Headless relatives and nominal subjects

This is a brief excursus to note an interesting consequence, and some interesting questions, arising from the way we've just decided to handle nominal subjects in relative clauses, in the context of the way we handle headless relatives specifically. This occurred to me with surprise just now while writing about something else and I didn't want to forget it!

To set this up, let's take a second to remember how predicates used adjectivally work in syntax. At base, the predicate simply follows the head it describes, without specifier:

ka sivu vihe
DEF leaf green
"the green leaves"

These adjectivals can also stand on their own with a specifier to describe an unnamed entity with the given characteristic, thus

ka vihe
DEF green
"the green one(s)"

It does occur to me to wonder in this moment if there's a "Ø" in the nominal slot in that kind of construction -- not the kind of thinking we've typically applied to the syntactic motivation of lexical class in Koa, but it's an interesting question:

?ka Ø vihe
DEF Ø green
"the green Ø"

Anyway, we can also give that adjectival a verbal force by turning it into a relative clause with u, thus

ka sivu u vihe
DEF leaf REL.CL green
"the leaves that are green"

So far so good, and uncontroversial in modern-day Koa. Now, looking at that last example, suppose we adjust the relative clause to have a nominal subject:

ka sivu [ le Kéoni u ako ]
DEF leaf [ NAME John REL.CL pluck ]
"the leaves John picked"

Here the entire clause le Kéoni u ako "that John picked" is functioning as an adjectival modifying the head ka sivu "the leaves." But what happens if, like ka vihe "the green ones" above, we want to delete the head noun? What if we just want to say "the ones John picked"? It would appear, keeping the structures parallel, that we would have to be left with the following!

ka [ le_Kéoni u ako ]
DEF [ John REL.CL pluck ]
"what John picked, the one(s) John picked"

Integrated into a matrix clause, we'd have:

Ai se-si-nae ka [ le_Kéoni u ako ] ?
QU 2SG-ANT-see [ John REL.CL pluck ]
"Have you seen the one(s) John picked?"

Even though this output follows logically, and I don't see anything wrong with the structure itself, it boggles my mind in a rather uncomfortable way. It's possible that I still need some time to get used to these newly-approved clause type markers and the way they show up in syntax; if that were older hat, perhaps the above wouldn't look surprising at all.

As an excursus to the excursus (an exexcursus?) I might mention that another allowable strategy for the relative clause up there would be to remove the clause marker and thereby make it non-finite; I've been saving the full discussion of these structures for some presumably upcoming post on nominalized clauses specifically. But just for completeness, this would be another way to say the same thing:

ka [ le_Kéoni ako ]
DEF [ John pluck ]
"the one(s) John picked"

Back to the plot, however weird it looks, ka le Kéoni u ako does again seem to be completely above board syntactically according to all the rules we've worked out to this point. Pushing into extremely speculative territory, though, I do have a tiny tingle of curiosity about whether -- in addition -- ka could itself show up as a clause type marker instead of u for this particular kind of construction. That would give e.g.

?le_Kéoni ka ako
John ?.CL pick
"what John picked, the one(s) John picked"

...or, in a larger clause,

?Ai se-si-nae [ le_Kéoni ka ako ] ?
QU 2SG-ANT-see [ John ?.CL pluck ]
"Have you seen the one(s) John picked?"

Honestly, I don't know! It's beautiful in the same way as the other dependent clauses, it just never, ever occurred to me that Koa syntax could possibly work this way. Since there's more than one possible specifier for the item(s) picked, though, would this logic take us into complete absurdity?

ti vihe
this green
"this green one"

po vihe
UNIV green
"green ones (in general)"

po_ka vihe
all green
"all the green ones"

Now with the standard headless relativization strategy:

ti [ le_Kéoni u ako ]
this [ John REL.CL pluck ]
"this one John picked"

po [ le_Kéoni u ako ]
UNIV [ John REL.CL pluck ]
"ones John picked (in general)"

po_ka [ le_Kéoni u ako ]
all [ John REL.CL pluck ]
"all the ones John picked, everything John picked"

So far so good...but

?le Kéoni ti ako
"this one John picked"

?le Kéoni po ako
"ones John picked (in general)"

