empty elements

Tue Aug 28 15:13:28 UTC 2001

Hi Berthold,

Berthold Crysmann wrote:
> while I do share your view that they are translatable, the issue is whether
> one really would want to do so, I mean from LR-based  to morpheme-based
> ....

I do not. But there are people who claim that it is impossible.

> >
> > Having an (empty) affix is similar to having empty element in syntax.
>
> Similar yes, but there are certain differences which I believe make the
> option of (zero-)morphemes quite unattractive: Typically, in syntax, there
> is only a single though largely underspecified empty element (e.g. trace)
> the porperties of which are filled by syntactic context in a principled
> way. In morpheme-based morphology, however, the purpose of introducing zero
> element is quite a different one: here, properties are intrinsic to the
> empty element (e.g. empty plural morphemes, empty agreement markers, empty
> tense/aspect morphemes, empty category converters.... ), and what one will
> end up with is a plethora of homophonous (all zero) , yet
> categorically/featurally  distinct morphemes. (Add to this the necessity of
> adding black hole morphemes, for subtractive morphology and the like).  It
> is probably quite telling that noone, as far as I'm aware, has introduced
> empty heads into HPSG. But this is exactly what people in morpheme-based
> morphology are doing. Worse, while in syntax empty elements are always
> filled by the context, this is not necessarily true of morphology.

Okay, one instance of an empty element that projects syntactic and
semantic information that is different from the elements it combines
with and that is not determined by its syntactic environment is the
empty relativizer that was suggested in PS94. Ivan showed how relative
clauses can
be analyzed with out the empty head in a construction based fashion. In
my German grammar I also assume a grammar rule instead of the empty
head. (In fact this makes it possible to change the feature geometry and
certain interactions of nonlocal dependencies are simplified. See
Mueller (1999, Ch. 10.3.2, 10.5))

As Emily noted there are also proposals for empty nominal heads (see for
instance Klaus Netter's dissertation and Nerbonne & Mullen).

In the Verbmobil grammar I cover the empty N/empty Det data by unary
projections.

The status of empty elements in these constructions has been discussed a
long time ago by Wunderlich (1987) and Olsen (1987). And Wunderlich also
suggested a unary projection.

> I feel that the equivalence is  only of a very  technical nature, then....

That is true. Things get really nasty if you have a grammar with several
empty elements and you want to get rid of them by using (unary)
projections. Consider a grammar for German that assumes a trace for
extraction and an empty element for verb movement.

For `She knows him' you get something like

(1) Ihn_i kennt_j sie _i _j.

The way to eleminate the verb movement trace is to assume schemata that
project potential complements or potential adjuncts of verbs into v's.

One way to get rid of the extraction trace is to project a head and
saturate the last element from its subcat list and put it into slash
(that's what the trace would do).

Since the verb movement trace cannot be applied to the extraction trace
(we want to eliminat it), we need a new rule for (1) that does two
things at once. It introduces the verb movement together with a nonlocal
dependency. Instead of projecting a verb form the accusative object (_i)
we project a verb from the subject (sie) that contains something in
slash.

While this works technically the actual rules are hardly understandable
for anybody except the one who wrote them.

Plus we need slash introduction rules for all schemata were a trace
could appear. Head-Complement, Head-Adjunct, Head-Cluster. These rules
yield other rules that incorporate the effect of the verb movement
trace.

However, my point is, that the accounts are transferable into each
other,
although this sometimes gets technical.

In the literature one finds very often arguments that say it is not
possible at all, but I think this is false. In the end it is a matter of
aesthetics in many cases.

Greetings

	Stefan

@Book{Mueller99a,
  author      =	{Stefan M{\"u}ller},
  title	      =	{{Deutsche Syntax deklarativ. Head-Driven Phrase
Structure Grammar f\"ur das
		Deutsche}},
  publisher   =	{Max Niemeyer Verlag},
  address     =	{T\"ubingen},
  series      =	{Linguistische Arbeiten},
  number      =	394,
  note	      =	{\url{http://www.dfki.de/~stefan/Pub/e_hpsg.html}},
  year	      =	"1999"
}

@Book{Netter98,
  author      =	"Klaus Netter",
  title	      =	"Functional Categories in an {HPSG} for {German}",
  series      =	{Saar\-br{\"u}\discretionary{k-}{k}{ck}en Dissertations
in Computational Linguistics
		and Language Technology},
  number      =	3,
  publisher   =	{Deutsches Forschungszentrum f\"ur K\"unstliche
Intelligenz und
		Uni\-ver\-si\-t{\"a}t des Saarlandes},
  address     =	{Saar\-br{\"u}\discretionary{k-}{k}{ck}en},
  year	      =	"1998",
}

@InCollection{NM99a,
  author      =	{John Nerbonne and Tony Mullen},
  title	      =	{Null-Headed Nominals in {German} and {English}},
  booktitle   =	{Computational Linguistics in the Netherlands 1998},
  editor      =	{Frank {van Eynde} and Ineke Schuurman and Ness
Schelkens},
  pages	      =	{143--164},
  note	      =
{\url{http://grid.let.rug.nl/~nerbonne/papers/null-n-lp-CLIN98.ps}.
		\urlchecked{27}{12}{99}},
  year	      =	1999
}

@Article{Olsen87a,
  author      =	{Susan Olsen},
  title	      =	{{Das substantivierte Adjektiv im Deutschen:
Attributierung vs.\ syntaktische
		Substantivierung}},
  journal     =	{Folia Linguistica},
  untertitel  =	{Acta Scietatis Linguisticae Europaeae},
  year	      =	1987
}

@Article{Sag97a,
  author      =	"Ivan A. Sag",
  title	      =	"English Relative Clause Constructions",
  journal     =	"Journal of Linguistics",
  volume      =	33,
  number      =	2,
  pages	      =	{431--484},
  note	      =
{\url{ftp://ftp-csli.stanford.edu/linguistics/sag/rel-pap.ps.gz}.
		\urlchecked{13}{04}{97}},
  year	      =	1997
}

@Article{Wunderlich87d,
  author      =	{Dieter Wunderlich},
  title	      =	{{Vermeide Pronomen -- Vermeide leere Kategorien}},
  journal     =	{Studium Linguistik},
  volume      =	21,
  pages	      =	{36--44},
  year	      =	1987
}

--
PD Dr. Stefan Müller
Institut für Germanistische Sprachwissenschaft
Friedrich Schiller Universität Jena
Fürstengraben 30
D-07743 Jena

Tel: (+49) (+3641) 9 44 320     http://www.dfki.de/~stefan/
Fax: (+49) (+3641) 9 44 322
http://www.dfki.de/~stefan/Babel/Interaktiv/Babajava/