Apposition with nouns such as 'agent', 'patient' etc could replace case morphology

[kodé]

As I understand it, cases are basically just the practical realisation of semantic roles, kind of how phones are the realisation of phonemes. If your language is completely regular, each role should always get the same case marking.

Kind of, but it would be more accurate to say that semantic roles apply to deep structure, while case applies to surface structure.

There are all sorts of ways case and semantic role may be different:

* passive
* dative movement
* applicatives
* raising ("I want him to go")
* politeness (e.g. Spanish's personal infinitive, in form equivalent to the dative)
* other odd reversals of valence (Sp. me gusta María vs. Pt. eu gosto Maria)
* other arbitrary stuff (e.g. Russian acc > gen in negative sentences)
* whatever is going on in English "him and I"
* case attraction

(Also, FWIW, the original idea of semantic roles was that they were universal. That's speculative, but if you accept it, individual languages may have other odd case divergences.)

Also Quirky Case, like dative subjects in Icelandic (I need to reacquaint myself with this, since it's been several years since I read the literature on it).

One distinction that's commonly made in the generative literature is between "Structural Case" and "Inherent Case". The idea is that Structural Case is assigned (mainly) based on syntax, while Inherent Case is assigned (mostly) based on semantics. Nominative, Accusative and Absolutive (at least) are Structural Cases; I can't think of any languages where these Cases are used for fewer than three semantic roles (which I'm taking to be theta-roles). A typical Minimalist analysis of a transitive verb in a Nom/Acc language is that Accusative Case is given within the Verb Phrase and Nominative Case is given to the argument that moves to the subject position in the Tense Phrase; this argument is typically the more agent-y one, which is argued to start out higher in the syntactic tree.

OTOH, things like the Instrumental, Inessive, Comitative, etc. are Inherent Cases: they're not assigned based on the argument being a syntactic subject or object, and they (generally) have only one possible theta-role. IMO Cases like Dative and Ergative are somewhere in between: they each have their own typical theta roles (goal or experiencer for Dat, agent or causer for Erg), but they are also relevant to the syntax. Moreover, Pre/postpositions and Adjectives can sometimes assign a specific Case, which could either be more toward Structural or more toward Inherent depending on the phenomenon. I think the Ancient Greek thing where a single preposition could take a Noun Phrase in the Genitive, Dative or Accusative, each with a different meaning, is more toward Inherent Case. Other languages have a specific Case that must occur on pre/postpositional complements, such as Accusative.

It gets wilder in languages where object Noun Phrases are marked with different Cases based on whether they are animate or definite; quite a few IE languages mark animate direct objects with the Dative (this seems to have been innovated independently a few times). If you squint you can sorta see a semantic reason for this (the goal/experiencer theta roles are more "animate-y" than the patient/theme theta roles), but this is pretty clearly more on the syntactic side of things.

[Salmoneus]

I think this is philosophically suspect. Unless you assume a specific universal grammar, you cannot count semantic roles. If one language treats 'two' roles as one, and another language treats that 'one' role as two - or even if such a disagreement could conceivably occur in human language - there is no absolute perspective from which to assign truth and falsehood. Whether there are two roles or one role, or a thousand roles (almost all of which these two languages happen to fail to distinguish between) is not a matter of fact, only of analysis. Conceptually speaking, the role of every possible thing in each possible scenario is unique, and hence there are an infinite number of roles, semantically - any list of semantic roles smaller than infinite is making (necessaily subjective and/or relative) personal judgements as to which roles should be grouped together, and requires a specific analytical framework. Each possible language imposes its own analytical framework, and there is no absolute vantage point from which to discern which framework is 'most accurate' (because none of them are).

[a concrete example: different languges draw the semantic line between comitative and instrumental in different places, which necessitates that there is a semantic difference in role corresponding to each position for that line; likewise, instrumentals and ergatives, and comitatives and genitives and locatives and... ...and of course if all our grammars were written by native speakers of Iaai, we might have to be constantly reminded that all thirty-six (or however many) semantic roles relating to possession are covered by a single genitive case...]