?le Kéoni po ka ako
?po le Kéoni ka ako
"everything John picked"

Uhh...my parser definitely just broke. My instinct is the clauses with ka kind of make sense -- though I would need to do some research to decide what even to call such a clause...just "headless?" -- but that it's madness to allow any and every possible specifier to sub in for a clause type marker. Given, though, that it's a basic Koa principle that where one particle/predicate/structure of a given type can go, any other such particle/predicate/structure can go, we would have to be careful with definitions here. We'd have to say...that ka as a clause type marker is homophonous with ka "the" but is in fact a different marker, in the same way that ko forms abstract nouns but also has a separate identity marking finite clauses used as nominals.

Going to have to sit with that.

In the mean time, let's try a more complex sentence and see whether that creates a train wreck with any of these strategies: "I gave you what John said Mary wanted." That has two levels of embedded clauses, one of which is a headless relative and both of which have nominal subjects...

1. ni.ána.se ka [ le_Kéoni u sano [ le_Meli ko halu ] ]
I.gave.you DEF [ John REL.CL say [ Mary NOM.CL want ] ]

2. ?ni.ána.se [ le_Kéoni ka sano [ le_Meli ko halu ] ]
I.gave.you [ John HDLS.CL say [ Mary NOM.CL want ] ]

3. ni.ána.se ka [ le_Kéoni sano ko [ le_Meli halu ] ]
I.gave.you DEF [ John say COMP [ Mary want ] ]

Example 1 uses the standard form with finite dependent clauses; example 2 incorporates the experimental new headless relative clause marker; and example 3, for contrast, shows the alternative strategy in which both subordinate clauses are nominalized (nonfinite).

Between (1) and (3) I feel like I'm getting a difference in register. With finite dependent clauses there's no question that the syntax is much more complex, in perhaps a slightly formal/scientific/legal/nerdy way? The closest similar distinction I can draw off the cuff in English would be between "the things that Mary said that she wanted" and "what Mary said she wanted," but I think in Koa the difference in register is wider. Interesting.

With (2)...I think this structure is still too experimental for me to have intuitive feelings about it. I can get at le Kéoni ka sano as "what John said," but once that clause starts to take its own complement I can't process it at all. This may be a place where we need either more examples in actual use, more crosslinguistic evidence, or both.

What about other kinds of headless relatives, though? What about where John said it? Oh dear. Traditionally I think we would have this, with a ke-compound; unfortunately the article on theta clauses had not one single example clause with a nominal subject, so my choice of ko as the clause type marker here is intuitive rather than examined!

kene le_Kéoni ko sáno-ta
location John NOM.CL say-3SG
"(the place) where John said it

or, non-finite,

kene ko le_Kéoni sáno-ta
location COMP John say-3SG
"(the place) where John said it

Actually I think we're okay. By analogy to the other headless relative types we've just been discussing I was wondering if we might also have e.g. ne ka le Kéoni u sano, but that's something different: it would mean "in the things John said."

To be fully transparent, I have to say I'm wincing a bit at sentences like niánase ka le Kéoni u sano le Meli ko halu; the syntax is complex at a level that younger Koa would have rejected reflexively as clearly, obviously unacceptable to the charter and to the spirit of the language. I feel a bit wistful, or regretful, for our now lost fully modular structures, which would have made this sentence come out simpler and much more recognizable to Koa's original creole muses:

ni.ána.se ka [ le_Kéoni i sano ko [ le_Meli i halu ] ]
I.gave.you DEF [ John FIN say COMP [ Mary FIN want ] ]
"I gave you what John said Mary wanted"

I don't know what to do with how much I like that...and how much more I like it than most of what we've been talking about here today. I wonder how much choice between structural options is reasonable in a language: could all of these options be acceptable depending on style and register? Swahili certainly has multiple allowable strategies for relativization, for example, each kind of singing its own song. This feels like it's verging onto artlang territory just a bit, but maybe that's inescapable as soon as any conlang is subjected to the expressive needs of actual use, whether or not its speakers admit it.

Well, that was much less brief than I expected or intended, and way to make myself question everything yet again. I guess it turns out this was a corner of Koa syntax that still needed some rigorous investigation! ...and as to that final point, some additional soul-searching.