How to define special “untokenizable” words for nltk.word_tokenize

问题

I'm using nltk.word_tokenize for tokenizing some sentences which contain programming languages, frameworks, etc., which get incorrectly tokenized.

For example:

>>> tokenize.word_tokenize("I work with C#.")
['I', 'work', 'with', 'C', '#', '.']

Is there a way to enter a list of "exceptions" like this to the tokenizer? I already have compiled a list of all the things (languages, etc.) that I don't want to split.

回答1:

The Multi Word Expression Tokenizer should be what you need.

You add the list of exceptions as tuples and pass it the already tokenized sentences:

tokenizer = nltk.tokenize.MWETokenizer()
tokenizer.add_mwe(('C', '#'))
tokenizer.add_mwe(('F', '#'))
tokenizer.tokenize(['I', 'work', 'with', 'C', '#', '.'])
['I', 'work', 'with', 'C_#', '.']
tokenizer.tokenize(['I', 'work', 'with', 'F', '#', '.'])
['I', 'work', 'with', 'F_#', '.']

来源：https://stackoverflow.com/questions/45618528/how-to-define-special-untokenizable-words-for-nltk-word-tokenize