Embedded clauses: the show so far

This is a big one.

I've been deliberately staying agnostic for the last 12 to 13 years, biding my time until I felt I had the wisdom or clarity to make a decision. Meanwhile, though, my interim strategies have been seeing so much use that they've actually been influencing other important choices that will not be easy to disentangle. It's clearly time to make up my mind.

The matter under consideration is that of embedded clauses. These fall into the two broad categories of modifiers (relative clauses) and nominals (complement and adverbial clauses), though as ever these categories are fluid in Koa.

Before diving into this discussion, I should first mention that one of Koa's relativization strategies, the internally-headed relative clause, is actually fairly uncontroversial. In this structure the particle ke marks the head while it remains in situ:

ke kane i si ipo ka sahi ni
QU man 3P PERF drink DEF wine 1SG
"the man who drank my wine"

le Keoni i si ipo ke sahi
NAME John 3P PERF drink QU wine
"the wine John drank"

nu ma asu ne ke talo
1PL IMPF dwell LOC QU house
"the house we live in"

If the head is focalized, note that this changes the sense from one of modification to identification:

ke kane sa si ipo ka sahi ni
QU man FOC PERF drink DEF wine 1SG
"which man drank my wine," as in "I don't know..."

ke sahi sa le Keoni si ipo
QU wine FOC NAME John PERF drink
"which wine John drank"

ke talo sa nu ma asu (ne ta)
QU house FOC 1PL IMPF dwell (LOC 3SG)
"which house we live in"

So far so good. This is easy to form, pretty unproblematic to parse, and works well much of the time. It feels a little odd to speakers of IE and neighboring languages, though, which is enough of a reason to have another option even without the fact that long chains of relative clauses can end up pretty unparseable using this strategy:

vo ke sene i si tapa ke hili i si suo ke lepa i tai ne le Iako i si tei ke talo
~ "this is which cat killed which mouse ate which bread was in Jack built which house"

What we need is a way for a clause to modify a nominal head that stays in its usual position within the matrix clause: that is, something more along the lines of a traditional Indo-European relative clause. This is easy when the head is the subject of the relative clause, because in that syntactic context the verb phrase can just as easily be considered adjectival anyway:

ka kane [ ma ipo ka sahi ni ] i ia paha
DEF man [ IMPF drink DEF wine 1SG ] 3P AFF evil
"the man drinking my wine is certainly evil"

We have the "relativizer" particle u at our disposal as well, which heretofore we've described as marking a phrase as both adjectival and pragmatically important, as in

ka sahi u puna
DEF wine REL red
"the red wine, the wine which is red, etc."

Our sample relative clause, then, could optionally also incorporate u:

ka kane [ u ma ipo ka sahi ni ] i ia paha
DEF man [ REL IMPF drink DEF wine 1SG ] 3P AFF evil
"the man who is drinking my wine is certainly evil"

I suspect that the difference between these would be about the same as the difference between "the man drinking my wine" and "the man who's drinking my wine" in English: in other words, pretty ethereal. One can envision contexts in which one or the other sounds better, but in general they're equivalent. The above is unproblematic, and indeed follows automatically from the basic principles of Koa structure.

Once the head occupies a position other than subject within the relative clause, though, we immediately run into apparently insoluble problems -- or at least, problems whose solutions have not seemed obvious to me for 12 to 13 years. Suppose, for example, that we want to say "The wine John drank had gone bad." Calquing the English structure would lead us to do something like this:

ka sahi [ u le Keoni i si ipo ] i si miláho
DEF wine [ REL NAME John 3P PERF drink ] 3P PERF INCEP-rot
"the wine John drank had gone bad"

It seems so normal that I might not even object, but then I remember the optionality of u in my previous examples. It was optional because the relative clause was "adjectival" on its own, with the same meaning. The same cannot be said here: there is no precedent anywhere in the language for a phrase like le Keoni i si ipo being able to directly modify anything.

Likewise, this clause doesn't seem to be modular, able to be slipped into any syntactic position, like all other parts of Koa are. For example, it should be possible to say this:

?ka [ le Keoni i si ipo ] i koa nai
DEF [ NAME John 3P PERF drink ] 3P good some
"the one John drank was pretty good"

...but I have no confidence in this at all. I don't see why I should expect that the bolded phrase should have the given English translation considering the remainder of Koa grammar, other than my English language intuition.

What we want is some way of forming a clause in Koa that would sound something like "the John-having-drunk(-it) wine" when translated literally into English. How would this be done?

