chris_notts wrote:Although the definition of subject is itself a bit hairy! If we take subject to merely mean the primary MAP, then this statement is a tautology.
I suppose so.
The working "definition" of "subject" I came up with myself and have usually been using is the "syntactically most-privileged argument" and/or the "most syntactically-privileged argument". If "most privileged" is what "primary" means for MAPs, then yes, it's nearly* a tautology to say that any language that has MAPs has subjects. *(You also have to assume that some clauses have only one MAP, and/or that for some one MAP is clearly more privileged than any others. But I think that's a safe assumption.)
The idea of Subjects as "core arguments", to me, means that their morphosyntax is mostly about their
syntactic position (that is, syntactic relation to the clause and/or the verb) as Subjects, and not so much about their semantic roles. Their case-endings (if they have any) and adpositions (if they have any) just tell that they're Syntactic Subjects; what semantic role they play is indicated by morphosyntax of the verb and VP rather than by the Subject's NP.
"Oblique arguments", OTOH, have a morphosyntax that's mostly about their semantic roles; their case endings (if they have any) and adpositions (if they have any) are mostly about their
semantic relation to other NPs, to the verb, to the clause, and to other clause elements; rather than to their
syntactic position in the clause (which can usually be summed up as "oblique argument").
The difference between core and oblique is likely IMO to be language-dependent; it's also IMO likely to be fuzzy. For example, there are some very well-known and well-studied languages about which it is still undecided whether they have 2 or 3 MAPs. "Secondary Object" or "Indirect Object" is, in many ways, a lot more like "Oblique Argument" than either "Subject" or "Primary or Direct Object" are.
There are people, inclulding some on the ZBB, who find the distinction between "Argument" and "Adjunct" fuzzy, too.
chris_notts wrote:If we take subject to actually have some specific semantic and syntactic properties associated with it, then your statement generally tends to be false unless we take the smorgasbord approach of defining a long list of properties we'd like subjects to have and accepting anything that covers at least some of them. Some of my recent reading (books on Construction Grammar in particular) have made me doubt whether this is the right thing to do.
There are many things in Linguistics and other fields that I think are best "defined" by a "Chinese-menu-type" or "DSM-IV-type" "definition". For instance, some physical disorders are diagnosed by listing 14 properties and saying anything that has 7 or more of those properties is that disorder.
IMO "Syntactic Subject" is one of these. I consider "Syntactic Subject", which is what I almost always mean when I say "Subject", to be best "defined" by Keenan's "Subject Properties List"; in any clause the participant which has a majority of those properties, and has at least as more of them than any other participant, is that clause's "Syntactic Subject". (If there is no such participant than there may be no "syntactic subject" for that clause; or, that clause may have more than one participant that are equally good candidates to be considered the "syntactic subject", in which case inter-clausal interactions may determine the issue (or may not).) Most of those properties are not semantic, and most are not pragmatic; most are strictly syntactic or almost-strictly syntactic.
The "smorgasbord approach" for Syntactic Subjecthood is a "smorgasbord" only cross-linguistically, AIUI. Each particular language, if it has subjects, has a particular subset of Keenan's properties list (or of some possibly expanded list perhaps including some properties Keenan omitted) that are relevant to subjecthood
in that language; and an ordering by relevance of those properties. That subset and that order are, AIUI, fixed
for that language. But they vary from language to language.
(As you can tell, that definition isn't precisely the same as "most syntactically-privileged argument or syntactically most-privileged argument". But I'd say it's close.)
Other linguistic ideas that apparently need a "smorgasbord" "definition" include Dowty's Proto-Agent Macro-Role and his Proto-Patient Macro-Role. He started with a list of four properties for each, and quickly expanded each list to five properties; since then other people (e.g. Jackendoff et al.) have expanded them to about eight properties each. Any participant that has at least as many proto-agent properties as any other is a good candidate to be the Agent (or at least the Actor) of the clause; any that has at least as many proto-patient properties as any other is a good candidate to be the Patient (or at least the Undergoer) of the clause. If a participant has more Proto-Agent properties than it lacks, and also has more than any other participant, then it is the Agent or Actor; sim for Proto-Patient properties and Patient or Undergoer.
