Python和NLTK:如何分析句子语法？

Question

我有这段代码，它应该根据定义的语法显示句子的句法结构。然而，它正在返回一个空的[]。我错过了什么或者做错了什么？

import nltk

grammar = nltk.parse_cfg("""
S -> NP VP 
PP -> P NP
NP -> Det N | Det N PP 
VP -> V NP | VP PP
N -> 'Kim' | 'Dana' | 'everyone'
V -> 'arrived' | 'left' |'cheered'
P -> 'or' | 'and'
""")

def main():
    sent = "Kim arrived or Dana left and everyone cheered".split()
    parser = nltk.ChartParser(grammar)
    trees = parser.nbest_parse(sent)
    for tree in trees:
        print tree

if __name__ == '__main__':
    main()

Answer 1

让我们做一些反向工程：

>>> import nltk
>>> grammar = nltk.parse_cfg("""
... NP -> Det N | Det N PP
... N -> 'Kim' | 'Dana' | 'everyone'
... """)
>>> sent = "Kim".split()
>>> parser = nltk.ChartParser(grammar)
>>> print parser.nbest_parse(sent)
[]

看上去这些规则连第一部作品都认不出是NP。所以让我们试试注射NP -> N

>>> import nltk
>>> grammar = nltk.parse_cfg("""
... NP -> Det N | Det N PP | N
... N -> 'Kim' | 'Dana' | 'everyone'
... """)
>>> sent = "Kim".split()
>>> parser = nltk.ChartParser(grammar)
>>> print parser.nbest_parse(sent)
[Tree('NP', [Tree('N', ['Kim'])])]

现在它开始工作了，让我们继续Kim arrived or Dana and

>>> import nltk
>>> grammar = nltk.parse_cfg("""
... S -> NP VP
... PP -> P NP
... NP -> Det N | Det N PP | N
... VP -> V NP | VP PP
... N -> 'Kim' | 'Dana' | 'everyone'
... V -> 'arrived' | 'left' |'cheered'
... P -> 'or' | 'and'
... """)
>>> sent = "Kim arrived".split()
>>> parser = nltk.ChartParser(grammar)
>>> print parser.nbest_parse(sent)
[]
>>> 
>>> sent = "Kim arrived or".split()
>>> parser = nltk.ChartParser(grammar)
>>> print parser.nbest_parse(sent)
[]

似乎没有办法得到VP与或不带P，因为V要么需要一个NP之后，或者它必须向上树成为一个VP在采取P之前，所以它是放松规则，说VP -> V PP而不是VP -> VP PP

>>> import nltk
>>> grammar = nltk.parse_cfg("""
... S -> NP VP
... PP -> P NP
... NP -> Det N | Det N PP | N
... VP -> V NP | V PP
... N -> 'Kim' | 'Dana' | 'everyone'
... V -> 'arrived' | 'left' |'cheered'
... P -> 'or' | 'and'
... """)
>>> sent = "Kim arrived or Dana".split()
>>> parser = nltk.ChartParser(grammar)
>>> print parser.nbest_parse(sent)
[Tree('S', [Tree('NP', [Tree('N', ['Kim'])]), Tree('VP', [Tree('V', ['arrived']), Tree('PP', [Tree('P', ['or']), Tree('NP', [Tree('N', ['Dana'])])])])])]

好吧，我们越来越接近了，但下一个词似乎又违反了cfg的规则：

>> import nltk
>>> grammar = nltk.parse_cfg("""
... S -> NP VP
... PP -> P NP
... NP -> Det N | Det N PP | N
... VP -> V NP | V PP
... N -> 'Kim' | 'Dana' | 'everyone'
... V -> 'arrived' | 'left' |'cheered'
... P -> 'or' | 'and'
... """)
>>> sent = "Kim arrived or Dana left".split()
>>> parser = nltk.ChartParser(grammar)
>>> print parser.nbest_parse(sent)
[]
>>> sent = "Kim arrived or Dana left and".split()
>>> parser = nltk.ChartParser(grammar)
>>> print parser.nbest_parse(sent)
[]
>>> 
>>> sent = "Kim arrived or Dana left and everyone".split()
>>> parser = nltk.ChartParser(grammar)
>>> print parser.nbest_parse(sent)
[]
>>> 
>>> sent = "Kim arrived or Dana left and everyone cheered".split()
>>> parser = nltk.ChartParser(grammar)
>>> print parser.nbest_parse(sent)
[]

所以，我希望上面的例子告诉你，试图改变规则，把语言现象从左到右结合起来是很困难的。

而不是从左到右，实现

[[[[[[[[Kim] arrived] or] Dana] left] and] everyone] cheered]

为什么你不试着制定更合理的语言规则来实现：

1. [[[Kim arrived] or [Dana left]] and [everyone cheered]]
2. [[Kim arrived] or [[Dana left] and [everyone cheered]]]

试一试：

import nltk
grammar = nltk.parse_cfg("""
S -> CP | VP 
CP -> VP C VP | CP C VP | VP C CP
VP -> NP V 
NP -> 'Kim' | 'Dana' | 'everyone'
V -> 'arrived' | 'left' |'cheered'
C -> 'or' | 'and'
""")

print "======= Kim arrived ========="
sent = "Kim arrived".split()
parser = nltk.ChartParser(grammar)
for t in parser.nbest_parse(sent):
    print t

print "\n======= Kim arrived or Dana left ========="
sent = "Kim arrived or Dana left".split()
parser = nltk.ChartParser(grammar)
for t in parser.nbest_parse(sent):
    print t 

print "\n=== Kim arrived or Dana left and everyone cheered ===="
sent = "Kim arrived or Dana left and everyone cheered".split()
parser = nltk.ChartParser(grammar)
for t in parser.nbest_parse(sent):
    print t

[out]

======= Kim arrived =========
(S (VP (NP Kim) (V arrived)))

======= Kim arrived or Dana left =========
(S (CP (VP (NP Kim) (V arrived)) (C or) (VP (NP Dana) (V left))))

=== Kim arrived or Dana left and everyone cheered ====
(S
  (CP
    (CP (VP (NP Kim) (V arrived)) (C or) (VP (NP Dana) (V left)))
    (C and)
    (VP (NP everyone) (V cheered))))
(S
  (CP
    (VP (NP Kim) (V arrived))
    (C or)
    (CP
      (VP (NP Dana) (V left))
      (C and)
      (VP (NP everyone) (V cheered)))))

上面的解决方案显示了您的CFG规则需要足够健壮，不仅能够捕获完整的句子，而且也能捕获句子的一部分。

Answer 2

您没有在语法中定义Det，但是每个NP (以及相应的S)都必须有一个语法定义。

相比较

>>> grammar = nltk.parse_cfg("""
... S -> NP VP
... NP -> Det N | Det N PP
... VP -> V NP | VP PP
... Det -> 'a' | 'the'
... N -> 'Kim' | 'Dana' | 'everyone'
... V -> 'arrived' | 'left' |'cheered'
... """)
>>>
>>> parser = nltk.ChartParser(grammar)
>>> parser.nbest_parse('the Kim left a Dana'.split())
[Tree('S', [Tree('NP', [Tree('Det', ['the']), Tree('N', ['Kim'])]), Tree('VP', [Tree('V', ['left']), Tree('NP', [Tree('Det', ['a']), Tree('N', ['Dana'])])])])]