Ergative-Accusative MSA (aka Tripartite Case)

[Carson]

It's been a while since I've lurked, let alone posted, here so I decided it was time to make a blip and share some stuff that I learned this year in my linguistic typology class.

I've been working on a conlang for some time, now. A few years, off and on, never adding very much at a time. One of the last additions to the grammar I made early last year was tripartition in its case marking system, i.e. marking Agents, Patients, and intransitive Subjects uniquely (rather than uniting two of them as do most languages). I found out after making this change that Frommer had also done this with Na'vi (which displeased kitty). Anyway, I presented this state of the grammar to my professor, who immediately challenged my decision to use Tripartition. Besides "'cause it's cool," I had no answer.

Enter the research phase.

As it happened, the next class after that discussion was all about ergativity, following the works of Dixon, Comrie, and others. After the usual definition of 'Ergative' was some discussion of split systems, and the tendency for ergative languages to incorporate some degree of accusativity. I did some independent research based on what I learned in class, which led me to the Ergative-Accusative morphosyntactic alignment system. This system characterizes a number of Australian languages, as well as a few Caucasian and Dravidian languages (and some others, no doubt). My research eventually focused on the West Australian language Warrongo (via Tsunoda), which is a geographic neighbor to the more familiar Djirbal (the poster child of Ergativity, mentioned in the same breath as Basque 8/10 times).

I mention Warrongo mostly because Tsunoda's impressively complete work on the language and culture pointed me to a 1974-ish paper by Silverstein. His work was what really unlocked split-systems for me, and will be the focus of what I describe to you.

Silverstein's NP hierarchy is an early description of the interaction between Person, Animacy, Number, and Case across languages. The basics are thus: Animacy follows a hierarchy wherein the First Person is at the top, followed by 2nd then 3rd, then whatever is the next lowest for that language, ending with inanimate/nonreal objects. You can already see how Person and Animacy are interrelated as the two are incorporated into one hierarchy. In Warrongo, there are three Persons in three Numbers each in the pronominal paradigm, as well as three Nominal classes which fit into different parts of the hierarchy. In my conlang, it so happens that there is a verb ending for each step down the hierarchy (I was still in high school at the time and didn't even know what linguistic typology is, let alone who on Earth is Silverstein). One last note of importance is that Silverstein observed Accusativity rolling down the Animacy hierachy (i.e. 1st person pronouns are the most likely to be Accusative cross linguistically), while Ergativity climbs up the hierarchy (thus languages with any degree of Ergativity are most likely to mark full NPs that do not denote living people as Ergative).

Warrongo, like Djirbal, Warlpiri, and Wangkumara, exhibits Tripartition in its pronouns and some noun classes, but not for all NPs. My first question, of course, was why only pronouns? It has to do with that animacy hierarchy. Pronouns are inherently high on the animacy scale because they more often than not refer to living creatures, and especially to living humans. Even English 'it' is used quite often for animals and bugs when their sex is not known (but oddly never for humans, even when their sex is not apparent). Pronouns, then, are far more frequently subject to Tripartition.

But why is there Tripartition in the first place? Whence comes the mixed Ergative-Accusative MSA?

In languages with split-ergativity, there is a "pivot." This is a particular feature which when negative licenses Accusativity and when positive licenses Ergativity. In Tamil (iirc), there is a pivot on the past tense. In Warrongo, and I think in most split languages (pending further research), the pivot is somewhere on the animacy hiearchy--specifically, at the 3rd person PL pronouns, beginning with DU. Interestingly, Warrongo's 3rd DU and PL are expected to pattern Accusatively, but have the OPTION of patterning Ergatively, ie marking them with Ergative case would be perceived as grammatical by a native (if you could find one; sadly, Palmer and Tsunoda may be the last people to speak the language). Also, Kinship terms are expected to pattern Ergatively, but have the OPTION to pattern Accusatively. In this space where you find optionality is where you find Tripatition. In other words, if the point on the hierarchy at which Accusativity stops is below the point at which Ergativity begins, the space between the two points will show morphological Tripartition.

