Perhaps it's considered "good enough" because users balk at performing NLP-style queries when they find they almost never work properly.