Halliday & Matthiessen (1999: ix-x):
… we have tried to represent language as the resource whereby the human species, and each individual member of that species, constructs the functioning mental map of their phenomenal world: of their experience of process, both what goes on out there and what goes on in the realms of their own consciousness.
It seems to us that our dialogue is relevant to current debates in cognitive science. In one sense, we are offering it as an alternative to mainstream currents in this area, since we are saying that cognition "is" (that is, can most profitably be modelled as) not thinking but meaning: the "mental" map is in fact a semiotic map, and "cognition" is just a way of talking about language. In modelling knowledge as meaning, we are treating it as a linguistic construct: hence, as something that is construed in the lexicogrammar. Instead of explaining language by reference to cognitive processes, we explain cognition by reference to linguistic processes. But at the same time this is an "alternative" only if it is assumed that the "cognitive" approach is in some sense natural, or unmarked. It seems to us that current approaches to neural networks, "connectionist" models and the like, are in fact more compatible with a semantic approach, where "understanding" something is transforming it into meaning, and to "know" is to have performed that transformation.