def args_syn_wn(self, lemmas_t, lemmas_h, score, id, type):
result = 0
wn = WNTools()
tool = RTETools()
tool.set_tokens(lemmas_t.split())
tool.quit_punct()
lemmas_t = tool.quit_sw()
tool.set_tokens(lemmas_h.split())
tool.quit_punct()
lemmas_h = tool.quit_sw()
(diff_ht, diff_th) = self.diff(lemmas_t, lemmas_h)
expand_diff_ht = wn.expand_bow_syns(list(diff_ht))
expand_diff_th = wn.expand_bow_syns(list(diff_th))
if len(expand_diff_ht) != 0 and len(expand_diff_th) !=0:
sim = SetMetrics(expand_diff_ht, expand_diff_th)
if sim.cosine() > 0:
result = 1
else:
result = 1
predicate = '>arg_relsyn\n%s %s %s'%(id, type, result)
return predicate
def args_syn_wn(self, lemmas_t, lemmas_h, score, id, type):
result = 0
wn = WNTools()
tool = RTETools()
tool.set_tokens(lemmas_t.split())
tool.quit_punct()
lemmas_t = tool.quit_sw()
tool.set_tokens(lemmas_h.split())
tool.quit_punct()
lemmas_h = tool.quit_sw()
(diff_ht, diff_th) = self.diff(lemmas_t, lemmas_h)
expand_diff_ht = wn.expand_bow_syns(list(diff_ht))
expand_diff_th = wn.expand_bow_syns(list(diff_th))
if len(expand_diff_ht) != 0 and len(expand_diff_th) !=0:
sim = SetMetrics(expand_diff_ht, expand_diff_th)
if sim.cosine() > 0:
result = 1
else:
result = 1
predicate = 'ArgRelSyn(%s, %s, %s)'%(id, type, result)
return predicate
def arg_proc(self, id, point, sep):
self.a_predicates = []
n = 3 #levels in wn tree
if 'verbs' in point:
verbs = point['verbs']
for i, verb in verbs.iteritems():
i = '%s.%s'%(id, i)
(vt, vh) = verb['tokens']
if 'ARG' in verb:
args = verb['ARG']
for type,arg in args.items():
w_t = arg['wordform-t'].split()
w_h = arg['wordform-h'].split()
l_t = arg['lemma-t'].split()
l_h = arg['lemma-h'].split()
p_t = arg['pos-t'].split()
p_h = arg['pos-h'].split()
c_t = arg['chunk-t'].split()
c_h = arg['chunk-h'].split()
n_t = arg['ne-t'].split()
n_h = arg['ne-h'].split()
score = arg['score']
w_t.extend(w_h)
l_t.extend(l_h)
p_t.extend(p_h)
#TODO quit stop words
for j, word in enumerate(w_t):
word_arg = '>token_word\n"%s" %s "%s"'%(type, j, self.clean_str(word))
lemma_arg = '>token_lemma\n"%s" %s "%s"'%(type, j, self.clean_str(l_t[j]))
pos_arg = '>token_pos\n"%s" %s "%s"'%(type, j, self.clean_str(p_t[j]))
lin = Lin()
sim_words = lin.expand_w(word)
wn = WNTools()
hyps = wn.get_mfs_hypernyms((l_t[j], p_t[j]))
self.a_predicates.append(word_arg)
self.a_predicates.append(lemma_arg)
self.a_predicates.append(pos_arg)
for j, sim_word in enumerate(sim_words):
lin_arg = '>token_lin\n"%s" %s "%s"'%(type, j, self.clean_str(sim_word))
self.a_predicates.append(lin_arg)
for key, tree in hyps:
j = 0
for category in tree[:n]:
hyp_arg = '>token_wn\n"%s" %s "%s"'%(type, j, self.clean_str(category))
j += 1
self.a_predicates.append(hyp_arg)
arg_id = '>arg\n"%s" "%s%s%s" %s'%(type, self.clean_str(vt), sep, self.clean_str(vh), id)
self.a_predicates.append(arg_id)
return self.a_predicates
