Ergativity for Novices

[bradrn]

I’ve had some more time to work on this lately, so I’ve gone back and made some minor updates to some of the previous posts:

• In the post on morphological ergativity, I’ve rewritten it slightly to clarify why I’m including word order (a seemingly syntactic phenomenon) in that particular section. (I remember getting some questions about that point earlier, so I thought it would be a good idea to clarify this.)
• In the same post, I added references to Canela, Halkomelem and Päri — three languages which seemingly defy ‘universals’ around agreement and case-marking. I also rewrote that particular section to make reference to those.
• In the post on animacy-based splits, I added a paragraph on the ways in which case concord interacts with such splits. (akam chinjir mentioned that point in a reply a while ago, but I only managed to incorporate that into the main post now.)

Finally, as for the next post (on syntactic ergativity): I’ve had a lot of work recently, so I haven’t started writing it yet, but I’ve been doing a lot of research on the topic and hope to start writing it soon now that I’ve got a bit more time. I’m hoping to finish that post within a few weeks, although it’s probably best to think of that as a rough timeframe rather than a deadline.

[jal]

Cool, looking forward to it!


JAL

[bradrn]

Syntactic Ergativity

Introduction

At this point, we have already covered several forms of ergativity: ergativity in case-marking, ergativity in verbal agreement, ergativity in word order, and the manifold forms of split ergativity. But there is one thing left which we haven’t yet covered: ergative phenomena in syntax. Now, it is important to be aware that, for the most part, the morphosyntactic alignment of a language doesn’t really change much about how its syntax works. However, there do seem to be some areas of syntax which genuinely do work differently between ergative languages and accusative languages. For instance, one often-quoted example is coordination between transitive and intransitive verbs in sentences like ‘mother saw father and returned’ — who returned? In English, it’s the mother: the S argument of returned is treated the same way as the A argument of saw. But in some ergative languages such as Dyirbal, it’s the father:

ŋuma

father.ABS

yabu-ŋgu

mother-ERG

bura-n

see-NONFUT

banaga-nʸu

return-NONFUT

Mother saw father and [he] returned

Languages which display such phenomena are said to show syntactic ergativity.

What is syntactic ergativity?

It might already seem clear from the above discussion what a syntactic ergative language is. Unfortunately, it’s not that simple.

As a first try, we might say that a language is syntactically ergative if its syntax treats S and O the same but A differently. However, it turns out that there are some important syntactic processes which — even in the most thoroughly ergative languages — always treat S and A the same but O differently. There are several prominent syntactic processes which work like this:

• Imperatives always treat S and A in the same way: both S and A are often restricted to second person, and both S and A may be omitted (especially if they are second person).
• In languages with reflexive pronouns, it is always the S or A argument of the verb which is the antecedent, while the O argument takes the reflexive pronouns.
• In languages where concepts such as ‘can’, ‘might’, ‘try’, ‘want’, ‘need’ etc. (what Dixon calls ‘secondary concepts’) are expressed using lexical verbs with a subordinate clause, the S or A argument of the main verb must be coreferential with the S or A argument of the subordinate clause, but O may not be coreferential.
• In control predicates, it seems that O may never be controlled, but S and A may. (This is controversial though; see below for more details.)

(Similarly, there are several processes which always treat S and O the same. For instance: noun incorporation always allows incorporation of S and O arguments, but never A; if a verb has multiple senses, then the choice of sense may relate to S or O but not A (e.g. The horse/tap is running, or I cut the paper/my finger). For a fuller list consult Dixon’s Basic Linguistic Theory, Section 13 Appendix 1.)

Now, partly due to the fact that these important syntactic processes always treat S and A in the same way, some authors have treated this as disproving the existence of syntactic ergativity altogether. However, as we have already seen with coordination, this cannot be correct: there do exist certain syntactic phenomena which operate ergatively in some morphologically ergative languages. Aside from coordination, such phenomena include Ā-movement, and definiteness and specificity. But even here, there is disagreement: not every ergative language has ergative syntax in all these areas, and so there is still some debate about exactly which of these phenomena are to be treated as being part of ‘syntactic ergativity’. For instance, Dixon places emphasis on coordination; Polinsky only includes Ā-movement; Manning includes both of those, plus definiteness and specificity; and Deal includes all of the above, plus control. As a result of this, it is remarkably difficult to figure out exactly what is meant by ‘syntactic ergativity’. Here I will adopt an inclusive view, attempting to cover all of these phenomena; a language will then be called ‘syntactically ergative’ if it exhibits ergativity in any one of those areas. (And conversely, ‘syntactically accusative’ only if it exhibits ergativity in none of those areas; possibly this isn’t the best of definitions, but it’s certainly convenient!) Note also that it is possible for a language to be neither syntactically ergative nor syntactically accusative; for instance, we will see that English is syntactically neutral with respect to Ā-movement.

An additional aspect of syntactic ergativity is the antipassive. This is a valency-changing operation which in many ways acts in an opposite manner to the passive voice: while the passive promotes the A argument to S, the antipassive instead promotes the O argument to S. The antipassive is not a syntactically ergative operation in and of itself — indeed, many accusative languages include an antipassive — but it is an important part of the morphosyntax of syntactically ergative languages, and so will be discussed here.

Ā-movement

One very important process exhibiting syntactic ergativity is that of ‘Ā-movement’ or ‘extraction’ (alternately spelled ‘A'-movement’; either way, it’s pronounced A-bar movement). Roughly speaking, Ā-movement is a type of movement which occurs for semantic or pragmatic reasons. (This is as opposed to A-movement, which is motivated on purely syntactic grounds. For more details see akam chinjir’s excellent overview of this subject.) Ā-movement does not refer to one single process, but is a general category including many different syntactic phenomena such relativisation, wh-question formation, focalisation and topicalisation.

