Halliday & Matthiessen (2014; 34):
Field, tenor and mode are thus sets of related variables, with ranges of contrasting values. Together they define a multi-dimensional semiotic space – the environment of meanings in which language, other semiotic systems and social systems operate. The combinations of field, tenor and mode values determine different uses of language — the different meanings that are at risk in a given type of situation. There are systematic correspondences between the contextual values and the meanings that are at risk in the contexts defined by these values. As Halliday (1978) suggested, field values resonate with ideational meanings, tenor values resonate with interpersonal meanings, and mode values resonate with textual meanings (see also Halliday & Hasan, 1985: 26).* In other words, the correspondences between context and language are based on the functional organisation of both orders of meaning.
* We use the term ‘resonate with’ because the relationship is not a one-way causal relationship, but rather a two-way realisational relationship (cf. Jay Lemke’s, 1984, notion of metaredundancy, discussed in Halliday, 1992). Contextual values influence linguistic choices but are also influenced by them.
 Note that this is the culture as meaning— as a semiotic system — not language as meaning (i.e. semantics). Confusion in this regard led to Martin's (1992) misconstrual of diatypic varieties of language (register, genre) as context, and subsequently, to the inability of his students to tell the difference.
 Note the contradiction inherent in Halliday's reviser, Matthiessen, claiming, on the one hand, that the relation between context and language is not causal, and on the other hand, that context 'determines' uses of language and that contextual values and linguistic choices 'influence' each other; 'influence' means to have an effect upon, and an 'effect' is the consequence of a cause. As the expression 'two-way realisational relation' suggests, context and language are construed together, since they are in a relation of identity at two levels of symbolic abstraction. As Halliday & Matthiessen (1999: 354) say:
we may model the relationship in one direction or the other, but text and context are construed together.
Causation can be more coherently associated with the process of instantiation, since the selection of features, and the activation of their realisation statements, can influence the selection of others, given the probabilistic nature of the system.
 The masturbatory metaphor 'meanings at risk' appears to have been introduced into SFL by Martin (1992: 464, 476, 477, 486-7, 489, 502, 508, 528, 531, 536, 538), drawn from a notion in Barthes (1977: 102):
However minimal its importance, a sequence, since it is made up of a small number of nuclei (that is to say, in fact of 'dispatchers'), always involves moments of risk and it is this that justifies analysing it. It might seem futile to constitute into a sequence the logical succession of trifling acts which go to make up the offer of a cigarette (offering, accepting, smoking, lighting), but precisely at every one of these points, an alternative — and hence a freedom of meaning — is possible. … A sequence is thus, one can say, a threatened logical unit, this being its justification a minimo.
In terms of SFL theory, Barthes simply means that the speaker is always free to instantiate a different option during logogenesis. The "risk" is to the feature and it is the "risk" of not being selected(!).