This was a non-issue which i have solved.
You would need to sort against a field that has not been analyzed, or at
least not tokenized (filters such as lowercase will still emit one token).
Technically, you can sort using a script that pulls the non-analyzed data
from the source, but it would be a big performance hit. Do not do it (I
should have not even mentioned it!)
If you still need the text to be analyzed for search, look into using
multi-fields, which is ideal in this scenario:
Note, your content will be to be reindexed.