def arg_proc(self, id, point, sep):
self.a_predicates = []
if 'verbs' in point:
verbs = point['verbs']
for i, verb in verbs.iteritems():
i = '%s.%s'%(id, i)
(vt, vh) = verb['tokens']
if 'ARG' in verb:
args = verb['ARG']
for type,arg in args.items():
w_t = arg['wordform-t'].split()
w_h = arg['wordform-h'].split()
l_t = arg['lemma-t'].split()
l_h = arg['lemma-h'].split()
p_t = arg['pos-t'].split()
p_h = arg['pos-h'].split()
c_t = arg['chunk-t'].split()
c_h = arg['chunk-h'].split()
n_t = arg['ne-t'].split()
n_h = arg['ne-h'].split()
score = arg['score']
w_t.extend(w_h)
l_t.extend(l_h)
p_t.extend(p_h)
#TODO quit stop words
for j, word in enumerate(w_t):
word_arg = 'Token(%s, %s, "%s")'%(type, id, self.clean_str(word))
lemma_arg = 'Token(%s, %s, "%s")'%(type, id, self.clean_str(l_t[j]))
pos_arg = 'Token(%s, %s, "%s")'%(type, id, self.clean_str(p_t[j]))
lin = Lin()
sim_words = lin.expand_w(word)
wn = WNTools()
hyps = wn.get_mfs_hypernyms((l_t[j], p_t[j]))
self.a_predicates.append(word_arg)
self.a_predicates.append(lemma_arg)
self.a_predicates.append(pos_arg)
for sim_word in sim_words:
lin_arg = 'Token(%s, %s, "%s")'%(type, id, self.clean_str(sim_word))
self.a_predicates.append(lin_arg)
for key, tree in hyps:
t = 0
for category in tree:
if t >= 3:
break
hyp_arg = 'Token(%s, %s, "%s")'%(type, id, self.clean_str(category))
self.a_predicates.append(hyp_arg)
t +=1
arg_id = 'ARG(%s, %s, %s)'%(type, i, id)
self.a_predicates.append(arg_id)
return self.a_predicates
def __init__(self,
frames_text={},
tokens_text=[],
chunks_text=[],
frames_hypo={},
tokens_hypo=[],
chunks_hypo=[],
e_type='simple',
verbose=1,
entailment=-1):
self.e_type = e_type
self.srl_t = SRLTools()
self.srl_h = SRLTools()
self.verb_net = VerbMetrics()
self.arg_sim = SetMetrics()
self.lin = Lin()
self.wn = WNTools()
self.frames_hypo = frames_text
self.frames_text = frames_hypo
self.tokens_hypo = tokens_hypo
self.tokens_text = tokens_text
self.chunks_hypo = chunks_hypo
self.chunks_text = chunks_text
self.args_text = {}
self.args_hypo = {}
self.verbs_text = []
self.verbs_hypo = []
self.edit_score = 0
self.verb_score = 0
self.chunk_score = 0
self.oper_type = {'del': 0, 'in': 0, 'sub': 0}
self.verbose = verbose
self.entailment = entailment
return
def args_hyp_wn(self, lemmas_t, pos_t, lemmas_h, pos_h, score, id, type):
result = 0
wn = WNTools()
tool = RTETools()
tool.set_tokens(lemmas_t)
tool.quit_punct()
lemmas_t = tool.quit_sw()
tool.set_tokens(lemmas_h)
tool.quit_punct()
lemmas_h = tool.quit_sw()
(diff_ht, diff_th) = self.diff(lemmas_t, lemmas_h)
#TODO
predicate = 'HypRelSyn(%s, %s)'%(id, type, result)
return predicate
def __init__(self, frames_text = {}, tokens_text = [], frames_hypo = {}, tokens_hypo = [], sim_type = 'Lin', verbose = 1):
self.srl_t = SRLTools()
self.srl_h = SRLTools()
self.verb_net = VerbMetrics()
self.arg_sim = SetMetrics()
