(Note: This chapter is intended as an overview of ergativity for those who know little to nothing about it, so it may be a bit boring if you are already aware of the fundamentals — but I would still recommend that you at least skim through the chapter below, since I do cover some concepts which are not often mentioned in discussions of this subject. Hopefully the next chapter will be more interesting! Also, I’m not entirely confident how clear the descriptions below are, so please tell me if something I’ve written seems unclear, or if you have any questions!)
What is ergativity?
The universal grammatical relations
Let’s start with the basics. Recall that all languages contain two main types of clauses:
- Intransitive clauses, which contain a verb plus one core argument (e.g. I laugh). Verbs which can occur in these clauses are also called intransitive verbs.
- Transitive clauses, which contain a verb plus two core arguments (e.g. I see you). Verbs which can occur in these clauses are also called transitive verbs.
(Some languages also have
ambitransitive or
labile verbs, which can be used both transitively and intransitively; I will return to these later.)
Based on this, we can classify noun phrases in a sentence as being in one of three ‘primitive’ grammatical relations:
- The sole argument of an intransitive verb, denoted S
- The subject of a transitive verb, denoted A
- The object of a transitive verb, denoted O
(Note that conventions differ for these abbreviations. The ones above are Dixon’s, and seem to be more standard, but in the
ALC, zompist prefers using
e,
a,
p. In this text I will use Dixon’s abbreviations.)
For instance, the noun phrases in the following sentences can be classified as:
I (S) laugh
I (A) see you (O)
I (S) go
I (A) go there (O)
These relations are almost always marked in some way: either through case-marking on the noun phrases, agreement markers on the verb, word order, or a combination of these. For instance, in English, A is the first argument, whereas O is the second argument. In a language like Latin, they would be determined by the fact that A is in the nominative case, while O is in the accusative case.
A, O and control
For a transitive verb with two arguments, one argument is mapped onto A, while the other is mapped to O. An important question about these is this: how, for a particular verb, can one determine which argument is A and which argument is O?
The answer has to do with the idea of
control over an action. In the majority of languages (those with what Dixon calls ‘syntactic-based marking’), given a verb, the arguments which get mapped to A and O roles can be determined by looking at a prototypical usage of that verb. Generally, in that prototypical usage, one argument will have control over the action, while another argument will be most affected by that action. The former argument is called the
agent and is assigned the A role, while the latter argument is the
patient and is assigned the O role. For instance, with the word
hit, a prototypical usage would be something like
He hit me; in this prototypical sentence, the hitter
he is in control of the action, while the ‘hittee’
me is most affected by the action. Thus, on the basis of this prototypical usage, the hitter is mapped to A role, while the hittee is mapped to O role. (English uses (mostly) word order to mark A and O, and so this is why the hitter comes before the verb while the hittee comes after it.) Note the use of the ‘prototypical’ meaning to determine A and O: even in a sentence like
The falling branch hit me, where the branch is not really in control of the action, the branch is still mapped to A and I am mapped to O, based on the mappings determined from the prototypical sentence.
The important takeaway here is this:
The A relation represents the argument which prototypically controls the event, while the O relation represents the argument which prototypically does not control the event. This is in fact an important statement about A and O, and will become very important later, when we look at split ergativity.
(Aside: The above procedure of using a prototypical sentence to find the agent and patient may seem a bit restrictive: in particular, many non-prototypical sentences do not have A as the agent and O as the patient — and what about verbs where there is no obvious agent or patient? This second problem in particular can be seen in English:
I like the painting has
I as A and
the painting as O, but
The painting pleases me does the opposite. There are several languages which solve this problem by using ‘semantically-based marking’: in these languages, nouns are marked for their semantic role in each individual sentence, rather than their semantic role in the prototypical sentence. For instance, the Tibeto-Burman language Manipuri has three nominal suffixes: (a)
-nə, marking the controller of an action; (b)
-bu, marking an animate affected by an action; (c)
-də /
-ŋondə, marking something indirectly involved in an action. These nominal suffixes apply at the level of individual sentences: Dixon gives the example
I-nə touched Tomba-bu, with the implication that I am in control i.e. I did it intentionally, vs
I-∅ touched Tomba-bu, with the implication that I am not in control i.e. I did it unintentionally. (This sort of duality in meaning will become very familiar later on, when we look at Fluid-S systems.) Languages with semantically-based marking are fascinating, but this topic is tangential to ergativity, so I will refrain from further discussion; for more information, refer to Dixon’s book, cited in the introductory post.)