A participant could be both the one that has more Proto-Agent properties than any other participant, and also the one that has more Proto-Patient properties than any other. I call such a participant the Semantic
Subject unless it's the only participant, in which case I call it the
Sole participant. That's possibly different from the Syntactic Subject.
The Proto-Agent, if there is one and only one, is usually the participant which is most controls or performs or effects or instigates the situation described in the clause, or acts most volitionally, or acts most directly upon the Proto-Patient; provided those ideas have any descriptive bearing on that particular clause. The Proto-Patient, if there is one and only one, is usually the participant which is most affected, particularly the one most visibly affected or most physically affected, by the situation described by the clause; again, provided those ideas have any descriptive bearing on that particular clause.
So if a clause with more than one participant has a Semantic Subject, it's describing a situation which is kind-of-sort-of reflexive or middle-diathesis or introversive; the same participant is most controlling (etc.) and most affected (etc.).
Primus came up with the idea of a Proto-Recipient (I think that's a bad name). If a participant other than the Agent/Actor/Patient/Undergoer/Subject is "highly involved", which basically means the sum of its Proto-Agent properties and its Proto-Patient properties is higher than any other participant with the possible exceptions of the Syntactic Subject and/or Proto-Agent and/or Proto-Patient, then, she calls it the Proto-Recipient. "Proto-Recipient" is another semantic term. It corresponds with what Rick Morneau calls "the Focus" and Barry Blake calls "dative"; it's highly, even crucially, involved, but is neither the most controlling nor the most affected. Recipients are entities conscious of being affected; and their participation is at least consistent with their volitional participation. So if the Recipient isn't the Proto-Patient it's usually the Proto-Recipient.
Well, in fact applicatives don't necessarily promote arguments to be THE direct object (or whatever the equivalent is - do syntactically ergative languages have a direct object MAP?), just A direct object. As I mentioned in my post, there are plenty of languages where the new object and the 'base' object appear to be more or less syntactically equivalent, so the 'base' object is not displaced. In your terminology, this would mean you have more than one argument occupying the same MAP. Even more confusing is when object properties are split between the base and applicative object, in which case what you have presumably is two new MAPs that make up the old MAP in terms of syntactic and morphological behaviour when added together.
I'm sorry I failed to fully absorb that on first reading; I'll read it again.
I do not believe applicativization and ergative alignment go together, do they? Are their any uncontroversially ergative languages that have a process that's uncontroversially applicativization?
Anyway.
Yes, there appear to be languages that have a non-promoting passivization, where the passive-voice verb still "has the same object" as its active-voice counterpart, but "has no subject". The question is, by which meaning of "subject" and which meaning of "object" is that participant still an object and not a subject? By the definition "most syntactically-privileged NP" it must now be the "syntactic subject". By Keenan's Subject Properties List, it may, or may not, now be the "Syntactic Subject". By the morphosyntax of the verb and of the object NP, though, it may still be an Object.
Also, there are sentences in some languages that some grammarians have analyzed as having
two Subjects. But many of these have also been alternatively analyzed as Topic-Comment sentences; one "subject" is the Topic, and the other is the subject of the Comment clause.
And, in some languages, ditransitive clauses appear to have two Objects of equal or nearly-equal rank. There's sometimes little or nothing in the morphosyntax to distinguish between them. My possibly-incorrect impression is that most MAP theorists would be able to find subtle reasons one was Primary or Direct and the other Secondary or Indirect, but I personally wouldn't know how, and I could be wrong about them too.
I know that some MAP theorists have analyzed some languages as having four MAPs; one Subject, one Primary Object, and two Secondary Objects of different kinds, (the way your examples have two or more (Direct?) Objects of different kinds).
