Halliday & Matthiessen (1999: 198):
But we also have to recognise that there is a systematic relationship among a set of terms such as bicycle, tandem, scooter, car, van, truck and bus. Here the form of wording gives no clue to any such relationship; all we have is an inventory of different words. Yet these also form some kind of a taxonomic set: they are all wheeled vehicles. We can have taxonomies of things without using the syntagmatic resources of the nominal group, simply by the device of naming — the organising principle is not syntagmatic but paradigmatic.But there is a difference. The paradigmatic strategy, that of inventing new names, typically construes sets of things which are systemically related but not in a relationship of strict taxonomy. This resource is typically associated with feature networks: that is, networks made up of systems of features, such that each lexical item (as the name of a thing) realises a certain combination of these features selected from different systems within the network — a particular clustering of values of systemic variables.