I am not sure that I understand what Dick is saying here. So I'll try to get
this down to what I think are basics. It seems to me that there are SoAs, and
that these may involve participants. In some (pretheoretical, trying to avoid
commitment to a particular position) sense you need (a) to establish those
participants in in the lexical representation, (b) to be able to manage
"grammatical" diatheses like passive as well as "lexical" diathese like
unaccusative, (c) to account for the kind of semantic decomposition that Chet
reminded us of, (d) to handle the fact that semantics is ontologically a
different kind of thing from syntax, (e) get from the lexical entry to an
utterance.
The question is in at least 2 parts. One part, it seems to me, comes down to
what linguistic semantics is. If you are Jackendoff or Culicover then semantics
is conceptual structure. But you cd reasonably adopt a position where you want
to exclude parts of conceptual structure from semantics because language is not
sensitive to that information. So, for example lg, is sensitive to a sg/pl
distinction, or various kinds of temporal distinctions, but not to colour
differences and the colour of a particular noun does not affect its expression.
The second part is a representational one. How do you want to represent semantic
information? This question is contingent on the first one because the more
linguisticaly relevant semantics there is, the less you can reduce it to
annotations on a syntactic representation.
So And wants to say that very little of conceptual structure is linguisticaly
relevant, and that which is linguistically relevant can be limited to
annotations in a syntactic representation (if you treat contentful gramrels as
such a species of annotation). I don't really understand what Dick thinks
because I think I've always thought of er and ee as he describes 1,2,3, so I
need to be walked round some examples showing the differences between what he
thinks er and ee are and 1,2,3 are. The consensus beyond the WG community is
that there is no clear consensus, though some people adopt discrete semantic
representations of some complexity (e.g. Levin and Rappaport Hovav) without
necessarily adopting the Jackendoff position.
Another issue concerns reference. What is it and where does it live? In an And
model is it part of syntax. I've suggested on this list before that reference
could be treated as part of syntax (about 4 or 5 years ago), and actually I
didn't think it's that whacky an idea; I don't know, however, if I can
reconstruct the argument(!)
I think BTW that force dynamic relations do participate in linguistically
relevant facts but are not tied to particular grammatical relations and
therefore need a domain of representation of their own. This is argued for in my
thesis and also in my Language Sciences paper from last year.
Nik.
Quoting Dick Hudson <[log in to unmask]>:
> Thanks, Chet, for bringing us back to the semantics question. This
> question about the status of Er and Ee has been rumbling on in the WG
> group
> for quite some years now and I've vacillated enormously. Like you I'm
> fond
> of Er and Ee, but I don't feel at all sure that they're justifiable.
> By chance Jasper and I have been talking about it just in the
> last few
> days, in connection with a paper I wrote a month or so ago about
> commercial-transaction verbs. (It's on my web site in
> ...dick/papers.htm.)
> I suggested there that we might use the Relational Grammar labels 1, 2,
> 3
> instead of Er and Ee to show what I call 'construal relations' - i.e.
> how
> you 'construe' the situation being described in terms of the kind of
> thing
> Joe has just been talking about: which argument do you focus on first,
> which second and so on. This question may not arise at all outside of
> speaking, so these relations may only exist in 'thinking for speaking'
> where you're moving towards a linear representation of the situation,
> so
> they're strongly oriented towards syntax; and indeed they provide the
> default linkage (1 = ref of subj, 2 = ref of ind.obj, 3 = ref of obj).
> Nevertheless they're purely semantic relations in as much as they're
> relations between elements of meaning rather than words, and (as in
> Chet's
> examples) they can exist without in fact being mapped onto syntax at
> all.
> So what I'm suggesting is that for each SoA (i.e. concept which
> is some
> type of state of affairs) we can identify its arguments in terms of
> some
> system such as force dynamics or gestalt categories, and we can label
> these
> relationships 1, 2, 3 according to their default lexical linkage. E.g.
> (1) Being: figure = 1, ground = ? (E.g. a fly is in my soup)
> (2) Having (which isa Being): figure = 2, ground = 1 (e.g. I have a
> fly).
