def main():
parser = StanfordServerParser()
while True:
try:
line = raw_input("Enter line: ")
tree = parser.parse(line)
tree.pretty_print()
except EOFError:
print "Bye!"
sys.exit(0)
def main():
parser = StanfordServerParser()
while True:
try:
line = input("Enter line: ")
tree = parser.parse(line)
tree.pretty_print()
except EOFError:
print("Bye!")
sys.exit(0)
class StanfordParserTest(object):
def __init__(self, host='localhost', port=9000, properties={}):
self.parser = StanfordServerParser(host, port, properties)
def query(self, sentence):
tree = self.parser.parse(sentence)
return tree
class QueryHandler(object):
"""
Processing natural language query and return answer.
"""
def __init__(self,
data_source='online',
host='localhost',
port=9000,
properties={}):
"""
Initialize query handler.
Args:
data_source (str): using 'online' zhishi.me API or 'local' sparql endpoint.
host (str): host of Stanford CoreNLP service
port (int): port of Stanford CoreNLP service
properties (dict): properties for Stanfoord CoreNLP service
"""
# Initialize Stanford CoreNLP Parser
self.parser = StanfordServerParser(host, port, properties)
# Define rules.
# query single entity
self.entity_rules = OrderedDict([
# 命名实体，如 周杰伦，微软
('( FRAG ( NR:subject-r ) )', {}),
# 普通名词，如 水果
('( NP ( NN:subject-r ) )', {}),
# 谁是周杰伦，什么是桃子
('( IP ( NP ( PN ) ) ( VP ( VC ) ( NP ( NN/NR:subject-r ) ) ) )',
{}),
# 周杰伦是谁，桃子是什么
('( IP ( NP ( NN/NR:subject-r ) ) ( VP ( VC ) ( NP ( PN ) ) ) )',
{}),
])
# query entity property
self.entity_property_rules = OrderedDict([
# 姚明身高
('( NP ( NP ( NP/NR:subject-o ) ) ( NP ( NN:property-r ) ) )', {}),
# 姚明的身高
('( NP ( DNP ( NP ( NN/NR:subject-r ) ) ( DEG ) ) ( NP ( NN:property-r ) ) )',
{}),
# 珠穆朗玛峰的海拔是多少
('( IP ( NP ( DNP ( NP ( NR:subject-r ) ) ( DEG ) ) ( NP ( NN:property-r ) ) ) ( VP ) )',
{}),
# 珠穆朗玛峰海拔是多少
('( IP ( NP ( NP/NR:s_type ) ( NN/NP:p_type ) ) ( VP ( VC ) ( QP/NP:q_type ) ) )',
{
's_type':
OrderedDict([
('( NR:subject-r )', {}),
('( NP ( NR:subject-r ) )', {}),
]),
'p_type':
OrderedDict([
('( NN:property-r )', {}),
('( NP ( NN:property-r ) )', {}),
]),
'q_type':
OrderedDict([
('( QP ( CD ) )', {}),
('( NP ( PN ) )', {}),
])
}),
])
# Initialize Knowledge Graph query client
self.data_source = data_source
if self.data_source == 'online':
self.sparql = SparqlOnline()
else:
self.local_sparql = SparqlLocal()
# Initialize logger
self.logger = logging.getLogger(self.__class__.__module__)
self.logger.setLevel(logging.DEBUG)
ch = logging.StreamHandler()
ch.setLevel(logging.DEBUG)
self.logger.addHandler(ch)
# Initialize synonyms handler
self.property_synonyms = PropertySynonyms()
def _entity_query(self, entity_name):
"""
Handle entity query.
Args:
entity_name (str): entity name
Returns:
text (str): abstract of entity
"""
if self.data_source == 'local':
return self.sparql.get_abstract(entity_name)
else:
baidu_result = self.sparql.get_abstract(entity_name,
baike='baidubaike')
if not baidu_result:
return self.sparql.get_abstract(entity_name, baike='zhwiki')
return baidu_result
def _eneity_property_query(self, entity_name, property_name):
"""
Handle entity property query.