For another example, in my conlang all nominals (including pronouns) which denote a biological thing (meaning it is an organism or product of an organism which takes in nutrients and excretes waste [automobiles don't exist in the surrounding conworld, so that bridge hasn't yet been reached]) are expected to pattern Accusatively. At the same time, any nominal which does not denote one of the participants in the speech act (i.e. 1st and 2nd person) is expected to pattern Ergatively. This creates a pretty broad range in which Tripartition can occur, thus living up to my early hopes for a Tripartite language while making enough concessions to reality that my professor can nod his head and go, "that's very interesting" in his posh Cambridge accent. 3rd person pronouns and NPs which denote any plants, animals, or humanoid creatures all exhibit Tripartition.

I have not looked much at how Frommer handled Tripartition in Na'vi; this is partly because technical documentation proved harder to find than I was willing to contend with while taking 17 units, and partly because I didn't want his solution to color my own solution (though I expect they're similar anyway--if I, an undergrad, found Silverstein's work, I'm sure a teaching professor with a lot of credentials in linguistics would have found it, as well). I thus welcome any discussion from anyone familiar with that conlang.

It is my hope that this information will contribute to understanding of split systems and aid anyone who is considering making a tripartite or split system for their conlang, as it helped me.

A couple end notes: pivots need not be on Animacy. As I said earlier, Tamil (iirc) pivots on the featue [+/-Past]. Dixon (1994) alluded to various Mood/Aspect pivots being attested here and there, as well. As conlangers, I'm sure this crowd will come up with some interesting ways to split morphosyntax. To get tripartition, all you need do is have that pivot not be a point but rather a range. Djirbal, for instance, also pivots on animacy but has no overlap and thus could not be truthfully called Tripartite (though Ergative-Accusative might apply); it pivots on a point (iirc, on third person, or Silverstein's [+/-Participant] feature). Warrongo, on the other hand, has a range where the pivot is present and, in that range, exhibits Tripartition.

[Radius Solis]

Split ergativity is always interesting, yes. I have not previously heard the term "pivot" used that way, though; usually I see "pivot" used as short for the syntactic pivot. Aside from that, what you've given is the classic description - that "split ergativity" involves a split based on either of two grammatical features, hierarchical animacy and aspect/tense.

Thomas Payne, in Describing Morphosyntax, adds another feature to the list by rolling active-stative languages into the same category: these languages can also be described as split ergative, but with the split based on the semantic role of the subject. There are two subtypes of this - 1. "split-S", where some verbs select an agent-like subject and an accusative alignment and the rest select a patient-like subject and an ergative alignment, this feature lexically built in to each verb - and 2. "fluid-S", where the split is based on whether the subject is an agent or a patient, regardless of the verb, in any given instance. Usually active-stative languages are treated separately from other forms of split ergativity, but after reading Payne I've never been able to fathom why. Their underlying logic is quite akin to the animacy-based splits. It's the tense/aspect splitting languages that are the odd man out.

Within the animacy-splitting languages, your treatment basically matches Payne's. Accusative alignment starts at the top of the hierarchy and is used for some ways down it, and ergative alignment starts at the bottom and is used for some ways up it. Typically where one stops the other begins. But there are two other possibilities - first, when the two alignments overlap each other so that both are in play over some middle range of the hierarchy, then that range is tripartite. Second, it's also possible for the two not to even meet, leaving a middle range that is not aligned in either manner - this is sometimes called the Monster Raving Loony system because no known language does it universally, it only shows up in the middle range of some animacy-based split systems.

[Carson]

... Monster Raving Looney system...

I love it!

I'll take a stab at why Active-Stative languages are treated differently from other Ergative languages. In what I've come to understand, Ergativity is something of a spectrum. At one end you have Latin with Accusative morphology and syntax, the other Avar with Ergative morphology and syntax. A language like Warrongo falls in between, having Accusative AND Ergative morphology with Ergative syntax, or a language like Warlpiri with Ergative morphology and Accusative syntax. Each of these is classifiable by their case marking or what effect is achieved with equi NP deletion.

