def _make_labels(nodes, links, vgen):
nids = [node.nodeid for node in nodes]
edges = []
for l in links:
if safe_int(l.start) == LTOP_NODEID:
nids = [l.start] + nids
vgen.vid = 0 # start at h0 for TOP
if l.post == EQ_POST:
edges.append((l.start, l.end))
labels = {}
# components return in order of nids
for component in _connected_components(nids, edges):
lbl = vgen.new(HANDLESORT)[0]
for nid in component:
labels[nid] = lbl
return labels
def to_triples(self, short_pred=True, properties=True):
"""
Encode the Dmrs as triples suitable for PENMAN serialization.
"""
ts = []
qs = set(self.nodeids(quantifier=True))
for n in nodes(self):
pred = n.pred.short_form() if short_pred else n.pred.string
ts.append((n.nodeid, 'predicate', pred))
if n.lnk is not None:
ts.append((n.nodeid, 'lnk', '"{}"'.format(str(n.lnk))))
if n.carg is not None:
ts.append((n.nodeid, 'carg', '"{}"'.format(n.carg)))
if properties and n.nodeid not in qs:
for key, value in n.sortinfo.items():
ts.append((n.nodeid, key.lower(), value))
for l in links(self):
if safe_int(l.start) == LTOP_NODEID:
ts.append((l.start, 'top', l.end))
else:
relation = '{}-{}'.format(l.rargname.upper(), l.post)
ts.append((l.start, relation, l.end))
return ts
def __init__(self,
nodes=None,
links=None,
top=None,
index=None,
xarg=None,
lnk=None,
surface=None,
identifier=None):
if nodes is None: nodes = []
if links is None: links = []
qeq = HandleConstraint.qeq
vgen = _VarGenerator()
# check this here to streamline things later
if top is not None:
links = [Link(LTOP_NODEID, top, None, H_POST)] + list(links)
top = None
labels = _make_labels(nodes, links, vgen)
qs = set(l.start for l in links
if (l.rargname or '').upper() == RSTR_ROLE)
ivs = _make_ivs(nodes, vgen, qs)
# initialize args with ARG0 for intrinsic variables
args = {nid: {IVARG_ROLE: iv} for nid, iv in ivs.items()}
hcons = []
for l in links:
if l.start not in args:
args[l.start] = {}
if safe_int(l.start) != LTOP_NODEID:
if not l.rargname or l.rargname.upper() == BARE_EQ_ROLE:
continue # don't make an argument for bare EQ links
if l.post == H_POST:
hole = vgen.new(HANDLESORT)[0]
hcons += [qeq(hole, labels[l.end])]
args[l.start][l.rargname] = hole
# if the arg is RSTR, it's a quantifier, so we can
# find its intrinsic variable now
if l.rargname.upper() == RSTR_ROLE:
ivs[l.start] = ivs[l.end]
args[l.start][IVARG_ROLE] = ivs[l.start]
elif l.post == HEQ_POST:
args[l.start][l.rargname] = labels[l.end]
else: # NEQ_POST or EQ_POST
args[l.start][l.rargname] = ivs[l.end]
# ignore top link if top is already set
elif top is None:
# The most explicit value of post for a link that denotes a
# TOP that is qeq to a label is H_POST, but I equally accept
# NIL_POST for backward compatibility. HEQ_POST denotes a TOP
# that selects a label directly (and this label equality would
# have been captured earlier)
top = labels[l.start]
if l.post == H_POST or l.post == NIL_POST:
hcons += [qeq(top, labels[l.end])]
eps = []
for node in nodes:
nid = node.nodeid
if node.carg is not None:
args[nid][CONSTARG_ROLE] = node.carg
ep = (nid, node.pred, labels[nid], args[nid], node.lnk,
node.surface, node.base)
eps.append(ep)
icons = None # future feature
super(Dmrs, self).__init__(top=top,
index=ivs.get(index),
xarg=ivs.get(xarg),
eps=eps,
hcons=hcons,
icons=icons,
vars=vgen.store,
lnk=lnk,
surface=surface,
identifier=identifier)
def test_safe_int():
assert safe_int('1') == 1
assert safe_int('1.0') == '1.0'
assert safe_int('-12345') == -12345
assert safe_int('1a') == '1a'
def test_safe_int():
assert safe_int('1') == 1
assert safe_int('1.0') == '1.0'
assert safe_int('-12345') == -12345
assert safe_int('1a') == '1a'
def convert(path,
source_fmt,
target_fmt,
select='result:mrs',
properties=True,
show_status=False,
predicate_modifiers=False,
color=False,
pretty_print=False,
indent=None):
"""
Convert between various DELPH-IN Semantics representations.