Now, as examples of Ā-movement, all of those processes have something obvious in common: an NP moves from one position to another. We might ask — which types of NPs can be moved? In English, it turns out that all of them can. Using relativisation and wh-question formation as examples:

NP	Starting sentence	Relativisation	Wh-question formation
Subject	I see you	the person who sees you	Who sees you?
Direct object	I see you	the person who you see	Who do you see?
Indirect object	I give this to you	the person who I give this to	Who do I give this to?
Oblique object	The cat is on the mat	the thing that the cat is on	What is the cat on?
Possessor	I see your car	the person whose car this is	Whose car is this?
Standard of comparison	I am taller than you	the person who I am taller than	Who am I taller than?

However, this is not the case in all languages. Keenan and Comrie found that languages tend to follow an accessibility hierarchy in terms of which NPs may be relativised (and by extension undergo other types of Ā-movement):

Subject > Direct Object > Indirect Object > Oblique Object > Possessor > Standard of Comparison

In this hierarchy, if a language allows extraction of one type of NP, then all the NPs to the left of it in the hierarchy may also undergo extraction. For instance, French and German allow extraction of everything except standards of comparison, Tamil allows only subjects, direct objects and indirect objects to be extracted, while Malagasy only allows subjects to be extracted (source).

Now, you may have noticed a problem with this hierarchy: it relies on notions of ‘subject’ and ‘object’. This works well enough for English, French and Malagasy — in which ‘subject’ and ‘direct object’ simply correspond to the nominative and accusative arguments — but breaks down with ergative languages, which don’t really have the quite same notion of a ‘subject’ (if indeed they have one at all). And this is where syntactic ergativity comes into play! Syntactically ergative languages are those in which the accessibility hierarchy is modified:

Absolutive > Ergative > Indirect Object > Oblique Object > Possessor > Standard of Comparison

So in syntactically ergative languages, it is now absolutive rather than nominative arguments which are most easily moved. This can be seen in languages such as Chukchi (Polinsky 2016), in which absolutives can undergo processes such as relativisation, while ergatives cannot:

ʔətt-e

dog-ERG

melotalɣ-ən

hare-ABS

piri-nin

catch-AOR.3s.3s

The dog caught a/the hare

[ʔətt-e

dog-ERG

piri-lʔ-ən]

catch-PTCP-ABS

melotalɣ-ən

hare-ABS

the hare that the dog caught

*[melotalɣ-ən

hare-ABS

piri-lʔ-ən]

catch-PTCP-ABS

ʔətt-ən

dog-Abs

Intended: the dog that caught the hare

In order to relativise on the ergative, the antipassive voice must be applied to promote the ergative argument to an accusative:

ʔətt-ən

dog-ABS

ine-piri-ɣʔi

ANTIP-catch-AOR.3s

(melotalɣ-tə)

hare-DAT

The dog caught (a/the hare)

[melotalɣ-tə

hare-DAT

ine-piri-lʔ-ən]

ANTIP-catch-PTCP-ABS

ʔətt-ən

dog-Abs

the dog that caught the hare

I will discuss the antipassive a bit more later.

Now, looking again at the accessibility hierarchy, consider the case of a language slightly more permissive than Chukchi — a language in which only ergatives and absolutives may undergo extraction. Basque is an example of such a language. It is important to realise that such a language can be said to be neither syntactically ergative nor syntactically accusative with respect to Ā-movement: the case where only ergatives and absolutives may undergo Ā-movement is indistinguishable from the case where only nominatives and accusatives may undergo Ā-movement. (English is such a language as well.) For this reason, when talking about Ā-movement, we restrict the category of ‘syntactically ergative’ languages only to languages such as Chukchi, where the absolutive argument is demonstrably treated differently to the ergative argument.

Another subtlety to be aware of is that not all constructions under the umbrella of ‘Ā-movement’ may act the same way. Above I mentioned relativisation, wh-question formation, focalisation and topicalisation as examples of Ā-movement. But in fact some languages express these constructions without movement — for instance, the English question ‘You did what‽’ involves no Ā-movement. Naturally, such constructions are not restricted by the animacy hierarchy (or at least do not have the same restrictions). Other similarly problematic cases to be aware of include (pseudo-)clefts and headless relative clauses (which in some languages also may not involve movement), or non-restrictive relative clauses (which Polinsky notes ‘may be subject to different principles’, with not all languages possessing such constructions). And even when all of these constructions can be shown to operate via Ā-movement, some may be syntactically ergative while others may not: for instance, Chukchi disallows relativisation of ergatives (as seen above), but allows wh-questioning of both the ergative and absolutive arguments — even though wh-questioning in Chukchi also utilises Ā-movement. Interestingly, it does seem that all syntactically ergative languages show ergativity at least in relativisation, even if they don’t have it in other areas with Ā-movement; we will discuss this further below.

Conjunction reduction

Another construction in which some languages demonstrate syntactic ergativity is coreferential deletion in conjunctions, in which a repeated NP in a conjunction gets deleted. This has already been briefly discussed using the sentence ‘mother saw father and returned’ — in English, the deleted NP is mother, but in Dyirbal it’s the father. For this reason, we say that English has an S/A pivot, meaning that corefential NPs may be deleted only when both NPs are in S or A function, but not when either is O. Enumerating all the possibilities, this can easily be seen to be correct, with conjunction reduction proving to be impossible with the intended meaning only in the case when one of the coreferential NPs is in O function:

1. (S, S) mother came and [mother] returned
2. (S, A) mother came and [mother] saw father
3. (S, O) *mother came and father saw [mother]
4. (A, S) mother saw father and [mother] returned
5. (A, A) mother saw father and [mother] heard John
6. (A, O) *mother saw father and John heard [mother]
7. (O, S) *father saw mother and [mother] returned
8. (O, A) *father saw mother and [mother] heard John
9. (O, O) *father saw mother and John heard [mother]

Similarly, syntactically ergative languages may operate according to an S/O pivot — the coreferential NPs must be in S or O function, but not A. In such a language, sentences 3,7,9 above would be grammatical, while sentences 2,4,5 would now be ungrammatical.

