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.

1472 lines
53 KiB
Python

# Natural Language Toolkit: Regular Expression Chunkers
#
# Copyright (C) 2001-2020 NLTK Project
# Author: Edward Loper <edloper@gmail.com>
# Steven Bird <stevenbird1@gmail.com> (minor additions)
# URL: <http://nltk.org/>
# For license information, see LICENSE.TXT
import re
from nltk.tree import Tree
from nltk.chunk.api import ChunkParserI
##//////////////////////////////////////////////////////
## ChunkString
##//////////////////////////////////////////////////////
class ChunkString(object):
"""
A string-based encoding of a particular chunking of a text.
Internally, the ``ChunkString`` class uses a single string to
encode the chunking of the input text. This string contains a
sequence of angle-bracket delimited tags, with chunking indicated
by braces. An example of this encoding is::
{<DT><JJ><NN>}<VBN><IN>{<DT><NN>}<.>{<DT><NN>}<VBD><.>
``ChunkString`` are created from tagged texts (i.e., lists of
``tokens`` whose type is ``TaggedType``). Initially, nothing is
chunked.
The chunking of a ``ChunkString`` can be modified with the ``xform()``
method, which uses a regular expression to transform the string
representation. These transformations should only add and remove
braces; they should *not* modify the sequence of angle-bracket
delimited tags.
:type _str: str
:ivar _str: The internal string representation of the text's
encoding. This string representation contains a sequence of
angle-bracket delimited tags, with chunking indicated by
braces. An example of this encoding is::
{<DT><JJ><NN>}<VBN><IN>{<DT><NN>}<.>{<DT><NN>}<VBD><.>
:type _pieces: list(tagged tokens and chunks)
:ivar _pieces: The tagged tokens and chunks encoded by this ``ChunkString``.
:ivar _debug: The debug level. See the constructor docs.
:cvar IN_CHUNK_PATTERN: A zero-width regexp pattern string that
will only match positions that are in chunks.
:cvar IN_CHINK_PATTERN: A zero-width regexp pattern string that
will only match positions that are in chinks.
"""
CHUNK_TAG_CHAR = r"[^\{\}<>]"
CHUNK_TAG = r"(<%s+?>)" % CHUNK_TAG_CHAR
IN_CHUNK_PATTERN = r"(?=[^\{]*\})"
IN_CHINK_PATTERN = r"(?=[^\}]*(\{|$))"
# These are used by _verify
_CHUNK = r"(\{%s+?\})+?" % CHUNK_TAG
_CHINK = r"(%s+?)+?" % CHUNK_TAG
_VALID = re.compile(r"^(\{?%s\}?)*?$" % CHUNK_TAG)
_BRACKETS = re.compile("[^\{\}]+")
_BALANCED_BRACKETS = re.compile(r"(\{\})*$")
def __init__(self, chunk_struct, debug_level=1):
"""
Construct a new ``ChunkString`` that encodes the chunking of
the text ``tagged_tokens``.
:type chunk_struct: Tree
:param chunk_struct: The chunk structure to be further chunked.
:type debug_level: int
:param debug_level: The level of debugging which should be
applied to transformations on the ``ChunkString``. The
valid levels are:
- 0: no checks
- 1: full check on to_chunkstruct
- 2: full check on to_chunkstruct and cursory check after
each transformation.
- 3: full check on to_chunkstruct and full check after
each transformation.
We recommend you use at least level 1. You should
probably use level 3 if you use any non-standard
subclasses of ``RegexpChunkRule``.
"""
self._root_label = chunk_struct.label()
self._pieces = chunk_struct[:]
tags = [self._tag(tok) for tok in self._pieces]
self._str = "<" + "><".join(tags) + ">"
self._debug = debug_level
def _tag(self, tok):
if isinstance(tok, tuple):
return tok[1]
elif isinstance(tok, Tree):
return tok.label()
else:
raise ValueError("chunk structures must contain tagged " "tokens or trees")
def _verify(self, s, verify_tags):
"""
Check to make sure that ``s`` still corresponds to some chunked
version of ``_pieces``.
:type verify_tags: bool
:param verify_tags: Whether the individual tags should be
checked. If this is false, ``_verify`` will check to make
sure that ``_str`` encodes a chunked version of *some*
list of tokens. If this is true, then ``_verify`` will
check to make sure that the tags in ``_str`` match those in
``_pieces``.
:raise ValueError: if the internal string representation of
this ``ChunkString`` is invalid or not consistent with _pieces.
"""
# Check overall form
if not ChunkString._VALID.match(s):
raise ValueError(
"Transformation generated invalid " "chunkstring:\n %s" % s
)
# Check that parens are balanced. If the string is long, we
# have to do this in pieces, to avoid a maximum recursion
# depth limit for regular expressions.
brackets = ChunkString._BRACKETS.sub("", s)
for i in range(1 + len(brackets) // 5000):
substr = brackets[i * 5000 : i * 5000 + 5000]
if not ChunkString._BALANCED_BRACKETS.match(substr):
raise ValueError(
"Transformation generated invalid " "chunkstring:\n %s" % s
)
if verify_tags <= 0:
return
tags1 = (re.split(r"[\{\}<>]+", s))[1:-1]
tags2 = [self._tag(piece) for piece in self._pieces]
if tags1 != tags2:
raise ValueError(
"Transformation generated invalid " "chunkstring: tag changed"
)
def to_chunkstruct(self, chunk_label="CHUNK"):
"""
Return the chunk structure encoded by this ``ChunkString``.
:rtype: Tree
:raise ValueError: If a transformation has generated an
invalid chunkstring.