Active languages, on the other hand, be they close to Latin or Avar, are singularly classifiable by the split intransitive. Even Warrongo has the same morpheme cover both the NOM and ABS cases (0 marked, distinct from both Erg and Acc, which are distinct from each other) such that intransitives are ambiguous about their Ergativity, whereas an Active language would mark the S as A or P accordingly.

I would sum it up by saying that Actives are treated differently because they have a feature which neither implies nor prevents the presence of ergativity, even if it can be said that the underlying cognitive motivation is similar.

As a quick aside, this question just flitted through my forebrain: could you wind up with a 4-way case system if you combined Tripartition with Active-Stative? Re-reading that question, it strikes me as both interesting and silly... It'd probably never happen on Earth

[Hallow XIII]

It depends. You could, probably, but it would be a system that would be extremely Split-S and also probably highly unstable. Also it would take a lot of very unorthodox grammar.

...

Yep, I'm making one now.

[Carson]

It depends. You could, probably, but it would be a system that would be extremely Split-S and also probably highly unstable. Also it would take a lot of very unorthodox grammar.

...

Yep, I'm making one now.

Lol, have fun!

I agree with you, though. Unorthodox grammar, indeed. I envision, as an idealized system, a grammar that has syntactic pivots on numerous places. For instance, declarative sentences may license Accusative equi NP deletion, while sentences with [+irrealis] may license Ergative deletion. This would, of course, be working alongside the morphological pivot of animacy, so you might end up with a sentence in the Subjunctive mood about you and your listener having Accusative morphology but Ergative syntax!

My head wants to explode just thinking about it, and I don't believe that any such language could exist in reality. It is idealized only for the purposes of making predictions, sort of like removing friction from kinematic equations so that undergrads can grasp concepts like elasticity or acceleration.

[zompist]

so you might end up with a sentence in the Subjunctive mood about you and your listener having Accusative morphology but Ergative syntax!

This happens in many Australian languages. E.g. subject assignment is ergative, and pronouns are accusative, so you have sentences like

He hit her, and then left.

where the person who left is understood to be "her".

[Carson]

so you might end up with a sentence in the Subjunctive mood about you and your listener having Accusative morphology but Ergative syntax!

This happens in many Australian languages. E.g. subject assignment is ergative, and pronouns are accusative, so you have sentences like

He hit her, and then left.

where the person who left is understood to be "her".

Ergative marking with accusative syntax is quite common, but I hear the reverse (as in my example) is unattested so far. But I might be mixing this up with verb agreement...

[2+3 clusivity]

Cool thread. All of my languages have used a split-ergative system, so I guess I will chime in.

First for a quip.

This system characterizes a number of Australian languages, as well as a few Caucasian and Dravidian languages (and some others, no doubt).

No, not in Dravidian languages. Split ergativity between perfective and all other aspects (i.e. imperfect, habitual) does, however, characterize many of the neighboring Indo-Aryan languages. Dravidian languages are Nom-Acc.

In Tamil (iirc), there is a pivot on the past tense.

Example?

---

Regarding why tripartite marking occurs in the "middle" of the animacy hierarchy is pretty easy if you graph it out. I am not sure why you are confused, because you laid out the reason.

Silverstein observed Accusativity rolling down the Animacy hierachy (i.e. 1st person pronouns are the most likely to be Accusative cross linguistically), while Ergativity climbs up the hierarchy (thus languages with any degree of Ergativity are most likely to mark full NPs that do not denote living people as Ergative).

if you take your hierarchy:

1st/2nd/3d & demonstratives/proper nouns & names/human nouns/animate nouns/inanimate nouns/abstract nouns
Decreasing likelihood of Nominative marking--------------------------------------------------------------------------------------->
<------------------------------------------------------------------------------------------Decreasing likelihood of Ergative marking

At some middle point--perhaps one that shifts rapidly--you might experience both [+nom/acc] and [+erg/abs] marking where the two systems are duking it out as part of language change.

