It may be worth changing the normalization process to strip these characters in pairs (only if there's a non "letter" at the start and end). What was the example case that triggered this?
The problem with that is there are a bunch of games & other media which put what looks like year in their actual title but aren't released in that year.
That format is also heavily used by Wikipedia and IMDB
I have changed this to only strip when the start and end has non-letters, which should resolve this case and hopefully most of the other incorrect cases.