empirical basis, typology, and gradience

[Raúl Aranovich]

Carl and Detmar --

At 04:42 PM 5/1/01 -0400, Carl Pollard wrote:
>Folks,
>
> >
> > Correct me if I am wrong again, but my impression is that work in HPSG has
> > concentrated too much on analyses of a single phenomenon in a single
> > language, too much for my taste, that is.
>
>I disagree with this assessment: good empirical work on theoretically
>interesting phenomena in my experience is exactly what makes it
>possible to have meaningful discussions across frameworks.
> >>
>
>I agree with Detmar here. Inductive generalization has to generalize
>FROM vast numbers of detailed, specific observations.

I wasn't implying that HPSG should give up empirical rigor. That is one of
the things that makes working within this community a rewarding experience:
the theory does not get ahead of the facts. What I had in mind as the
subject of my comment is something like this: let's say someone introduces
a new feature in an analysis of a particular phenomenon in a particular
language. For the sake of argument, let's say that an attribute CLTS is
proposed, the value of which is a list of pronominal NPs, and that this
feature makes an analysis of clitic placement in Romanian work, and
accounts for the facts. That's great, but it could be taken one more step
ahead: Is this feature universal? Do all verbs have a feature CLTS, which
in some languages can only have the empty list as its value, or is this
feature only appropriate for Romanian verbs? How about other languages that
have clitics?  Does the same feature exist in the descriptions of those
languages' verbs? How much variation in the appropriateness of features and
values does the theory allow? These are the questions that for me need to
be asked, and this can be done without sacrificing empirical rigor.

As R. Borsley said briefly in his posting on the subject, HPSG is sometimes
perceived as being too "descriptive", and not "explanatory" enough. I had
to argue against this view many times, but I see where the criticism comes
from. One the one hand, part of the "culture" of HPSG (probably inherited
from GPSG) is not to sacrifice the exceptional on the altar of the
linguistic generalization. This is a very healthy practice, I agree, and it
may be the case that there aren't as many generalizations about language to
capture as we may want to think. But, on the other hand, if one can
introduce a new feature to account for the data 'on demand', so to speak,
without a clear conception of what is language-particular and what is
language-universal, then the perception may be that the analysis is only
aimed at describing the phenomenon. This perception, mind you, is not only
prevalent among Chomskyans, but among practitioners of other frameworks as
well. I defend HPSG citing things like the constraint that the value of
valence features be a SYNSEM object, etc., but I have nothing to say when
asked about how HPSG accounts for language variation in a constrained way.

sorry about the long answer, it is not easy for me to make myself clear
with fewer words!

The question of language variation, and its import for what many
undersrtand by explanation in linguistics, brings me to the second part of
my comments and your replies. I should have probably not brought it up,
since I am just starting to work on these issues. I'll give you a short
answer: I think there is room to adapt the results of Aissen and Bresnan's
recent work on Optimality Syntax to HPSG, to have a theory of language
variation. Rob Maluof's paper on word order and constituency, which he
presented at the last HPSG conference in Berkeley, is a brilliant example
of how this can be done (Help me here if you are reading this, Rob; am I
interpreting you correctly?). I think the idea is that default unification
can resolve conflicts between constraints (finding the 'optimal'
candidate), and that language variation is the result of 're-ranking' the
types in the hierarchy. But this is probably too dense to post to the list
in an intelligible manner.

> >
> > don't know enough about it). The hierarchical lexicon is a very powerful
> > tool to do typology and contrastive linguistics,
>
>Has this "very powerful tool" ever been used for typological work?  If
>so, I'd appreciate a pointer.
> >>
>
> > to it one can have a beautiful theory of markedness.
>
>To have a theory of markedness, there must be something called
>markedness to have a theory of. It has never been clear to me that
>there is such a thing. Do any branches of science outside of
>linguistics make use of a notion of markedness? Most mammals bear live
>young, but the platypus and the echidna lay eggs.  So are they MARKED
>mammals? Is giving birth to live young a DEFAULT for mammals? I am not
>aware that these notions have found their way into the vocabulary of
>zoological taxonomy; instead one just sets up a subsort of mammals --
>monotremes -- with only two species. How are words different from
>mammals in this respect?
>
>Carl

In am *not* going to get into that discussion now! No, sir (but how old
were you, Carl, when you discovered that dolphins weren't just fish that
had to come out of the water for air?)! Markedness, true, can be a slippery
topic. I think I see it as relating to the fact that relativizing a direct
object is more "marked" than relativizing a subject, or that plural is more
marked than singular. When I have the relationship between markedness and
the hierarchical lexicon worked out, I'll let you know.

To sum up: I think HPSG has the potential to answer very broad questions
about language (not just grammar), and that it is important to do so, to
bring the formal discipline of HPSG to other areas of linguistics, and
to     keep HPSG relevant as a theory of grammar outside a reduced circle.

Just my two cents. And I appreciate all the information I am getting out of
this discussion. Now I know what a 'monotreme' is!