* While fixing #17 this bug was introduced * The answer returned by the model has white-spaces in between letters of a single word, like 'apple' would be returned as 'a p p l e' * It cannot handle the cases where start token occurs after the end token and thus no answer is returned in such cases.