Even if you discover empirically a universal grammar to provide an absolute reference point, it would only apply to real-world humans, not to language not set on earth.


I think it would be much more helpful to think of a language's case system as being a mapping between surface realisations (syntactic cases, as represented through word order behaviour and/or overt marking morphemes) and semantic territory (a mass noun, rather than a discrete and countable inventory of 'real' roles).



Saying that a language encodes semantic roles instead of cases is like saying that a language has no phonemes, only phones, or that your language has no words for phenomena, only for things-in-themselves.

[mèþru]

Well there can be rules for certain roles that don't care about cases, especially in animacy hierarchies

[Pabappa]

Instead of using verb inflections to form for example passive voice, we can just drop either argument. So for a simplified example:
AGENT Alice lick PATIENT Paul = "Alice licks Paul"
AGENT Alice lick = "Alice licks (someone)"
PATIENT Paul lick = "Paul gets licked (by someone)"

what would be the meaning of the sentence if neither the agent nor patient morphemes were present? To me this seems like a tripartite setup, as those are the languages that discretely mark both agents and patients.

[Chuma]

what would be the meaning of the sentence if neither the agent nor patient morphemes were present? To me this seems like a tripartite setup, as those are the languages that discretely mark both agents and patients.

In this case, that would mean "someone licks someone". Most likely, you'd be expected to know which someone from context, so you can think of it as prodropping:
AGENT Alice buy PATIENT ice-cream . lick = "Alice buys an ice-cream. She licks it."

Would it make sense to call it tripartite if there are only two cases? To me it seems most similar to a fluid-s system, but not quite the same as any attested system.

Since all verbs are effectively optionally transitive, there's a decision to be made for the ones that don't have an obvious transitive interpretation. What does "A sneeze B" mean – "A causes B to sneeze", or "A sneezes onto B"? I find that the former comes more naturally, partly because it makes sense to me to think of most intransitive actors as patients, and partly because it's always possible to add a "cause to" actor. So I call the cases absolutive and ergative, although that's a somewhat arbitrary choice.

EDIT: Oh wait, I might have misunderstood you – you mean if the noun phrases are present but not the case morphemes? In that case, they wouldn't be noun phrases; I guess you could say the case marking is fused with the noun marking, or something. So you can't really say that one case is more marked than the other.

[Pabappa]

you mean if the noun phrases are present but not the case morphemes? In that case, they wouldn't be noun phrases; I guess you could say the case marking is fused with the noun marking, or something. So you can't really say that one case is more marked than the other.

yes, thanks. I think it's unrealistic, or at least unattested, to have mandatory marking of both roles with non-fused morphemes, but looking back, you did say it was a logical language, so it makes sense that it might have a few unattested traits.

[Chuma]

Yes, I mean, sort of; it might not always be clear what's marked and what's not. Example, from German:
er ist der Hund "he is the dog"
er isst den Hund "he eats the dog"
Different cases, but you can't immediately see from this example which one is marked, right? A reasonable interpretation would be that there's one morpheme that says "accusative" and one that says "noun" (or "definite", I guess), but since they can be fused, there's no way of telling which particle is two morphemes.

Another example, which I found on WALS just now, from Latvian:
Bērn-s zīmē sun-i.
child-NOM draw.PRES.3 dog-ACC
"The child is drawing a dog."

WALS lists only 3 languages that have (only) marked nominative, compared to 61 that mark the accusative, but sadly doesn't comment on how many of the 61 mark the nominative as well.


Actually, this simplified example doesn't accurately describe my conlang; in fact, you can tell which one is unmarked, and surprise, it's... the dative. Pretty sure that's unattested, but it stems from another interesting observation about case: The form we think of as the unmarked one, the "dictionary form", is often not the "default" one. So the answer to "Who wants ice cream?" is typically not "I!" but "Me!". Zamenhof made the seemingly logical decision in Esperanto to use the nominative as the default, so you say "give it to I", etc. But as far as my grammar professor was aware, this is actually also unattested in natlangs – prepositional phrases will always take the marked form, if there is one.