Now, there are certainly differences between languages as to the exact operation of these pivots. For instance, Dixon notes that the pivot in Dyirbal is much ‘stronger’ than that in English: in Dyirbal, ‘two clauses cannot be coordinated to form a complex sentence construction … unless they have a common NP and the pivot condition is satisfied’ (emphasis not mine). On the other hand, English freely allows any two NPs to be coreferential if the appropriate pronoun is used: a sentence such as ‘mother saw father and John heard her’ (with A, O coreferential) is fine in English, but Dixon implies that such a sentence is forbidden in Dyirbal even when using a pronoun.

There are also more complex cases of pivots. For instance, in the Australian language Yidinʸ, the pivot depends on the alignment of the NPs involved in the conjunction. Recall from earlier that Yidinʸ uses accusative marking for first and second person pronouns, ergative marking for common nouns, and tripartite marking for nouns between those points on the animacy hierarchy. The rules for coreferential deletion reflect this — when the common NP is pronominal, it may be deleted only when both NPs are S or A, but when the common NP is non-pronominal, it may be deleted only when both NPs are S or O. An interesting case arises with sentences such as ‘I followed the man and fell down’ — the subject has an S/A pivot, but the object has an S/O pivot, so the sentence is ambiguous. An even worse case arises with sentences such as ‘The woman saw me and was frightened’ — in this case, neither argument has the appropriate pivot to be the S argument of the second clause! In Yidinʸ, an S/A pivot is used to resolve this situation.

And there exist even more complex cases. For instance, the Cariban language Panare has four verbal suffixes indicating subordination, each of which uses a slightly different pivot:

Suffix	Meaning	Pivot
-séjpe	‘and then, in order to’	S₁/A₁ = S₂/A₂
-séʼñape	‘as a result’	S₁/O₁ = O₂
-ñépe	‘and then/in order to’	S₁/O₁ = S₂/A₂
-pómën	‘after/because’	S₁/A₁ = S₂/A₂

(Interestingly, Dyirbal has a similar construction: applying -ŋurra to the second verb in a conjunction indicates firstly that the action of the second verb occurs immediately after that of the first verb, and secondly that A of the first verb is coreferential with S/O of the second verb. It is noted that, unlike the antipassivisation construction, covered below, this construction may be used ‘without forward planning’, though it appears to be used relatively rarely.)

Of course, a pivot strategy isn’t the only way to approach conjunction reduction. Some languages (e.g. Limbu, Samoan, possibly Fijian) simply leave such cases ambiguous, with no apparent restrictions on coreferential deletion. Other languages (e.g. Kannada) have restrictions which do not seem to use the same notion of ‘pivot’ — for instance, in Kannada, the deleted NP in the second clause must be coreferential with the first constituent of the first clause (it helps that word order is rather free). Another common method is through ‘switch-reference marking’ (e.g. in Diyari), where verbs may be inflected to show which of their arguments are coreferential with the next clause. Naturally, none of these strategies qualify as being either syntactically ergative or syntactically accusative.

Other areas of syntactic ergativity

In addition to the two syntactically ergative phenomena covered so far — Ā-movement and conjunction reduction — there do exist other syntactic phenomena which have been claimed to operate ergatively. These include the following:

• In some languages, the absolutive argument must be definite (or be generic, or have wide scope; the exact characterisation has been debated — see (Manning 1994), for more details). In that case a compensatory strategy may be used if the absolutive is not definite: in the (seemingly) most common case, some languages have a form of split ergativity in which A receives absolutive and O receives an oblique case when O is not definite. (Woolford (2015) provides a most interesting explanation of this: Woolford proposes that, in ergative languages, the object may move out of the verb phrase, and A may only receive ergative marking when this movement occurs. Additionally, in some languages, the object moves out of the verb phrase only when it is definite or specific. For instance, in Niuean, the word order with a nonspecific object is [V O] S, with O being unmarked and S being marked with absolutive; when O is specific, it moves out of the VP to get [V] S O, and now that it has moved, S may receive ergative marking and O gets absolutive marking. Similarly, in Inuit, the underlying structure is proposed to be S [O V], with S receiving absolutive and O receiving an oblique case; O moves out of the VP when it is specific, to get a structure of S O [V], with S receiving ergative and O receiving absolutive case.)
• Several languages have been claimed to exhibit ergativity in control, but most of these claims are doubtful for one reason or another. ‘Control’ refers to constructions such as ‘we would like to stay’ and ‘I asked him to say that’, in which an argument of the main verb determines an argument of the subordinate clause. English allows the controlled argument to be S or A, but not O: a sentence such as ‘*I asked him the doctor to see’ is ungrammatical. (Interestingly, there are no such restrictions on the controller, which may be any of S, A, O.) A language with syntactic ergativity in control would be the opposite: S and O could be controlled, but A could not. The most prominent claimed case of such a language is Dyirbal: a ‘purposive’ inflection on the subordinate verb is used for control, with this construction operating ergatively. However Manning notes that this construction has many other uses besides control, and seems more likely to be just another instance of conjunction reduction (which as we already saw operates ergatively). Other claimed cases include Kalkatungu and Sama Southern; more research is needed to determine if these really are instances of ergative control. Whatever the case, ergativity in control is certainly very rare, with only three or four potential examples being identified. For more information see (Manning 1994), (Kazenin 1994), (Deal 2015).

Distribution

As it turns out, the various areas of syntactic ergativity are not distributed randomly, but relate to each other in various ways. Most prominently, there is a very strong generalisation that a language can only show syntactic ergativity if it shows some form of morphological ergativity. ‘Morphological ergativity’ here includes split ergative systems, though curiously enough there seem to be no active-stative (i.e. split-S or fluid-S) languages with syntactic ergativity. (This is one of the reasons why many people treat active-stative languages as being non-ergative). In fact, split ergativity usually has no effect whatsoever on syntax, with a construction remaining equally syntactically ergative (or accusative) no matter whether it is used with ergative or accusative case-marking. (Yidinʸ, mentioned above, is a rare exception to this rule.)

