def __init__(self, grammar, drt_parser):
assert isinstance(grammar, str) and grammar.endswith('.fcfg'), \
"%s is not a grammar name" % grammar
self.drt_parser = drt_parser()
self.presupp_parser = PresuppDrtParser()
self.logic_parser = LogicParser()
self.parser = load_parser(grammar, logic_parser=self.drt_parser)
class Tester(object):
INFERROR = {
3 : AdmissibilityError,
2 : InformativityError,
1 : ConsistencyError
}
WORD_SPLIT = re.compile(" |, |,")
EXCLUDED_NEXT = re.compile("^ha[sd]|is|was|not|will$")
EXCLUDED = re.compile("^does|h?is|red|[a-z]+ness$")
SUBSTITUTIONS = [
(re.compile("^died$"), ("did", "die")),
(re.compile("^([A-Z][a-z]+)'s?$"), lambda m: (m.group(1), "s")),
(re.compile("^(?P<stem>[a-z]+)s$"), lambda m: ("does", m.group("stem"))),
(re.compile("^([a-z]+(?:[^cvklt]|lk|nt))ed|([a-z]+[cvlkt]e)d$"), lambda m: ("did", m.group(1) if m.group(1) else m.group(2))),
(re.compile("^([A-Z]?[a-z]+)one$"), lambda m: (m.group(1), "one")),
(re.compile("^([A-Z]?[a-z]+)thing$"), lambda m: (m.group(1), "thing")),
(re.compile("^bit$"), ("did", "bite")),
(re.compile("^bought$"), ("did", "buy")),
(re.compile("^wrote$"), ("did", "write")),
]
def __init__(self, grammar, drt_parser):
assert isinstance(grammar, str) and grammar.endswith('.fcfg'), \
"%s is not a grammar name" % grammar
self.drt_parser = drt_parser()
self.presupp_parser = PresuppDrtParser()
self.logic_parser = LogicParser()
self.parser = load_parser(grammar, logic_parser=self.drt_parser)
def _split(self, sentence):
words = []
exlude_next = False
for word in Tester.WORD_SPLIT.split(sentence):
match = None
if Tester.EXCLUDED_NEXT.match(word):
exlude_next = True
words.append(word)
continue
if exlude_next or Tester.EXCLUDED.match(word):
exlude_next = False
words.append(word)
continue
for pattern, replacement in Tester.SUBSTITUTIONS:
match = pattern.match(word)
if match:
if isinstance(replacement, LambdaType):
words.extend(replacement(match))
else:
words.extend(replacement)
break
if not match:
words.append(word)
return words
def parse(self, text, **args):
sentences = text.split('.')
utter = args.get("utter", True)
verbose = args.get("verbose", False)
drs = (utter and self.drt_parser.parse('DRS([n],[])')) or []
for sentence in sentences:
sentence = sentence.lstrip()
if sentence:
words = self._split(sentence)
if verbose:
print words
trees = self.parser.nbest_parse(words)
try:
new_drs = trees[0].node['SEM'].simplify()
except IndexError:
raise UngrammaticalException()
if verbose:
print(new_drs)
if drs:
drs = (drs + new_drs).simplify()
else:
drs = new_drs
if verbose:
print drs
return drs
def test(self, cases, **args):
verbose = args.get("verbose", False)
for number, sentence, expected in cases:
expected_drs = []
if expected:
for item in expected if isinstance(expected, list) else [expected]:
expected_drs.append(self.presupp_parser.parse(item, verbose))
try:
expression = self.parse(sentence, **args)
readings, errors = expression.resolve(lambda x: (True, None), verbose)
if len(expected_drs) == len(readings):
for index, pair in enumerate(zip(expected_drs, readings)):
if pair[0] == pair[1]:
print("%s. %s -- Reading (%s): %s\n" % (number, sentence, index + 1, pair[1]))
else:
print("%s. !!!failed reading (%s)!!!\n\n%s\n\nExpected:\t%s\n\nReturns:\t%s\n" %
(number, index + 1, sentence, pair[0], pair[1]))
else:
print("%s. !!!comparison failed!!!\n\n%s\n" % (number, sentence))
except Exception as e:
if type(e) is ResolutionException:
print("%s. *%s -- Exception:%s\n" % (number, sentence, e))
else:
print("%s. !!!unexpected error!!!\n%s\n%s" % (number, sentence, e))
def interpret(self, expr_1, expr_2, background=None, verbose=False, test=False):
"""Interprets a new expression with respect to some previous discourse
and background knowledge. The function first generates relevant background
knowledge and then performs inference check on readings generated by
the resolve() method. It returns a list of admissible interpretations in
the form of DRSs."""