Args:
entity_name (str)
property_name (str)
Returns:
text (str): entity property value
"""
if self.data_source == 'local':
return self.sparql.get_property(entity_name, property_name)
else:
corrected_property = self.property_synonyms.get_synonyms(
property_name)
self.logger.debug('Corrected property:\n%s' % corrected_property)
baidu_result = self.sparql.get_property(entity_name,
property_name,
baike='baidubaike')
if not baidu_result:
return self.sparql.get_property(entity_name,
property_name,
baike='zhwiki')
return baidu_result
def query(self, sentence):
"""
Answers a query
Args:
sentence (str): query sentence
Returns:
ans(str): answer text
"""
tree = self.parser.parse(sentence)
self.logger.debug('Dependence parse tree: \n%s' % tree)
info = match_rules(tree, self.entity_rules)
if info:
# entity query
self.logger.debug('Entity match:\n%s' % info)
ans = self._entity_query(info['subject'])
return ans
info = match_rules(tree, self.entity_property_rules)
if info:
# entity proprety query
self.logger.debug('Entity property match:\n%s' % info)
ans = self._eneity_property_query(info['subject'],
info['property'])
return ans
return 'rule not match'
def __init__(self,
data_source='online',
host='localhost',
port=9000,
properties={}):
"""
Initialize query handler.
Args:
data_source (str): using 'online' zhishi.me API or 'local' sparql endpoint.
host (str): host of Stanford CoreNLP service
port (int): port of Stanford CoreNLP service
properties (dict): properties for Stanfoord CoreNLP service
"""
# Initialize Stanford CoreNLP Parser
self.parser = StanfordServerParser(host, port, properties)
# Define rules.
# query single entity
self.entity_rules = OrderedDict([
# 命名实体，如 周杰伦，微软
('( FRAG ( NR:subject-r ) )', {}),
# 普通名词，如 水果
('( NP ( NN:subject-r ) )', {}),
# 谁是周杰伦，什么是桃子
('( IP ( NP ( PN ) ) ( VP ( VC ) ( NP ( NN/NR:subject-r ) ) ) )',
{}),
# 周杰伦是谁，桃子是什么
('( IP ( NP ( NN/NR:subject-r ) ) ( VP ( VC ) ( NP ( PN ) ) ) )',
{}),
])
# query entity property
self.entity_property_rules = OrderedDict([
# 姚明身高
('( NP ( NP ( NP/NR:subject-o ) ) ( NP ( NN:property-r ) ) )', {}),
# 姚明的身高
('( NP ( DNP ( NP ( NN/NR:subject-r ) ) ( DEG ) ) ( NP ( NN:property-r ) ) )',
{}),
# 珠穆朗玛峰的海拔是多少
('( IP ( NP ( DNP ( NP ( NR:subject-r ) ) ( DEG ) ) ( NP ( NN:property-r ) ) ) ( VP ) )',
{}),
# 珠穆朗玛峰海拔是多少
('( IP ( NP ( NP/NR:s_type ) ( NN/NP:p_type ) ) ( VP ( VC ) ( QP/NP:q_type ) ) )',
{
's_type':
OrderedDict([
('( NR:subject-r )', {}),
('( NP ( NR:subject-r ) )', {}),
]),
'p_type':
OrderedDict([
('( NN:property-r )', {}),
('( NP ( NN:property-r ) )', {}),
]),
'q_type':
OrderedDict([
('( QP ( CD ) )', {}),
('( NP ( PN ) )', {}),
])
}),
])
# Initialize Knowledge Graph query client
self.data_source = data_source
if self.data_source == 'online':
self.sparql = SparqlOnline()
else:
self.local_sparql = SparqlLocal()
# Initialize logger
self.logger = logging.getLogger(self.__class__.__module__)
self.logger.setLevel(logging.DEBUG)
ch = logging.StreamHandler()
ch.setLevel(logging.DEBUG)
self.logger.addHandler(ch)
# Initialize synonyms handler
self.property_synonyms = PropertySynonyms()
from collections import OrderedDict
import os
from lango.parser import StanfordServerParser
from lango.matcher import match_rules
parser = StanfordServerParser()
sents = ['What religion is the President of the United States?']
