syncretism w/o paradigms

[Jonathan David Bobaljik]

>
>Oh good, I thought this would get a rise out of you!  In the original
>Halle & Marantz story, Agr is inserted post-syntactically.  If so,
>Impoverishment *could* occur before agreement (in the sense of
>detailed feature-copying, not in the sense of abstract syntactic
>agreement).  If Impoverishment *must* occur before feature-copying
>(or *must* affect copied features as well as the originals), then the
>theory is stronger -- more potential evidence for Impoverishment.  I
>don't know if there's evidence against the stronger theory.  Do you?
>
>Cheers,
>Martha

One of the standard arguments for 
unification-based approaches to 
underspecification in theoretical treatments of 
agreement (see Chapter 2 of Pollard & Sag, later: 
Kathol; Wechsler & Zlatic's book, etc) is that 
systematically non-expressed features enter into 
agreement relations. As Steve Wechsler puts it in 
a recent paper on this: "agreement systems can 
evince distinctions that are not reflected in the 
morphological paradigms themselves."

The example Pollard & Sag start with is 1,2 
person pronouns, which are unspecified for gender 
in many I-E languages, but trigger gender 
agreement:

I could say:		Je suis intelligent.
Martha could say:	Je suis intelligente.

There appears to be a markedness generalization, 
say a filter of the Noyer kind: *[person, gender] 
= gender distinctions restricted to the third 
person. This filter constrains possible 
vocabulary items in these languages, both 
pronouns and agreement morphemes (no word class 
marks both person and gender). But--on the 
assumption that agreement is 
copying/matching--the controller must be fully 
specified for features. The subject must be [1 sg 
f] when Martha is speaking, this is matched on 
the targets, even though no single vocabulary 
item can spell out all of the features.

This is, perhaps, evidence that agreement is not 
sensitive to the effects of Noyer-filters. If 
Noyer filters are instantiated via impoverishment 
(gender --> Ø / person; certainly not the only 
way to do this), then impoverishment happens 
after agreement.

I suspect this is general. One standard type of 
argument for underspecification is that features 
are not morphologically signalled on (some class 
of) controllers, but enter into agreement 
nevertheless.


(What I have suggested above is of course not the 
only way of looking at this data. Pollard & Sag 
are ambivalent about these examples, using them 
to motivate unification rather than 
copying/matching on the one hand (which has the 
same effect as agreement before impoverishment; 
hence the same examples are used by Stump to 
argue for realization = underspecified vocabulary 
items partially spelling out a fully-specified 
syntax), but as a hybrid system of grammatical 
and pragmatic agreement later in the chapter.)

Is this more or less what you were looking for?

-Jonathan



>
>mcginnis at ucalgary.ca


--
_______________________
Jonathan David Bobaljik
University of Connecticut
Department of Linguistics, Unit 1145
337 Mansfield Road
Storrs, CT 06269-1145
USA

tel: (860) 486-0153
fax: (860) 486-0197

http://bobaljik.uconn.edu/