accusative and ergative languages

[Larry Trask]

On Tue, 22 Jun 1999, Patrick C. Ryan wrote:

[PCR]

>>> I am almost sure that you will want to refine my understanding of
>>> the word "simplicity" but for whatever it may be worth, I would
>>> characterize an isolating language as simpler than a flectional one.

[LT]

>> If this were true, then a good reference grammar of Chinese or
>> Vietnamese ought to be shorter than a good reference grammar of Russian
>> or Latin.  And this does not appear to be the case.

> Actually, I think it may well be the case for Chinese. I have a copy
> of Chao's _Mandarin Primer_ (336 pages), which has a Chapter III -
> Grammar, beginning on page 33, and *ending* on page 59. I can
> compare this to Forbes' _Russian Grammar_, which contains 436 pages
> of grammar and indices.

It is hardly reasonable to adduce one small and elderly primer of
Chinese and compare it with a large reference grammar of Russian.
In any case, I have seen at least one very large reference grammar of
Mandarin, though I can't recall who wrote it.

One way of appreciating the impossibility of a language with a "simple"
grammar is to browse through the Comrie-Smith questionnaire, or through
one of the grammars based on it.  This shows quickly that every language
has a host of grammatical functions to provide grammatical forms for.
A few examples of the kinds of thing that every language must provide
grammatical constructions for:

	"That house is made of bricks."
	"The gum is stuck to the bottom of the chair."
	"These shoes are too big for me."
	"It's obvious that she's drunk."
	"The more he eats, the fatter he gets."
	"Two of the boys have long legs."
	"The train was not as late as last time."
	"The woman I was talking to is always complaining."
	"I forgot how to open the lock."

And so on, and so on.  Whether a given language, or does not, use
inflections to do some of this work has little effect on the overall
grammar: all of these zillions of things have to be expressed somehow,
and grammatical devices must exist to provide for them.  And, of course,
a learner of the language must learn every one of those devices and
learn when to use it.

> Even if the earliest morphemes of language werenot recoverable as
> you would maintain, logic tells us that monosyllables would have, at
> least, predominated.

I'm afraid that "logic" tells us no such thing.  This is no more than a
wild guess.

> The syntax of these monosyllables would have had to convey whatever
> grammar the language had at that point; and this would certainly
> resemble languages which are currently termed "isolating".

An isolating language is indeed isolating: no dispute there.  But even a
monosyllabic language need not lack inflections altogether: there exist
languages with internal inflection, as in English `sing', `sang',
`sung'.

> I prefer (though, of course, you and many others may not) to
> distinguish the terms: "isolating", referring to a language for
> which we can reconstuct no flectional or aggltuinating stage; and
> "analytic", for a language we can.

Bizarre.  What is the point of trying to classify languages on the basis
of what we can reconstruct for their ancestors?

> Of course we have no data on the earliest human language in the same
> way we have no data on IE but, by analysis, we can reasonably form
> an opinion as to what IE must have been like. In the same way, we
> can analyze current linguistic data, and form an opinion as to what
> earliest language must have been like though, I admit, the process
> is much more doubtful.

Hardly comparable.  With PIE, we are reconstructing only 2000-3000 years
earlier than our earliest substantial texts.  Trying to reconstruct
50-150,000 years back is a whole nother ballgame.

> I do believe that the term 'ergative language' has a real value in
> linguistic analysis --- to differentiate two basic approaches to
> transitive constructions:

> Language A: Noun(A)+erg. noun(B)+abs. verb
> will be interpreted as A performs an activity on B.

> Language B: Noun(A)+nom. noun(B)+acc. verb
> will be interpreted as A performs an activity on B.

> However, in Language A, noun(B)+abs. verb will be interpreted as an
> activity is performed by an unspecified agent on B

No, not at all.  The facts vary from language to language.  In Basque,
for example, the absolutive is interpreted as the performer of the
action if the verb is intransitive, as the patient if the verb is
transitive.  Same appears to be true of Dyirbal.

