Split ergativity — Part 3
Splits on tense, aspect and mood
At this point, we have covered two types of split ergativity: namely, splits on the nature of the verb (i.e. active-stative splits) and splits on the nature of the NP (i.e. animacy-based splits). Now let’s investigate another very common type of split ergativity: splits on tense, aspect or mood (abbreviated TAM). In a language with a TAM split, one morphosyntactic alignment is used in a clause with some TAMs, while another morphosyntactic alignment is used in a clause with other TAMs.
Earlier, it was noted that, due to their nature, active-stative and animacy-based splits usually occur with a specific type of morphological ergativity: active-stative splits depend on the specific verb which is being used, and so these splits most often occur with verbal agreement, while animacy-based splits depend on the specific NP which is being used, and so most often occur with case-marking. Since tense, aspect and mood aren’t particularly intimately related to either the verb or the NP to the same extent as the two other split systems mentioned, we might expect that TAM splits could occur using either tense-marking affixes or verbal agreement affixes. And this is indeed the case: languages such as Burushaski and Hindi show a TAM split using case-marking affixes, while languages such as Yucatec and Chukchi show a TAM split using verbal agreement.
The split systems which have already been covered clearly show a strong ‘preference’ (for lack of a better word) in terms of which part of the system gets which type of marking. For instance, animacy-based splits almost always have ergative marking for less animate NPs and absolutive marking for more animate NPs. Similarly, splits on tense, aspect or mood also have a strong preference for using ergative marking in some situations and accusative marking in others:
- In a split on tense, the past tense will be ergative and the present tense will be accusative.
- In a split on aspect, the perfective will be ergative and the imperfective will be accusative.
- Splits on mood are rarer and show fewer clear patterns, but one relatively common pattern is for the imperative to be accusative while most or all other moods are ergative. Another attested pattern is for negative clauses to be accusative while positive clauses are ergative, although strictly speaking this isn’t quite a modal distinction. (Note: this category of ‘modal splits’ seem to have quite a substantial overlap with ‘splits by clause type’, as presented below; there are quite a few splits which can be categorised as both.)
Of course, there are variations on these patterns. Most prominently, it is relatively common for a language with a TAM split to use direct marking (i.e. leave out cases entirely) rather than accusative marking; this is the case in Burushaski, for instance, where there is an ergative system of case-marking in ‘past-based tenses’ (as Dixon calls them), but no case-marking in ‘non-past tenses’. Basque and Lhasa Tibetan also show this pattern with an aspectual split, albeit with a few subtleties (see below for details). Since the absolutive is usually unmarked anyway in ergative-absolutive systems, such a system could also be considered to be a form of optional ergative marking, where ergative case-marking is present in past or perfective but not present or imperfective.
Another variation with this sort of split is the exact point where ergative changes to absolutive. Above I stated that these splits always occur at a certain point, but Jessica Coon suggests that aspectual splits follow a hierarchy, in a similar way to animacy-based splits, where the left side is accusative (or at least non-ergative) and the right side is ergative:
progressive > imperfective > perfective
One might expect a similar hierarchy with tense splits — most probably ‘future > present > past’ — but I haven’t seen such a hierarchy described anywhere. As for modal splits, it unfortunately seems much harder to draw a clear hierarchy for those, given the absence of many clear patterns.
And finally, as with the other types of splits we have seen so far, there are always some languages which don’t follow the rules, although these are of course quite rare. The most notable example is the Macro-Jê language Shokleng, which has an aspectual split in main clauses — but it is the ‘stative aspect’ (as McGregor calls it; it is presumably related to the perfective) which gets ergative marking, and the ‘active aspect’ which gets an accusative system (which interestingly enough is marked nominative). This is a fairly unusual system, although it’s worth mentioning that the split system in Shokleng is unusual in other ways as well (see below for details).
We saw in previous sections that split systems are generally pretty easy to explain: accusative marking is used in more agentive situations, while ergative marking is used in less agentive situations. Unfortunately, splits on tense and aspect seem to be far more difficult to explain, with several different explanations having been proposed. There have been too many explanations proposed to list them all, so I won’t go into details; I will just say that none of the explanations proposed (including my own) seem particularly plausible.
