AW: Increasing interest in the HPSG conference

[Carl Pollard]

Hi Andrea,

In reply to my assertion

> A more important distinction than whether the tree is gussied up into a
> feature structure is whether the tree is viewed as (a) a model
> (structural representation) of a sign, or (b) as a proof (derivation)
> of a sign. I believe this is a (maybe even THE) fundamental issue that
> has to be clarified

you said:

  My personal answer to Carl's question is as follows: quite
  obviously, trees are what McCawley conceived them to be,
  i.e. structural representations.

But if it were so obvious, it would not be such a bone of contention.
It SEEMED obvious to me too for many years, then I wondered about it
for a few years, and now I believe it is false.

You went on to say:

  In the 1980s, it was somewhat vague whether GB should be considered
  derivational, and explicitly representational variants like Koster's
  existed, but then Chomsky and Pollock argued that certain well-known
  facts about adverb position, negation and the like could not be
  handled in a representational framework. An actual 'proof' in the
  broadest sense of this term has never been given (and actually
  cannot be given, cf. Kim and Sag).


There are no proofs in science, except for refutations in the form of
counterexamples.

>
What is more, Lasnik (2000, cf. my short notice in JL) has pointed out that
the derivational analyses offered by Chomsky and others do not work!
>>

The MP conception of derivation is not the only one. For example,
categorial grammar is also derivational.

In response to my assertion

> [GPSG] wasn't, as Uszkoreit and Peters proved. (The context-
> freeness of GKPS was bought at the cost of imposing a linguistically
> unmotivated prohibition on application of a given metarule more than
> once in a derivation.)


you replied

  You seem to assume that the price to be paid for the restrictiveness was to
  high.

It is too high because it renders ungrammatical sentences like

  A violin this well-crafted, even the most difficult sonatas are
  easy [to play __ on __].

  I have a personnel issue that I'm not sure who
  [to talk to __ about __].

Moreover GKPS generated only CFL's, and as far as I know nobody has ever
found a mistake in Shieber's proof that Swiss German (as a stringset) is
not a CFL. So GKPS as a package deal wasn't sustainable. (I don't think
HPSG is either but for different reasons. )

>
I flatly reject this conclusion.
>>

I urge you to reconsider.

>
In any case, restricting metarules is
much more restrictive than a bag of completely unrestricted lexical rules,
inheritance hierarchies, features and types of dubious origin, let alone
list- and set-valued features.
>>

But how should they be restricted? You can't just rule out good sentences
in the interest of keeping your formalism lean. AS for the bag you mention,
in spite of having omce presented a WCCFL paper called "PSG without metarules,
I no longer see much difference between metarules and lexical rules. And there
are ways of doing the other things you mention (inheritance hierarchies, features
and types, list and set values) that make them not seem so louche. Which would
you give up?


>
If I remember correctly, there was no question of type vs. token-identity
cropping up in GPSG.
There weren't even AVMs in GPSG.
>>

This is true and to GPSG's credit, and I would advocate neither the type vs.
token identity distinction nor AVMs.

But I would not advocate trees qua structural representations either.


Best

T.