In my conlang, the prototypical nominal phrase is marked by both a preposition / case particle and a determiner / pronoun, but the common combinations are typically fused (in completely unnaturalistic ways), so you can't tell that they're combinations unless you look them up in the grammar. The separate words are only used for rare combinations.

So I thought, should there be an unmarked case, and if so, which one? In effect, for my conlang, that comes down to: Which of the nominal particles should be officially declared as the actual determiner, rather than a prep/det combination, and therefore used with rare prepositions? Keep in mind, that "unmarked" case would not at all need to be any simpler than the others.

If there's no unmarked case, that would mean that the actual determiners would only get used in those rare cases, which seemed like a bad idea. At first I thought that Zamenhof's way made sense – use the most common, simplest form as the default. But then I figured, there's actually a point to having a less common form be the default. The most common case is likely to be present in the same clause, so using a less common case for an adjunct would add a little redundancy and reduce the risk of mishearing. It's kind of a middle ground between the Esperanto way and having a separate "adjunct case". English picks the second case – the object form – whereas my conlang takes it a step further to the third.

So the dative, which is phonetically more marked than abs/erg, becomes the technically "unmarked" form. Such is life in the maddening world of completely unnaturalistic conlangs.

[Richard W]

Actually, this simplified example doesn't accurately describe my conlang; in fact, you can tell which one is unmarked, and surprise, it's... the dative. Pretty sure that's unattested, but it stems from another interesting observation about case: The form we think of as the unmarked one, the "dictionary form", is often not the "default" one. So the answer to "Who wants ice cream?" is typically not "I!" but "Me!". Zamenhof made the seemingly logical decision in Esperanto to use the nominative as the default, so you say "give it to I", etc. But as far as my grammar professor was aware, this is actually also unattested in natlangs – prepositional phrases will always take the marked form, if there is one.

That paragraph's crawling with nits. I shall now pick at them.

In general, the unmarked form in English is the accusative, rather than the nominative. 'Who' v. 'whom' feels exceptional, but that probably reflects the importance of the register that lacks 'whom'.

You seem pretty sure that the Englisn possessive is not a case form. English prepositional phrases taking the possessive are fairly restricted.

Are you counting the zero suffix on Russian feminine and neuter genitive plurals as a marked form? There are Russian prepositions that use the genitive.

Persian has a definite accusative suffix. The objects of prepositions have no more marking than the subject of a sentence.

PIE seems to have had a locative singular with zero ending.

[zompist]

English prepositional phrases taking the possessive are fairly restricted.

I assume you're talking about pronouns. But we have two pronominal possessives. So far as I can see, forms like "your" can't occur after a preposition, but forms like "yours" can occur quite freely.

[Salmoneus]

English prepositional phrases taking the possessive are fairly restricted.

I assume you're talking about pronouns. But we have two pronominal possessives. So far as I can see, forms like "your" can't occur after a preposition, but forms like "yours" can occur quite freely.

That's not what "taking the possessive" means. Chuma is saying that prepositions always require their objects to be in a marked case if there is one. Richard is saying that in English, most prepositions do not 'take' (i.e. 'govern') the marked possessive. He's right. They do take the accusative, but I'm not sure if he's right in saying that the accusative is "the unmarked case", given that English case marking is by suppletion.

It's true that English prepositions CAN take the "yours" form, but that's because "yours" works like a substantivised form of the adjective "your", and as a result, like other substantive forms of adjectives in English, can go anywhere, even in subject position. "Yours was the pint, right?" It can even be made possessive - "that dog's head is orange but mine's is purple". I'm not sure that it's really best treated as a case-form at all - or at least, not always. I think maybe it's a case-form when it's a form of "your" used in predications, but not a case-form when it's used as a substantive.