Until you mentioned your examples I'd never heard of a language with a grammatical clause that had
three or more Primary or Direct Objects, nor three or more Secondary or Indirect Objects.
But if non-demoting applicativization can be applied twice to a clause that already had a Primary or Direct Object to begin with, especially in a 2-MAP language (if there even is such a thing -- MAP theorists think so, but I suppose there must be linguists who think "all MAP theory is total bullshit"), the resulting sentence would have three Primary or Direct Objects; and, depending on how applicativization works in that language, there might be nothing in the verb's morphology or the noun-phrases' morphology, and little or nothing in the syntax of the clause, to distinguish one of them as being syntactically Primary or Direct or Secondary or Indirect, in contrast to the others.
See above for why this number may be large. One of the examples Peterson uses, Hakha Lai, has multiple applicative constructions, with different divisions of object properties for different constructions. The properties which define objects in monotransitive clauses don't divide up neatly in the same way across all applicative constructions.
And in fact, I believe that in most languages you could define many more than two or three MAPs if you really wanted to, especially if any particular MAP need not occur in all clauses (which you must accept as soon as you argue for more than one MAP, since there are clearly intransitive verbs). There are always subtle criteria which can be used to split further if you want to.
I'm not really sure that's not due to a misunderstanding on someone's part; likely yours or Peterson's, maybe someone else's (unless
I'm the one that's not getting it!)
AIUI anything in a MAP is a core arguments, and any core argument is in an MAP.
Anything occupying a morphosyntactically-assigned argument position must be an argument, of course; "argument" is what the "A" in "MAP" stands for.
The fact that it's a "position" rather than a semantic role, and that it's "morphosyntactically assigned" rather than pragmatically or semantically assigned, is what makes it a "core position".
I think, maybe, (I'll re-read to check whether I'm talking through my hat now), most of what you're looking at in those subtle distinctions, are more pragmatic and/or semantic than morphosyntactic; and/or, that they apply not to morphosyntactic "positions", but rather to semantic or pragmatic roles or relations.
(If I've misunderstood I'd love to have it cleared up! It would be a rewarding "learning experience".)
However:
I remember reading someone's claim that some North American Native language had grammatical clauses whose verbs had to agree with seven participants. If "the verb has to agree with whatever's in that slot" qualifies as an MAP -- and why shouldn't it? -- then, if that's what that author actually claimed, and if his/her claim was actually true, the stated upper limit of at most 4 MAPs per clause per language would be exceeded. I don't know.
As for "What is a (Syntactic) Object (of a clause)?", one of Trask's definitions -- the one I think is most pertinent here -- is something like "Any ((core argument) or MAP) that isn't the Syntactic Subject". AIUI MAP theory then says a (syntactic) Primary Object is any syntactic Object that's more syntactically-privileged than any other Object; and a (syntactic) Secondary Object is any syntactic Object that isn't a Primary Object.
OTOH you could use
Keenan's Correlates of the Absolutive. You could say: "Any participant which has more of these properties than it lacks, and also has more than any other participant, is either the Absolutive Subject or the Direct (or Primary) Object, depending on alignment and on what's happening to the Syntactic Subject and on the clause's case-frame."
If these properties are split up among the non-Subject participants of a clause; or if more than one of them ties at having the most; then maybe the clause has no Primary/Direct Object, or has more than one Primary/Direct Object. I do not know of a theoretical dogma that there must be one and ony one such Primary or Direct Object of any clause that has a core participant or MAP that isn't a or the Subject.
(MAP theory would require that if there is an occupied MAP other than the Subject MAP, then at least one of them must be a or the Primary or Direct Object; but AFAIK it allows more than one, and AFAIK the "Keenan Correlates of the Absolutive" don't require that there be even one.)
Obviously, this is trivially true for languages with two or less MAPs.
If the starting and ending positions of the promoted and demoted arguments must be MAPs, it's trivially true even for languages with three MAPs.
