• The Morphology of German Particle Verbs: Solving the ...


  •   
  • FileName: Mueller-pn.pdf [read-online]
    • Abstract: [ SUBCAT. CONT 1. SUBJ. particle. SUBCAT. VCOMP. CONT. ARG. 1. begin. local. This particle modifies an ... SUBCAT. 3. CONT 4. stem. 2nd-inflected-verb. This lexical rule produces a finite form from the ...

Download the ebook

The Morphology of German Particle Verbs:
Solving the Bracketing Paradox
Stefan Müller
DFKI GmbH
Friedrich-Schiller-Universität Jena
Language Technology Lab
Fürstengraben 30
Stuhlsatzenhausweg 3
D-07743 Jena, Germany
66 123 Saarbruecken, Germany
E-mail: [email protected]
November 15, 2001
1 Introduction
Inflectional affixes are sensitive to morphological properties of the elements they attach
to. Therefore Bierwisch (1987) suggested that the inflectional material is combined
with both the verbal stem of simplex verbs and the verbal stem of particle verbs. He
argued that this leads to the bracketing paradox in the case of particle verbs since the
semantic contribution of the inflectional information scopes over the semantic contri-
bution of the complete particle verb. This paradox will be discussed in section 2.1.
In section 2.2, I will discuss nominalizations and adjective derivation, which are also
problematic because of various bracketing paradoxes. In section 3 I will suggest a solu-
tion to these apparent paradoxes that assumes that inflectional and derivational prefixes
and suffixes always attach to a form of a stem that contains the information about a pos-
sible particle already, but without containing a phonological realization of the particle.
The particle is a dependent of the verb and is combined with its head after inflection
and derivation. With such an approach no rebracketing for the analysis of particle verbs
mechanisms are necessary.
2 The Phenomena
2.1 Inflection
Both particle and prefix verbs always have the same inflection class as their base verb.
This means that the inflectional suffix has to have access to the morphological features
of the stem. This is accounted for easily with a structure like the one in figure 1a.
Bierwisch (1987, p. 163) argues that the meaning of the verb aufhören (‘end’) is not
Proceedings of the 8th International HPSG Conference, Norwegian University of Science and
Technology (3-5 August 2001)
Frank van Eynde, Lars Hellan and Dorothee Beermann (editors)
2002.CSLI Publications. http://csli-publications.stanford.edu/
247
a. V b. V
P V V en
auf V en P V
hör auf hör
Figure 1: Alternative Structures for aufhören
transparent with regard to the combination of auf and hör-, but combinations of the
form auf-hör-t-est and auf-ge-hör-t are transparent with regard to the combination of
the meaning end and the conceptual content of the inflectional affixes. He claims that
one needs structures like the one in figure 1b because of this, and hence he has a struc-
tural paradox. Bierwisch (1987, p. 165) and Stiebels (1996, p. 46) suggest rebracketing
mechanisms to derive the structure in figure 1a from the one in figure 1b. However,
this paradox is not a real one, since the situation with idioms is similar as far as com-
positionality is concerned.1 It is not justified that a head that is part of an idiomatic
expression is combined with all parts of the idiom before it is inflected. So one can
stick to the structure in figure 1a; assuming that the semantics of non-transparent par-
ticle verbs is constructed parallel to the semantics of idioms.
For transparent particle verb combinations I also assume the structure in figure 1a. I
assume that the inflectional affix attaches to a stem that contains the information that it
will combine with a particle, i.e., a stem that is subcategorized for a particle. This stem
has the meaning of the complete particle verb combination although the exact meaning
is not fully instantiated until the particle combines with the (inflected) stem. Since the
semantic information that will be contributed by the particle is accessible in the stem
entry already, the ending can scope over it.
2.2 Derivation
Similar bracketing paradoxes seem to arise in derivational morphology. Some deriva-
