Getting Koa's valence-decreasing house in order

It's become increasingly clear this week that I don't really understand valence-decreasing operations in Koa. I created pa very early on as a passive marker -- whatever that exactly means in practice -- and seem to have identified hi as a reflexive prefix, but that about sums it up: no further exploration or description has taken place in all these years. Here are some questions that need to be answered:

1. Can the agent of a passive verb be overtly indicated? If so, how?
2. How does the passive work with ditransitive verbs?
3. How do reflexives work? Is it just a verbal prefix? Or is it a pronoun? If a pronoun, can it appear in other syntactic positions?
4. Can pronominal objects in fact appear preverbally?
5. Is there a way to background the subject, or make it indefinite/impersonal, without using the passive? (Like aquí se habla español "Spanish is spoken here" or ווערטער זאָל מען װעגן און ניט צײלן verter zol men vegn un nit tseyln "words should be weighed, not counted)
6. If there's some kind of impersonal construction, does it matter if the logical agent is human or not?
7. What's the relationship, if any, between reflexives and impersonals?
8. How do we represent reciprocal action, and how is that different from reflexives?

There's a lot of material here and I'm not sure if this is destined to be a single post or several. I guess we might as well go through them one at a time...

1. Indicating agents of passive verbs

It's been a long time since the typology class where we looked at a thorough cross-linguistic survey of agents in passive constructions, so I've been extremely grateful for this paper by Edward Keenan and Matthew Dryer providing exactly that. I needed some help with agent demotion because I knew it was most common for them to be relegated to an oblique position, but it was extremely unclear which Koa adjunct particle would be appropriate. Meanwhile, the only passive agent indication in current active use has been phrases like

pa-lóha-ni
PASS-love-1SG
"my beloved"

From the above it should be possible to say

se palóhani
2SG PASS-love-1SG
"you are my beloved"

...and if this is written out with the particles separated, you end up with

se pa loha ni
2SG PASS love 1SG
"you are my beloved" = "you are loved by me"

This makes it look like passivization in Koa -- at least by means of the particle pa -- is in fact not a valence-decreasing operation at all! Instead, Loglan-style, the subject and object are simply swapped, with pragmatic consequences. Further examples if this is true:

se loha iu poli pi mehe
2SG love EXT many QUANT person
"you love so many people"

se pa loha iu poli pi mehe
2SG PASS love EXT many QUANT person
"you are loved by so many people"

Initially I figured this kind of linguistic backflip wasn't really possible for humans, and therefore should be considered inadmissible. The best I could do was to suppose that palóhani breaks down to [[palóha]ni], like we start out with "beloved person" -- palóha -- and add personal possession: now it's ka palóha kémeni, "my beloved person." Maybe it doesn't literally mean "the one who is loved by me"...as much as it really, really, really, really looks like it does.

Keenan and Dryer, though, point out that some languages really do do this, particularly in the Bantu family. Here's one of their Swahili examples:

maji ya-meenea nchi
water it-cover land
"the water covers the land"

nchi i-meenea maji
land it-cover water
"the land is covered by water"

Here there isn't even any kind of passive marker, just verb agreement with a different subject. Bottom line, this may in fact be a viable Koa strategy. What if we do want an adjunct, though, whether for clarity or for semantic or pragmatic reasons? One strategy is an ablative -- so o in Koa -- but that seems a little ad hoc and apparently some kind of instrumental is more common cross-linguistically.

Only...Koa doesn't have an unambiguous instrumental. I've deferred serious consideration over the years, imagining (or hoping) that other particles might be sufficient: going in a car, or writing with a pen, etc. When it comes down to it, though, I really do want to be able to talk straightforwardly about means. If kelo is "reason" and kemo is "manner," what would "means" be?

