str.translate gives TypeError - Translate takes one argument (2 given), worked in Python 2

Question

I have the following code

import nltk, os, json, csv, string, cPickle
from scipy.stats import scoreatpercentile

lmtzr = nltk.stem.wordnet.WordNetLemmatizer()

def sanitize(wordList): 
answer = [word.translate(None, string.punctuation) for word in wordList] 
answer = [lmtzr.lemmatize(word.lower()) for word in answer]
return answer

words = []
for filename in json_list:
    words.extend([sanitize(nltk.word_tokenize(' '.join([tweet['text'] 
                   for tweet in json.load(open(filename,READ))])))])

I've tested lines 2-4 in a separate testing.py file when I wrote

import nltk, os, json, csv, string, cPickle
from scipy.stats import scoreatpercentile

wordList= ['\'the', 'the', '"the']
print wordList
wordList2 = [word.translate(None, string.punctuation) for word in wordList]
print wordList2
answer = [lmtzr.lemmatize(word.lower()) for word in wordList2]
print answer

freq = nltk.FreqDist(wordList2)
print freq

and the command prompt returns ['the','the','the'], which is what I wanted (removing punctuation).

However, when I put the exact same code in a different file, python returns a TypeError stating that

File "foo.py", line 8, in <module>
  for tweet in json.load(open(filename, READ))])))])
File "foo.py", line 2, in sanitize
  answer = [word.translate(None, string.punctuation) for word in wordList]
TypeError: translate() takes exactly one argument (2 given)

json_list is a list of all the file paths (I printed and check that this list is valid). I'm confused on this TypeError because everything works perfectly fine when I'm just testing it in a different file.

Answer 1

If all you are looking to accomplish is to do the same thing you were doing in Python 2 in Python 3, here is what I was doing in Python 2.0 to throw away punctuation and numbers:

text = text.translate(None, string.punctuation)
text = text.translate(None, '1234567890')

Here is my Python 3.0 equivalent:

text = text.translate(str.maketrans('','',string.punctuation))
text = text.translate(str.maketrans('','','1234567890'))

Basically it says 'translate nothing to nothing' (first two parameters) and translate any punctuation or numbers to None (i.e. remove them).

Answer 2

I suspect your issue has to do with the differences between str.translate and unicode.translate (these are also the differences between str.translate on Python 2 versus Python 3). I suspect your original code is being sent unicode instances while your test code is using regular 8-bit str instances.

I don't suggest converting Unicode strings back to regular str instances, since unicode is a much better type for handling text data (and it is the future!). Instead, you should just adapt to the new unicode.translate syntax. With regular str.translate (on Python 2), you can pass an optional deletechars argument and the characters in it would be removed from the string. For unicode.translate (and str.translate on Python 3), the extra argument is no longer allowed, but translation table entries with None as their value will be deleted from the output.

To solve the problem you'll need to create an appropriate translation table. A translation table is a dictionary mapping from Unicode ordinals (that is, ints) to ordinals, strings or None. A helper function for making them exists in Python 2 as string.maketrans (and Python 3 as a method of the str type), but the Python 2 version of it doesn't handle the case we care about (putting None values into the table). You can build an appropriate dictionary yourself with something like {ord(c): None for c in string.punctuation}.