1
|
|
|
#!/usr/bin/env python |
2
|
|
|
# -*- coding: utf-8 -*- |
3
|
|
|
from __future__ import unicode_literals |
4
|
|
|
from .utils import post_json |
5
|
|
|
from .ds import Document, Interval, NLPDatum |
6
|
|
|
from termcolor import colored |
7
|
|
|
import re |
8
|
|
|
import json |
9
|
|
|
|
10
|
|
|
|
11
|
|
|
class OdinHighlighter(object): |
12
|
|
|
|
13
|
|
|
@staticmethod |
14
|
|
|
def LABEL(token): |
15
|
|
|
return colored(token, color="red", attrs=["bold"]) |
16
|
|
|
|
17
|
|
|
@staticmethod |
18
|
|
|
def ARG(token): |
19
|
|
|
return colored(token, on_color="on_green", attrs=["bold"]) |
20
|
|
|
|
21
|
|
|
@staticmethod |
22
|
|
|
def TRIGGER(token): |
23
|
|
|
return colored(token, on_color="on_blue", attrs=["bold"]) |
24
|
|
|
|
25
|
|
|
@staticmethod |
26
|
|
|
def CONCEAL(token): |
27
|
|
|
return colored(token, on_color="on_grey", attrs=["concealed"]) |
28
|
|
|
|
29
|
|
|
@staticmethod |
30
|
|
|
def MENTION(token): |
31
|
|
|
return colored(token, on_color="on_yellow") |
32
|
|
|
|
33
|
|
|
@staticmethod |
34
|
|
|
def highlight_mention(mention): |
35
|
|
|
""" |
36
|
|
|
Formats text of mention |
37
|
|
|
""" |
38
|
|
|
text_span = mention.sentenceObj.words[:] |
39
|
|
|
# format TBM span like an arg |
40
|
|
|
if mention.type == "TextBoundMention": |
41
|
|
|
for i in range(mention.start, mention.end): |
42
|
|
|
text_span[i] = OdinHighlighter.ARG(text_span[i]) |
43
|
|
|
if mention.arguments: |
44
|
|
|
for (role, args) in mention.arguments.items(): |
45
|
|
|
for arg in args: |
46
|
|
|
for i in range(arg.start, arg.end): |
47
|
|
|
text_span[i] = OdinHighlighter.ARG(text_span[i]) |
48
|
|
|
# format trigger distinctly from args |
49
|
|
|
if mention.trigger: |
50
|
|
|
trigger = mention.trigger |
51
|
|
|
for i in range(trigger.start, trigger.end): |
52
|
|
|
text_span[i] = OdinHighlighter.TRIGGER(text_span[i]) |
53
|
|
|
|
54
|
|
|
# highlight tokens contained in mention span |
55
|
|
|
for i in range(mention.start, mention.end): |
56
|
|
|
text_span[i] = OdinHighlighter.MENTION(text_span[i]) |
57
|
|
|
mention_span = OdinHighlighter.MENTION(" ").join(text_span[mention.start:mention.end]) |
58
|
|
|
# highlight spaces in mention span |
59
|
|
|
formatted_text = " ".join(text_span[:mention.start]) + " " + mention_span + " " + " ".join(text_span[mention.end:]) |
60
|
|
|
return formatted_text.strip() |
61
|
|
|
|
62
|
|
|
class Mention(NLPDatum): |
63
|
|
|
""" |
64
|
|
|
A labeled span of text. Used to model textual mentions of events, relations, and entities. |
65
|
|
|
|
66
|
|
|
Parameters |
67
|
|
|
---------- |
68
|
|
|
token_interval : Interval |
69
|
|
|
The span of the Mention represented as an Interval. |
70
|
|
|
sentence : int |
71
|
|
|
The sentence index that contains the Mention. |
72
|
|
|
document : Document |
73
|
|
|
The Document in which the Mention was found. |
74
|
|
|
foundBy : str |
75
|
|
|
The Odin IE rule that produced this Mention. |
76
|
|
|
label : str |
77
|
|
|
The label most closely associated with this span. Usually the lowest hyponym of "labels". |
78
|
|
|
labels: list |
79
|
|
|
The list of labels associated with this span. |
80
|
|
|
trigger: dict or None |
81
|
|
|
dict of JSON for Mention's trigger (event predicate or word(s) signaling the Mention). |
82
|
|
|
arguments: dict or None |
83
|
|
|