A more interesting question is, why don't both types of marking simply engulf the whole spectrum. Since Accusative and Ergative (v. nom or abs) marking tend to block verb concord, I would suspect that this middle area--or even the whole spectrum--becomes/would become difficult to manage from an information structure standpoint. Sure, word order or case affixes might sort things out quickly, but the reduction in redundancy otherwise expected in the language is probably problematic.

when the two alignments overlap each other so that both are in play over some middle range of the hierarchy, then that range is tripartite. Second, it's also possible for the two not to even meet, leaving a middle range that is not aligned in either manner - this is sometimes called the Monster Raving Loony system because no known language does it universally, it only shows up in the middle range of some animacy-based split systems.

"Monster-raving looney system" or Transative-intransitive alignment is another "transitional system." You can see it occasionally in some of the Eastern-Iranian languages/dialects like Rushani. http://en.wikipedia.org/wiki/Rushani_language. Some of the lit--including Wiki--describes it as "double oblique," because languages in that body of literature often attach their ergative or accusative marking to an oblique stem. The information structure problem discussed above seems similar there in at least transitive structures.

---

Ergative marking with accusative syntax is quite common, but I hear the reverse (as in my example) is unattested so far. But I might be mixing this up with verb agreement...

I think verb concord v. inter-clausal or syntactic ergativity needs to be distinguished. Zomp gave an example of the latter and Dixon has a chapter on it in "Ergativity." At least regarding nom-acc noun phrase marking v. erg-abs verb agreement, there have been a couple threads that have mentioned this--at least obliquely. viewtopic.php?f=7&t=41693. viewtopic.php?f=4&t=40917.

WALS seems to be broken right now, but if you combines features 98A or 99A with 100A you should find that ergative person marking and nom-acc pronoun marking do not co-occur. http://wals.info/feature. http://wals.info/feature/combined/98A/100A.

This actually makes sense, I think its Comrie--maybe Dixon--who offers a variant on the animacy hierarchy:

Verb agreement/1st/2nd/3d & demonstratives/proper nouns & names/human nouns/animate nouns/inanimate nouns/abstract nouns
Decreasing likelihood of Nominative marking--------------------------------------------------------------------------------------------------------------->
<-----------------------------------------------------------------------------------------------------------Decreasing likelihood of Ergative marking

About a year ago I found this interesting paper: http://web.mit.edu/pritty/www/patel2007.pdf, where the fieldworker suggests they found a language showing in some cases: "Nominative/accusative case with ergative/absolutive agreement." Which is the reverse of what you would expect from the hierarchy immediately above, and what the field worker suggests they expected.

[zompist]

so you might end up with a sentence in the Subjunctive mood about you and your listener having Accusative morphology but Ergative syntax!

This happens in many Australian languages. E.g. subject assignment is ergative, and pronouns are accusative, so you have sentences like

He hit her, and then left.

where the person who left is understood to be "her".

Ergative marking with accusative syntax is quite common, but I hear the reverse (as in my example) is unattested so far. But I might be mixing this up with verb agreement...

Er, the example I just gave is precisely ergative syntax and accusative morphology. What you're saying is unattested is in fact attested. But perhaps you meant something else...

[2+3 clusivity]

Er, the example I just gave is precisely ergative syntax and accusative morphology. What you're saying is unattested is in fact attested. But perhaps you meant something else...

I think we are speaking about different things.

I do not disagree that accusative marking and what Dixon would call "inter-clausal or syntactic ergativity" can co-occur. What I was referring to was ergative verb agreement within the same clause (to my knowledge) does not co-occur with accusative morphology.

I.e.

A[nom] O[acc] Verb[agrees with A, i.e. accusative verb agreement]
A[erg] O[abs] Verb[agrees with O, i.e. ergative verb agreement]
A[erg] O[abs] Verb[agrees with A, i.e. accusative verb agreement]

but

*A[nom] O[acc] Verb[agrees with O, i.e. ergative verb agreement]

[Hallow XIII]

Er, the example I just gave is precisely ergative syntax and accusative morphology. What you're saying is unattested is in fact attested. But perhaps you meant something else...

I think we are speaking about different things.