(Causativization involves putting something that wasn't in the original clause at all, into the top MAP. If it were put into the middle MAP instead, that would be a counter-example. Likewise, if an oblique were promoted into the middle MAP. But, unless there are at least four MAPs, you can't promote or demote anything from one non-top non-bottom MAP into another non-top non-bottom MAP.)
I'm not sure, though, that I correctly stated the remark. It may have been that promotion and demotion had to start at the top or bottom slot; or that promotion and demotion had to end at the top or bottom slot; or some other formula having an "and" where I said "or" or "or" where I said "and".
Given their definition of MAP, is their sample size of languages with >= 3 MAPs large enough to make any significant conclusions?
I don't know, but I think so.
You may recall that when Comrie and Keenan first proposed the Noun-Phrase Accessibility Hierarchy (which governs which NPs in a clause can be relativized), they thought it also governed which "slots" had to be filled and which "slots" a demoted NP could wind up in. They thought that each clause's nominals had to fill a top segment of that hierarchy, and that when any valency-changing or valency-rearranging operation demoted a participant, it had to be demoted to the highest empty position.
It's now accepted that that's true only of a minority of languages. I don't know what evidence makes that be "now accepted"; but I suspect it involves several 3-MAP languages, since "Indirect Object" is the third item on the NPAH.
Also, Comrie was the one who published evidence and arguments for saying that many languages don't have a third MAP; or, truer to his actual words, that there's no convincing reason to decide that "Indirect Object" should be a separate category in these languages. So, I doubt that whatever evidence makes some of Comrie's and Keenan's initial conjectures about the NPAH no-longer-accepted, would pass Comries (nor even Keenan's) eye unless it included a suitably large sample of languages that uncontroversially do have three MAPs.
Also, do they have robust definitions for syntactically more or less privileged?
I don't know, but I'm going to guess.
If you don't accept "smorgasbords" as "robust", I'm guessing the answer is "no".
Otherwise, I'm guessing the answer depends on who "they" are. I don't know of one, other than the various "smorgasbords" I've referred to in this post, and
maybe others that haven't come to mind at the moment.
See above. There are languages where it's not obvious that anything has necessarily been demoted, and where applicatives are strictly valence increasing.
I will have to do a better job of "seeing above".
I guess a key part of this thread is "what counts as oblique in the first place?".
Yes; I was trying to say "core argument" is "any argument in an MAP" and "oblique argument" is "any argument that's not in an MAP". You can re-work that definition to still mean something without accepting any of MAP theory. However you have to either assume that people already know what an argument is, or you have to define it.
As I've been implying, I do not know, or haven't made up my mind about, the answer to your OP. But, I strongly suspect the answer is:
1. The distinction occurs in almost all languages, if not all;
2. The particular definition of the distinction is partly language-specific;
3. In some or many languages the distinction is more-or-less fuzzy.
I think "Topic vs Comment" and "Focus vs Ground" are likelier to be widely-acceptable as "universal" distinctions than "core vs oblique".
Well, language does have centralising tendencies, obvious. People like to give similar things similar encodings, because obviously we have limited bandwidth, memory and processing power.
...
Well, as mentioned above, another perfectly viable solution is just to have more clauses with fewer arguments. This appears to be the case in some american indian languages, where very few clauses have more than one or two overt arguments. This is partly what I was hoping to read more about in the article by Mithun.
Quoted just to say I think you're right.
A similar solution occurs in languages that like serial verb constructions - normally the SVC is considered to be a single clause, but the fact remains that in many such languages each verb in the clause can potentially introduce an additional argument. In a language which purely uses SVCs, you don't need any verbs that can have more than two arguments, because you can just glue verbs together to get more.
And in symmetric SVCs at least, the object-like arguments introduced by the various verbs are more or less syntactically equal when it comes to the syntactic criteria typically used to define things like MAPs.
I'd love to know more about that.
Everything I've read about SVCs, I've read because you recommended it.
I can see why what you're saying makes sense.