After a few days of pondering, I think this is important enough to assign one of our newly available c- particles to. Initially I'd chosen ca but no amount of familiarity seemed able to make that feel right, so I switched it to ci at about 3am last night and I'm pretty happy with that. So:

ni kanu ka tue ci kivi
1SG injure DEF finger INSTR rock
"I hurt my finger with/on a rock"

se ia te puhu ci le Koa
2SG CERT ABIL speak INSTR NAME Koa
"you really can speak Koa"*

ta miilo ve ka kala ta i si pa iune ci mola
3SG INCEPT-know REL DEF fish 3SG VP ANT PASS steal INSTR bear
"he discovered his fish had been stolen by a bear"

*Usage here hasn't been formally decided. Should it be ci le Koa "by means of Koa" like Hungarian, or mo le Koa "in the manner of Koa, Koa-ly" like Polish or Latin? Or (probably not) ne le Koa like English? Seems like ci is most appropriate semantically but a final decision can come later.

By the way, I may have also decided in the midst of this instrumental study that the word for "and" should just be e plain and simple. I've had me conjoining NP's -- influence from Swahili -- but I'm not sure what that gets me in exchange for additional semantic ambiguity (did I really buy a house with a car, or just a house and a car separately?). I can't imagine that letting go of this would upset any learner regardless of linguistic origin.

Getting back to the original question, yes, this kind of passive can indeed have an overt agent phrase: (1) definitely headed by ci, our new instrumental, and (2) maybe even as a plain NP that looks like a core argument. Still unanswered is what's really going on here with valence at a lower level, particularly in view of option (2), so we'll need to come back to this.

2. The passive of ditransitives

In English a person can be given a book, and a book can be given to a person, within the same apparent passive structure. What about Koa? The answer seems to hinge on how ditransitives are handled in active clauses, about which I'm honor-bound to admit sheepishly that I'm not sure. If both the direct and indirect objects are full NP's I think one of them has to be phrased as an adjunct to avoid modifying the other:

A. ta ana ka lelu la ka toto
3SG give DEF toy DAT DEF child
"he gave a toy to the child"

B. ta ana ka toto pe ka lelu
3SG give DEF child APPL toy
"he gave the child a toy"

??ta ana ka toto ka lelu
3SG give DEF child DEF toy
"he gave the toy's child..."

What's less clear is what happens when the indirect object is a pronoun. I'm not sure why this shouldn't work:

ta ana ni ka lelu
3SG give 1SG DEF toy
"he gave me the toy"

I guess we're kind of begging the question, though, because as soon as we're saying that both of those ana clauses above are acceptable, it's clear that it must be possible to promote either the direct or indirect object. I guess I'll go ahead and say yes, this is permissible, until I find a good reason to forbid it. For what it's worth I think Yoruba agrees. So then, the passives would look...like this, I think?

A. ka lelu i pa ana la ka toto (ci ta)
DEF toy VP PASS give DAT DEF child (INSTR 3SG)
"the toy was given to the child (by him)"

B. ka toto i pa ana (pe) ka lelu (ci ta)
DEF child VP PASS give (APPL) DEF toy (INSTR 3SG)
"the child was given the toy (by him)"

Oof, this is getting complicated. I'm not sure whether the pe in structure B is obligatory, despite the fact that it would have been in the active sentence -- once the two objects aren't piled up anymore, the ambiguity disappears! And while we're trying to figure that out, if it might be possible to free agents from an adjunct phrase like we were saying before, might this be an acceptable rephrasing of type A?

A. ka lelu i pa ana ta la ka toto
DEF toy VP PASS give 3SG DAT DEF child
"the toy was given by him to the child"

In this case, the literal translation would look something like "the toy was his given thing (i.e. his gift) to the child." Even though this surprises my IE intuition, I really don't see why this shouldn't be okay! And I actually kind of love it.

Getting back to B above, I think it's better with pe omitted: included, it's acceptable but not really helpful. It's the same as the fact that we can theoretically say

ni ipo pe ka cai
1SG drink APPL DEF tea
"I drank [with respect to] the tea"

...but the circumstances where a speaker would naturally choose to do that would be pretty specific. At least 99% of the time you'd just get ni ipo ka cai.

This is already long enough that I think I should leave 3-8 for separate posts...and I haven't forgotten that there's some really important material still waiting to be written about nominalization and relativization! I'll do my best over Thanksgiving break.

