What technique would make Snips accurate at reading long strings?
If I had several training statements that looked like:
The [reader:name](boy) read "[title:book_title](A Connecticut Yankee in King Arthur's Court)". The [reader](girl) read "[title]The Scarlet Letter".
A chatbot that heard:
The man read “There and Back Again: A Hobbit’s Tale”
would likely return three “title” slots, containing “there”, “and Back Again”, “Hobbit’s Tale”.
Note what I did with the capitalization and dropping a few pieces, because this is what I’m actually seeing. Snips will split the book’s name between multiple slots, do random things to capitalization, and even drop random pieces.
Is there a technique or way to tag the training data so that Snips doesn’t get “creative”, and just returns the whole string in the same way it was passed in?