Passive voice in construction-style HPSG

[Stefan Müller]

Hi,

> I'm looking for a description of the passive voice (in English or other
> languages) done in construction-grammar style HPSG.

If you are looking for something in terms of Linking Constructions, as
they were mentioned in Kay/Fillmore 1999

http://www.icsi.berkeley.edu/~kay/wxdy.ps

and explained in more detail in Michaelis/Ruppenhofer, 2000

http://cslipublications.stanford.edu/site/1575863308.html

you will probably not find anything.

The reason is that the understanding of sets that is assumed in these
publications is different from the one assumed in HPSG (Pollard/Sag,
1987, p. 47--49). Kay and Fillmore assume that the unification of two
sets containing incompatible information is the set union:

{ a } ^ { b } = { a, b }

In the HPSG understanding of sets the unification of { a } ^ { b } is
not defined unless a and b are unifyable.

The Linking Constructions in Michaelis/Ruppenhofer specify a singleton
element of the valence list. So for instance

(1) [ val { [ da +,
               syn 0/PP ] } ]

says that a valence element that is specified to be the designated
argument (roughly the subject of the active clause) is not realized or
is realized as oblique PP. (I do not have the book here, so maybe the
formulation is slightly different).

So this specification is unified with an existing lexical entry. If a
[da +] element exists in the valence set of the lexical entry, it has to
be realized as PP or is suppressed.

Note, that this analysis has problems with subjectless verbs like German
  `grauen' (`to dread'). If the passive linking construction is combined
with `grauen' the unification of the valence sets would result in a
bigger set that includes a designated argument that can be realized as a
PP. To fix this, one had to have a means of saying: I am not allowed to
introduce new material, but have to match existing material.

On the other hand the linking constructions have to be able to intrdocue
material into valence sets. For instance the German middle constructions
contains a reflexive pronoun.

(2) a. Er     liest den Aufsatz.
        he-nom reads the essay-acc

     b. Der Aufsatz   liest sich leicht.
        the essay-nom reads self easily

So the Middle Linking Construction had to extend the valence list of
`liest' by introducing `sich'.

The Linking Construction approach could be made compatible to the HPSG
understanding of sets by reformulating the Linking Constraints as
membership tests.

So the Linking Construction in (1) would be:

(3) [ val { [ da +,
               syn 0/PP ] } u X ]

That is, we would say, that the valence set of a certain item is a set
that consists of one element that is the designated argument plus a
possibly empty other set X (`u' stands for set union). This approach
would not wrongly introduce arguments, but it would also fail to explain
the German middle, since in the HPSG understanding of sets, sets cannot
be extended by unification.

In HPSG, passive is usually done by lexical rules (Pollard & Sag, 1987).

Alternative proposals based on argument attraction techniques were
suggested by Kathol and by Pollard. Both papers are contained in the
HPSG for German book from 1994 (CSLI).

> I imagine this
> could be found in the LinGO source code, but I don't have access to that
> right now.

LinGO (2004-10-17) does it by a lexical rule as well.

> Are there papers or books out there where such a description
> has been published?

You may read a whole chapter about passive, modal infinitives and
interactions with adjective derivation in

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

The analysis there uses blocking/deblocking techniques first suggested
by Hubert Haider in several papers in the eighties. It is a lexical rule
based analysis in which the auxiliaries play a crucial role (The
deblocking of blocked arguments is done by auxiliaries). The analysis
differs from the analysis in the LinGO grammar in that it has only one
lexical item for the participle that is used both in perfect and in
passive sentences.

I also discuss Kathol's and Pollard's proposals there.


Best wishes


	Stefan



--
Stefan Müller

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

http://www.stefan-müller.net

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