This paper considers situations for which detailed metadata descriptions of learning resources are necessary, and focuses on human generation of such metadata. It describes a model which facilitates human production of good quality metadata by the development and use of structured vocabularies. Using examples, this model is applied to single and multiple communities of metadata creators. The approach for transferring vocabularies across communities is related to similar work on the use of ontologies to support the development of the semantic web. Notable conclusions from this work are the need to encourage collaboration between the metadata specialists, content authors and system designers, and the scope for using accurate and consistent metadata created for one context in another context by producing descriptions of the relationships between those contexts.