I have mentioned Serikats a few times on the list before, but
searching the archives indicates that the last time I mentioned it was
in 2009, so, quick background: Serikats was one of my very early
conlanging attempts from back before I really Knew What I Was Doing;
its a mildly esoteric language and ended up being more than I could
handle at the time, and so has languished in a semi-consistent
unfinished state for quite some time.

The central idea behind Serikats was that tense would be obligatorily
carried on all noun phrases, while case would be indicated by
inflections on verbs.

Though I did not know it at the time, marking tense on noun phrases is
apparently Not Uncommon in western american languages, and the way I
did it is actually fairly similar to how natlangs do it, so there's no
real problem there. Figuring out how to handle case marking via verbal
inflections is much more of a challenge. However, I believe I have had
some useful insight that might simultaneously solve the problem of how
to make the synchronic system work, and provide a plausible historical
explanation for how it came about.

Step 1 ("Proto-Serikats") is a fairly normal VSO language (except that
I will assume nominal tense already exists at this point) that marks
cases by adpositions. Things end up a little simpler later on if there
at least exists an adposition for every core argument, even if they
are not obligatory (like "by" for demoted subjects in English
passives, or "to" for English indirect objects), but I think it still
works if some of those are missing.

The voice system is then simplified to make use of adpositions even in
core arguments- if you want to put something other than the normal,
unmarked subject in the subject position, you just *do*, and use the
appropriate adposition to signal the change. Original unmarked core
arguments either become obligatorily marked, or are re-analyzed as
using a null adposition. At this point, the language is similar in
style to Gogido or WSL, but with stricter word order.

Next, we take advantage of explicit case marking with adpositions to
free up the word order. Specifically, we allow fronting of noun
phrases for focus / topicalization. As in English, however, relocated
NPs that start inside PPs can leave their governing prepositions
behind. This leaves bare prepositions sitting around next to the verb,
where they are ripe for being incorporated as clitics or affixes.

At that stage, if you only allow fronting the subject, then
incorporating the left-behind subject prepositions ends up looking a
lot like a trigger system. If you only front the object, then nothing
special happens- you just end up with dangling prepositions at the
ends of clauses, like English. But if it is possible to move both
subjects and objects at the same time, two things happen: we can
incorporate multiple left-behind bits to get a sort of polypersonal
agreement, but we also end up with some serious ambiguity as to which
preposition goes with which moved NP.

The original design for Serikats matched up verbal suffixes with noun
phrases according to their order in the clause, ignoring the position
of the verb; that logically works, but it feels very clunky and
unnatural. A slightly better option is to associate a specific mapping
with each of the three possible verb positions- say, NNV = OSV, NVN =
SVO and VNN = VSO (the original basic pattern). English kind of does
this with object-focus constructions (e.g., "Him, I like; you, I
don't."), which is largely why Yoda speak works. I think I'm going to
sticky-note that for a potential sister language.

The approach Serikats ends up taking, however, is inserting resumptive
pronouns, which again can become incorporated as part of the verbal
inflection. This results in a situation where clauses with two 2nd
person or two 3rd person arguments with the same number/gender/etc.
are ambiguous, but no other combinations are. And plenty of natlangs
get along fine with that level of ambiguity. It feels very Salishan to
me. If I really wanted to nail it down, I could use that NNV/NVN/VNN
pattern system solely to deal with the ambiguous cases... but nah, a
little ambiguity is OK. In extended discourse, making use of a lot of
conjoined clauses with elided syntactic pivots is likely to eliminate
the ambiguity in most cases.

Verbal morphology is simplified a little bit if there is at least one
null resumptive pronoun-turned-agreement-marker, but that's also easy
to arrange.

Now it's possible to re-order both noun phrases and the inflections on
the verb independently, and it seems to me that that ought to be
exploited to fulfil additional pragmatic functions. I'm thinking we
end up with a hierarchy of more-to-less subject-like role inflections
where any two must occur in one particular unmarked order, and where
switching them has some highly marked discourse function (like, say,
indicating that a prototypical patient is highly agentive in this
case, or that a more-prototypical agent is acting non-volitionally).
The relative positions of noun and verb phrases can then be used to
indicate contrastive focus and/or topicality.

I don't have all the little details nailed down yet, and it will be
interesting to see how well I can retrofit this new system onto the
existing language with a maximum of reanalysis and a minimum of
actually-having-to-change-things (fortunately, the existing corpus is
quite small), but I'm feeling pretty happy with the outline so far.