Another strong generalisation is that if a language shows syntactic ergativity, it will always show it at least in Ā-movement. A corollary of this is that, if a language has syntactic ergativity in some other area, it will also have syntactically ergative Ā-movement. Within the category of Ā-movement, all syntactically ergative languages seem to display ergativity at least with relativisation, even if they do not display it with other types of Ā-movement.

Thirdly, it seems that all syntactically ergative languages have some compensatory strategy to allow ergatives to be extracted. Most prominently, an antipassive may be used to promote an ergative to an absolutive, but there are other methods as well. For instance, some languages have a special Agent Focus construction allowing the ergative to be extracted; some languages use a resumptive pronoun when the ergative is extracted; some languages use ‘anti-agremeent’, in which verbal agreement with the ergative is supressed (or sometimes altered) when the subject is extracted; and some languages use a nominalisation to convert the agent from a verbal argument to a nominal complement, which may then be extracted. But no matter which method is used, there is always some way of extracting ergatives.

Finally, it is remarkable that the majority of morphologically ergative languages show some sort of syntactic ergativity (again discounting active-stative languages like Tibetan and Basque, which never show syntactic ergativity). For instance, using the WALS language sample, Polinsky finds only seven ergative languages which do not display syntactic ergativity. This sample indicates that having morphological ergativity but not syntactic accusativity appears to be an areal feature: of these seven languages, two are Nakh-Dagestanian, three are Australian, and the last is the language isolate Burushaski, surrounded by the syntactically accusative languages of India and Tibet. But the WALS sample is admittedly small, and misses some other languages (e.g. the Peruvian language Shipibo) which are known to be morphologically ergative but syntactically accusative; Polinsky notes that ‘we do not have sufficient information … so any judgment calls about the relative frequency of certain patterns over others are necessarily preliminary’. Still, it is remarkable that — at least when it comes to relativisation — it is far easier to find morphologically ergative languages which are syntactically ergative than it is to find morphologically accusative languages which are syntactically accusative.

Passive and antipassive

No discussion of syntactic ergativity would be complete without a discussion of the antipassive voice. We are all familiar with the passive voice: a valency-changing operation which promotes O to S, and demotes A to an (optional) oblique argument. (There do exist claimed ‘passives’ which do not change valency; here we will consider only valency-changing operations.) The antipassive voice is exactly the reverse: a valency-changing operation which promotes A to S, and demotes O to an (optional) oblique argument.

Now, from a nominative-accusative perspective, this feels deeply weird. A and S have the same case and role in the sentence — how is it possible to ‘promote’ one to the other‽ And the accusative is usually the more marked case, so O is already pretty demoted with respect to S and A — why would you want to demote it further? But from an ergative-absolutive perspective, the antipassive makes perfect sense: A (the more marked ergative argument) is already pretty demoted, so it makes sense to be able to make it an absolutive, promoting it to S, while doing so means moving the former absolutive argument O out of the way by demoting it to an oblique, or even getting rid of it altogether. (And, for this reason, it is the passive voice which is weird in ergative languages.)

But why would any language want an antipassive (or, for that matter, a passive) in the first place? The answer lies with syntactic ergativity. As mentioned above, all syntactically ergative languages need some sort of compensatory strategy — there always needs to be some method to extract or coordinate ergatives. And, fairly often, the antipassive provides just that method! By promoting the ergative argument to the absolutive, that argument may now be extracted and coordinated in exactly the same way as all other absolutives.

Perhaps a concrete example may help. Let’s take another look at those Chukchi sentences from earlier:

ʔətt-e

dog-ERG

melotalɣ-ən

hare-ABS

piri-nin

catch-AOR.3s.3s

The dog caught a/the hare

[ʔətt-e

dog-ERG

piri-lʔ-ən]

catch-PTCP-ABS

melotalɣ-ən

hare-ABS

the hare that the dog caught

ʔətt-ən

dog-ABS

ine-piri-ɣʔi

ANTIP-catch-AOR.3s

(melotalɣ-tə)

hare-DAT

The dog caught (a/the hare)

[melotalɣ-tə

hare-DAT

ine-piri-lʔ-ən]

ANTIP-catch-PTCP-ABS

ʔətt-ən

dog-Abs

the dog that caught the hare

The first sentence is normal — the dog is in the ergative, the hare is in the absolutive. Thus, by relativising on the absolutive argument, we can say ‘the hare that the dog caught’. But we can’t say ‘the dog that the hare caught’, because the dog is in the ergative, which of course cannot be relativised. However, by applying the antipassive, we can move the dog into the absolutive, with the hare becoming a dative, letting us relativise on the dog to say ‘the dog that caught the hare’. Dixon refers to this as ‘feeding the pivot’ — the S/A pivot here cannot work with an O, so the antipassive needs to be applied, giving an A which may be used to metaphorically ‘feed’ the pivot.

But pivot-feeding is not the only place where the antipassive and passive are applied. If we look at the passive in English, it certainly may be used in sentences like Father was seen by mother and returned, where we want coreference between S and O — but most of the time the passive is used for completely different reasons. This use is pervasive in English, and there are many examples even in this very post: ‘Languages which display such phenomena are said to show syntactic ergativity’. It is easy to see why the passive may be used for other reasons: it may be used to focus on O rather than A, or to avoid mentioning A altogether. It may also be used to when O is more animate or more definite than A (‘I was hit by a tree branch’ sounds a bit more natural than ‘A tree branch hit me’). The passive also has aspectual connotations: it often has a perfective (or even perfect) meaning, focusing on the result of the action rather than the action itself.

