Stripping proper nouns from text

Question

I have a df with several thousand rows of text data. I'm using spaCy to do some NLP on a single column of that df and and am trying to remove proper nouns, stop words, and punctuation from my text data using the following:

tokens = []
lemma = []
pos = []

for doc in nlp.pipe(df['TIP_all_txt'].astype('unicode').values, batch_size=9845,
                        n_threads=3):
    if doc.is_parsed:
        tokens.append([n.text for n in doc if not n.is_punct and not n.is_stop and not n.is_space and not n.is_propn])
        lemma.append([n.lemma_ for n in doc if not n.is_punct and not n.is_stop and not n.is_space and not n.is_propn])
        pos.append([n.pos_ for n in doc if not n.is_punct and not n.is_stop and not n.is_space and not n.is_propn])
    else:
        tokens.append(None)
        lemma.append(None)
        pos.append(None)

df['s_tokens_all_txt'] = tokens
df['s_lemmas_all_txt'] = lemma
df['s_pos_all_txt'] = pos

df.head()

But I get this error and I'm not sure why:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-34-73578fd46847> in <module>()
      6                         n_threads=3):
      7     if doc.is_parsed:
----> 8         tokens.append([n.text for n in doc if not n.is_punct and not n.is_stop and not n.is_space and not n.is_propn])
      9         lemma.append([n.lemma_ for n in doc if not n.is_punct and not n.is_stop and not n.is_space and not n.is_propn])
     10         pos.append([n.pos_ for n in doc if not n.is_punct and not n.is_stop and not n.is_space and not n.is_propn])

<ipython-input-34-73578fd46847> in <listcomp>(.0)
      6                         n_threads=3):
      7     if doc.is_parsed:
----> 8         tokens.append([n.text for n in doc if not n.is_punct and not n.is_stop and not n.is_space and not n.is_propn])
      9         lemma.append([n.lemma_ for n in doc if not n.is_punct and not n.is_stop and not n.is_space and not n.is_propn])
     10         pos.append([n.pos_ for n in doc if not n.is_punct and not n.is_stop and not n.is_space and not n.is_propn])

AttributeError: 'spacy.tokens.token.Token' object has no attribute 'is_propn'

If I take out the not n.is_propn the code runs as expected. I've googled around and read the spaCy documentation, but haven't been able to find an answer thus far.

Answer 1

I don't see is_propn attribute available on the Token object.

I think you should be checking the Part of Speech type to be PROPN (reference):

from spacy.parts_of_speech import PROPN

def is_proper_noun(token):
    if token.doc.is_tagged is False:  # check if the document was POS-tagged
        raise ValueError('token is not POS-tagged')

    return token.pos == PROPN

Answer 2

Adding on to @alecxe answer.

There's no need to

• populate all the rows of dataframe at one go.
• get separate tokens, lemmas and pos lists when populating the dataframe.

You can try:

df = pd.DataFrame(columns=['tokens', 'lemmas', 'pos'])

annotated_docs = nlp.pipe(df['TIP_all_txt'].astype('unicode').values,
                          batch_size=9845, n_threads=3)

for doc in annotated_docs:
    if doc.is_parsed:
        # Remove the tokens that you don't want.
        tokens, lemmas, pos = zip(*[(tok.text, tok.lemma_, tok.pos_) 
                                    for tok in doc if not
                                    (tok.is_punct or tok.is_stop 
                                     or tok.is_space or is_proper_noun(tok) )
                                   ]
                                  )
        # Populate the DataFrame.
        df.append({'tokens':tokens, 'lemmas':lemmas, 'pos':pos})

And here's a neater pandas trick from how to split column of tuples in pandas dataframe? but the dataframe will take up more memory:

df = pd.DataFrame(columns=['Tokens'])

annotated_docs = nlp.pipe(df['TIP_all_txt'].astype('unicode').values,
                          batch_size=9845, n_threads=3)

for doc in annotated_docs:
    if doc.is_parsed:
        # Remove the tokens that you don't want.
        df.append([(tok.text, tok.lemma_, tok.pos_) 
                    for tok in doc if not
                    (tok.is_punct or tok.is_stop 
                     or tok.is_space or is_proper_noun(tok) )
                   ]
                  )

df[['tokens', 'lemmas', 'pos']] = df['Tokens'].apply(pd.Series)