Backing up: the illusion of null derivation in Koa

The following is a slightly modified transcript of a correspondence with Allison regarding the previous post.

I think I should clarify something that I've clearly been too lax about explaining in the past: this "singer" issue that I know people have had problems with intuitively.

The thing about Koa is that it doesn't actually have parts of speech that correspond to European languages at all. I keep cavalierly using terms like "nominalization" as if it were entirely clear to everyone else exactly what I mean, but in fact you don't really have nominalization in Koa. In the same way, there's no verbalization, or "adjectivalization": just words used as predicates or modifiers.

My suggesting that ka lalu is the nominalized form of the verb lalu, then, is actually almost completely unhelpful. I'm realizing that it goes beyond that to the point of being positively deceptive.

The idea behind "words" in Koa is that they take their apparent lexical class -- their sense -- from the place they're used in a clause, but that none of this is inherent. If we take a simple clause like ka kane mata i ma luke "the short man is/was reading," we could equally visualize the "nominal" here as

a noun: the man
an adjective: the male one
a verb: the one-who-is-male, the one being male

likewise, the "adjective" mata could be seen as

an adjective: short
a verb: who-is-short, being short, "shorting"
a noun: a short one (appositive)

and the "predicate" could be framed as

a verb: is reading
a noun: is being a reader
an adjective: is a reading one, is one who is reading

The above is true for Koa not just in theory but in practice. There is unavoidably a question of arbitrariness in terms of what arrangement of semantic roles to foist onto each lexeme in order to make these interpretations what they are, and this is something I want to talk about more in a minute. There is not, however, anything arbitrary about the way the words resolve into each apparent lexical class once you know their basic meaning.

This is all building to an attempt to get at the feeling that it doesn't make intuitive sense for the "nominalized" form of "sing" to mean "singer." I think it's important to note that it actually doesn't mean "singer" with any of the semantic or aspectual/modal sense of the English word. It might be helpful to look at what the verb lalu actually means in a sentence like le Keoni i lalu. I've translated it as "John sings," but that statement is aspectually ambiguous in English.

The best example I can come up with is in the context of, say, a party, at which a certain portion of the contingent has decided it's time to start singing some songs. The folks who want to sing start asking less-well-known attendees whether they'd like to participate by saying "Do you sing?" to which they might respond, "Yeah, I sing" or "No, I don't sing." In Koa, these would get translated as Ai se lalu? Ia, ni lalu and Na, ni na lalu. In a sense, then, le Keoni i lalu might be more helpfully translated as "John is willing to sing," or "John can sing." The assertion is that John has the general potential to sing, whatever the specific realization of that fact in context might be.

I tend to translate ka into English as "the." It's more accurate to say that ka placed before a Koa predicate ("predicate" just meaning "content word" in the Koa grammatical tradition) gives it the sense "a definite instantiation of the meaning of the root," in other words "the one which..." From i puna "is red," then, we get ka puna "the one which is red," "the red one." In the same way, i lalu "sings" gives us ka lalu "the one which can/will sing, the one which sings, the 'singer'." Looking at it another way, i lalu could be equally correctly translated as "is one who can/will sing, is one who sings, is a 'singer'."

Given the semantics of a particular predicate, then, there is only one possible meaning that it could have in its nominal, adjectival or verbal role. There's no question of what semantic role to "promote" during the apparent process of nominalization: it's the same semantic role that it has everywhere else as well. What I've been calling null derivation should really be called "apparent null derivation," because there isn't actually any derivation going on here. Traditional concepts of lexical class just don't apply.

At least, that's the way it's always been; what I brought up in my previous post was the possibility of adding another layer of arbitrariness that would firmly reintroduce traditional lexical classes into the language for the theoretical benefit of greater intuitiveness, and/or greater word-worthiness of basic roots in each lexical class guise. Or reducing the average number of morphemes per word to give the language more of the feel of a creole. The disadvantage, beyond the increase in arbitrariness itself, would be that it would break the carefully constructed elegant system above.

When it comes to which thematic roles to map onto arguments for a given word, of course it's unavoidably true that it's philosophically an arbitrary choice. My guiding strategy, though, is based in the assumption that pragmatically most of that theoretical arbitrariness disappears. Once in a while one sees a language make some bizarre encoding choices -- Maltese where "thief" means literally "one who is considered a thief," for example, being my favorite -- but by and large this isn't what I've seen languages do. Dogs are nominal. Killing is verbal, and the agent will be the subject. Inasmuch as a language has a class anything like adjectives, "big" will be an adjective. I would suggest that, statistically, there are good and bad choices where these assignments are concerned.

Where there is genuine disagreement across large chunks of the language spectrum, I've tried to make my decisions at least consistent and predictable. Experiencers are encoded as subjects in Koa, for example. Bodily substances are their own nominal subjects (i.e. i taku means "is blood" not "bleeds"). I do actually intend to present the dictionary in a somewhat Loglanny way, demonstrating a clause frame for each word to illustrate the semantic structure; ideally my choices would feel intuitive or at least reasonable to the greatest percentage of humans, and where there is guesswork for a learner, there is at least a system on which to lean. This kind of arbitrariness is, I feel, pretty distinct from Esperanto's where kombo is the verbal noun "combing" but broŝo is the instrument "brush."

I've been relying on my intuition based on a very large number of studied languages in making these choices, but I really ought to be more scientific about it. In general I'm being pretty agnostic about the thematic role/argument deployment of potentially problematic words, with the understanding that I'll firm that up (or not -- some ambiguity is probably okay) later on through philological investigation.

I wouldn't want anyone to think I'm getting caught up in my own idea or anything. I'm firmly grounded in the fact that IAL design is always fundamentally going to be a variety of intellectual navel-gazing. The process is fascinating to me, though, and I really do have the conceit of thinking, or at least hoping, that this particular language, when it's "finished," would be easier to learn for a much greater number of people than Esperanto -- the founding goal back in 1999.

The question I was posing in the previous post, then, was that of whether this goal would be best served by coherence to an established and predictable, though more complex, system, or by the addition of ambiguity in order that more forms might be morphologically simpler and more typologically neutral. I'm leaning strongly towards the former at this point, if only for the reason that Koa was designed from the bottom up with the existing system informing every choice, and this change would destroy all internal consistency. It would be better, in terms of optimality, to start over from scratch if this were going to be a primary design goal.