-
Notifications
You must be signed in to change notification settings - Fork 6
/
make_deep_disfluency_tagging.py
251 lines (213 loc) · 10.5 KB
/
make_deep_disfluency_tagging.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
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
from argparse import ArgumentParser
import pandas as pd
import json
from lib.babi import get_files_list, read_task, extract_slot_values, extract_slot_value_pps
UTTERANCES_TEST = [('can you book a table in a expensive price range sorry in a cheap price range in rome no in madrid for four people with italian cuisine', ''),
('for two people no for six people please', ''),
('i i um i want a a place with italian sorry spanish cuisine', ''),
('i want a place with italian oh no spanish cuisine', ''),
('i want a place with with let me check italian cuisine', ''),
('i want a place with italian cuisine um sorry with spanish cuisine', ''),
('i want um yeah i want a place with spanish cuisine', '')]
class SWDADisfluencyTagger(object):
def __init__(self, in_slot_values, in_slot_value_phrases, in_action_templates):
self.slot_values = in_slot_values
self.slot_value_phrases = set([tuple(self.delexicalize_slot_values(phrase))
for phrase in in_slot_value_phrases])
self.action_templates = []
for key, templates in in_action_templates.iteritems():
self.action_templates += map(tuple, templates)
def reset_state(self):
self.state = 'FLUENT'
def delexicalize_slot_values(self, in_tokens):
return ['<value>' if token in self.slot_values else token
for token in in_tokens]
def tag_utterance(self, in_utterance):
self.reset_state()
utterance_tokens_original = in_utterance.split()
utterance_tokens = self.delexicalize_slot_values(utterance_tokens_original)
fluent_buffer = []
reparandum_buffer = []
interregnum_buffer = []
repair_buffer = []
tags = []
for token in utterance_tokens:
if self.state == 'FLUENT':
if matches_template_prefix([token], self.action_templates):
interregnum_buffer.append(token)
self.state = 'INSIDE_INTERREGNUM'
else:
fluent_buffer.append(token)
if matches_template_prefix(reparandum_buffer + [token], self.slot_value_phrases):
reparandum_buffer.append(token)
self.state = 'INSIDE_REPARANDUM'
elif self.state == 'INSIDE_INTERREGNUM':
if matches_template_prefix(interregnum_buffer + [token], self.action_templates):
interregnum_buffer.append(token)
else:
if matches_template_prefix(repair_buffer + [token], [tuple(reparandum_buffer)]):
repair_buffer.append(token)
self.state = 'INSIDE_REPAIR'
elif token == '<value>':
repair_buffer = [token]
else:
tags += self.flush_tags(fluent_buffer,
reparandum_buffer,
interregnum_buffer,
repair_buffer)
fluent_buffer = [token]
reparandum_buffer = []
interregnum_buffer = []
repair_buffer = []
self.state = 'FLUENT'
elif self.state == 'INSIDE_REPARANDUM':
if matches_template_prefix(reparandum_buffer + [token], self.slot_value_phrases):
reparandum_buffer.append(token)
fluent_buffer.append(token)
elif matches_template_prefix(interregnum_buffer + [token], self.action_templates):
interregnum_buffer.append(token)
self.state = 'INSIDE_INTERREGNUM'
else:
tags += self.flush_tags(fluent_buffer,
reparandum_buffer,
interregnum_buffer,
repair_buffer)
fluent_buffer = [token]
reparandum_buffer = []
interregnum_buffer = []
repair_buffer = []
self.state = 'FLUENT'
if matches_template_prefix(reparandum_buffer + [token],
self.slot_value_phrases):
reparandum_buffer.append(token)
self.state = 'INSIDE_REPARANDUM'
elif self.state == 'INSIDE_REPAIR':
if matches_template_prefix(repair_buffer + [token], [tuple(reparandum_buffer)]):
repair_buffer.append(token)
else:
tags += self.flush_tags(fluent_buffer,
reparandum_buffer,
interregnum_buffer,
repair_buffer)
reparandum_buffer = []
interregnum_buffer = []
repair_buffer = []
fluent_buffer = []
if matches_template_prefix([token], self.action_templates):
interregnum_buffer.append(token)
self.state = 'INSIDE_INTERREGNUM'
elif matches_template_prefix(reparandum_buffer + [token], self.slot_value_phrases):
reparandum_buffer.append(token)
fluent_buffer.append(token)
self.state = 'INSIDE_REPARANDUM'
else:
