an aspectual view of OBJtheta

[Miriam Butt]

Hi,

I followed the discussion about Obj thetas with some interest, but
haven't had a chance to put in my two cents until now. 

I have only recently begun thinking about OBJ vs. OBJtheta since in my
dissertation, I just assumed the most standard version of LMT that I
could find, which seemed to work quite well.  

However, I have been very interested in Alex Alsina's arguments for a
rethinking of the grouping of grammatical functions as implied by the
[+/- r] and [+/- o] features, and it seems to me that he has very
strong arguments for the view he has articulated in his dissertation
and subsequent work. In particular, I find that I agree very much with
his statement that

    "there is only 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."

However, I have not been able to become convinced of Alex's particular
approach to solving the problem, i.e., the adaptation of Dowty's
proto-roles and the additional feature [R] that he talks about for the
OBJ2s.  Something like [R] just seems to me to be non-explanatory,
though it may work technically.

The reason I got thinking about these creatures is that I've been
trying to visualize how one might integrate the findings as to
compositional aspect into LFG (I got there through my private hobby,
complex predicates, but the issue is more global, I think). That is,
there has been quite a lot of work done on LEXICAL aspect within LFG
(Annie Zaenen's work, for example) and related approaches (like Van
Valin's, for example), where it is attempted to derive certain
syntactic patterns of NP arguments through a lexically decomposed
structure which includes CAUSEs, BECOMEs, DOs, etc.  This works pretty
well for many cases, but does fall short (as was the case in Dowty's
(1979) original work as well).  

Famous contrasts like 
   (1) Hans ate the/an apple  (telic --- achievement)
   (2) Hans ate apples        (atelic -- activity)

show that one and the same verb can in fact have differing aspectual
interpretation, and that this kind of pattern is systematic for a
large number of verbs in the lexicon.  

One approach to take would be that this kind of difference in semantic
interpretation is the domain of semantic interpretation and should be
left to that module exclusively.  I.e., it has absolutely no bearing
on the OBJtheta issue.  

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.

Helen de Hoop proposes a fundamental distinction between two types of
objects:  "Strong" and "Weak" Case objects.  In her structural
account, these sit in different phrase structure positions -- the Weak
Case ones are direct sisters of the verbs (and can't scramble).  In
fact, Veerle van Geenhoven in some recent work on Noun Incorporation
in Greenlandic argues that incorporated structures can be thought of
as special instances of Weak Objects (the N must not just be a sister of
the V, but be morphologically attached to it), and draws a parallel
between the semantics of Germanic bare plurals like (2), and the West
Greenlandic incorporated structures (also arguing for a base-generated
account of incorporation, as opposed to head-to-head movement account
like Baker's).  

Now, this Weak vs. Strong Object notion has recently been gaining some
ground in the literature, since it does seem to be a notion that can
capture quite a bit of nifty data.  People tend to disagree precisely
on how to interpret these two types of objects, but the
differentiation itself seems to be a useful concept.

Now, in trying to think how I could fit in this very useful notion
into an LFG approach (I was working with Tracy King on bare objects in
Urdu and Turkish and their interaction with preverbal focus, where
this distinction is incredibly handy), it occurred to me that LFG in
fact already had this distinction in that it did posit a split between
differing kinds of objects: OBJ and OBJtheta. However, once one takes
a close look at the correspondences between OBJ, OBJtheta, Strong
Objects, and and Weak objects, the correspondence between the concepts
seems to go in a way rather differently from what one might initially
assume.

OBJtheta in fact must be taken to correspond to the Strong Object, and
OBJ to the Weak Object. That is, OBJtheta is the more "normal" kind of
object, in that it makes a semantic contribution to the clause in
terms of telicity, specificity, etc., while the OBJ just hangs around
being unmarked. OBJtheta is thus "the apple" in (1), and OBJ is
"apples" in (2).

Now, this view is not so strange if one takes a slightly different
slant on what "semantically restricted" is supposed to mean. If
"semantically restricted" is taken to be equivalent to: this object
makes a semantic contribution in terms of marking specificity,
partitivity, or even affectedness, then in fact, that is exactly what
many objects seem to do, and it would only make sense to note that
down in the form of OBJtheta. Thus, in fact, what we have is that OBJ
and OBJtheta are actually one and the same GF, and OBJthetas are
simply annotated to reflect the fact that they are contributing some
semantics to the clause, while the OBJs are "underspecified" and
therefore Weak, in that they have nothing much to contribute (and
hence, are also susceptible to noun incorporation and the like).

