I have a few intents in my setup, such as affirm, deny, greet, and other my work specific intents. I recently added capability to handle fallback. Sometimes, I see that nlu_fallback gets higher match compared to other intents such as deny, affirm, even when I enter the exact same words as example in nlu training.Example: I entered ‘hello’, nlu_fallback got 0.67 and greet got 0.33. I fixed these issue, by adding more examples to those intents. My question: is this expected behaviour, what are the common practices to handle this problem.