Given all this, it should come as no surprise to hear that the antipassive has the opposite implications. The antipassive is often used to focus on A rather than O, or to avoid mentioning O altogether. As with the passive, it may be used when O and A are the ‘wrong way round’ on the animacy hierarchy. With regard to aspect, the antipassive often has an atelic or imperfective (or even progressive/inchoative/inceptive/durative/iterative) meaning, focusing on the action itself rather than the result of that action; it may also imply that the patient was not affected by the action. (A nice Dyirbal example: biya Jani-ŋgu gunyjan ‘John is drinking beer’ places more focus on the beer, whereas the antipassive Jani gunyjalŋanʸu (biya-gu) ‘John is drinking (beer)’ focuses on the drinking; the English translation gives some flavour of that as well.) WALS notes that, depending on the language, the antipassive may be preferred or required if the object is ‘plural (Bezhta), indefinite (West Greenlandic), non-specific (Archi), generic (Diyari) or implicit (Mayan)’. The antipassive may have other language-dependent interpretations as well — for instance, in the North-east Caucasian language Bezhta, Dixon notes that the antipassive has a potential meaning, so it is required in sentences like ‘Brother can boil the water’.

Finally, there is the issue of the distribution of the antipassive. From the discussion above, it may sound like antipassives are restricted to ergative language, and passives are restricted to accusative languages. Certainly this is true as a correlation: antipassives are less common in non-ergative languages than ergative ones. But antipassives are by no means restricted to ergative languages, with plenty of accusative languages having an antipassive. In fact, many languages have both passives and antipassives. And the reverse is also true: passives are not restricted to accusative languages, with some ergative languages containing passives as well. Some ergative languages even have more than one passive! For example, the ergative Mayan language Mam has no less than four passives, with differing semantics: one implies that the agent has lost control over the action, or never had control over it in the first place, another implies motion is involved in the action, and so on. (Lest this sound exotic, note that even English has two types of passives: ‘is X-ed’ focuses on the result of the action, while ‘got X-ed’ focuses on the action itself.) Interestingly though, antipassives tend to be fairly uncommon in general: less than half of ergative languages have an antipassive, and even some syntactically ergative languages lack an antipassive. (For more detail than you would ever want to know on the distribution and typology of the antipassive — and also ergativity more generally — see Chapters 5 and 6 of (Heaton 2013).)

[evmdbm]

This is great. Really clear, but I do have a question about questions (or wh-questions). If A-bar movement involves (well) movement then "John kicked the cat".... "Who kicked the cat?" involves no movement; the subject is in the same place. So in a language with syntactic ergativity and AVO word order replacing John-erg with who-erg involves no movement. It is only "What-abs did John kick?" that involves movement (say). Are we treating ergative and absolutive demonstrably differently here? Is that absolutive extraction only? I just wonder if you might say a bit more about how syntactic ergativity would work here. Do you have any examples in Chukchi or Basque?

Incidentally "You did what?" is a bad example in English as what - I think - stands for a VP. "He kicked what?" is better, but only, I think, used for emphasis.

[bradrn]

This is great. Really clear, but I do have a question about questions (or wh-questions). If A-bar movement involves (well) movement then "John kicked the cat".... "Who kicked the cat?" involves no movement; the subject is in the same place. So in a language with syntactic ergativity and AVO word order replacing John-erg with who-erg involves no movement. It is only "What-abs did John kick?" that involves movement (say). Are we treating ergative and absolutive demonstrably differently here? Is that absolutive extraction only? I just wonder if you might say a bit more about how syntactic ergativity would work here. Do you have any examples in Chukchi or Basque?

Ah, interesting question! I actually don’t have a clue (I’m not too good at syntax myself), but probably someone else would know. Maybe try asking this in the Syntax Random Thread.

Incidentally "You did what?" is a bad example in English as what - I think - stands for a VP. "He kicked what?" is better, but only, I think, used for emphasis.

Yes, but I believe what itself is still an NP.

[evmdbm]

Ah, interesting question! I actually don’t have a clue (I’m not too good at syntax myself),

Well there's true honesty

[bradrn]

Ah, interesting question! I actually don’t have a clue (I’m not too good at syntax myself),

Well there's true honesty

Well, it’s hardly a secret that I’m bad at syntax! (Most recently, I was asking for book recommendations about how to learn it.)

[jal]

"even some syntactically accusative languages lack an antipassive" - shouldn't that be "syntactically ergative"?


JAL

[bradrn]

"even some syntactically accusative languages lack an antipassive" - shouldn't that be "syntactically ergative"?

Yep — thanks for finding this! Should be fixed now.

[bradrn]

Diachronics of ergativity

At this point, we have covered most of the areas in which ergativity can be observed. Yet there is still one last relevant topic: namely, how languages can gain and lose ergativity and accusativity.

Perhaps the simplest method for morphological ergativity to develop is from another case-marker — most commonly some sort of instrumental. For instance, the Hittite ergative case-marker is thought to have developed from the instrumental case-marker -anza. To understand this development, note that an instrument can alternately be conceived as an agent. For instance, a sentence such as John cut the bread with the knife can be rephrased as The knife cut the bread. In a language in which the instrumental is expressed with case-marking, this results in a development John-NOM knife-INSTR bread-ACC cut → knife-INSTR bread-ACC cut. The instrumental can then be easily reinterpreted as an ergative. Note that this diachronic pathway naturally leads to a split system where only inanimates can take ergative alignment, since inanimates are most likely to be given instrumental marking — and indeed this is exactly the situation found in Hittite. This pathway will also result the ergative and instrumental case-markers being formally identical — and recall from earlier that ergative case-markers very often do exactly this, taking on other roles in addition to the ergative.

This sort of development seems common. For instance, an ablative→ergative pathway is attested in a few different languages: as well as Hittite (in which the instrumental does double duty as an ablative), this pathway is also hypothesised for Basque and Trumai. And even more interestingly, some Australian languages allow the agent to be focused by replacing the ergative with the ablative! (A possible explanation: the ablative marks a ‘source’ or ‘origin’, and the agent of an action is also the metaphorical ‘source’ of that action.) Other case-markers, including genitives and locatives, have also been known to become ergative case-markers.

