Question

I’m writing some code that iterates a set of POS tags (generated by pos_tag in NLTK) to search for POS patterns. Matching sets of POS tags are stored in a list for later processing. Surely a regex-style pattern filter already exists for a task like this, but a couple of initial google searches didn’t give me anything.

Are there any code snippets out there that can do my POS pattern filtering for me?

Thanks,
Dave

EDIT: Complete solution (using RegexParser, and where messages is any string)

text = nltk.word_tokenize(message)
tags = nltk.pos_tag(text)
grammar = r"""
    RULE_1: {<JJ>+<NNP>*<NN>*}
    """
chunker = nltk.RegexpParser(grammar)
chunked = chunker.parse(tags)
def filter(tree):
    return (tree.node == "RULE_1")
for s in chunked.subtrees(filter):
    print s

Check out http://nltk.googlecode.com/svn/trunk/doc/book/ch07.html and http://www.regular-expressions.info/reference.html for more on creating the rules.

Answer 1

I think you’re looking for RegexpChunkParser.