tional affixes are sensitive to the argument structure of the head they combine with and
some others are sensitive to the semantics of the heads they combine with, some affixes
1 Bierwisch gives examples from compounding that suggest that rebracketing may be needed and, of
course, there are famous examples of a similar kind from English; but for the cases at hand a rebracketing
mechanism is not necessary as will be shown in section 3.1.
Stump (1991) discusses a wide variety of morphosemantic mismatches in English, Breton, Georgian,
and Sanskrit and suggests paradigm functions that allow inflectional or derivational material to attach to a
head that is contained inside of other material, i.e., he assumes a structure like figure 1b. On page 714 he
remarks that in derivational paradigms in which the derived member belongs to a syntactic category distinct
from that of the base member, the derived member generally fails to allow this kind of structure where
the inflectional or derivational material attaches to the head. He remarks that nouns derived from particle
verbs are exceptions (hang on → hanger on, pass by → passer-by). With my analysis particle verbs can be
analyzed without a paradox and therefore they do not constitute an exception to his generalization. I will
discuss his approach in section 4.2.
248
are sensitive for both kinds of properties. In sections 2.2.1 and 2.2.2, I will examine
the relevant forms of nominalization and adjective formation.
2.2.1 Ge- -e-Nominalizations
The Ge- -e-nominalization is the only discontinuous or combinatorial noun derivation
in German, consisting of the prefix Ge- and the suffix -e. Ge- -e-derivation is quite pro-
ductive for transitive as well as for intransitive simplex verbs. Deverbal Ge- -e-nouns
have the meaning of ‘to V constantly/repeatedly’ and usually they have the connotation
that the constant V-ing is unpleasant.
Particle verbs also allow for Ge- -e-derivation. It is interesting that the ge- separates
particle and base verb: Herumgerenne (‘repeated aimless running’). Ge- -e-nominal-
izations of particle verbs with the particle herum (‘around’) follow a productive pattern
and are quite frequent.
As Lüdeling (2001, p. 106) notes, the interesting thing about these Ge- -e-nomi-
nalizations is that there again seems to be a bracketing paradox: If one combines the
stem renn- with Ge- and -e one gets Gerenne, which means ‘repeated or constant run-
ning’, or more technically ‘repeated running events’. However, Herumgerenne means
‘repeated instances of aimless running events’. The ‘aimless’ part of the meaning is
contributed by herum. This meaning of Herumgerenne would be expected if the Ge- -e
were combined with the whole particle verb combination.
Lüdeling considers for a moment the introduction of an abstract predicate for the
form of rennen, but dismisses this suggestion since, according to her, this solution
would not extend to listed particle verb combinations. I do not understand this argu-
mentation, since the non-transparent forms are always the unproblematic ones in terms
of scope relations. The particle verb selects the particle and the complete semantics
is represented at the entry of the verb. Lüdeling suggests the analysis in figure 2b. It
a. N b. N
P N V Ge- -e
herum V Ge- -e P V
renn herum renn
Figure 2: Alternative Structures for Herumgerenne
is unclear how the prefix ge- is supposed to get in-between the particle and the verb
without the assumption of rebracketing. In what follows I will assume the structure in
figure 2a. I assume that the stem renn- that is used to derive Herumgerenne already
contains the information that it combines with a particle, although the exact semantic
and syntactic contribution of the particle is still underspecified. The Ge- -e-nominal-
ization can therefore access the semantic contribution that will be instantiated by the
particle and the right scope relations can be established.
249
2.2.2 Adjective Derivation with -bar
-bar-derivation applies to transitive or ditransitive verbs that have an accusative object.
The logical subject of the verb is suppressed and the accusative object is promoted
