----------
> From: Dick Hudson
> To: [log in to unmask]
> Subject: RE: er and ee
> Date: 01 February 2000 19:38
>
> And on me:
> >> ## There's another way to look at er and ee which doesn't require them
to
> >> have any common defining semantic features ('content') at all. Suppose
we
> >> have a super-category, Sao (State of affairs, as in HPSG).
> >
> >[psoa (propositional s. o. a.)]
> ## Woops. Where did sao come from? Oh well ....
stamped addressed onvelope? standard average oropean?
> >> That interpretation means that there's no reason why ees should have
> >> anything in common across all Saos (though they should have common
> >features
> >> across some subset of Saos). It also allows the choice of er and ee to
be
> >> to some extent arbitrary when the typical single-argument
characteristics
> >> are split between the arguments, as with verbs like LIKE, PLEASE and
> >> RECEIVE.
> >
> >This on its own is not linguistically relevant; it needs to be tied to
> >an account of linking, say.
> ## I don't see how it can fail to be linguistically relevant;
a cognitive category is linguistically relevant only if it is the sense
of some linguistic entity (or involved in some other kind of semantic
rule). You seemed to be debating how to cognitively model soa arguments,
without showing a correspondence to linguistic categories.
> but it's
> really a question of how we organise such things cognitively. If we do
have
> a hierarchical classifcation of PSOAs and if inheritance applies, then
> different characteristics will automatically be inherited by the er (the
> only argument of a single-argument PSOA) according to which PSOA type it's
> inherited from.
I agree. But I think that it's not an issue that matters for language.
Also, I have a problem with this Er category that generalizes over the
only arg of 1-arg soas and one of the args of many-arg soas.
>
> It's not even obvious that this model would
> >adequately predict 1st arg linking to subj and 2nd arg to obj, for
> >transitives.
> ## Of course not - that's why I say below that you need different linking
> rules for different kinds of languages; e.g. 2nd arg linking to subject in
> an ergative language.
>
> >
> >And it is unclear why you identify the argument of single argument
> >soas with one of the arguments of multiple argument soas.
> ## Because it will happen automatically by inheritance if single-argument
> SOAS are the default: they will have just a single argument, which we're
> calling er. So if Action (i.e. Do-ing) has just an er, and Kicking isa
> Doing, then Kicking will have an er too.
First, it does not seem to me at all self evident that the subj-expressed
arg relation of a 2-arg soa isa arg relation of a 1-arg soa. In other words:
monotrans subjects have on average at least as much in common with patients
as they do with agents.
Second, it is also not obvious that that arguments should automatically
inherit when a 2-arg soa isa a 1-arg soa. It seems perfectly plausible
that the 2-arghood could override the inherited 1-arghood, so that
neither of the two arg relations would necessarily isa the overridden
arg relation. It might happen that one of the two arg relations might
have properties in common with the one overridden relation, in which case
they could be subtypes of the same category (e.g. Agent). I realize that
it's hard to model in WG what I'm suggesting, but I think that that's
a problem with WG more than with my suggestion.
Look at it this way. 'er' for 1-arg soas could simply be renamed 'Argument'.
Then for multiple-arg soas, **EACH** of their argument relations could
isa Argument. Hence properties of Argument would inherit to every Argument.
This to me looks not only equally but also more plausible than what you're
suggesting. For 2 arg soas you would set up subtypes of Argument, such as
Agent and Patient.
> >> This is where I think Dowty comes in.
> >
> >I don't see what you're getting at. Dowty tells you how to map from
> >a group of arguments to Subject and Object. As far as I can see, for
> >D's scheme (of which I am a fan) to work all that matters is the
arguments'
> >content. To the extent that er/ee lack content, they would be
linguistically
> >irrelevant.
> ## Fair point - Dowty's system is much too rigid, in my opinion, because
it
> doesn't generalise easily to PSOAs that aren't actions.
Ahhh -- I think I see what you're trying to achieve. You're trying to get
a generalizable account of linking that doesn't break down on the hard
cases (like LIKE and PRECEDE etc.)?
Currently it looks to me simply as though you have a stipulative account
of linking, disguised by being shifted from the lexicon to conceptual
structure.
> >I'm probably failing to grasp your point, but if er/ee don't have
constant
> >content then they must be identified by a separate stipulation for each
> >soa. The grounds for identifying er vs ee, if they have no constant
content,
> >is which links to subj and which to obj, and hence your two Linking Rules
> >are tautologously vacuous.
> ## Yes, I think you're missing the significance of the isa hierarchy. It
> has SOA at the top, and maybe SOA has an er, but the er has virtually no
> content - i.e. nothing is inherited from that level. But under SOA you
have
> States, Actions and so on, each of which adds more to the definition of
er;
> and below them you have further subtypes (e.g. Going, Feeling, whatever),
> and so on till you get to the most specific SOAs. Each level in the isa
> hierarchy adds a different set of characteristics to what is inherited
from
> higher up and of course, depending on which branch of the tree you go
down,
> you inherit a very different set of characteristics.
It's not clear to me why the subject-expressed arguments of BUY, SELL, LIKE,
PRECEDE, UNDERGO, etc. are all ers, though. That is (a) I'm wondering
why they're not, say, ees, and why it's not the case that some other
arg is er, and (b) I'm wondering about the validity of inheriting 'er'
to multiple-arg soas, as I said above.
Your proposals rely on there being a robust cognitive case for Er
independently of linking facts, and I don't think you've made it.
--And.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|