subject: objects

[Alex Alsina]

There are several comments I would like to make about Miriam Butt's
first posting on the issue of objects.  They come under the headings:
(1) Should the OBJ/OBJtheta distinction be a multi-purpose
distinction?  (2) Is semantics autonomous (of syntax), and why does
that matter?  and (3) what is explanatory?

(1) Should the OBJ/OBJtheta distinction be a multi-purpose
distinction? 

Standard LFG proposals about grammatical functions posit a distinction
between the notions of restricted and unrestricted object in terms of
grammatical functions, currently notated as OBJ and OBJtheta (the
"theta" being a subscript).  (I've stated what I have to say about
this proposal in my previous message, so I won't insist now.)  When
one examines the facts of a language, one may find a syntactic
behavioral distinction between two kinds of syntactic elements that we
can reasonably analyze as objects.  Thus, in a sense, we have two
different kinds of objects.  A big temptation is to say: "I have two
kinds of objects in this language, and LFG gives us two distinct
formal attributes (OBJ and OBJtheta) for different kinds of objects;
so, let's use this formal distinction to capture the behavioral
distinction that I have observed."  This move will often be
disastrous, and I will use Miram's proposal to illustrate this point.

Work such as Bresnan and Moshi 1990 and Alsina and Mchombo 1989/1993
constitutes the beginnings of a substantive, cross-linguistic theory
of object asymmetries.  The essence of this theory is that some
languages, such as Chichewa, exhibit an asymmetry among objects
whereby one of the objects in a double object construction lacks
certain properties that the other object has.  We can call the object
lacking some properties the restricted object (for me, the intuitive
motivation for this name being that the object is restricted in its
syntactic behavior), and the other object the unrestricted object,
regardless of how we represent this distinction.  (In the work
mentioned, this distinction is doubly represented: respectively, as
the [+o]/[-r] distinction at a-structure, and as the OBJtheta/OBJ
distinction in terms of GFs.)  The theoretical and empirical substance
of this distinction is that the restricted object cannot alternate
with the subject function, as in passivization, cannot be
reciprocalized, and cannot be expressed by means of an object marker
(a verbal affix agreeing with the object or functioning as a
pronominal object), in addition to possibly other properties.  Thus,
if we want to preserve the essence of this theory, when we extend it
to another language, we should make sure (a) that the theoretical
distinction corresponds to the predicted behavioral asymmetries, and
(b) that these behavioral asymmetries are not best accounted for in
some other independently required or motivated way.

If my understanding of Miriam's proposal is correct (I do share some
of Chris Pinon's feelings about it), the behavioral properties
associated with the two kinds of objects, OBJ and OBJtheta, in her
proposal have absolutely nothing to do with the properties predicted
by the theory of object asymmetries of B&M, A&M, etc. summarized in
the previous paragraph.  On the other hand, the properties associated
with OBJ and OBJtheta in Miriam's proposal correlate with semantic
properties such as specificity, partitivity, affectedness, which have
absolutely no correspondence with the two kinds of objects according
to the theory of object asymmetries.  

We are thus faced with a dilemma: either (a) the OBJ/OBJtheta
distinction is a multi-purpose distinction, in other words, a formal
distinction that you can use at your convenience to analyze any
syntactic behavioral asymmetry among objects, or (b) it is not a
multi-purpose distinction, but is associated with empirical,
cross-linguistically valid, consequences.  It is clear that, whereas
option (b) has the potential to yield an interesting, cross-linguistic
theory of object asymmetries, option (a) does not have this potential
and can only yield a hodgepodge.  I think most of us would be
inclined to reject option (a) and go for option (b).  Within option
(b), of course, alternative theories could be developed.  One theory
that has been developed within option (b) is the one I call the theory
of object asymmetries, described two paragraphs up.  Another
alternative could be the one Miriam proposes.  What needs to be made
clear is that the theory of object asymmetries and Miriam's proposal
are incompatible theories (within option (b)): we cannot keep both; we
have to choose one and bin the other.  Until Miriam makes a very
persuasive case for her proposal, my choice is made.


(2) Is semantics autonomous (of syntax), and why does that matter?