"""
if self._debug > 0:
self._verify(self._str, 1)
# Use this alternating list to create the chunkstruct.
pieces = []
index = 0
piece_in_chunk = 0
for piece in re.split("[{}]", self._str):
# Find the list of tokens contained in this piece.
length = piece.count("<")
subsequence = self._pieces[index : index + length]
# Add this list of tokens to our pieces.
if piece_in_chunk:
pieces.append(Tree(chunk_label, subsequence))
else:
pieces += subsequence
# Update index, piece_in_chunk
index += length
piece_in_chunk = not piece_in_chunk
return Tree(self._root_label, pieces)
def xform(self, regexp, repl):
"""
Apply the given transformation to the string encoding of this
``ChunkString``. In particular, find all occurrences that match
``regexp``, and replace them using ``repl`` (as done by
``re.sub``).
This transformation should only add and remove braces; it
should *not* modify the sequence of angle-bracket delimited
tags. Furthermore, this transformation may not result in
improper bracketing. Note, in particular, that bracketing may
not be nested.
:type regexp: str or regexp
:param regexp: A regular expression matching the substring
that should be replaced. This will typically include a
named group, which can be used by ``repl``.
:type repl: str
:param repl: An expression specifying what should replace the
matched substring. Typically, this will include a named
replacement group, specified by ``regexp``.
:rtype: None
:raise ValueError: If this transformation generated an
invalid chunkstring.
"""
# Do the actual substitution
s = re.sub(regexp, repl, self._str)
# The substitution might have generated "empty chunks"
# (substrings of the form "{}"). Remove them, so they don't
# interfere with other transformations.
s = re.sub("\{\}", "", s)
# Make sure that the transformation was legal.
if self._debug > 1:
self._verify(s, self._debug - 2)
# Commit the transformation.
self._str = s
def __repr__(self):
"""
Return a string representation of this ``ChunkString``.
It has the form::
<ChunkString: '{<DT><JJ><NN>}<VBN><IN>{<DT><NN>}'>
:rtype: str
"""
return "<ChunkString: %s>" % repr(self._str)
def __str__(self):
"""
Return a formatted representation of this ``ChunkString``.
This representation will include extra spaces to ensure that
tags will line up with the representation of other
``ChunkStrings`` for the same text, regardless of the chunking.
:rtype: str
"""
# Add spaces to make everything line up.
str = re.sub(r">(?!\})", r"> ", self._str)
str = re.sub(r"([^\{])<", r"\1 <", str)
if str[0] == "<":
str = " " + str
return str
##//////////////////////////////////////////////////////
## Chunking Rules
##//////////////////////////////////////////////////////
class RegexpChunkRule(object):
"""
A rule specifying how to modify the chunking in a ``ChunkString``,
using a transformational regular expression. The
``RegexpChunkRule`` class itself can be used to implement any
transformational rule based on regular expressions. There are
also a number of subclasses, which can be used to implement
simpler types of rules, based on matching regular expressions.
Each ``RegexpChunkRule`` has a regular expression and a
replacement expression. When a ``RegexpChunkRule`` is "applied"
to a ``ChunkString``, it searches the ``ChunkString`` for any
substring that matches the regular expression, and replaces it
using the replacement expression. This search/replace operation
has the same semantics as ``re.sub``.
Each ``RegexpChunkRule`` also has a description string, which
gives a short (typically less than 75 characters) description of
the purpose of the rule.
This transformation defined by this ``RegexpChunkRule`` should
only add and remove braces; it should *not* modify the sequence
of angle-bracket delimited tags. Furthermore, this transformation
may not result in nested or mismatched bracketing.
"""
def __init__(self, regexp, repl, descr):
"""
Construct a new RegexpChunkRule.
:type regexp: regexp or str
:param regexp: The regular expression for this ``RegexpChunkRule``.
When this rule is applied to a ``ChunkString``, any
substring that matches ``regexp`` will be replaced using
the replacement string ``repl``. Note that this must be a
normal regular expression, not a tag pattern.
:type repl: str
:param repl: The replacement expression for this ``RegexpChunkRule``.
When this rule is applied to a ``ChunkString``, any substring
that matches ``regexp`` will be replaced using ``repl``.
:type descr: str
:param descr: A short description of the purpose and/or effect
of this rule.
"""
if isinstance(regexp, str):
regexp = re.compile(regexp)
self._repl = repl
self._descr = descr
self._regexp = regexp
def apply(self, chunkstr):
# Keep docstring generic so we can inherit it.
"""
Apply this rule to the given ``ChunkString``. See the
class reference documentation for a description of what it
means to apply a rule.
:type chunkstr: ChunkString
:param chunkstr: The chunkstring to which this rule is applied.
:rtype: None
:raise ValueError: If this transformation generated an
invalid chunkstring.
"""
chunkstr.xform(self._regexp, self._repl)
def descr(self):
"""
Return a short description of the purpose and/or effect of
this rule.
:rtype: str
"""
return self._descr
def __repr__(self):
"""
Return a string representation of this rule. It has the form::
<RegexpChunkRule: '{<IN|VB.*>}'->'<IN>'>
Note that this representation does not include the
description string; that string can be accessed
separately with the ``descr()`` method.
:rtype: str
"""
return (
"<RegexpChunkRule: "
+ repr(self._regexp.pattern)
+ "->"
+ repr(self._repl)
+ ">"
)
@staticmethod
def fromstring(s):
"""
Create a RegexpChunkRule from a string description.
