In this section only a few very general recommendations can be given, as the decision how to annotate a resource or which annotation to use highly depends on the data, annotators, available tools and on the task at hand.
The most general but very important recommendation is to provide metadata for each annotation of a resource. For already existing resources, not all of the relevant information might still be known or can be reconstructed, but as much as possible of the following information should be provided together with the annotation:
information on how the annotation was created, i.e., manually, semi-automatically or automatically,
the annotation guidelines, in case of a manual or semi-automatic annotation,
the information which tools and tool versions have been applied in an automatic or semi-automatic annotation,
the tagset and the explanations of the tags, preferrably linked to a data category registry such as ISOcat,
information about the quality of the annotations, e.g. inter-annotator agreement or scores of automatic processing tools, when applied to gold standards,
explicit dependencies, e.g., if an annotation relies on other annotation layers or a specific version of the primary data,
version information that identifies the existing annotation and allows for differentiation of versions
When creating a new set of annotations, it is also recommended to opt for sustainability and reusability, for example by choosing a state-of-the-art representation format and a common encoding such as UTF-8 for textual data. It is also recommended to choose a fine-grained basic segmentation that fits the annotation but also allows for other annotations to build on the same segmentation layer.