def convconll(goldconll, parsesconll):
"""Copy columns from parsesconll to goldconll file; overwrite in-place.
Files should contain only a single document."""
goldconlldata = next(iter(readconll(goldconll).values()))
header = open(goldconll).readline().rstrip()
parsesconlldata = next(iter(readconll(parsesconll).values()))
if len(goldconlldata) != len(parsesconlldata):
raise ValueError('mismatch in number of sentences')
for gchunk, pchunk in zip(goldconlldata, parsesconlldata):
if len(gchunk) != len(pchunk):
raise ValueError('Sentence length mismatch')
if len(pchunk[0]) < 13:
raise ValueError('Not enough fields for gold CoNLL 2012 file')
if len(gchunk[0]) < 13:
raise ValueError('Not enough fields for parses CoNLL 2012 file')
with open(goldconll, 'w') as out:
print(header, file=out)
for gchunk, pchunk in zip(goldconlldata, parsesconlldata):
for gline, pline in zip(gchunk, pchunk):
gline[5] = pline[5]
gline[6] = pline[6]
gline[11] = pline[11]
print('\t'.join(gline[1:]), file=out)
print('', file=out)
print('#end document', file=out)
def compare(cmd, goldfile, respfile, hidecorrectlinks=False, out=sys.stdout):
"""Compare mentions and links across CoNLL 2012 files."""
print('comparing gold file:', goldfile, file=out)
print('against system output:', respfile, file=out)
golddocs = readconll(goldfile)
respdocs = readconll(respfile)
for docname in golddocs:
print('\ndocument:', *docname)
gold = golddocs[docname]
resp = respdocs[docname]
goldspansforcluster = conllclusterdict(gold)
respspansforcluster = conllclusterdict(resp)
goldspans = {span for spans in goldspansforcluster.values()
for span in spans}
respspans = {span for spans in respspansforcluster.values()
for span in spans}
if cmd == 'mentions':
comparementions(gold, resp, goldspans, respspans, out=out)
elif cmd == 'links':
comparecoref(resp, goldspans, respspans, goldspansforcluster,
respspansforcluster, hidecorrectlinks, out)
else:
raise ValueError('unknown cmd: %s' % cmd)
def parseclindata(pattern, outdir):
"""Parse the CLIN dataset."""
origdir = os.getcwd()
filenames = glob(os.path.abspath(pattern))
os.mkdir(outdir)
os.chdir(outdir)
for n, conllfile in enumerate(filenames, 1):
data = next(iter(readconll(conllfile).values()))
fname = os.path.basename(conllfile)
docname = fname[:fname.index('_')]
tokenidx = 3
print('Parsing %d/%d: %s' % (n, len(filenames), docname))
parse(data, docname, tokenidx)
os.chdir(origdir)
def parsesemeval(path, outdir):
"""Parse the SemEval dataset."""
path = os.path.abspath(path)
origdir = os.getcwd()
os.mkdir(outdir)
os.chdir(outdir)
with open(path) as inp:
data = inp.read()
docnames = re.findall(r'#begin document ([\w_]+)', data)
docs = readconll(path)
for n, docname in enumerate(docnames, 1):
data = docs[docname, 0]
tokenidx = 2
print('Parsing %d/%d: %s' % (n, len(docnames), docname))
parse(data, docname, tokenidx)
os.chdir(origdir)
def convalpino(conllfile, parsesdir):
"""Add parse bits to a single file, overwrite in-place.