assert(not expr_1 or isinstance(expr_1, str)), "Expression %s is not a string" % expr_1
assert(isinstance(expr_2, str)), "Expression %s is not a string" % expr_2
assert(not background or isinstance(background, dict)), "Background knowledge is not in dictionary format"
try:
if expr_1:
discourse = self.parse(expr_1, utter=True)
expression = self.parse_new(discourse, expr_2)
else:
discourse = None
expression = self.parse(expr_2, utter=True)
interpretations, errors = self.interpret_new(discourse, expression, background=background, verbose=verbose)
if test:
return interpretations, errors
else:
return interpretations
except IndexError:
print "Input sentences only!"
except ValueError as e:
print "Error:", e
def collect_background(self, discourse, background, verbose=False):
background_knowledge = None
for formula in get_bk(discourse, background):
try:
parsed_formula = self.presupp_parser.parse(formula).fol()
except ParseException:
try:
parsed_formula = self.logic_parser.parse(formula)
except Exception as e:
print "Error: %s" % e
if background_knowledge:
background_knowledge = AndExpression(background_knowledge, parsed_formula)
else:
background_knowledge = parsed_formula
if verbose:
print "Generated background knowledge:\n%s" % background_knowledge
return background_knowledge
def parse_new(self, discourse, expression_str):
"""parse the new expression and make sure that it has unique variables"""
expression = self.parse(expression_str, utter=False)
for ref in set(expression.get_refs(True)) & set(discourse.get_refs(True)):
newref = DrtVariableExpression(unique_variable(ref))
expression = expression.replace(ref, newref, True)
return expression
def interpret_new(self, discourse, expression, background=None, verbose=False):
"""Interprets a new expression with respect to some previous discourse
and background knowledge. The function first generates relevant background
knowledge and then performs inference check on readings generated by
the resolve() method. It returns a list of admissible interpretations in
the form of DRSs."""
try:
if discourse:
new_discourse = (NewInfoDRS([], [expression]) + discourse).simplify()
else:
new_discourse = expression
if background:
background_knowledge = self.collect_background(new_discourse, background, verbose)
else:
background_knowledge = None
return new_discourse.resolve(lambda x: inference_check(x, background_knowledge, verbose), verbose)
except IndexError:
print "Input sentences only!"
except ValueError as e:
print "Error: %s" % e
def inference_test(self, cases, bk, verbose=False):
for number, discourse, expression, judgement in cases:
print "\n%s. %s %s" % (number, discourse, expression)
interpretations, errors = self.interpret(discourse, expression, bk, verbose=False, test=True)
for interpretation in interpretations:
print "\nAdmissible interpretation: ", interpretation
if judgement:
if not isinstance(judgement, list):
judgement = [judgement]
if len(judgement) == len(errors):
for index, error in enumerate(errors):
error_message = Tester.INFERROR.get(judgement[index], False)
if verbose:
print "\nexpected error:%s" % error_message
print "\nreturned error:%s" % error[1]
if type(error[1]) is error_message:
print "\nInadmissible reading %s returns as expected:\n\t%s" % (error[0], error_message.__name__)
else:
print "\n#!!!#: Inadmissible reading %s returned with unexpected error: %s" % (error[0], error[1])
else:
print "\n#!!!#: !Unexpected error! #!!!#"
else:
print "\nNo inadmissible readings"
