def demo():
"""
A demonstration showing how C{Grammar}s can be created and used.
"""
from nltk import cfg
# Create some nonterminals
S, NP, VP, PP = cfg.nonterminals('S, NP, VP, PP')
N, V, P, Det = cfg.nonterminals('N, V, P, Det')
VP_slash_NP = VP/NP
print 'Some nonterminals:', [S, NP, VP, PP, N, V, P, Det, VP/NP]
print ' S.symbol() =>', `S.symbol()`
print
print cfg.Production(S, [NP])
# Create some Grammar Productions
grammar = cfg.parse_grammar("""
S -> NP VP
PP -> P NP
NP -> Det N
NP -> NP PP
VP -> V NP
VP -> VP PP
Det -> 'a'
Det -> 'the'
N -> 'dog'
N -> 'cat'
V -> 'chased'
V -> 'sat'
P -> 'on'
P -> 'in'
""")
print 'A Grammar:', `grammar`
print ' grammar.start() =>', `grammar.start()`
print ' grammar.productions() =>',
# Use string.replace(...) is to line-wrap the output.
print `grammar.productions()`.replace(',', ',\n'+' '*25)
print
def demo():
"""
A demonstration showing how C{Grammar}s can be created and used.
"""
from nltk import cfg
# Create some nonterminals
S, NP, VP, PP = cfg.nonterminals('S, NP, VP, PP')
N, V, P, Det = cfg.nonterminals('N, V, P, Det')
VP_slash_NP = VP / NP
print 'Some nonterminals:', [S, NP, VP, PP, N, V, P, Det, VP / NP]
print ' S.symbol() =>', ` S.symbol() `
print
print cfg.Production(S, [NP])
# Create some Grammar Productions
grammar = cfg.parse_grammar("""
S -> NP VP
PP -> P NP
NP -> Det N
NP -> NP PP
VP -> V NP
VP -> VP PP
Det -> 'a'
Det -> 'the'
N -> 'dog'
N -> 'cat'
V -> 'chased'
V -> 'sat'
P -> 'on'
P -> 'in'
""")
print 'A Grammar:', ` grammar `
print ' grammar.start() =>', ` grammar.start() `
print ' grammar.productions() =>',
# Use string.replace(...) is to line-wrap the output.
print ` grammar.productions() `.replace(',', ',\n' + ' ' * 25)
print
def parse(self, p_string):
"""
Parses a string and stores the resulting hierarchy of "domains"
"hierarchies" and "tables"
For the sake of NLP I've parsed the string using the nltk
context free grammar library.
A query is a "sentence" and can either be a domain, hierarchy or a table.
A domain is simply a word.
A hierarchy is expressed as "domain/domain"
A table is exressed as "table(sentence, sentence, sentence)"
Internally the query is represented as a nltk.parse.tree
Process:
1. string is tokenized
2. develop a context free grammar
3. parse
4. convert to a tree representation
"""
self.nltktree = None
# Store the query string
self.string = p_string
# Tokenize the query string, allowing only strings, parentheses,
# forward slashes and commas.
re_all = r'table[(]|\,|[)]|[/]|\w+'
data_tokens = tokenize.regexp(self.string, re_all)
# Develop a context free grammar
# S = sentence, T = table, H = hierarchy, D = domain
O, T, H, D = cfg.nonterminals('O, T, H, D')
# Specify the grammar
productions = (
# A sentence can be either a table, hierarchy or domain
cfg.Production(O, [D]),
cfg.Production(O, [H]),
cfg.Production(O, [T]),
# A table must be the following sequence:
# "table(", sentence, comma, sentence, comma, sentence, ")"
cfg.Production(T, ['table(', O, ',', O, ',', O, ')']),
# A hierarchy must be the following sequence:
# domain, forward slash, domain
cfg.Production(H, [D, '/', D]),
# domain, forward slash, another operator
cfg.Production(H, [D, '/', O]))
# Add domains to the cfg productions
# A domain is a token that is entirely word chars
re_domain = compile(r'^\w+$')
# Try every token and add if it matches the above regular expression
for tok in data_tokens:
if re_domain.match(tok):
prod = cfg.Production(D, [tok]),
productions = productions + prod
# Make a grammar out of our productions
grammar = cfg.Grammar(O, productions)
rd_parser = parse.RecursiveDescentParser(grammar)
# Tokens need to be redefined.
# It disappears after first use, and I don't know why.
tokens = tokenize.regexp_tokenize(self.string, re_all)
toklist = list(tokens)
# Store the parsing.
# Only the first one, as the grammar should be completely nonambiguous.
try:
self.parseList = rd_parser.get_parse_list(toklist)[0]
except IndexError:
print "Could not parse query."
return
# Set the nltk.parse.tree tree for this query to the global sentence
string = str(self.parseList)
string2 = string.replace(":", "").replace("')'", "").replace(
"table(", "").replace("','", "").replace("'", "").replace("/", "")
self.nltktree = parse.tree.bracket_parse(string2)
# Store the resulting nltk.parse.tree tree
self.parseTree = QuerySentence(self.nltktree)
self.xml = self.parseTree.toXML()
def demo():
"""
A demonstration of the probabilistic parsers. The user is
prompted to select which demo to run, and how many parses should
be found; and then each parser is run on the same demo, and a
summary of the results are displayed.
