Metadata is linked data which directly describes and/or structures data elements in a specific context.
Metadata may be linked to data either internally-- i.e., within the same data set-- or externally.
A specific instance of metadata may be purely descriptive or purely structural on a specific hierarchical level; however, it may also perform both functions. For example, we often descriptively emphasize text in a document by assigning italics or bold font. We may later perform a batch process upon all emphasized elements, such as changing the font size.
Descriptive metadata links classifying and/or dimensional descriptions to data, or (secondarily) evaluates linked descriptions.
Structural metadata groups data elements, or sequences of elements, into subforms within a format. Structural metadata enables processes, such as display formatting and the assignment of descriptive metadata, to be assigned uniformly to all elements within identified subsets.
About automated generation (creation) of metadata: Some types of metadata can be automatically generated by algorithms. That’s a subject with great potential, as well as potentially dehumanizing risks. MOT emphatically prioritizes metadata and linked data which agents intentionally create via direct personal inputs. This author (Greg) believes that the current potentials of directly created metadata are tragically overlooked and unrealized.