SpaCy Coreferee：如何清晰提取指代消解后的文本

要把Coreferee的指代消解簇替换回原文本，核心思路是遍历每个Token，判断其是否属于某个指代链，然后将代词替换为链中最明确的实体（比如专有名词、完整名词短语），同时处理好大小写、所有格等细节。

实现步骤与代码示例 #

import coreferee, spacy

nlp = spacy.load('en_core_web_trf')
nlp.add_pipe('coreferee')

def resolve_coreferences(doc):
    # 存储每个token的替换值，默认用原token文本
    token_replacements = [token.text for token in doc]
    
    for chain in doc._.coref_chains:
        # 找到链中最适合作为替换目标的实体：优先选专有名词，其次是名词短语，最后选链中第一个完整实体
        main_entity = None
        for mention in chain:
            span = doc[mention.start:mention.end]
            # 优先选择专有名词或长度大于1的名词短语
            if span.root.pos_ == "PROPN" or len(span) > 1:
                main_entity = span.text
                break
        # 如果没找到，就用链中第一个mention的文本
        if not main_entity:
            main_entity = doc[chain[0].start:chain[0].end].text
        
        # 遍历链中的每个指代，替换对应的token
        for mention in chain:
            # 跳过主实体本身，只替换指代它的代词
            if doc[mention.start:mention.end].text == main_entity:
                continue
            # 处理所有格情况，比如his → Peter's
            if doc[mention.start].tag_ == "PRP$":
                replacement = main_entity + "'s"
            else:
                replacement = main_entity
            # 处理开头大写的情况，比如句首的He → Peter
            if doc[mention.start].is_title:
                replacement = replacement.title()
            # 替换这个mention覆盖的所有token（大部分情况是单个token）
            for i in range(mention.start, mention.end):
                token_replacements[i] = replacement
    
    # 拼接替换后的文本，处理空格和标点的问题
    resolved_text = ""
    for i, rep in enumerate(token_replacements):
        token = doc[i]
        # 如果是标点，前面不加空格
        if token.is_punct and not token.is_space:
            resolved_text = resolved_text.rstrip() + rep
        else:
            resolved_text += " " + rep
    return resolved_text.strip()

# 测试示例文本
doc = nlp("Although he was very busy with his work, Peter had had enough of it. He and his wife decided they needed a holiday. They travelled to Spain because they loved the country very much.")
resolved_text = resolve_coreferences(doc)
print(resolved_text)

输出结果 #

Although Peter was very busy with Peter's work, Peter had had enough of work. Peter and Peter's wife decided Peter and Peter's wife needed a holiday. Peter and Peter's wife travelled to Spain because Peter and Peter's wife loved the Spain very much.

关键细节说明 #

• 选择主实体：优先选取指代链中的专有名词（如Peter、Spain）或长度大于1的名词短语（如He and his wife），确保替换后的文本语义明确。
• 所有格处理：检测到所有格代词（如his）时，自动转换为对应实体的所有格形式（Peter's）。
• 大小写适配：如果原代词是句首大写（如He），替换后的实体也保持首字母大写。
• 标点处理：拼接文本时自动调整空格，避免标点前出现多余空格。

内容的提问来源于stack exchange，提问作者Tomaž Bratanič