Of course, this is not the only way in which ergativity can develop. Another well-known pathway is through re-interpretation of the passive. If one is most familiar with nominative-accusative languages, the similarities between the passive voice and ergative alignment can seem striking: when a sentence is passivised, the unmarked ‘nominative’ argument is always underlying O, whereas the marked ‘indirect object’ is always the underlying A argument. Compare this to an intransitive sentence (which in most languages cannot be passivised), in which the nominative case marks S. So if a language develops a situation in which transitive clauses are consistently passivised, then in this situation, O will be marked the same as S while A is marked differently — an ergative system. A well-known example of this pathway can be seen in the development from the accusative case-marking pattern seen in Vedic Sanskrit to the split-ergative pattern seen in modern Indo-Aryan languages (Butt & Deo 2017):

1. In Vedic Sanskrit, the passive participle was marked with -ta. (Note that this is the passive participle, rather than the passive voice -ya.) In this form, the patient (promoted to A) is expressed with the nominative and triggers verbal agreement, while the agent (now demoted to patient and usually omitted entirely) is expressed with the instrumental. As with the passive participle in English, this form was used to describe the resulting state of an action. It was subsequently generalised to additionally mark the perfect aspect and the past perfective.
2. In Middle Indo-Aryan, the tense-aspect system was heavily simplified. In particular, the inflectional past tenses disappeared, leaving the passive participle as the only remaining tense-aspect affix referring to past events. Due to this, the passive participle was then reanalysed as a perfective marker. The clause is correspondingly reanalysed as having active voice with ergative marking — A is marked with instrumental, while S and O are marked with nominative. The verb still agrees uniformly with the nominative argument — S/A in the present, but S/O in the past.
3. The case system then underwent syncretism: nominative and accusative case-markers merged, and nominative and instrumental merged for 1p and 2p pronouns. (Thus ergative/instrumental marking is now present only on third person arguments.) However agreement is still with the ‘nominative’ argument, thus making it the only marker of ergative alignment in most clauses.
4. Many New Indo-Aryan languages (e.g. Hindi) then innovated a new ergative marker -ne, leading to a thoroughly ergative pattern in the perfective aspect. (Others, such as Marathi, simply kept the older system.)

Similar results could arise from the development of ablatives to ergatives. Trask and Millar (2015) note that possessive constructions can be used to express statives, e.g. I have a broken window (for My window has been broken), and from there develop into perfects. But in the case of languages which express possessives using constructions like A broken window is to me, the ‘subject’ is now an oblique, and the ‘object’ is now a direct subject. This can then develop into a perfect, giving another source for a split ergative system with ergative alignment in the perfect but accusative alignment otherwise.

A less well-known source of ergative case-markers is from pronominals and determiners. This development can be observed in many Australian languages (McGregor 2017), in which the ergative case-marker is formally similar or identical to a pronominal or determiner (e.g. Jingulu, Wangkumara, Pantyikali). One possible explanation for this trend is in terms of discourse. There is a tendency in discourse for subjects (especially A arguments) to give old information. It may have been the case that apposition of a pronominal or determiner was used to mark new information in A position. (An English example gives something of the flavour: ‘It wasn’t Jane who washed the dishes; that John did it’. Or consider a cleft: ‘It’s John who washed the dishes’.) It is then a natural next step for the pronoun to affix or cliticise onto the head NP, becoming an ergative marker. Evidence for this comes from the fact that many of these languages have optional ergative marking, in which the ergative case can be omitted only with expected agents.

A similar case can be observed in the Trans–New Guinea language Tauya. In this language, agents (both transitive subjects and some intransitive agents) are optionally marked with the ergative case-marker -ni (MacDonald 1985). This is very similar to the third person singular pronoun ne, also phonetically [ni]. Note also that as with most Papuan languages, Tauya has a basic word order of SOV. Due to this, MacDonald proposes that, originally, Tauya had sentences of the form A ne V, with a third person singular object. The pronoun was then reanalysed as a suffix, A-ni V, which then spread to other agents. Finally, other object pronouns were reanalysed as verbal prefixes. As unusual as this development sounds, it does nonetheless explain several other features of Tauya. For instance, the object prefixes are identical in form to the personal pronouns, except for third person singular objects, which correspond to a null prefix. This agrees with the explanation above, which suggests that the 3s pronoun ne cliticised to the agent before the other pronouns became verbal prefixes. Additionally, it turns out that the ergative case-marker is obligatory only with human objects; this corresponds to the fact that Tauya object prefixes are used only when O is human.

In some cases, the transition from an accusative to an ergative pattern may be more complex. A good example of this is given by the Carib languages of South America. These languages show two construction types: an accusative type, in which word order is accusative, and case-marking and agreement are neutral; and an ergative type, in which word order is ergative, case-marking is given on A arguments, and the verb takes an S/O proclitic. There is a distinctive split between the use of these construction types: ergative constructions are used consistently in subordinate clauses (which often take the form of nominalisations), while depending on the language, main clauses may use either accusative or ergative constructions. Given the generalisation that grammatical changes apply first to main clauses and only later to subordinate clauses, we might expect proto-Carib to have originally been entirely ergative, with accusative main clauses being innovated later. But in fact, on close examination we find a quite different situation — proto-Carib was originally accusative with ergative subordinate clauses, with ergativity in main clauses being an innovation. (This is as of 1994, based on Dixon (1994) citing Gildea (1992); Dixon does state that ‘more work is needed fully to validate [this conclusion]’, but even after 26 years I can find no such future work.)

While the preceding cases have outlined accusative→ergative changes with clear semantics, occasionally languages can become ergative without any semantic motivation whatsoever. Such a case has been outlined by Anderson (1992), which I present with slight modifications. Consider a VSO language in which the nominative is marked with -s, and the accusative is marked with -t. If this language then loses all clause-final obstruents, S and O both lose their marking, with -s remaining only on A — i.e. an ergative system. It is possible to conceive of a similar situation with verbal agreement: if a language with polypersonal agreement uses subject and object agreement suffixes which are identical in form, and subsequently undergoes a sound change at the right edge of the word, the markers used for S and O will then become different from the markers used for A. However, I am not aware of any language which has become ergative purely from these sorts of sound changes. (Panye (1983) reportedly discusses such a case in Yagua, but I am unable to confirm this as I cannot access that paper.)

