如何用Spacy识别英文名词性别？Python NLP新手技术求助

我是Python Spacy NLP领域的新手，目前在处理主语对应代词的问题时遇到瓶颈：已实现I、we、they等代词的主语匹配，但无法处理he/she/it这类第三人称单数代词对应的名词性别判断。根据语法规则，第三人称单数主语需根据名词性别选择he/she/it，我编写了如下代码：

import spacy

nlp = spacy.load("en_core_web_sm")

def get_gender(sent:str):
    doc = nlp(sent)
    for token in doc:
        if token.dep_ == "nsubj":
            if token.morph.get("Gender") == ["Masc"]:
                return token.text + "-male"
            elif token.morph.get("Gender") == ["Fem"]:
                return token.text + "-female"
            else:
                return token.text + "-unknown"
    return "unknown"

sent1="He is honest"
sent2="She is honest"
sent3="The man is honest"

print(f"{sent1}: {get_gender(sent1)}")
# Output: He is honest: He-male
print(f"{sent2}: {get_gender(sent2)}")
# Output: She is honest: She-female
print(f"{sent3}: {get_gender(sent3)}")
# Output: The man is honest: man-unknown

但该代码仅能识别代词的性别，无法识别名词（如man）的性别。我需要判断主语名词的性别（如Father-阳性、Mother-阴性、Table-中性、Student-通用），请问Spacy是否有内置方法？若没有，能否提供分性别的名词词典？

回答： #

1. Spacy内置能力说明 #

Spacy的默认模型（比如en_core_web_sm）没有内置的名词性别识别功能，它的morph属性仅针对代词（如he/she/it）提供性别标记，对普通名词不会返回性别信息。这是因为英文中大部分名词本身没有语法性别，语义性别需要额外的词典或规则来判断。

2. 解决方案：自定义性别词典 #

你可以通过构建自定义性别词典来实现名词性别判断，结合Spacy的主语识别逻辑即可解决问题。以下是优化后的代码示例：

import spacy

nlp = spacy.load("en_core_web_sm")

# 自定义性别词典，可根据需求扩展
GENDER_DICT = {
    # 阳性名词
    "man": "male",
    "father": "male",
    "son": "male",
    "brother": "male",
    "uncle": "male",
    "king": "male",
    # 阴性名词
    "woman": "female",
    "mother": "female",
    "daughter": "female",
    "sister": "female",
    "aunt": "female",
    "queen": "female",
    # 中性名词
    "table": "neutral",
    "chair": "neutral",
    "book": "neutral",
    "computer": "neutral",
    # 通用名词（无特定性别）
    "student": "neutral",
    "teacher": "neutral",
    "doctor": "neutral",
    "person": "neutral"
}

def get_gender(sent: str):
    doc = nlp(sent)
    for token in doc:
        if token.dep_ == "nsubj":
            # 先检查代词的morph性别
            morph_gender = token.morph.get("Gender")
            if morph_gender == ["Masc"]:
                return f"{token.text}-male"
            elif morph_gender == ["Fem"]:
                return f"{token.text}-female"
            elif morph_gender == ["Neut"]:
                return f"{token.text}-neutral"
            # 再检查词典中的名词性别
            elif token.text.lower() in GENDER_DICT:
                return f"{token.text}-{GENDER_DICT[token.text.lower()]}"
            # 未匹配到的情况返回unknown
            else:
                return f"{token.text}-unknown"
    return "unknown"

# 测试用例
sent1 = "He is honest"
sent2 = "She is honest"
sent3 = "The man is honest"
sent4 = "My mother is kind"
sent5 = "The table is wooden"
sent6 = "The student studies hard"

print(f"{sent1}: {get_gender(sent1)}")
# Output: He is honest: He-male
print(f"{sent2}: {get_gender(sent2)}")
# Output: She is honest: She-female
print(f"{sent3}: {get_gender(sent3)}")
# Output: The man is honest: man-male
print(f"{sent4}: {get_gender(sent4)}")
# Output: My mother is kind: mother-female
print(f"{sent5}: {get_gender(sent5)}")
# Output: The table is wooden: table-neutral
print(f"{sent6}: {get_gender(sent6)}")
# Output: The student studies hard: student-neutral

3. 扩展建议 #

• 可以通过添加更多名词到GENDER_DICT来覆盖更多场景，比如职业、亲属称谓等。
• 对于模糊性的名词（如"doctor"），可根据需求定义为通用中性或保留为unknown。
• 如果需要更全面的语义性别识别，可以手动整理WordNet等语义库中的性别标记来扩展词典。

内容的提问来源于stack exchange，提问作者Shahriar Zarir