On Mon, Feb 02, 2004 at 11:55:33AM -0800, Philippe Caquant wrote:
> This is semantic. What makes us confused is that we
> are conditionned by the grammar we learned at school.
> We think in terms of subject, and direct object, and
> indirect object, and transitivity, and passive, and so
> on.

I believe the original post makes a point that verb arguments can be
elided if they are clear from context. So to examine the "true" valency of
a verb, one would have to consider it in isolation, eg. at the beginning
of a conversation where there is no prior context.  In that sense, one may
infer that certain verbs have certain (semantic) valencies that the
grammar of a language would have to account for. Intransitive verbs are
those that would be semantically complete with only one argument;
transitive verbs are those that need two arguments to be semantically
complete. (And IMHO, there are the trivalent verbs like "give" which
strictly speaking require at least three arguments.)

My point then, was that the arguments of verbs with different valencies
may not necessarily map to each other in an obvious way (or at all). For
example, consider three verbs of differing valencies:
a)      sit:    (1) the person/thing which sits
b)      love:   (1) the lover (2) the object of love
c)      give:   (1) the giver (2) the thing given (3) the recipient

Nothing says that (a)(1) is the same as (a)(2), or that (b)(2) is the same
as (c)(2). It could easily be the case, in a particular language, that
(a)(1) corresponds with (b)(2), and (b)(2) corresponds with (c)(3).

One may say that the case system of any language can be represented by a
mapping of each of (a)(1), (b)(1), (b)(2), ... (c)(3) to some set of 3 or
more semantic roles. For example, in English (as well as most IE
languages, I suspect) the semantic roles are S(ubject), O(bject), and
(I)ndirect object. (You can label them arbitrarily; the point is that
there are three distinct roles.) The mapping for the above three verbs are
        (a)(1) -> S
        (b)(1) -> S     (b)(2) -> O
        (c)(1) -> S     (c)(2) -> O     (c)(3) -> I

For another language, it is quite possible that the mappings for trivalent
verbs is instead:
        (c)(1) -> S     (c)(2) -> I     (c)(3) -> O

Furthermore, not all languages necessarily share this same set of semantic
roles; ergative languages can be said to have a 4-element set of roles,
A(bsolutive), E(rgative), P(atient), I(ndirect object), and have the
following mappings:
        (a)(1) -> A
        (b)(1) -> E     (b)(2) -> P
        (c)(1) -> E     (c)(2) -> P     (c)(3) -> I

There may also be different classes of verbs with the same valencies, with
different mappings for each class. For example, in some (hypothetical)
language, the two arguments for "love" may not be the same as the two
arguments for "kick":
        love:   (b)(1) -> A     (b)(2) -> A
        kick:   (b)(1) -> A     (b)(2) -> O

where the repeated use of A in "love" implies mutuality, whereas "kick" is
a directed action and so uses a different role O for its second argument.

Ebisedian's seemingly strange case system is actually just a simple
application of this model: it dispenses with the requirement that one of
the semantic roles (eg. the subject in accusative langs) is mandatory, and
uses a 5-element semantic role set: originative (source), receptive
(destination), instrumental (means), conveyant (that which was affected),
locative (locus). (Arguably, the locative is a secondary role and needn't
be considered here.) The Ebisedian mappings for "kick", "see", and "give",
for example, are:

        kick: kicker -> O       person kicked -> R
        see:  seer -> R         thing seen -> O
        give: giver -> O        thing given -> C        recipient -> R

The mappings for "see" may seem odd, but only because the accusative
mindset is used to thinking about agents and patients. In Ebisedian,
agents and patients are irrelevant; the semantically important concepts
are source and destination: from whence did the event originate, and to
what is it directed at, rather than who/what did it, and who/what
underwent it. So actually, "kick" and "see" share the same mappings; it's
just a matter of ordering verb arguments the Ebisedian way.


I am a consultant. My job is to make your job redundant. -- Mr Tom