-
Notifications
You must be signed in to change notification settings - Fork 44
/
conllu-formconvert.py
executable file
·64 lines (57 loc) · 3.02 KB
/
conllu-formconvert.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
import sys
import re
import file_util
from file_util import ID,FORM,HEAD,DEPREL,DEPS #column index for the columns we'll need
import argparse
def w2t(wtree):
tokens=get_tokens(wtree)
word_ids=[u"0"] #root remains 0
line_idx=0 #index of the line in wtree we are editing
for token_idx,(b,e) in enumerate(tokens): #go over all token ranges and produce new IDs for the words involved
wtree[line_idx][ID]=unicode(token_idx+1) #Renumber the ID field of the token
if b==e: #token==word
word_ids.append("%d"%(token_idx+1))
line_idx+=1
else:
#We have a range, renumber the words
line_idx+=1
for word_idx,_ in enumerate(range(b,e+1)): #consume as many lines as there are words in the token
word_ids.append("%d.%d"%(token_idx+1,word_idx+1))
wtree[line_idx][ID]=word_ids[-1]
line_idx+=1
#word_ids is now a list with 1-based indexing which has the new ID for every single word
#the ID column has been renumbered by now
#now we can renumber all of the HEAD columns
for cols in wtree:
if cols[HEAD]==u"_": #token
continue
cols[HEAD]=word_ids[int(cols[HEAD])]
if cols[DEPS]!=u"_": #need to renumber secondary deps
new_pairs=[]
for head_deprel in cols[DEPS].split(u"|"):
head,deprel=head_deprel.split(u":")
new_pairs.append(word_ids[int(head)]+u":"+deprel)
cols[DEPS]=u"|".join(new_pairs)
if __name__=="__main__":
opt_parser = argparse.ArgumentParser(description='Conversion script from word-based CoNLL-U to other formats.')
opt_parser.add_argument('input', nargs='?', help='Input file name, or "-" or nothing for standard input.')
opt_parser.add_argument('output', nargs='?', help='Output file name, or "-" or nothing for standard output.')
opt_parser.add_argument('-f','--output-format', default="dgraph", help='Output format. Currently supported: dgraph (CoreNLP dep output). Default: %(default)s.')
args = opt_parser.parse_args() #Parsed command-line arguments
inp,out=file_util.in_out(args)
for comments,tree in file_util.trees(inp):
deps=set() #A set of (gov,dep,dType) where gov and dep are zero-based indices
for line in tree:
if not line[ID].isdigit(): #token line, skip
continue
if line[HEAD] not in (u"_",u"0"):
deps.add((int(line[HEAD])-1,int(line[ID])-1,line[DEPREL]))
#Process also the DEPS field
if line[DEPS]!=u"_":
for head_col_deprel in line[DEPS].split(u"|"):
head,deprel=head_col_deprel.split(u":",1)
deps.add((int(head)-1,int(line[ID])-1,line[DEPREL]))
#Done. Maybe these should be sorted somehow? Also, what to do if we have no deps?
for gov,dep,deprel in sorted(deps):
print >> out, u"%s(%s-%d, %s-%d)"%(deprel,tree[gov][FORM],gov+1,tree[dep][FORM],dep+1)
print >> out