self.lin = Lin()
self.wn = WNTools()
self.frames_hypo = frames_text
self.frames_text = frames_hypo
self.tokens_hypo = tokens_hypo
self.tokens_text = tokens_text
self.args_text = {}
self.args_hypo = {}
self.verbs_text = []
self.verbs_hypo = []
self.tine_score = 0
self.verb_score = 0
self.arg_score = 0
self.sim_type = sim_type
self.verbose = verbose
#self.pos_text = pos_text
#self.pos_hypo = pos_hypo
return
def main(args):
pickle_file = args[0]
print 'loading file:',pickle_file
with open(pickle_file, 'r') as pf:
pairs = pickle.load(pf)
k = 0
for pair in pairs:
print 'id:', pair.get_id()
print 's1:', pair.get_text()
print 's2:', pair.get_hypo()
print 'features:', pair.get_features_text_type()
print 'set-metrics, cos test'
lemmas_text = pair.get_feature_text('lemmas')
lemmas_hypo = pair.get_feature_hypo('lemmas')
set_th = SetMetrics(lemmas_text, lemmas_hypo)
cos = set_th.cosine()
#print cos
print 'SRL tools'
frames_text = pair.get_feature_text('frames')
print frames_text
print '################'
srl = SRLTools(lemmas_text, frames_text)
word_to_frame = srl.get_words_frame()
print word_to_frame
print '################'
print srl.get_verbs()
print '################'
#print 'verb-metrics, '
pos_text = pair.get_feature_text('pos')
pos_hypo = pair.get_feature_hypo('pos')
verbs = VerbMetrics()
lin = Lin()
vectors = VectorMetrics()
hyper = WNTools()
for i, pos_tuple_t in enumerate(pos_text):
(token, pos_t) = pos_tuple_t
if pos_t.startswith('V'):
for j, pos_tuple_h in enumerate(pos_hypo):
(token, pos_h) = pos_tuple_h
if pos_h.startswith('V'):
verbs.set_text_verb(lemmas_text[i])
verbs.set_hypo_verb(lemmas_hypo[j])
#print 'verbs test t:%s h:%s'%(lemmas_text[i], lemmas_hypo[j])
vn_isec = verbs.vn_isec()
#print 'verb net isec: %d'%vn_isec
#print 'lin(%s):'%lemmas_text[i], '\n', lin.n_similar_words(lemmas_text[i])
#print 'lin(%s):'%lemmas_hypo[j], '\n', lin.n_similar_words(lemmas_hypo[j])
t_sim = lin.n_similar_words(lemmas_text[i])
h_sim = lin.n_similar_words(lemmas_hypo[j])
t_score = [float(score) for word,score in t_sim]
h_score = [float(score) for word,score in h_sim]
vectors.set_vectors(t_score, h_score)
#print 'cos_vect: ', vectors.cosine()
elif pos_h.startswith('N'):
#print 'wn test hypernyms'
trees = hyper.get_mfs_hypernyms((lemmas_hypo[j], pos_h))
#print trees
k += 1
if k >= 10:
break
pf.close
return
def arg_proc(self, id, point, sep):
self.a_predicates = []
if 'verbs' in point:
verbs = point['verbs']
for i, verb in verbs.iteritems():
(vt, vh) = verb['tokens']
if 'ARG' in verb:
args = verb['ARG']
for type,arg in args.items():
w_t = arg['wordform-t']
w_h = arg['wordform-h']
l_t = arg['lemma-t']
l_h = arg['lemma-h']
p_t = arg['pos-t']
p_h = arg['pos-h']
c_t = arg['chunk-t']
c_h = arg['chunk-h']
n_t = arg['ne-t']
n_h = arg['ne-h']
score = arg['score']
token_arg = '>token_arg\n%s %s "%s%s%s"'%(id, type, self.clean_str(w_t), sep, self.clean_str(w_h))
