Halliday & Matthiessen (1999: 271-2):
Metaphors are dangerous, however, they have too much power, and grammatical metaphor is no different in this respect. Because it leaves the relations within a figure almost totally inexplicit, this demands that they should be in some sense already in place. In the typical rhetorical context for the highly favoured 'backgrounding' type, as we have seen, the configurational relations have been established in the preceding discourse: cf. ... if one takes alcohol one's brain rapidly becomes dull. Alcohol's rapid dulling effect on the brain... Here by the time we reach the metaphor they are already in place: we know that rapid dulling effect means 'causes ... rapidly to become dull', not any of the other things it might mean such as 'has an effect which soon becomes dull, or blunted'.
But this is an idealised example, constructed for the purpose. Usually the configurational pattern will have been built up over long stretches of the text, or (especially if it is a technical form of discourse) over a great variety of different texts — for example, a series of textbooks used in teaching a science subject throughout a school. Very often the learner has to construct the configurational relations from various sources without their being made fully explicit in any one place; and in the limiting (but by no means unusual) case they have never been made explicit at all, so that the figure has to be construed from the metaphor — a very difficult task indeed.