At this point, we could follow Alex's suggestion and simply choose to
dispense with the theta subscripts on OBJ, letting the [+/- r,o]
features speak for themselves, and avoid duplication of information in
the grammar, as Alex argues. However, I have been trying to take this
business with the Weak and Strong Objects one step further and have
been checking to see whether we could, in fact, incorporate into LFG
some treatment of compositional aspect as has been worked out in
contributions by Manfred Krifka and Hank Verkuyl over the years. In
fact, Krifka proposes the use of "Aspectual Roles" which appear to
function very much like the Theta Roles we are used to, but which
enter into a compositional semantic interpretation in a well-defined
way. Gillian Ramchand, in her work on Scottish Gaelic, picks up on
Krifka's ideas, and extends the set of formally defined "Aspectual
Roles" to account for a wider variety of verbal frames. Now, it turns
out that these Aspectual Roles are actually quite handy. They are
well-defined, and people (Krifka) have worked out a semantics that
knows what to do with them. They also simultaneously provide the kind
of information that is crucial for linking. For example, in Urdu and
Turkish (examples for Urdu given below) specific, affected objects
must be marked with an accusative "-ko", and simultaneously must carry
along the information that the aspect of the VP must be interpreted as
telic). On the other hand, unmarked objects as in (4) give rise to
ambiguity (see T.Mohanan 1995 in NLLT on this), which also affects the
aspectual nature of the clause.

  (3) Nadyaa-ne ghore-ko bheje
      Nadya-Erg horses-Acc sold
      `Nadya sold the horses'   (telic)

  (4) Nadyaa-ne ghore      bheje
      Nadya-Erg horses-Nom sold
      `Nadya sold (the) horses' (telic)
      `Nadya did horse-sellinug' (atelic)

The proposal I'm trying to work out currently (partly in collaboration
with Mary Dalrymple) is the following:  we use these aspectual roles
to guide the linking of lexical semantic information to syntax.  In
particular, what is relevant for linking turns out to be divided into
two classes:  aspectually "contentful" roles, and those that are not. 
In fact, this turns out to correspond to the distinction between
Strong (OBJtheta) and Weak (OBJ) objects.  The two-way distinction is
sufficient for linking:  Strong objects are linked via [+o] or [+r]
(right now, I'm tending towards [+r]), weak objects via [-r].  
However, in order to be able to do justice to further semantic
interpretation at some point, especially in terms of figuring out
aspect compositionally, finer-grained information is needed.  This
information is in fact contained in the various types of aspectually
contentful roles formulated by Ramchand (based on Krifka).  So, I
suggest we use the theta subscript on the OBJ as a way to pipe
relevant information to the module that will do semantic
interpretation.  As such, then, the theta subscript is not functioning
merely to reduplicate information already encoded in the [+/- r]
business, but is instead encoding semantically relevant information
that will be taken up by the component that does semantic
interpretation.  

I was a bit mystified by Joan's comment that:

   "Moreover, "restricted" can be defined without any reference to
thematic roles:  for example, in the interpretation suggested above,
it simply means that the function must be indexed to a lexical role
within the argument list brackets.  This of course has empirical
content when coupled with the semantics, as in the recent work of
Dalrymple, Lamping, Pereira, and Saraswat."

I do agree with the first part, but I'm not sure how the indexation to
a lexical role at arg. structure can be smoothly coupled to the type
of semantics introduced by Dalrymple et al.  Mary Dalrymple and I
have, in fact, been talking quite a bit about what needs to be done in
order to really have the linear logic semantics (which I think is
pretty neat) work together with the assumptions of linking theory in a
meaningful way.  It has long been clear (from work by Dowty, mostly, I
think) that one does not want to do semantic interpretation directly
off of thematic roles -- things simply turn out incoherent.  Also,
there turn out to be some incompatibilities between the assumptions
made by Dalrymple et al. about linking theory, and the way it is
actually working in LFG.  

One good way to think about the relationship between semantic
interpretation and linking, and one that avoids both theoretical and
technical confusion, I think, is the approach I've sketched
above.  


Miriam

P.S. Note on the Persian for Siamak Rezaei. The Persian use of -ra on
objects is very reminiscent of what happens in Urdu (Butt 1993) and
Turkish (Enc 1991, -- also for Turkish, you could ask Beryl Hoffman
about it, who's spending this year in Edinburgh). What I am mystified
and very intrigued by is the use of "-ra" as a topic marker as well,
and I'll go check it out.