[Chuma]

Hah, yes, technically Richard is of course correct in that the possessive/genitive forms are marked, more so than the accusative. It should be fairly obvious that I wasn't trying to claim otherwise.

So we should restate the rule. How about: "In accusative languages with a marked accusative, the default case for prepositional phrases will be the accusative." Is that true? What happens in ergative languages, and those with a marked nominative? True universals seem to be rare, so chances are there's some exception to this rule, too.

Flattering as the assumption may be, I don't actually speak Russian. I do speak Swedish and German, and those also have prepositions that govern the genitive (a quick googling also finds Czech and Polish), but it's only for a few pronouns. Didn't some older version of English do something similar? What about "a friend of mine"?

[Salmoneus]

I see a reference in an 18th century grammar saying that some prepositions govern the nominative in pronouns in French (although one of the ones listed is 'chez', which even to this non-French speaker seems to govern the genitive now); I see a claim that Finnish Romani has/had prepositions governing the nominative that are being/have been replaced due to contact with Finnish.

Ahha! Albanian has two prepositions that govern the nominative. ka/nga (from, to, at, out of - originally from a phrase meaning "in which this") and te (to, at, which wikipedia claims governs the nominative because it's only recently become a preposition).

[akam chinjir]

(although one of the ones listed is 'chez', which even to this non-French speaker seems to govern the genitive now)

Pretty sure "chez" gets the tonic pronouns ("chez moi").

[Salmoneus]

Oh, that's what that is? I know literally nothing about French.

[zompist]

English prepositional phrases taking the possessive are fairly restricted.

I assume you're talking about pronouns. But we have two pronominal possessives. So far as I can see, forms like "your" can't occur after a preposition, but forms like "yours" can occur quite freely.

That's not what "taking the possessive" means. Chuma is saying that prepositions always require their objects to be in a marked case if there is one. Richard is saying that in English, most prepositions do not 'take' (i.e. 'govern') the marked possessive.

Um... Richard said that English PPs taking the possessive* are "fairly restricted". The "fairly restricted" part surprised me, because I can't think of any instances where they can occur. If he had some examples, I'd like to see them.

*Using "possessive" to mean pro-Adj's like your just to keep the statement simple. I don't see any principled reason to consider the pro-NPs like yours to not be case forms.

I'm baffled by your first statement. What difference are you trying to establish between (say) "the objective case can occur after with" and "with governs the objective case"?

[zompist]

How about: "In accusative languages with a marked accusative, the default case for prepositional phrases will be the accusative." Is that true?

Nope— completely falls apart for Russian. Up to four cases may appear in PPs; not infrequently, the same preposition may be used with two different cases. Accusative is one of the possibilities, but by no means predominant.

[Salmoneus]

I assume you're talking about pronouns. But we have two pronominal possessives. So far as I can see, forms like "your" can't occur after a preposition, but forms like "yours" can occur quite freely.

That's not what "taking the possessive" means. Chuma is saying that prepositions always require their objects to be in a marked case if there is one. Richard is saying that in English, most prepositions do not 'take' (i.e. 'govern') the marked possessive.

Um... Richard said that English PPs taking the possessive* are "fairly restricted". The "fairly restricted" part surprised me, because I can't think of any instances where they can occur. If he had some examples, I'd like to see them.

Well, "of".. "A friend of his", not "a friend of him". Although sure, we need to take "fairly" here as litotic for "extremely"...

[I believe genitives are also found after some other prepositions in some non-standard dialects, but I don't have any data to hand]

*Using "possessive" to mean pro-Adj's like your just to keep the statement simple.

Well that seems to be reading more into what he said than is merited.

I don't see any principled reason to consider the pro-NPs like yours to not be case forms.

I'm baffled by your first statement. What difference are you trying to establish between (say) "the objective case can occur after with" and "with governs the objective case"?

