def clean(self, sys_act=None):
"""
"""
#TODO - deal with duplicates, probabilities, others?
tempActs = {}
#first add context to user acts
for act in self.semanticActs:
[act
] = contextUtils._add_context_to_user_act(sys_act, [[act, 1.0]],
self.domainTag)
act = act[0]
intent, slotValues = self._parseDialogueAct(act)
if intent not in tempActs:
tempActs[intent] = slotValues
else:
for slot in slotValues:
if slot in tempActs[intent]:
if slotValues[slot] != tempActs[intent][slot]:
logger.warning(
'Sematic decoding of input lead to different interpretations within one hypothesis. Slot {} has values {} and {}.'
.format(slot, slotValues[slot],
tempActs[intent][slot]))
else:
tempActs[intent][slot] = slotValues[slot]
#x
hypos = []
for intent in tempActs:
if len(tempActs[intent]):
(key, value) = tempActs[intent].items()[0]
if value is not None:
hypos.append('{}({})'.format(
intent,
','.join('%s=%s' % (key, value)
for (key,
value) in tempActs[intent].items())))
else:
hypos.append('{}({})'.format(
intent,
','.join('%s' % (key)
for (key,
value) in tempActs[intent].items())))
else:
logger.warning("intent {} found in input without arguments")
self.semanticActs = "|".join(hypos)
def simulate_add_context_to_user_act(self, sys_act, user_acts, domainTag):
'''
NB: only for Simulate.py
While for simulation no semantic decoding is needed, context information needs to be added in some instances. This is done with this method.
:param sys_act: the last system act
:type sys_act: str
:param user_acts: user act hypotheses
:type user_acts: list
:param domainTag: the domain of the dialogue
:type domainTag: str
'''
# simulate - will only pass user_act, and we will return contextual user act
self.active_domain = domainTag # this is required below in checking binary slots of given domain
hyps = contextUtils._add_context_to_user_act(
sys_act, hyps=user_acts, active_domain=self.active_domain)
return hyps # just the act
def simulate_add_context_to_user_act(self, sys_act, user_acts, eType):
'''
NB: only for Simulate.py
While for simulation no semantic decoding is needed, context information needs to be added in some instances. This is done with this method.
:param sys_act: the last system act
:type sys_act: str
:param user_acts: user act hypotheses
:type user_acts: list
:param domainTag: the domain of the dialogue
:type domainTag: str
'''
# simulate - will only pass user_act, and we will return contextual user act
hyps = contextUtils._add_context_to_user_act(sys_act.to_string_plain(),
hyps=user_acts,
active_domain=eType)
return hyps # just the act
def decode(self, ASR_obs, sys_act, domainTag, turn=None):
'''
The main method for semantic decoding. It takes the ASR input and returns a list of semantic interpretations. To decode, the task is delegated to the respective domain semantic decoder
:param ASR_obs: ASR hypotheses
:type ASR_obs: list
:param sys_act: is the system action prior to collecting users response in obs.
:type sys_act: str
:param domainTag: is the domain we want to parse the obs in
:type domainTag: str
:param turn: the turn id, this parameter is optional
:type turn: int
:return: None
'''
# use the correct domain:
self.active_domain = domainTag # used down call chain in adding context
self._ensure_booted(domainTag)
# --------
# explore how to clean IBM asr - "i dont care" problem ...
for i in range(len(ASR_obs)):
was = ASR_obs[i][0]
fix = [
str(c) for c in was if c.isalpha() or c.isspace() or c == "'"
or c != "!" or c != "?"
] # slow way - initial fix #lmr46 06/09/16 this filtering filters question marks that are important!!!! syj requirement
res = ''.join(fix)
ASR_obs[i] = (res.rstrip(), ASR_obs[i][1])
#---------------------------------------------------
# Additional argument turn as described above
hyps = self.semiManagers[domainTag].decode(ASR_obs, sys_act, turn=turn)
logger.info(hyps)
# add context if required
hyps = contextUtils._add_context_to_user_act(sys_act, hyps,
self.active_domain,
self.ontology)
return hyps