On the other hand, modal splits often have much more obvious explanations than tense and aspect splits. For instance, one somewhat common modal split is a situation where the imperative gets accusative marking, but non-imperative moods get ergative marking. This split has a clear explanation: the imperative emphasises the fact that S or A is in control of the activity (and thus can be ordered to perform it), whereas non-imperative moods place no such emphasis on this agentivity.
Splits by clause type
A rarer type of split ergativity consists of a split between different types of clause. This is sometimes summarised as a split between ‘main’ and ‘subordinate’ clauses; however, strictly speaking this description is incorrect, since languages with this sort of split usually have splits between specific types of subordinate clauses rather than splits between
all main and
all subordinate clauses. Generally, these splits operate on semantic grounds:
- Purporsive subclauses (expressed in English as infinitives or clauses starting with ‘in order to …’) and imperative clauses generally describe an action which is under the control of the agent; thus, they will get accusative marking.
- By contrast, relative clauses (e.g. ‘I saw a tree [which fell over]’) and non-purposive complement clauses (e.g. ‘I think that [puppies are cute]’) simply describe an event without ascribing agency or absence of agency; thus they may recieve either accusative or ergative marking.
- Another way of viewing relative clauses is through their association with the animacy hierarchy. In English, relative clauses are common with humans, animates and inanimates, but are almost never used with pronouns and demonstratives; this is especially prominent with restrictive relative clauses, which may not occur with first and second person singular pronouns. Other languages show a similar pattern. So relative clauses are associated with the right-hand side of the animacy hierarchy, and thus can get ergative marking.
So if, for instance, there is a split between purporsive clauses and other clauses, purporsives will get accusative marking while other clauses will be ergative. Conversely, if there is a split between relative clauses and other clauses, relative clauses will get ergative marking while other clauses will get accusative marking. In general, in a split between one type of subclause and all other clauses, the subclause will get one morphosyntactic alignment while the other clause types will get the opposite alignment.
By and large, the attested examples of splits by clause type support this reasoning. For instance, Päri has a split between imperative and non-imperative clauses; as expected, imperative clauses are accusative (albeit in a marked nominative pattern), whereas all other clauses get ergative marking. (Although note that this has also been described as a split on mood; as noted above, there is a substantial overlap between splits by mood and splits by clause type). Similarly, Tsimshian has a split between ‘subjunctive’ subordinate clauses and other clauses (Boas 1911, cited in Dixon): here, subordinate clauses get ergative marking, whereas main clauses get accusative marking. (Interestingly, recall that non-purposive complement clauses can be either ergative or accusative, so theoretically there should be no problem with the split going the other way around here.)
Of course, there are always exceptions. In this case, one of the more interesting languages with a main/subordinate clause split is Shokleng. Unusually, Shokleng does have a split between
all subordinate clauses — which get ergative verbal agreement — and
all main clauses — which get ergative or accusative verbal agreement depending on their aspect (this split was described above). McGregor notes that this unusual alignment may be a consequence of the unusual aspectual split in Shokleng, as described above: he theorises that since both subordinate clauses and stative aspect typically describe backgrounded information, it makes sense that if one has an unusual alignment, then the other will too.
Combinations of splits
The four splits already covered suffice to explain the majority of split ergative systems. However, many more complex split systems don’t just use one type of split; instead, they use combinations of two (or more rarely three) different types of split. Some examples (from Dixon) will illustrate:
- Balochi combines an animacy-based split and a split on TAM: ergative case is marked only in perfective aspect, and only on third person pronouns and nouns. When either of these conditions are not satisfied, Balochi uses a neutral system instead (S,A,O are all unmarked). Burushaski and Sumerian are similar.
- Kuikúro also combines an animacy-based split and a TAM split, but in a slightly different way. In Kuikúro, moods are divided into ‘interactive’ (including imperative, hortative, intentional) and ‘descriptive’ (presumably everything else). As expected, the descriptive moods get obligatory ergative marking (since they’re less agentive than the interactive moods), but the marking used for the interactive moods is split by animacy: ergative marking is obligatory for third person A, optional for second person or exclusive first person plural A, and completely disallowed for first person singular or inclusive first person plural A.