Currently, the following formats are supported::
{regexp} # chunk rule
}regexp{ # chink rule
regexp}{regexp # split rule
regexp{}regexp # merge rule
Where ``regexp`` is a regular expression for the rule. Any
text following the comment marker (``#``) will be used as
the rule's description:
>>> from nltk.chunk.regexp import RegexpChunkRule
>>> RegexpChunkRule.fromstring('{<DT>?<NN.*>+}')
<ChunkRule: '<DT>?<NN.*>+'>
"""
# Split off the comment (but don't split on '\#')
m = re.match(r"(?P<rule>(\\.|[^#])*)(?P<comment>#.*)?", s)
rule = m.group("rule").strip()
comment = (m.group("comment") or "")[1:].strip()
# Pattern bodies: chunk, chink, split, merge
try:
if not rule:
raise ValueError("Empty chunk pattern")
if rule[0] == "{" and rule[-1] == "}":
return ChunkRule(rule[1:-1], comment)
elif rule[0] == "}" and rule[-1] == "{":
return ChinkRule(rule[1:-1], comment)
elif "}{" in rule:
left, right = rule.split("}{")
return SplitRule(left, right, comment)
elif "{}" in rule:
left, right = rule.split("{}")
return MergeRule(left, right, comment)
elif re.match("[^{}]*{[^{}]*}[^{}]*", rule):
left, chunk, right = re.split("[{}]", rule)
return ChunkRuleWithContext(left, chunk, right, comment)
else:
raise ValueError("Illegal chunk pattern: %s" % rule)
except (ValueError, re.error):
raise ValueError("Illegal chunk pattern: %s" % rule)
class ChunkRule(RegexpChunkRule):
"""
A rule specifying how to add chunks to a ``ChunkString``, using a
matching tag pattern. When applied to a ``ChunkString``, it will
find any substring that matches this tag pattern and that is not
already part of a chunk, and create a new chunk containing that
substring.
"""
def __init__(self, tag_pattern, descr):
"""
Construct a new ``ChunkRule``.
:type tag_pattern: str
:param tag_pattern: This rule's tag pattern. When
applied to a ``ChunkString``, this rule will
chunk any substring that matches this tag pattern and that
is not already part of a chunk.
:type descr: str
:param descr: A short description of the purpose and/or effect
of this rule.
"""
self._pattern = tag_pattern
regexp = re.compile(
"(?P<chunk>%s)%s"
% (tag_pattern2re_pattern(tag_pattern), ChunkString.IN_CHINK_PATTERN)
)
RegexpChunkRule.__init__(self, regexp, "{\g<chunk>}", descr)
def __repr__(self):
"""
Return a string representation of this rule. It has the form::
<ChunkRule: '<IN|VB.*>'>
Note that this representation does not include the
description string; that string can be accessed
separately with the ``descr()`` method.
:rtype: str
"""
return "<ChunkRule: " + repr(self._pattern) + ">"
class ChinkRule(RegexpChunkRule):
"""
A rule specifying how to remove chinks to a ``ChunkString``,
using a matching tag pattern. When applied to a
``ChunkString``, it will find any substring that matches this
tag pattern and that is contained in a chunk, and remove it
from that chunk, thus creating two new chunks.
"""
def __init__(self, tag_pattern, descr):
"""
Construct a new ``ChinkRule``.
:type tag_pattern: str
:param tag_pattern: This rule's tag pattern. When
applied to a ``ChunkString``, this rule will
find any substring that matches this tag pattern and that
is contained in a chunk, and remove it from that chunk,
thus creating two new chunks.
:type descr: str
:param descr: A short description of the purpose and/or effect
of this rule.
"""
self._pattern = tag_pattern
regexp = re.compile(
"(?P<chink>%s)%s"
% (tag_pattern2re_pattern(tag_pattern), ChunkString.IN_CHUNK_PATTERN)
)
RegexpChunkRule.__init__(self, regexp, "}\g<chink>{", descr)
def __repr__(self):
"""
Return a string representation of this rule. It has the form::
<ChinkRule: '<IN|VB.*>'>
Note that this representation does not include the
description string; that string can be accessed
separately with the ``descr()`` method.
:rtype: str
"""
return "<ChinkRule: " + repr(self._pattern) + ">"
class UnChunkRule(RegexpChunkRule):
"""
A rule specifying how to remove chunks to a ``ChunkString``,
using a matching tag pattern. When applied to a
``ChunkString``, it will find any complete chunk that matches this
tag pattern, and un-chunk it.
"""
def __init__(self, tag_pattern, descr):
"""
Construct a new ``UnChunkRule``.
:type tag_pattern: str
:param tag_pattern: This rule's tag pattern. When
applied to a ``ChunkString``, this rule will
find any complete chunk that matches this tag pattern,
and un-chunk it.
:type descr: str
:param descr: A short description of the purpose and/or effect
of this rule.
"""
self._pattern = tag_pattern
regexp = re.compile("\{(?P<chunk>%s)\}" % tag_pattern2re_pattern(tag_pattern))
RegexpChunkRule.__init__(self, regexp, "\g<chunk>", descr)
def __repr__(self):
"""
Return a string representation of this rule. It has the form::
<UnChunkRule: '<IN|VB.*>'>
Note that this representation does not include the
description string; that string can be accessed
separately with the ``descr()`` method.
:rtype: str
"""
return "<UnChunkRule: " + repr(self._pattern) + ">"
class MergeRule(RegexpChunkRule):
"""
A rule specifying how to merge chunks in a ``ChunkString``, using
two matching tag patterns: a left pattern, and a right pattern.
When applied to a ``ChunkString``, it will find any chunk whose end
matches left pattern, and immediately followed by a chunk whose
beginning matches right pattern. It will then merge those two
chunks into a single chunk.
"""
def __init__(self, left_tag_pattern, right_tag_pattern, descr):
"""
Construct a new ``MergeRule``.
:type right_tag_pattern: str
:param right_tag_pattern: This rule's right tag
pattern. When applied to a ``ChunkString``, this
rule will find any chunk whose end matches
``left_tag_pattern``, and immediately followed by a chunk
whose beginning matches this pattern. It will
then merge those two chunks into a single chunk.
:type left_tag_pattern: str
:param left_tag_pattern: This rule's left tag
pattern. When applied to a ``ChunkString``, this
rule will find any chunk whose end matches
this pattern, and immediately followed by a chunk
whose beginning matches ``right_tag_pattern``. It will
then merge those two chunks into a single chunk.
:type descr: str
:param descr: A short description of the purpose and/or effect
of this rule.
