-
Notifications
You must be signed in to change notification settings - Fork 1
/
diff_tokenize.py
166 lines (134 loc) · 5.27 KB
/
diff_tokenize.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
# -*- coding: utf-8 -*-
import os
import re
import fileinput
import spacy
import time
from collections import defaultdict
from nltk.tokenize import sent_tokenize, word_tokenize
from spacy.lang.en import English
from spacy.attrs import ORTH
import kenlm
KnModel = kenlm.Model('../jjc/1b.bin')
SPACY_MODEL = os.environ.get('SPACY_MODEL', 'en')
nlp = spacy.load(SPACY_MODEL)
tokenizer = English().Defaults.create_tokenizer(nlp)
edit_re = re.compile('\[\-(((?!\[\-).)*?)\-\]|\{\+(((?!\{\+).)*?)\+\}')
sep_sents_re = r'\w\.\w'
def init_tokenizer_option():
# add special segmenting case for spacy tokenizer
tokenizer.add_special_case("{}", [{ORTH: "{}"}])
tokenizer.add_special_case("{{", [{ORTH: "{{"}])
tokenizer.add_special_case("}}", [{ORTH: "}}"}])
def restore_line_break(text):
return text.replace('<br/>', '\n').replace('<br>', '\n')
def restore_xmlescape(text):
while '&' in text:
text = text.replace('&', '&')
text = text.replace('"e;', '"')
text = text.replace('"', '"')
text = text.replace(' ', ' ')
text = text.replace('<', '<')
text = text.replace('>', '>')
return text
def tokenize_edit(edit_token):
def _tokenize(text):
# remove nested edit tokens
while edit_re.search(text):
text = edit_re.sub(' ', text).strip()
# nltk word_tokenize
return '\u3000'.join(token for token in word_tokenize(text))
# spacy word_tokenize (TODO: clean fixed space token)
# return '\u3000'.join(token.text for token in tokenizer(text.replace('\u3000', ' ')))
if edit_token.startswith('[-'):
if edit_token.endswith('+}'): # replace
delete, insert = edit_token[2:-2].rsplit('-]{+', 1)
delete, err_type = delete.rsplit('//', 1)
insert, err_type = insert.rsplit('//', 1)
delete, insert = _tokenize(delete), _tokenize(insert)
return '[-{0}//{2}-]{{+{1}//{2}+}}'.format(delete, insert, err_type)
else: # delete
delete, err_type = edit_token[2:-2].rsplit('//', 1)
delete = _tokenize(delete)
return '[-{0}//{1}-]'.format(delete, err_type)
else: # insert
insert, err_type = edit_token[2:-2].rsplit('//', 1)
insert = _tokenize(insert)
return '{{+{0}//{1}+}}'.format(insert, err_type)
def mask_edits(text):
edits = []
tokens = []
for token in text.split(' '):
token = token.strip()
if token.startswith('{+') or token.startswith('[-'):
tokens.append('{}')
edits.append(tokenize_edit(token))
elif token:
tokens.append(token.replace('{', ' {{ ').replace('}', ' }} '))
return ' '.join(token.strip() for token in tokens), edits
def seperate_sents(last_sent):
seg_sents = []
keep_seg = True
bef_sent = ''
while keep_seg:
com_sent, last_sent = last_sent.split('.', 1)
com_sent = bef_sent + com_sent
com_sent += '.'
if not last_sent: # prevent it is the last sent
seg_sents.append(reorganize_sent(com_sent))
break
com_token = ' '.join(word_tokenize(com_sent))
last_token = ' '.join(word_tokenize(last_sent))
last_token = last_token.capitalize()
com_token = com_token.capitalize()
sep_scores = KnModel.score(com_token) + KnModel.score(last_token)
total_scores = KnModel.score(com_token + ' ' + last_token)
if sep_scores > total_scores: # should seperate these sentences
bef_sent = ''
seg_sents.append(reorganize_sent(com_sent))
else:
bef_sent = com_sent
if not re.search(r'\.', last_sent):
keep_seg = False
seg_sents.append(reorganize_sent(last_sent))
return seg_sents
def reorganize_sent(sent):
return ' '.join(token.text for token in tokenizer(sent) if token.text).strip()
def tokenize_doc(text):
text = restore_line_break(text)
text = restore_xmlescape(text)
# mask edit tokens first to prevent being segmented
# I have {+a+} pen. => I have {} pen.
text_masked, edits = mask_edits(text)
tokenized_sents = []
for line in text_masked.splitlines():
# sentence tokenize (using nltk)
for sent in sent_tokenize(line.strip()):
# print(sent)
# sent sep again (using nltk)
# if re.search(sep_sents_re, sent):
# tokenized_sents += seperate_sents(sent)
# else:
# word tokenize (using spacy)
# else:
# text = ' '.join(token.text for token in tokenizer(sent) if token.text).strip()
# tokenized_sents.append(reorganize_sent(sent))
tokenized_sents.append(sent)
# restore masked edit
return '\n'.join(tokenized_sents).format(*edits)
def main():
revise_type = defaultdict(int)
for doc in fileinput.input():
doc = doc.strip()
# print('#', 'doc', '=', doc)
print(tokenize_doc(doc))
return revise_type
if __name__ == '__main__':
start = time.time()
init_tokenizer_option()
main()
end = time.time()
#print('Time Cost: ', end - start)
#print('edit count')
#for key, count in revise_type.items():
#print('edit_count: ', key,'times: ', count)