What does the parameter 'clean_up_tokenization_spaces' do in the tokenizer.decode function?

[ivnle]

As titled. The documentation is a bit vague.

Whether or not to clean up the tokenization spaces.