"""
# Ensure that the individual patterns are coherent. E.g., if
# left='(' and right=')', then this will raise an exception:
re.compile(tag_pattern2re_pattern(left_tag_pattern))
re.compile(tag_pattern2re_pattern(right_tag_pattern))
self._left_tag_pattern = left_tag_pattern
self._right_tag_pattern = right_tag_pattern
regexp = re.compile(
"(?P<left>%s)}{(?=%s)"
% (
tag_pattern2re_pattern(left_tag_pattern),
tag_pattern2re_pattern(right_tag_pattern),
)
)
RegexpChunkRule.__init__(self, regexp, "\g<left>", descr)
def __repr__(self):
"""
Return a string representation of this rule. It has the form::
<MergeRule: '<NN|DT|JJ>', '<NN|JJ>'>
Note that this representation does not include the
description string; that string can be accessed
separately with the ``descr()`` method.
:rtype: str
"""
return (
"<MergeRule: "
+ repr(self._left_tag_pattern)
+ ", "
+ repr(self._right_tag_pattern)
+ ">"
)
class SplitRule(RegexpChunkRule):
"""
A rule specifying how to split chunks in a ``ChunkString``, using
two matching tag patterns: a left pattern, and a right pattern.
When applied to a ``ChunkString``, it will find any chunk that
matches the left pattern followed by the right pattern. It will
then split the chunk into two new chunks, at the point between the
two pattern matches.
"""
def __init__(self, left_tag_pattern, right_tag_pattern, descr):
"""
Construct a new ``SplitRule``.
:type right_tag_pattern: str
:param right_tag_pattern: This rule's right tag
pattern. When applied to a ``ChunkString``, this rule will
find any chunk containing a substring that matches
``left_tag_pattern`` followed by this pattern. It will
then split the chunk into two new chunks at the point
between these two matching patterns.
:type left_tag_pattern: str
:param left_tag_pattern: This rule's left tag
pattern. When applied to a ``ChunkString``, this rule will
find any chunk containing a substring that matches this
pattern followed by ``right_tag_pattern``. It will then
split the chunk into two new chunks at the point between
these two matching patterns.
:type descr: str
:param descr: A short description of the purpose and/or effect
of this rule.
"""
# Ensure that the individual patterns are coherent. E.g., if
# left='(' and right=')', then this will raise an exception:
re.compile(tag_pattern2re_pattern(left_tag_pattern))
re.compile(tag_pattern2re_pattern(right_tag_pattern))
self._left_tag_pattern = left_tag_pattern
self._right_tag_pattern = right_tag_pattern
regexp = re.compile(
"(?P<left>%s)(?=%s)"
% (
tag_pattern2re_pattern(left_tag_pattern),
tag_pattern2re_pattern(right_tag_pattern),
)
)
RegexpChunkRule.__init__(self, regexp, r"\g<left>}{", descr)
def __repr__(self):
"""
Return a string representation of this rule. It has the form::
<SplitRule: '<NN>', '<DT>'>
Note that this representation does not include the
description string; that string can be accessed
separately with the ``descr()`` method.
:rtype: str
"""
return (
"<SplitRule: "
+ repr(self._left_tag_pattern)
+ ", "
+ repr(self._right_tag_pattern)
+ ">"
)
class ExpandLeftRule(RegexpChunkRule):
"""
A rule specifying how to expand chunks in a ``ChunkString`` to the left,
using two matching tag patterns: a left pattern, and a right pattern.
When applied to a ``ChunkString``, it will find any chunk whose beginning
matches right pattern, and immediately preceded by a chink whose
end matches left pattern. It will then expand the chunk to incorporate
the new material on the left.
"""
def __init__(self, left_tag_pattern, right_tag_pattern, descr):
"""
Construct a new ``ExpandRightRule``.
:type right_tag_pattern: str
:param right_tag_pattern: This rule's right tag
pattern. When applied to a ``ChunkString``, this
rule will find any chunk whose beginning matches
``right_tag_pattern``, and immediately preceded by a chink
whose end matches this pattern. It will
then merge those two chunks into a single chunk.
:type left_tag_pattern: str
:param left_tag_pattern: This rule's left tag
pattern. When applied to a ``ChunkString``, this
rule will find any chunk whose beginning matches
this pattern, and immediately preceded by a chink
whose end matches ``left_tag_pattern``. It will
then expand the chunk to incorporate the new material on the left.
:type descr: str
:param descr: A short description of the purpose and/or effect
of this rule.
"""
# Ensure that the individual patterns are coherent. E.g., if
# left='(' and right=')', then this will raise an exception:
re.compile(tag_pattern2re_pattern(left_tag_pattern))
re.compile(tag_pattern2re_pattern(right_tag_pattern))
self._left_tag_pattern = left_tag_pattern
self._right_tag_pattern = right_tag_pattern
regexp = re.compile(
"(?P<left>%s)\{(?P<right>%s)"
% (
tag_pattern2re_pattern(left_tag_pattern),
tag_pattern2re_pattern(right_tag_pattern),
)
)
RegexpChunkRule.__init__(self, regexp, "{\g<left>\g<right>", descr)
def __repr__(self):
"""
Return a string representation of this rule. It has the form::
<ExpandLeftRule: '<NN|DT|JJ>', '<NN|JJ>'>
Note that this representation does not include the
description string; that string can be accessed
separately with the ``descr()`` method.
:rtype: str
"""
return (
"<ExpandLeftRule: "
+ repr(self._left_tag_pattern)
+ ", "
+ repr(self._right_tag_pattern)
+ ">"
)
class ExpandRightRule(RegexpChunkRule):
"""
A rule specifying how to expand chunks in a ``ChunkString`` to the
right, using two matching tag patterns: a left pattern, and a
right pattern. When applied to a ``ChunkString``, it will find any
chunk whose end matches left pattern, and immediately followed by
a chink whose beginning matches right pattern. It will then
expand the chunk to incorporate the new material on the right.