rules = {
'( SBARQ ( WHNP/WHADVP:wh_t ) ( SQ ( VBZ ) ( NP:np_t ) ) )': {
'np_t': {
'( NP ( NP:subj-o ) ( PP ( IN:subj_in-o ) ( NP:obj-o ) ) )': {},
'( NP:subj-o )': {},
},
'wh_t': {
'( WHNP:whnp ( WDT ) ( NN:prop-o ) )': {},
'( WHNP/WHADVP:qtype-o )': {},
}
},
'( SBARQ:subj-o )': {},
}
keys = ['subj', 'subj_in', 'obj', 'prop', 'qtype']
for sent in sents:
tree = parser.parse(sent)
contexts = match_rules(tree, rules, multi=True)
for context in contexts:
print(", ".join(['%s:%s' % (k, context.get(k)) for k in keys]))
"""
def __init__(self, host='localhost', port=9000, properties={}):
LoggingInterface.__init__(self)
self.parser = StanfordServerParser(host, port, properties)
"""
Rule for matching a subject and/or property of NP
Matches:
- subject : Subject to get property of
- prop (optional): Propert to get of subject
Examples:
- Obama born
- Obama
- Obama's birthday
- Barack Obama's wife
"""
self.subj_rules = OrderedDict([
# When was (Obama born)
('( NP ( NP:subject-o ) ( VP:prop-o ) )', {}),
# What is (the birth day of Obama)
('( NP ( NP:prop-o ) ( PP ( IN ) ( NP:subject-o ) ) )', {}),
# What is (Obama's birthday)
('( NP ( NP:subject-o ( NNP ) ( POS ) ) ( NN/NNS:prop-o ) $ )', {}
),
# What is (Obama's birth day)
('( NP ( NP:subject-o ( NNP ) ( POS ) ) ( NN/JJ:prop-o ) ( NN/NNS:prop2-o ) )',
{}),
# What is (Barrack Obama's birthday)
('( NP ( NP:subject-o ( NNP ) ( NNP ) ( POS ) ) ( NN/NNS:prop-o ) $ )',
{}),
# What is (Barack Obama's birth day)
('( NP ( NP:subject-o ( NNP ) ( NNP ) ( POS ) ) ( NN/JJ:prop-o ) ( NN/NNS:prop2-o ) )',
{}),
# What was (the Soviet Union’s motto)
('( NP ( NP:subject-o ( DT ) ( NNP ) ( NNP ) ( POS ) ) ( NN:prop-o ) )',
{}),
('( NP:subject-o )', {}),
])
"""
Rule for matching subject property query
Matches:
- qtype : Question type (who, where, what, when)
- subject : Subject to get property of
- prop (optional): Property to get of subject
- prop2 (optional): Second part of property
- prop3 (optional): Overwrite property
- jj (optional): Adjective that will be property (e.g. many/tall/high)
Examples:
- What religion is Obama?
- Who did Obama marry?
- Who is Obama?
- Who is Barack Obama's wife?
- How tall is Mt. Everest?
"""
self.subject_prop_rules = {
'( SBARQ ( WHNP/WHADVP/WHADJP:qtype_t ) ( SQ:sq_t ) )': {
'qtype_t':
OrderedDict([
# What religion
('( WHNP ( WDT:qtype-o=what ) ( NN:prop3-o ) )', {}),
# What genre
('( WHNP ( WDT:qtype-o=what ) ( NP ( NN:prop3-o ) ) )', {}
),
# How many/tall
('( WHADJP ( WRB:qtype-o ) ( JJ:jj-o ) )', {}),
# which country
('( WHNP ( WDT:qtype-o=which ) ( NN ) )', {}),
# What/where/who
('( WHNP/WHADVP:qtype-o )', {}),
]),
'sq_t': {
# What ethnicity is Obama
'( SQ ( VP ( ADVP:prop-o ) ) ( VBZ ) ( VP:suject-o ) )':
