CHILDESCorpusReader: multiple stems correspond to their source token; keep suffix PoS info

[jacksonllee]

Fixes for _get_words in the class CHILDESCorpusReader

After some standard code like the following is run:

>>> import nltk
>>> corpus_root = nltk.data.find('corpora/childes/data-xml/Eng-NA-MOR/')
>>> from nltk.corpus.reader import CHILDESCorpusReader
>>> valian = CHILDESCorpusReader(corpus_root, "Valian/.*.xml")

The corpus reader does not seem to handle word tokens with multiple stems properly (note "Lastname's" below):

>>> valian.words('Valian/01a.xml')[:5]
['at', 'Parent', "Lastname's", 'house', 'with']
>>> valian.words('Valian/01a.xml', stem=True)[:5]
['at', 'Parent', 'Lastname', 's', 'house']
>>> valian.tagged_words('Valian/01a.xml')[:5]
[('at', 'prep'), ('Parent', 'n:prop'), ("Lastname's", 'n:prop'), ('house', 'n'), ('with', 'prep')]
>>> valian.tagged_words('Valian/01a.xml', stem=True)[:5]
[('at', 'prep'), ('Parent', 'n:prop'), ('Lastname', 'n:prop'), 's', ('house', 'n')]

Issues:

• When stem is True, stems are all treated as distinct strings, making it difficult or impossible to align which stems correspond to which words.
• If the stem is treated as a suffix in the XML data, then its PoS tag info is lost.

Results after the fixes are implemented:

>>> valian.words('Valian/01a.xml')[:5]
['at', 'Parent', "Lastname's", 'house', 'with']
>>> valian.words('Valian/01a.xml', stem=True)[:5]
['at', 'Parent', 'Lastname~s', 'house', 'with']
>>> valian.tagged_words('Valian/01a.xml')[:5]
[('at', 'prep'), ('Parent', 'n:prop'), ("Lastname's", 'n:prop~poss'), ('house', 'n'), ('with', 'prep')]
>>> valian.tagged_words('Valian/01a.xml', stem=True)[:5]
[('at', 'prep'), ('Parent', 'n:prop'), ('Lastname~s', 'n:prop~poss'), ('house', 'n'), ('with', 'prep')]

(~ is used as the separator for multiple "stems" of a word, following the CHILDES convention.)

[stevenbird]

@tomonori-nagano or @alexisdimi – would you please review this?

[stevenbird]

These changes appear reasonable, and in the absence of any objection from @tomonori-nagano or @alexisdimi I'm merging them