- Mawayana combines an active-stative system and a TAM split. In this system, intransitive verbs are divided into two classes. One set of verbs is always S=O; for these, the S argument is marked on the verb with the same set of suffixes used for O. The other set is split on TAM: the O agreement suffixes are used in ‘present continuous and near past clauses’, but the A agreement suffixes are used ‘in remote past and future clauses’.
- Yukulta combines an animacy-based split and (debatably) a TAM split. There are two different transitive constructions: construction 1 (as I will call it) marks A with the ergative and O with the absolutive, and marks the verb with transitive suffixes, whereas construction 2 marks A with the nominative and O with the dative, and marks the verb with intransitive suffixes (Legate 2017, Dixon 1994). Construction 2 is used in situations when O is less affected by the action: in particular, it is required for negative clauses and desideratives. (Dixon analyses this as a realis/irrealis mood split, but Legate disputes this.) It is also required when A and O are both pronouns and O outranks A on the animacy hierarchy. In all other situations, construction 1 is used.
- As mentioned above, Shokleng combines a TAM split with a split by clause type. In subordinate clauses, ergative is always marked. But in main clauses with ‘stative aspect’, the ergative is marked via a postposition, whereas in main clauses with ‘active aspect’, the nominative is marked via a (different) postposition.
- Tsimshian apparently combines an animacy-based split with a split by clause type, but Dixon doesn’t go into details.
- Georgian combines an active-stative split, animacy-based split, and TAM split: aorist and perfect verbs get split-S marking, but the ‘ergative’ case (the one used for A and S=A) is only marked for nouns and third person pronouns.
- Cavineña combines an animacy-based split, TAM split and split by clause type. In ‘high activity’ transitive clauses, the ergative case is always marked on nouns, but it is often unmarked on pronouns. But in ‘low activity’ transitive clauses (i.e. those which express ‘negation, potentiality, intention, sensation and contrary to fact’ (Camp 1985, quoted in Dixon), i.e. ‘mood types that downplay the role of the agent’ (Dixon)), the ergative is almost always marked, even on pronouns.
- Some Tupí-Guaraní languages use a combination of an active-stative split, animacy-based split and split on clause type. These languages have four different sets of verbal agreement prefixes: set 1 is used for A and S=A, set 2 is used for O and S=O, set 3 is used for certain serial verb constructions, and set 4 (which only marks number) is used for O. In intransitive main clauses, set 1 is used for S=A verbs and set 2 is used for S=O verbs. In transitive main clauses, the choice of sets depends on the animacy of A and O: when O is third person, then the verb agrees with A (using set 1) and O (using set 2); when O is higher on the animacy hierarchy then A, the verb only agrees with O (using set 2); and when A is first person and O is second person, then the verb again only agrees with O (but using set 4). But in temporal and conditional subordinate clauses (which in fact make up all subordinate clauses in these languages), the verb agrees only with O and all S arguments (both S=A and S=O) using the set 2 prefixes. So subordinate clauses are fully ergative, while main clauses combine an active-stative split (for intransitive verbs) with an animacy-based split (for transitive verbs).
Note that this means that all combinations of two of the split types which have been described are attested; any type of split ergativity may co-occur with any other type of split ergativity. Similarly, almost every combination of three types of split ergativity is attested, the only unattested combination being a combination of an active-stative system + TAM split + split on clause type.
Rarer types of split
(Much of the information in this section is from the 2017 article
Split Ergativity in Syntax and at Morphological Spellout, by Ellen Woolford.)
The split types listed above are sufficient to describe the vast majority of known split ergative systems. However, there also do exist some more unusual types of split, including the following:
- Nepali apparently has a split between ‘individual level predicates’, which predicate ‘a lasting/inherent property of a referent’, and ‘stage level predicates’, which ‘say something about a property of a referent that holds for a slice of that referent’s … existence’ (Butt and Poudel 2007, quoted in Woolford). Individual level predicates, as in the sentence ‘Ram knows English’, get ergative alignment; stage level predicates, as in the sentence ‘Ram will speak English (today)’, get accusative alignment. Woolford notes that in copular clauses, Nepali uses two different copulas for these two different predicate types; outside of copular clauses, the distinction is maintained using an alignment split instead.