Of course, it is also entirely possible for ergative languages to become accusative. Though this seems to be less common than the reverse, it is certainly attested. One of the most striking examples of this development can be seen in the Pamir sub-group of the Iranian languages. Though Proto-Iranian was accusative, this developed into an ergative system though reinterpretation of the passive as described above, leading to a situation with three cases: in the present tense, nominative was used for S/A and accusative for O, while in the past tense, nominative was used for S/O and an oblique case (the old genitive) for A. Proto-Pamir then merged the marked accusative and oblique cases, giving a split system with just two cases: a ‘direct’ case used for S/A in present tense and S/O in past tense, and an ‘oblique’ case used for O in present tense and A in past tense. The modern Pamir languages have then modified this system in various ways. Yazgulyam, for instance, has developed a new accusative case in both tenses, giving nominative-accusative alignment in the present tense and tripartite alignment in the past. Rushani, by contrast, expanded the use of the oblique marker for O into past tense, giving its infamous ‘monster raving loony’ alignment: in the past tense, ‘oblique’ is now used for A and O, while ‘direct’ is restricted to S. Of course, this means that A and O are no longer distinguished in past tense, so younger speakers tend to use the direct case for A in past tense as well as present tense; they also use the preposition az before O. These changes have gone even farther in Bartang, and az has developed into an accusative case-marker a- — a nominative-accusative system.

Similar developments — expanding the range of one case-marker to include more NPs — seem to be common. For instance, amongst the Kartvelian languages, Georgian and Svan reportedly have ergative case-marking only in the aorist (Dixon 1994). In Mingrelian, the ergative case-marker has been extended to cover S, giving accusative case-marking; meanwhile, in Laz, the ergative case-marker has been extended to A in tenses other than the aorist, making the case-marking even more strongly ergative. Similar developments appear to have occurred in Tibeto-Burman. Mayan verbal agreement also shows similar developments, where agreement markers which formerly agreed with A only have been extended to certain S markers. And, in languages such as Bengali and Oriya, the minimal ergative case-marking of Middle Indo-Aryan (outlined above) has completely disappeared to be replaced by the nominative case, with the associated ergative agreement system being similarly replaced by a nominative one (Butt & Deo 2017).

Another possible source for accusative marking is the reinterpretation of an antipassive. This proceeds via a pathway analogous to that outlined above, in which a passive construction (with patient promoted to nominative and agent demoted to oblique) becomes reanalysed as active (with agent given ‘oblique’/ergative marking and patient given ‘nominative’/absolutive marking). In an ergative language, it is similarly possible for an antipassive construction (with agent promoted to absolutive and patient demoted to oblique) to be reanalysed as active (with agent given ‘absolutive’/nominative marking and patient given ‘oblique’/accusative marking). Such a change appears to have taken place in the Tangkic languages Lardil and Kayardild. As I mentioned earlier, the closely related language Yukulta has two constructions: in one, A is ergative, O absolutive, and the verb takes transitive suffixes, while in the other, A is absolutive, O dative, and the verb takes intransitive suffixes. The latter antipassive-like construction is used in most clauses; the former is used only with statements of past fact and future intention (and even then not all such statements). It is clear that, if generalised to all clauses, the latter construction would give a nominative-accusative case-marking system, and that appears to be exactly what has happened in Lardil and Kayardild. For instance, the Lardil accusative case is cognate with the Yukulta dative case. Additional confirmation comes from imperatives, which have unusual case-marking: the O argument takes no inflection, rather than the accusative as expected. This is related to the fact that antipassives are rarely applied to imperatives, and how imperatives generally tend to preserve more conservative features.)

_____________________

Summary and future work

Well, this thread took longer than I expected (to say the least), but I think I’ve now managed to cover all the main areas involved in ergativity. Given the length of this series, however, I think it would be useful to provide a summary of the key takeaways:

• S is the sole argument of an intransitive verb; A and O are the subject and object respectively of transitive verbs.
• A is typically the agent which initiates an action, whereas O is typically the patient which is affected by an action. S can be either.
• In a nominative-accusative (a.k.a ‘accusative’) alignment, S and A (‘nominative’ arguments) are marked the same, whereas O (‘accusative’ argument) is marked differently.
• This contrasts with an ergative-absolutive (a.k.a ‘ergative’) alignment, in which S and O (‘absolutive’ arguments) are marked the same, whereas A (‘ergative’ argument) is marked differently.
• This marking can be via nominal case-markers, verbal agreement, or (more rarely) word order.
• In case-marking, the case used for S (nominative or absolutive) is typically unmarked or less marked, with the other case (accusative or ergative) being more marked. (Note that this has the effect of explicitly marking agentivity in ergative systems and patientivity in accusative systems; the unmarked semantic role is the patient in ergative systems and the agent in accusative systems.)
• In verbal agreement, it is possible to agree only with S/A (for accusative alignment) or S/O (for ergative alignment); agreement only with the ergative or only with the accusative is rare, if not nonexistent.
• It is possible to have nominative agreement with ergative case-marking; the reverse is unattested or nearly so.
• Pure accusative or ergative systems are practically impossible; more usually, there is a ‘split system’, which uses ergative agreement in some places and accusative agreement in others.
• One common type of split alignment is ‘active-stative’ alignment. In this alignment, A and O are marked differently to each other, while S is marked like A when it is an agent, and like O when it is a patient. In some active-stative systems, the marking of S is lexically determined (‘Split-S’); in others, it can be different for each sentence (‘Fluid-S’). Active-stative systems are common in verbal agreement, but apparently unattested with case-marking.
• Another type of split alignment is ‘animacy-based’. In such systems, ergative alignment is used for less animate/agentive NPs (ensuring they are most marked when used as agents), while accusative alignment is used for more animate/agentive NPs (so they are most marked when patients). This is done through an ‘animacy hierarchy’: 1 > 2 > 3/demonstratives > proper nouns > humans > animates > inanimates. Items on the left of the hierarchy are accusative, whereas items on the right are ergative, with the exact dividing point being language-specific. Interesting alignment systems are frequently present around the centre of the hierarchy, where the two alignments meet.
• Many split systems have a split based on tense or aspect. In such systems, ergative alignment is used for past/perfective, while accusative alignment is used for present/imperfective. (Unlike most splits, this seems to have little clear semantic basis, though the pattern is still remarkably consistent.)
• A few languages have a split based around mood or clause type (the two can be difficult to distinguish). There are few patterns, except for the consistent syntactic basis: clauses where agentivity is expected get accusative, clauses where patientivity is expected get ergative.
• A notable subtype of split ergativity, cross-cutting the above types, is ‘optional ergativity’, in which the ergative case-marker can be omitted under certain circumstances. This usually is related to the animacy difference between A and O or the expectedness of the A argument in discourse, but in various languages can also occur under numerous different circumstances.
• And, of course, all the various types of split systems above can be combined, occasionally creating very complicated conditions for splits to occur.
• Though morphosyntactic alignment does not affect most parts of syntax, there are some areas where it has an impact. Most commonly, in Ā-movement, many ergative languages forbid movement of the ergative argument; only the absolutive may move. Conjunction reduction is similar in some ergative languages: in a sentence like mother saw father and returned, it is the mother who returned in an accusative language, but the father in some syntactically ergative languages.
• To cope with this, many ergative languages have an antipassive: a valency-changing operation which promotes A (ergative) to S (absolutive), while demoting O to an oblique.
• Diachronically, ergative case-marking can originate from other case-markers (usually instrumentals, ablatives and genitives), reinterpretation of a passive construction, affixation of pronouns, or occasionally phonological changes. It can be lost through generalisation of another case-marker at the expense of the ergative, or through reinterpretation of an antipassive construction.

