The simplest way of reducing valence, if you want to call it that, is by deleting the object:
u sótama i tuho ka lina
DEF.PL soldier VP IMPF destroy DEF city
"the soldiers destroyed the city"
u sótama i ma tuho Ø
DEF.PL soldier VP IMPF destroy Ø
"the soldiers were destroying [things]"
Valence in Koa is pretty free compared to English. You've got a verb, and the subject position is required (unless it was previously stated or can be assumed, like i kuma he tana "it's hot today"), but beyond that you can kind of free-form the participants in the way you choose and anything omitted either (1) becomes indefinite or (2) is recovered anaphorically. This latter reading could therefore yield "the soldiers were destroying it" as an alternative interpretation to the previous sentence, if "it" were some referent already established in discourse.
It turns out that at some point we earmarked pa as a suffix to indicate an indefinite object, which means we could resolve the ambiguity, if necessary, by saying
u sótama i ma túhopa
DEF.PL soldier VP IMPF destroy-INDEF
"the soldiers were destroying things, the soldiers were going around destroying"
That's clear enough, but it's not 100% obvious what that pa actually is...is it some kind of pronoun? I thought so at one point. Or is it more derivational, like "do X here and there/haphazardly/with undefined results?" These days I think hi can actually fulfill the role I'd envisioned for pa back then (see below) so I'll leave it as a suffix for the moment with no further pretensions toward independent status.
I was surprised to discover last night that I thought I already understood this once upon a time, but I'm not at all satisfied with a lot of those conclusions, and I'll start from scratch here.
The most neutral way of indicating that a verb's object is identical with its subject is via the particle hi. This is genuine valence reduction in that the verb's object slot for the repeated participant disappears when this particle is used.
ni hi loha
1SG REFL love
"I love myself"
ni hi ana ti kánute
1SG REFL give this injury
"I gave myself this injury"
Note that ditransitive verbs can become reflexive with respect to either the direct or indirect object.
In a somewhat more marked way, the valence of the verb can be left alone but a reflexive construction used for the object:
ni loha niími
1SG love 1SG-self
"I love myself"
ni ana ti kánute la niími
1SG give this injury DAT 1SG-self
"I gave this injury to myself"
There's an added sense of contrast or emphasis here, so depending on the context this structure might tend to show up with focalization as well. Combining the two strategies is also possible with a serial verb, with even greater centering of the reflexiveness:
ni hi loha i imi
1SG REFL love VP self
"I love myself"
Semantic reflexiveness can come in via an incorporated object as well:
ni loha imi
1SG love self
"I love myself," "I self-love"
This gives a slightly different meaning, though: we're implying that self-love is a thing, and that the speaker practices it.
There's one final undecided point, though, of just how pronominal hi is allowed to get within VP arguments: in other words, can/should we say this instead for "I love myself"?
?ni loha hiími
1SG love REFL-self
"I love myself"
On purely aesthetic grounds I do greatly prefer it when languages have a reflexive pronoun (e.g. Polish golę się, lit. "I shave oneself" vs Spanish me afeito, lit. "I shave me") but this does potentially introduce a host of complications. Which of these is "I love my cat?"
ni loha ka sene ni
1SG love DEF cat 1SG
ni loha ka sene hi
1SG love DEF cat REFL
At the moment the sentence with hi is feeling pretty weird to me so I think we can nix that unless some emergent circumstances demand it in the future.
In reflexive verbs the totality of the action applies to all subjects individually. Reciprocals, in which the action travels from each of a plurality of subjects to each other subject, are formally very similar. The least marked structure is actually the same as for reflexives, and context disambiguates 95% of the time:
nu hi loha
1PL REFL love
"we love each other" OR "we love ourselves"
The possibilities for overt disambiguation exactly parallel those for reflexives with the same minor differences in sense, using the word kahi "reciprocal":
nu loha nukahi
1PL love 1PL-reciprocal
"we love each other," lit. "we love our reciprocal ones"
nu loha kahi
1PL love reciprocal
"idem," lit. "we love reciprocally"
nu hi loha i kahi
1PL REFL love VP reciprocal
"we love each other [reciprocally]"
When a transitive verb has a general, non-referential object, the most neutral structure is to incorporate it: in Koa terms, to omit the article so that it modifies the verb rather than standing on its own as an independent object. This reduces the verb's valence by one slot and is explored pretty thoroughly (with surprisingly few differences from modern usage) in this post from 2011, so I'll let that explanation stand in the interest of brevity. This gives us clauses like
u sótama i ma tuho lina
DEF.PL soldier VP IMPF destroy city
"the soldiers were destroying cities," "the soldiers were engaging in city-wrecking"
Passive/Middle Voice and Argument Shifts
Though traditionally labeled "passive," the particle pa affects argument structure in a manner that's pretty unlike the passive voice in IE languages. pa "promotes" a verbal object to subject position, but it also optionally "demotes" the subject to postverbal -- i.e. object -- position. The structure thereby created allows the agent/experiencer can be elided in a way it cannot in preverbal position, and valence is not technically decreased (depending on how you want to define object deletion).
