A rare phenomenon in
language, investigated by
Bob Dixon in the Queensland language
Dyirbal. There are two common ways of marking
subject and
object, an
accusative system or an
ergative system. Quite a few languages have ergative
marking, but almost none extend the
ergativity throughout the
syntax, but Dyirbal is one that does. Psychologically it seems to be quite unnatural.
I don't know whether I'll be able to find actual examples of Dyirbal on the Web, and have no other reference for it, so for now I'm going to explain the difference schematically. Later: Found a real example. I'll add the actual Dyirbal after my explanation.
Almost all European languages work on this pattern:
(1) The man-SUBJ sleeps.
(2) The man-SUBJ kicks the dog-OBJ.
In Latin or German or Greek the SUBJ is in a case called the nominative in both (1) and (2), and the OBJ which only occurs in (2) is in a case called the accusative. This is reflected in the endings they are marked with.
An ergative language has a different pattern. The sole actor in (1) is aligned with the OBJ in (2), not with the SUBJ:
(3) The man-ABS sleeps.
(4) The man-ERG kicks the dog-ABS.
The ABS stands for absolutive. It is not the order of words which is important in this pattern, it is that the sole actor in a one-actor sentence (actually the correct term is actant) is marked the same as the thing that is suffering the action in (2) or (4). The difference between accusative and ergative patterning is one of marking.
The above ergative pattern is common enough across the world, though it does not occur in any familiar Western language. What is very rare is to take it to its logical conclusion in all syntactic respects.
Consider relative clauses.
(5) I see the man. The man sleeps.
--> I see the man who sleeps.
(6) I see the man. The man kicks the dog.
--> I see the man who kicks the dog.
In these, 'the man' is object of 'see' but subject of both 'sleeps' and 'kicks'. The pattern is this:
(5') I see the man-OBJ. The man-SUBJ sleeps.
--> I see the man-OBJ who-SUBJ sleeps.
(6') I see the man-OBJ. The man-SUBJ kicks the dog.
--> I see the man-OBJ who-SUBJ kicks the dog.
When two sentences have something in common you can leave one of the elements out and join them by 'and':
(7) The man kicks the dog. The man sleeps.
--> The man kicks the dog and sleeps.
(8) I see the man who kicks the dog and sleeps.
What is in common is not just 'the man' but 'the man-SUBJ'. You can't omit 'the man' if one is SUBJ and one is OBJ. This is true in English, even though we don't have overt markers of SUBJ and OBJ the way Latin does.
(9) The woman punches the man. The man sleeps.
--> *The woman punches the man and sleeps?
--> *I see the man who(m) the woman punches and sleeps?
Clearly that doesn't work. You can't just leave out the second 'the man' entirely. You have to keep in 'he' or a second 'who', or turn the first sentence into the passive.
Now consider how this would work in an ergative language.
(10) The woman-ERG punches the man-ABS. The man-ABS sleeps.
--> The woman-ERG punches the man-ABS and sleeps.
--> I see the man-ABS who the woman punches and sleeps.
This ought to work in an ergative language, because it's 'the man-ABS' in common. But in fact it never does. Almost all ergative languages treat this construction with accusative syntax, as in English, even though the nouns are marked with an ergative pattern.
Dyirbal is remarkable because it really does use a pure ergative syntax, and does allow relative clauses to be conjoined in that seemingly unnatural way. That's what we call syntactic ergativity.
Here is an example. 'The man' is bayi yara and 'the woman' is balan dyugumbil in the absolutive. (I explain the noun class system, and why 'the' has different forms, in the node for Dyirbal.) So bayi yara baninyu 'the man came', where the man is absolutive subject. In banggun dyugumbiryu bayi yara balgan 'the woman hit the man' he is absolutive object, and the woman is ergative. The same thing can also be said in the order bayi yara banggun dyugumbiryu balgan. The man, as object, can be coordinated with his role as subject of an intransitive verb: bayi yara baninyu banggun dyugumbiryu balgan 'the man came and the woman hit (him)', or 'the man came and was hit by the woman', though in the Dyirbal sentence there is no passive voice: both parts are active.
The learned Gone Jackal has caused bits of my brain to fly out of the back of my skull emitting coloured sparks by pointing out that Sumerian was syntactically ergative in its early days; and lost this by contact with the Semitic (and therefore accusative) language Akkadian.