Just to be clear on this issue, I think the intention here is for the UC where the user explicitly marks text to create triple statements, i.e., bottom up approach to lift the document. This is different than the two other possibilities (and we can create separate issues for them):
Automated RDFa embeds e.g., when a sub-section is created, dokieli takes care of that by creating a part-of relationship with the parent section. Or any part of the interface in which the RDFa is deterministic based on the type of text that's entered e.g., author profile, references, section heading.
Top down approach where a semi-automated process can be used to suggest and populate possible properties/classes for verbs and entities.
Just to be clear on this issue, I think the intention here is for the UC where the user explicitly marks text to create triple statements, i.e., bottom up approach to lift the document. This is different than the two other possibilities (and we can create separate issues for them):