In order to train Bixby to correctly select values of variable lengths, simply provide Bixby with examples of such cases.
It is good practice to provide Bixby with example utterances where the value that needs to be collected as the input is the minimum required, maximum required, a few in the middle. Once Bixby sees examples of inputs with differing values, it will generalize that learning to capture inputs of variable lengths.
For example, suppose we have a capsule that needs to recognize hobbies that a user mentions and tell the user that Bixby likes to do those hobbies too. Since hobbies can be described with a single word such as 'hiking' or multiple words such as 'making pillow forts', we need to train Bixby to expect utterances of different lengths.
In order to accomplish this, we need to train Bixby to understand that hobbies can be of variable lengths.
Now that we have trained Bixby to understand utterances with different length inputs, we can attempt an input of a size that Bixby has not encountered in any of our utterance trainings to confirm that it will pick up any new hobbies regardless of the size of the description.
As expected, Bixby was able to accurate find the full input from the user's utterance despite it being a length that had not been explicitly trained by us.