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.

714 lines
27 KiB
Python

"""
Porter Stemmer
This is the Porter stemming algorithm. It follows the algorithm
presented in
Porter, M. "An algorithm for suffix stripping." Program 14.3 (1980): 130-137.
with some optional deviations that can be turned on or off with the
`mode` argument to the constructor.
Martin Porter, the algorithm's inventor, maintains a web page about the
algorithm at
http://www.tartarus.org/~martin/PorterStemmer/
which includes another Python implementation and other implementations
in many languages.
"""
from __future__ import print_function, unicode_literals
__docformat__ = 'plaintext'
import re
from nltk.stem.api import StemmerI
from nltk.compat import python_2_unicode_compatible
@python_2_unicode_compatible
class PorterStemmer(StemmerI):
"""
A word stemmer based on the Porter stemming algorithm.
Porter, M. "An algorithm for suffix stripping."
Program 14.3 (1980): 130-137.
See http://www.tartarus.org/~martin/PorterStemmer/ for the homepage
of the algorithm.
Martin Porter has endorsed several modifications to the Porter
algorithm since writing his original paper, and those extensions are
included in the implementations on his website. Additionally, others
have proposed further improvements to the algorithm, including NLTK
contributors. There are thus three modes that can be selected by
passing the appropriate constant to the class constructor's `mode`
attribute:
PorterStemmer.ORIGINAL_ALGORITHM
- Implementation that is faithful to the original paper.
Note that Martin Porter has deprecated this version of the
algorithm. Martin distributes implementations of the Porter
Stemmer in many languages, hosted at:
http://www.tartarus.org/~martin/PorterStemmer/
and all of these implementations include his extensions. He
strongly recommends against using the original, published
version of the algorithm; only use this mode if you clearly
understand why you are choosing to do so.
PorterStemmer.MARTIN_EXTENSIONS
- Implementation that only uses the modifications to the
algorithm that are included in the implementations on Martin
Porter's website. He has declared Porter frozen, so the
behaviour of those implementations should never change.
PorterStemmer.NLTK_EXTENSIONS (default)
- Implementation that includes further improvements devised by
NLTK contributors or taken from other modified implementations
found on the web.
For the best stemming, you should use the default NLTK_EXTENSIONS
version. However, if you need to get the same results as either the
original algorithm or one of Martin Porter's hosted versions for
compatibility with an existing implementation or dataset, you can use
one of the other modes instead.
"""
# Modes the Stemmer can be instantiated in
NLTK_EXTENSIONS = 'NLTK_EXTENSIONS'
MARTIN_EXTENSIONS = 'MARTIN_EXTENSIONS'
ORIGINAL_ALGORITHM = 'ORIGINAL_ALGORITHM'
def __init__(self, mode=NLTK_EXTENSIONS):
if mode not in (
self.NLTK_EXTENSIONS,
self.MARTIN_EXTENSIONS,
self.ORIGINAL_ALGORITHM,
):
raise ValueError(
"Mode must be one of PorterStemmer.NLTK_EXTENSIONS, "
"PorterStemmer.MARTIN_EXTENSIONS, or "
"PorterStemmer.ORIGINAL_ALGORITHM"
)
self.mode = mode
if self.mode == self.NLTK_EXTENSIONS:
# This is a table of irregular forms. It is quite short,
# but still reflects the errors actually drawn to Martin
# Porter's attention over a 20 year period!
irregular_forms = {
"sky": ["sky", "skies"],
"die": ["dying"],
"lie": ["lying"],
"tie": ["tying"],
"news": ["news"],
"inning": ["innings", "inning"],
"outing": ["outings", "outing"],
"canning": ["cannings", "canning"],
"howe": ["howe"],
"proceed": ["proceed"],
"exceed": ["exceed"],
"succeed": ["succeed"],
}
self.pool = {}
for key in irregular_forms:
for val in irregular_forms[key]:
self.pool[val] = key
self.vowels = frozenset(['a', 'e', 'i', 'o', 'u'])
def _is_consonant(self, word, i):
"""Returns True if word[i] is a consonant, False otherwise
A consonant is defined in the paper as follows:
A consonant in a word is a letter other than A, E, I, O or
U, and other than Y preceded by a consonant. (The fact that
the term `consonant' is defined to some extent in terms of
itself does not make it ambiguous.) So in TOY the consonants
are T and Y, and in SYZYGY they are S, Z and G. If a letter
is not a consonant it is a vowel.
