Theiling Online    Sitemap    Conlang Mailing List HQ   

THEORY: Connolly: Interpreting ergative sentences

From:Lars Henrik Mathiesen <thorinn@...>
Date:Friday, July 16, 1999, 20:19

This is a little essay that I found on the Indoeuropean list. It's all
about the relations between case roles, syntactic subjects,
morphological subjects and ergative agreement systems, all of which
seem to be subjects of some interest to this list.

It is written in response to a claim by another poster there (the Pat
mentioned in the text --- yes, that Pat, Patrick Ryan of ProtoLanguage
fame) that not only are ergative constructions often diachronically
related to earlier passive ones, they are _inherently_ passive.

This is especially recommended if you think Rick Morneau's syntax
theory is the final word on the subject.

Lars Mathiesen (U of Copenhagen CS Dep) <thorinn@...> (Humour NOT marked)
Content-Type: message/rfc822

Return-Path: <indoeuropean-request@...>
Received: from ( [])
        by (8.9.0/8.9.0) with SMTP id DAA03406
        for <thorinn@...>; Wed, 14 Jul 1999 03:07:04 +0200 (METDST)
Received: from by with TCP; Sun, 11 Jul 1999 22:03:00 -0700 (PDT)
Received: from LATTE.MEMPHIS.EDU by LATTE.MEMPHIS.EDU (PMDF V5.1-12 #D3067)
 id <01JDGGHLJ8409WZFEY@...> for; Mon,
 12 Jul 1999 00:01:00 CDT
Date: Mon, 12 Jul 1999 00:01:00 -0500 (CDT)
Subject: Interpreting ergative sentences
Message-id: <01JDGGHLJA029WZFEY@...>
MIME-version: 1.0
Sender: Indo-European mailing list <Indo-European@...>

Dear Pat and everyone else,

I've been reading the recent (well, actually not so recent) discussion
of ergativity with more than a little interest, since I've been working
on the problem off and on for years -- strictly from the synchronic
point of view.  Please bear with me -- this will be long, since I have
to do some theoretical stuff to work through, some of which will be old
hat to many among us.

Much of the discussion has centered around the interpretation of
ergative sentences in Sumerian, of all things, which only Pat seems to
know at all.  (I sure don't.)  But I think it would be worth while to
consider various phenomena in modern languages, some of them ergative,
others not.

First, what is a "subject"?  Pat and the others have been talking past
each other on this, to a considerable extent.  The reason is simple:
there's great disagreement on what it means.

Dixon, in his work on ergativity, has consistently equated "subject"
with "morphological subject".  That is, if a language has a rule that
it must (usually) have one noun or pronoun which gets some specified
treatment not accorded to other items in the sentence, that item is the
morphological subject.  If there is case, it will be called
"nominative" in "accusative" languages, but (usually, but IMHO
unnecessarily) "absolutive" in "ergative" languages. -- Not all
languages have such a rule.  Those that do not are called "active"
languages, where casemarking (or equivalent) is not uniform for
one-argument sentences.

Pat definitely agrees with Dixon here, and so (but for very different
reasons) do I.  Most modern syntacticians do not: for them, the subject
is the noun phrase (NP) which, in an accusative language, would (a) be
the morphological subject and (b) have the *syntactic* privileges of a
subject.  That is, it would occupy "subject position", if the language
has such a thing, or it might be the only item to which a reflexive
pronoun might refer, or the only one which might be replaced by a
relative pronoun -- this is all highly language-specific, of course.

The catch is: for some verbs in (as far as I can tell) all languages,
and for many or most transitive verbs in ergative languages, the thing
that "ought" to be the subject might not be.  Then some funny things
can happen, which again are language specific.  But first: how can we
tell which NP "ought" to be the subject?

"Case Grammar" of the sort first proposed by Charles Fillmore is
helpful here.  It assumes that various NPs have semantico-syntactic
relations to verbs, which he foolishly called "deep cases".  There is a
hierarchy of these cases, which I believe is as follows:

        1.  Agent
        2.  Experiencer; beneficiary (including owner)
        3.  Instrument; effector
        4.  Patient (aka "theme")
        5.  Oblique cases (including place, time, goal, and a few

In an accusative language, the NP which ranks highest in this hierarchy
is normally made the morphological subject.  In addition, it has the
syntactic privileges associated with it.  For convenience, I call this
one "HRNP", which stands for "highest-ranking NP".

But not all verbs in these languages actually "select" (to use
Fillmore's term) the normal subject.  Repeatedly we see that verbs with
experiencers or beneficiaries instead of agents may make the patient
the morphological subject.  It then often turns out that while the
morphogical subject (patient) has some syntactic privileges, the HRNP
has others.  Some examples:

English _own_ and _belong_ have Beneficiary and Patient as arguments.
_Own_ selects the Beneficiary as subject, but _belong_ has the Patient.
The result: sentences with _own_ can be made passive, but those with
_belong_ cannot.

        The Rockefellers own that railroad. ==>
        That railroad is owned by the Rockefellers.

        That railroad belongs to the Rockefellers.
        *The Rockefellers are belonged to by that railroad.

Why no passive for _belong_?  I suspect the reason is that
passivization serves to demote the "correct" subject (HRNP) and use the
"wrong" one instead -- but _belong_ already has the wrong subject!

German _helfen_ 'help' and _schmecken_ 'taste (good)' both have dative
objects.  Normally, German nominative nouns must precede dative nouns
when the NPs are contiguous:

        Will der Professor dem Studenten helfen?
             Agent         Beneficiary
        'Does the professor wish to help the student?'

        *Will dem Studenten der Professor helfen?

