문장 BLEU: {score:.4f}")

# 개별 n-gram 가중치 지정
weights_1gram = (1, 0, 0, 0)  # BLEU-1
weights_2gram = (0.5, 0.5, 0, 0)  # BLEU-2
print(f"BLEU-1: {sentence_bleu(reference, candidate, weights=weights_1gram):.4f}")
print(f"BLEU-2: {sentence_bleu(reference, candidate, weights=weights_2gram):.4f}")

# 2. Smoothing (짧은 문장이나 n-gram 미일치 시)
smoother = SmoothingFunction()
short_candidate = ['the', 'cat']
smoothed_score = sentence_bleu(
    reference, short_candidate,
    smoothing_function=smoother.method1
)
print(f"Smoothed BLEU: {smoothed_score:.4f}")

# 3. Corpus 단위 BLEU (여러 문장)
references_corpus = [
    [['the', 'cat', 'is', 'on', 'the', 'mat']],
    [['there', 'is', 'a', 'dog', 'in', 'the', 'house']]
]
candidates_corpus = [
    ['the', 'cat', 'sat', 'on', 'the', 'mat'],
    ['a', 'dog', 'is', 'in', 'the', 'house']
]

corpus_score = corpus_bleu(references_corpus, candidates_corpus)
print(f"Corpus BLEU: {corpus_score:.4f}")

# 4. SacreBLEU (표준화된 BLEU, 권장)
# pip install sacrebleu
import sacrebleu
refs = [["The cat is on the mat."]]
hyp = "The cat sat on the mat."
bleu = sacrebleu.sentence_bleu(hyp, refs)
print(f"SacreBLEU: {bleu.score:.2f}")