"""
if word[i] in self.vowels:
return False
if word[i] == 'y':
if i == 0:
return True
else:
return not self._is_consonant(word, i - 1)
return True
def _measure(self, stem):
"""Returns the 'measure' of stem, per definition in the paper
From the paper:
A consonant will be denoted by c, a vowel by v. A list
ccc... of length greater than 0 will be denoted by C, and a
list vvv... of length greater than 0 will be denoted by V.
Any word, or part of a word, therefore has one of the four
forms:
CVCV ... C
CVCV ... V
VCVC ... C
VCVC ... V
These may all be represented by the single form
[C]VCVC ... [V]
where the square brackets denote arbitrary presence of their
contents. Using (VC){m} to denote VC repeated m times, this
may again be written as
[C](VC){m}[V].
m will be called the \measure\ of any word or word part when
represented in this form. The case m = 0 covers the null
word. Here are some examples:
m=0 TR, EE, TREE, Y, BY.
m=1 TROUBLE, OATS, TREES, IVY.
m=2 TROUBLES, PRIVATE, OATEN, ORRERY.
"""
cv_sequence = ''
# Construct a string of 'c's and 'v's representing whether each
# character in `stem` is a consonant or a vowel.
# e.g. 'falafel' becomes 'cvcvcvc',
# 'architecture' becomes 'vcccvcvccvcv'
for i in range(len(stem)):
if self._is_consonant(stem, i):
cv_sequence += 'c'
else:
cv_sequence += 'v'
# Count the number of 'vc' occurences, which is equivalent to
# the number of 'VC' occurrences in Porter's reduced form in the
# docstring above, which is in turn equivalent to `m`
return cv_sequence.count('vc')
def _has_positive_measure(self, stem):
return self._measure(stem) > 0
def _contains_vowel(self, stem):
"""Returns True if stem contains a vowel, else False"""
for i in range(len(stem)):
if not self._is_consonant(stem, i):
return True
return False
def _ends_double_consonant(self, word):
"""Implements condition *d from the paper
Returns True if word ends with a double consonant
"""
return (
len(word) >= 2
and word[-1] == word[-2]
and self._is_consonant(word, len(word) - 1)
)
def _ends_cvc(self, word):
"""Implements condition *o from the paper
From the paper:
*o - the stem ends cvc, where the second c is not W, X or Y
(e.g. -WIL, -HOP).
"""
return (
len(word) >= 3
and self._is_consonant(word, len(word) - 3)
and not self._is_consonant(word, len(word) - 2)
and self._is_consonant(word, len(word) - 1)
and word[-1] not in ('w', 'x', 'y')
) or (
self.mode == self.NLTK_EXTENSIONS
and len(word) == 2
and not self._is_consonant(word, 0)
and self._is_consonant(word, 1)
)
def _replace_suffix(self, word, suffix, replacement):
"""Replaces `suffix` of `word` with `replacement"""
assert word.endswith(suffix), "Given word doesn't end with given suffix"
if suffix == '':
return word + replacement
else:
return word[: -len(suffix)] + replacement
def _apply_rule_list(self, word, rules):
"""Applies the first applicable suffix-removal rule to the word
Takes a word and a list of suffix-removal rules represented as
3-tuples, with the first element being the suffix to remove,
the second element being the string to replace it with, and the
final element being the condition for the rule to be applicable,
or None if the rule is unconditional.
