Martin (1992: 228-9):
With additives, one useful rule of thumb is to check whether the second of the messages involved in the relation is potentially "branched". If so, the relation can be taken as external. Thus [4:159] can be read as external, because of the missing Subject and Finite elements; there is no possibility of paratactic ellipsis of this kind in [4:160], which can accordingly be taken as internal (this reading is reinforced by the clause complex initial position of and).
EXTERNAL ADDITIVE (paratactic ellipsis)[4:159] Such assumptions are embedded in language, learnt through language, and reinforced in language use.
INTERNAL ADDITIVE (no branching possible)[4:160] Language fixes a world that is so much more stable and coherent than that we actually see that it takes its place in our consciousness and becomes what we think we have seen.
And since normal perception works by constant feed-back, the gap between the real world and the actually constructed world is constantly being reduced, so that what we do 'see' tends to become what we can say.
 The contrast being construed here as external vs internal additive relations is, in SFL theory, the contrast between a structural additive relation (logical metafunction) and a cohesive additive relation (textual metafunction). The cohesive additive relation is not internal because it relates the experiential meanings realised by the clauses, not the interpersonal meanings.
 The clause complex initial position of the linker 'and' does not "reinforce" an internal reading of the additive relation. This is where structural conjunctions operate, even when functioning cohesively to mark a textual transition between two groups of messages.