to the subject of the adjective. There are also a few -bar-adjectives like brennbar
(‘flammable’) that have an intransitive base verb, but these are listed in the lexicon
Riehemann (1998a) and not derived by the productive rules. The -bar-suffix adds a
modal meaning, usually possibility, but sometimes also necessity.
Particle verb combinations that are the result of a productive process can take part
in -bar-derivations as is shown by (1).
(1) Die Geschäfte müssen anfahrbar bleiben.
the shops must PART (to).drivable remain
‘The shops must remain accessible by car.’
The particle an combines with intransitive verbs and licenses an additional argument.
(2) a. Er fährt.
he drives
b. * Er fährt die Geschäfte.
he drives the shops
c. Er fährt die Geschäfte an.
he drives the shops towards
‘He drives towards the shops.’
As (2b) shows die Geschäfte is not an argument of the base verb fährt. This NP is
licenced in the particle verb construction only. The pattern is productive.
This seems to result in another bracketing paradox: There are particles that only
combine with intransitive verbs and add another argument. On the other hand, -bar
combines only with transitive verbs productively. If one assumes the structure in fig-
ure 3a with fahr- being the stem of the intranitive version of fahren one has to explain
why -bar can combine with intransitive verbs. Furthermore the modal operator that is
contributed by -bar scopes over the complete meaning of the particle verb. In the light
of pairs like (3), the structure in figure 3a seems implausible, since there is no way of
deriving the meaning of the second word from the meaning of the first:
(3) a. schaffbar (‘do-able’) wegschaffbar (‘possible to be got rid of’, ‘dispos-
able’)
b. greifbar (‘reachable’) angreifbar (‘possible to be attacked’)
c. stellbar (‘possible to stand/set up’) darstellbar (‘possible to be repre-
sented’, ‘representable’), einstellbar (‘possible to set’, ‘employable’), her-
stellbar (‘possible to manufacture’), vorstellbar (‘imaginable’)
Even worse, a bar-adjective without particle does not exist for the examples in (4).
(4) a. * guckbar anguckbar (‘possible to look at’)2
2 (Bierwisch, 1987, p. 163)
250
b. * gleichbar ausgleichbar (‘possible to even out’)
c. * weisbar nachweisbar (‘possible to prove’)
At first glance figure 3b seems to be the only option. For reasons of uniformity I will
assume the structure in figure 3a. While at first glance this may seem to be problematic
a. A b. A
P A V bar
an V bar P V
fahr an fahr
Figure 3: Alternative Structures for anfahrbar (‘reachable by car’)
for the reasons mentioned above, it is not in constraint-based theories. I assume that
the stem in figure 3a contains a slot for the particle that will be added in a later step.
The valence and the semantics of the whole combination is represented at the stem so
that -bar may access it.
3 The Analysis
In (Müller, 2000a,b, To Appear) I discussed constituent order data that showed that
particles behave like other elements that take part in predicate complex formation
(verbal complexes, subject and object predicative constructions, and resultative con-
structions).3 I therefore suggested a complex predicate analysis that uses argument
attraction techniques which were introduced into the HPSG framework by Hinrichs
and Nakazawa (1989). I followed (Chung, 1993; Rentier, 1994; Müller, 1997), who
suggested using a separate valence feature for the selection of dependents that are part
of the predicate complex. So particles are selected by their head via the valence feature
VCOMP . Verbs that can function as heads in productive particle verb combinations are
licensed by the following lexical rule:
3 Tilman Höhle suggested using the same rule for the combination of particle and verb as for the verbal
complex in his 1976 dissertation. The chapter of his dissertation that deals with this issue was published as
Höhle (1982). Höhle deals mainly with morphological problems. The syntactic properties of the particle
verb constructions are not explored in detail.
251
(5) Lexical Rule for Productive Particle Verb Combinations:
   