lemma_arg = '>lemma_arg\n%s %s "%s%s%s"'%(id, type, self.clean_str(l_t), sep, self.clean_str(l_h))
pos_arg = '>pos_arg\n%s %s "%s%s%s"'%(id, type, self.clean_str(p_t), sep, self.clean_str(p_h))
#expand content words Lin
sw_tool = RTETools()
sw_tool.set_tokens(l_t.split())
sw_tool.quit_sw()
l_tsw = sw_tool.quit_punct()
sw_tool.set_tokens(l_h.split())
sw_tool.quit_sw()
l_hsw = sw_tool.quit_punct()
content_arg = '>cont_arg\n%s %s "%s%s%s"'%(id, type, self.clean_str(' '.join(l_tsw)),
sep, self.clean_str(' '.join(l_hsw)))
sim_diff = self.sim_dif(l_tsw, l_hsw)
diff_arg = '>diff_arg\n%s %s "%s"'%(id, type, self.clean_str(' '.join(sim_diff)))
if score >= 0.4:
score_arg = '>sim_arg\n%s %s 1'%(id, type)
else: #TODO find threshold from data statistics mf
score_arg = '>sim_arg\n%s %s 0'%(id, type)
lin = Lin()
bow_t = lin.expand_bow(l_tsw)
bow_h = lin.expand_bow(l_hsw)
lin_arg = '>lin_arg\n%s %s "%s%s%s"'%(id, type, self.clean_str(' '.join(bow_t)),
sep, self.clean_str(' '.join(bow_h)))
wn = WNTools()
wns_t = zip(l_t.split(), p_t.split())
wns_h = zip(l_h.split(), p_h.split())
bow_wnt = wn.expand_bow_tree(wns_t)
bow_wnh = wn.expand_bow_tree(wns_h)
wn_arg = '>wn_arg\n%s %s "%s%s%s"'%(id, type, self.clean_str(' '.join(bow_wnt)),
sep, self.clean_str(' '.join(bow_wnh)))
rel_syn_args = self.args_syn_wn(l_t, l_h, score, id, type)
self.a_predicates.append(token_arg)
self.a_predicates.append(score_arg)
self.a_predicates.append(lemma_arg)
self.a_predicates.append(rel_syn_args)
self.a_predicates.append(content_arg)
self.a_predicates.append(diff_arg)
self.a_predicates.append(pos_arg)
self.a_predicates.append(lin_arg)
self.a_predicates.append(wn_arg)
return self.a_predicates
def arg_proc(self, id, point, sep):
self.a_predicates = []
if 'verbs' in point:
verbs = point['verbs']
for i, verb in verbs.iteritems():
(vt, vh) = verb['tokens']
if 'ARG' in verb:
args = verb['ARG']
for type,arg in args.items():
w_t = arg['wordform-t']
w_h = arg['wordform-h']
l_t = arg['lemma-t']
l_h = arg['lemma-h']
p_t = arg['pos-t']
p_h = arg['pos-h']
c_t = arg['chunk-t']
c_h = arg['chunk-h']
n_t = arg['ne-t']
n_h = arg['ne-h']
score = arg['score']
token_arg = 'TokenArg(%s, %s, "%s %s %s")'%(id, type, self.clean_str(w_t), sep, self.clean_str(w_h))
lemma_arg = 'LemmaArg(%s, %s,"%s %s %s")'%(id, type, self.clean_str(l_t), sep, self.clean_str(l_h))
pos_arg = 'PosArg(%s, %s, "%s %s %s")'%(id, type, self.clean_str(p_t), sep, self.clean_str(p_h))
#expand content words Lin
sw_tool = RTETools()
sw_tool.set_tokens(l_t.split())
sw_tool.quit_sw()
l_tsw = sw_tool.quit_punct()
sw_tool.set_tokens(l_h.split())
sw_tool.quit_sw()
l_hsw = sw_tool.quit_punct()
content_arg = 'ContArg(%s, %s, "%s %s %s")'%(id, type, self.clean_str(' '.join(l_tsw)),
sep, self.clean_str(' '.join(l_hsw)))
sim_diff = self.sim_dif(l_tsw, l_hsw)
