Default unification: online vs. offline

[Lyne Da Sylva]

Robert Levine wrote:
>
> *Is* there either data or psycholinguistic evidence which actually point
> unequivocally to defaults of any kind as anything other than an
> abbreviatory device?
>
> Bob Levine


I argued in my thesis (1998) that there are syntactic phenomena
of French which are best handled by default values.
The most prominent example is that of agreement features :

(i) in most cases, words with agreement features (mainly
    adjectives, determiners and verbs) do agree with a
    so-called controlling NP in number, gender and person
    (person being restricted to verbs, and gender to non-verbs)

(ii) in a *large* number of specific contexts, no NP is
    present to determine the agreement features (this includes
    cases where an adjective is used as a noun or an adverb,
    where the subject of the verb is a clause rather than an NP,
    where semantically null pronouns are required by the syntax, etc.);
    in those cases, the agreeing forms systematically appear
    in the 3rd person, masculine, singular form (where applicable)

There are other ways of getting the agreement features
right in all of these various contexts, but they involve duplicating
the agreement features in all contexts (and are unable to explain the
fact that in all the contexts the features are the same). They lack
the overall generalization that in the absence of evidence
to the contrary, agreeing forms should be in the 3rd person
masculine singular.

This is not a lexical default: lexical items are not 3rd-person-
masculine-singular by default, only in certain (wide-ranging)
syntactic contexts. Thus, I can see no justification that the default
in this case be an abbreviatory measure: it seems to describe
a default reasoning rule and not a static descriptive statement.

These defaults need to be handled as persistent defaults (as in
Lascarides and Copestake, 1999) because of the more or less wide
syntactic scope of the defaulting contexts.

I am embarking on a research project which aims to provide an
evaluation of default-free and defaulting HPSG-style grammars
describing the same language phenomena, in order to assess
what are the costs and gains of using defaults. I do hope
to offer, sometime soon, positive evidence as to the usefulness
of defaults in syntactic descriptions.

My thesis is available online (but in French only) at the following
address :
http://www.pum.umontreal.ca/theses/pilote/dasylva/these.pdf


Lyne Da Sylva
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Lyne.Da.Sylva.vcf
Type: text/x-vcard
Size: 449 bytes
Desc: Card for Lyne Da Sylva
URL: <http://listserv.linguistlist.org/pipermail/hpsg-l/attachments/20011023/8d8951a2/attachment.vcf>