Halliday & Matthiessen (1999: 237-8):
Why then in our present interpretation have we to recognise two parts, one a lexicogrammar and one a semantics? Because the system continued to evolve beyond that point [congruent construals], enriching itself (i.e. engendering a richer model of experience) by forcing apart the two ‘facets’ of the sign so that each could take on a new partner — sequences could be realised by other things than clause complexes, processes could be realised by other things than verbs, and so on. […] It is this step that gives rise to grammatical metaphor. When a sequence is realised as a clause complex, or a process as a verb, this is congruent: it is the clause complex, and the verb, in the function in which it evolved. When a sequence is realised as something other than a clause complex, or a process as something other than a verb, this is metaphorical. Some other grammatical unit is supplanting them in these functions.