There is no one-to-one relation between referential terms and types of access to the referents (referring modes) in multimodal human-computer interaction. We propose a classification of referring modes implying sub-contexts. We describe in detail the nature of these contextual subsets and we show their importance for each type of referring action. Then we define a relation between terms and modes, and we deduce a list of disambiguation principles for the computation of referential terms in order to identify the correct referring mode, the correct sub-context, and the correct referent.