{},
# Who did Obama marry
'( SQ ( VBD:action-o ) ( NP:subj_t ) ( VP:prop-o ) )': {
'subj_t': self.subj_rules
},
# Who did
'( SQ ( VP ( VBZ/VBD/VBP:action-o ) ( NP:subj_t ) ) )': {
'subj_t': self.subj_rules
},
# Who is Edward Thatch known as
'( SQ ( VBZ:action-o ) ( NP:subj_t ) ( VP:prop-o ) )': {
'subj_t': self.subj_rules,
},
# Which country is Deshauna Barber a citizen of?
'( SQ ( VBZ:action-o ) ( NP:subject-o ( NNP ) ( NNP ) ) ( NP ( NP ( DT ) ( NN:prop-o ) ) ( PP ( IN ) ) ) )':
{},
# What is Obama
'( SQ ( VBZ/VBD/VBP:action-o ) ( NP:subj_t ) )': {
'subj_t': self.subj_rules
}
}
}
}
"""
Rule for getting property of NP or VP
Matches:
prop : Property of instance to match
op : Operation to match property
value: Value of property
Examples:
- born in 1950
- have population over 100,000
"""
self.prop_rules = OrderedDict([
#
('( SQ/VP ( VB/VBP/VBD ) ( VP ( VBN:prop-o ) ( PP ( IN:op-o ) ( NP:value-o ) ) ) )',
{}),
# are in Asia
('( SQ/VP ( VB/VBP/VBD=are ) ( PP ( IN:op-o ) ( NP:value-o ) ) )',
{}),
# died from laryngitis
('( SQ/VP ( VB/VBP/VBD:prop-o ) ( PP ( IN:op-o ) ( NP:value-o ) ) )',
{}),
# have population over 1000000
('( SQ/VP ( VB/VBP/VBD ) ( NP ( NP:prop-o ) ( PP ( IN:op-o ) ( NP/CD/JJ:value-o ) ) ) )',
{}),
('( SQ/VP ( VB/VBP/VBD ) ( NP:prop-o ) ( NP ( QP ( JJR:op-o ) ( IN ) ( CD:value-o ) ) ) )',
{}),
('( SQ/VP ( VB/VBP/VBD ) ( NP ( QP ( JJR:op-o ) ( IN=than ) ( NP/CD/JJ:value-o ) ) ( NNS:value_units-o ) ) )',
{}),
('( PP ( IN:op-o ) ( NP ( NP:value-o ) ( PP:pp_t ) ) )', {}),
('( PP ( IN:op-o ) ( NP:value-o ) )', {}),
])
"""
Rules for finding entity queries
Matches:
qtype : question type (how many, which)
inst : instance of entity to match
prop_match_t (optional): Parse tree for first property match
prop_match2_t (optional): Parse tree for second property match
Examples:
- How many POTUS are there?
- Which POTUS are born in 1950?
- How many books are written by George Orwell?
- How many countries are in Asia and have population over 100,000?
"""
self.find_entity_rules = OrderedDict([
('( SBARQ ( WHNP ( WHNP ( WHADJP:qtype-o ) ( NNS:inst-O ) ) ( PP:prop_match_t ) ) )',
{}),
('( SBARQ ( WHNP:qtype-o=who ) ( SQ:sq_t ) )', {
'sq_t': {
'( SQ ( VBD/VBZ ) ( NP ( NP:inst-O ) ( PP:prop_match_t ) ) )':
{}
},
}),
(
'( SBARQ ( WHNP ( WHADJP/WDT/WHNP:qtype-o ) ( NNS/NN/NP:inst-O ) ) ( SQ:sq_t ) )',
{
'sq_t':
OrderedDict([
# are there
('( SQ ( VBP ) ( NP ( EX=there ) ) )', {}),
# are in Asia and have population over 100,000
('( SQ ( VP ( VP:prop_match_t ) ( CC ) ( VP:prop_match2_t ) ) )',
{}),
('( SQ ( VP:prop_match_t ) )', {}),
('( SQ:prop_match_t )', {}),
])
}),
])
self.wd = WikiData()
class NLQueryEngine(LoggingInterface):
"""
Grammar mapping for knowledge queries of the form:
- What is the X of Y
- What is X's Y
"""
def __init__(self, host='localhost', port=9000, properties={}):
LoggingInterface.__init__(self)
self.parser = StanfordServerParser(host, port, properties)
"""
Rule for matching a subject and/or property of NP
Matches:
- subject : Subject to get property of
- prop (optional): Propert to get of subject
Examples:
- Obama born
- Obama
- Obama's birthday
- Barack Obama's wife
"""
self.subj_rules = OrderedDict([
# When was (Obama born)
('( NP ( NP:subject-o ) ( VP:prop-o ) )', {}),
# What is (the birth day of Obama)
('( NP ( NP:prop-o ) ( PP ( IN ) ( NP:subject-o ) ) )', {}),
# What is (Obama's birthday)
('( NP ( NP:subject-o ( NNP ) ( POS ) ) ( NN/NNS:prop-o ) $ )', {}
),
# What is (Obama's birth day)
('( NP ( NP:subject-o ( NNP ) ( POS ) ) ( NN/JJ:prop-o ) ( NN/NNS:prop2-o ) )',
{}),
# What is (Barrack Obama's birthday)
('( NP ( NP:subject-o ( NNP ) ( NNP ) ( POS ) ) ( NN/NNS:prop-o ) $ )',
{}),
# What is (Barack Obama's birth day)
('( NP ( NP:subject-o ( NNP ) ( NNP ) ( POS ) ) ( NN/JJ:prop-o ) ( NN/NNS:prop2-o ) )',
{}),
# What was (the Soviet Union’s motto)
('( NP ( NP:subject-o ( DT ) ( NNP ) ( NNP ) ( POS ) ) ( NN:prop-o ) )',
{}),
('( NP:subject-o )', {}),
])
"""
Rule for matching subject property query
Matches:
- qtype : Question type (who, where, what, when)
- subject : Subject to get property of
- prop (optional): Property to get of subject
- prop2 (optional): Second part of property
- prop3 (optional): Overwrite property
- jj (optional): Adjective that will be property (e.g. many/tall/high)
Examples:
- What religion is Obama?
