/
parse_transcript_xml.py
216 lines (180 loc) · 7.47 KB
/
parse_transcript_xml.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
from operator import itemgetter
import re
import sys
import xml.etree.cElementTree as ET
def fix_ocr_errors_in_line(line_in):
fixed = line_in
# fix Urn = Um
fixed = re.sub(r'\b([Uu])rn\b', r'\1m', fixed)
# fix rna' am = ma'am
fixed = re.sub('rna\' am', 'ma\'am', fixed)
# fix IS = 's
fixed = re.sub(r'\bIS\b', '\'s', fixed)
# fix 1s = is and 1n = in
fixed = re.sub(r'\b1s\b', r'is', fixed)
fixed = re.sub(r'\b1n\b', r'in', fixed)
# fix other i = 1 variations
fixed = re.sub('1ng', 'ing', fixed)
fixed = re.sub('r1e', 'rie', fixed)
fixed = re.sub('p1e', 'pie', fixed)
fixed = re.sub('r1m', 'rim', fixed)
fixed = re.sub('r1o', 'rio', fixed)
fixed = re.sub('g1v', 'giv', fixed)
fixed = re.sub('exam1n', 'examin', fixed)
fixed = re.sub('aga1n', 'again', fixed)
fixed = re.sub('excess1ve', 'excessive', fixed)
fixed = re.sub('1gnore', 'ignore', fixed)
fixed = re.sub('1mage', 'image', fixed)
fixed = re.sub('1ssue', 'issue', fixed)
fixed = re.sub('magaz1ne', 'magazine', fixed)
fixed = re.sub('mus1c', 'music', fixed)
fixed = re.sub('n1ne', 'nine', fixed)
fixed = re.sub('occas1on', 'occasion', fixed)
fixed = re.sub('op1n1on', 'opinion', fixed)
fixed = re.sub('se1ze', 'seize', fixed)
fixed = re.sub('s1gn', 'sign', fixed)
fixed = re.sub('s1nce', 'since', fixed)
fixed = re.sub('s1r', 'sir', fixed)
fixed = re.sub('s1x', 'six', fixed)
fixed = re.sub('s1ze', 'size', fixed)
fixed = re.sub('superv1sor', 'supervisor', fixed)
fixed = re.sub('v1ce', 'vice', fixed)
# fix some one-offs
fixed = re.sub('Sox-rays', 'So x-rays', fixed)
fixed = re.sub('pun ked', 'punked', fixed)
fixed = re.sub(ur' \u2022', '', fixed)
return fixed
def parseXML(xml_in):
'''
parse the document XML
'''
# if two fragments of text are within LINE_TOLERANCE of each other they're on the same line
LINE_TOLERANCE = 4
# if two fragments of text are within INDENT_TOLERANCE of each other they're at the same level of indentation
INDENT_TOLERANCE = 40
# list of cover page numbers
COVER_PAGES = [1,83,277,489,670,956,1226,1441,1675,1781,1923,2121,2376,2634,2744,2975,3139,3394,3580,3850,4054,4279,4388,4654]
# page bounding box
FORMAT_PAGE = {'from_page': 1, 'to_page': 4799, 'left': 0.0, 'top': 792.0, 'right': 614.0, 'bottom': 0.0}
# dicts of format ranges and associated bounding boxes
FORMAT_RANGES = [
{'from_page': 1, 'to_page': 2743, 'left': 160.0, 'top': 712.0, 'right': 586.0, 'bottom': 84.0},
{'from_page': 2744, 'to_page': 4653, 'left': 125.0, 'top': 712.0, 'right': 555.0, 'bottom': 77.0},
{'from_page': 4654, 'to_page': 4799, 'left': 118.0, 'top': 753.0, 'right': 588.0, 'bottom': 40.0}
]
# if true, capture all text, including that outside the focus areas
CAPTURE_ALL_TEXT = False
# :NOTE: set page offset to 500 when parsing sample.xml
PAGE_OFFSET = 0
# get the page elements
tree = ET.ElementTree(file=xml_in)
pages = tree.getroot()
if pages.tag != "pages":
sys.exit("ERROR: pages.tag is %s instead of pages!" % pages.tag)
# all the document lines
all_lines = []
# step through the pages
for page in pages:
# get all the textline elements
textlines = page.findall("./textbox/textline")
#print "found %s textlines" % len(textlines)
