The plain text parser shouldn't use regular expressions to detect special characters

While the improvement should be safe, I'm not applying it to LTS branches for now as the regular expression was exposed as internal API and the change doesn't seem important enough to risk any breakage.