Morphosyntactic alignment: tripartite, nominative-accusative, and ergative-absolutive languages
A quick recap: In the previous sections, we have established that the core arguments of verbs can be analysed in terms of three ‘primitive’ grammatical relations, these being the sole intransitive argument S, the transitive agent A, and the transitive patient O. I now want to focus on the way in which these relations are marked, called the
morphosyntactic alignment of a language.
The ‘simplest’ morphosyntactic alignment is a
tripartite alignment, where S, A and O are all marked differently. This is the approach taken by the Nez Perce language. Nez Perce has three noun cases: -∅ for S,
-nim for A, and
-ne for O. In an intransitive sentence, there is only one argument, marked with -∅; in a transitive sentence, there are two arguments, one with
-nim and one with
-ne.
However, this approach is quite redundant — if we see a core argument of an intransitive sentence, we automatically know that it must be in S role! So, if we merge the marking of S with A, or S with O, we can only get by with only two different types of marking without any ambiguity. This is the approach taken by the majority of the world’s languages.
If S is merged with A, the resulting alignment is called
nominative-accusative. This alignment has two different ways of marking arguments: one is used for S and A, and the other is used for O. If a language has noun cases, the S/A case is called the
nominative case, while the O case is called the
accusative case. If it isn’t obvious, English is an example of a nominative-accusative language, as shown by its word order: S and A both go before the verb, while O goes after. Insofar as English has noun cases, they are also nominative-accusative: the nominative pronouns
I,
us,
he,
she,
they are used for S and A, while the accusative pronouns
me,
we,
him,
her,
them are used for O.
Perhaps a clearer example of a nominative-accusative language is Latin. To borrow an example from Dixon:
- domin-us
- master-NOM
- veni-t
- come-3s
The master comes
- domin-us
- master-NOM
- serv-um
- slave-ACC
- audi-t
- hear-3s
The master hears the slave
- serv-us
- slave-NOM
- veni-t
- come-3s
The slave comes
Note how the nominative case (here
-us) is used for the S argument of
veni- ‘come’ and the A argument of
audi- ‘hear’, while the accusative case (here
-um) is used only for the O argument of
audi-.
Alternatively, S can be merged with O instead of A. The resulting alignment is called
ergative-absolutive. As with nominative-accusative alignment, there are two different ways of marking arguments, but in ergative-absolutive languages one is used for A, and the other is used for S and O. A noun case which covers only A is called
ergative, while the noun case which marks S and O is called
absolutive. An example of an ergative language is the Australian language Dyirbal; to take examples from Dixon again:
- ŋuma-∅
- father-ABS
- banaga-nʸu
- return-NONFUT
Father returned
- ŋuma-∅
- father-ABS
- yabu-ŋgu
- mother-ERG
- bura-n
- see-NONFUT
Mother saw father
- yabu-∅
- mother-ABS
- banaga-nʸu
- return-NONFUT
Mother returned
Note how the ergative case is used only for the A argument of
banaga ‘return’, while the absolutive case is used for the S argument of
banaga ‘return’ and the O argument of
bura ‘see’.
(Note: Dyirbal also has ‘noun markers’ which agree with each noun in case and noun class, as well as indicating deixis. Dixon notes that in all the Dyirbal examples he gives, he has omitted the noun markers for clarity: the sentences above should really be
bayi ŋuma banaga-nʸu,
bayi ŋuma baŋgun yabu-ŋgu bura-n,
balan yabu banaga-nʸu.)
A further example of an ergative phenomenon can be found in English, where some ambitransitive verbs like
break equate S with O rather than with A:
I break the window, but
the window breaks. This phenomenon is common even in otherwise fully nominative-accusative languages, and has been called an
unaccusative verb.
