I'm failing to stem certain Arabic terms using the SnowballStemmer. Many terms are stemmed successfully but some terms cause an AttributeError to be raised. Please see below for a minimal example that fails on the term 'from'.
(anaconda2-4.4.0) richard-balmer-macbook:~ richardbalmer$ pip freeze | grep nltk
nltk==3.2.5
(anaconda2-4.4.0) richard-balmer-macbook:~ richardbalmer$ ipython
Python 2.7.13 |Anaconda custom (x86_64)| (default, Dec 20 2016, 23:05:08)
Type "copyright", "credits" or "license" for more information.
IPython 5.3.0 -- An enhanced Interactive Python.
? -> Introduction and overview of IPython's features.
%quickref -> Quick reference.
help -> Python's own help system.
object? -> Details about 'object', use 'object??' for extra details.
In [1]: from nltk.stem.snowball import SnowballStemmer
In [2]: stemmer = SnowballStemmer('arabic')
In [3]: stemmer.stem(u'تسدد')
Out[3]: u'\u062a\u0633\u062f\u062f'
In [4]: stemmer.stem(u'من')
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
<ipython-input-4-ffa733106049> in <module>()
----> 1 stemmer.stem(u'من')
/Users/richardbalmer/.pyenv/versions/anaconda2-4.4.0/lib/python2.7/site-packages/nltk/stem/snowball.pyc in stem(self, word)
762 modified_word = self.__Suffix_Verb_Step2b(modified_word)
763 if not self.suffix_verb_step2b_success:
--> 764 modified_word = self.__Suffix_Verb_Step2a(modified_word)
765 if self.is_noun:
766 modified_word = self.__Suffix_Noun_Step2c2(modified_word)
/Users/richardbalmer/.pyenv/versions/anaconda2-4.4.0/lib/python2.7/site-packages/nltk/stem/snowball.pyc in __Suffix_Verb_Step2a(self, token)
533 break
534
--> 535 if suffix in self.__conjugation_suffix_verb_present and len(token) > 5:
536 token = token[:-2] # present
537 self.suffix_verb_step2a_success = True
AttributeError: 'ArabicStemmer' object has no attribute '_ArabicStemmer__conjugation_suffix_verb_present'
@richbalmer Thanks for reporting the issue.
@LBenzahia Could you help to look into this? Thanks in advance!
Hi @richbalmer thank you for reporting, First word 'تسدد' is the best possible stem because Snowball arabic stemmer based on light stemming algorithm deals with prefixes/suffixes, if you are looking for the root of "تسدد" you can use ISRI (root-based stemmer/deep stemming), The second word 'من' is a stop word, you should use stop word filter before start using Snowball ArabicStemmer, Also this stemmer doesn't deal with the case when the word have 2 letters.
Anyways, I've fixed the problem in this PR #1856.
Thank you again !
@LBenzahia thanks for looking into this so quickly! I'm getting:
File "/Users/richardbalmer/src/nltk/nltk/stem/util.py", line 24
arabic_stopwords = ['إذ',
^
SyntaxError: Non-ASCII character '\xd8' in file /Users/richardbalmer/src/nltk/nltk/stem/util.py on line 24, but no encoding declared; see http://python.org/dev/peps/pep-0263/ for details
Which also appears to be causing the tests to fail on Jenkins (https://nltk.ci.cloudbees.com/job/pull_request_tests/454/TOXENV=py27-jenkins,jdk=jdk8latestOnlineInstall/testReport/nose.failure/Failure/runTest/). I think all you need to do is put # -*- coding: utf-8 -*-
at the top of stem/util.py
.
Also, after fixing that locally I get a UnicodeWarning:
/Users/richardbalmer/src/nltk/nltk/stem/snowball.py:748: UnicodeWarning: Unicode equal comparison failed to convert both arguments to Unicode - interpreting them as being unequal
if word in arabic_stopwords:
It might be worth making those stopwords unicode strings.
Other than that it looks like your fix works nicely for me - thanks again!
p.s. One other suggestion: testing set inclusion is quite a lot faster than list inclusion, so it might be worth making that stopword list a set instead.
@richbalmer are you using python2.7 ? ,
It might be worth making those stopwords unicode strings.
done for python2.7 , test it again and tell me,It works fine for me. i've updated the PR
Yup I'm using 2.7. Looking good @LBenzahia - thanks again!
Still having the error :
AttributeError: 'ArabicStemmer' object has no attribute '_ArabicStemmer__conjugation_suffix_verb_present'
I'm using python 3
@NouraAls solved in PR
Most helpful comment
@richbalmer Thanks for reporting the issue.
@LBenzahia Could you help to look into this? Thanks in advance!