As such, pa verbs are able to say something about their patient without inferring anything at all about their agent or experiencer, or even implying their existence. In this way it's more similar to a middle voice:
to puku i pa tino mo vake
that costume VP PASS take.off ADV difficult
"that costume 'takes off' with difficulty," "that costume is hard to take off"
A more passive reading is possible as well:
ka lina i pa tuho he to kesa
DEF city VP PASS destroy TIME that summer
"the city was destroyed that summer"
The wild difference from IE passives and middles, and one that I would consider the most Loglanesque operation in the whole of Koa if it hadn't been rescued by Bantu, is that there is still an argument slot for the notional subject after the verb:
ka lina i pa tuho u sótama
DEF city VP PASS destroy DEF.PL soldier
"the city was destroyed by the soldiers"
For speakers of case languages, though, there's another way of experiencing this sentence that has nothing to do with passivity. It's hard to get at in English except in the most poetic of contexts; maybe if you start with "bright waxed the moon" (a predicate, not an object, I know) and then parse "the city destroyed the soldiers" the same way you can sort of feel it? But if you've got an accusative like Latin, you can just say
urbem dēlēvērunt mīlitēs
city-SG.ACC destroy-PERF-3PL soldier-PL.ACC
"the soldiers destroyed the city"
The pragmatic or stylistic circumstances where one might make this choice in Koa will still need to be established, probably in the very long term. In the mean time, we can also make the agent oblique with ci (instrumental) and give the structure more of a traditional passive feel:
ka lina i pa tuho ci u sótama
DEF city VP PASS destroy INSTR DEF.PL soldier
"the city was destroyed by [the actions of] the soldiers"
Passive Agent Incorporation
In the same way that a general predicate in the object position of an active can be incorporated, a general agent/experiencer in postverbal position in a passive sentence can be as well:
ka lina i pa tohu sótama
DEF city VP PASS destroy soldier
"The city was destroyed by soldiers"
Note that the grouping here is [[pa tohu] sótama] "be destroyed by soldiers," not [pa [tohu sótama]] "be destroy soldier-ed" -- we're incorporating a predicate into a passive verb, not passivizing an incorporated verb which would make no sense because it's presumably intransitive.
The subject position can't be left vacant the way the object position can, but we can use an impersonal construction to code the subject as unknown/indefinite/general/unnamed/irrelevant. This is accomplished with hi -- the same particle used for reflexives -- but in a pronominal position with respect to the verb. I don't think we've ever actually thoroughly mapped out all the preverbal slots and clearly should, but for now the general shape of things is
(SUBJECT) - PRONOUN/i - (NEG) - TAM - VAL - VERB
Thus with hi in the pronoun slot, we can say
hi tohu ka lina
INDEF destroy DEF city
"they destroyed the city," "the city got destroyed"
This is actually very often the best Koa translation for an English passive without expressed agent, because the implication of an actor -- whether intentional or not -- is still present. Note that the actor is not necessarily implied to be animate. Another example of usage:
ai hi puhu le Níkili ne tia?
QU INDEF speak NAME English LOC this
"is English spoken here?"
It's possible for a verb to have a hi in two different positions with different (but related) meanings!
ha hi va hi kanu, hi vi hake ko apu
if INDEF HAB REFL injure, INDEF IMPER seek ABS help
"if one injures oneself, one should seek help"
Okay this was long (again), but I believe I've covered all the questions except for #4, "can pronominal objects in fact appear preverbally?" Despite my certainty in past years that this should be possible, the short answer here is no. There's simply no reason to dream this into being, even if there's a way you could justify it as equal treatment for all predicates. I'm down to reconsider someday should usage seem to demand it for some reason, and in the meantime I'm pleased to say I think we've got a pretty solid system of valence decreasing devices developed at this point!