Indeed you are, as he was mostly addressing Carson.

*A[nom] O[acc] Verb[agrees with O, i.e. ergative verb agreement]

Yeah. That seems... Pretty totally unlikely. Good concept for an alien language, maybe.

[Vardelm]

*A[nom] O[acc] Verb[agrees with O, i.e. ergative verb agreement]

Yeah. That seems... Pretty totally unlikely. Good concept for an alien language, maybe.

I would assume that it is non-existent, or at least extremely rare. However, it seems like it makes sense in a way. In nom/acc alignment, you mark the object noun, and this would be similar to marking the object, if that makes sense. The opposite does occur though, doesn't it, in languages w/ ergative marking but accusative syntax? So:

A[erg] O[abs] Verb[agrees with A, i.e. accusative verb agreement]

Even if accusative marking w/ ergative verb agreement is unattested, it doesn't seem like it's a huge stretch to be believable.


EDIT: Also....

Good concept for an alien language, maybe.

Countdown to DePaw using this for his Narin conlang in 3.... 2.... 1......

[Hallow XIII]

*A[nom] O[acc] Verb[agrees with O, i.e. ergative verb agreement]

Yeah. That seems... Pretty totally unlikely. Good concept for an alien language, maybe.

I would assume that it is non-existent, or at least extremely rare. However, it seems like it makes sense in a way. In nom/acc alignment, you mark the object noun, and this would be similar to marking the object, if that makes sense. The opposite does occur though, doesn't it, in languages w/ ergative marking but accusative syntax? So:

A[erg] O[abs] Verb[agrees with A, i.e. accusative verb agreement]

Even if accusative marking w/ ergative verb agreement is unattested, it doesn't seem like it's a huge stretch to be believable.

Well no. Like many other features that don't occur or haven't been observed in human languages, such marking is not found in large languages because it does not rank high on our arbitrary simplicity scale, not because it's in any way impracticable. In fact, you could probably find multiple ways in which such marking might develop pretty easily.

EDIT: Also....

Good concept for an alien language, maybe.

Countdown to DePaw using this for his Narin conlang in 3.... 2.... 1......

[2+3 clusivity]

The opposite does occur though, doesn't it, in languages w/ ergative marking but accusative syntax? So:

A[erg] O[abs] Verb[agrees with A, i.e. accusative verb agreement]

Even if accusative marking w/ ergative verb agreement is unattested, it doesn't seem like it's a huge stretch to be believable.

If you wanted examples.

A[nom] O[acc] Verb[agrees with A, i.e. accusative verb agreement], e.g. English, most acc languages.
A[erg] O[abs] Verb[agrees with O, i.e. ergative verb agreement], e.g. Hindi in perfective, most erg languages.
A[erg] O[abs] Verb[agrees with A, i.e. accusative verb agreement], e.g. Nepali in perfective (according to: http://web.mit.edu/pritty/www/patel2007.pdf, see ex. p. 11)
*A[nom] O[acc] Verb[agrees with O, i.e. ergative verb agreement], e.g. apparently a dialect of Gujarati and maybe some dialects of Rajasthani in perf. (same, see ex. pp. 13, 70-71)

It is my hope that this information will contribute to understanding of split systems and aid anyone who is considering making a tripartite or split system for their conlang, as it helped me.

Going back to the OP, I think the real takeaway from all of these systems (tripartite, split-erg, nominal v. verb agreement) is that at a certain point in language change odd things can pop up. Perhaps using something funky discussed here in a proto-language would be a useful tool for developing diverse daughter languages as the transitional system collapses.

[Carson]

Er, the example I just gave is precisely ergative syntax and accusative morphology. What you're saying is unattested is in fact attested. But perhaps you meant something else...

I think we are speaking about different things.

I do not disagree that accusative marking and what Dixon would call "inter-clausal or syntactic ergativity" can co-occur. What I was referring to was ergative verb agreement within the same clause (to my knowledge) does not co-occur with accusative morphology.

I.e.