dict of JSON for Mention's arguments. |
84
|
|
|
paths: dict or None |
85
|
|
|
dict of JSON encoding the syntactic paths linking a Mention's arguments to its trigger (applies to Mentions produces from `type:"dependency"` rules). |
86
|
|
|
doc_id: str or None |
87
|
|
|
the id of the document |
88
|
|
|
|
89
|
|
|
Attributes |
90
|
|
|
---------- |
91
|
|
|
tokenInterval: processors.ds.Interval |
92
|
|
|
An `Interval` encoding the `start` and `end` of the `Mention`. |
93
|
|
|
start : int |
94
|
|
|
The token index that starts the `Mention`. |
95
|
|
|
end : int |
96
|
|
|
The token index that marks the end of the Mention (exclusive). |
97
|
|
|
sentenceObj : processors.ds.Sentence |
98
|
|
|
Pointer to the `Sentence` instance containing the `Mention`. |
99
|
|
|
characterStartOffset: int |
100
|
|
|
The index of the character that starts the `Mention`. |
101
|
|
|
characterEndOffset: int |
102
|
|
|
The index of the character that ends the `Mention`. |
103
|
|
|
type: Mention.TBM or Mention.EM or Mention.RM |
104
|
|
|
The type of the `Mention`. |
105
|
|
|
|
106
|
|
|
See Also |
107
|
|
|
-------- |
108
|
|
|
|
109
|
|
|
[`Odin` manual](https://arxiv.org/abs/1509.07513) |
110
|
|
|
|
111
|
|
|
Methods |
112
|
|
|
------- |
113
|
|
|
matches(label_pattern) |
114
|
|
|
Test if the provided pattern, `label_pattern`, matches any element in `Mention.labels`. |
115
|
|
|
|
116
|
|
|
overlaps(other) |
117
|
|
|
Test whether other (token index or Mention) overlaps with span of this Mention. |
118
|
|
|
|
119
|
|
|
copy(**kwargs) |
120
|
|
|
Copy constructor for this Mention. |
121
|
|
|
|
122
|
|
|
words() |
123
|
|
|
Words for this Mention's span. |
124
|
|
|
|
125
|
|
|
tags() |
126
|
|
|
Part of speech for this Mention's span. |
127
|
|
|
|
128
|
|
|
lemmas() |
129
|
|
|
Lemmas for this Mention's span. |
130
|
|
|
|
131
|
|
|
_chunks() |
132
|
|
|
chunk labels for this Mention's span. |
133
|
|
|
|
134
|
|
|
_entities() |
135
|
|
|
NE labels for this Mention's span. |
136
|
|
|
""" |
137
|
|
|
|
138
|
|
|
TBM = "TextBoundMention" |
139
|
|
|
EM = "EventMention" |
140
|
|
|
RM = "RelationMention" |
141
|
|
|
|
142
|
|
|
def __init__(self, |
143
|
|
|
token_interval, |
144
|
|
|
sentence, |
145
|
|
|
document, |
146
|
|
|
foundBy, |
147
|
|
|
label, |
148
|
|
|
labels=None, |
149
|
|
|
trigger=None, |
150
|
|
|
arguments=None, |
151
|
|
|
paths=None, |
152
|
|
|
keep=True, |
153
|
|
|
doc_id=None): |
154
|
|
|
|
155
|
|
|
NLPDatum.__init__(self) |
156
|
|
|
self.label = label |
157
|
|
|
self.labels = labels if labels else [self.label] |
158
|
|
|
self.tokenInterval = token_interval |
159
|
|
|
self.start = self.tokenInterval.start |
160
|
|
|
self.end = self.tokenInterval.end |
161
|
|
|
self.document = document |
162
|
|
|
self._doc_id = doc_id or hash(self.document) |
163
|
|
|
self.sentence = sentence |
164
|
|
|
if trigger: |
165
|
|
|
# NOTE: doc id is not stored for trigger's json, |
166
|
|
|
# as it is assumed to be contained in the same document as its parent |
167
|
|
|
trigger.update({"document": self._doc_id}) |
168
|
|
|
self.trigger = Mention.load_from_JSON(trigger, self._to_document_map()) |
169
|
|
|
else: |
170
|
|
|
self.trigger = None |
171
|
|
|
# unpack args |
172
|
|
|
