Trees, pheno, tectogrammar

Thu Jul 1 13:00:04 UTC 2004

Hi Tibor, Carl, and everybody else,

> For scrambling, the effects on quantification and variable binding are
> dramatic (and have been completely ignored in e.g. Kathol's thesis).
>
> To cut a long story short: Separating tectogrammar from phenogrammar does
> not work.

Although I argued for linearization accounts for over a decade, I now
also believe that a linearization based treatement of German is
empirically not adequate. But I think that the quantification problem
can be solved by referring to the order of the elements of constituent
order domains (basically giving up the strict separation of meaing
composition/linearization).

But I think that there is data that poses a more severe problem for
linearization approaches: Examples where there seem to be two (or more)
constituents infront of the finite verb in German.

(1) [Dauerhaft] [mehr Arbeitspltze] gebe es erst, wenn sich eine
Wachstumsrate von mindestens 2,5 Prozent ber einen Zeitraum von drei
oder vier Jahren halten lasse. taz, 19.04.2000, p. 5
      lasting     more jobs           give it first when self a
growth.rate from at.least   2.5 percent over a     period from three or
four years hold let

`A long-term fall in unemployment can only be expected if a growth rate
of at least 2.5 percent can be maintained over a period of three or four
years.'

Linearization based theories cannot explain why the material infront of
the finite verb behaves like a verbal projection. Theories that assume
verb movement can use the same machinery to account for the fronted
verbal projection (dauerhaft mehr Arbeitsplätze) as is used for verb
movement in general (verb-initial vs. verb-last).

(2) a. weil Klaus das Buch kennt.
     b. Kennt_i Klaus das Buch _i?

I agree on Carl's point regarding empty determiners vs. unary branching
rules, but I did not understand the statement that there is no
difference between flat and binary branching structures. If we account
for (2) by assuming flat structures, we can save the empty element in
(2b), but if we do not have this empty element, we have no way to
account for multiple frontings as in (1). The data shows that a variety
of material can appear there (adjuncts and arguments, several arguments,
several adjuncts). If we assume flat structures or linearization domains
and if we want to capture the fact that the material infornt of the
finite verb behaves like a verbal projection, we have two options:

1) We stipulate a special purpose empty head. (This is what I did in
earlier proposals).

2) We stipulate tons of headless constructions.

The option 2 is the result of a partial evaluation of all immediate
dominance schemata with the empty verbal head that one would assume in
option 1. (Basically the same trick as Carl mentioned with regard to
empty determiners).

If we have a trace for verb movement as in (2b), we can use the very
same trace to account for the multiple frontings, i.e. we do not have to
stipulate it for this special phenomenon.

Another disadvantage of accounts that treat constituent order via
linearization domains is that they cannot account for the sentences in
(3) in obvious ways:

(3) a. [Den Wählern erzählen] sollte man diese Geschichte nicht.
     b. [Märchen erzählen] sollte man den Wählern nicht.

erzählen is usually subcategorized as <NP[nom],NP[dat],NP[acc]> or
<NP[nom],NP[acc],NP[dat]>. If one accounts for order variation in the
Mittelfeld by free linearization in an order domain. This does not give
an account for (3). If we asume the first subcat frame, we can analyze
(3b) but not (3a), since the dative cannot be combined with the verb
directly. If we assume the second frame, we cannot analyze (3b).

Andreas solved this problem by assuming both subcategorization frames
(He used shuffle in the representation). However this results in
spurious ambiguities in the analysis of sentences like (4):

(4) weil man den Wählern diese Geschichte nicht erzählen sollte.

The reason is that we now have too much freedom. We have freedom in
choosing an element from the subcat list and freedom in serialization.

The solution that I used in implementations was a special purpose rule,
which is a hack.

So, currently, I really believe that there is only one option left for
treating German: the verb movement analysis.

If you want to read more about this you may consider the following papers:

The data:
http://www.cl.uni-bremen.de/~stefan/Pub/mehr-vf-ds.html

The analysis:
http://www.cl.uni-bremen.de/~stefan/Pub/mehr-vf-lb.html

Includes a discussion of previous proposals to verb position and
constituent order in German.

Unfortunately these papers are in German. An earlier version in English
can be found in the Formal Grammar paper at:

http://www.cl.uni-bremen.de/~stefan/Pub/mehr-vf.html

In this early paper I assumed a linearization-based account + empty head
for the Vorfeld material. But as mentioned above, assuming such an empty
head in a linearization setting is just a stipulation. The paper also
differs from the final version in that I managed to get rid of VCOMP as
additional selectional feature. Therefore it becomes possible to treat
optional coherence in verbal complexes properly and the lexical rule
that licences multiple frontings is a subcase of the general verb
movement rule.

Greetings

	Stefan

@TechReport{Frank94,
   author      =	{Anette Frank},
   title	      =	{Verb Second by Lexical Rule or by Underspecification},
   institution =	{IBM Deutschland GmbH},
   address     =	{Heidelberg},
   type	      =	{Arbeitspapiere des SFB 340},
   number      =	{\Nr 43},
   note	      =
{\url{ftp://ftp.ims.uni-stuttgart.de/pub/papers/anette/v2-usp.ps.gz}.
		\urlchecked{20}{08}{2002}},
   year	      =	1994
}

@Book{Kathol2000a,
   author      =	{Andreas Kathol},
   title	      =	{Linear Syntax},
   publisher   =	{Oxford University Press},
   address     =	{New York, Oxford},
   optnote     =	{\url{http://www.oup.co.uk/isbn/0-19-823734-0}.
\urlchecked{18}{08}{2002}},
   year	      =	2000
}

@Book{Kiss95a,
   author      =	{Tibor Kiss},
   title	      =	{{Infinite Komplementation. Neue Studien zum deutschen
Verbum infinitum}},
   publisher   =	{Max Niemeyer Verlag},
   address     =	{T\"ubingen},
   series      =	la,
   number      =	{333},
   preis	      =	{118,- DM},
   optisbn     =	{3484303336},
   year	      =	1995
}

@Book{Kiss95b,
   author      =	"Kiss, Tibor",
   title	      =	"Merkmale und Repr{\"a}sentationen",
   address     =	"Opladen/Wiesbaden",
   publisher   =	"Westdeutscher Verlag",
   year	      =	"1995"
}

--
Stefan Müller

Universität Bremen/Fachbereich 10      Tel: (+49) (+421) 218-8601
Postfach 33 04 40
D-28334 Bremen

http://www.cl.uni-bremen.de/~stefan/

http://www.cl.uni-bremen.de/~stefan/Babel/Interaktiv/