"""
def __init__(self, left_tag_pattern, right_tag_pattern, descr):
"""
Construct a new ``ExpandRightRule``.
:type right_tag_pattern: str
:param right_tag_pattern: This rule's right tag
pattern. When applied to a ``ChunkString``, this
rule will find any chunk whose end matches
``left_tag_pattern``, and immediately followed by a chink
whose beginning matches this pattern. It will
then merge those two chunks into a single chunk.
:type left_tag_pattern: str
:param left_tag_pattern: This rule's left tag
pattern. When applied to a ``ChunkString``, this
rule will find any chunk whose end matches
this pattern, and immediately followed by a chink
whose beginning matches ``right_tag_pattern``. It will
then expand the chunk to incorporate the new material on the right.
:type descr: str
:param descr: A short description of the purpose and/or effect
of this rule.
"""
# Ensure that the individual patterns are coherent. E.g., if
# left='(' and right=')', then this will raise an exception:
re.compile(tag_pattern2re_pattern(left_tag_pattern))
re.compile(tag_pattern2re_pattern(right_tag_pattern))
self._left_tag_pattern = left_tag_pattern
self._right_tag_pattern = right_tag_pattern
regexp = re.compile(
"(?P<left>%s)\}(?P<right>%s)"
% (
tag_pattern2re_pattern(left_tag_pattern),
tag_pattern2re_pattern(right_tag_pattern),
)
)
RegexpChunkRule.__init__(self, regexp, "\g<left>\g<right>}", descr)
def __repr__(self):
"""
Return a string representation of this rule. It has the form::
<ExpandRightRule: '<NN|DT|JJ>', '<NN|JJ>'>
Note that this representation does not include the
description string; that string can be accessed
separately with the ``descr()`` method.
:rtype: str
"""
return (
"<ExpandRightRule: "
+ repr(self._left_tag_pattern)
+ ", "
+ repr(self._right_tag_pattern)
+ ">"
)
class ChunkRuleWithContext(RegexpChunkRule):
"""
A rule specifying how to add chunks to a ``ChunkString``, using
three matching tag patterns: one for the left context, one for the
chunk, and one for the right context. When applied to a
``ChunkString``, it will find any substring that matches the chunk
tag pattern, is surrounded by substrings that match the two
context patterns, and is not already part of a chunk; and create a
new chunk containing the substring that matched the chunk tag
pattern.
Caveat: Both the left and right context are consumed when this
rule matches; therefore, if you need to find overlapping matches,
you will need to apply your rule more than once.
"""
def __init__(
self,
left_context_tag_pattern,
chunk_tag_pattern,
right_context_tag_pattern,
descr,
):
"""
Construct a new ``ChunkRuleWithContext``.
:type left_context_tag_pattern: str
:param left_context_tag_pattern: A tag pattern that must match
the left context of ``chunk_tag_pattern`` for this rule to
apply.
:type chunk_tag_pattern: str
:param chunk_tag_pattern: A tag pattern that must match for this
rule to apply. If the rule does apply, then this pattern
also identifies the substring that will be made into a chunk.
:type right_context_tag_pattern: str
:param right_context_tag_pattern: A tag pattern that must match
the right context of ``chunk_tag_pattern`` for this rule to
apply.
:type descr: str
:param descr: A short description of the purpose and/or effect
of this rule.
"""
# Ensure that the individual patterns are coherent. E.g., if
# left='(' and right=')', then this will raise an exception:
re.compile(tag_pattern2re_pattern(left_context_tag_pattern))
re.compile(tag_pattern2re_pattern(chunk_tag_pattern))
re.compile(tag_pattern2re_pattern(right_context_tag_pattern))
self._left_context_tag_pattern = left_context_tag_pattern
self._chunk_tag_pattern = chunk_tag_pattern
self._right_context_tag_pattern = right_context_tag_pattern
regexp = re.compile(
"(?P<left>%s)(?P<chunk>%s)(?P<right>%s)%s"
% (
tag_pattern2re_pattern(left_context_tag_pattern),
tag_pattern2re_pattern(chunk_tag_pattern),
tag_pattern2re_pattern(right_context_tag_pattern),
ChunkString.IN_CHINK_PATTERN,
)
)
replacement = r"\g<left>{\g<chunk>}\g<right>"
RegexpChunkRule.__init__(self, regexp, replacement, descr)
def __repr__(self):
"""
Return a string representation of this rule. It has the form::
<ChunkRuleWithContext: '<IN>', '<NN>', '<DT>'>
Note that this representation does not include the
description string; that string can be accessed
separately with the ``descr()`` method.
:rtype: str
"""
return "<ChunkRuleWithContext: %r, %r, %r>" % (
self._left_context_tag_pattern,
self._chunk_tag_pattern,
self._right_context_tag_pattern,
)
##//////////////////////////////////////////////////////
## Tag Pattern Format Conversion
##//////////////////////////////////////////////////////
# this should probably be made more strict than it is -- e.g., it
# currently accepts 'foo'.
CHUNK_TAG_PATTERN = re.compile(
r"^((%s|<%s>)*)$" % ("([^\{\}<>]|\{\d+,?\}|\{\d*,\d+\})+", "[^\{\}<>]+")
)
def tag_pattern2re_pattern(tag_pattern):
"""
Convert a tag pattern to a regular expression pattern. A "tag
pattern" is a modified version of a regular expression, designed
for matching sequences of tags. The differences between regular
expression patterns and tag patterns are:
- In tag patterns, ``'<'`` and ``'>'`` act as parentheses; so
``'<NN>+'`` matches one or more repetitions of ``'<NN>'``, not
``'<NN'`` followed by one or more repetitions of ``'>'``.
- Whitespace in tag patterns is ignored. So
``'<DT> | <NN>'`` is equivalant to ``'<DT>|<NN>'``
- In tag patterns, ``'.'`` is equivalant to ``'[^{}<>]'``; so
``'<NN.*>'`` matches any single tag starting with ``'NN'``.