"""
for rule in rules:
suffix, replacement, condition = rule
if suffix == '*d' and self._ends_double_consonant(word):
stem = word[:-2]
if condition is None or condition(stem):
return stem + replacement
else:
# Don't try any further rules
return word
if word.endswith(suffix):
stem = self._replace_suffix(word, suffix, '')
if condition is None or condition(stem):
return stem + replacement
else:
# Don't try any further rules
return word
return word
def _step1a(self, word):
"""Implements Step 1a from "An algorithm for suffix stripping"
From the paper:
SSES -> SS caresses -> caress
IES -> I ponies -> poni
ties -> ti
SS -> SS caress -> caress
S -> cats -> cat
"""
# this NLTK-only rule extends the original algorithm, so
# that 'flies'->'fli' but 'dies'->'die' etc
if self.mode == self.NLTK_EXTENSIONS:
if word.endswith('ies') and len(word) == 4:
return self._replace_suffix(word, 'ies', 'ie')
return self._apply_rule_list(
word,
[
('sses', 'ss', None), # SSES -> SS
('ies', 'i', None), # IES -> I
('ss', 'ss', None), # SS -> SS
('s', '', None), # S ->
],
)
def _step1b(self, word):
"""Implements Step 1b from "An algorithm for suffix stripping"
From the paper:
(m>0) EED -> EE feed -> feed
agreed -> agree
(*v*) ED -> plastered -> plaster
bled -> bled
(*v*) ING -> motoring -> motor
sing -> sing
If the second or third of the rules in Step 1b is successful,
the following is done:
AT -> ATE conflat(ed) -> conflate
BL -> BLE troubl(ed) -> trouble
IZ -> IZE siz(ed) -> size
(*d and not (*L or *S or *Z))
-> single letter
hopp(ing) -> hop
tann(ed) -> tan
fall(ing) -> fall
hiss(ing) -> hiss
fizz(ed) -> fizz
(m=1 and *o) -> E fail(ing) -> fail
fil(ing) -> file
The rule to map to a single letter causes the removal of one of
the double letter pair. The -E is put back on -AT, -BL and -IZ,
so that the suffixes -ATE, -BLE and -IZE can be recognised
later. This E may be removed in step 4.
"""
# this NLTK-only block extends the original algorithm, so that
# 'spied'->'spi' but 'died'->'die' etc
if self.mode == self.NLTK_EXTENSIONS:
if word.endswith('ied'):
if len(word) == 4:
return self._replace_suffix(word, 'ied', 'ie')
else:
return self._replace_suffix(word, 'ied', 'i')
# (m>0) EED -> EE
if word.endswith('eed'):
stem = self._replace_suffix(word, 'eed', '')
if self._measure(stem) > 0:
return stem + 'ee'
else:
return word
rule_2_or_3_succeeded = False
for suffix in ['ed', 'ing']:
if word.endswith(suffix):
intermediate_stem = self._replace_suffix(word, suffix, '')
if self._contains_vowel(intermediate_stem):
rule_2_or_3_succeeded = True
break
if not rule_2_or_3_succeeded:
return word
return self._apply_rule_list(
intermediate_stem,
[
('at', 'ate', None), # AT -> ATE
('bl', 'ble', None), # BL -> BLE
('iz', 'ize', None), # IZ -> IZE
# (*d and not (*L or *S or *Z))
# -> single letter
(
'*d',
intermediate_stem[-1],
lambda stem: intermediate_stem[-1] not in ('l', 's', 'z'),
),
# (m=1 and *o) -> E
(
'',
'e',
lambda stem: (self._measure(stem) == 1 and self._ends_cvc(stem)),
),
],
)
def _step1c(self, word):
"""Implements Step 1c from "An algorithm for suffix stripping"
From the paper:
Step 1c
(*v*) Y -> I happy -> happi
sky -> sky
"""
def nltk_condition(stem):
"""
This has been modified from the original Porter algorithm so
that y->i is only done when y is preceded by a consonant,
but not if the stem is only a single consonant, i.e.
(*c and not c) Y -> I
So 'happy' -> 'happi', but
'enjoy' -> 'enjoy' etc
This is a much better rule. Formerly 'enjoy'->'enjoi' and
'enjoyment'->'enjoy'. Step 1c is perhaps done too soon; but
with this modification that no longer really matters.
Also, the removal of the contains_vowel(z) condition means
that 'spy', 'fly', 'try' ... stem to 'spi', 'fli', 'tri' and
conflate with 'spied', 'tried', 'flies' ...
"""
return len(stem) > 1 and self._is_consonant(stem, len(stem) - 1)
def original_condition(stem):
return self._contains_vowel(stem)
return self._apply_rule_list(
word,
[
(
'y',
'i',
nltk_condition
if self.mode == self.NLTK_EXTENSIONS
else original_condition,
)
],
)
def _step2(self, word):
"""Implements Step 2 from "An algorithm for suffix stripping"
From the paper:
Step 2
(m>0) ATIONAL -> ATE relational -> relate
(m>0) TIONAL -> TION conditional -> condition
rational -> rational
(m>0) ENCI -> ENCE valenci -> valence
(m>0) ANCI -> ANCE hesitanci -> hesitance
(m>0) IZER -> IZE digitizer -> digitize
(m>0) ABLI -> ABLE conformabli -> conformable
(m>0) ALLI -> AL radicalli -> radical
(m>0) ENTLI -> ENT differentli -> different
(m>0) ELI -> E vileli - > vile
(m>0) OUSLI -> OUS analogousli -> analogous
(m>0) IZATION -> IZE vietnamization -> vietnamize
(m>0) ATION -> ATE predication -> predicate
(m>0) ATOR -> ATE operator -> operate
(m>0) ALISM -> AL feudalism -> feudal
(m>0) IVENESS -> IVE decisiveness -> decisive
(m>0) FULNESS -> FUL hopefulness -> hopeful
(m>0) OUSNESS -> OUS callousness -> callous
(m>0) ALITI -> AL formaliti -> formal
(m>0) IVITI -> IVE sensitiviti -> sensitive
(m>0) BILITI -> BLE sensibiliti -> sensible
"""
if self.mode == self.NLTK_EXTENSIONS:
# Instead of applying the ALLI -> AL rule after '(a)bli' per
# the published algorithm, instead we apply it first, and,
# if it succeeds, run the result through step2 again.
if word.endswith('alli') and self._has_positive_measure(
self._replace_suffix(word, 'alli', '')
):
return self._step2(self._replace_suffix(word, 'alli', 'al'))
bli_rule = ('bli', 'ble', self._has_positive_measure)
abli_rule = ('abli', 'able', self._has_positive_measure)
rules = [
('ational', 'ate', self._has_positive_measure),
('tional', 'tion', self._has_positive_measure),
('enci', 'ence', self._has_positive_measure),
('anci', 'ance', self._has_positive_measure),
('izer', 'ize', self._has_positive_measure),
abli_rule if self.mode == self.ORIGINAL_ALGORITHM else bli_rule,
('alli', 'al', self._has_positive_measure),
('entli', 'ent', self._has_positive_measure),
('eli', 'e', self._has_positive_measure),
('ousli', 'ous', self._has_positive_measure),
('ization', 'ize', self._has_positive_measure),
('ation', 'ate', self._has_positive_measure),
('ator', 'ate', self._has_positive_measure),
('alism', 'al', self._has_positive_measure),
('iveness', 'ive', self._has_positive_measure),
('fulness', 'ful', self._has_positive_measure),
('ousness', 'ous', self._has_positive_measure),
('aliti', 'al', self._has_positive_measure),
('iviti', 'ive', self._has_positive_measure),
('biliti', 'ble', self._has_positive_measure),
]
if self.mode == self.NLTK_EXTENSIONS:
rules.append(('fulli', 'ful', self._has_positive_measure))
# The 'l' of the 'logi' -> 'log' rule is put with the stem,
# so that short stems like 'geo' 'theo' etc work like
# 'archaeo' 'philo' etc.
rules.append(
("logi", "log", lambda stem: self._has_positive_measure(word[:-3]))
)
if self.mode == self.MARTIN_EXTENSIONS:
rules.append(("logi", "log", self._has_positive_measure))
return self._apply_rule_list(word, rules)
def _step3(self, word):
"""Implements Step 3 from "An algorithm for suffix stripping"
From the paper:
Step 3
(m>0) ICATE -> IC triplicate -> triplic
(m>0) ATIVE -> formative -> form
(m>0) ALIZE -> AL formalize -> formal
(m>0) ICITI -> IC electriciti -> electric
(m>0) ICAL -> IC electrical -> electric
(m>0) FUL -> hopeful -> hope
(m>0) NESS -> goodness -> good
"""
return self._apply_rule_list(
word,
[
('icate', 'ic', self._has_positive_measure),
('ative', '', self._has_positive_measure),
('alize', 'al', self._has_positive_measure),
('iciti', 'ic', self._has_positive_measure),
('ical', 'ic', self._has_positive_measure),
('ful', '', self._has_positive_measure),
('ness', '', self._has_positive_measure),
],
)
def _step4(self, word):
"""Implements Step 4 from "An algorithm for suffix stripping"
Step 4
(m>1) AL -> revival -> reviv
(m>1) ANCE -> allowance -> allow
(m>1) ENCE -> inference -> infer
(m>1) ER -> airliner -> airlin
(m>1) IC -> gyroscopic -> gyroscop
(m>1) ABLE -> adjustable -> adjust
(m>1) IBLE -> defensible -> defens
(m>1) ANT -> irritant -> irrit
(m>1) EMENT -> replacement -> replac
(m>1) MENT -> adjustment -> adjust
(m>1) ENT -> dependent -> depend
(m>1 and (*S or *T)) ION -> adoption -> adopt
(m>1) OU -> homologou -> homolog
(m>1) ISM -> communism -> commun
(m>1) ATE -> activate -> activ
(m>1) ITI -> angulariti -> angular
(m>1) OUS -> homologous -> homolog
(m>1) IVE -> effective -> effect
(m>1) IZE -> bowdlerize -> bowdler
The suffixes are now removed. All that remains is a little
tidying up.