A Koa lullaby

On the evening of August 22nd, 2012 when Callie was five months old and I was trying to get her to sleep, this lullaby somehow spontaneously came out of me fully formed. That was the heyday of real Koa usage, and I frequently spoke it to her in the course of daily life; for a minute it seemed not inconceivable that it might become a living language. That hasn't quite happened (yet), but the song itself -- Aika ko Nuku, "Time for Sleep" or "Sleepytime" -- has stuck around and both girls know it word for word, even if they can't parse it.

Given that this is still the only existing Koa text, I'm kind of appalled I've never posted anything about it! Here's the score, followed by an interlinear translation.

We usually sing it twice through, repeating the last two measures more slowly on the final repetition.

Over the years I realized I've been imagining the spelling as Aika Konuku rather than Aika ko Nuku as I had it in my score from 2012, but I'm not sure why. There's really no prescriptive...anything in Koa about word grouping or capitalization at this point, so everything is reflecting perceived aesthetics of the moment. Another topic for someday. Anyway, the words:

aika ko nuku la ka piku ni
time ABS sleep DAT DEF little 1SG
"it's sleepytime for my little one"

vo se io maka ne ni áheki
PRESENT 2SG TRANS lie LOC 1SG arm-DIM
"here you are now lying in my sweet arms"

toa pi kiuni i hala pai pui
that QUANT need-rest VP after day long
"so tired after the long day"

aika ko nuku la se
time ABS sleep DAT 2SG
"it sleepytime for you."

The third line has been a bit problematic. Through most of the song's life it existed as

toa pi kiuni hala a pai pui
that QUANT need-rest after INDEF day long
"so tired after a long day"

First of all, though hala has meant "after" since about 2011, it wasn't until this year that I finally figured out how to use it, and this isn't it -- there was nothing integrating it into syntax! This has been changed into its proper verbal form in the words above (a post about this is forthcoming). A is not at all the right particle, either...if anything it might make sense to say ...hala ti pai pui "after this long day," but really the long day is not being spoken of in any kind of specific way: we're referring to a kind of tiredness one feels after a long day in general, and as such the right way to express it is via object incorporation: no article.

Still an issue, though, is the translation of so tired. I picked a word for this out of a kind of Esperanto correlative logic, where you would genuinely say tiom laca "that amount of tired," but it doesn't actually make sense in Koa: toa pi kiuni suggests a real referent in the world for toa "that" which clearly doesn't exist ("I'm not this tired, I'm that tired."). What we want is iu kiuni "so tired," but iu didn't exist back then, and it doesn't fit the meter. We need more syllables.

se iu kiuni "you're so tired" (accent in the wrong place)
kiuni poli "very tired" (words too drawn out and accent on poli instead of kiuni. I think we need 3 syllables)
tótoki kiuni "tired little child" (maybe okay! Sounds a lot like the original; two successive ki's but it's still not hard to sing)
toto iu kiuni "such a tired child" (also possible but I like it less for some reason)
néneki kiuni "tired little baby" (since Callie was one when this was composed. Maybe?)

I don't think I'm making a decision today, but I potentially like some of these options. There may be a revolt, though...the change from hala a pai to i hala pai wasn't even noticeable to the little ones, but this one -- when I figure out a replacement -- won't be so easy to sneak past them.

Incidentally I mentioned to them at some point that I was working on a second verse, and since then they ask me about it every few months. I think I'm having some performance anxiety, as usual for me with songwriting, but I know it'll be something like

Aika ko nuku la ka piku ni (same as first verse)
Ka esi me aimo something ("the moon and stars will watch over you through the night?")
Something about no cause for worry when you're so loved
Aika ko nuku la se (same as first verse)

Uh...still some big gaps in there. I'll get back to you.

The two faces of ke- compounds

There's an additional bit of preparation we need before diving into that thorough discussion of nominalization, relativization and focus that I keep promising: an understanding of the formative ke-.

This particle, by default carrying interrogative meaning, can be productively preposed to a second particle to form compounds that look and behave like predicates. There has traditionally been a sharp divide between two classes of resultant forms, though, depending on whether the second particle is an article (a, ka, le, and most recently u) or not.

1. When added to an article, ke- forms an interrogative pronoun. These solicit a response of the type indicated by the article itself, so:

kea = indefinite, "what?"
keka = definite, "who/which one?"
keu = definite plural, "who all, which ones?"
kele = name, "what is the name?"

