I was pondering the differences in demonstrative usage between Russian
and English, and it gave me an idea for a unique (as far as I know)
three-way definiteness distinction for the as-yet-unnamed
Salish-inspired language.

So far, I have the articles <txe> and <ta> for singular definite and
singular indefinite, respectively. Already, these work a little bit
different from "the" and "a" in English, because, inspired by Salish,
this language cannot encode presupposition. Thus <txe> does not
indicate "a specific thing that I expect to be commonly understood",
but rather "a specific thing that I have in mind, regardless of
whether or not you know about it".

The third category, for which I have made up the singular article
<kwe> (incidentally homophonous with the antipassive verb suffix, but
that's fine, because homophones are naturalistic, and these are
unlikely to be confused), indicates " a specific thing which I intend
to specify for you". I.e., it specifically presages further
description. This is sort of like a presupposition, but opposite- it
doesn't assert that you should already know something, but rather that
you should know something soon. Grammatically, it has the effect of
requiring[1] the relative clause that it heads to contain an
additional embedded relative clause that co-references the same
subject. When used with transitives, that means that an explicit
object cannot be expressed (although it could be brought back in as an
oblique), because the single core argument slot in the syntax must be
taken up by the subject.

Some examples should make this clearer:

xe-hentqeka ta swetqe-la
xe=hentqe-nk ta swetqe-0=la
1sg=spy-1/3p IND man-3sg=ART
"I see a man (but I don't know who it is)"

xe-hentqeka txe swetqe-la
1sg=spy-1/3p DEF1 man-3sg=ART
"I see the man (and I know who it is, whether or not you do)"

xe-hentqeka kwe swetqe txe tupund ta nk'ap-la
1sg=spy-1/3p DEF2 man-3sg DEF1 hit.TRANS IND nk'ap=ART
"I see the man who hit a coyote (and you ought to recognize him from that)"

To demonstrate how the second definite can mess with argument
structure, we can switch things around a little to get:

xe-hentqeka txe tupund ta swetqe-la
"I see the one who hit a man."

xe-hentqeka kwe tupund ta swetqe-la
"I see the hitter who is a man" / "I see the male hitter".

xe-hentqeka kwe tupund ta swetqe-la va ta nk'ap-la
"I see the coyote-hitter who is a man" / "I see the male hitter of a coyote".

This also demonstrates some of the tricky-to-pin down behavior of the
article enclitic <=la>. In this case, it serves to terminate the
relative clause "ta swetqe-la", which means that "va ta nk'ap-la"
cannot be an oblique argument to "swetqe". That oblique phrase instead
has to attach to one of the previous verbs- "tupund" / "hit" or
"hentqe" / "spy". If we leave it out, we get

xe-hentqeka kwe tupund ta swetqe va ta nk'ap-la
"I see the hitter who is a man of a coyote"

which does not make a whole lot of sense, unless "coyote man" is some
idiomatic, culturally-specific thing. There is still attachment
ambiguity there, though, such that "xe-hentqenk kwe tupund ta
swetqe-la va ta nk'ap-la" could also mean "I spy the male hitter
around/behind a coyote". Context should easily disambiguate, though-
if there is no coyote currently present behind which you are hiding,
then clearly that is not the intended interpretation!

[1] Except when it doesn't. If an embedded argument is provided, then
it must be co-referential with the subject, but given the right
discourse structure you could leave out the defining embedded clause
entirely, as long as you provide some other defining information
(probably restricted to within the same sentence). That way, <kwe> can
be used to establish a sense of tensions and expectation. Thus,

*xe-hentqeka kwe swetqe-la

would be ungrammatical, because it sounds unfinished. But,

xe-hentqeka kwe swetqe-la agu le-tupund ta nk'ap-la
"I see a man, and he is the one who hit a coyote"

would be fine. (Cf. the earlier "I see the man who hit a coyote").

This structure can be used to chain together a bunch of different
referents who can all be identified by the same descriptive clause.

xe-hentqeka kwe swetqe-la agu kwe lwuuxa-la agu he-tupund ta nk'ap-la
"I spy a man and a woman- the ones who hit a coyote".

(This construction also nicely side-steps the fact that I have yet to
invent any plural determiners, so I cannot yet do an example of "I spy
the coyote-hitters"!)

The fact that this new second definite article is not a bit of verbal
morphology, but nevertheless impacts argument structure further
inspired me to create another isolating particle that messes with
argument structure: an "infinitive", <at>.

If "swetqe" means "man" (which it does), then "at swetqe" means "to be
a man", "the state of being a man". If "nbatqe" means "move under
one's own power", then "at nbatqe" means "voluntary movement" or

I general, <at> takes any verb and turns it into a intransitive whose
subject is the nominalization of that verb. Such a verb can, however,
still take obliques to indicate a patient, or be explicitly
transitivized (and inverted, if the agent is not fourth person) or
applicativized in order to re-introduce an agent or some other

The infinitive particle does not attach to the verb directly, but does
cliticize to preceding determiners; thus: <kwe-ot> (recall that
rounded <a> becomes <o>), <tx-at> (deletion of final <e> since
unrounded vowels cannot be in hiatus), <ta-t>.

With these, we can make phrases like:

ta-t matqe-la "automation"
ta-t tupund va ta nk'ap-la "to hit a coyote"
tx-at tupund va ta nk'ap-la "the hitting of a coyote"
ta-t tupuntsa txe nBale-la "one time that Mary hit something"
tx-at tupuntsa txe nBale-la va ta nk'ap-la "Mary's hitting of a coyote"
kwe-ot tupund txe hentqesk-la "the impact you saw"
kwe-ot tupuntsa txe hentqesk-la "the beating you saw done by someone"