In particular, ``tag_pattern2re_pattern`` performs the following
transformations on the given pattern:
- Replace '.' with '[^<>{}]'
- Remove any whitespace
- Add extra parens around '<' and '>', to make '<' and '>' act
like parentheses. E.g., so that in '<NN>+', the '+' has scope
over the entire '<NN>'; and so that in '<NN|IN>', the '|' has
scope over 'NN' and 'IN', but not '<' or '>'.
- Check to make sure the resulting pattern is valid.
:type tag_pattern: str
:param tag_pattern: The tag pattern to convert to a regular
expression pattern.
:raise ValueError: If ``tag_pattern`` is not a valid tag pattern.
In particular, ``tag_pattern`` should not include braces; and it
should not contain nested or mismatched angle-brackets.
:rtype: str
:return: A regular expression pattern corresponding to
``tag_pattern``.
"""
# Clean up the regular expression
tag_pattern = re.sub(r"\s", "", tag_pattern)
tag_pattern = re.sub(r"<", "(<(", tag_pattern)
tag_pattern = re.sub(r">", ")>)", tag_pattern)
# Check the regular expression
if not CHUNK_TAG_PATTERN.match(tag_pattern):
raise ValueError("Bad tag pattern: %r" % tag_pattern)
# Replace "." with CHUNK_TAG_CHAR.
# We have to do this after, since it adds {}[]<>s, which would
# confuse CHUNK_TAG_PATTERN.
# PRE doesn't have lookback assertions, so reverse twice, and do
# the pattern backwards (with lookahead assertions). This can be
# made much cleaner once we can switch back to SRE.
def reverse_str(str):
lst = list(str)
lst.reverse()
return "".join(lst)
tc_rev = reverse_str(ChunkString.CHUNK_TAG_CHAR)
reversed = reverse_str(tag_pattern)
reversed = re.sub(r"\.(?!\\(\\\\)*($|[^\\]))", tc_rev, reversed)
tag_pattern = reverse_str(reversed)
return tag_pattern
##//////////////////////////////////////////////////////
## RegexpChunkParser
##//////////////////////////////////////////////////////
class RegexpChunkParser(ChunkParserI):
"""
A regular expression based chunk parser. ``RegexpChunkParser`` uses a
sequence of "rules" to find chunks of a single type within a
text. The chunking of the text is encoded using a ``ChunkString``,
and each rule acts by modifying the chunking in the
``ChunkString``. The rules are all implemented using regular
expression matching and substitution.
The ``RegexpChunkRule`` class and its subclasses (``ChunkRule``,
``ChinkRule``, ``UnChunkRule``, ``MergeRule``, and ``SplitRule``)
define the rules that are used by ``RegexpChunkParser``. Each rule
defines an ``apply()`` method, which modifies the chunking encoded
by a given ``ChunkString``.
:type _rules: list(RegexpChunkRule)
:ivar _rules: The list of rules that should be applied to a text.
:type _trace: int
:ivar _trace: The default level of tracing.
"""
def __init__(self, rules, chunk_label="NP", root_label="S", trace=0):
"""
Construct a new ``RegexpChunkParser``.
:type rules: list(RegexpChunkRule)
:param rules: The sequence of rules that should be used to
generate the chunking for a tagged text.
:type chunk_label: str
:param chunk_label: The node value that should be used for
chunk subtrees. This is typically a short string
describing the type of information contained by the chunk,
such as ``"NP"`` for base noun phrases.
:type root_label: str
:param root_label: The node value that should be used for the
top node of the chunk structure.
:type trace: int
:param trace: The level of tracing that should be used when
parsing a text. ``0`` will generate no tracing output;
``1`` will generate normal tracing output; and ``2`` or
higher will generate verbose tracing output.
"""
self._rules = rules
self._trace = trace
self._chunk_label = chunk_label
self._root_label = root_label
def _trace_apply(self, chunkstr, verbose):
"""
Apply each rule of this ``RegexpChunkParser`` to ``chunkstr``, in
turn. Generate trace output between each rule. If ``verbose``
is true, then generate verbose output.
:type chunkstr: ChunkString
:param chunkstr: The chunk string to which each rule should be
applied.
:type verbose: bool
:param verbose: Whether output should be verbose.
:rtype: None
"""
print("# Input:")
print(chunkstr)
for rule in self._rules:
rule.apply(chunkstr)
if verbose:
print("#", rule.descr() + " (" + repr(rule) + "):")
else:
print("#", rule.descr() + ":")
print(chunkstr)
def _notrace_apply(self, chunkstr):
"""
Apply each rule of this ``RegexpChunkParser`` to ``chunkstr``, in
turn.
:param chunkstr: The chunk string to which each rule should be
applied.
:type chunkstr: ChunkString
:rtype: None
"""
for rule in self._rules:
rule.apply(chunkstr)
def parse(self, chunk_struct, trace=None):
"""
:type chunk_struct: Tree
:param chunk_struct: the chunk structure to be (further) chunked
:type trace: int
:param trace: The level of tracing that should be used when
parsing a text. ``0`` will generate no tracing output;
``1`` will generate normal tracing output; and ``2`` or
highter will generate verbose tracing output. This value
overrides the trace level value that was given to the
constructor.
:rtype: Tree
:return: a chunk structure that encodes the chunks in a given
tagged sentence. A chunk is a non-overlapping linguistic
group, such as a noun phrase. The set of chunks
identified in the chunk structure depends on the rules
used to define this ``RegexpChunkParser``.
