def pronominalise(np, *features):
"""Create a pronoun for the corresponding noun phrase. """
# features can be: person, gender, subject|object (case),
# possessive determiner, possessive pronoun, reflexive
get_log().info('Doing pronominalisation on {0}'.format(repr(np)))
tmp = [x for x in features if str(Gender) == x[0]]
if len(tmp) == 1:
gender = tmp[0]
else:
gender = lexicon.guess_phrase_gender(np)
all_features = list(features)
if gender == Gender.epicene:
all_features = [x for x in all_features if x != Number.singular]
all_features.append(Number.plural)
else:
all_features.append(gender)
all_features.extend(list(np._features.items()))
get_log().debug('Phrase features for pronominalisation:\n\t{}'
.format(all_features))
res = lexicon.pronoun_for_features(*all_features)
get_log().debug('\tresult:{}'.format(res))
return res
def optimise_ref_exp(phrase, context):
"""Replace anaphoric noun phrases with pronouns when possible. """
# TODO: include Number in the dicision process (it vs they)
# FIXME: Coordinated elements need some special attention
result = copy(phrase)
# test for selecting phrases taht can be processed
test = lambda x: isinstance(x, NounPhrase) or isinstance(x, Coordination)
# reverse so that we start with large phrases first (eg CC)
get_log().debug('-='*40)
get_log().debug('constituents:')
for x in phrase.constituents():
get_log().debug('\t {}'.format(' '.join(repr(x).split())))
nps = [x for x in phrase.constituents() if test(x)]
uttered = []
processed_ids = set()
for np in nps:
replaced = False
get_log().debug('current NP:\n{}'.format(np))
gender = lexicon.guess_phrase_gender(np)
get_log().debug('gender of NP: {}'.format(gender))
number = lexicon.guess_phrase_number(np)
get_log().debug('number of NP: {}'.format(number))
if not np.has_feature('PERSON'):
if context.is_last_speaker(np):
person = Person.first
else:
person = Person.third
else:
person = ('PERSON', np.get_feature('PERSON'))
phrases = [x for x in (context.np_stack + uttered)
if lexicon.guess_phrase_gender(x) == gender]
# get_log().debug('distractors of NP:\n\t{}'.format(distractors))
if id(np) in processed_ids:
get_log().debug('current NP: {} was already processed'.format(np))
continue
# if ((np in context.np_stack or np in uttered) and np == phrases[-1]):
if (np in phrases[-1:]):
# this np is the most salient so pronominalise it
if isinstance(phrase, Clause):
if id(np) == id(phrase.subj):
pronoun = pronominalise(np, gender, PronounUse.subjective, person)
elif (np in phrase.subj.constituents() and
np in phrase.vp.constituents()):
pronoun = pronominalise(np, gender, PronounUse.reflexive, person)
# TODO: implement -- possessive will be used if it is a complement of an NP?
# elif any(id(np) in [id(x) for x in pp.constituents()]
# for pp in pps):
# pronoun = pronominalise(np, gender, PronounUse.possessive)
elif (np in phrase.vp.constituents()):
pronoun = pronominalise(np, gender, PronounUse.objective, person)
else:
pronoun = pronominalise(np, gender, PronounUse.subjective, person)
else:
pronoun = pronominalise(np, gender, PronounUse.subjective, person)
get_log().debug('replacing {}:{} with {}'.format(id(np), np, pronoun))
replace_element_with_id(result, id(np), pronoun)
replaced = True
# if you replace an element, remove all the subphrases from the list
processed = [y for y in np.constituents()]
processed_ids.update([id(x) for x in processed])
unspec_np = deepcopy(np)
unspec_np.spec = Element()
uttered.append(unspec_np)
if not replaced:
# fix determiners in the processed NP
optimise_determiner(np, phrases, context)
context.add_sentence(phrase)
return result