But watch what happens with _schmecken_ and many other verbs that
select Patient as subject over the higher-ranking Experiencer:

        ?Hat der Kaffee dem Professor geschmeckt?
               Patient  Experiencer
        'Did the coffee taste good to the professor?'

        Hat dem Professor der Kaffee geschmeckt?
        'Did the professor like the coffee?'

The "normal" subject+object version of this sentence, while surely
acceptable, is simply not as good as the "abnormal" object+subject
version.  But why?   The reason is simply that the "abnormal" version
puts the HRNP ahead of the lower-ranking Patient-subject: the NP that
"ought" to be the subject is put where the subject "ought" to be!  And
it's no accident that the English gloss which puts _professor_ ahead of
_coffee_ sounds better than the other -- for exactly the same reason.

Accusative languages very greatly in which syntactic provileges are
attached to the morphological subject and which to the HRNP.  It may
well be true that in accusative languages, if any NP is marked on the
verb, the morphological subject must be; but otherwise it's a very
mixed bag.  In Icelandic, the HRNP seems to have *all* the syntactic
"subject" properties, regardless of what case it's in.  In English, the
morphological subject is king (except re. passivization).

A similar situation obtains in ergative constructions.  In Dyirbal, the
HRNP is has no syntactic privileges whatsoever; everything depends on
the morphological subject.  But most ergative languages do exactly the
opposite: the HRNP has all the syntactic properties, and in some it is
even the only one which controls verb agreement.  The Patient is then
still the "morphological subject" in that it's in the absolutive case,
but that's all.

That being said, we must ask what is the most appropriate way to
translate a "normal" transitive sentence in a "typical" ergative
language?  Larry Trask has already pointed out that Basque (like many
other ergative languages) has a passive formation.  But since the
Patient is already the morphological subject, what does the passive
accomplish?  It simply demotes the HRNP, so that it no longer has any
syntactic "subject" properties and must (in many other ergative
languages: may) be omitted.

Many also have an "antipassive" formation.  This makes the HRNP the
morphological subject, demoting the Patient.  The HRNP now has all the
syntactic subject properties, as it typically does in typical sentences
in accusative languages.  Different languages have different reasons
for doing this; Dyirbal uses it mainly as a technique for stringing
sentences together that can share a single absolutive.

A "normal" ergative sentence -- no passive or antipassive, and agent NP
present in the ergative case -- corresponds most closely to the English
active voice; the agent should then be the English subject.  A passive
translation would be inappropriate, since passivization downplays the
agent and makes it syntactically irrelevant, making the translation
quite different in tone (and content) than the original.  But a passive
sentence in an ergative language might best be translated with an
English passive.

This is actually quite like what often happens with experiencer and
beneficiary verbs in accusative languages.  The fact that e.g. Spanish
makes the patient the morphological and syntactic subject of a verb
does not mean that an English translator must do the same.  _No me
gusta la mu'sica_ does not mean 'the music doesn't please me' -- that
translation suggests that someone was trying to please (or perhaps
annoy) me with the music.  Rather, it means, very precisely, 'I don't
like the music': the Spanish doesn't insinuate that anyone was thinking
of me, so neither should the translation.

Similarly, I had no hesitation above in glossing _Hat dem Professor der
Kaffee geschmeckt?_ as 'Did the professor like the coffee?', even
though (a) the two languages have different morphological subjects, and
(b) _like_ is not an exact translation of _schmecken_, which always
means 'taste'. I also changed the tenses: present perfect in German,
past in English, since each is the normal tense for past time in the
respective language.  And this is legitimate, since in tone and meaning
it most closely approximates the German original.  Another example:

        Der Wagen ist nicht mehr zu reparieren.
        the car   is  not   more to repair

A good translation would be 'The car can't be fixed any more' -- or
'The car is beyond repair' -- or 'They can't fix the car this time'
etc.  They're all pretty good, but they're not literal renderings.

I could have said instead: 'The car is no longer to be repaired.'  That
seems closer -- except that I will have used a *passive* infinitive
instead of the German active.  And let's face it: it sounds stilted,
while the original doesn't.

I can take these "liberties" because German is a modern language that I
know well.  There are very few who could do this for a dead language,
even one as well known and thoroughly studied as Latin or Greek.  And
that's the trap: because we don't know the nuances, we want to be as
literal as possible, in the hope that a literal interpretation might be
correct.  Sumerian is even worse, since we don't have all that much of
it (no poetry, no menus, nothing that seems to be finely nuanced
speech.  And on top of that, it's ergative, and most of us (including
me) are *not* used to dealing with such languages.

So how should we render Sumerian sentences of the type

        verb = 'distributed'
        NP absolutive = 'camels'
        NP dative = 'heirs'

(Sorry I didn't cite the actual sentence, but this time it really is
the thought that counts!)  Not knowing Sumerian nuances, we can only
draw on seemingly similar sentences in other ergative languages.
Despite what Pat says, it is far from clear that the English must have
a passive of the type 'the camels were distributed to the heirs by some
unknown agent'.  That is a plausible attempt at a rendering -- but
since we do not know the effect of omitting the agent, and there seems
to be no hint of a passive transformation being done, there's no reason
to think it's "better" or "more precise" than any of the following:

        The heirs shared the camels.
        The heirs divided the camels aong them.
        The camels were shared by the heirs.
        "They" divided the camels among the heirs.
        All the heirs got some camels.
        The camels (jewels, books etc.) went to the heirs.

We don't know, because we don't know Sumerian the way we know English
and other modern languages.

So please, caution!  And take due note of what has been discovered
about *living* ergative languages.  You can't hope to understand the
dead ones without them.


Leo A. Connolly                         Foreign Languages & Literatures              University of Memphis