"""
if len(chunk_struct) == 0:
print("Warning: parsing empty text")
return Tree(self._root_label, [])
try:
chunk_struct.label()
except AttributeError:
chunk_struct = Tree(self._root_label, chunk_struct)
# Use the default trace value?
if trace is None:
trace = self._trace
chunkstr = ChunkString(chunk_struct)
# Apply the sequence of rules to the chunkstring.
if trace:
verbose = trace > 1
self._trace_apply(chunkstr, verbose)
else:
self._notrace_apply(chunkstr)
# Use the chunkstring to create a chunk structure.
return chunkstr.to_chunkstruct(self._chunk_label)
def rules(self):
"""
:return: the sequence of rules used by ``RegexpChunkParser``.
:rtype: list(RegexpChunkRule)
"""
return self._rules
def __repr__(self):
"""
:return: a concise string representation of this
``RegexpChunkParser``.
:rtype: str
"""
return "<RegexpChunkParser with %d rules>" % len(self._rules)
def __str__(self):
"""
:return: a verbose string representation of this ``RegexpChunkParser``.
:rtype: str
"""
s = "RegexpChunkParser with %d rules:\n" % len(self._rules)
margin = 0
for rule in self._rules:
margin = max(margin, len(rule.descr()))
if margin < 35:
format = " %" + repr(-(margin + 3)) + "s%s\n"
else:
format = " %s\n %s\n"
for rule in self._rules:
s += format % (rule.descr(), repr(rule))
return s[:-1]
##//////////////////////////////////////////////////////
## Chunk Grammar
##//////////////////////////////////////////////////////
class RegexpParser(ChunkParserI):
"""
A grammar based chunk parser. ``chunk.RegexpParser`` uses a set of
regular expression patterns to specify the behavior of the parser.
The chunking of the text is encoded using a ``ChunkString``, and
each rule acts by modifying the chunking in the ``ChunkString``.
The rules are all implemented using regular expression matching
and substitution.
A grammar contains one or more clauses in the following form::
NP:
{<DT|JJ>} # chunk determiners and adjectives
}<[\.VI].*>+{ # chink any tag beginning with V, I, or .
<.*>}{<DT> # split a chunk at a determiner
<DT|JJ>{}<NN.*> # merge chunk ending with det/adj
# with one starting with a noun
The patterns of a clause are executed in order. An earlier
pattern may introduce a chunk boundary that prevents a later
pattern from executing. Sometimes an individual pattern will
match on multiple, overlapping extents of the input. As with
regular expression substitution more generally, the chunker will
identify the first match possible, then continue looking for matches
after this one has ended.
The clauses of a grammar are also executed in order. A cascaded
chunk parser is one having more than one clause. The maximum depth
of a parse tree created by this chunk parser is the same as the
number of clauses in the grammar.
When tracing is turned on, the comment portion of a line is displayed
each time the corresponding pattern is applied.
:type _start: str
:ivar _start: The start symbol of the grammar (the root node of
resulting trees)
:type _stages: int
:ivar _stages: The list of parsing stages corresponding to the grammar
"""
def __init__(self, grammar, root_label="S", loop=1, trace=0):
"""
Create a new chunk parser, from the given start state
and set of chunk patterns.
:param grammar: The grammar, or a list of RegexpChunkParser objects
:type grammar: str or list(RegexpChunkParser)
:param root_label: The top node of the tree being created
:type root_label: str or Nonterminal
:param loop: The number of times to run through the patterns
:type loop: int
:type trace: int
:param trace: The level of tracing that should be used when
parsing a text. ``0`` will generate no tracing output;
``1`` will generate normal tracing output; and ``2`` or
higher will generate verbose tracing output.
"""
self._trace = trace
self._stages = []
self._grammar = grammar
self._loop = loop
if isinstance(grammar, str):
self._read_grammar(grammar, root_label, trace)
else:
# Make sur the grammar looks like it has the right type:
type_err = (
"Expected string or list of RegexpChunkParsers " "for the grammar."
)
try:
grammar = list(grammar)
except:
raise TypeError(type_err)
for elt in grammar:
if not isinstance(elt, RegexpChunkParser):
raise TypeError(type_err)
self._stages = grammar
def _read_grammar(self, grammar, root_label, trace):
"""
Helper function for __init__: read the grammar if it is a
string.
"""
rules = []
lhs = None
for line in grammar.split("\n"):
line = line.strip()
# New stage begins if there's an unescaped ':'
m = re.match("(?P<nonterminal>(\\.|[^:])*)(:(?P<rule>.*))", line)
if m:
# Record the stage that we just completed.
self._add_stage(rules, lhs, root_label, trace)
# Start a new stage.
lhs = m.group("nonterminal").strip()
rules = []
line = m.group("rule").strip()
# Skip blank & comment-only lines
if line == "" or line.startswith("#"):
continue
# Add the rule
rules.append(RegexpChunkRule.fromstring(line))
# Record the final stage
self._add_stage(rules, lhs, root_label, trace)
def _add_stage(self, rules, lhs, root_label, trace):
"""
Helper function for __init__: add a new stage to the parser.
"""
if rules != []:
if not lhs:
raise ValueError("Expected stage marker (eg NP:)")
parser = RegexpChunkParser(
rules, chunk_label=lhs, root_label=root_label, trace=trace
)
self._stages.append(parser)
def parse(self, chunk_struct, trace=None):
"""
Apply the chunk parser to this input.
:type chunk_struct: Tree
:param chunk_struct: the chunk structure to be (further) chunked
(this tree is modified, and is also returned)
:type trace: int
:param trace: The level of tracing that should be used when
parsing a text. ``0`` will generate no tracing output;
``1`` will generate normal tracing output; and ``2`` or
highter will generate verbose tracing output. This value
overrides the trace level value that was given to the
constructor.
:return: the chunked output.
:rtype: Tree
"""
if trace is None:
trace = self._trace
for i in range(self._loop):
for parser in self._stages:
chunk_struct = parser.parse(chunk_struct, trace=trace)
return chunk_struct
def __repr__(self):
"""
:return: a concise string representation of this ``chunk.RegexpParser``.