A third common possibility for alignment is
split ergativity. In split ergative alignments, S is sometimes merged with A and sometimes with O: the language is ‘split’ between nominative-accusative and ergative-absolutive alignments. This often has a fundamentally semantic basis: when the S is more agentive, S=A is used, but when S is less agentive, S=O is used. (Recall the earlier discussion about how the A argument is prototypically ‘agentive’, i.e. in control of the action.) Split ergativity is very common, much more so than ‘normal’ ergativity: in fact, almost all so-called ‘ergative’ languages are actually split ergative. For instance, Dyirbal (mentioned above) is in fact split ergative rather than ergative: pronouns use nominative-accusative alignment, but other nouns use ergative-absolutive alignment. Split ergativity is an important topic and will be covered in its own section.
It may be helpful to see a diagram of the various alignments. Here’s one, from Wikipedia:
This diagram shows how nominative-accusative alignment equates S and A, whereas ergative-absolutive alignment equates S and O. Unfortunately the diagram doesn’t show split ergativity, but you could picture it as having S divided up between A and O.
Other alignments
I would be remiss if I didn’t quickly cover the other possible alignments mentioned in the diagram, as well as some other common ones:
- Direct alignment makes no distinctions whatsoever between S, A and O. This alignment is understandably very rare, but it does occur as part of some more complex split ergative systems.
- Transitive alignment equates A and O, while keeping S separate. You could say that it combines the worst parts of tripartite and direct alignment, by redundantly marking S separately while not distinguishing between A and O. According to zompist, Justin Rye has called this alignment Monster Raving Loony, and I don’t disagree. Transitive alignment is found only in the Iranian language Rushani, and is a product of the fact that Rushani is in the middle of a transition from ergative-absolutive to nominative-accusative alignment; this will be covered in a later chapter.
- Austronesian (or Phillipine-type) alignment is an unusual type of alignment, found in many Austronesian languages, in which every verb takes a ‘trigger’ affix; this affix determines the case-marking of the verb’s arguments.
- Direct-inverse alignment uses an ‘animacy hierarchy’ to infer the roles in which the arguments of a verb are. If the arguments take opposite roles to those expected, a special ‘inverse’ marker is added to the verb. This alignment will be discussed later in the section on split ergativity, along with the animacy hierarchy itself.
- Semantically-based marking (covered earlier), in which each noun is marked for its semantic role rather than S/A/O status, could be considered a type of alignment. Refer to the discussion above for more details.
What is ergativity?
So now we have established what ergative-absolutive alignment is: it’s a method of marking arguments in which S and O are grouped together, and A is marked separately. Now, if that were all there is to the concept of ‘ergativity’, there wouldn’t be much need for a thread like this. However, it turns out that there are many more different ‘ergative’ phenomena which equate S=O and separate A:
- Ergative case-marking, discussed above
- Ergative verbal agreement, in which one set of affixes is used for S=O arguments and another is used for A arguments
- Ergative word order, where S and O have the same location in a sentence but A does not (e.g. a language which is SVO in transitive clauses but VS in intransitive clauses)
- The many varieties of split ergativity, where the split between ergative-absolutive and nominative-accusative depends variously on the volitionality of the verb, the nature of the individual arguments, the tense/aspect of the clause, main vs subordinate clauses, or a combination of the above
- ‘Syntactic ergativity’, in which syntactic operations such as coordination and relativisation treat S and O the same and A differently
- The antipassive, a voice which is the ‘opposite’ of passive and is often found in ergative languages
Now, one might reasonably wonder whether it is even sensible to lump all these disparate phenomena together, based only on the fact that they treat S=O differently from A. This is the position taken by Scott DeLancey in his article
The Blue Bird of Ergativity, in which he argues that these phenomena, while interesting on their own merits, do not deserve to be combined into a single typological category of ‘ergativity’; such a classification, he argues, would be as useless as a group consisting of all blue birds. I would disagree with this: many of the phenomena above co-occur with each other with a frequency which would be unusual if they were separate phenomena (e.g. all languages with syntactic ergativity also show either ergative case-marking or ergative verbal agreement). Additionally, I aim to show in future posts that many of the phenomena above share a common semantic basis.