2. When added to any other kind of particle, however, these lose interrogative force and instead give us a sort of meta-description of the meaning of that particle. There are lots of these, so for example:

kene = location
kela = destination
keme = attribute
kemo = manner
kehe = time (at which something occurs), moment
kepe = subject, topic, respect, regard
kelo = reason, cause
kesi = that which went before, past
kepi = amount, quantity
kecu = that which will be, future
kema = that which is ongoing; current?
kete = possibility
keki = necessity
kelu = a desire
kena = that which is not; negative?
keha = condition (that which "ifs")

These should all be spelled out and explained at some point -- I was sure I wrote a post about this back in the 20-teens, but it would appear I never got around to it! And it would be good to do this before I accidentally assign those roots to other meanings (see point 5 below).

3. In one case the resultant form has been modified a bit:

kia = affirmative (not keia which is of an inadmissible form for a predicate; curiously this was not an intentional choice, but an accident based loosely on Finnish kyllä "certainly"!)

4. Forms with other specifiers would have no identifiable meaning and so aren't included in this set: keko, kehu, kepo, keti, keto.

5. Note that there are other words of this apparent form which in fact are their own bisyllabic roots:

keli "language"
kevi "light in weight"

This was not planned out particularly well, and in fact keli could and really should also mean "hypothetical" and kevi "a command." I'm not sure what to do about this...either we could have homonyms in a breathtaking change of allegiance in favor of Koa's rights as an artlang, or maybe those roots need to be reconsidered. I honestly don't love keli -- it somehow utterly fails to capture what I love about Finnish kieli -- and kevi could become kevu. Or something. OR I could create different roots for the meanings that either or both of these "should" have with the ke- formative: I think the "command" root should start with transitive verbal meaning, for one, rather than "that which is commanded."

Incidentally, I don't think I've ever mentioned out loud just how many Koa roots are in fact unapologetically borrowed from Finnish. Paa "head," kume "ten," sata "hundred," tuha "thousand," ela "live," kusu "ask," nuku "sleep," poi "away," voi "be able," tule "come," mene "go," lahe "leave," soi "sound, ring," iso "big," suli "great," sini "blue," puna "red," lepa "bread," vai "butter," vate "cloth," puhu "speak," sano "say," hulu "crazy," ike "cry," pime "dark," valo "light," pai "day," suva "deep," vake "difficult," vami "ready," ovi "door," ava "open," asu "dwell," suo "eat," vela "even," paha "evil," pele "family," vaha "few," luta "find," kala "fish," hisi "mist," uto "foreign," uno "forget," ana "give," hei "hello," moi "goodbye," vihe "green," sivu "leaf," pusu "gun," kova "hard," kulu "hear," apu "help," koke "high," maki "hill," koto "home," talo "house," asi "idea," tapa "kill," maa "land," keli "language," vime "last," liu "lead," opi "learn," kile "write," vesi "liquid," nae "see," hake "look for," mata "low," kone "machine," and on and on. In some cases the form or meaning has obviously been adapted. I'm not sure why, but something about Finnish phonology really lends itself to the vibe I've been going for with Koa from the beginning.

ANYWAY, it occurred to me the other day that if kea literally breaks down to ke a -- in other words, "which indefinite thing beginning with a?" -- is it possible that non-article compounds should have interrogative force as well? Couldn't kene mean "which phrase beginning with ne," i.e. "where?" Suddenly a whole cast of what Esperanto might call correlatives effortlessly unfolds:

kene - tine - tone = where - here - there
kela - tila - tola = whither - hither - thither
kehe - tihe - tohe = when - now - then
kemo - timo - tomo = how - like this - like that
kelo - tilo - tolo = why - for this reason - for that reason
kepi - tipi - topi = how much - this much - that much

For a moment it seemed like, despite the fact that I kind of hate nearly every single one of those forms with those meanings on aesthetic grounds, it may be a logical necessity to allow this. In other words, "why" would also mean "reason," as in "let me tell you about the how and why." It felt inelegant and unappealing, but maybe important in the service of internal consistency.