> In Chet's example of Sinking, the result isa Being whose figure = 1 by
> inheritance from (1), even though it doesn't in this case end up as
> subject
> of anything.
> In short, I think my current position is that although I
> couldn't defend
> Er and Ee as such, I think I could defend 1, 2, 3, and I hope these
> would
> provide roughly the same analytical benefits that Er and Ee used to.
> No doubt others will disagree more or less strongly with this
> view, so I
> look forward to either abandoning this view or refining it.
> Dick
>
>
> At 06:53 07/03/2002 -0500, you wrote:
> >The syntactic side of the discussion has been excellent, but the
> semantic
> >side remains almost untouched:
> >
> >And:
> >> #> (iii) I don't see any place in the grammar for Er and Ee.
> Outside
> >> #> the gramamr, word senses define their own participant role, so
> the=3D20
> >> #> encylopedia entry for Kiss will include definitions of Kisser and
> =3D
> >> #> Kissee.=3D20
> >> #> Inside the grammar, some syntactic categories (such as
> dependencies)
> >> #> have semantic content; dependencies with semantic content are
> >> #> 'theta-roles'. There is no need for additional layers of structure
> to
> >> #> mediate between syntax's theta roles and the interpretative
> interface
> >> #> with pragmatics.
> >
> >Chet:
> >> #Isn't this a bit extreme without explicit consideration of the
> roles
> >> #Er and Ee play in the semantics -- e.g. as anchor points for
> referent
> >> #arcs?
> >
> >And:
> >> An interesting question, but can you give me specific examples?
> >
> >I confess that I was not prepared for And's proposal that Er and Ee
> be
> >dispensed with. There are now two questions:
> >
> >I. Can er's and ee's be eliminated entirely from the grammar
> >
> >II. If they are two be retained is there any motivation for giving
> them
> >any semantic content of the sort implied by their names?
> >
> >W.r.t. I., remember that one of the features of WG's severe syntactic
> >component is that no indices are allowed there. One role played by
> >Er/Ee (or whatever other label(s) are proposed to denote arguments in
> >the semantics) is to carry referent indices. From the point of view
> of
> >the consideration of this point alone, er's and ee's could be
> replaced
> >by indices alone.
> >
> >W.r.t. II. Even if WG has a kind of default projection principle such
> >that each valent in the syntax has an associated argument in the
> >semantics, there are still the cases which Dick has discussed under
> the
> >rubric of raising (and Jasper in his dissertation) in the semantics
> >where there are additional arguments in the the semantics which are
> not
> >in the syntax. Thus in the semantics of (1) there are a movement
> process
> >and a result process, both of which have first-arguments, i.e. Er's.
> >
> >(1) The submarine sank.
> >
> >In the semantics of (2) there is also a result process such that the
> >location of the swallow-ee is in the stomach of the swallow-er.
> >
> >(2) Marina swallowed.
> >
> >Note also that although both (1) and (2) have processes which are not
> >present in the semantics, the first-argument of the two processes in
> >(1) is the same, whereas that of the eating process in (2) is not
> >the same as that of the result process in (2). It is evidence of
> this
> >type which presumably led Dick to use the mnemonic (?) labels Er and
> Ee.
> >
> >My original semantic question therefore was, do these Spanish data
> >provide any additional evidence for argument labels with semantic
> >content? My tentative answer is no and that this kind of information
> >is already provided in the lexical class information inherited by
> psych
> >and unaccusative verbs [And's discussion above refers only to
> >individual lexical entries, so perhaps the question of the role of
> >semantic classes in the grammar should also be discussed]. The same
> >would be true for the semantic facts implied by (1) and (2).
> >
> >I confess to a certain fondness for Er and Ee, and have found that
> >their pedagogical utility is immense. It would seem, however, that
> >from a consideration of simplicity, they can be dispensed with.
> >
> >Chet
> >
> >
>
> Richard (= Dick) Hudson
>
> Phonetics and Linguistics, University College London,
> Gower Street, London WC1E 6BT.
> +44(0)20 7679 3152; fax +44(0)20 7383 4108;
> http://www.phon.ucl.ac.uk/home/dick/home.htm
>
|