- Who did Obama marry?
- Who is Obama?
- Who is Barack Obama's wife?
- How tall is Mt. Everest?
"""
self.subject_prop_rules = {
'( SBARQ ( WHNP/WHADVP/WHADJP:qtype_t ) ( SQ:sq_t ) )': {
'qtype_t':
OrderedDict([
# What religion
('( WHNP ( WDT:qtype-o=what ) ( NN:prop3-o ) )', {}),
# What genre
('( WHNP ( WDT:qtype-o=what ) ( NP ( NN:prop3-o ) ) )', {}
),
# How many/tall
('( WHADJP ( WRB:qtype-o ) ( JJ:jj-o ) )', {}),
# which country
('( WHNP ( WDT:qtype-o=which ) ( NN ) )', {}),
# What/where/who
('( WHNP/WHADVP:qtype-o )', {}),
]),
'sq_t': {
# What ethnicity is Obama
'( SQ ( VP ( ADVP:prop-o ) ) ( VBZ ) ( VP:suject-o ) )':
{},
# Who did Obama marry
'( SQ ( VBD:action-o ) ( NP:subj_t ) ( VP:prop-o ) )': {
'subj_t': self.subj_rules
},
# Who did
'( SQ ( VP ( VBZ/VBD/VBP:action-o ) ( NP:subj_t ) ) )': {
'subj_t': self.subj_rules
},
# Who is Edward Thatch known as
'( SQ ( VBZ:action-o ) ( NP:subj_t ) ( VP:prop-o ) )': {
'subj_t': self.subj_rules,
},
# Which country is Deshauna Barber a citizen of?
'( SQ ( VBZ:action-o ) ( NP:subject-o ( NNP ) ( NNP ) ) ( NP ( NP ( DT ) ( NN:prop-o ) ) ( PP ( IN ) ) ) )':
{},
# What is Obama
'( SQ ( VBZ/VBD/VBP:action-o ) ( NP:subj_t ) )': {
'subj_t': self.subj_rules
}
}
}
}
"""
Rule for getting property of NP or VP
Matches:
prop : Property of instance to match
op : Operation to match property
value: Value of property
Examples:
- born in 1950
- have population over 100,000
"""
self.prop_rules = OrderedDict([
#
('( SQ/VP ( VB/VBP/VBD ) ( VP ( VBN:prop-o ) ( PP ( IN:op-o ) ( NP:value-o ) ) ) )',
{}),
# are in Asia
('( SQ/VP ( VB/VBP/VBD=are ) ( PP ( IN:op-o ) ( NP:value-o ) ) )',
{}),
# died from laryngitis
('( SQ/VP ( VB/VBP/VBD:prop-o ) ( PP ( IN:op-o ) ( NP:value-o ) ) )',
{}),
# have population over 1000000
('( SQ/VP ( VB/VBP/VBD ) ( NP ( NP:prop-o ) ( PP ( IN:op-o ) ( NP/CD/JJ:value-o ) ) ) )',
{}),
('( SQ/VP ( VB/VBP/VBD ) ( NP:prop-o ) ( NP ( QP ( JJR:op-o ) ( IN ) ( CD:value-o ) ) ) )',
{}),
('( SQ/VP ( VB/VBP/VBD ) ( NP ( QP ( JJR:op-o ) ( IN=than ) ( NP/CD/JJ:value-o ) ) ( NNS:value_units-o ) ) )',
{}),
('( PP ( IN:op-o ) ( NP ( NP:value-o ) ( PP:pp_t ) ) )', {}),
('( PP ( IN:op-o ) ( NP:value-o ) )', {}),
])
"""
Rules for finding entity queries
Matches:
qtype : question type (how many, which)
inst : instance of entity to match
prop_match_t (optional): Parse tree for first property match
prop_match2_t (optional): Parse tree for second property match
Examples:
- How many POTUS are there?
- Which POTUS are born in 1950?
- How many books are written by George Orwell?
- How many countries are in Asia and have population over 100,000?
