Re: Core case roles

En réponse à Josh Roth <Fuscian@...>:

>
> I think the problem is that an agent has to be *willing*.
Only if you give an extra-linguistic definition of "agent", which is exactly
what DeLancey's article is fighting against (with good reasons I believe). In
linguistics, everything is defined only in relation with other things: phonemes
are defined only in opposition with each other, not for any immanent quality,
grammatical roles are defined in contrast with each other, even the semantic
content of words is defined only on relation with the semantic contents of
other words. So a correct theory of semantic roles must do the same to even
have a chance of being correct. It won't mean it's necessarily correct, but any
theory that has to come up with extra-linguistic definition which don't fit in
a system of relationships will necessarily be wrong.

Of course, in his theory Agent is something very different from what is usually
viewed as agent. So what? In his article the evidence he gives to unify patient
and theme in one category Theme is quite compelling, so I'm pretty sure we can
unify agent and instrument under one category too. His theory intends to show
things at a very low level (hence its reference to the Figure and Ground theory
of psychology). Segregations like agent and instrument, just like segregations
between theme and patient, occur at a higher level.

Unfortunately the site lacks his lecture about the Agent thematic relation of
his theory. Too bad, his lecture about Theme and Location is pretty insightful,
and I'm pretty sure his lecture about Agent addresses all the issues that have
been raised here.

>
> A stone should work the same way. If it breaks a window, is semantically
> in
> the same role whether someone threw it or not.
Why? Semantic roles are not things that exist independently out there. They
exist only in relationship with each other. Why pose the existence of an agent
when none is present, even in context? You can do this extrapolation if you
want, but it's not something basic, like a Universal Grammar of semantic roles
should be. It's something you add from your experience and your point of view.
You pose the existence of an agent because in your experience a stone doesn't
throw itself through a window. It's completely extra-linguistic and thus
needn't be part of the theory. My example showed that this agent may be quite
hard to define too, so hard that it's quite unnatural to pose its necessary
existence.

I think your problem is not with the concept that you can reduce core semantic
cases to only three, but with the fact that one is called Agent, and you want
to give to it the meaning that *you* defined for it, rather than the meaning
the theory in which it's used gives to it. The problem is that you miss the
point then.

 It is not automatically
> an
> agent just because it's the subject (the two are independent - you can
> change
> word order and phrasing, but the semantic roles stay the same - for
> example,
> "the window broke from the stone").
Only if you define agent the way you do, extra-linguistically. The whole point
of this is that it's wrong to do so.

Kar Marinam would treat the stone as an instrument in
> both
> instances, and a speaker would probably say that there is an agent too,
> that
> really caused the action, though we just don't know or want to say what
> it is
> (maybe a spirit or something).
But all this is extra-linguistic again, and thus shouldn't be part of a
linguistic theory of semantic role.

 Even if a language doesn't have that
> excuse,
> it shouldn't matter - not every sentence has to have an agent.
Very true. But that's another point.

 If I say,
> "I
> sneezed," there's no agent expressed, and is there really a need for
> one?
>
Indeed there's no agent. Not even in DeLancey's theory, since there's nothing
stated which "caused the change of state to happen" (his "simple" definition of
Agent: the external cause of the change of state (or Location, in a very broad
sense) of a Theme). On the other hand, the Theme-Location relationship is the
primary one for core cases and is always present, even when only one of them
seems to be present in the sentence. In that case, it is always possible to
argue that the other one is present in the verb itself (which can be made clear
by making equivalent sentences in which the other part of the relationship
appears independently from the verb again. For instance, in "I sneeze", "I" is
Location - like any experiencer - while the Theme is included in the verb
itself. It gets clear when you rephrase your sentence as "I had a sneeze",
which though not synomymous is equivalent in meaning to the first one).

Of course, DeLancey's theory may be proved wrong. But it's pretty insightful
and people have until now been unable to prove it wrong, except by giving
to "agent" their own definition, which of course doesn't prove anything.

Christophe.

http://rainbow.conlang.free.fr

Take your life as a movie: do not let anybody else play the leading role.