Merge takes two elements and combines them into a set. Say it takes a DP "Luke" and a V "run". We can then define a VP by going merge(DP,V) = {DP,V} "Luke run". But since sets are un-ordered, {DP,V} = {V,DP}. So what's determining the fact that {V,DP} = "Luke run" and not "run Luke"?
(note: assuming the vp-internal subject hypothesis, and that this will be merged with TP later on).
Because there is structural asymmetry of some sort in the resulting structure. I don't think you ever get to linearize just the {V, DP}. Instead, be it phase-size or complete CP structure, the linearization is sensitive to the resulting hierarchical structure.
There's a few accounts, probably the most abstract (and famous) being Richie Kayne's Antisymmetry (1994). It states that a node higher in the structure will linearly precede the one lower in the structure (Linear Correspondence Axiom). It is X-bar style, and basically gives you spec-head-complement order (but it's not dependent on X-bar schema really, i.e. can be carried over to MP style bare structure).
Most versions of minimalism assume linearization constraints that are independent of the structure building operation. This becomes obvious when you see that different languages have different V S O orders. Linearization is usually taken to be a more language specific thing.
There have been several syntactic theories based on unordered constituents, among them Relational Grammar (with Arc Pair Grammar) and GPSG. The idea is that tree sisters are fundamentally without any intrinsic order, but that eventually, at a superficial level, for certain languages and certain constituents, the order gets fixed. I think that Gerald Sanders was the first to suggest such an explanation for the generalization in Ross's Gapping and the Order of Constituents.
I know beans about "merge", but maybe it is that old idea about separating ordering principles from constituency principles cropping up again.
Merge does not directly produce surface utterances. Fixing linear order can be treated as phonological (a terrible misinterpretation of phonology in my opinion), and there is also Move.
