about OBJ(theta) as a controller

[Alex Alsina]

Lori Levin gives some motivation for the claim that OBJ(theta) should be 
grouped with OBL(theta) on the basis that both kinds of grammatical 
functions are restricted to specific thematic roles:

> 
> OBJ(theta) shares some properties with OBJ and some properties with
> OBL(theta).  The property it shares with OBJ is its syntactic encoding
> without a semantically restricted preposition or case marker.  The
> property it shares with OBL(theta) is its restrictions in linking only
> to specific semantic roles.

The claim that OBJ(theta) is restricted in linking only to specific 
thematic roles is true for English, at least, true for the analyses of 
English that treat the second object in ditransitive constructions as 
being obligatorily an OBJ(theta).  In such analyses, the OBJ(theta) can 
only be a theme or patient role, whereas the OBJ can be a goal or a 
beneficiary, as well as a theme or patient.  On the basis of these 
analyses of English, one could claim that the OBJ(theta) can link to a 
proper subset of the semantic roles that an OBJ (or a SUBJ) can link to.  
This claim received some cross-linguistic support from the analysis of 
the Bantu language Chichewa (and possibly other languages), in which the 
OBJ(theta), although not so restricted thematically as its English 
counterpart (as it can link to instrumentals and locatives, in addition 
to themes and patients), is still restricted to link to a proper subset 
of the semantic roles that the OBJ can link to.  The OBJ can also link to 
goals, beneficiaries, and agents.

Assuming that all languages could fit into the English and Chichewa
pattern, there would be some motivation for the idea that both OBJ(theta)s
and OBL(theta)s are semantically restricted.  Even so, if this grouping 
is supposed to have some theoretical significance, there should be 
principles, generalizations, rules, what have you, that treat these two 
types of grammatical functions as a class.  I haven't seen any such 
principle, etc.

Anyway, the major problem for the claim that OBJ(theta) is semantically 
restricted in that it links to a proper subset of the roles that the OBJ 
links to came out of the analyses that Gregorio Firmino did of the Bantu 
language Gitonga and that Carolyn Harford did of the Bantu language 
Kitharaka, more or less at the same time.  (Carolyn's work was published 
in 1991 "Object Asymmetries in Kitharaka", BLS 17: Special Session, pp. 
98-105; Gregorio's work, I believe remains unpublished, although I would 
like someone to correct me if I'm wrong.)  Both of these analyses argue, 
very conclusively in my opinion, that in both Kitharaka and Gitonga there 
is a need to distinguish between OBJ and OBJ(theta) and that the 
OBJ(theta) has the same range of thematic roles as the OBJ.  In other 
words, in these two languages the OBJ(theta) does not link to a proper 
subset of the thematic roles that the OBJ links to: it links to the same 
set of roles as the OBJ.

Given this, the claim that OBJ(theta) is semantically restricted is only 
true for some languages and thus cannot be proposed as a universal 
characterization of the OBJ(theta).  Sometimes people working within LFG 
fault theories and analyses in other frameworks for being parochial, that 
is, for proposing notions that are narrowly motivated by the facts of 
a small number of languages (generally, English) for the analysis of all 
languages, even though there are languages that can be shown not fit into 
the proposed scheme.  We don't want our proposals to suffer from this 
defect, do we?

Finally, a note about the subscripted (theta).  Isn't it high time we got 
rid of it?  It appears to me to be some kind of intrusion of semantics 
into the representation of syntax.  A framework that claims to factor out 
different types of information into different formal types or levels of 
representation has no need for such leakages of information.  In the 
initial stages in the development of LFG, people concentrated almost 
exclusively on the strictly syntactic aspects, the c-structure and the 
f-structure.  Lexical semantics (or lexical conceptual structure, or 
whatever) was out there, but no one got near it.  So, at this point it 
was understandable that certain semantic notions should show up in the 
f-structure.  As time went by, people started thinking of ways of 
representing semantics alongside the syntactic levels.  At present, I 
don't think anyone believes thematic roles should be in f-structure: 
thematic roles (or what may correspond to them) should be represented at 
some level of semantic structure.  Given that the f-structure and the 
semantic structure should be linked, it would be unnecessary to specify 
the thematic role of an oblique in the f-structure, since this 
information should be found in the semantic structure.

Alex Alsina

fasaa at leonis.nus.sg