A[nom] O[acc] Verb[agrees with A, i.e. accusative verb agreement]
A[erg] O[abs] Verb[agrees with O, i.e. ergative verb agreement]
A[erg] O[abs] Verb[agrees with A, i.e. accusative verb agreement]

but

*A[nom] O[acc] Verb[agrees with O, i.e. ergative verb agreement]

This is what I was thinking of in my last post. My notes on the topic are not very clearly arranged, it seems, and I think I misread Zompist's post as well.

Well no. Like many other features that don't occur or haven't been observed in human languages, such marking is not found in large languages because it does not rank high on our arbitrary simplicity scale, not because it's in any way impracticable. In fact, you could probably find multiple ways in which such marking might develop pretty easily.

I thought simplicity was gauged by processing time and average accuracy, the way they have rated the difficulty of various relativization strategies by having adults and children reproduce them. Maybe it was just that one experiment?

But as for why it may be that there are no languages with overt ergativity and accusative subject agreement, my guess right now is that ergativity forms a spectrum across languages. Ergative syntax is very close to the extreme end, where you find a language like Avar with ergativity in both morphology and syntax. At the middle range, you find fewer and fewer languages with ergative syntax (and more languages with Ergative-Accusative MSA), and at the Accusative end... If you think of it like a hierarchy, my suspicion is that ergative syntax is very low in the hierarchy and needs to be licensed by a bunch of other things.

Damn, I feel like I could turn this discussion into an honors thesis next year XD

[Drydic]

Damn, I feel like I could turn this discussion into an honors thesis next year XD

Do it.

[Radius Solis]

But as for why it may be that there are no languages with overt ergativity and accusative subject agreement, my guess right now is that ergativity forms a spectrum across languages. Ergative syntax is very close to the extreme end, where you find a language like Avar with ergativity in both morphology and syntax. At the middle range, you find fewer and fewer languages with ergative syntax (and more languages with Ergative-Accusative MSA), and at the Accusative end... If you think of it like a hierarchy, my suspicion is that ergative syntax is very low in the hierarchy and needs to be licensed by a bunch of other things.

Maybe, but I'm not sure it completely works that way. If you want to try typologizing languages along such a spectrum, you'll have to weigh what it gains you against there being a lot of mess and noise in the signal. At present, the best well-accepted way to look at it is that languages have various behaviors that can be aligned accusatively or ergatively, and that most of the time these need not co-occur, save that some of them are rarely ergatively aligned and may depend on the alignment of other behaviors in the language. So the only spectrum that's widely accepted is that behaviors can be ranked according to how frequently they are aligned one way or the other. For example:

subject assignment - rarely ergative
verb agreement - occasionally ergative
case marking - often ergative but still mostly accusative
noun incorporation - always ergative

Obviously a language need not have systematic noun incorporation before it can have ergative case marking, for example. So it's not as simple as the lower things on the list licensing higher ones. Another interesting point: all alignable behaviors are attested to be ergative in at least some language, but noun incorporation is not attested to be accusative in any language.

[zompist]

subject assignment - rarely ergative
verb agreement - occasionally ergative
case marking - often ergative but still mostly accusative
noun incorporation - always ergative

Neat, but a little confusing, as I believe your descriptors are universal, rather than relative to the next rung down. That is, ergative syntax is "rare" worldwide. I don't think it's rare in morphologically ergative languages.

Also, I wouldn't be so sure of that "always" on noun incorporation, in light of such English terms as employee-run, moth-eaten.

[Vuvuzela]

Nez Perce is a tripartite language where the "pivot" seems to be a single "point", direct vs. inverse. this paper, the ergative and accusative cases were each originally limited to specific circumstances, but in Nez Perce (and to some extent in Sahaptin) they've both broadened over time.
The accusative comes from an earlier "directional" or "allative", which was first extended to indirect objects (compare English to), and then to human direct objects (compare Spanish a), and in Nez Perce (also optionally in Sahaptin) to nonhuman direct objects. In sentences unmarked for voice, the primary object takes the accusative case, while secondary objects and subjects are unmarked.
The ergative comes from an earlier "cislocative" meaning "hither".When it first became a case marker, was only used to mark third-person subjects of sentences with first- or second-person objects. However, in Nez Perce, it is now used for all third person subjects in inverse constructions, in which the primary object is also marked as accusative.
Note that, while there are Nominative-Accusative constructions in the language, there are no purely Ergative-Absolutive constructions.