SUBCAT 2 ⊕ 3
      
 
  MOD 1 
        
    
 HEAD SUBJ 2  

  
    
 

 
CAT  
LOC CAT 
 particle 
 
 SYNSEM|LOC  VCOMP     

     SUBCAT 3  
      
 
    VCOMP  
    
 
  CONT 4 
  
 
 CONT 4 
   
 
 LOC | CAT
HEAD verb 
 SYNSEM 1  
 LEX - DTR  VCOMP  
   
 
 stem 
stem
The rule applies to all verbs with an empty VCOMP value. The output of the rule is a
verb that selects a particle. Whether the resulting verb is actually used in an analysis
depends on the presence of a particle that can be combined with this verb. The valence
requirements of the output verb are determined by the particle: The SUBCAT and SUBJ 4
value of the particle attracted by the output verb. The rule licenses verbal stems that
select particles that modify the base verb semantically. This is indicated by the structure
sharing of the MOD value of the particle and the SYNSEM value of the input verb ( 1 ).5
Particles like los in losrennen (‘start to run’) and an in anfahren (‘drive towards
sthg.’) have the form of adjuncts. They select their head via MOD. The entry for los is
shown in (6).
(6) los (aspectual marker):
    
MOD V[SUBCAT , CONT 1]
    
 HEAD SUBJ  
CAT  
 particle 
  
  
 SUBCAT 
 
 VCOMP 
 
 
CONT ARG 1 
 begin

 
local
This particle modifies an intransitive verb (SUBCAT = ) and encapsulates the seman-
tics of this verb ( 1 ) under the relation it contributes (begin). When lexical items that are
licensed by the lexical rule in (5) are combined with the particle, they take the semantic
contribution from the particle. This is ensured by the structure sharing 4 in (5).
4Ifollow (1992; 1995) in assuming that SUBJ is a head feature.
5 Thisrule is in a certain way similar to the adjunct introduction lexical rule that van Noord and Bouma
(1994) use: An adjunct is introduced into a valence feature list.
252
As an example, consider what happens, if the lexical rule applies to the entry of the
base verb rennen (‘to run’).
(7) renn- (‘run’):
 

SUBJ NP[str] 1
 HEAD 
CAT  verb 
 
  
 SUBCAT 
 
 VCOMP 
 
 
 AGENT 1 
CONT 
 rennen 
local
The result is shown in (8). This entry has to be inflected in order to be usable in syntax.
Instead of undergoing inflectional rules, (8) may be input to derivational rules.
(8) renn- (‘run’ + subcategorized for particle):
 
phon renn
synsem|loc 
   
 subj NP[str ] 1 
 

  head verb  
 
    
  subcat  
   2 ⊕ 3
        
    
   subj NP[str ] 1
 
        
h
   
       c  verb    
         subcat    
           
      mod|l     
        vcomp    
     h      
 c          
    c   agent 1
   
   l    cont    
  vcomp     rennen    
          
         
      subj 2
  
        
      particle
  
     subcat 3   
       
       
     vcomp   
  
  
  cont 4 
  
 cont 4 
 
 local 
stem
In the following I will use the entry in (8) to explain the syntactic combination of
particle and verb. When the inflected form of the entry in (8) is combined with the
particle in (6), the structure under CAT|VCOMP gets instantiated in the following way:
253
(9) rennen (‘run’ + particle los result of the unification in VCOMP):
 
  
VFORM bse
 

 HEAD SUBJ


NP[str] 1
 
 
  
  verb 
  
 SUBCAT 2 ⊕ 3 
         
  
  SUBJ NP[str] 1 
       H  
        
      CAT  verb  

        
      SUBCAT  
     MOD|LOC   
        
       VCOMP  
CAT 
   H 
 
  

      
   CAT   CONT 4 AGENT 1  
      rennen  
  
 LOC     
 VCOMP     
      
     SUBJ 2  
       
    particle   
    

 
  
    
    SUBCAT 3   
     
    VCOMP  
     
    
     
  CONT 5 ARG 4  
 begin
 
 
 
CONT 5 
local
The information that was added by the particle is the structure sharing 4 between the
semantics of the original base verb that was the input to the lexical rule (5) and the
argument of the relation contributed by the particle. The semantics of the combination
of rennen and los is taken from the adjunct ( 5 ) and is also represented as the semantics
of the complete combination. The SUBJ value of los is raised to the SUBCAT list of
rennen. Since los does not have a subject, the combination of los and rennen remains
intransitive. The result of combining the particle with the verb is shown in (10).
(10) rennen (‘run’ + Particle los):
  