diff_arg = 'DiffArg(%s, %s, "%s")'%(id, type, self.clean_str(' '.join(sim_diff)))
if score >= 0.4:
score_arg = 'SimArg(%s, %s, 1)'%(id, type)
else: #TODO find threshold from data statistics mf
score_arg = 'SimArg(%s, %s, 0)'%(id, type)
lin = Lin()
bow_t = lin.expand_bow(l_tsw)
bow_h = lin.expand_bow(l_hsw)
lin_arg = 'LinArg(%s, %s, "%s %s %s")'%(id, type, self.clean_str(' '.join(bow_t)),
sep, self.clean_str(' '.join(bow_h)))
wn = WNTools()
wns_t = zip(l_t.split(), p_t.split())
wns_h = zip(l_h.split(), p_h.split())
bow_wnt = wn.expand_bow_tree(wns_t)
bow_wnh = wn.expand_bow_tree(wns_h)
wn_arg = 'WnArg(%s, %s, "%s %s %s")'%(id, type, self.clean_str(' '.join(bow_wnt)),
sep, self.clean_str(' '.join(bow_wnh)))
rel_syn_args = self.args_syn_wn(l_t, l_h, score, id, type)
self.a_predicates.append(token_arg)
self.a_predicates.append(score_arg)
self.a_predicates.append(lemma_arg)
self.a_predicates.append(rel_syn_args)
self.a_predicates.append(content_arg)
self.a_predicates.append(diff_arg)
self.a_predicates.append(pos_arg)
self.a_predicates.append(lin_arg)
self.a_predicates.append(wn_arg)
return self.a_predicates
def arg_proc(self, id, point, sep):
self.a_predicates = []
n = 3 #levels in wn tree
if 'verbs' in point:
verbs = point['verbs']
for i, verb in verbs.iteritems():
i = '%s.%s' % (id, i)
(vt, vh) = verb['tokens']
if 'ARG' in verb:
args = verb['ARG']
for type, arg in args.items():
w_t = arg['wordform-t'].split()
w_h = arg['wordform-h'].split()
l_t = arg['lemma-t'].split()
l_h = arg['lemma-h'].split()
p_t = arg['pos-t'].split()
p_h = arg['pos-h'].split()
c_t = arg['chunk-t'].split()
c_h = arg['chunk-h'].split()
n_t = arg['ne-t'].split()
n_h = arg['ne-h'].split()
score = arg['score']
w_t.extend(w_h)
l_t.extend(l_h)
p_t.extend(p_h)
#TODO quit stop words
for j, word in enumerate(w_t):
word_arg = '>token_word\n"%s" %s "%s"' % (
type, j, self.clean_str(word))
lemma_arg = '>token_lemma\n"%s" %s "%s"' % (
type, j, self.clean_str(l_t[j]))
pos_arg = '>token_pos\n"%s" %s "%s"' % (
type, j, self.clean_str(p_t[j]))
lin = Lin()
sim_words = lin.expand_w(word)
wn = WNTools()
hyps = wn.get_mfs_hypernyms((l_t[j], p_t[j]))
self.a_predicates.append(word_arg)
self.a_predicates.append(lemma_arg)
self.a_predicates.append(pos_arg)
for j, sim_word in enumerate(sim_words):
lin_arg = '>token_lin\n"%s" %s "%s"' % (
type, j, self.clean_str(sim_word))
self.a_predicates.append(lin_arg)
for key, tree in hyps:
j = 0
for category in tree[:n]:
hyp_arg = '>token_wn\n"%s" %s "%s"' % (
type, j, self.clean_str(category))
j += 1
self.a_predicates.append(hyp_arg)
arg_id = '>arg\n"%s" "%s%s%s" %s' % (
type, self.clean_str(vt), sep, self.clean_str(vh),
id)
self.a_predicates.append(arg_id)
return self.a_predicates
class TineVN:
def __init__(self, frames_text = {}, tokens_text = [], frames_hypo = {}, tokens_hypo = [], sim_type = 'Lin', verbose = 1):
