###############################################################################
# PyDial: Multi-domain Statistical Spoken Dialogue System Software
###############################################################################
#
# Copyright 2015 - 2019
# Cambridge University Engineering Department Dialogue Systems Group
#
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
###############################################################################
'''
RuleTopicTrackers.py - Rule based topic trackers
==========================================================================
Copyright CUED Dialogue Systems Group 2015 - 2017
.. seealso:: CUED Imports/Dependencies:
import :mod:`utils.Settings` |.|
import :mod:`utils.ContextLogger` |.|
import :mod:`ontology.OntologyUtils` |.|
************************
'''
__author__ = "cued_dialogue_systems_group"
'''
Modifications History
===============================
Date Author Description
===============================
Jul 20 2016 lmr46 Inferring only the domains configured in the config-file
Note that keywords for domains are set in the dictionary here (handcoded)
TODO: What happen when the same keyword apply for different domains?
'''
from utils import Settings, ContextLogger
from ontology import OntologyUtils
logger = ContextLogger.getLogger('')
class TopicTrackerInterface(object):
"""Template for any Topic Tracker for the cued-python system
.. Note:
To dynamically load a class, the __init__() must take one argument: domainString.
"""
def infer_domain(self,userActHyps=None):
pass # Define in actual class. Must set and also return self.current_tracking_result
def restart(self):
pass # Define in actual class. May be some notion of state etc to be reset in more advanced topic trackers
class TextBasedSwitchTopicTracker(TopicTrackerInterface):
"""When using texthub, you can enter: switch("CamRestaurants") which will change domains to CamRestaurants for example.
-- if switch("XX") not entered, assumes you want to stay in domain of previous turn
"""
def __init__(self):
self.restart()
def restart(self):
self.current_tracking_result = None
self.FOUND_DOMAIN = False
def infer_domain(self, userActHyps=None):
"""userActHyps : [(text, prob)]
"""
if 'switch("' in userActHyps[0][0]:
candidateDomain = userActHyps[0][0].split('"')[1] # a little fragile -
if candidateDomain in OntologyUtils.available_domains:
self.current_tracking_result = candidateDomain
self.FOUND_DOMAIN = True
else:
logger.warning("Not a valid domain tag in your switch('X') command - remain with previous domain")
elif not self.FOUND_DOMAIN:
msg = '\nSWITCH TOPIC TRACKER USAGE: When using the texthub switch topic tracker '
msg += '-You should start by saying which domain to switch to.\n'
msg += 'Enter exactly (where DOMAINTAG is CamRestaurants,Laptops6 etc): switch("DOMAINTAG")\n'
msg += 'You can continue on directly by entering for example: switch("DOMAINTAG")i want a cheap one\n'
msg += 'Alternatively, use a different topic tracker.'
exit(msg)
else:
logger.info('Switch("DOMAINTAG") not detected - staying with previous domain')
return self.current_tracking_result
class KeywordSpottingTopicTracker(TopicTrackerInterface):
""" Just a hacky topic tracker to develop voicehub with.
:: Assumptions/Notes
-- To resolve resturants and hotels will also have to spot location
-- Assume we will stick with last domain unless we detect one of our keywords
"""
def __init__(self):
self.current_tracking_result = None
self.keywords = dict.fromkeys(OntologyUtils.available_domains, None)
#lmr46: added some keywords or lexical units ('food')
#consider to have a Lexicon that groups words per concepts, there are available lexica for English
#lmr46: Adapting only the domains available in the config file
domains = Settings.config.get("GENERAL",'domains') # a Hub has checked this exists
possible_domains = domains.split(',')
for dom in possible_domains:
kwds=[]
if dom=="CamRestaurants":
kwds=["cambridge","restaurant",'food','eat']
elif dom=="CamHotels":
kwds=["cambridge","hotel", "guest house", "guesthouse"]
elif dom=="SFRestaurants":
kwds=["san francisco","restaurant", "food","place to eat"]
elif dom=="SFHotels":
kwds=["san francisco","hotel", "guest house", "guesthouse", "hostel", "motel", "place to stay"]
elif dom=="wikipedia":
kwds=["wiki"]
self.keywords[dom]=kwds
# self.keywords["CamRestaurants"] = ["cambridge","restaurant",'food']
# self.keywords["CamHotels"] = ["cambridge","hotel", "guest house", "guesthouse"]
# self.keywords["SFRestaurants"] = ["san francisco","restaurant", "food","book"] # ASR cant recognise much at present -- will develop
# # system using CamRestaurants and CamHotels
# self.keywords["SFHotels"] = ["san francisco","hotel", "guest house", "guesthouse", "hostel", "motel", "book"]
# self.keywords["wikipedia"] = ["wiki"] # this could be used like "OK Google" or "Alexa"
def restart(self):
self.current_tracking_result = None
def infer_domain(self,userActHyps=None):
"""
-- Assumptions: Only working with the top hypothesis from ASR
-- Stick to last domain if nothing spotted in this turn
-- ORDER IS IMPORTANT -- ie it will hand off to FIRST domain a keyword is spotted in
"""
# TODO - could require all keywords to be present - e.g to disambiguate cam hotels from SFHotels
#lmr46: allowing overlapping keywords between domains
#su259: making current_tracking_result a local variable. method returns none if no new domain has been identified.
current_tracking_result = None
overlappindomains=[]
for dstring in list(self.keywords.keys()):
if self._is_a_keyword_in_sentence(self.keywords[dstring],userActHyps[0][0]):
logger.info(dstring + " keyword found in: " + userActHyps[0][0])
if "i(=" in userActHyps[0][0] or "inform(query=" in userActHyps[0][0]:
current_tracking_result = "wikipedia" # this is just a hack so i can wiki things like hotels!
else:
overlappindomains.append(dstring)
current_tracking_result = dstring
#break #TODO: Not handling overlapping of keywords between domains - it has to disambiguate!!!
if len(overlappindomains) > 1:
current_tracking_result = None
return current_tracking_result
def _is_a_keyword_in_sentence(self,keywords, sentence):
"""Note keywords just use the first spotted one ... this needs to be a little more sophisticated to resolve
SF hotel versus Cambridge hotel
"""
#TODO - will need changing if/when ASR is good enough to decode LOCATIONS - so that a match will require e.g
# "CAMBRIDGE" + "RESTAURANT" to count for TT domain
if keywords is not None:
for keyword in keywords:
if keyword in sentence.lower():
return True
return False
#END OF FILE