VFORM bse
   
 HEAD SUBJ NP[str] 1 
CAT 



  verb 
  
 SUBCAT 
 
 VCOMP 
 
   
 AGENT 1 
 ARG 
CONT  rennen 
 
 
 
 begin 
local
If one combines (8) with the an that is used in anfahren instead of los one gets a
different result, since the lexical entry of this an differs from the entry for los in that it
has an element on SUBJ and therefore introduces an additional element to the SUBCAT
list of the verb it combines with.
254
3.1 Morphology
There are two basic approaches to inflectional and derivational morphology. The first
is called “Item-and-Arrangement (IA) approach”, “Morpheme-based approach”, or
“Word Syntax approach”. It is assumed that words consist of morphems that are form
meaning pairs. Such morphems are combined in a way that is similar to what is known
from syntax. The alternative proposal is called “Item-and-Process (IP) approach”. Here
it is assumed that stems are related to other stems or to words by realizational rules.
Affixes are not elements of the lexicon. The phonological material that is contributed
by an affix in the Item-and-Arrangement model is introduced in the process that derives
a form from a given stem. For a comparison of the two approaches see (Hockett, 1954)
and (Anderson, 1988).
As an example consider the inflected form fragt (’asks’) which consists of the stem
frag- and the ending -t. In a morpheme-based approach both the stem and the ending
are morphems and it is assumed that both bear meaning. The word fragt has the struc-
ture frag + t. In a Item-and-Process approach there is no lexical entry for -t. Instead
the form fragt is licensed by a process that relates the stem to the fully inflected word
(frag ⇒ fragt). The information that t is an appropriate ending for the present tense is
contained in the definition of the relation that relates the stem to the word.
In the HPSG paradigm both Item-and-Arrangement and Item-and-Process analyses
have been developed: Trost (1991), Krieger and Nerbonne (1993), Krieger (1994),
van Eynde (1994, Chapter 4), and Lebeth (1994) suggest an affix-based approach and
Pollard and Sag (1987, Chapter 8.2), Orgun (1996); Riehemann (1998b); Ackerman
and Webelhuth (1998); Kathol (1999), Koenig (1999)),6 use lexical rules that relate
stems to other stems or words.
One advantage of the IP view is that one does not have to stipulate zero morphems
for cases of zero inflection or conversion. Another advantage is that the stipulation of
subtractive morphems is not necessary. Hockett (1954, p. 224) discusses cases from
Chinese and French where a shorter form is regarded as derived from a longer more
basic one (bon vs. bonne is the French example). A morpheme-based analysis would
have to stipulate an abstract entity that has some meaning, but no phonological form.
If it is combined with some other element, phonological material of this element is
deleted. In the IP view on the other hand there is just a mapping from bonne to bon
and the fact that something is deleted is encoded in this mapping. In what follows I
therefore suggest a lexical rule-based analysis.
3.1.1 Inflection
The lexical rule in (11) is used to derive inflected lexical items from entries that are
listed in the lexicon or that have been derived by other lexical rules that map uninflected
lexical items to other uninflected lexical items. So it can be used to derive rennst from
various forms of renn- (‘run’). One entry for renn- is the one that is listed. Another
one is derived by the rule for productive particle verb combinations (see (5) on page 5),
and can be used in sentences like er rennt los (‘he starts to run’).
6 For non-HPSG-based approaches see for instance (Dowty, 1979, p. 304; Stump, 1991; Aronoff, 1994).
255
(11) Lexical rule for the 2nd person singular, present:
 
PHON 1 ⊕ st  
    
 VFORM fin 
   HEAD   
    SUBJ  
  CAT   
    verb  
 
SYNSEM
 LOC 
  

   SUBCAT 2 ⊕ 3  
  
    
  CONT PSOA 4  
   


Use: 0.4213