- Folopa and Mongsen Ao (mentioned earlier) have an unusual ‘socially-controlled split’, in which the ergative case ‘is interpreted in certain contexts as asserting or emphasizing the subject’s control/ownership of the object, which can be rude in some social situations, and indicates theft in others’ (Woolford). In particular, Folopa is unusual in that it appears to have a fluid-S system with transitive verbs rather than intransitive verbs: the A argument may be marked either with the ergative or nominative case depending on the connotation which the speaker wants to convey.
- Dixon describes Yagua and Pajonal Campa as also having an unusual type of fluid-S system. In these languages, the accusative seems to be the unmarked case for S arguments; the absolutive is used mainly ‘to mark a change of state, a change of location, or a point of episodic climax’; that is, ‘to highlight some new information (but not a new participant)’.
Case Studies
To finish our discussion of split ergativity, I now want to present a couple of more in-depth case studies of two interesting split ergative systems — namely those of Basque and Lhasa Tibetan. The examples of split systems presented above are all fairly easy to understand; their splits only have one or two parts to them, without many complicating factors. But languages don’t always do this; often, split ergative systems can become much more difficult to analyse (as in Basque). And even languages with fairly ‘normal’ split ergative systems can have many subtleties (as in Lhasa Tibetan).
Basque
Basque is well-known as the only ergative language in western Europe. This is manifested in Basque through both case-marking — where A is marked with
-ak and O/S are marked with
-a — as well as verbal agreement (Coon 2013):
- Ehiztari-ak
- hunter-ART.ERG
- otso-a
- wolf-ART.ABS
- harrapatu
- caught
- du.
- AUX(have)
The hunter has caught the wolf.
- Otso-a
- wolf-ART.ABS
- etorri
- arrived
- da.
- AUX(be)
The wolf has arrived.
However, Basque also shows an aspectual split, where the perfective and imperfective aspects receive ergative marking, while the progressive aspect receives neutral marking, in accordance with the aspectual hierarchy given above:
- Emakume-a
- woman-ART.ABS
- ogi-a
- bread-ART.ABS
- ja-te-n
- eat-NML-LOC
- ari
- PROG
- da.
- AUX(be)
The woman is eating the bread.
- Emakume-a
- woman-ART.ABS
- dantza-n
- dance-LOC
- ari
- PROG
- da.
- AUX(be)
The woman is dancing.
At first, this appears to be a fairly standard split ergative system. However, Coon (citing Laka (2006) as the original proposer of this idea) suggests that Basque is not split ergative at all, but rather uses a thoroughly non-split ergative alignment. According to Coon, the apparent split is caused by the Basque progressive being a biclausal construction, with the ‘progressive marker’
ari rather than
ja- eat being the main verb of this sentence. In this analysis, the first sentence above — namely
The woman is eating the bread — is in fact biclausal, with
eating the bread given as a locative. This makes the ‘progressive’ verb
ari intransitive, and hence the ‘subject’
emakume woman must be in the absolutive rather than the otherwise expected ergative. An English equivalent would be something like
The woman is engaged [at eating the bread]; obviously the verb
engaged is intransitive here, but since English is nominative-accusative, both the A argument of
eating and the S argument of
the woman get the same case-marking (which Coon suggests could be why mainly nominative-accusative languages don’t normally exhibit aspectual splits). Clearly, on this basis, it is difficult to categorise Basque in terms of whether it is split ergative or just plain ergative with an interesting progressive construction.
(Coon lists many other language with this variety of split, including Chol, Qʼanjobʼal, and some Nakh-Daghestanian languages. Based on this data, she then goes on to suggest that in fact all aspectual splits are of this type, although I’m not too sure about this conclusion.)