Let's leave this for a moment and turn to the other embedded clause type. Complement and adverbial clauses occur in environments in which they are formally nominal -- that is to say, ordinarily one would see "nouns" in those contexts -- so a reasonable starting assumption might be that clauses of this type would have some kind of specifier. In fact, ko works very well for this purpose when, as with relative clauses, there isn't a subject expressed within the embedded phrase:

le Keoni i halu ko [ ipo a sahi koa ]
NAME John 3P want ABSTR [ drink INDEF wine good ]
"John wants to drink some good wine"

le Keoni i na ma mai koa lo ko [ si ipo a sahi pua ]
NAME John 3P NEG IMPF feel good REASON ABSTR [ PERF drink INDEF wine bad ]
"John doesn't feel well because he drank some bad wine"

The second example, which in English is an adverbial clause, might just as well be translated "John doesn't feel well because of having drunk some bad wine," and as such demonstrates why ko is the appropriate particle: ko si ipo a sahi pua really does mean "[the idea/state of] having drunk some bad wine." It's a little more of a stretch as a complement clause: I'm not sure it's as obvious that the literal "John wants [ drinking some good wine ]" ought to have the given meaning. Nonetheless, it's the only remotely reasonable way of doing this that I've ever come up with.

Having used this kind of structure for some time now, I found it easy enough to start regarding ko as a kind of complementizer, having in its scope an entire following clause. It seemed like this was a reasonable extension of its usual role of marking abstract concepts. Using it this way, we might see phrases like this:

ni si kulu ko [ le Keoni i na ma mai koa ]
1SG PERF hear ABSTR [ NAME John 3P NEG IMPF feel good ]
"I heard that John isn't feeling well"

You may be noticing a similarity developing with what happens with relative clauses. The problem is that ko marks the abstraction of a root. That's its one function. Every particle in Koa has one function. In using it this way I've done a very natural, linguistically neutral thing, but a fundamentally very un-Koa thing. Given the meaning of ko everywhere else, does it make any sense to express "John not feeling well" as ko le Keoni i na ma mai koa, in the same way that ko puna means "redness?" I'm not convinced that it does.

Scope is definitely part of this feeling. Even though the languages I know best don't have any problem parsing the appropriate scope of a complementizer, I feel uncomfortable assuming that everyone should just understand where this ko-phrase ends.

Another spot of discomfort is in the fact that, by preposing ko, I'm making this clause into a nominal. The clause, especially with that i in there, feels awfully finite for a nominalization.

This also fails the modularity test. If ko mevúa means "raininess," and pai mevúa means "rainy day," we should be able to say:

pai [ le Keoni i na ma mai koa ]
day [ NAME John 3P NEG IMPF feel good ]
"A John-not-feeling-well day"

Similarly, since I can say ti pai i mevúa "this day is rainy," why not:

ti pai i [ le Keoni i na ma mai koa ]
this day 3P [ NAME John 3P NEG IMPF feel good ]
"this day is John-not-feeling-well-y"

I don't think either of these is very well motivated. Although I wish I could clearly articulate why, my instinct is strong enough that I don't think I can use this kind of structure moving forward...unless I decide that it's okay for ko to lead a double life as a complementizer, in which its structures are not modular in the same way as other sort-of nominals.

If I'm opening up that line of inquiry, there's also the option of using one of my few remaining particles as a bona fide complementizer: perhaps ve, in homage to Bislama:

ni si kulu ve [ le Keoni i na ma mai koa ]
1SG PERF hear COMP [ NAME John 3P NEG IMPF feel good ]
"I heard that John isn't feeling well"

and even

pai ve [ le Keoni i na ma mai koa ]
day COMP [ NAME John 3P NEG IMPF feel good ]
"A John-not-feeling-well day"

It doesn't seem to work as well without an overt subject on the embedded clause, I suppose because what follows ve here is supposed to be a fully formed finite expression. We'd need to use structures like

...lo ve [ ta si ipo a sahi pua ]
REASON COMP [ PERF drink INDEF wine bad ]
"...because he drank some bad wine"

which seems to work reasonably well. What about this, though?

?le Keoni i halu ve [ ta ipo a sahi koa ]
NAME John 3P want COMP [ 3SG drink INDEF wine good ]
"John wants to drink some good wine"

Not so much. Whether we read this as "John wants himself to drink..." or "John wants him to drink," it's not inspiring any applause. I guess its structure is parallel to the same kind of clause in Greek/Romanian/Bulgarian/etc., though:

ο Γιάννης θέλει να πιει καλό κρασί
DEF John want-3SG COMP drink-3SG good wine
"John wants to drink good wine"

Anyway, before looking much more closely at that kind of strategy, or giving up on my principles, I would like to see if it might be possible to come up with a way of doing all this that really does work the way I had been envisioning. This is already a ridiculously long post, so we'll go on to that in the next one.