> whereas in Language B: noun(B)+acc. verb is *ungrammatical*.

I take it you've never come across Spanish, Italian, Russian or Latin.

Spanish: <Ha visto la pelicula> `S/he has seen the film.'
Latin: <Caesarem vidit> `S/he saw Caesar.'

Accusative noun, verb, no overt subject, perfectly normal.

> I would characterize Language A as (at least, essentially "ergative").

But which languages are like this?

[LT]

>> I know no Sumerian.  But, speaking generally, the `passive'
>> interpretation of ergativity, once so popular, was discredited years
>> ago.

[PAR]

> Perhaps to your satisfaction but not to mine.

Not much of an answer, I'm afraid.  The "passive" interpretation of
ergativity was based squarely on morphology, with no attention to syntax
-- even though a passive is a syntactic form.  More particularly, it was
based on the confused and erroneous notion that a grammatical subject
must always stand in the same case.

For Basque, and for other ergative languages, the "passive" view of
transitive sentences can be shredded, point by devastating point.

[LT]

>> In very many ergative languages, it is trivial to demonstrate that
>> transitive sentences are active, not passive.  I myself have done this
>> for Basque.

> Perhaps you have done this for Basque but it is certainly not
> trivial to demonstrate this for Sumerian, where Thomsen
> characterizes: "The Sumerian verbal root is in principle neither
> transitive nor intransitive, but neutral in this respect".

I don't know any Sumerian.  But what is true of Sumerian is not
necessarily true of any other language.

[PAR]

>>> I assert that the relationship is closer between an object and
>>> transitive verb in an ergative language because the relationship
>>> (frequently OV) and is immediate and direct. The ergative agent,
>>> IMHO, can best be regarded as a *missible* adverbial adjunct to the
>>> verbal phrase, consisting of the Obj + V, having no more importance
>>> to the verb than an adverbial phrase denoting the target or manner
>>> of an intransitive verb of motion.

[LT]

>> This may or may not be so in some ergative languages, but it is
>> certainly not true for ergative languages generally.

> Is there an 'ergative language', presuming by your definition some
> such exists, where this is not true?

Sure, taking `ergative language' to mean `language in which ergativity
is prominent'.

In Basque, for example, intransitive subjects (absolutive), transitive
subjects (ergative) and direct objects (absolutive) are all equally
optional:

	Gizonak mutila jo zuen.
	man-the-Erg boy-the-Abs hit Aux
	`The man hit the boy.'

	Gizonak jo zuen.
	`The man hit him.'

	Mutila jo zuen.
	`He hit the boy.'

	Jo zuen.
	`He hit him.'

All perfectly normal in context.

[LT]

>>>> Unsubstantiated assertion.  You might, with equal justification, assert
>>>> that accusativity must precede ergativity in all cases.  If anything, it
>>>> is this last statement which is better supported by the evidence.

> Pat asks:
> What evidence might that be?

The fact that we know of a few cases -- Indo-Iranian is one -- in which
ergativity has arisen in languages which formerly lacked it, while we
have hardly any examples of languages in which accusativity has arisen
in languages which formerly lacked it.

[on my observation that children acquiring English do not go through an
ergative stage]

> I am sure you are better read on child language acquisition patterns
> than I. Based on what I have observed personally, I doubt your
> assertion but if studies have shown this (could you name one?), how
> can I dispute it.

We now have a vast body of data on children acquiring English.  And I
know of no study, not one, which recognizes an ergative stage during
acquisition.

> A sequence like: "Me hurt. Tommy did it" is a virtual ergative so
> far as I would judge. I have heard children speak in this way.

Yes, sure, but a single datum proves nothing.  Children at the two-word
stage also say things like `Mommy get', meaning `[I want] Mommy to get
the ball.'  This should be impossible in an "ergative" view.

Larry Trask
COGS
University of Sussex
Brighton BN1 9QH
UK

larryt at cogs.susx.ac.uk