(De-)constructing evidentiality