:rtype: str
"""
return "<chunk.RegexpParser with %d stages>" % len(self._stages)
def __str__(self):
"""
:return: a verbose string representation of this
``RegexpParser``.
:rtype: str
"""
s = "chunk.RegexpParser with %d stages:\n" % len(self._stages)
margin = 0
for parser in self._stages:
s += "%s\n" % parser
return s[:-1]
##//////////////////////////////////////////////////////
## Demonstration code
##//////////////////////////////////////////////////////
def demo_eval(chunkparser, text):
"""
Demonstration code for evaluating a chunk parser, using a
``ChunkScore``. This function assumes that ``text`` contains one
sentence per line, and that each sentence has the form expected by
``tree.chunk``. It runs the given chunk parser on each sentence in
the text, and scores the result. It prints the final score
(precision, recall, and f-measure); and reports the set of chunks
that were missed and the set of chunks that were incorrect. (At
most 10 missing chunks and 10 incorrect chunks are reported).
:param chunkparser: The chunkparser to be tested
:type chunkparser: ChunkParserI
:param text: The chunked tagged text that should be used for
evaluation.
:type text: str
"""
from nltk import chunk
from nltk.tree import Tree
# Evaluate our chunk parser.
chunkscore = chunk.ChunkScore()
for sentence in text.split("\n"):
print(sentence)
sentence = sentence.strip()
if not sentence:
continue
gold = chunk.tagstr2tree(sentence)
tokens = gold.leaves()
test = chunkparser.parse(Tree("S", tokens), trace=1)
chunkscore.score(gold, test)
print()
print("/" + ("=" * 75) + "\\")
print("Scoring", chunkparser)
print(("-" * 77))
print("Precision: %5.1f%%" % (chunkscore.precision() * 100), " " * 4, end=" ")
print("Recall: %5.1f%%" % (chunkscore.recall() * 100), " " * 6, end=" ")
print("F-Measure: %5.1f%%" % (chunkscore.f_measure() * 100))
# Missed chunks.
if chunkscore.missed():
print("Missed:")
missed = chunkscore.missed()
for chunk in missed[:10]:
print(" ", " ".join(map(str, chunk)))
if len(chunkscore.missed()) > 10:
print(" ...")
# Incorrect chunks.
if chunkscore.incorrect():
print("Incorrect:")
incorrect = chunkscore.incorrect()
for chunk in incorrect[:10]:
print(" ", " ".join(map(str, chunk)))
if len(chunkscore.incorrect()) > 10:
print(" ...")
print("\\" + ("=" * 75) + "/")
print()
def demo():
"""
A demonstration for the ``RegexpChunkParser`` class. A single text is
parsed with four different chunk parsers, using a variety of rules
and strategies.
"""
from nltk import chunk, Tree
text = """\
[ the/DT little/JJ cat/NN ] sat/VBD on/IN [ the/DT mat/NN ] ./.
[ John/NNP ] saw/VBD [the/DT cats/NNS] [the/DT dog/NN] chased/VBD ./.
[ John/NNP ] thinks/VBZ [ Mary/NN ] saw/VBD [ the/DT cat/NN ] sit/VB on/IN [ the/DT mat/NN ]./.
"""
print("*" * 75)
print("Evaluation text:")
print(text)
print("*" * 75)
print()
grammar = r"""
NP: # NP stage
{<DT>?<JJ>*<NN>} # chunk determiners, adjectives and nouns
{<NNP>+} # chunk proper nouns
"""
cp = chunk.RegexpParser(grammar)
demo_eval(cp, text)
grammar = r"""
NP:
{<.*>} # start by chunking each tag
}<[\.VI].*>+{ # unchunk any verbs, prepositions or periods
<DT|JJ>{}<NN.*> # merge det/adj with nouns
"""
cp = chunk.RegexpParser(grammar)
demo_eval(cp, text)
grammar = r"""
NP: {<DT>?<JJ>*<NN>} # chunk determiners, adjectives and nouns
VP: {<TO>?<VB.*>} # VP = verb words
"""
cp = chunk.RegexpParser(grammar)
demo_eval(cp, text)
grammar = r"""
NP: {<.*>*} # start by chunking everything
}<[\.VI].*>+{ # chink any verbs, prepositions or periods
<.*>}{<DT> # separate on determiners
PP: {<IN><NP>} # PP = preposition + noun phrase
VP: {<VB.*><NP|PP>*} # VP = verb words + NPs and PPs
"""
cp = chunk.RegexpParser(grammar)
demo_eval(cp, text)
# Evaluation
from nltk.corpus import conll2000
print()
print("Demonstration of empty grammar:")
cp = chunk.RegexpParser("")
print(chunk.accuracy(cp, conll2000.chunked_sents("test.txt", chunk_types=("NP",))))
print()
print("Demonstration of accuracy evaluation using CoNLL tags:")
grammar = r"""
NP:
{<.*>} # start by chunking each tag
}<[\.VI].*>+{ # unchunk any verbs, prepositions or periods
<DT|JJ>{}<NN.*> # merge det/adj with nouns
"""
cp = chunk.RegexpParser(grammar)
print(chunk.accuracy(cp, conll2000.chunked_sents("test.txt")[:5]))
print()
print("Demonstration of tagged token input")
grammar = r"""
NP: {<.*>*} # start by chunking everything
}<[\.VI].*>+{ # chink any verbs, prepositions or periods
<.*>}{<DT> # separate on determiners
PP: {<IN><NP>} # PP = preposition + noun phrase
VP: {<VB.*><NP|PP>*} # VP = verb words + NPs and PPs
"""
cp = chunk.RegexpParser(grammar)
print(
cp.parse(
[
("the", "DT"),
("little", "JJ"),
("cat", "NN"),
("sat", "VBD"),
("on", "IN"),
("the", "DT"),
("mat", "NN"),
(".", "."),
]
)
)
if __name__ == "__main__":
demo()