self.arguments = {role:[Mention.load_from_JSON(a, self._to_document_map()) for a in args] for (role, args) in arguments.items()} if arguments else None |
173
|
|
|
self.paths = paths |
174
|
|
|
self.keep = keep |
175
|
|
|
self.foundBy = foundBy |
176
|
|
|
# other |
177
|
|
|
self.sentenceObj = self.document.sentences[self.sentence] |
178
|
|
|
self.text = " ".join(self.sentenceObj.words[self.start:self.end]) |
179
|
|
|
# recover offsets |
180
|
|
|
self.characterStartOffset = self.sentenceObj.startOffsets[self.tokenInterval.start] |
181
|
|
|
self.characterEndOffset = self.sentenceObj.endOffsets[self.tokenInterval.end - 1] |
182
|
|
|
# for later recovery |
183
|
|
|
self.id = None |
184
|
|
|
self.type = self._set_type() |
185
|
|
|
|
186
|
|
|
def __str__(self): |
187
|
|
|
return "{}: {}".format(OdinHighlighter.LABEL(self.label), OdinHighlighter.highlight_mention(self)) |
188
|
|
|
|
189
|
|
|
def __eq__(self, other): |
190
|
|
|
if isinstance(other, self.__class__): |
191
|
|
|
return self.__dict__ == other.__dict__ |
192
|
|
|
else: |
193
|
|
|
return False |
194
|
|
|
|
195
|
|
|
def __ne__(self, other): |
196
|
|
|
return not self.__eq__(other) |
197
|
|
|
|
198
|
|
|
def __hash__(self): |
199
|
|
|
return hash(self.to_JSON()) |
200
|
|
|
|
201
|
|
|
def to_JSON_dict(self): |
202
|
|
|
m = dict() |
203
|
|
|
m["id"] = self.id |
204
|
|
|
m["type"] = self.type |
205
|
|
|
m["label"] = self.label |
206
|
|
|
m["labels"] = self.labels |
207
|
|
|
m["tokenInterval"] = self.tokenInterval.to_JSON_dict() |
208
|
|
|
m["characterStartOffset"] = self.characterStartOffset |
209
|
|
|
m["characterEndOffset"] = self.characterEndOffset |
210
|
|
|
m["sentence"] = self.sentence |
211
|
|
|
m["document"] = self._doc_id |
212
|
|
|
# do we have a trigger? |
213
|
|
|
if self.trigger: |
214
|
|
|
m["trigger"] = self.trigger.to_JSON_dict() |
215
|
|
|
# do we have arguments? |
216
|
|
|
if self.arguments: |
217
|
|
|
m["arguments"] = self._arguments_to_JSON_dict() |
218
|
|
|
# handle paths |
219
|
|
|
if self.paths: |
220
|
|
|
m["paths"] = self.paths |
221
|
|
|
m["keep"] = self.keep |
222
|
|
|
m["foundBy"] = self.foundBy |
223
|
|
|
return m |
224
|
|
|
|
225
|
|
|
def startOffset(self): |
226
|
|
|
return self.sentenceObj.endOffsets[self.start] |
227
|
|
|
|
228
|
|
|
def endOffset(self): |
229
|
|
|
return self.sentenceObj.endOffsets[self.end -1] |
230
|
|
|
|
231
|
|
|
def words(self): |
232
|
|
|
return self.sentenceObj.words[self.start:self.end] |
233
|
|
|
|
234
|
|
|
def tags(self): |
235
|
|
|
return self.sentenceObj.tags[self.start:self.end] |
236
|
|
|
|
237
|
|
|
def lemmas(self): |
238
|
|
|
return self.sentenceObj.lemmas[self.start:self.end] |
239
|
|
|
|
240
|
|
|
def _chunks(self): |
241
|
|
|
return self.sentenceObj._chunks[self.start:self.end] |
242
|
|
|
|
243
|
|
|
def _entities(self): |
244
|
|
|
return self.sentenceObj._entities[self.start:self.end] |
245
|
|
|
|
246
|
|
|
def copy(self, **kwargs): |
247
|
|
|
""" |
248
|
|
|
Copy constructor for mention |
249
|
|
|
""" |
250
|
|
|
# return new instance |
251
|
|
|
return self.__class__( |
252
|
|
|
label=kwargs.get("label", self.label), |
253
|
|
|
labels=kwargs.get("label", self.labels), |
254
|
|
|
token_interval=kwargs.get("token_interval", self.tokenInterval), |
255
|
|
|
sentence=kwargs.get("sentence", self.sentence), # NOTE: this is the sentence idx |
256
|
|
|
