Help with finding "Parent" word of exceptions/contractions during tokenization

[irowberry]

I'm having an issue with the tokenization of contractions in English using spaCy. I know the normal behavior for a sentence like
"I've wondered about how she's doing." results in [I, 've, wondered, about, how, she, 's, doing, .]

However, I want to be able to check the tokens and see that [I, 've] both came from the word I've, likewise that [she, 's] came from she's.
Is there an attribute that already does this that I haven't seen? I've tried looking at the custom exceptions, but it only allows changing the attributes of ORTH and NORM.

I also saw that nlp.tokenizer.explain(text) returns
[('SPECIAL-1', 'I'), ('SPECIAL-2', "'ve"), ('TOKEN', 'wondered'), ('TOKEN', 'about'), ('TOKEN', 'how'), ('SPECIAL-1', 'she'), ('SPECIAL-2', "'s"), ('TOKEN', 'doing'), ('SUFFIX', '.')]

So I know that during the tokenization process, it's identifying exceptions, I just need to access them as an attribute of the tokens. Any help would be appreciated.

[adrianeboyd]

The main Tokenizer used in the pipeline doesn't track this information at all (it would affect the performance), so there's no way to access/store this.

Tokenizer.explain is a much slower implementation of the same algorithm just for debugging purposes.