The work presented in this paper aims at covering several domains: hypervideo modelling, document annotation, and practices sharing and emergence. It is based on the Advene project, providing a model and a prototype for creating, rendering and sharing annotations of audiovisual documents. After a presentation of the notion of multi-structurality in documents and a definition of hypervideos, we present the original Advene model. We then discuss some limitations observed in our model, and introduce a new model for hypervideos. We finally discuss related work with respect to our model, and raise the problem of the emergence of semantics in videos and hypervideos.