document=kwargs.get("document", self.document), |
257
|
|
|
foundBy=kwargs.get("foundBy", self.foundBy), |
258
|
|
|
trigger=kwargs.get("trigger", self.trigger), |
259
|
|
|
arguments=kwargs.get("arguments", self.arguments), |
260
|
|
|
paths=kwargs.get("paths", self.paths), |
261
|
|
|
keep=kwargs.get("keep", self.keep), |
262
|
|
|
doc_id=kwargs.get("doc_id", self._doc_id) |
263
|
|
|
) |
264
|
|
|
|
265
|
|
|
def overlaps(self, other): |
266
|
|
|
""" |
267
|
|
|
Checks for overlap. |
268
|
|
|
""" |
269
|
|
|
if isinstance(other, int): |
270
|
|
|
return self.start <= other < self.end |
271
|
|
|
elif isinstance(other, Mention): |
272
|
|
|
# equiv. sentences + checks on start and end |
273
|
|
|
return (self.sentence.__hash__() == other.sentence.__hash__()) and \ |
274
|
|
|
((other.start <= self.start < other.end) or (self.start <= other.start < self.end)) |
275
|
|
|
else: |
276
|
|
|
return False |
277
|
|
|
|
278
|
|
|
def matches(self, label_pattern): |
279
|
|
|
""" |
280
|
|
|
Test if the provided pattern, `label_pattern`, matches any element in `Mention.labels`. |
281
|
|
|
|
282
|
|
|
Parameters |
283
|
|
|
---------- |
284
|
|
|
label_pattern : str or _sre.SRE_Pattern |
285
|
|
|
The pattern to match against each element in `Mention.labels` |
286
|
|
|
|
287
|
|
|
Returns |
288
|
|
|
------- |
289
|
|
|
bool |
290
|
|
|
True if `label_pattern` matches any element in `Mention.labels` |
291
|
|
|
""" |
292
|
|
|
return any(re.match(label_pattern, label) for label in self.labels) |
293
|
|
|
|
294
|
|
|
def _arguments_to_JSON_dict(self): |
295
|
|
|
return dict((role, [a.to_JSON_dict() for a in args]) for (role, args) in self.arguments.items()) |
296
|
|
|
|
297
|
|
|
def _paths_to_JSON_dict(self): |
298
|
|
|
return {role: paths.to_JSON_dict() for (role, paths) in self.paths} |
299
|
|
|
|
300
|
|
|
@staticmethod |
301
|
|
|
def load_from_JSON(mjson, docs_dict): |
302
|
|
|
# recover document |
303
|
|
|
doc_id = mjson["document"] |
304
|
|
|
doc = docs_dict[doc_id] |
305
|
|
|
labels = mjson["labels"] |
306
|
|
|
kwargs = { |
307
|
|
|
"label": mjson.get("label", labels[0]), |
308
|
|
|
"labels": labels, |
309
|
|
|
"token_interval": Interval.load_from_JSON(mjson["tokenInterval"]), |
310
|
|
|
"sentence": mjson["sentence"], |
311
|
|
|
"document": doc, |
312
|
|
|
"doc_id": doc_id, |
313
|
|
|
"trigger": mjson.get("trigger", None), |
314
|
|
|
"arguments": mjson.get("arguments", None), |
315
|
|
|
"paths": mjson.get("paths", None), |
316
|
|
|
"keep": mjson.get("keep", True), |
317
|
|
|
"foundBy": mjson["foundBy"] |
318
|
|
|
} |
319
|
|
|
m = Mention(**kwargs) |
320
|
|
|
# set IDs |
321
|
|
|
m.id = mjson["id"] |
322
|
|
|
m._doc_id = doc_id |
323
|
|
|
# set character offsets |
324
|
|
|
m.character_start_offset = mjson["characterStartOffset"] |
325
|
|
|
m.character_end_offset = mjson["characterEndOffset"] |
326
|
|
|
return m |
327
|
|
|
|
328
|
|
|
def _to_document_map(self): |
329
|
|
|
return {self._doc_id: self.document} |
330
|
|
|
|
331
|
|
|
def _set_type(self): |
332
|
|
|
# event mention |
333
|
|
|
if self.trigger != None: |
334
|
|
|
return Mention.EM |
335
|
|
|
# textbound mention |
336
|
|
|
elif self.trigger == None and self.arguments == None: |
337
|
|
|
return Mention.TBM |
338
|
|
|
else: |
339
|
|
|
return Mention.RM |
340
|
|
|
|