nlp - 添加 SpaCy Tokenizer 异常 : Do not split '>>'

Question

我正在尝试添加一个异常(exception)来识别 '>>' 和 '>> ' 作为开始一个新句子的指示符。例如，

import spacy

nlp = spacy.load('en_core_web_sm')
doc = nlp(u'>> We should. >>No.')

for sent in doc.sents:
    print (sent)

它打印出:

>> We should.
>
>
No.

但是，我希望它打印出来:

>> We should.
>> No. 

提前感谢您的时间!

Answer 1

您需要创建一个 custom component .代码示例提供自定义句子切分example .从文档中，该示例执行以下操作:

Example of adding a pipeline component to prohibit sentence boundaries before certain tokens.

代码(根据您的需要调整示例):

import spacy


def prevent_sentence_boundaries(doc):
    for token in doc:
        if not can_be_sentence_start(token):
            token.is_sent_start = False
    return doc


def can_be_sentence_start(token):
    if token.i > 0 and token.nbor(-1).text == '>':
        return False
    return True

nlp = spacy.load('en_core_web_sm')
nlp.add_pipe(prevent_sentence_boundaries, before='parser')

raw_text = u'>> We should. >> No.'
doc = nlp(raw_text)
sentences = [sent.string.strip() for sent in doc.sents]
for sentence in sentences:
    print(sentence)

输出

>> We should.
>> No.