self.srl_t = SRLTools()
self.srl_h = SRLTools()
self.verb_net = VerbMetrics()
self.arg_sim = SetMetrics()
self.lin = Lin()
self.wn = WNTools()
self.frames_hypo = frames_text
self.frames_text = frames_hypo
self.tokens_hypo = tokens_hypo
self.tokens_text = tokens_text
self.args_text = {}
self.args_hypo = {}
self.verbs_text = []
self.verbs_hypo = []
self.tine_score = 0
self.verb_score = 0
self.arg_score = 0
self.sim_type = sim_type
self.verbose = verbose
#self.pos_text = pos_text
#self.pos_hypo = pos_hypo
return
def get_tine_score(self, frames_text = {}, tokens_text = [], frames_hypo = {}, tokens_hypo = []):
if frames_text:
self.frames_text = frames_text
if frames_hypo:
self.frames_hypo = frames_hypo
if tokens_text:
self.tokens_text = tokens_text
if tokens_hypo:
self.tokens_hypo = tokens_hypo
self.srl_t.set_frames(self.frames_text)
self.srl_h.set_frames(self.frames_hypo)
self.srl_t.set_tokens(self.tokens_text)
self.srl_h.set_tokens(self.tokens_hypo)
self.args_text = self.srl_t.get_words_frame()
self.args_hypo = self.srl_h.get_words_frame()
sum_verb = 0
num_verbs_h = len(self.args_text.keys())
self.__p_stderr('TINE VerbNet\n')
self.__p_stderr('T: %s \n H: %s\n'%(self.args_text, self.args_hypo))
self.__p_stderr('T: %s \n H: %s\n'%(self.args_text.keys(), self.args_hypo.keys()))
for verb_t, args_t in self.args_text.items():
for verb_h, args_h in self.args_hypo.items():
sim_verbs = self.__simVerbs(verb_t, verb_h)
if sim_verbs == 1:
self.__p_stderr('verbs(%s, %s)\n'%(verb_t, verb_h))
args_score = self.__simArgs(args_t, args_h)
sum_verb += args_score
self.tine_score = float(sum_verb) / num_verbs_h
self.__p_stderr('score:%s\n'%(self.tine_score))
return self.tine_score
def __simVerbs(self, verb_t = '', verb_h = ''):
if verb_t == verb_h:
return 1
self.verb_net.set_text_verb(verb_t)
self.verb_net.set_hypo_verb(verb_h)
isec = self.verb_net.vn_isec()
if isec == 0:
vo = self.verb_net.vo()
return vo
else:
return isec
return isec
def __simArgs(self, args_t = [], args_h = []):
sum_args = 0
num_args_h = len(args_h)
for tag_t, tokens_t in args_t:
for tag_h, tokens_h in args_h:
if tag_t == tag_h:
expand_t = []
expand_h = []
if self.sim_type == 'Lin':
expand_t = self.lin.expand_bow(tokens_t)
expand_h = self.lin.expand_bow(tokens_h)
elif self.sim_type == 'WN':
expand_t = self.wn.expand_bow_tree(tokens_t)
expand_h = self.wn.expand_bow_tree(tokens_h)
self.arg_sim.set_text(expand_t)
self.arg_sim.set_hypo(expand_h)
self.arg_score = self.arg_sim.cosine()
self.__p_stderr('\t[%s|%s] %s %s\n'%(tag_t, self.arg_score, expand_t, expand_h))
sum_args += self.arg_score
if num_args_h == 0:
return 0
else:
self.verb_score = float(sum_args) / num_args_h
return self.verb_score
def __p_stderr(self, text = ''):
if self.verbose == 1:
sys.stderr.write(text)
return
def get_verb_score(self):
return self.verb_score
def get_arg_score(self):
return self.arg_score