Answer to Roger and Shuichi (was: Trees, pheno, tectogrammar)

[Carl Pollard]

Hi everyone,

I'm afraid this pheno-tecto discussion has gotten past the e-mail
bandwidth. (By the way, focused workshops are good settings for this
sort of thing.) When I introduced Curry's terms into this discussion
I tried (and maybe failed) to make clear I was assuming

(1) that tectogrammatical, phenogrammatical, and semantic entities are
three different kinds of things, and

(2) that there are functions (call them phn and sem) that interpret
the first into the second two:

   phn: Tecto -> Pheno              sem: Tecto -> Sem

The first of these assumptions is possibly shared by
linearization-style HPSG (depending on whether you are working in a version
of it where there are actual data structures with edges called DAUGHTERS,
HEAD-DTRS or some such thing), but the latter definitely isn't. It is
this latter assumption that I referred to informally as tectostructure
"mediating" between phenostructure and semantics.

[I also think this is what Curry meant, but he wasn't INTERESTED in
phenogrammatical structure, so didn't say much about it. Jim Lambek tells
the story of being buttonholed by Curry at a meeting. Curry asked him
what he (Lambek) thought was so interesting about the order the words
happen to come out in. Lambek pointed to Curry's nametag and said,
"If word order didn't matter, this would say

             HASKELL CURRY
]             STATE PEN[N]

I owe you an answer.  First Roger, then Shuichi.


When Tibor says in his reply to Roger:

  ... I am quite sure that you would agree that phenogrammatical analyses
  of scrambling would not provide a tecto-grammatical (and hence:
  interpretational) difference between the grammatical and the ungrammatical
  case.

I read him as rejecting assumption 2.

Tibor goes on:

 Since you have alluded to a possible straw man in your email, let me
 try to make clear against which position I am arguing. If I remember
 correctly, Dowty's initial paper contains _minimalist_ in his title,
 and he is quite explicit in claiming that tectogrammatical combination
 and phenogrammatical reordering can be separated *and* are not
 mediated by syntax.

That way of describing Dowty's setup might be misconstrued.  Dowty
very clearly uses a tectogrammatical skeleton where each node is
phenogrammatically interpreted as a multiset of objects each of which
is either a word or a sequence of words fused (or "frozen") together
by an operation called ATTACHMENT (written +).  But there can also be
LP rules (as in G/HPSG) that constrain how the multisets can be
linearized. So the LP rules know nothing about tectostucture, but the
multisets to which they appy are "read off of" tectostructure. If you
use Dowty's architecture, then a good definition of what it means for
a phnenomenon (say topicalization, extraposition, or scrambling) to
"be phenogrammatical" would be that the topicalized, extraposed, or
scrambled sentence and a "canonical" counterpart come from the same
tectostructure, and therefore the same multiset, linearized (subject
to the LP rules) in two different ways.

Since (in this architecture) semantics is still read off of
tectostructure, it follows that such phenogrammatical phenomena have
no semantic effect. Given information-structure considerations, many
would argue that (assuming one wants to use this architecture)
topicalization and some things that have been called scrambling, such
as Mittelfeld ordering) are NOT phenogrammatical in this sense, and
therefore would have to be registered somehow in the
tectostructure. Thus the only purely phenogrammatical phenomena are
ones that are the word-order analogs of free variants. Some kinds of
extraposition might fall into this category.


Tibor continued:

  If I understood and remembered their proposals correctly, neither
  Dowty nor Kathol or Reape has claimed that there is a mediating
  syntactic component which records and transmits phenogrammatical
  information to tectogrammar  and vice versa. If somebody assumed
  that there *is* indeed a mediating syntactic component, I would
  argue that such a component is what other people assume under
  syntax proper and hence that these so-called minimalist approaches
  are in fact not really minimalist (i.e. they are not parsimonious).

Based on my sketch above of what Dowty actually proposed, this
characterization seems to me an oversimplication of Dowty's view.


  To sum up: Conceptually, the position I am attacking assumes
  phenogrammar , tectogrammar and no mediation. Empirically, it
  assumes that topicalization, extraposition, and scrambling are
  phenogrammatical operations.

I agree with Tibor here (except possibly about some kinds of
semantically/informationally non-potent extraposition and scrambling,
if they really exist), but

  If there is a straw man involved, please feel free to inform
  me.

think that Dowty was made into a straw man.

At the end of his reply to Roger, Tibor noted the contrast

(1) *Etwas     zugefl=FCstert, der dort  steht, hat sie dem Mann.
     Something whispered-to  who there stands has she the man
(2) Sie hat dem Mann etwas     zugefl=FCstert, der dort steht.
    She has the man  something whispered-to  who there stands

I wasn't sure what theoretical point was at stake here. Tibor,
can you clarify?

Finally, with respect to Tibor's reply to Shuichi:

  With respect to Shuichi's email, I would like to raise attention to
  the following type of example found in German (and presumably other
  languages like Korean and Japanese, according to Hoji (1986)):

  (3) QP_1 QP_2 V
  (4) QP_2 QP_1 V

  In (3) and (4), QP designates a quantificational NP and 1 and 2
  indicate = an ordering of the quantifiers according to some
  hierarchy (let's assume obliqueness, so that 1 is IndObj and 2 is
  DObj). The funny thing is that = (3) is in fact not ambiguous, but
  (4) is, allowing both a reading of QP_2 > Q= P_1 and vice versa.

If my memory is right, back in the early 1990s Todd Yampol showed
in a very detailed study of this phenomenon (he also looked at
three-quantifier combinations) that, at least for Japanese, the
empirical generalization stated above does not hold up. Todd
got out of this line of work (last I heard he was applying to
business schools), but I'll try to dig out the paper (which
i don't think was ever published) and summarize its conclusions.

Carl