"""
import sys, time
from nltk.tokenizer import WhitespaceTokenizer
# Define some nonterminals
S, VP, NP, PP = nonterminals("S, VP, NP, PP")
V, N, P, Name, Det = nonterminals("V, N, P, Name, Det")
# Define a PCFG
grammar_productions1 = [
PCFGProduction(NP, [Det, N], prob=0.5),
PCFGProduction(NP, [NP, PP], prob=0.25),
PCFGProduction(NP, ["John"], prob=0.1),
PCFGProduction(NP, ["I"], prob=0.15),
PCFGProduction(Det, ["the"], prob=0.8),
PCFGProduction(Det, ["my"], prob=0.2),
PCFGProduction(N, ["dog"], prob=0.5),
PCFGProduction(N, ["cookie"], prob=0.5),
PCFGProduction(VP, [VP, PP], prob=0.1),
PCFGProduction(VP, [V, NP], prob=0.7),
PCFGProduction(VP, [V], prob=0.2),
PCFGProduction(V, ["ate"], prob=0.35),
PCFGProduction(V, ["saw"], prob=0.65),
PCFGProduction(S, [NP, VP], prob=1.0),
PCFGProduction(PP, [P, NP], prob=1.0),
PCFGProduction(P, ["with"], prob=0.61),
PCFGProduction(P, ["under"], prob=0.39),
]
pcfg1 = PCFG(S, grammar_productions1)
# Define a second, more extensive, grammar.
lexicon = [
PCFGProduction(V, ["saw"], prob=0.21),
PCFGProduction(V, ["ate"], prob=0.51),
PCFGProduction(V, ["ran"], prob=0.28),
PCFGProduction(N, ["boy"], prob=0.11),
PCFGProduction(N, ["cookie"], prob=0.12),
PCFGProduction(N, ["table"], prob=0.13),
PCFGProduction(N, ["telescope"], prob=0.14),
PCFGProduction(N, ["hill"], prob=0.50),
PCFGProduction(Name, ["Jack"], prob=0.52),
PCFGProduction(Name, ["Bob"], prob=0.48),
PCFGProduction(P, ["with"], prob=0.61),
PCFGProduction(P, ["under"], prob=0.39),
PCFGProduction(Det, ["the"], prob=0.41),
PCFGProduction(Det, ["a"], prob=0.31),
PCFGProduction(Det, ["my"], prob=0.28),
]
grammar_productions2 = lexicon + [
PCFGProduction(S, [NP, VP], prob=1.00),
PCFGProduction(VP, [V, NP], prob=0.59),
PCFGProduction(VP, [V], prob=0.40),
PCFGProduction(VP, [VP, PP], prob=0.01),
PCFGProduction(NP, [Det, N], prob=0.41),
PCFGProduction(NP, [Name], prob=0.28),
PCFGProduction(NP, [NP, PP], prob=0.31),
PCFGProduction(PP, [P, NP], prob=1.00),
]
pcfg2 = PCFG(S, grammar_productions2)
# Define two demos. Each demo has a sentence and a grammar.
demos = [
("I saw John with my cookie", pcfg1),
("the boy saw Jack with Bob under the table with a telescope", pcfg2),
]
# Ask the user which demo they want to use.
print
for i in range(len(demos)):
print "%3s: %s" % (i + 1, demos[i][0])
print " %r" % demos[i][1]
print
print "Which demo (%d-%d)? " % (1, len(demos)),
try:
snum = int(sys.stdin.readline().strip()) - 1
s, pcfg = demos[snum]
except:
print "Bad sentence number"
return
# Tokenize the sentence.
token = Token(TEXT=s)
WhitespaceTokenizer(SUBTOKENS="WORDS").tokenize(token, add_locs=True)
