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.

235 lines
8.4 KiB
Plaintext

.. Copyright (C) 2001-2019 NLTK Project
.. For license information, see LICENSE.TXT
=======
Chat-80
=======
Chat-80 was a natural language system which allowed the user to
interrogate a Prolog knowledge base in the domain of world
geography. It was developed in the early '80s by Warren and Pereira; see
`<http://acl.ldc.upenn.edu/J/J82/J82-3002.pdf>`_ for a description and
`<http://www.cis.upenn.edu/~pereira/oldies.html>`_ for the source
files.
The ``chat80`` module contains functions to extract data from the Chat-80
relation files ('the world database'), and convert then into a format
that can be incorporated in the FOL models of
``nltk.sem.evaluate``. The code assumes that the Prolog
input files are available in the NLTK corpora directory.
The Chat-80 World Database consists of the following files::
world0.pl
rivers.pl
cities.pl
countries.pl
contain.pl
borders.pl
This module uses a slightly modified version of ``world0.pl``, in which
a set of Prolog rules have been omitted. The modified file is named
``world1.pl``. Currently, the file ``rivers.pl`` is not read in, since
it uses a list rather than a string in the second field.
Reading Chat-80 Files
=====================
Chat-80 relations are like tables in a relational database. The
relation acts as the name of the table; the first argument acts as the
'primary key'; and subsequent arguments are further fields in the
table. In general, the name of the table provides a label for a unary
predicate whose extension is all the primary keys. For example,
relations in ``cities.pl`` are of the following form::
'city(athens,greece,1368).'
Here, ``'athens'`` is the key, and will be mapped to a member of the
unary predicate *city*.
By analogy with NLTK corpora, ``chat80`` defines a number of 'items'
which correspond to the relations.
>>> from nltk.sem import chat80
>>> print(chat80.items) # doctest: +ELLIPSIS
('borders', 'circle_of_lat', 'circle_of_long', 'city', ...)
The fields in the table are mapped to binary predicates. The first
argument of the predicate is the primary key, while the second
argument is the data in the relevant field. Thus, in the above
example, the third field is mapped to the binary predicate
*population_of*, whose extension is a set of pairs such as
``'(athens, 1368)'``.
An exception to this general framework is required by the relations in
the files ``borders.pl`` and ``contains.pl``. These contain facts of the
following form::
'borders(albania,greece).'
'contains0(africa,central_africa).'
We do not want to form a unary concept out the element in
the first field of these records, and we want the label of the binary
relation just to be ``'border'``/``'contain'`` respectively.
In order to drive the extraction process, we use 'relation metadata bundles'
which are Python dictionaries such as the following::
city = {'label': 'city',
'closures': [],
'schema': ['city', 'country', 'population'],
'filename': 'cities.pl'}
According to this, the file ``city['filename']`` contains a list of
relational tuples (or more accurately, the corresponding strings in
Prolog form) whose predicate symbol is ``city['label']`` and whose
relational schema is ``city['schema']``. The notion of a ``closure`` is
discussed in the next section.
Concepts
========
In order to encapsulate the results of the extraction, a class of
``Concept``\ s is introduced. A ``Concept`` object has a number of
attributes, in particular a ``prefLabel``, an arity and ``extension``.
>>> c1 = chat80.Concept('dog', arity=1, extension=set(['d1', 'd2']))
>>> print(c1)
Label = 'dog'
Arity = 1
Extension = ['d1', 'd2']
The ``extension`` attribute makes it easier to inspect the output of
the extraction.
>>> schema = ['city', 'country', 'population']
>>> concepts = chat80.clause2concepts('cities.pl', 'city', schema)
>>> concepts
[Concept('city'), Concept('country_of'), Concept('population_of')]
>>> for c in concepts: # doctest: +NORMALIZE_WHITESPACE
... print("%s:\n\t%s" % (c.prefLabel, c.extension[:4]))
city:
['athens', 'bangkok', 'barcelona', 'berlin']
country_of:
[('athens', 'greece'), ('bangkok', 'thailand'), ('barcelona', 'spain'), ('berlin', 'east_germany')]
population_of:
[('athens', '1368'), ('bangkok', '1178'), ('barcelona', '1280'), ('berlin', '3481')]
In addition, the ``extension`` can be further
processed: in the case of the ``'border'`` relation, we check that the
relation is **symmetric**, and in the case of the ``'contain'``
relation, we carry out the **transitive closure**. The closure
properties associated with a concept is indicated in the relation
metadata, as indicated earlier.