Well, for example, it's correct to say that the plural can occur after 'with', but not correct to say that 'with' governs (or 'takes') the plural. "Can occur after" denotes possibility, whereas "governs" denotes necessity. [some pronouns are said to govern or take two different cases, but in those cases the choice of case determines the meaning of the pronoun - that is, effectively there are two different (semantically and syntactically) but homophonous pronouns]. Where the choice of case depends not on any property or meaning of the preposition, but purely on external factors, I would not say the preposition governs the choice the case.


In the case of substantive "yours" (synonymous with "your one" or "your ones"), this "case" is not governed by the preposition. Or, indeed, by anything syntactic - it can appear anywhere, and can in turn take the case-marking possessive clitic, strongly suggesting it's not "possessive" in the first place. Indeed, I can't see any reason to think that substantive "yours" is a case-form. It seems simply like a person-indexed noun form, and deixis is orthogonal to case, even if some forms of deixis can sometimes be marked through case.

There are even cases where the semantic difference can be seen. Compare the semantics of "when I bought her, Caesar's daughter became mine" with those of "after her molecules were divided by decay, floated through the ecosystem for two thousand years, and coincidentally reformed in my wife's womb, Caesar's daughter became mine". The former, the possessive, ascribes to Caesar's daughter a property, de dicto - whereas the latter, the deictic, ascribes to Caesar's daughter an identity de re. Similarly, when a person says "Chicago is your hometown but Beijing is mine", "mine" has a totally different meaning from when Genghis Khan cries "Beijing is mine!" - the former is the deictic (which hometown is your one?) while the latter is the possessive (who controls Beijing?). Or indeed take the phrase "a friend of mine" - it normally has a possessive sense (he is my friend), but where context redirects the deixis for us, both meanings are possible: "he may not be a friend of your wife, but he is a friend of mine". We can interpret this one sentence with either a possessive 'mine' or a deictic 'mine', and the referent is different - with the possessive, he is my friend, and with the deictic, he is my wife's friend.

There are two different senses that are encoded very similarly in many languages. But in English, they're not quite the same. The possessive is syntactically governed - there are places it cannot occur, and places it must occur. This is why the ambiguity arises in the above sentence - the possessive should 'logically' be just "a friend of me" as I(/me) am the the referent, but because 'of' governs the possessive, the possessive is imposed for syntactic reasons. As the possessive form of the pronoun is the same as the deictic form, ambiguity arises. This homophony does not occur, at least in standard English, with nouns, so the readings remain clear: "he's not a friend of your wife but he is a friend of Fred" vs "he's not a friend of your wife but he is a friend of Fred's" (i.e. of Fred's wife). Contrariwise, because the deictic is not syntactically governed, it can take subject position, which the possessive cannot. Thus, although both readings are available for "the cat is mine" (the cat belongs to me vs the cat is my thing), only the deictic reading is allowed (in modern English!) for "mine is the cat".

The possessive is obviously a case form of the personal pronoun - it's governed by "of" and cannot take subject position. The deictic pronoun, however, is not a case form, can go anywhere that any other noun can, and itself has its own possessive case-form. The possessive of the personal pronoun "I" is "mine" - "your dog's head is purple but mine is blue"; the possessive of the deictic pronoun "mine" is "mine's" - "your dog's head is purple but mine's is blue".



EDIT: well I explained that terribly, didn't I? I would go back and set it out more clearly from the beginning, but it's probably not worth it.

[zompist]

Richard said that English PPs taking the possessive* are "fairly restricted". The "fairly restricted" part surprised me, because I can't think of any instances where they can occur. If he had some examples, I'd like to see them.

Well, "of".. "A friend of his", not "a friend of him". Although sure, we need to take "fairly" here as litotic for "extremely"...

But that's the pro-NP, not the pro-Adj. Compare "a friend of hers / a friend of mine / a friend of ours", "*a friend of her / a friend of my / a friend of our."

"Can occur after" denotes possibility, whereas "governs" denotes necessity. [some pronouns are said to govern or take two different cases, but in those cases the choice of case determines the meaning of the pronoun - that is, effectively there are two different (semantically and syntactically) but homophonous pronouns]. Where the choice of case depends not on any property or meaning of the preposition, but purely on external factors, I would not say the preposition governs the choice the case.