Current linguistic theory can be seen as being split between two ways
of conceiving the relationship between semantics and syntax.  The
Chomskyan school of thought claims that semantics (or semantic
interpretation) is derived in a mechanical way from syntactic
structures.  Under this view, semantics is not represented as an
autonomous level of structure; therefore, we can call this hypothesis
the "Non-Autonomy of Semantics", or NAS, for short.  The NAS is an
essential part of GB, the minimalist program, and is built into the
GB/MP model by having the grammatically relevant semantic
representation expressed as LF, a syntactic level of representation.
A consequence of the NAS is that any semantic distinction with a
bearing on syntax must be represented in the syntax; if it weren't,
it would not be possible to capture the correlation between syntax and
semantics, because all that the semantic component does is interpret
or read the semantics off the syntax.  The alternative school of
thought has it that semantics is autonomous of syntax, in the sense
that principles specific to the semantic structure generate this
structure, which is put into correspondence with syntax by means of
principles that establish well-formed linkings (mappings,
correspondences) between pieces of the semantic structure and pieces
of syntactic structures.  We can refer to this hypothesis as the
"Autonomy of Semantics", or AS for short.  The AS allows semantic
information, that is, information represented only at semantic
structure, to constrain the syntactic structures that a given semantic
structure is linked to.  A consequence of the AS is that a semantic
distinction with a bearing on syntax need not (i.e., must not) be
represented in the syntax, since it should be represented in the
semantic structure, where it can constrain syntax by means of
correspondence principles.

Within LFG one could choose to adopt either the NAS or the AS; I
believe that the framework as such does not force the choice on us and
is consistent with either hypothesis.  However, given the general
architecture of LFG, in which grammatical information is factored into
different levels of representation related to each other by
correspondence principles, it seems that the AS is the natural
assumption for LFG, and I'm inclined to believe that it is the more
general position in work in this framework.  We could say that there
is a default assumption with LFG, namely, that you subscribe to the
AS.  If that is not the case, one should come out in the open about
it.  I think this is a problem with Miriam's proposal: it only makes
sense if one adopts the NAS, but she doesn't give any explicit
indication about this.

Consider the following paragraph from Miriam's first message:

>However, there then is work like that of Helen de Hoop, who examined a
>whole slew of languages and showed that in language after language the
>kind of telic/atelic contrast is also associated with case-marking
>(Finnish is the classic case). Gillian Ramchand, taking up this
>general line of investigation, further found that in Scottish Gaelic,
>the contrast is also encoded by word order: NPs giving rise to atelic
>readings have a differing distribution from those giving rise to telic
>readings (postverbal vs. preverbal). Trying to account for
>case-marking on objects and their distribution within the clause in
>general puts us back squarely within the domain of syntax, and in
>particular, of linking.

I'm not sure about Helen de Hoop (pardon my ignorance), but Gillian
Ramchand, at least the last time I saw her, was working within GB.
People working in GB are totally out of the closet about adopting the
NAS and don't even have to say it.  If you adopt the NAS and find that
a semantic contrast, such as the telic/atelic contrast, has a
syntactic correlate, such as a case-marking contrast, or a word order
contrast, you are forced to assume that this semantic contrast is
syntactically represented, which is very different from saying that it
has syntactic correlates.  And this is what Helen de Hoop and Gillian
Ramchand seem to do, according to Miriam.  However, if you do not
adopt the NAS, but adopt the AS instead, you do not need to assume
that the semantic contrast is syntactically represented.  In fact, the
idea of factorization of information and the availability of
correspondence principles make it undesirable to make this assumption.
The idea is that the telic/atelic distinction is a semantic one and
therefore should be represented in a semantic structure and a
principle of correspondence with syntax would make that semantic
distinction constrain the syntactic representations in the appropriate
way.

The point is that one should be clear about whether we assume that
semantics is autonomous or not, because this choice has consequences
on the way we write our grammars, and, hopefully, has also empirical
consequences.  This choice between the autonomy and the non-autonomy
of semantics is one of the topics that should come out in the workshop
on semantics organized by Tara Mohanan and K.P. Mohanan in the
forthcoming LFG colloquium in Grenoble.


(3) What is explanatory?

My last comment is prompted by Miriam's remark concerning my proposal
about the formal representation of the distinction between restricted
and unrestricted object:

> Something like [R] just seems to me to be non-explanatory,
>though it may work technically.

As far as capturing the distinction between restricted and
unrestricted object is concerned, the presence vs. absence of the
feature [R] at a-structure is equivalent to the feature [-r] vs. the
feature [+o] at a-structure respectively.  The latter is the proposal
in Bresnan and Moshi 1990 and others, and also in Miriam's own
proposal.  Given that the two alternatives are equivalent in capturing
the relevant distinction, both are equally explanatory, or
non-explanatory, as you prefer.  If Miriam were not using the
[-r]/[+o] distinction, or something like it, I would be looking
forward to a more explanatory proposal.  But she is using this
distinction.  So, we haven't made any improvement.

Maybe Miriam means something different by explanatory: perhaps that
the features used should correspond to some semantic concept.  (Yet,
associating a semantic concept with a syntactic feature can be done
equally well with either of the two formal proposals being compared.)
But if this is what she means, then we run into the two problems
discussed earlier in this message: the distinction between restricted
and unrestricted object becomes a multi-purpose one, and we duplicate
in the syntax what the semantics has to express anyway.

Alex Alsina