File should contain only a single document."""
from lxml import etree
try:
from discodop.tree import writebrackettree
from discodop.treebank import AlpinoCorpusReader
from discodop.treetransforms import raisediscnodes
except ImportError:
print('Install https://github.com/andreasvc/disco-dop')
return
conlldata = next(iter(readconll(conllfile).values()))
header = open(conllfile).readlines()[0].rstrip()
treebank = AlpinoCorpusReader(parsesdir + '/*.xml',
morphology='replace',
headrules='../disco-dop/alpino.headrules')
for chunk, (_key, item) in zip(conlldata, treebank.itertrees()):
if len(chunk) != len(item.sent):
raise ValueError('length mismatch')
if len(chunk[0]) < 12:
raise ValueError('Not enough fields for gold CoNLL 2012 file')
with open(conllfile, 'w') as out:
print(header, file=out)
for chunk, (_key, item) in zip(conlldata, treebank.itertrees()):
raisediscnodes(item.tree)
for n, (_, postag) in enumerate(item.tree.pos()):
if len(chunk[n]) < 13: # kludge
chunk[n] = chunk[n][:-1] + ['-', chunk[n][-1]]
# NB: parens as square brackets: N[eigen,...]
chunk[n][5] = postag
splitparse(writebrackettree(item.tree, item.sent), chunk)
xmltree = etree.fromstring(item.block)
addner(xmltree, chunk)
for line in chunk:
print('\t'.join(line[1:]), file=out)
print('', file=out)
print('#end document', file=out)
def getstats(args, parsesdir=None):
"""Print stats for a list of CoNLL 2012 files."""
import os
from glob import glob
from lxml import etree
import coref
sents = tokens = nummentions = numentities = numlinks = 0
pronouns = nominals = names = 0
ngdata, gadata = coref.readngdata()
for fname in args:
data = []
try:
docs = coref.readconll(fname)
except Exception as err:
print('file:', fname)
print(err)
return
for (docname, part), data in docs.items():
try:
goldspansforcluster = coref.conllclusterdict(data)
except Exception as err:
print('file:', fname)
print(err)
return
if parsesdir is not None:
# given docname, read <parsesdir>/<docname>/*.xml
path = os.path.join(parsesdir, docname, '*.xml')
filenames = sorted(glob(path), key=coref.parsesentid)
if len(data) != len(filenames):
raise ValueError(
'filename: %s; document %s %s; '
'sentences in CoNLL (%d) '
'and number of .xml parses (%d) not equal' %
(fname, docname, part, len(data), len(filenames)))
trees = [(coref.parsesentid(filename), etree.parse(filename))
for filename in filenames]
mentions = coref.extractmentionsfromconll(
data, trees, ngdata, gadata)
pronouns += sum(m.type == 'pronoun' for m in mentions)
nominals += sum(m.type == 'noun' for m in mentions)
names += sum(m.type == 'name' for m in mentions)
sents += len(data)
tokens += sum(len(sent) for sent in data)
nummentions += len({
span
for spans in goldspansforcluster.values() for span in spans
})
numentities += len(goldspansforcluster)
numlinks += sum(
int((len(cluster) * (len(cluster) - 1)) / 2)
for cluster in goldspansforcluster.values())
print('sents:', sents)
print('tokens:', tokens)
print('mentions:', nummentions)
print('entities:', numentities)
print('links:', numlinks)
print('tok/sent:', tokens / sents)
print('mentions / tokens:', nummentions / tokens)
print('entities / tokens:', numentities / tokens)
print('links / tokens:', numlinks / tokens)
print('mentions / entities:', nummentions / numentities)
print('links / entities:', numlinks / numentities)
if parsesdir is None:
print('specify --parses to get % pronouns, nominals, names')
else:
print('% pronouns', 100 * pronouns / nummentions)
print('% nominals', 100 * nominals / nummentions)
print('% names', 100 * names / nummentions)
"""Load CoNLL 2012 files and report any errors or warnings."""
import sys
from coref import readconll, conllclusterdict, setverbose
setverbose(True, sys.stdout)
for filename in sys.argv[1:]:
try:
for docname, conlldata in readconll(filename).items():
print('\n', filename, *docname)
conllclusterdict(conlldata)
except Exception as err:
print(err)
print('NB: Not checking for further errors in this file.')