要根据给定英文句子通过替换其中积极情感词为消极情感词来生成对应反事实样本，可以借助 Python 中的`nltk`库和情感词典来实现。以下是一个示例代码： ```python import nltk from nltk.corpus import sentiwordnet as swn from nltk.tokenize import word_tokenize nltk.download('sentiwordnet') nltk.download('punkt') nltk.download('wordnet') # 定义积极和消极情感词的映射，这里只是简单示例 positive_to_negative = { "happy": "sad", "good": "bad", "wonderful": "terrible", "amazing": "awful" } def generate_counterfactual(sentence): tokens = word_tokenize(sentence) counterfactual_tokens = [] for token in tokens: # 检查是否为积极情感词 synsets = list(swn.senti_synsets(token)) if synsets: pos_score = synsets[0].pos_score() neg_score = synsets[0].neg_score() if pos_score > neg_score and token in positive_to_negative: counterfactual_tokens.append(positive_to_negative[token]) else: counterfactual_tokens.append(token) else: counterfactual_tokens.append(token) counterfactual_sentence = " ".join(counterfactual_tokens) return counterfactual_sentence # 示例句子 example_sentence = "I am very happy today." counterfactual = generate_counterfactual(example_sentence) print(counterfactual) ``` 在上述代码中，首先定义了一个简单的积极情感词到消极情感词的映射字典。然后，对输入的句子进行分词，对于每个词，使用`nltk`的`sentiwordnet`库来判断其情感极性。如果是积极情感词且在映射字典中，则将其替换为对应的消极情感词，最后将处理后的词重新组合成句子。