Lhasa Tibetan
(Source: most of the information in this section comes from Scott DeLancey’s article
The Blue Bird of Ergativity.)
Unlike Basque, Lhasa Tibetan doesn’t have any particularly hard-to-classify features; instead its alignment is fairly straightforward and similar to other ergative languages. However, I cover it due to the complexity of its alignment system, which involves syntactic, semantic and pragmatic features, with some subtleties.
Most prominently, Lhasa Tibetan has a split on aspect (briefly mentioned earlier), where perfective clauses get obligatory ergative marking while imperfective clauses get optional ergative marking:
- nga-s
- 1s-ERG
- stag
- tiger
- bsad-pa yin
- killed-PERF/CONJ
I killed a tiger.
- nga(-s)
- 1s(-ERG)
- stag
- tiger
- gsod-kyi yod
- ?
I am killing a tiger.
- nga
- 1s
- bod-la
- Tibet-LOC
- >gro-gyi yod
- ?
I am going to Tibet.
(DeLancey provides no gloss for the last two sentences, although I have provided a gloss for some words where I can.)
However, Lhasa Tibetan also shows a curious phenomenon: the so-called ‘ergative’ case can also be used to optionally mark S! This only occurs in the perfective aspect, and for certain ‘volitional intransitive verbs’, particularly motion verbs (Tournadre 1995):
- nga(-s)
- 1s-ERG
- bod-la
- Tibet-LOC
- phyin-ba yin
- went-PERF/CONJ
I went to Tibet.
(Terminological note: given that
-s is used to mark A and S, it seems like it should be called a nominative rather than ergative case. But all the sources I’ve seen call it ergative, and I will do the same.)
It thus seems that Lhasa Tibetan is also an active-stative language, albeit one where the vast majority of verbs are in the S=O category. (Such systems seem to be not uncommon; another example is Warlpiri, where the ergative case is used to mark the S argument of only a very small set of intransitive verbs (Nash 1980).)
Now, I am sure that there are several ways to analyse this system, but I will only cover two. The obvious ‘analysis’ (insofar as the word can be applied) is to simply list how this system can be expressed as a combination of various types of split ergativity: Lhasa Tibetan has an aspectually split system where non-perfective clauses get optional ergative marking, and perfective clauses get active-stative marking. This seems clear enough, with the only problems being terminological: the same case is used to mark A in non-perfective clauses and S/A in perfective clauses, so it is hard to know whether to call this an ergative or nominative case.
On the other hand, DeLancey claims that this analysis obscures what is really happening in Lhasa Tibetan. Rather, he prefers to invoke ‘the Hopper and Thompson notion of global transitivity’: a prototypically transitive verb has two arguments, perfective aspect, and volition, whereas a prototypically intransitive verb has one argument, non-perfective aspect and less volition. In Lhasa Tibetan, the closer a verb is to prototypical transitivity, the more likely it is that its argument will be marked ergative — for instance, DeLancey says that ‘A two-argument clause can have an ergative argument even if it is not perfective or active … But a formally intransitive clause must be both active and perfective to qualify for ergative marking’.
However, DeLancey notes that neither of these analyses address a fundamental fact about the Lhasa Tibetan case system: the ‘optional’ ergative marker, as described above, is not in free variation between present and absent, but in fact serves a pragmatic role, as is often found in systems with optional ergative marking. Specifically, the presence of the ergative marker on A or S gives contrastive focus to that argument. (An example was given previously in the section on optional ergative marking.)
So to conclude, the distribution of the Tibetan ‘ergative case’ depends on a number of syntactic, semantic and pragmatic factors:
- Obligatory on the A argument of perfective transitive verbs
- Optional on the A argument of non-perfective transitive verbs and the S argument some particularly agentive perfective intransitive verbs; the presence of the ergative marker here indicates contrastive emphasis
- Disallowed on the S argument of non-perfective intransitive verbs.
If we insist on a further analysis — as DeLancey seems to do — I am sure that we can derive all sorts of interesting conclusions (as DeLancey indeed does), but I don’t see such an analysis as necessary, given that each point the description above is already pretty similar to the simpler split systems we’ve seen before.