Is there any real value in annotating Sub-type, Role and Attributes?

On this forum (and now on stack overflow) there is a discussion on the use of Role Definitions and Attributes relating to NLU (

The general answer was that NLU did not expose Role or Attributes. What is not clear anywhere is WHY we bother to annotate them in the first place. Thus:
1. Are Role and Attribute used to assist in building a more accurate machine model?
2. Are Role and Attribute relevent when deploying a custom model to other API’s such as Watson Discovery?

If the answer to each of the above is ‘No’. then it could svae us all a vast amount of time in our annotations….?