def __init__(self, sense_table, wiki_table, rel_vocab):
self.amr = penman.AMRCodec()
self.sense_table = dict()
for line in open(sense_table).readlines():
x, y = line.strip().split('\t')
self.sense_table[x] = y
self.wiki_table = dict()
for line in open(wiki_table).readlines():
x, y = line.strip().split('\t')
self.wiki_table[x] = y
self.rel_vocab = rel_vocab
def test_AMRCodec():
c = penman.AMRCodec()
assert c.invert_relation('ARG0') == 'ARG0-of'
assert c.invert_relation('ARG0-of') == 'ARG0'
assert c.invert_relation('domain') == 'mod'
assert c.invert_relation('mod') == 'domain'
assert c.invert_relation('consist-of') == 'consist-of-of'
assert c.invert_relation('consist-of-of') == 'consist-of'
with pytest.raises(penman.PenmanError):
c.invert_relation('instance')
assert c.encode(
penman.Graph([('w', 'instance', 'want-01'), ('w', 'ARG0', 'b'),
('w', 'ARG1', 'g'), ('b', 'instance', 'boy'),
('g', 'instance', 'go'),
('g', 'ARG0', 'b')])) == ('(w / want-01\n'
' :ARG0 (b / boy)\n'
' :ARG1 (g / go\n'
' :ARG0 b))')
g = penman.Graph([('g', 'instance', 'gold'), ('g', 'consist-of-of', 'r'),
('r', 'instance', 'ring')])
assert c.encode(g) == ('(g / gold\n' ' :consist-of-of (r / ring))')
assert c.encode(g, top='r') == ('(r / ring\n' ' :consist-of (g / gold))')
g = penman.Graph([('w', 'instance', 'white'), ('w', 'domain', 'c'),
('c', 'instance', 'cat')])
assert c.encode(g) == ('(w / white\n' ' :domain (c / cat))')
assert c.encode(g, top='c') == ('(c / cat\n' ' :mod (w / white))')
assert c.decode('(g / go)').triples() == [('g', 'instance', 'go')]
# example adapted from https://github.com/goodmami/penman/issues/17
assert c.decode('(g / go :null_edge (x20 / 876-9))').triples() == [
('g', 'instance', 'go'), ('x20', 'instance', '876-9'),
('g', 'null_edge', 'x20')
]
with pytest.raises(penman.DecodeError):
c.decode('(g)') # no concept or relations
with pytest.raises(penman.DecodeError):
c.decode('(g :ARG0 b)') # no concept
with pytest.raises(penman.DecodeError):
c.decode('(g :ARG0 (b / boy) / go)') # concept after relations
with pytest.raises(penman.DecodeError):
c.decode('(1 / one)') # bad variable form
with pytest.raises(penman.DecodeError):
c.decode('(g / go : (b / boy))') # anonymous relation
import penman
import networkx as nx
from stog.data.vocabulary import DEFAULT_PADDING_TOKEN, DEFAULT_OOV_TOKEN
from stog.data.dataset_readers.amr_parsing.graph_repair import GraphRepair
from stog.utils.string import find_similar_token, is_abstract_token, is_english_punct
from stog.utils import logging
logger = logging.init_logger()
# Disable inverting ':mod' relation.
penman.AMRCodec._inversions.pop('domain')
penman.AMRCodec._deinversions.pop('mod')
from penman import Triple
amr_codec = penman.AMRCodec(indent=6)
WORDSENSE_RE = re.compile(r'-\d\d$')
QUOTED_RE = re.compile(r'^".*"$')
class AMR:
def __init__(self,
id=None,
sentence=None,
graph=None,
tokens=None,
lemmas=None,
pos_tags=None,
ner_tags=None,
abstract_map=None,
def delinearize(linearized_amr):
tokens = linearized_amr.split(' ')
stack = []
concept_var = {}
triplets = []
rel = None
def get_var(tok, concept_var):
vars = concept_var.values()
if tok[0] in 'abcdefghijklmnopqrstuvwxyz':
t = tok[0]
else:
t = 'x'
if t not in vars:
return t
else:
count = 1
while t + str(count) in vars:
count += 1
return t + str(count)
#print(linearized_amr)
for tok in tokens:
if '(' in tok or ')' in tok:
#print('Tok contains parenthesis')
break
if tok.startswith(':'):
# A relation
if rel:
#print('Two relations %s - %s'%('..',rel))
break
else:
rel = tok[1:]
else:
# A concept
if len(stack) == 0:
#print('Length of stack = 0')
var = get_var(tok, concept_var)
triplets.append((var, 'instance', tok))
concept_var[tok] = var
stack.append((var, tok))
root = var
else:
top_var, top_concept = stack[-1]
if rel:
#print(concept_var)
if tok not in concept_var.keys():
#print('%s not in %s'%(tok, str(concept_var)))
var = get_var(tok, concept_var)
triplets.append((var, 'instance', tok))
concept_var[tok] = var
else:
var = concept_var[tok]
triplets.append((top_var, rel, var))
stack.append((var, tok))
rel = None
else:
if top_concept == tok:
stack.pop()
if len(stack) == 0:
break
else:
#print('Two concepts: %s - %s'%(top_concept, tok))
break
codec = penman.AMRCodec()
graph = penman.Graph(triplets)
#print(codec.encode(graph, top=root))
return codec.encode(graph, top=root)
def __init__(self, rel_vocab):
self.amr = penman.AMRCodec()
self.rel_vocab = rel_vocab
