Halliday & Matthiessen (1999: 555-6):
Synchronically (that is, viewed synoptically in this way as a meaning potential) a language is, as we have said, a probabilistic system: if we say that, in the grammar, there is a system of primary tenses past/present/future, we assume the rider 'with a certain probability attached to them'. But we do not, of course, speak or write with one grammatical system at a time. Systems intersect with each other simultaneously (we choose tense along with voice, polarity, mood, transitivity and so on), and they follow each other in linear succession (we choose tense in clause 1, again in clause 2, again in clause 3 and so on). Each instance has its environment, both of previous instances, and of simultaneous instances of systems with their own sets of probabilities.
We shall not attempt to discuss this issue here; except to refer briefly to what is one important aspect of indeterminacy, namely partial association between systems. We model the grammar as if each of these choices was independent that the choice of tense, say, is not affected by the simultaneous choice of mood. This may, in fact, be so; but when choice is made in two systems simultaneously, such that each serves as environment for the other, there is often a conditioning effect on the probabilities. This may be an indication of a change in progress, or it may be a stable feature of the overall system.