But thankfully for my aesthetic sensibilities, these correlatives weren't meant to be. The reason it works with kea - tia - toa and friends is that the resultant pronouns have a semantic that allows them to be integrated into syntax just like any other predicate (albeit with the article integrated inside of itself, so to speak). We can say

ni na suo toa
1SG NEG eat that
"I didn't eat that"

...in exactly the same way we can say

ni na suo lepa
1SG NEG eat bread
"I didn't eat bread"

But the compounds with other particle types would produce something that otherwise does not exist as a category anywhere in Koa: adverbials! If one said

ni si asu tone
1SG ANT dwell 'there'
"I used to live there"

...in order to parse it correctly, they would have to know that tone should not be interpreted according to the ordinary rules of Koa: that is, not as a direct object or a modifier, as one predicate following another. To allow these would be to introduce a genuine lexical class division among predicates into the language for the first time, and a completely externally unidentifiable one at that. Ick. Might as well hang it all up and start over if I'm going to throw out the most basic guiding light of the language. It has to be, as it always has been,

ni si asu ne toa
1SG ANT dwell LOC that
"I used to live there"

So then, feeling pretty solid about the system of ke- compounds and why things mean what they do, let's remember all those words under point 2 above (kene "location," kemo "manner," and so on), because they're going to be critical when we start trying to nominalize more complex clauses.

P.S. I'm feeling less and less sure about u as a plural definite article. I realized today that this would make "everyone" have to be pou instead of poka, which makes me sad, but beyond that I'm having some trouble being convinced that it feels very much like Koa. No final decisions but that's where I'm at. Note that if we keep it, we'll have series like this: poa "everything," poka "all of it," pou "everyone."

Kion fari?

A small break in the action: "what to do?"

This has been irritating me of late, because it's such a simple little structure in the languages I know best, but I could not figure out how to express it in Koa. First attempt: Kea sa kipaete? Literally "what is to be done?" But this has...so many morphemes. I'm wondering whether there might be something like Kea sa ete? I think this would be a focalization of ete kea? meaning literally "do what?" Actually either of those Koa sentences seem like they could work, looking at it like that, and in fact the latter feels like it might retain more of the vibe of the original: it's not really a focused sentence pragmatically, more a general neutral statement about a situation.

Incidentally, in the realm of little useful Koa phrases, I forgot to mention the translation of "OMG" that Allison and I worked out back in April! The Koa version is OVN, short for oo vala ni. I meant to start integrating that into my texting but somehow haven't managed it yet...

Representing verbal focus and lexical class

Based on the conclusions of the previous post, for about five minutes I thought we might be seeing the end of sa as a focus particle in favor of constructions with i ka. It was a little scary but also bold and exciting...I wrote:

"Kea sa se ma sano? is exactly equivalent to
     Ka se ma sano i kea?
     Kea i ka se ma sano?

If we do this, movement rules are completely eliminated. You can superficially 'front' things, as above, but it's then following the same grammatical/syntactic rules, not inventing a new one. We've never fully explored the implications of our ability to add specifiers to clauses, and this is potentially one of them."

Before that line of thought had really gotten going, though, I realized that it ran into irreconcilable difficulties if the focused constituent is a verb rather than its subject or object. For example, if we start with ka tálate i neni "the attempt was in vain," how do we focus "in vain?"

Neni sa ka tálate =
     Neni i ka ka tálate...............?

We end up with "Vain is what the attempt..." and then nothing. What should go in that space? A similar problem comes up in other circumstances when using a specifier with a verb, for example:

Kea sa ta? "What's he like?" -> Ni na ilo ka ta...........? "I don't know what he's like"

What completes these sentences? Is there a dummy verb? If there is one, then there must be a way to use it in non-nominalized clauses too, like

ta ila koke "he be's tall" (ila borrowed from Lithuanian yra incidentally)
ni na ilo ka ta ila "I don't know what he's like"

The only thing about this is that if ta ila koke means the same as ta koke, then technically shouldn't ta lalu be equivalent to ta ila lalu...like "he's a singer," "he's a singing one," rather than just "he sings"? I suppose ila could have an underlying meaning something like "X is a member of set Y."

