Halliday & Matthiessen (1999: 20-1):
(iii) There is also a third kind of semogenic process which arises from the nature of the sign itself. Our "sign" is not the Saussurean sign: we are not talking about the relationship between a word and its phonological representation (between content and expression, in Hjelmsiev's terms). The relationship is within the content plane, between a meaning and a wording — the non-arbitrary relationship between the system of semantics and the system of lexicogrammar.
This process, then, takes the form of deconstructing the two components of the sign. How is this possible? This can happen because, once a 'pair' of this kind has come into being, each component takes on an existence of its own. To pursue the example of the complex 'participant ^ noun' above: the category of 'participant becomes detached from that of noun, so that we can have participants realised by other things than nouns, and nouns realising other things than participants.
We now have three meanings instead of one: participants constructed by nouns, as hitherto, but now contrasting (a) with participants constructed by something other than nouns, and (b) with nouns constructing something other than participants. (Of course, neither of these two entails the other; there might be just many to one, not many to many. To take this actual example, there are many cases of nouns realising things other than participants; but relatively few cases of participants being realised otherwise than by a noun.)