On the other hand, it is worth noting that the ‘ergativity’ of a language has little to no effect on any aspect of the language outside the list above: DIxon reports that ergativity has little correlation — if any — to whether a language has voice, a distinct class of adjectives, or a large number of ambitransitive/labile verbs; the only correlation he can find is that ergativity may be more common in dependent-marking languages.
(The mention of ambitransitive verbs is worth elaborating on. Some ambitransitive verbs have S=A, for instance
I go there / I go; others have S=O, for instance
I break it / It breaks. Despite the fact that the former verb appears ‘accusative’, whereas the later appears ‘ergative’, the number and type of ambitransitive verbs has no relation to whether a language is mainly accusative or ergative. For instance, Latin is accusative, whereas Dyirbal is ergative, but both have a nearly complete lack ambitransitive verbs; Fijian is not ergative, but almost all verbs in Fijian are ambitransitive, and there is a nearly equal split between S=A and S=O ambitransitive verbs, with only a slight bias (3%) towards S=A.)
The distribution of ergativity
Although not particularly relevant for conlanging, it can sometimes be useful to know where in the world ergative languages occur:
- In Europe, almost all languages are nominative-accusative. The only exceptions are Basque, in western Europe, and the three Caucasian families (Northwest Caucasian, Northeastern Caucasian/Nakh-Dagestanian, South Caucasian/Kartvelian) of northeastern Europe.
- Africa is remarkably devoid of ergative languages, but ergativity has been found in several Western Nilotic languages (spoken in southern Sudan), most notably Päri.
- In Asia, many Tibeto-Burman languages are ergative, and so are many Indic and Iranian languages. So is the language isolate Burushaski. Another cluster of ergative languages can be found in the ancient Middle East: Sumerian, Hurrian, Urartian, Hattic, Elamite and the Anatolian languages all show ergative characteristics. Some Siberian families, namely Chukotko-Kamchatkan and Yukaghir, are also ergative.
- Many Austronesian languages show some ergativity, particularly the Polynesian languages Tongan and Samoan. In Māori the passive voice is apparently used so often that it can appear ergative in some ways, although at the moment it is still very much a nominative-accusative language. Apparently the Tamanic subgroup (on Borneo) and the South Sulawesi subgroup also display some ergativity. Many Phillipine and Formosan languages, while not ergative, show an unusual ‘Austronesian’ morphosyntactic alignment (see above). Acehnese has a form of active-stative split ergativity known as ‘Fluid-S’.
- The greatest concentration of ergative languages is probably in Australia: the great majority of Australian languages show some ergative characteristics, and many have highly ergative grammars. Of particular note is the Dyirbal language, spoken south-west of Cairns (in north-eastern Queensland); we will be seeing a lot more of Dyirbal later, as it is a pretty much perfect example of a highly ergative language. Several Papuan languages, including Enga, Hua, Yimas, Yawa, Koiari, Kaluli, Ku Wara and Kanum, show ‘superficial’ ergative features.
- North America has few highly ergative languages, the main exception being the Eskimo-Aleut family. Tsimshian has also been reported to be ergative. Nez Perce is not ergative but has an unusual ‘tripartite’ alignment (see above); its sibling language Sahaptin has a slightly more complex system. Many North American languages have a form of split ergativity known as active-stative alignment, the usual examples being Lakota and Eastern Pomo; see https://en.wikipedia.org/wiki/Active%E2 ... _languages for a fuller list.
- In Central America, many Mayan languages are highly ergative; Mayan languages often show many interesting features related to ergativity. The Mixe-Zoque family also shows ergativity.
- Many South American language families, including Jê, Arawak, Tupí-Guaraní, Panoan, Tacanan, Chibchan, Maku and Carib, show complex split ergative systems. The isolates Trumai and Jabuti also show ergative features.
One interesting thing to note about this list is that ergativity is often more of an areal feature than a genetic one. For instance, one large cluster of ergative languages exists around the Himalayas, comprising Tibeto-Burman languages, some languages in the Iranian and Indic subgroups of Indo-European, and the isolate Burushaski; another obvious one is in the Caucasus, comprising the three Caucasian families.
playing with alignment is always fun.
Glad you find it helpful!