still more on OBJ(theta)

[Alex Alsina]

In connection with the status of OBJ(theta), one of the points that I
touched on in my previous posting, but which is worth emphasizing, is
the lack of theoretical and empirical motivation for the grouping of
OBJ(theta) and OBL(theta).  A theoretical construct (such as a
feature, a category, a natural class, etc.) must be motivated; that
is, the construct in question must be required for the statement of
some principle, regularity, etc. of the theory.  If it is not required
for such a purpose, it is not necessary, which means that it should be
dispensed with: it has no place in the theory.  As I said earlier, I
still have to see a principle or rule that refers to the putative
class of GF(theta)s.

The analytical fact that SUBJ and OBJ do function as a natural class
does not imply that the GFs that are not in this class also form a
natural class.  I don't think anyone would accept the idea that a
heterogenous ensemble of elements could form a natural class *by
exclusion*, i.e., by not belonging to a motivated natural class.

Joan Bresnan, in her posting on this topic, says:

> In her 1986 dissertation Lori Levin first proposed eliminating the
> primitive syntactic distinction between primary and secondary
> objects by reducing it to an independently motivated property of
> semantic restrictedness.

Let's look at the independent motivation for the property of semantic
restrictedness.  Around 1982, the GFs that were assumed to have the
property of semantic restrictedness were exclusively obliques.  I
won't try to think of all the evidence that motivated this property,
but it would no doubt include: the inability of semantically
restricted GFs to functionally control.  There is probably other
evidence of this kind, some of which will be language-particular.
However, as has been pointed out earlier, OBJ(theta)s can be
controllers in functional control.  So, as far as I can see, the
independent motivation for the property of semantic restrictedness
disappears the moment we claim that this property characterizes both
OBJ(theta) and OBL(theta).  We are back to the point I was making at
the beginning of this posting: there is no motivation for grouping
these two GFs as a class.

Joan goes on to suggest that there are two ways we can interpret the
notion of semantic restrictedness such that it can define this
proposed class of GFs:

> (i) restrictedness means having a specified or fixed semantic role;
> (ii) restrictedness means having *some* semantic role (i.e. not being
> athematic, as are expletives, raised subjects, etc.). 

As Joan observes, the first interpretation has to be rejected, given
the evidence from Kitharaka and Gitonga that I mentioned.  What about
the second interpretation?  (It is obviously not the intended
interpretation in the texts that present the LMT, but it might be a
legitimate one, as long as the classification it tries to preserve
does some work: we don't want unemployed (chomeur?) theoretical
constructs.)  The observation that OBJ(theta)s cannot be athematic has
no known counterexample; it therefore needs to be explained.*

	*Thanks to Chris Culy for reminding us that there are
counterexamples to the claim that obliques cannot be athematic.  If we
cannot explain away these counterexamples, the last attempt to
motivate the grouping of obliques and OBJ(theta)s will have failed.  I
will proceed, for the purpose of the discussion, as if there were some
way to explain away those counterexamples.

The observation that OBJ(theta)s and OBL(theta)s cannot be athematic
could perhaps be explained by having a principle that prevents
GF(theta)s from linking to a semantic argument, which may be what Joan
has in mind.  This would also provide the desired motivation for the
class of GF(theta)s.  However, this principle would be redundant as
far as obliques are concerned, if we make the reasonable and standard
assumption that an oblique, because of its morphology, is compatible
only with a specific kind of semantic argument: it is semantically
restricted in the original interpretation.  The principle would then
just be an ad hoc principle for OBJ(theta)s.  Do we still want it?
No, it is in fact even unnecessary for OBJ(theta)s, since the
obligatory thematicity of OBJ(theta)s can be independently explained.
I will try to indicate how, as briefly as possible.