"""
self.find_entity_rules = OrderedDict([
('( SBARQ ( WHNP ( WHNP ( WHADJP:qtype-o ) ( NNS:inst-O ) ) ( PP:prop_match_t ) ) )',
{}),
('( SBARQ ( WHNP:qtype-o=who ) ( SQ:sq_t ) )', {
'sq_t': {
'( SQ ( VBD/VBZ ) ( NP ( NP:inst-O ) ( PP:prop_match_t ) ) )':
{}
},
}),
(
'( SBARQ ( WHNP ( WHADJP/WDT/WHNP:qtype-o ) ( NNS/NN/NP:inst-O ) ) ( SQ:sq_t ) )',
{
'sq_t':
OrderedDict([
# are there
('( SQ ( VBP ) ( NP ( EX=there ) ) )', {}),
# are in Asia and have population over 100,000
('( SQ ( VP ( VP:prop_match_t ) ( CC ) ( VP:prop_match2_t ) ) )',
{}),
('( SQ ( VP:prop_match_t ) )', {}),
('( SQ:prop_match_t )', {}),
])
}),
])
self.wd = WikiData()
def subject_query(self,
qtype,
subject,
action,
jj=None,
prop=None,
prop2=None,
prop3=None):
"""Transforms matched context into query parameters and performs query
Args:
qtype: Matched type of query (what, who, where, etc.)
subject: Matched subject (Obama)
action: Matched verb action (is, was, ran)
jj (optional): Matched adverb
prop (optional): Matched prop
prop2 (optional): Matched prop
prop3 (optional): Matched prop
Returns:
Answer: Answer from query, or empty Answer if None
"""
print()
print(qtype)
print(subject)
print(action)
print(jj)
print(prop)
print(prop2)
print(prop3)
print()
if jj == 'old':
# How old is Obama?
prop = 'age'
if jj in ['tall', 'high']:
# How tall is Yao Ming / Eifel tower?
prop = 'height'
if prop2:
prop = prop + ' ' + prop2
if prop3 and not prop:
prop = prop3
if not prop:
if action not in ['is', 'was']:
prop = action
if qtype == 'who' and prop.endswith('ed'):
prop += ' by'
if qtype == 'where' and prop == 'founded':
prop = 'location of formation'
if qtype == 'where' and action == 'is':
prop = 'located'
if qtype == 'how' and prop == 'died':
prop = 'manner of death'
if qtype == 'where' and prop == 'died':
prop = 'place of death'
if prop in ['birth day', 'birthday']:
prop = 'date of birth'
ans = self.get_property(qtype, subject, prop)
if not ans:
ans = Answer()
ans.params = {
'qtype': qtype,
'subject': subject,
'prop': prop,
}
if (prop == "population"):
ans.data = ans.data[-1]
print(ans.params)
return ans
def get_prop_tuple(self,
prop=None,
value=None,
op=None,
value_units=None,
pp_t=None):
"""Returns a property tuple (prop, value, op). E.g. (population, 1000000, >)
Args:
prop (str): Property to search for (e.g. population)
value (str): Value property should equal (e.g. 10000000)
op (str): Operator for value of property (e.g. >)
Returns:
tuple: Property tuple, e.g: (population, 10000000, >)
"""
self.info('Prop tuple: {0},{1},{2},{3},{4}', prop, value, op,
value_units, pp_t)
if op in ['in', 'by', 'of', 'from']:
oper = op
elif op in ['over', 'above', 'more', 'greater']:
oper = '>'
elif op in ['under', 'below', 'less']:
oper = '<'
else:
self.error('NO OP {0}', op)
return None
# Infer property to match value
if prop is None:
if value_units is not None:
if value_units in ['people']:
prop = 'population'
if not prop:
return None
props = [(prop, value, oper)]
if pp_t:
prop_tuple = match_rules(pp_t, self.prop_rules,
self.get_prop_tuple)
if not prop_tuple:
return None
props += prop_tuple
return props
def find_entity_query(self,
qtype,
inst,
prop_match_t=None,
prop_match2_t=None):
"""Transforms matched context into query parameters and performs query for
queries to find entities
Args:
qtype (str): Matched type of query (what, who, where, etc.)
inst (str): Matched instance of entity to match (Obama)
action (str): Matched verb action (is, was, ran)
prop_match_t (Tree): Matched property Tree
prop_match2_t (Tree): Matched property Tree
Returns:
Answer: Answer from query, or empty Answer if None
"""
props = []
if prop_match_t:
prop = match_rules(prop_match_t, self.prop_rules,
self.get_prop_tuple)
if not prop:
return
props += prop
if prop_match2_t:
prop = match_rules(prop_match2_t, self.prop_rules,
self.get_prop_tuple)
if not prop:
return
props += prop
if not inst.isupper():
inst = singularize(inst)
ans = self.wd.find_entity(qtype, inst, props)
if not ans:
ans = Answer()
ans.params = {
'qtype': qtype,
'inst': inst,
'props': props,
}
return ans
def get_property(self, qtype, subject, prop):
"""Gets property of a subject
Example:
get_property('who', 'Obama', 'wife') = 'Michelle Obama'
Args:
subject: Subject to get property of
prop: Property to get of subject
Todo:
* Add other APIs here
Returns:
Answer: Answer from query
"""
return self.wd.get_property(qtype, subject, prop)
def preprocess(self, sent):
"""Preprocesses a query by adding punctuation"""
if sent[-1] != '?':
sent = sent + '?'