# is this a cover page?
is_cover_page = int(page.attrib["id"]) + PAGE_OFFSET in COVER_PAGES
focus_bbox = FORMAT_PAGE
# get the bounding box for the transcript area of the page
if not is_cover_page:
for page_format in FORMAT_RANGES:
if page_format['from_page'] <= int(page.attrib["id"]) + PAGE_OFFSET <= page_format['to_page']:
focus_bbox = page_format
# step through the textlines
lines = []
for textline in textlines:
# get the boundaries of the textline [left,bottom,right,top]
line_bounds = [float(s) for s in textline.attrib["bbox"].split(',')]
#print "line_bounds: %s" % line_bounds
# is this line inside the focus area?
is_in_focus = line_bounds[3] < focus_bbox['top'] and line_bounds[3] > focus_bbox['bottom']
# get all the characters in this textline
chars = list(textline)
#print "found %s characters in this line." % len(chars)
# combine all the characters into a single string
line_text = ""
line_left = line_bounds[2]
for char in chars:
try:
char_left = float(char.attrib["bbox"].split(',')[0])
# if the line's inside the focus box, respect the left bounds
if is_in_focus:
if char_left > focus_bbox['left']:
line_text = line_text + char.text
line_left = min(char_left, line_left)
# otherwise, just add the character
elif CAPTURE_ALL_TEXT:
line_text = line_text + char.text
line_left = min(char_left, line_left)
# skip empty <text> elements
except KeyError:
pass
# strip edge & multiple spaces
line_text = re.sub(' +', ' ', line_text.strip())
# fix OCR errors
line_text = fix_ocr_errors_in_line(line_text)
# ignore page numbers
if not CAPTURE_ALL_TEXT:
if re.match('Page', line_text):
line_text = ""
# if line text survived, save a description of the line
if line_text != "":
line = {'left': line_left, 'top': line_bounds[3], 'text': line_text}
lines.append(line)
#print "page %s has %s lines" % (page.attrib["id"], len(lines))
# sort the lines by top position (reverse because measures in PDF/XML are from the bottom of the document)
lines.sort(key=itemgetter('top'), reverse=True)
# consolidate lines that have the same top (within tolerance)
consolidated_lines = []
line_collection = []
line_top = lines[0]['top']
for line in lines:
if abs(line['top'] - line_top) < LINE_TOLERANCE:
line_collection.append(line)
else:
# assure that text segments appear in the correct order
line_collection.sort(key=itemgetter('left'))
consolidated_lines.append(line_collection)
# reset
line_collection = [line]
line_top = line['top']
# add the last line collection to the consolidated lines
line_collection.sort(key=itemgetter('left'))
consolidated_lines.append(line_collection)
# merge each line's text
for line_segments in consolidated_lines:
merged_line = dict(line_segments[0])
merged_line['text'] = " ".join([segment['text'] for segment in line_segments])
all_lines.append(merged_line)
# now clean and categorize the text
last_line_left = 0.0
last_paragraph_left = 0.0
for line in all_lines:
# check for a name (name_match.group(1) will contain the name if there's a match)
name_match = re.match(r'(?:BY )?(M[RS]\. [A-Z]{2,}|M[RS]\.|THE WITNESS|THE FBI|THE COURT|GRAND JUROR|WITNESS 10|DETECTIVE|CHIEF):?', line['text'])
# check for Q/A or QUESTION: (will give false positive for lines of text starting with 'A ...' that aren't answers)
qa_match = re.match(r'[QA] |QUESTION:', line['text'])
# check for an open parens
parens_match = re.match(r'\(', line['text'])
# get the diffs for the position of this line
line_diff = line['left'] - last_line_left
paragraph_diff = line['left'] - last_paragraph_left
# check for a significant change in the indentation
indent_change = abs(line_diff) > INDENT_TOLERANCE
if name_match or qa_match or parens_match or (indent_change and line['left'] > last_line_left):
message = "--++ new paragraph!"
if abs(paragraph_diff) > INDENT_TOLERANCE:
message = message + " (at new indentation)"
print message
last_paragraph_left = line['left']
last_line_left = line['left']
print "(%s,%s) %s" % (line['left'], line['top'], line['text'].encode('utf-8'))
if __name__ == "__main__":
# :NOTE: if parsing sample.xml, set PAGE_OFFSET above to 500
parseXML("files/ferguson_grand_jury_testimony.xml")