Depending on the size of your input document set, why not creat a combination of documents with spelling alternatives then do some kind of state space search with a goal of like maximum entities recognized
--- Sent from VMware Boxer Original Quoted context has been removed as per policy
