Greedy matching of complex names in CoreNLP XML

gistfile1.py

# This script matches names like "Mr. Bennet" (and not just "Bennet")
# in the XML output of CoreNLP, which has one word per token

import xml.etree.ElementTree as ET

e = ET.parse('test.xml').getroot()

word_list = []
character_dict = dict()

for line in open('characters.tsv','r'):
    for character in line.split('\t'):
        character_parts = character.strip().encode('utf8').split()
        element = character_dict
        for part in reversed(character_parts):
            if part not in element:
                element[part] = dict()
            element = element[part]
        element['character_name'] = character.strip().encode('utf8')
        element['mentions'] = []

for index,token_tag in enumerate(e.iter('token')):
    word_tag = token_tag[0]
    element = character_dict
    word = word_tag.text
    while word in element:
        element = element[word]
        if index == 0:
            break
        word = word_list[index-1]
    if 'mentions' in element:
        element['mentions'].append(index)
    word_list.append(word_tag.text.encode('utf8'))

def get_mentions(element):
    for key in element:
        if key != 'mentions' and key != 'character_name':
            get_mentions(element[key])
    if 'character_name' in element:
        print element['character_name'],element['mentions']

get_mentions(character_dict)