return sent
def query(self, sent, format_='plain'):
"""Answers a query
If format is plain, will return the answer as a string
If format is raw, will return the raw context of query
Args:
sent: Query sentence
format_: Format of answer to return (Default to plain)
Returns:
dict: Answer context
str: Answer as a string
Raises:
ValueError: If format_ is incorrect
"""
sent = self.preprocess(sent)
tree = self.parser.parse(sent)
context = {'query': sent, 'tree': tree}
self.info(tree)
ans = first([
match_rules(tree, self.find_entity_rules, self.find_entity_query),
match_rules(tree, self.subject_prop_rules, self.subject_query),
])
if not ans:
ans = Answer()
ans.query = sent
ans.tree = str(tree)
if format_ == 'raw':
return ans.to_dict()
elif format_ == 'plain':
return ans.to_plain()
else:
raise ValueError('Undefined format: %s' % format_)
def __init__(self, host='localhost', port=9000, properties={}):
self.parser = StanfordServerParser(host, port, properties)
class QueryParser:
"""Parse an English-language question.
``QueryParser`` uses part-of-speech tagging and pattern matching to
turn a natural-language question into a structured query.
Parameters
----------
port : int, optional
The port on which the Stanford CoreNLP Server listens. [default: 9501]
"""
def __init__(self, port=9501):
self.parser = StanfordServerParser(port=port)
self.rules = self.get_rules()
def get_rules(self):
rules_outcome = {
'( SBARQ ( WHNP ( WDT ) ( NP:np ) ) ( SQ ( VP ( VBZ:action-o ) ( PP:pp1 ) ( PP:pp2 ) ) ) )': {
'np': {
'( NP:relation-o )': {}
},
'pp1': {
'( PP ( TO=to ) ( NP:to_object-o ) )': {},
'( PP ( IN=from ) ( NP:from_object-o ) )': {}
},
'pp2': {
'( PP ( TO=to ) ( NP:to_object-o ) )': {},
'( PP ( IN=from ) ( NP:from_object-o ) )': {}
}
},
'( SBARQ ( WHNP ( WDT ) ( NP:np1 ) ) ( SQ ( VP ( VBZ:action-o ) ( NP:np2 ) ( PP:pp ) ) ) )': {
'np1': {
'( NP:relation-o )': {}
},
'np2': {
'( NP:from_object-o )': {}
},
'pp': {
'( PP ( TO=to ) ( NP:to_object-o ) )': {}
}
},
'( SBAR ( WHNP ( WDT ) ) ( S ( NP:np1 ) ( VP ( VBZ:action-o ) ( NP:np2 ) ( PP:pp ) ) ) )': {
'np1': {
'( NP:relation-o )': {}
},
'np2': {
'( NP:from_object-o )': {}
},
'pp': {
'( PP ( TO=to ) ( NP ( NN:to_object-o ) ( NN ) ) )': {}
}
},
'( SBARQ ( WHNP ( WP ) ) ( SQ ( VBZ:action-o ) ( NP ( NP:np ) ( PP:pp ) ) ) )': {
'np': {
'( NP:relation-o )': {}
},
'pp': {
'( PP ( IN=between ) ( NP ( NN:from_object-o ) ( CC=and ) ( NN:to_object-o ) ) )': {}
}
},
'( S ( VP ( VB:action-o ) ( NP ( NP:np ) ( PP:pp ) ) ) )': {
'np': {
'( NP:relation-o )': {}
},
'pp': {
'( PP ( IN=between ) ( NP ( NN:from_object-o ) ( CC=and ) ( NN:to_object-o ) ) )': {}
}
},
'( NP ( NP ( NN:action-o ) ) ( NP ( NP:np ) ( PP:pp ) ) )': {
'np': {
'( NP:relation-o )': {}
},
'pp': {
'( PP ( IN=between ) ( NP ( NN:from_object-o ) ( CC=and ) ( NN:to_object-o ) ) )': {}
}
},
'( SBARQ ( WHNP ( WP ) ) ( SQ ( VBZ:action-o ) ( NP ( NP:np ) ( PP:pp ) ) ) )': {
'np': {
'( NP:relation-o )': {}
},
'pp': {
'( PP ( IN=between ) ( NP:compound_object-o ) )': {}
}
}
}
rules_protects = {
'( SBARQ ( WHNP ( WDT ) ( NP:np ) ) ( SQ ( VP ( VBP:relation-o ) ( PP:pp ) ) ) )': {
'np': {
'( NP:from_object-o )': {}
},
'pp': {
'( PP ( IN=from ) ( NP:to_object-o ) )': {}
}
},
'( SBARQ ( WHNP ( WDT ) ( NP:np ) ) ( SQ ( VP ( VBZ:relation-o ) ( PP:pp ) ) ) )': {
'np': {
'( NP:from_object-o )': {}
},
'pp': {