__________
One final thing, before I go. It’s just so happened that I’ve managed to find quite a bit of useful information on ergativity after writing the relevant posts, quite a bit of which is actually pretty relevant to a complete understanding of ergativity in general and its application to conlanging in particular. (This is especially so for the earlier parts, for which my research was a bit sloppy; also because I’ve started reading a lot more of the linguistic literature since then.) I’d quite like to integrate this information into the earlier parts, but I’m not quite sure how I could do it. One option is to progressively edit that information into the relevant posts, with regular updates as to what’s changed. Another option would be to rewrite this series as an actual typeset book which properly covers this information, especially since it’s ended up practically book-length anyway. (I personally prefer this option, as it would also give me an opportunity to clean up the posts, expand the focus of this series a bit, etc.) Or, of course, I could just refrain from messing any more with this thread, though my perfectionism doesn’t like that idea. But what I want is irrelevant: which option would you, as readers, like best?

And… that’s about it, I think. Thanks everyone for reading!

[Ketsuban]

The book option sounds interesting, especially if there's more to say about active-stative as a subset of ergativity - it might be worth asking Zomp about publishing under the Yonagu Books imprint.

[bradrn]

The book option sounds interesting, especially if there's more to say about active-stative as a subset of ergativity - it might be worth asking Zomp about publishing under the Yonagu Books imprint.

Oh, I wasn’t seriously thinking of actually publishing it — I was mostly thinking of turning the posts into a properly typeset PDF document which you could download. (Though if someone were to approach me about publishing it, I would hardly say no!)

As for active-stativity: That is indeed an interesting topic… I haven’t seen much about it, but I really should look into the area some more. (Particularly interesting is its distribution compared to other types of split ergativity — indeed, many deny that it should be considered ergative at all.) I’ll be sure to write more about it if I find anything.

[mèþru]

Really useful stuff! Thanks for posting this guide

[bradrn]

Really useful stuff! Thanks for posting this guide

You’re very welcome! I’m glad you found it useful.

[jal]

I once developed a conlang that got it's ergativity from moving from SVO to SOV. It already had the possibility to use SvOV, where "v" was a forward referencing pronoun, but that got regularized and "v" became an ergative clitic. In pseudo-English:

Parent language:
Man dog see
Man it see dog-with-very-large-ears

Decendent:
Man it see dog
Man it see dog-with-very-large-ears

Final:

Manit see dog
Manit see dog-with-very-large-ears

I wonder whether any natlang ever had such a development.


JAL

[Vardelm]

I wonder whether any natlang ever had such a development.

Not that I've seen (which isn't saying too much) but it looks pretty natural!

[bradrn]

I once developed a conlang that got it's ergativity from moving from SVO to SOV. It already had the possibility to use SvOV, where "v" was a forward referencing pronoun, but that got regularized and "v" became an ergative clitic. In pseudo-English:

Parent language:
Man dog see
Man it see dog-with-very-large-ears

Decendent:
Man it see dog
Man it see dog-with-very-large-ears

Final:

Manit see dog
Manit see dog-with-very-large-ears

I wonder whether any natlang ever had such a development.

I don’t know of any language with this exact development, but it’s pretty similar to what happened in Tauya (summarised above).

[evmdbm]

You posted on Christmas day? Dedication. It was really interesting and helpful though

[bradrn]

You posted on Christmas day? Dedication. It was really interesting and helpful though

I’m pleased you liked it! (Though as it happens, it wasn’t Christmas in my timezone… and I’m not Christian anyway, so it’s not a particularly important day to me.)

[willm]

I’m sorry if this has already been addressed in this thread, but I couldn’t find mention of it. Does anyone know if the objects of adpositions in ergative-absolutive languages tend to be in the ergative case or the absolutive case? I would expect the absolutive to be used, since it is the default and includes the accusative, or perhaps for it to depend on the semantics of the adposition. This would only be apparent, I would expect, in languages that use affixes to mark case. Georgian and Basque seem to mostly use additional cases, but I’m curious about what is usual in languages that use the ergative or absolutive case.