"""
measure_gt_1 = lambda stem: self._measure(stem) > 1
return self._apply_rule_list(
word,
[
('al', '', measure_gt_1),
('ance', '', measure_gt_1),
('ence', '', measure_gt_1),
('er', '', measure_gt_1),
('ic', '', measure_gt_1),
('able', '', measure_gt_1),
('ible', '', measure_gt_1),
('ant', '', measure_gt_1),
('ement', '', measure_gt_1),
('ment', '', measure_gt_1),
('ent', '', measure_gt_1),
# (m>1 and (*S or *T)) ION ->
(
'ion',
'',
lambda stem: self._measure(stem) > 1 and stem[-1] in ('s', 't'),
),
('ou', '', measure_gt_1),
('ism', '', measure_gt_1),
('ate', '', measure_gt_1),
('iti', '', measure_gt_1),
('ous', '', measure_gt_1),
('ive', '', measure_gt_1),
('ize', '', measure_gt_1),
],
)
def _step5a(self, word):
"""Implements Step 5a from "An algorithm for suffix stripping"
From the paper:
Step 5a
(m>1) E -> probate -> probat
rate -> rate
(m=1 and not *o) E -> cease -> ceas
"""
# Note that Martin's test vocabulary and reference
# implementations are inconsistent in how they handle the case
# where two rules both refer to a suffix that matches the word
# to be stemmed, but only the condition of the second one is
# true.
# Earlier in step2b we had the rules:
# (m>0) EED -> EE
# (*v*) ED ->
# but the examples in the paper included "feed"->"feed", even
# though (*v*) is true for "fe" and therefore the second rule
# alone would map "feed"->"fe".
# However, in THIS case, we need to handle the consecutive rules
# differently and try both conditions (obviously; the second
# rule here would be redundant otherwise). Martin's paper makes
# no explicit mention of the inconsistency; you have to infer it
# from the examples.
# For this reason, we can't use _apply_rule_list here.
if word.endswith('e'):
stem = self._replace_suffix(word, 'e', '')
if self._measure(stem) > 1:
return stem
if self._measure(stem) == 1 and not self._ends_cvc(stem):
return stem
return word
def _step5b(self, word):
"""Implements Step 5a from "An algorithm for suffix stripping"
From the paper:
Step 5b
(m > 1 and *d and *L) -> single letter
controll -> control
roll -> roll
"""
return self._apply_rule_list(
word, [('ll', 'l', lambda stem: self._measure(word[:-1]) > 1)]
)
def stem(self, word):
stem = word.lower()
if self.mode == self.NLTK_EXTENSIONS and word in self.pool:
return self.pool[word]
if self.mode != self.ORIGINAL_ALGORITHM and len(word) <= 2:
# With this line, strings of length 1 or 2 don't go through
# the stemming process, although no mention is made of this
# in the published algorithm.
return word
stem = self._step1a(stem)
stem = self._step1b(stem)
stem = self._step1c(stem)
stem = self._step2(stem)
stem = self._step3(stem)
stem = self._step4(stem)
stem = self._step5a(stem)
stem = self._step5b(stem)
return stem
def __repr__(self):
return '<PorterStemmer>'
def demo():
"""
A demonstration of the porter stemmer on a sample from
the Penn Treebank corpus.
"""
from nltk.corpus import treebank
from nltk import stem
stemmer = stem.PorterStemmer()
orig = []
stemmed = []
for item in treebank.fileids()[:3]:
for (word, tag) in treebank.tagged_words(item):
orig.append(word)
stemmed.append(stemmer.stem(word))
# Convert the results to a string, and word-wrap them.
results = ' '.join(stemmed)
results = re.sub(r"(.{,70})\s", r'\1\n', results + ' ').rstrip()
# Convert the original to a string, and word wrap it.
original = ' '.join(orig)
original = re.sub(r"(.{,70})\s", r'\1\n', original + ' ').rstrip()
# Print the results.
print('-Original-'.center(70).replace(' ', '*').replace('-', ' '))
print(original)
print('-Results-'.center(70).replace(' ', '*').replace('-', ' '))
print(results)
print('*' * 70)