Joe:
> > > > Although in principle I exempt myself from responsibility
> > > > for mentalist issues, I do in fact have two answers:
> > > > (i) invisible words occur where there are apparent gaps in paradigms
> > >
> > > Without counting reference to the situational or speech context, these
> > > gaps can be filled in either internally to a node (as in, e.g., HPSG)
> > > or externally as an extra node (as P&P does in abundance). This is
> > > what the debate is all about, I think.
> >
> > I don't understand what you mean.
>
> Correct me if I'm wrong, but here's what I think that I think:
>
> In WG, as I see it, each node on the dependency tree is really a rich
> network of information. Loosely speaking, you can represent just about
> anything in these nodes (except perhaps for linear precedence?). So
> why not take advantage of the (potential) richness of each network
> and represent the information of invisible words in the relevant
> nodes? This way you get the best of both worlds, which should be
> very appealing.

You'll have to sketch me a concrete example for me to see what you
mean.

> You could also do "coenunciation" this way (think of Chet's work on hybrid
> categories).

I'm not sure what you mean. Are you thinking of things like articled
prepositions, and thinking that these could somehow be both prepositions
and determiners? If so, then although that could be got to work, it
would be far less explanatory than the analysis me & Dick advocate.
I would only go for the hybrid analysis if you wanted to have a model
of grammar wherein compositional semantic structure is read off
compositional phonological structure -- a model that seems impracticable,
though it has a lot to commend it conceptually.

> > Here's an example of what I had in mind:
> >
> >     this book    these books
> >     that book    those books
> >     the  book    the   books
> >     a    book          books
>
> Actually, the last line should be, no?
>
>       a    book    some  books

Maybe. In which case you get the following gappy paradigm:

     this milk    this book    these books
     that milk    that book    those books
     the  milk    the  book    the   books
     some milk    a    book    some  books
          milk    ?????????          books
     ?????????    some book    ???????????

In contrast to:

     this milk    this book    these books
     that milk    that book    those books
     the  milk    the  book    the   books
          milk    a    book          books
     some milk    some book    some  books

which I prefer, and assumes (i) a phonologically null
plural indefinite article, (ii) the indef article imposes
count interpretation (cf. e.g. EVERY), (iii) there is some
semantic irregularity somewhere in the paradigm for SOME.

> (Spanish has a very clear possibility in this case:
>
>       un libro    unos libros
>        a book     a-PL books
>       'a book'   'some books'
>
> Things are more complicated than this because you also have
> algun(o) 'some-SG' ~ algunos 'some-PL.')

What happens with mass nouns in Spanish? What indefinite
determiner is used?

> Anyway, gaps occur all over the place:
>
>      parent             grandparent            ???
>      /    \                /  \               /   \
> mother   father  grandmother  grandrather  aunt  uncle
>
> So I'm not sure what gaps really show.

For some but not all gaps, they have to be stipulated; that
is, without the stipulation, the gap would be filled. "me's"
is an example of such a gap. For other gaps, they don't
have to be stipulated, but the overall paradigm requires
more stipulations (e.g. that AN is singular only, and that
that a bare singular noun can't receive a count interpretation).

> > -- there's a surprising gap where you'd expect to find the plural
> > indefinite article. Everything looks a bit more regular if you take
> > there to be a phonologically invisible plural indefinite article.
> >
> > > > (ii) syntactic structure being semantic structure, invisible words
> > > > are detectable from semantic structure, e.g. when the compositional
> > > > meaning of the phonologically visible words in a sentence does not
> > > > add up to the actual meaning.
> > >
> > > Wow! Can you go into this in a bit more detail? It sounds as if
> > > you've given up (rather radically) on modular syntax.
> >
> > I haven't. I broadly agree with GB/Minimalism, which has syntax
> interfacing
> > with the 'conceptual-intentional' system.
> >
> > I have gone into a bit more detail in another recent posting, with a
> > diagram & stuff.
>
> Yes, but in the posting, which is very interesting (give me some time
> to digest it, though), you don't really justify the move of treating
> syntax and semantics on the same plane.

That's right. The justification would have to be simplicity; avoiding
duplication of structure and multiplication of grammatical entities.

> Perhaps by "semantics" you mean LF, and that you don't take LF to be part
> of conceptual structure. This would be very much like Jackendoff's
> _Architecture_ book, which conjectures the following structure:
>
> Syntax module<--->LF interface<--->Concept. struc. mega-metamodule(!)

That's right. Exactly so. Note there's no terminological sleight of hand
going on here, because like TG, I recognize nothing within the grammar
besides LF that could be called 'semantics'.

> I don't think that this is very WGish, personally, though I'm almost
> certain that Nik might disagree.

(Nik tends to be one of that kind of person who by constitution is agnostic,
understanding a multiplicity of doctrines but refraining from investing
belief in none of them...)

Anyway, what I propose is definitely not very WGish. Here are some reasons
why.

* Because WG gives no formal status to the external boundaries of the
  grammar, WG feels obliged to model conceptual structure - encylopedic
  knowledge and all elements of linguistically encoded meaning.
  In contrast the model of language that I favour is obliged to say
  nothing about encyclopedic knowledge, and relatively little about
  linguistically encoded meaning (e.g. it must say that the sense of
  DOG is 'dog', but needn't define 'dog').
* WG has been against the mixing of levels, mainly on the grounds of
  simplicty (avoidance of redundancy). For example, given that WG
  seeks to model a fully explicit semantic structure (that is, the
  conceptual/cognitive/ideational structure that a sentence encodes/
  communicates), the idea of representing things like theta roles and
  quantifier scope in syntactic structure seems an unwarranted mixing
  of levels, entailing the repetion of semantic information in both
  the syntactic and the semantic structures.

WG is not alone in these views. Autolexical Grammar definitely agrees
with WG. I don't know enough about other models' view of semantics to
say which camp they're in. Categorial Grammar is obviously in the TG
camp.

The difference between camps is very abstract and not easy to grasp.
You might find useful an analogy at the other end of grammar: imagine
that a linguistic theory felt obliged to model phonetic structure, the motor
activity and so on. Advocates of such a model would probably want to exclude
all information about phonetic realization from phonological structure,
since this would be duplicated in phonetic structure.

--And.



%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%