'( PP ( IN=from ) ( NP:to_object-o ) )': {}
}
},
'( FRAG ( SBAR ( WHNP ( WDT ) ) ( S ( NP:np ) ( VP ( VBP:relation-o ) ( PP:pp ) ) ) ) )': {
'np': {
'( NP:from_object-o )': {}
},
'pp': {
'( PP ( IN=from ) ( NP:to_object-o ) )': {}
}
}
}
return({**rules_outcome, **rules_protects})
def process_matches(self, relation, from_object=None, to_object=None, compound_object=None):
if compound_object is not None:
(to_object, from_object) = compound_object.split(' and ')
return({'from':{'term':from_object}, 'to':{'term':to_object}, 'relation':{'term':relation}})
def parse(self, question):
"""Parse a natural-language question.
Parameters
----------
question : str
The input question.
Returns
-------
terms : dict
A dictionary of parsed terms.
"""
tree = self.parser.parse(question)
terms = match_rules(tree, self.rules, self.process_matches)
if terms is None:
return {}
mesh = MeshTools()
terms['from'].update({k:v for k,v in mesh.get_best_term_entity(terms['from']['term']).items() if k in ('entity', 'bound')})
terms['to'].update({k:v for k,v in mesh.get_best_term_entity(terms['to']['term']).items() if k in ('entity', 'bound')})
# if terms['from']['entity'] is None:
# tm = NCITTermMapper()
# terms['from']['entity'] = tm.get_entity(terms['from']['term'])
# terms['from']['bound'] = True
# if terms['to']['entity'] is None:
# tm = NCITTermMapper()
# terms['to']['entity'] = tm.get_entity(terms['to']['term'])
# terms['to']['bound'] = True
# if terms['relation']['term'] == 'clinical outcome pathway' and terms['to']['entity'] in ('GeneticCondition', 'Symptom'):
# terms['to']['entity'] = 'Disease'
## cheat for proof-of-concept
## if automatic entity parsing fails, get the entities from the original definitions in the question files
if terms['from']['entity'] is None:
db = 'data/reasoner_data.sqlite'
conn = sqlite3.connect(db)
c = conn.cursor()
query = 'SELECT entity FROM ncats_entity_map WHERE term = "%s"' % terms['from']['term']
result = next(iter(c.execute(query).fetchall()), [])
if len(result) > 0:
terms['from']['entity'] = result[0]
terms['from']['bound'] = True
if terms['to']['entity'] is None:
db = 'data/reasoner_data.sqlite'
conn = sqlite3.connect(db)
c = conn.cursor()
query = 'SELECT entity FROM ncats_entity_map WHERE term = "%s"' % terms['to']['term']
result = next(iter(c.execute(query).fetchall()), [])
if len(result) > 0:
terms['to']['entity'] = result[0]
terms['to']['bound'] = True
terms['from']['bound'] = bool(terms['from']['bound'])
terms['to']['bound'] = bool(terms['to']['bound'])
return terms
def __init__(self, port=9501):
self.parser = StanfordServerParser(port=port)
self.rules = self.get_rules()
from collections import OrderedDict
import os
from lango.parser import StanfordServerParser
from lango.matcher import match_rules
parser = StanfordServerParser()
sents = [
'Call me an Uber.',
'Get my mother some flowers.',
'Find me a pizza with extra cheese.',
'Give Sam\'s dog a biscuit from Petshop.'
]
"""
me.call({'item': u'uber'})
my.mother.get({'item': u'flowers'})
me.order({'item': u'pizza', u'with': u'extra cheese'})
sam.dog.give({'item': u'biscuit', u'from': u'petshop'})
"""
subj_obj_rules = {
'subj_t': OrderedDict([
# my brother / my mother
('( NP ( PRP$:subject-o=my ) ( NN:relation-o ) )', {}),
# Sam's dog
('( NP ( NP ( NNP:subject-o ) ( POS ) ) ( NN:relation-o ) )', {}),
# me