You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

125 lines
4.0 KiB
Python

5 years ago
# Natural Language Toolkit: Cooper storage for Quantifier Ambiguity
#
# Copyright (C) 2001-2020 NLTK Project
5 years ago
# Author: Ewan Klein <ewan@inf.ed.ac.uk>
# URL: <http://nltk.org/>
# For license information, see LICENSE.TXT
from nltk.sem.logic import LambdaExpression, ApplicationExpression, Variable
from nltk.parse import load_parser
from nltk.parse.featurechart import InstantiateVarsChart
class CooperStore(object):
"""
A container for handling quantifier ambiguity via Cooper storage.
"""
def __init__(self, featstruct):
"""
:param featstruct: The value of the ``sem`` node in a tree from
``parse_with_bindops()``
:type featstruct: FeatStruct (with features ``core`` and ``store``)
"""
self.featstruct = featstruct
self.readings = []
try:
self.core = featstruct["CORE"]
self.store = featstruct["STORE"]
5 years ago
except KeyError:
print("%s is not a Cooper storage structure" % featstruct)
def _permute(self, lst):
"""
:return: An iterator over the permutations of the input list
:type lst: list
:rtype: iter
"""
remove = lambda lst0, index: lst0[:index] + lst0[index + 1 :]
if lst:
for index, x in enumerate(lst):
for y in self._permute(remove(lst, index)):
yield (x,) + y
else:
yield ()
def s_retrieve(self, trace=False):
"""
Carry out S-Retrieval of binding operators in store. If hack=True,
serialize the bindop and core as strings and reparse. Ugh.
Each permutation of the store (i.e. list of binding operators) is
taken to be a possible scoping of quantifiers. We iterate through the
binding operators in each permutation, and successively apply them to
the current term, starting with the core semantic representation,
working from the inside out.
Binding operators are of the form::
bo(\P.all x.(man(x) -> P(x)),z1)
"""
for perm, store_perm in enumerate(self._permute(self.store)):
if trace:
print("Permutation %s" % (perm + 1))
term = self.core
for bindop in store_perm:
# we just want the arguments that are wrapped by the 'bo' predicate
quant, varex = tuple(bindop.args)
# use var to make an abstraction over the current term and then
# apply the quantifier to it
term = ApplicationExpression(
quant, LambdaExpression(varex.variable, term)
)
if trace:
print(" ", term)
term = term.simplify()
self.readings.append(term)
def parse_with_bindops(sentence, grammar=None, trace=0):
"""
Use a grammar with Binding Operators to parse a sentence.
"""
if not grammar:
grammar = "grammars/book_grammars/storage.fcfg"
5 years ago
parser = load_parser(grammar, trace=trace, chart_class=InstantiateVarsChart)
# Parse the sentence.
tokens = sentence.split()
return list(parser.parse(tokens))
def demo():
from nltk.sem import cooper_storage as cs
sentence = "every girl chases a dog"
# sentence = "a man gives a bone to every dog"
print()
print("Analyis of sentence '%s'" % sentence)
print("=" * 50)
trees = cs.parse_with_bindops(sentence, trace=0)
for tree in trees:
semrep = cs.CooperStore(tree.label()["SEM"])
5 years ago
print()
print("Binding operators:")
print("-" * 15)
for s in semrep.store:
print(s)
print()
print("Core:")
print("-" * 15)
print(semrep.core)
print()
print("S-Retrieval:")
print("-" * 15)
semrep.s_retrieve(trace=True)
print("Readings:")
print("-" * 15)
for i, reading in enumerate(semrep.readings):
print("%s: %s" % (i + 1, reading))
if __name__ == "__main__":
5 years ago
demo()