# Define a list of parsers. We'll use all parsers.
parsers = [
ViterbiPCFGParser(pcfg, LEAF="TEXT", SUBTOKENS="WORDS"),
InsidePCFGParser(pcfg, LEAF="TEXT", SUBTOKENS="WORDS"),
RandomPCFGParser(pcfg, LEAF="TEXT", SUBTOKENS="WORDS"),
UnsortedPCFGParser(pcfg, LEAF="TEXT", SUBTOKENS="WORDS"),
LongestPCFGParser(pcfg, LEAF="TEXT", SUBTOKENS="WORDS"),
BeamPCFGParser(len(token["WORDS"]) + 1, pcfg, LEAF="TEXT", SUBTOKENS="WORDS"),
]
# Run the parsers on the tokenized sentence.
times = []
average_p = []
num_parses = []
all_parses = {}
for parser in parsers:
print "\ns: %s\nparser: %s\ngrammar: %s" % (S, parser, pcfg)
parser.trace(3)
t = time.time()
parses = parser.get_parse_list(token)
times.append(time.time() - t)
if parses:
p = reduce(lambda a, b: a + b.prob(), parses, 0) / len(parses)
else:
p = 0
average_p.append(p)
num_parses.append(len(parses))
for p in parses:
all_parses[p.freeze(lambda t: t.freeze())] = 1
# Print some summary statistics
print
print " Parser | Time (secs) # Parses Average P(parse)"
print "-------------------+------------------------------------------"
for i in range(len(parsers)):
print "%18s |%11.4f%11d%19.14f" % (parsers[i].__class__.__name__, times[i], num_parses[i], average_p[i])
parses = all_parses.keys()
if parses:
p = reduce(lambda a, b: a + b.prob(), parses, 0) / len(parses)
else:
p = 0
print "-------------------+------------------------------------------"
print "%18s |%11s%11d%19.14f" % ("(All Parses)", "n/a", len(parses), p)
# Ask the user if we should draw the parses.
print
print "Draw parses (y/n)? ",
if sys.stdin.readline().strip().lower().startswith("y"):
import nltk.draw.tree
print " please wait..."
nltk.draw.tree.draw_trees(*parses)
# Ask the user if we should print the parses.
print
print "Print parses (y/n)? ",
if sys.stdin.readline().strip().lower().startswith("y"):
for parse in parses:
print parse
def parse(self, p_string):
"""
Parses a string and stores the resulting hierarchy of "domains"
"hierarchies" and "tables"
For the sake of NLP I've parsed the string using the nltk
context free grammar library.
A query is a "sentence" and can either be a domain, hierarchy or a table.
A domain is simply a word.
A hierarchy is expressed as "domain/domain"
A table is exressed as "table(sentence, sentence, sentence)"
Internally the query is represented as a nltk.parse.tree
Process:
1. string is tokenized
2. develop a context free grammar
3. parse
4. convert to a tree representation
"""
self.nltktree = None
# Store the query string
self.string = p_string
# Tokenize the query string, allowing only strings, parentheses,
# forward slashes and commas.
re_all = r'table[(]|\,|[)]|[/]|\w+'
data_tokens = tokenize.regexp(self.string, re_all)
# Develop a context free grammar
# S = sentence, T = table, H = hierarchy, D = domain
O, T, H, D = cfg.nonterminals('O, T, H, D')
# Specify the grammar
productions = (
# A sentence can be either a table, hierarchy or domain
cfg.Production(O, [D]), cfg.Production(O, [H]), cfg.Production(O, [T]),
# A table must be the following sequence:
# "table(", sentence, comma, sentence, comma, sentence, ")"
cfg.Production(T, ['table(', O, ',', O, ',', O, ')']),
# A hierarchy must be the following sequence:
# domain, forward slash, domain
cfg.Production(H, [D, '/', D]),
# domain, forward slash, another operator
cfg.Production(H, [D, '/', O])
)
# Add domains to the cfg productions
# A domain is a token that is entirely word chars
re_domain = compile(r'^\w+$')
# Try every token and add if it matches the above regular expression
for tok in data_tokens:
if re_domain.match(tok):
prod = cfg.Production(D,[tok]),
productions = productions + prod
# Make a grammar out of our productions
grammar = cfg.Grammar(O, productions)
rd_parser = parse.RecursiveDescentParser(grammar)
# Tokens need to be redefined.
# It disappears after first use, and I don't know why.
tokens = tokenize.regexp_tokenize(self.string, re_all)
toklist = list(tokens)
# Store the parsing.
# Only the first one, as the grammar should be completely nonambiguous.
try:
self.parseList = rd_parser.get_parse_list(toklist)[0]
except IndexError:
print "Could not parse query."
return
# Set the nltk.parse.tree tree for this query to the global sentence
string = str(self.parseList)
string2 = string.replace(":","").replace("')'","").replace("table(","").replace("','","").replace("'","").replace("/","")
self.nltktree = parse.tree.bracket_parse(string2)
# Store the resulting nltk.parse.tree tree
self.parseTree = QuerySentence(self.nltktree)
self.xml = self.parseTree.toXML()