This is getting a little sidetracked from the original topic, but speaking of dummy verbs, if I ask Kea sa se ete? "What are you doing?", why can't the answer be an adjective? In other words, why should we assume that the predicate ete is replacing must be verbal rather than adjectival? Why does Kea sa se ete not just mean "what predicate defines the set you are a part of?", just the same as ila above?

BECAUSE, I realize, ete doesn't exactly mean "do." It means "verb"!!!! Without intending to, I created words that allow specificity with respect to the semantic of lexical class, since Koa entirely lacks this concept formally. Check this out:

ete "do the action of predicate X" = "verb"
ila "be a member of set X" = "adjective"
mea "an instantiation of predicate X" = "noun"

This is really pretty exciting. Suddenly we can say things like

na vi ila hulu
NEG IMP 'adjective' crazy
"don't be crazy," as in "don't act crazy"

na vi hulu
NEG IMP crazy
"don't (actually) be crazy"

na vi ila toa
NEG IMP 'adjective' that
"don't be like that"

And there's a fine distinction that can be made between things like

ta ete lalu
3SG 'verb' sing
"he does singing," "he does that singing thing," "the action he engages in is singing," "he sings"

versus

ta ila lalu
3SG 'adjective' sing
"he's one of those singers," "the set he belongs to is the singing one," "he's a singer"

In other words, we can clarify between what he is and what he does. Furthermore, these words also give us Koa-native meta-terminology for these kinds of usages of predicates:

étema "verb"
ílama "adjective"
méama "noun"
nóama "name"

I had never thought before about the need to be able to talk about Koa in Koa, but clearly yes, we should have our own words for the concepts most relevant to Koa grammar. Now I really want words for "predicate" and "particle."

SO THEN, getting back to what we were talking about here, it turns out that we actually can potentially focus verbs using that same i ka structure:

neni i ka ka tálate i ila
vain VP DEF DEF try.instance VP 'adjective'
"vain is what the attempt was," "the attempt was in vain"

ka ka tálate i ila i neni
DEF DEF try.instance VP 'adjective' VP vain
"idem"

Do we want to, though? That's a different question. Honestly...not really. Partly because I like the brevity and flow of neni sa ka tálate over the necessarily syntactically complete versions above, partly because I dread the proliferation of /k/s that this structure would ensure, e.g. keka i ka ka lúlema i kusu? "whom did the judge ask?", partly because I honestly have some loyalty to sa as one of my very first particles. But also, allowing multiple ways of more or less saying the same things also gives us more nuance of sense in a super useful way for a living language:

kea sa se ma sano?
what FOC 2SG IMPF say
"what are you saying?"

ka se ma sano i kea?
DEF 2SG IMPF say VP what
"what you're saying is...what, exactly?"

kea i ka se ma sano?
what VP DEF 2SG IMPF say
"what is it that you're saying?"

We do still need to talk in detail about what verbal focus looks like in practice. Is "I kissed it, I didn't eat it!" suso sa ni ete ta, na suo sa? Whoa...I was expecting that to be weird, but actually I think that's exactly right. Anyway, more to come on that front. But meanwhile, I think we finally have just about everything we need to lay down some principles of focus/relativization/nominalization, hopefully the next time I find some spare time to write.

Unrelated note: tai should stop meaning "stand." I'm not sure why I decided it should have this double life, but it's kind of weird and I don't like it. It just means "be/exist."

Focus without movement

This is the first in a series of posts about focus, nominalized clauses and relativization, which in Koa are all closely related. I'm hoping that by the end of it we'll have cleared up a whole suite of muddlements that have persisted since the early days.