fluent_buffer = [token]
self.state = 'FLUENT'
else:
raise NotImplementedError
tags += self.flush_tags(fluent_buffer,
reparandum_buffer,
interregnum_buffer,
repair_buffer)
assert len(tags) == len(utterance_tokens)
return tags
def flush_tags(self,
in_fluent_buffer,
in_reparandum_buffer,
in_interregnum_buffer,
in_repair_buffer):
result = []
result += ['<f/>'] * len(in_fluent_buffer)
if matches_template(in_interregnum_buffer, self.action_templates):
result += ['<e/>'] * len(in_interregnum_buffer)
else:
result += ['<f/>'] * len(in_interregnum_buffer)
if 1 == len(in_repair_buffer):
result.append('<rm-{}/><rpEndSub/>'.format(len(in_interregnum_buffer) + 1))
elif 1 < len(in_repair_buffer):
result.append('<rm-{}/><rpMid/>'.format(len(in_interregnum_buffer) + 1))
for _ in xrange(len(in_repair_buffer) - 2):
result.append('<f/>')
result.append('<rpEnd/>')
return result
def matches_template_prefix(in_buffer, in_templates):
if not len(in_buffer):
return False
buffer_tuple = tuple(in_buffer)
for template in in_templates:
if template[:len(in_buffer)] == buffer_tuple:
return True
return False
def matches_template(in_buffer, in_templates):
buffer_tuple = tuple(in_buffer)
if not len(in_buffer):
return None
for template in in_templates:
if template == buffer_tuple:
return True
return None
def collect_babi_slot_values(in_babi_root):
dataset_files = get_files_list(in_babi_root, 'task1-API-calls')
babi_files = [(filename, read_task(filename)) for filename in dataset_files]
full_babi = reduce(lambda x, y: x + y[1],
babi_files,
[])
slots_map = extract_slot_values(full_babi)
return reduce(lambda x, y: list(x) + list(y), slots_map.values(), [])
def collect_babi_slot_value_pps(in_babi_root, in_slot_values):
dataset_files = get_files_list(in_babi_root, 'task1-API-calls')
babi_files = [(filename, read_task(filename)) for filename in dataset_files]
full_babi = reduce(lambda x, y: x + y[1],
babi_files,
[])
return extract_slot_value_pps(full_babi, in_slot_values)
def get_action_templates(in_config):
action_templates = dict(in_config['action_templates'])
for key in action_templates.keys():
value = action_templates[key]
value = map(lambda x: x.split(), value)
value = [filter(lambda x: not x.startswith('$'), tokens) for tokens in value]
action_templates[key] = value
del action_templates['NULL']
return action_templates
def configure_argument_parser():
parser = ArgumentParser(description='Tag disfluencies SWDA-style')
parser.add_argument('parallel_file')
parser.add_argument('result_file')
parser.add_argument('config_file')
parser.add_argument('babi_folder',
help='original bAbI Dialog dataset root',
default='dialog-bAbI-tasks')
parser.add_argument('--test', default=False, action='store_true')
return parser
def test(in_config, in_babi_folder):
slot_values = collect_babi_slot_values(in_babi_folder)
slot_value_pps = collect_babi_slot_value_pps(in_babi_folder, slot_values)
action_templates = get_action_templates(in_config)
tagger = SWDADisfluencyTagger(slot_values, slot_value_pps, action_templates)
for idx, (utterance_disfluent, utterance_fluent) in enumerate(UTTERANCES_TEST):
tags = tagger.tag_utterance(utterance_disfluent)
print utterance_disfluent
print ' '.join(tags)
def main():
parser = configure_argument_parser()
args = parser.parse_args()
with open(args.config_file) as config_in:
config = json.load(config_in)
if args.test:
test(config, args.babi_folder)
return
dataset = pd.read_csv(args.parallel_file, delimiter=';')
slot_values = collect_babi_slot_values(args.babi_folder)
slot_value_pps = collect_babi_slot_value_pps(args.babi_folder, slot_values)
action_templates = get_action_templates(config)
result_utterances, result_tags = [], []
tagger = SWDADisfluencyTagger(slot_values, slot_value_pps, action_templates)
for idx, (utterance_disfluent, utterance_fluent) in dataset.iterrows():
tags = tagger.tag_utterance(utterance_disfluent)
result_utterances.append(utterance_disfluent.split())
result_tags.append(tags)
result = pd.DataFrame({'utterance': result_utterances, 'tags': result_tags})
result.to_json(args.result_file)
if __name__ == '__main__':
main()