An interesting idea that is clearly presented in Bresnan and Moshi
(1990) is that the so-called intrinsic classifications of arguments
are part of the argument structure.  For internal arguments (or
object-like arguments), this means that the features [-r] and [+o] are
represented at a-structure.  One can therefore read these features as
imposing a constraint on the mapping between arguments and GFs.  A
[-r] argument must link to SUBJ or OBJ, and, given the default mapping
principles, a [+o] argument must link to an OBJ(theta).  What this
means is that, in B&M, the two types of objects--OBJ and
OBJ(theta)--are distinguished at the level of a-structure, in addition
to being distinguished at the level of grammatical functions.  A
restricted object is a [+o] argument at a-structure and an OBJ(theta)
at the level of GFs; an unrestricted object is a [-r] argument at
a-structure and an OBJ at the level of GFs.  So, we have the same
distinction expressed twice: [+o] vs. [-r], and OBJ(theta) vs. OBJ.
In a framework such as LFG that factors information into different
levels of representation, this is an undesirable situation because the
same theoretical distinction is being represented at two different
levels of representation: a failure of factorization.

The solution to this problem is to represent the distinction between
the two kinds of objects only at one level.  Given the architecture of
the framework, the information at one level constrains the information
at other levels, so that there is not only no need to copy formal
distinctions from one level into another, but it goes against the very
spirit of LFG.  The evidence that the distinction between the two
kinds of objects must be represented at a-structure is compelling and
I will not repeat it here.  (See, for example, B&M.)  Consequently, if
the distinction is represented at a-structure, it need not, i.e.,
should not, be duplicated at the level of GFs.  What this means is
that there is one single kind of grammatical function at the level of
GFs, which we may call OBJ, and it will have predictably different
empirical properties depending on whether it maps onto a [+o] argument
or not. 

A formal implementation of these ideas can be found in my
dissertation, which you can refer to if you are interested in the
details.  I will just mention two assumptions of this theory that
explain why restricted objects cannot be athematic, whereas
unrestricted objects can be.  I assume that the distinction between
the two kinds of objects is captured by means of the privative feature
[R] at the level of a-structure.  (By the way, this feature is not
part of the representation of obliques.)  An OBJ that maps onto an [R]
argument is a restricted object, whereas an OBJ that does not map onto
an [R] argument is an unrestricted object.  The second assumption is
that arguments at a-structure are semantically contentful; they map
onto semantic structure.  What this means is that expletives and other
athematic functions are not represented at a-structure.  Consequently,
a restricted object cannot be athematic, since it is an OBJ that maps
onto an [R] argument, and arguments are by definition thematic.  On
the other hand, an unrestricted object, as it is defined as an OBJ
that does not map onto an [R] argument, need not map onto any argument
at all, i.e., may be athematic.

A clarification.  Joan writes:

> Alsina also argues that by admitting restricted objects into the
> theory we are parochially assuming that what appears in English is 
> universal,

Did I say that?  I certainly didn't intend whatever I said to be
interpreted in this way.  I argue, in this message, as I do elsewhere,
that restricted objects are part of the theory (although not all
languages have restricted objects--but that's another matter).  Where
my proposal differs from previous ones is that restricted object--or
some abbreviated form--is not the name of a grammatical function, but
is a linguistic category that is defined by looking at two levels of
representation at the same time: an OBJ (at the level of GFs) and an
[R] argument (at the level of a-structure).  What I argue would be
parochial is to claim that, just because restricted objects in English
link to a proper subset of the thematic roles that unrestricted
objects link to, this property of restricted objects in English should
be proposed as a universal characterization of restricted objects.  I
think we agree on this point.

One of the consequences of the theory outlined earlier is that there
is only one grammatical function label (OBJ) for the two kinds of
objects.  Having successfully got rid of the semantic-looking
appendage (theta) of restricted objects, it is obvious that we can do
the same with obliques.  The standard interpretation of the
subscripted theta is that it is a cover term for any thematic role
(locative, agent, instrumental, etc.).  So, OBL(theta) is normally
understood to be a notation for collapsing the various oblique
functions: OBL(loc), OBL(ag), OBL(inst), etc.  Let's think about what
this means: an OBL(loc) is an oblique that links onto an argument that
bears a locative semantic role.  If semantic roles are defined and
represented at some level of semantic structure, the information that
an OBL(loc) links onto a locative semantic role is represented at this
semantic structure, which is a level of representation parallel to
f-structure.  Clearly, then, by looking at semantic structure we can
see that an OBL(loc) links onto a locative role; the subscripted (loc)
does not add any information--it merely seems to replicate an
information available elsewhere.  So, we can dispense with the
subscripted (loc), and, generalizing, with the subscripted (theta).

Alex Alsina

fasaa at leonis.nus.sg