Koa's focus particle, sa, has its origins in Yoruba ni with the same function. In Yoruba, the focalized constituent is moved to the front of the clause, followed by ni (li or l' before an oral vowel), and leaving a gap in its original position:

kíl'o rà níbẹ̀?
what.FOC-2SG buy there
"what did you buy there?"

aṣọ ni mo rà
cloth FOC 1SG buy
"it was cloth I bought"

This is exactly parallel, at least superficially, to the structures in Koa:

kea sa se kou ne toa?
what FOC 2SG buy LOC there
"what did you buy there?"

vate sa ni kou
cloth FOC 1SG buy
"it was cloth I bought"

Movement rules like this seemed plausible enough given Yoruba's permission and I didn't think that much about it until I started trying to translate headless relative clauses. If "What do you want?" is Kea sa se halu?, then how do you say "I don't know what you want"? I recall going through a whole bunch of contortions trying to figure this out:

?ni na ilo [ kea sa se halu ]
1SG NEG know [ what FOC 2SG want ]
A word-for word calque of the English structure. Can embedded clauses can be focused just like main clauses, and without overt marking? Is this how embedded questions should work? This feels very natural, obviously, but that's not necessarily a good thing: it needs to make sense in terms of Koa, not in terms of English.

?ni na ilo ko [ kea sa se halu ]
1SG NEG know COMP [ what FOC 2SG want ]
This has a complementizer to set off the embedded clause, which is how Hungarian does it. Still feeling really nervous about the way focus works in the sub-clause, and also the way the question is embedded.

?ni na ilo ko [ se halu kea ]
1SG NEG know COMP [ 2SG want what ]
This gets rid of the worrisome focus issue, but somehow feels even worse.

I think it was nagging at me that (A) I wasn't sure I really liked my not-particularly-examined movement rules after all, and (B) accepting the IE way of thinking of these as "embedded questions" in the first place felt like sloppy, circular thinking. If there were a book titled What I Think, would the Koa translation genuinely be Kea Sa Ni Lule, using a question word -- and literally exactly the same sequence of words as the question "what do I think?" -- even though there is not really any kind of question being asked? PLUS we're explicitly not supposed to be forced to rely on intonation for basic functional distinctions, and that's exactly what this would require.

It also made me feel a little squirmy that sa was such an anomaly in every way. It's a particle that goes after the constituent it applies to? What in fact was going on here below the surface?

All this led me to remember my Nahuatl, which handles focus in a pretty astonishingly different way. Note that ca is a statement marker, contrasting with e.g. cuix which would turn these into questions:

ca cihuātl in cochi
STMT woman DEF sleep.3SG
"it is a woman who is sleeping," lit. "the she-sleeps-one is a woman"

ca ātl in niqui
STMT water DEF 1SG-drink
"it's water that I'm drinking," lit. "the I-drink-one is water"

Calquing these into Koa, we'd end up with:

mina i ka nuku
woman VP DEF sleep
"it is a woman who is sleeping"

anu i ka ni ma ipo
water VP DEF 1SG IMPF drink
"it's water that I'm drinking"

Several pretty noticeable things came out of this right away.

1) Focus requires no underlying movement rules, which is frankly awesome in a denying-Chomsky-his-invisible-branching-structures kind of way

2) Which constituent counts as the "NP" and which as the "VP" is a little arbitrary; both of these could be flipped around while retaining the focus:

ka nuku i mina
DEF sleep VP woman
"the one sleeping is a woman"

ka ni ma ipo i anu
DEF 1SG IMPF drink VP water
"the thing I'm drinking is water"

3) These structures are exactly parallel to relative clauses:

ka mina ve nuku
DEF woman REL sleep
"the woman who's sleeping"

ka anu ve ni ma ipo
DEF water REL 1SG IMPF drink
"the water I'm drinking"

4) They would give us a Koa-native way of doing "embedded questions," without having to think of them as questions at all:

ka nuku
DEF sleep
"who is sleeping"

ka ni ma ipo
DEF 1SG IMPF drink
"what I'm drinking"

Ka Ni Lule
DEF 1SG think
"What I Think"

5) Most fascinating of all, i ka in the original calqued-from-Nahuatl examples can be replaced with sa to yield identical meanings:

mina sa nuku
woman FOC sleep
"it is a woman who is sleeping"

anu sa ni ma ipo
water FOC 1SG IMPF drink
"it's water that I'm drinking"

Nahuatl made it possible to work backwards up to that Koa structure with sa in such a way that we can understand exactly what it's doing there without having to infer movement at all, AND fixed the headless relative clause problem, all in one fell swoop.

The natural follow-up question, in the face of this, is whether we actually need sa at all! And the answer is yes, for interesting reasons that we'll get to next time...