Provoke the NLTK part-of-speech tagger to report a plural proper noun

Question

Let's try out Python's renouned part-of-speech tagger in the nltk package.

import nltk
# You might also need to run nltk.download('maxent_treebank_pos_tagger') 
#  even after installing nltk

string = 'Buddy Billy went to the moon and came Back with several Vikings.'
nltk.pos_tag(nltk.word_tokenize(string))

This gives me

[('Buddy', 'NNP'), ('Billy', 'NNP'), ('went', 'VBD'), ('to', 'TO'), ('the', 'DT'), ('moon', 'NN'), ('and', 'CC'), ('came', 'VBD'), ('Back', 'NNP'), ('with', 'IN'), ('several', 'JJ'), ('Vikings', 'NNS'), ('.', '.')]

You can interpret the codes here. I'm slightly disappointed that 'Back' got categorized as a proper noun (NNP), although the confusion is understandable. I'm more upset that 'Vikings' got called a simple plural noun (NNS) instead of a plural proper noun (NNPS). Can anyone come up with a single example of a brief input that leads to at least one NNPS tag?

Sixteenth

Answer 1

There seems to be some problems with the tags in NLTK brown corpus that tags NNPS as NPS (Possibly the NLTK tagset is an updated/outdated tags that is different from https://www.ling.upenn.edu/courses/Fall_2003/ling001/penn_treebank_pos.html)

Here's an example of plural proper nouns:

>>> from nltk.corpus import brown
>>> for sent in brown.tagged_sents():
...     if any(pos for word, pos in sent if pos == 'NPS'):
...             print sent
...             break
... 
[(u'Georgia', u'NP'), (u'Republicans', u'NPS'), (u'are', u'BER'), (u'getting', u'VBG'), (u'strong', u'JJ'), (u'encouragement', u'NN'), (u'to', u'TO'), (u'enter', u'VB'), (u'a', u'AT'), (u'candidate', u'NN'), (u'in', u'IN'), (u'the', u'AT'), (u'1962', u'CD'), (u"governor's", u'NN$'), (u'race', u'NN'), (u',', u','), (u'a', u'AT'), (u'top', u'JJS'), (u'official', u'NN'), (u'said', u'VBD'), (u'Wednesday', u'NR'), (u'.', u'.')]

But if you tag with nltk.pos_tag, you'll get NNPS:

>>> for sent in brown.tagged_sents():
...     if any(pos for word, pos in sent if pos == 'NPS'):
...             print " ".join([word for word, pos in sent])
...             break
... 
Georgia Republicans are getting strong encouragement to enter a candidate in the 1962 governor's race , a top official said Wednesday .
>>> from nltk import pos_tag
>>> pos_tag("Georgia Republicans are getting strong encouragement to enter a candidate in the 1962 governor's race , a top official said Wednesday .".split())
[('Georgia', 'NNP'), ('Republicans', 'NNPS'), ('are', 'VBP'), ('getting', 'VBG'), ('strong', 'JJ'), ('encouragement', 'NN'), ('to', 'TO'), ('enter', 'VB'), ('a', 'DT'), ('candidate', 'NN'), ('in', 'IN'), ('the', 'DT'), ('1962', 'CD'), ("governor's", 'NNS'), ('race', 'NN'), (',', ','), ('a', 'DT'), ('top', 'JJ'), ('official', 'NN'), ('said', 'VBD'), ('Wednesday', 'NNP'), ('.', '.')]

Mowry

I verified that the Georgia Republicans are ... input successfully provokes the NNPS tag. – Sixteenth

To your point, "There seems to be some problems with the tags ...", I am not surprised that the tags returned by brown tagger (comp.leeds.ac.uk/ccalas/tagsets/brown.html) are different from the tags produced by nltk.pos_tag since the latter is based on the Penn Treebank corpus, which uses a completely different tagset (cs.nyu.edu/grishman/jet/guide/PennPOS.html) – Sixteenth