[Radius Solis]

Also, I wouldn't be so sure of that "always" on noun incorporation, in light of such English terms as employee-run, moth-eaten.

That was silly of me - what I should have said was "never accusative". Obviously subject incorporation is possible. There just aren't any languages where subjects but not objects can be incorporated.

[2+3 clusivity]

I found this, If you need reading: http://web.mit.edu/pritty/www/ergativity/index.htm. It is a pile of links to papers on ergativity.

[Radius Solis]

subject assignment - rarely ergative
verb agreement - occasionally ergative
case marking - often ergative but still mostly accusative
noun incorporation - always ergative

Neat, but a little confusing, as I believe your descriptors are universal, rather than relative to the next rung down. That is, ergative syntax is "rare" worldwide. I don't think it's rare in morphologically ergative languages.

Wikipedia claims that it is, though it doesn't provide an in-line cite. "... few morphologically ergative languages have ergative syntax" is all it says on the matter. In any case, the dependence of "syntactic ergativity" (i.e. ergative syntactic pivot) on the presence of morphological ergativity is the only such dependence I know of among the whole list of potentially ergative behaviors, so describing them relative to the next rung on the ladder might be even more confusing. There's quite a lot of things that can be aligned ergatively or accusatively. Besides the aforementioned, there's also basic word order; which arguments can be relativized; switch reference marking; which arguments can be subclauses; the entire list all over again for behaviors within subclauses (subclauses sometimes preserve older alignments just as they preserve older word orders); and nominalizing operations (like how English -er is accusative in that it names a subject or agent, but -ee is ergative in that it names a subject or patient).

[zompist]

The accusative comes from an earlier "directional" or "allative", which was first extended to indirect objects (compare English to), and then to human direct objects (compare Spanish a), and in Nez Perce (also optionally in Sahaptin) to nonhuman direct objects. In sentences unmarked for voice, the primary object takes the accusative case, while secondary objects and subjects are unmarked.
The ergative comes from an earlier "cislocative" meaning "hither".When it first became a case marker, was only used to mark third-person subjects of sentences with first- or second-person objects. However, in Nez Perce, it is now used for all third person subjects in inverse constructions, in which the primary object is also marked as accusative.

I've highlighted some text above, because it fits in with a principle that helps explain why the agentivity hierarchy works as it does.

The basic idea is, there's something unexpected about
* high-animacy things being patients
* low-animacy things being agents

So e.g. pronouns and proper nouns are likely to refer to high-animacy things, i.e. people, and we want to flag the rarer situation of them being acted on by making them accusative. Whereas (say) inanimate nouns are low-animacy, so we flag the rarer situation of them being the actors by making them ergative.

(Of course this doesn't tell you where the split in the hierarchy comes-- that varies by language. It's also mysterious why sometimes ergativity is limited to certain tenses, but this seems to be quite rare.)

[hwhatting]

It's also mysterious why sometimes ergativity is limited to certain tenses, but this seems to be quite rare.)

At least in the cases of Iranian ergativity, the reason is simply historical - a construction of the past passive participle plus copula first became a resultative perfect and then the simple past tense. In the Iranian languages, like in most IE languages, the past passive participle of intransitive verbs has an active meaning (cf. German er ist geschlagen "he has been beaten", but er ist gekommen "he has come" (not *"been come (by somebody)"). That's why this tense has an ergative structure, with the subject of intransitive verbs being interpreted as agent and the subject of transitive verbs being interpreted as patient. What happened then is that the resultative past tense ("perfect") became the regular past tense, which is a frequently observed development. IIRC, the history of the ergative past tenses in Modern Indo-Aryan is similar. I'd wager that, if we knew more of the history of ergative past tenses in other language families, we'd find similar developments out of resultative constructions.