But "can occur after" includes all instances of "governs". It's like if I complained about the animal on the table and you said "No, it's not an animal, it's a cat." Since I wasn't sure what phenomena Richard was talking about, I didn't want to assume that the genitive was required.

In the case of substantive "yours" (synonymous with "your one" or "your ones"), this "case" is not governed by the preposition. Or, indeed, by anything syntactic - it can appear anywhere, and can in turn take the case-marking possessive clitic, strongly suggesting it's not "possessive" in the first place. Indeed, I can't see any reason to think that substantive "yours" is a case-form. It seems simply like a person-indexed noun form, and deixis is orthogonal to case, even if some forms of deixis can sometimes be marked through case.

...That is, it's a pro-NP. An anaphor that can occur where an NP can. And of course it's possessive! It represents something owned by or associated with someone.

I'll grant you that the pro-NP isn't chosen by structural position (like an accusative) or simply by the preposition. It adds an inherent meaning (i.e. possession) of its own. But, well, so does the pro-Adj.

That is, sometimes all the meaning of a pronoun is its reference-- you can take an underlying structure "X gave Y some help on Z", and assign cases just based on argument structure, or specific governors (here "on"). On the other hand, if you have (say) Latin "responsio X", neither structure nor any governor tells you if its "responsio militis" (the soldier's answer) or "responsio militi" (an answer to the soldier). Case can absolutely add some extra information so the meaning is not just "X".

Edit: In other languages, we can decide that derivational processes apply to pronouns, creating forms that aren't simply case forms of the pronoun. E.g. Latin has a possessive adjective meus alongside the genitive mei. But that doesn't help us with English. Maybe it's useful to think of mine as derivational; but then why not my? I think my formulation-- that one is a pro-NP and the other is a pro-Adj-- is far simpler.

There are even cases where the semantic difference can be seen.[...] Similarly, when a person says "Chicago is your hometown but Beijing is mine", "mine" has a totally different meaning from when Genghis Khan cries "Beijing is mine!" - the former is the deictic (which hometown is your one?) while the latter is the possessive (who controls Beijing?). Or indeed take the phrase "a friend of mine" - it normally has a possessive sense (he is my friend), but where context redirects the deixis for us, both meanings are possible: "he may not be a friend of your wife, but he is a friend of mine". We can interpret this one sentence with either a possessive 'mine' or a deictic 'mine', and the referent is different - with the possessive, he is my friend, and with the deictic, he is my wife's friend.

Um, you're rediscovered the polysemy of case. All this happens in Latin and Greek, with nice names for each usage-- Genitive of Possession, etc. Note that there is no semantic difference between the pro-Adj and pro-PP here: "Beijing is my city", "he is my friend".

The deictic pronoun, however, is not a case form, can go anywhere that any other noun can, and itself has its own possessive case-form. The possessive of the personal pronoun "I" is "mine" - "your dog's head is purple but mine is blue"; the possessive of the deictic pronoun "mine" is "mine's" - "your dog's head is purple but mine's is blue".

I'm not sure this is valid IMD, and in any case doubled-up cases are not that weird, though you can't do it in Latin.

Can you say "..but the childrens's is blue"?

[Richard W]

Well, "of".. "A friend of his", not "a friend of him". Although sure, we need to take "fairly" here as litotic for "extremely"...

That's the example I had in mind, though I was think more of the use with nouns, as in a friend of Peter's roughly meaning one of Peter's friends and being a replacement for *a Peter's friend, which is ungrammatical unless the indefinite article applies to Peter.

[Richard W]

Can you say "..but the childrens's is blue"?

That falls foul of the rule that rather than have two identical suffixes, one is dropped, as in the 3s of the verb to McDonald's meaning to eat at or from McDonald's. It's inelegant at best. I think there's a possibility that the final sibilant will be lengthened, but that may be a performance issue.