Args:
path (str, file): filename, testsuite directory, open file, or
stream of input representations
source_fmt (str): convert from this format
target_fmt (str): convert to this format
select (str): TSQL query for selecting data (ignored if *path*
is not a testsuite directory; default: `"result:mrs"`)
properties (bool): include morphosemantic properties if `True`
(default: `True`)
show_status (bool): show disconnected EDS nodes (ignored if
*target_fmt* is not `"eds"`; default: `False`)
predicate_modifiers (bool): apply EDS predicate modification
for certain kinds of patterns (ignored if *target_fmt* is
not an EDS format; default: `False`)
color (bool): apply syntax highlighting if `True` and
*target_fmt* is `"simplemrs"` (default: `False`)
pretty_print (bool): if `True`, format the output with
newlines and default indentation (default: `False`)
indent (int, optional): specifies an explicit number of spaces
for indentation (implies *pretty_print*)
Returns:
str: the converted representation
"""
if source_fmt.startswith('eds') and not target_fmt.startswith('eds'):
raise ValueError(
'Conversion from EDS to non-EDS currently not supported.')
if indent:
pretty_print = True
indent = 4 if indent is True else safe_int(indent)
if len(tsql.inspect_query('select ' + select)['projection']) != 1:
raise ValueError('Exactly 1 column must be given in selection query: '
'(e.g., result:mrs)')
# read
loads = _get_codec(source_fmt)
if path is None:
xs = loads(sys.stdin.read())
elif hasattr(path, 'read'):
xs = loads(path.read())
elif os.path.isdir(path):
ts = itsdb.TestSuite(path)
xs = [next(iter(loads(r[0])), None) for r in tsql.select(select, ts)]
else:
xs = loads(open(path, 'r').read())
# write
dumps = _get_codec(target_fmt, load=False)
kwargs = {}
if color: kwargs['color'] = color
if pretty_print: kwargs['pretty_print'] = pretty_print
if indent: kwargs['indent'] = indent
if target_fmt == 'eds':
kwargs['pretty_print'] = pretty_print
kwargs['show_status'] = show_status
if target_fmt.startswith('eds'):
kwargs['predicate_modifiers'] = predicate_modifiers
kwargs['properties'] = properties
# this is not a great way to improve robustness when converting
# many representations, but it'll do until v1.0.0. Also, it only
# improves robustness on the output, not the input.
# Note that all the code below is to replace the following:
# return dumps(xs, **kwargs)
head, joiner, tail = _get_output_details(target_fmt)
parts = []
if pretty_print:
joiner = joiner.strip() + '\n'
def _trim(s):
if head and s.startswith(head):
s = s[len(head):].lstrip('\n')
if tail and s.endswith(tail):
s = s[:-len(tail)].rstrip('\n')
return s
for x in xs:
try:
s = dumps([x], **kwargs)
except (PyDelphinException, KeyError, IndexError):
logging.exception('could not convert representation')
else:
s = _trim(s)
parts.append(s)
# set these after so head and tail are used correctly in _trim
if pretty_print:
if head:
head += '\n'
if tail:
tail = '\n' + tail
return head + joiner.join(parts) + tail