>>> borders = set([('a1', 'a2'), ('a2', 'a3')])
>>> c2 = chat80.Concept('borders', arity=2, extension=borders)
>>> print(c2)
Label = 'borders'
Arity = 2
Extension = [('a1', 'a2'), ('a2', 'a3')]
>>> c3 = chat80.Concept('borders', arity=2, closures=['symmetric'], extension=borders)
>>> c3.close()
>>> print(c3)
Label = 'borders'
Arity = 2
Extension = [('a1', 'a2'), ('a2', 'a1'), ('a2', 'a3'), ('a3', 'a2')]
The ``extension`` of a ``Concept`` object is then incorporated into a
``Valuation`` object.
Persistence
===========
The functions ``val_dump`` and ``val_load`` are provided to allow a
valuation to be stored in a persistent database and re-loaded, rather
than having to be re-computed each time.
Individuals and Lexical Items
=============================
As well as deriving relations from the Chat-80 data, we also create a
set of individual constants, one for each entity in the domain. The
individual constants are string-identical to the entities. For
example, given a data item such as ``'zloty'``, we add to the valuation
a pair ``('zloty', 'zloty')``. In order to parse English sentences that
refer to these entities, we also create a lexical item such as the
following for each individual constant::
PropN[num=sg, sem=<\P.(P zloty)>] -> 'Zloty'
The set of rules is written to the file ``chat_pnames.fcfg`` in the
current directory.
SQL Query
=========
The ``city`` relation is also available in RDB form and can be queried
using SQL statements.
>>> import nltk
>>> q = "SELECT City, Population FROM city_table WHERE Country = 'china' and Population > 1000"
>>> for answer in chat80.sql_query('corpora/city_database/city.db', q):
... print("%-10s %4s" % answer)
canton 1496
chungking 1100
mukden 1551
peking 2031
shanghai 5407
tientsin 1795
The (deliberately naive) grammar ``sql.fcfg`` translates from English
to SQL:
>>> nltk.data.show_cfg('grammars/book_grammars/sql0.fcfg')
% start S
S[SEM=(?np + WHERE + ?vp)] -> NP[SEM=?np] VP[SEM=?vp]
VP[SEM=(?v + ?pp)] -> IV[SEM=?v] PP[SEM=?pp]
VP[SEM=(?v + ?ap)] -> IV[SEM=?v] AP[SEM=?ap]
NP[SEM=(?det + ?n)] -> Det[SEM=?det] N[SEM=?n]
PP[SEM=(?p + ?np)] -> P[SEM=?p] NP[SEM=?np]
AP[SEM=?pp] -> A[SEM=?a] PP[SEM=?pp]
NP[SEM='Country="greece"'] -> 'Greece'
NP[SEM='Country="china"'] -> 'China'
Det[SEM='SELECT'] -> 'Which' | 'What'
N[SEM='City FROM city_table'] -> 'cities'
IV[SEM=''] -> 'are'
A[SEM=''] -> 'located'
P[SEM=''] -> 'in'
Given this grammar, we can express, and then execute, queries in English.
>>> cp = nltk.parse.load_parser('grammars/book_grammars/sql0.fcfg')
>>> query = 'What cities are in China'
>>> for tree in cp.parse(query.split()):
... answer = tree.label()['SEM']
... q = " ".join(answer)
... print(q)
...
SELECT City FROM city_table WHERE Country="china"
>>> rows = chat80.sql_query('corpora/city_database/city.db', q)
>>> for r in rows: print("%s" % r, end=' ')
canton chungking dairen harbin kowloon mukden peking shanghai sian tientsin
Using Valuations
-----------------
In order to convert such an extension into a valuation, we use the
``make_valuation()`` method; setting ``read=True`` creates and returns
a new ``Valuation`` object which contains the results.
>>> val = chat80.make_valuation(concepts, read=True)
>>> 'calcutta' in val['city']
True
>>> [town for (town, country) in val['country_of'] if country == 'india']
['bombay', 'calcutta', 'delhi', 'hyderabad', 'madras']
>>> dom = val.domain
>>> g = nltk.sem.Assignment(dom)
>>> m = nltk.sem.Model(dom, val)
>>> m.evaluate(r'population_of(jakarta, 533)', g)
True