Quoting Mangiat <[log in to unmask]>:

> Christophe wrote:
>
> > If this seems unclear to you, that's because you're reasoning with the=20
> > concepts of "subject" and "object", which fit only accusative languages=20
> > (and then, only those that neatly use the nominative for subjects and
> the=20
> > accusative for objects). If you try to apply them to ergative languages,
> it=
> > =20
> > becomes automatically a mess (unless you treat the term in the
> absolutive=20
> > as "subject", which is a valid way to think about it, but then you have
> to=
> > =20
> > remember that the subject of transitive sentences is the *patient*
> rather=20
> > than the agent). So what you need to try is to stop thinking in these=20
> > terms, and only think in terms of syntactic cases.
>
> So, is it possible to have a coherent definition of subject and object which
> fits both accusative and ergative languages? One of my university text books
> (an introduction to linguistics) said it is the argument the verb agrees
> with- a definition which works fairly well with Italian, but... what about
> Swedish, for instance, where the verb does not inflect for person and
> number? And what about many Northern Italian dialects, where verbs agree in
> number with a 3rd person subject only when they follow it (when the verb
> precede  the subject it takes 3 sg agreement, even when the subject is
> plural)?

I think the notion of "subject" makes very limited sense outside accusative
systems. "Object", I guess, might reasonably be defined as the (normally) more
patient-like argument of a transitive.

However, if it's true that all ergative languages have non-insignificant
traces of accusativity, then of course the accusative notion of subject should
be relevant wrt to (parts of) all ergative languages.

                                                        Andreas

PS If I've got it right, all ergative languages are really split-ergative, but
plenty of accusative languages have only the faintest traces of ergativity
(like the -ee suffix in English). If so, it would seem to suggest that
accusativity, for some reason, is the "default" for human language, yes? What
about active, tripartite, clairvoyant and MRL languages - do they too more or
less universally contain bits of accusativity?